1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * NET3 Protocol independent device support routines.
5 * Derived from the non IP parts of dev.c 1.0.19
7 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
8 * Mark Evans, <evansmp@uhura.aston.ac.uk>
11 * Florian la Roche <rzsfl@rz.uni-sb.de>
12 * Alan Cox <gw4pts@gw4pts.ampr.org>
13 * David Hinds <dahinds@users.sourceforge.net>
14 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
15 * Adam Sulmicki <adam@cfar.umd.edu>
16 * Pekka Riikonen <priikone@poesidon.pspt.fi>
19 * D.J. Barrow : Fixed bug where dev->refcnt gets set
20 * to 2 if register_netdev gets called
21 * before net_dev_init & also removed a
22 * few lines of code in the process.
23 * Alan Cox : device private ioctl copies fields back.
24 * Alan Cox : Transmit queue code does relevant
25 * stunts to keep the queue safe.
26 * Alan Cox : Fixed double lock.
27 * Alan Cox : Fixed promisc NULL pointer trap
28 * ???????? : Support the full private ioctl range
29 * Alan Cox : Moved ioctl permission check into
31 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
32 * Alan Cox : 100 backlog just doesn't cut it when
33 * you start doing multicast video 8)
34 * Alan Cox : Rewrote net_bh and list manager.
35 * Alan Cox : Fix ETH_P_ALL echoback lengths.
36 * Alan Cox : Took out transmit every packet pass
37 * Saved a few bytes in the ioctl handler
38 * Alan Cox : Network driver sets packet type before
39 * calling netif_rx. Saves a function
41 * Alan Cox : Hashed net_bh()
42 * Richard Kooijman: Timestamp fixes.
43 * Alan Cox : Wrong field in SIOCGIFDSTADDR
44 * Alan Cox : Device lock protection.
45 * Alan Cox : Fixed nasty side effect of device close
47 * Rudi Cilibrasi : Pass the right thing to
49 * Dave Miller : 32bit quantity for the device lock to
50 * make it work out on a Sparc.
51 * Bjorn Ekwall : Added KERNELD hack.
52 * Alan Cox : Cleaned up the backlog initialise.
53 * Craig Metz : SIOCGIFCONF fix if space for under
55 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
56 * is no device open function.
57 * Andi Kleen : Fix error reporting for SIOCGIFCONF
58 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
59 * Cyrus Durgin : Cleaned for KMOD
60 * Adam Sulmicki : Bug Fix : Network Device Unload
61 * A network device unload needs to purge
63 * Paul Rusty Russell : SIOCSIFNAME
64 * Pekka Riikonen : Netdev boot-time settings code
65 * Andrew Morton : Make unregister_netdevice wait
66 * indefinitely on dev->refcnt
67 * J Hadi Salim : - Backlog queue sampling
68 * - netif_rx() feedback
71 #include <linux/uaccess.h>
72 #include <linux/bitops.h>
73 #include <linux/capability.h>
74 #include <linux/cpu.h>
75 #include <linux/types.h>
76 #include <linux/kernel.h>
77 #include <linux/hash.h>
78 #include <linux/slab.h>
79 #include <linux/sched.h>
80 #include <linux/sched/mm.h>
81 #include <linux/mutex.h>
82 #include <linux/rwsem.h>
83 #include <linux/string.h>
85 #include <linux/socket.h>
86 #include <linux/sockios.h>
87 #include <linux/errno.h>
88 #include <linux/interrupt.h>
89 #include <linux/if_ether.h>
90 #include <linux/netdevice.h>
91 #include <linux/etherdevice.h>
92 #include <linux/ethtool.h>
93 #include <linux/skbuff.h>
94 #include <linux/kthread.h>
95 #include <linux/bpf.h>
96 #include <linux/bpf_trace.h>
97 #include <net/net_namespace.h>
99 #include <net/busy_poll.h>
100 #include <linux/rtnetlink.h>
101 #include <linux/stat.h>
104 #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>
149 #include <linux/pm_runtime.h>
150 #include <linux/prandom.h>
152 #include "net-sysfs.h"
154 #define MAX_GRO_SKBS 8
156 /* This should be increased if a protocol with a bigger head is added. */
157 #define GRO_MAX_HEAD (MAX_HEADER + 128)
159 static DEFINE_SPINLOCK(ptype_lock);
160 static DEFINE_SPINLOCK(offload_lock);
161 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
162 struct list_head ptype_all __read_mostly; /* Taps */
163 static struct list_head offload_base __read_mostly;
165 static int netif_rx_internal(struct sk_buff *skb);
166 static int call_netdevice_notifiers_info(unsigned long val,
167 struct netdev_notifier_info *info);
168 static int call_netdevice_notifiers_extack(unsigned long val,
169 struct net_device *dev,
170 struct netlink_ext_ack *extack);
171 static struct napi_struct *napi_by_id(unsigned int napi_id);
174 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
177 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
179 * Writers must hold the rtnl semaphore while they loop through the
180 * dev_base_head list, and hold dev_base_lock for writing when they do the
181 * actual updates. This allows pure readers to access the list even
182 * while a writer is preparing to update it.
184 * To put it another way, dev_base_lock is held for writing only to
185 * protect against pure readers; the rtnl semaphore provides the
186 * protection against other writers.
188 * See, for example usages, register_netdevice() and
189 * unregister_netdevice(), which must be called with the rtnl
192 DEFINE_RWLOCK(dev_base_lock);
193 EXPORT_SYMBOL(dev_base_lock);
195 static DEFINE_MUTEX(ifalias_mutex);
197 /* protects napi_hash addition/deletion and napi_gen_id */
198 static DEFINE_SPINLOCK(napi_hash_lock);
200 static unsigned int napi_gen_id = NR_CPUS;
201 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
203 static DECLARE_RWSEM(devnet_rename_sem);
205 static inline void dev_base_seq_inc(struct net *net)
207 while (++net->dev_base_seq == 0)
211 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
213 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
215 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
218 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
220 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
223 static inline void rps_lock(struct softnet_data *sd)
226 spin_lock(&sd->input_pkt_queue.lock);
230 static inline void rps_unlock(struct softnet_data *sd)
233 spin_unlock(&sd->input_pkt_queue.lock);
237 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
240 struct netdev_name_node *name_node;
242 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
245 INIT_HLIST_NODE(&name_node->hlist);
246 name_node->dev = dev;
247 name_node->name = name;
251 static struct netdev_name_node *
252 netdev_name_node_head_alloc(struct net_device *dev)
254 struct netdev_name_node *name_node;
256 name_node = netdev_name_node_alloc(dev, dev->name);
259 INIT_LIST_HEAD(&name_node->list);
263 static void netdev_name_node_free(struct netdev_name_node *name_node)
268 static void netdev_name_node_add(struct net *net,
269 struct netdev_name_node *name_node)
271 hlist_add_head_rcu(&name_node->hlist,
272 dev_name_hash(net, name_node->name));
275 static void netdev_name_node_del(struct netdev_name_node *name_node)
277 hlist_del_rcu(&name_node->hlist);
280 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
283 struct hlist_head *head = dev_name_hash(net, name);
284 struct netdev_name_node *name_node;
286 hlist_for_each_entry(name_node, head, hlist)
287 if (!strcmp(name_node->name, name))
292 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
295 struct hlist_head *head = dev_name_hash(net, name);
296 struct netdev_name_node *name_node;
298 hlist_for_each_entry_rcu(name_node, head, hlist)
299 if (!strcmp(name_node->name, name))
304 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
306 struct netdev_name_node *name_node;
307 struct net *net = dev_net(dev);
309 name_node = netdev_name_node_lookup(net, name);
312 name_node = netdev_name_node_alloc(dev, name);
315 netdev_name_node_add(net, name_node);
316 /* The node that holds dev->name acts as a head of per-device list. */
317 list_add_tail(&name_node->list, &dev->name_node->list);
321 EXPORT_SYMBOL(netdev_name_node_alt_create);
323 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
325 list_del(&name_node->list);
326 netdev_name_node_del(name_node);
327 kfree(name_node->name);
328 netdev_name_node_free(name_node);
331 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
333 struct netdev_name_node *name_node;
334 struct net *net = dev_net(dev);
336 name_node = netdev_name_node_lookup(net, name);
339 /* lookup might have found our primary name or a name belonging
342 if (name_node == dev->name_node || name_node->dev != dev)
345 __netdev_name_node_alt_destroy(name_node);
349 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
351 static void netdev_name_node_alt_flush(struct net_device *dev)
353 struct netdev_name_node *name_node, *tmp;
355 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
356 __netdev_name_node_alt_destroy(name_node);
359 /* Device list insertion */
360 static void list_netdevice(struct net_device *dev)
362 struct net *net = dev_net(dev);
366 write_lock_bh(&dev_base_lock);
367 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
368 netdev_name_node_add(net, dev->name_node);
369 hlist_add_head_rcu(&dev->index_hlist,
370 dev_index_hash(net, dev->ifindex));
371 write_unlock_bh(&dev_base_lock);
373 dev_base_seq_inc(net);
376 /* Device list removal
377 * caller must respect a RCU grace period before freeing/reusing dev
379 static void unlist_netdevice(struct net_device *dev)
383 /* Unlink dev from the device chain */
384 write_lock_bh(&dev_base_lock);
385 list_del_rcu(&dev->dev_list);
386 netdev_name_node_del(dev->name_node);
387 hlist_del_rcu(&dev->index_hlist);
388 write_unlock_bh(&dev_base_lock);
390 dev_base_seq_inc(dev_net(dev));
397 static RAW_NOTIFIER_HEAD(netdev_chain);
400 * Device drivers call our routines to queue packets here. We empty the
401 * queue in the local softnet handler.
404 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
405 EXPORT_PER_CPU_SYMBOL(softnet_data);
407 #ifdef CONFIG_LOCKDEP
409 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
410 * according to dev->type
412 static const unsigned short netdev_lock_type[] = {
413 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
414 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
415 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
416 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
417 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
418 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
419 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
420 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
421 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
422 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
423 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
424 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
425 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
426 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
427 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
429 static const char *const netdev_lock_name[] = {
430 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
431 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
432 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
433 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
434 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
435 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
436 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
437 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
438 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
439 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
440 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
441 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
442 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
443 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
444 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
446 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
447 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
449 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
453 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
454 if (netdev_lock_type[i] == dev_type)
456 /* the last key is used by default */
457 return ARRAY_SIZE(netdev_lock_type) - 1;
460 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
461 unsigned short dev_type)
465 i = netdev_lock_pos(dev_type);
466 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
467 netdev_lock_name[i]);
470 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
474 i = netdev_lock_pos(dev->type);
475 lockdep_set_class_and_name(&dev->addr_list_lock,
476 &netdev_addr_lock_key[i],
477 netdev_lock_name[i]);
480 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
481 unsigned short dev_type)
485 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
490 /*******************************************************************************
492 * Protocol management and registration routines
494 *******************************************************************************/
498 * Add a protocol ID to the list. Now that the input handler is
499 * smarter we can dispense with all the messy stuff that used to be
502 * BEWARE!!! Protocol handlers, mangling input packets,
503 * MUST BE last in hash buckets and checking protocol handlers
504 * MUST start from promiscuous ptype_all chain in net_bh.
505 * It is true now, do not change it.
506 * Explanation follows: if protocol handler, mangling packet, will
507 * be the first on list, it is not able to sense, that packet
508 * is cloned and should be copied-on-write, so that it will
509 * change it and subsequent readers will get broken packet.
513 static inline struct list_head *ptype_head(const struct packet_type *pt)
515 if (pt->type == htons(ETH_P_ALL))
516 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
518 return pt->dev ? &pt->dev->ptype_specific :
519 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
523 * dev_add_pack - add packet handler
524 * @pt: packet type declaration
526 * Add a protocol handler to the networking stack. The passed &packet_type
527 * is linked into kernel lists and may not be freed until it has been
528 * removed from the kernel lists.
530 * This call does not sleep therefore it can not
531 * guarantee all CPU's that are in middle of receiving packets
532 * will see the new packet type (until the next received packet).
535 void dev_add_pack(struct packet_type *pt)
537 struct list_head *head = ptype_head(pt);
539 spin_lock(&ptype_lock);
540 list_add_rcu(&pt->list, head);
541 spin_unlock(&ptype_lock);
543 EXPORT_SYMBOL(dev_add_pack);
546 * __dev_remove_pack - remove packet handler
547 * @pt: packet type declaration
549 * Remove a protocol handler that was previously added to the kernel
550 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
551 * from the kernel lists and can be freed or reused once this function
554 * The packet type might still be in use by receivers
555 * and must not be freed until after all the CPU's have gone
556 * through a quiescent state.
558 void __dev_remove_pack(struct packet_type *pt)
560 struct list_head *head = ptype_head(pt);
561 struct packet_type *pt1;
563 spin_lock(&ptype_lock);
565 list_for_each_entry(pt1, head, list) {
567 list_del_rcu(&pt->list);
572 pr_warn("dev_remove_pack: %p not found\n", pt);
574 spin_unlock(&ptype_lock);
576 EXPORT_SYMBOL(__dev_remove_pack);
579 * dev_remove_pack - remove packet handler
580 * @pt: packet type declaration
582 * Remove a protocol handler that was previously added to the kernel
583 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
584 * from the kernel lists and can be freed or reused once this function
587 * This call sleeps to guarantee that no CPU is looking at the packet
590 void dev_remove_pack(struct packet_type *pt)
592 __dev_remove_pack(pt);
596 EXPORT_SYMBOL(dev_remove_pack);
600 * dev_add_offload - register offload handlers
601 * @po: protocol offload declaration
603 * Add protocol offload handlers to the networking stack. The passed
604 * &proto_offload is linked into kernel lists and may not be freed until
605 * it has been removed from the kernel lists.
607 * This call does not sleep therefore it can not
608 * guarantee all CPU's that are in middle of receiving packets
609 * will see the new offload handlers (until the next received packet).
611 void dev_add_offload(struct packet_offload *po)
613 struct packet_offload *elem;
615 spin_lock(&offload_lock);
616 list_for_each_entry(elem, &offload_base, list) {
617 if (po->priority < elem->priority)
620 list_add_rcu(&po->list, elem->list.prev);
621 spin_unlock(&offload_lock);
623 EXPORT_SYMBOL(dev_add_offload);
626 * __dev_remove_offload - remove offload handler
627 * @po: packet offload declaration
629 * Remove a protocol offload handler that was previously added to the
630 * kernel offload handlers by dev_add_offload(). The passed &offload_type
631 * is removed from the kernel lists and can be freed or reused once this
634 * The packet type might still be in use by receivers
635 * and must not be freed until after all the CPU's have gone
636 * through a quiescent state.
638 static void __dev_remove_offload(struct packet_offload *po)
640 struct list_head *head = &offload_base;
641 struct packet_offload *po1;
643 spin_lock(&offload_lock);
645 list_for_each_entry(po1, head, list) {
647 list_del_rcu(&po->list);
652 pr_warn("dev_remove_offload: %p not found\n", po);
654 spin_unlock(&offload_lock);
658 * dev_remove_offload - remove packet offload handler
659 * @po: packet offload declaration
661 * Remove a packet offload handler that was previously added to the kernel
662 * offload handlers by dev_add_offload(). The passed &offload_type is
663 * removed from the kernel lists and can be freed or reused once this
666 * This call sleeps to guarantee that no CPU is looking at the packet
669 void dev_remove_offload(struct packet_offload *po)
671 __dev_remove_offload(po);
675 EXPORT_SYMBOL(dev_remove_offload);
677 /******************************************************************************
679 * Device Boot-time Settings Routines
681 ******************************************************************************/
683 /* Boot time configuration table */
684 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
687 * netdev_boot_setup_add - add new setup entry
688 * @name: name of the device
689 * @map: configured settings for the device
691 * Adds new setup entry to the dev_boot_setup list. The function
692 * returns 0 on error and 1 on success. This is a generic routine to
695 static int netdev_boot_setup_add(char *name, struct ifmap *map)
697 struct netdev_boot_setup *s;
701 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
702 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
703 memset(s[i].name, 0, sizeof(s[i].name));
704 strlcpy(s[i].name, name, IFNAMSIZ);
705 memcpy(&s[i].map, map, sizeof(s[i].map));
710 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
714 * netdev_boot_setup_check - check boot time settings
715 * @dev: the netdevice
717 * Check boot time settings for the device.
718 * The found settings are set for the device to be used
719 * later in the device probing.
720 * Returns 0 if no settings found, 1 if they are.
722 int netdev_boot_setup_check(struct net_device *dev)
724 struct netdev_boot_setup *s = dev_boot_setup;
727 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
728 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
729 !strcmp(dev->name, s[i].name)) {
730 dev->irq = s[i].map.irq;
731 dev->base_addr = s[i].map.base_addr;
732 dev->mem_start = s[i].map.mem_start;
733 dev->mem_end = s[i].map.mem_end;
739 EXPORT_SYMBOL(netdev_boot_setup_check);
743 * netdev_boot_base - get address from boot time settings
744 * @prefix: prefix for network device
745 * @unit: id for network device
747 * Check boot time settings for the base address of device.
748 * The found settings are set for the device to be used
749 * later in the device probing.
750 * Returns 0 if no settings found.
752 unsigned long netdev_boot_base(const char *prefix, int unit)
754 const struct netdev_boot_setup *s = dev_boot_setup;
758 sprintf(name, "%s%d", prefix, unit);
761 * If device already registered then return base of 1
762 * to indicate not to probe for this interface
764 if (__dev_get_by_name(&init_net, name))
767 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
768 if (!strcmp(name, s[i].name))
769 return s[i].map.base_addr;
774 * Saves at boot time configured settings for any netdevice.
776 int __init netdev_boot_setup(char *str)
781 str = get_options(str, ARRAY_SIZE(ints), ints);
786 memset(&map, 0, sizeof(map));
790 map.base_addr = ints[2];
792 map.mem_start = ints[3];
794 map.mem_end = ints[4];
796 /* Add new entry to the list */
797 return netdev_boot_setup_add(str, &map);
800 __setup("netdev=", netdev_boot_setup);
802 /*******************************************************************************
804 * Device Interface Subroutines
806 *******************************************************************************/
809 * dev_get_iflink - get 'iflink' value of a interface
810 * @dev: targeted interface
812 * Indicates the ifindex the interface is linked to.
813 * Physical interfaces have the same 'ifindex' and 'iflink' values.
816 int dev_get_iflink(const struct net_device *dev)
818 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
819 return dev->netdev_ops->ndo_get_iflink(dev);
823 EXPORT_SYMBOL(dev_get_iflink);
826 * dev_fill_metadata_dst - Retrieve tunnel egress information.
827 * @dev: targeted interface
830 * For better visibility of tunnel traffic OVS needs to retrieve
831 * egress tunnel information for a packet. Following API allows
832 * user to get this info.
834 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
836 struct ip_tunnel_info *info;
838 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
841 info = skb_tunnel_info_unclone(skb);
844 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
847 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
849 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
851 static struct net_device_path *dev_fwd_path(struct net_device_path_stack *stack)
853 int k = stack->num_paths++;
855 if (WARN_ON_ONCE(k >= NET_DEVICE_PATH_STACK_MAX))
858 return &stack->path[k];
861 int dev_fill_forward_path(const struct net_device *dev, const u8 *daddr,
862 struct net_device_path_stack *stack)
864 const struct net_device *last_dev;
865 struct net_device_path_ctx ctx = {
869 struct net_device_path *path;
872 stack->num_paths = 0;
873 while (ctx.dev && ctx.dev->netdev_ops->ndo_fill_forward_path) {
875 path = dev_fwd_path(stack);
879 memset(path, 0, sizeof(struct net_device_path));
880 ret = ctx.dev->netdev_ops->ndo_fill_forward_path(&ctx, path);
884 if (WARN_ON_ONCE(last_dev == ctx.dev))
887 path = dev_fwd_path(stack);
890 path->type = DEV_PATH_ETHERNET;
895 EXPORT_SYMBOL_GPL(dev_fill_forward_path);
898 * __dev_get_by_name - find a device by its name
899 * @net: the applicable net namespace
900 * @name: name to find
902 * Find an interface by name. Must be called under RTNL semaphore
903 * or @dev_base_lock. If the name is found a pointer to the device
904 * is returned. If the name is not found then %NULL is returned. The
905 * reference counters are not incremented so the caller must be
906 * careful with locks.
909 struct net_device *__dev_get_by_name(struct net *net, const char *name)
911 struct netdev_name_node *node_name;
913 node_name = netdev_name_node_lookup(net, name);
914 return node_name ? node_name->dev : NULL;
916 EXPORT_SYMBOL(__dev_get_by_name);
919 * dev_get_by_name_rcu - find a device by its name
920 * @net: the applicable net namespace
921 * @name: name to find
923 * Find an interface by name.
924 * If the name is found a pointer to the device is returned.
925 * If the name is not found then %NULL is returned.
926 * The reference counters are not incremented so the caller must be
927 * careful with locks. The caller must hold RCU lock.
930 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
932 struct netdev_name_node *node_name;
934 node_name = netdev_name_node_lookup_rcu(net, name);
935 return node_name ? node_name->dev : NULL;
937 EXPORT_SYMBOL(dev_get_by_name_rcu);
940 * dev_get_by_name - find a device by its name
941 * @net: the applicable net namespace
942 * @name: name to find
944 * Find an interface by name. This can be called from any
945 * context and does its own locking. The returned handle has
946 * the usage count incremented and the caller must use dev_put() to
947 * release it when it is no longer needed. %NULL is returned if no
948 * matching device is found.
951 struct net_device *dev_get_by_name(struct net *net, const char *name)
953 struct net_device *dev;
956 dev = dev_get_by_name_rcu(net, name);
962 EXPORT_SYMBOL(dev_get_by_name);
965 * __dev_get_by_index - find a device by its ifindex
966 * @net: the applicable net namespace
967 * @ifindex: index of device
969 * Search for an interface by index. Returns %NULL if the device
970 * is not found or a pointer to the device. The device has not
971 * had its reference counter increased so the caller must be careful
972 * about locking. The caller must hold either the RTNL semaphore
976 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
978 struct net_device *dev;
979 struct hlist_head *head = dev_index_hash(net, ifindex);
981 hlist_for_each_entry(dev, head, index_hlist)
982 if (dev->ifindex == ifindex)
987 EXPORT_SYMBOL(__dev_get_by_index);
990 * dev_get_by_index_rcu - find a device by its ifindex
991 * @net: the applicable net namespace
992 * @ifindex: index of device
994 * Search for an interface by index. Returns %NULL if the device
995 * is not found or a pointer to the device. The device has not
996 * had its reference counter increased so the caller must be careful
997 * about locking. The caller must hold RCU lock.
1000 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
1002 struct net_device *dev;
1003 struct hlist_head *head = dev_index_hash(net, ifindex);
1005 hlist_for_each_entry_rcu(dev, head, index_hlist)
1006 if (dev->ifindex == ifindex)
1011 EXPORT_SYMBOL(dev_get_by_index_rcu);
1015 * dev_get_by_index - find a device by its ifindex
1016 * @net: the applicable net namespace
1017 * @ifindex: index of device
1019 * Search for an interface by index. Returns NULL if the device
1020 * is not found or a pointer to the device. The device returned has
1021 * had a reference added and the pointer is safe until the user calls
1022 * dev_put to indicate they have finished with it.
1025 struct net_device *dev_get_by_index(struct net *net, int ifindex)
1027 struct net_device *dev;
1030 dev = dev_get_by_index_rcu(net, ifindex);
1036 EXPORT_SYMBOL(dev_get_by_index);
1039 * dev_get_by_napi_id - find a device by napi_id
1040 * @napi_id: ID of the NAPI struct
1042 * Search for an interface by NAPI ID. Returns %NULL if the device
1043 * is not found or a pointer to the device. The device has not had
1044 * its reference counter increased so the caller must be careful
1045 * about locking. The caller must hold RCU lock.
1048 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
1050 struct napi_struct *napi;
1052 WARN_ON_ONCE(!rcu_read_lock_held());
1054 if (napi_id < MIN_NAPI_ID)
1057 napi = napi_by_id(napi_id);
1059 return napi ? napi->dev : NULL;
1061 EXPORT_SYMBOL(dev_get_by_napi_id);
1064 * netdev_get_name - get a netdevice name, knowing its ifindex.
1065 * @net: network namespace
1066 * @name: a pointer to the buffer where the name will be stored.
1067 * @ifindex: the ifindex of the interface to get the name from.
1069 int netdev_get_name(struct net *net, char *name, int ifindex)
1071 struct net_device *dev;
1074 down_read(&devnet_rename_sem);
1077 dev = dev_get_by_index_rcu(net, ifindex);
1083 strcpy(name, dev->name);
1088 up_read(&devnet_rename_sem);
1093 * dev_getbyhwaddr_rcu - find a device by its hardware address
1094 * @net: the applicable net namespace
1095 * @type: media type of device
1096 * @ha: hardware address
1098 * Search for an interface by MAC address. Returns NULL if the device
1099 * is not found or a pointer to the device.
1100 * The caller must hold RCU or RTNL.
1101 * The returned device has not had its ref count increased
1102 * and the caller must therefore be careful about locking
1106 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
1109 struct net_device *dev;
1111 for_each_netdev_rcu(net, dev)
1112 if (dev->type == type &&
1113 !memcmp(dev->dev_addr, ha, dev->addr_len))
1118 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1120 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1122 struct net_device *dev, *ret = NULL;
1125 for_each_netdev_rcu(net, dev)
1126 if (dev->type == type) {
1134 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1137 * __dev_get_by_flags - find any device with given flags
1138 * @net: the applicable net namespace
1139 * @if_flags: IFF_* values
1140 * @mask: bitmask of bits in if_flags to check
1142 * Search for any interface with the given flags. Returns NULL if a device
1143 * is not found or a pointer to the device. Must be called inside
1144 * rtnl_lock(), and result refcount is unchanged.
1147 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1148 unsigned short mask)
1150 struct net_device *dev, *ret;
1155 for_each_netdev(net, dev) {
1156 if (((dev->flags ^ if_flags) & mask) == 0) {
1163 EXPORT_SYMBOL(__dev_get_by_flags);
1166 * dev_valid_name - check if name is okay for network device
1167 * @name: name string
1169 * Network device names need to be valid file names to
1170 * allow sysfs to work. We also disallow any kind of
1173 bool dev_valid_name(const char *name)
1177 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1179 if (!strcmp(name, ".") || !strcmp(name, ".."))
1183 if (*name == '/' || *name == ':' || isspace(*name))
1189 EXPORT_SYMBOL(dev_valid_name);
1192 * __dev_alloc_name - allocate a name for a device
1193 * @net: network namespace to allocate the device name in
1194 * @name: name format string
1195 * @buf: scratch buffer and result name string
1197 * Passed a format string - eg "lt%d" it will try and find a suitable
1198 * id. It scans list of devices to build up a free map, then chooses
1199 * the first empty slot. The caller must hold the dev_base or rtnl lock
1200 * while allocating the name and adding the device in order to avoid
1202 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1203 * Returns the number of the unit assigned or a negative errno code.
1206 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1210 const int max_netdevices = 8*PAGE_SIZE;
1211 unsigned long *inuse;
1212 struct net_device *d;
1214 if (!dev_valid_name(name))
1217 p = strchr(name, '%');
1220 * Verify the string as this thing may have come from
1221 * the user. There must be either one "%d" and no other "%"
1224 if (p[1] != 'd' || strchr(p + 2, '%'))
1227 /* Use one page as a bit array of possible slots */
1228 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1232 for_each_netdev(net, d) {
1233 struct netdev_name_node *name_node;
1234 list_for_each_entry(name_node, &d->name_node->list, list) {
1235 if (!sscanf(name_node->name, name, &i))
1237 if (i < 0 || i >= max_netdevices)
1240 /* avoid cases where sscanf is not exact inverse of printf */
1241 snprintf(buf, IFNAMSIZ, name, i);
1242 if (!strncmp(buf, name_node->name, IFNAMSIZ))
1245 if (!sscanf(d->name, name, &i))
1247 if (i < 0 || i >= max_netdevices)
1250 /* avoid cases where sscanf is not exact inverse of printf */
1251 snprintf(buf, IFNAMSIZ, name, i);
1252 if (!strncmp(buf, d->name, IFNAMSIZ))
1256 i = find_first_zero_bit(inuse, max_netdevices);
1257 free_page((unsigned long) inuse);
1260 snprintf(buf, IFNAMSIZ, name, i);
1261 if (!__dev_get_by_name(net, buf))
1264 /* It is possible to run out of possible slots
1265 * when the name is long and there isn't enough space left
1266 * for the digits, or if all bits are used.
1271 static int dev_alloc_name_ns(struct net *net,
1272 struct net_device *dev,
1279 ret = __dev_alloc_name(net, name, buf);
1281 strlcpy(dev->name, buf, IFNAMSIZ);
1286 * dev_alloc_name - allocate a name for a device
1288 * @name: name format string
1290 * Passed a format string - eg "lt%d" it will try and find a suitable
1291 * id. It scans list of devices to build up a free map, then chooses
1292 * the first empty slot. The caller must hold the dev_base or rtnl lock
1293 * while allocating the name and adding the device in order to avoid
1295 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1296 * Returns the number of the unit assigned or a negative errno code.
1299 int dev_alloc_name(struct net_device *dev, const char *name)
1301 return dev_alloc_name_ns(dev_net(dev), dev, name);
1303 EXPORT_SYMBOL(dev_alloc_name);
1305 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1310 if (!dev_valid_name(name))
1313 if (strchr(name, '%'))
1314 return dev_alloc_name_ns(net, dev, name);
1315 else if (__dev_get_by_name(net, name))
1317 else if (dev->name != name)
1318 strlcpy(dev->name, name, IFNAMSIZ);
1324 * dev_change_name - change name of a device
1326 * @newname: name (or format string) must be at least IFNAMSIZ
1328 * Change name of a device, can pass format strings "eth%d".
1331 int dev_change_name(struct net_device *dev, const char *newname)
1333 unsigned char old_assign_type;
1334 char oldname[IFNAMSIZ];
1340 BUG_ON(!dev_net(dev));
1344 /* Some auto-enslaved devices e.g. failover slaves are
1345 * special, as userspace might rename the device after
1346 * the interface had been brought up and running since
1347 * the point kernel initiated auto-enslavement. Allow
1348 * live name change even when these slave devices are
1351 * Typically, users of these auto-enslaving devices
1352 * don't actually care about slave name change, as
1353 * they are supposed to operate on master interface
1356 if (dev->flags & IFF_UP &&
1357 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1360 down_write(&devnet_rename_sem);
1362 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1363 up_write(&devnet_rename_sem);
1367 memcpy(oldname, dev->name, IFNAMSIZ);
1369 err = dev_get_valid_name(net, dev, newname);
1371 up_write(&devnet_rename_sem);
1375 if (oldname[0] && !strchr(oldname, '%'))
1376 netdev_info(dev, "renamed from %s\n", oldname);
1378 old_assign_type = dev->name_assign_type;
1379 dev->name_assign_type = NET_NAME_RENAMED;
1382 ret = device_rename(&dev->dev, dev->name);
1384 memcpy(dev->name, oldname, IFNAMSIZ);
1385 dev->name_assign_type = old_assign_type;
1386 up_write(&devnet_rename_sem);
1390 up_write(&devnet_rename_sem);
1392 netdev_adjacent_rename_links(dev, oldname);
1394 write_lock_bh(&dev_base_lock);
1395 netdev_name_node_del(dev->name_node);
1396 write_unlock_bh(&dev_base_lock);
1400 write_lock_bh(&dev_base_lock);
1401 netdev_name_node_add(net, dev->name_node);
1402 write_unlock_bh(&dev_base_lock);
1404 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1405 ret = notifier_to_errno(ret);
1408 /* err >= 0 after dev_alloc_name() or stores the first errno */
1411 down_write(&devnet_rename_sem);
1412 memcpy(dev->name, oldname, IFNAMSIZ);
1413 memcpy(oldname, newname, IFNAMSIZ);
1414 dev->name_assign_type = old_assign_type;
1415 old_assign_type = NET_NAME_RENAMED;
1418 pr_err("%s: name change rollback failed: %d\n",
1427 * dev_set_alias - change ifalias of a device
1429 * @alias: name up to IFALIASZ
1430 * @len: limit of bytes to copy from info
1432 * Set ifalias for a device,
1434 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1436 struct dev_ifalias *new_alias = NULL;
1438 if (len >= IFALIASZ)
1442 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1446 memcpy(new_alias->ifalias, alias, len);
1447 new_alias->ifalias[len] = 0;
1450 mutex_lock(&ifalias_mutex);
1451 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1452 mutex_is_locked(&ifalias_mutex));
1453 mutex_unlock(&ifalias_mutex);
1456 kfree_rcu(new_alias, rcuhead);
1460 EXPORT_SYMBOL(dev_set_alias);
1463 * dev_get_alias - get ifalias of a device
1465 * @name: buffer to store name of ifalias
1466 * @len: size of buffer
1468 * get ifalias for a device. Caller must make sure dev cannot go
1469 * away, e.g. rcu read lock or own a reference count to device.
1471 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1473 const struct dev_ifalias *alias;
1477 alias = rcu_dereference(dev->ifalias);
1479 ret = snprintf(name, len, "%s", alias->ifalias);
1486 * netdev_features_change - device changes features
1487 * @dev: device to cause notification
1489 * Called to indicate a device has changed features.
1491 void netdev_features_change(struct net_device *dev)
1493 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1495 EXPORT_SYMBOL(netdev_features_change);
1498 * netdev_state_change - device changes state
1499 * @dev: device to cause notification
1501 * Called to indicate a device has changed state. This function calls
1502 * the notifier chains for netdev_chain and sends a NEWLINK message
1503 * to the routing socket.
1505 void netdev_state_change(struct net_device *dev)
1507 if (dev->flags & IFF_UP) {
1508 struct netdev_notifier_change_info change_info = {
1512 call_netdevice_notifiers_info(NETDEV_CHANGE,
1514 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1517 EXPORT_SYMBOL(netdev_state_change);
1520 * __netdev_notify_peers - notify network peers about existence of @dev,
1521 * to be called when rtnl lock is already held.
1522 * @dev: network device
1524 * Generate traffic such that interested network peers are aware of
1525 * @dev, such as by generating a gratuitous ARP. This may be used when
1526 * a device wants to inform the rest of the network about some sort of
1527 * reconfiguration such as a failover event or virtual machine
1530 void __netdev_notify_peers(struct net_device *dev)
1533 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1534 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1536 EXPORT_SYMBOL(__netdev_notify_peers);
1539 * netdev_notify_peers - notify network peers about existence of @dev
1540 * @dev: network device
1542 * Generate traffic such that interested network peers are aware of
1543 * @dev, such as by generating a gratuitous ARP. This may be used when
1544 * a device wants to inform the rest of the network about some sort of
1545 * reconfiguration such as a failover event or virtual machine
1548 void netdev_notify_peers(struct net_device *dev)
1551 __netdev_notify_peers(dev);
1554 EXPORT_SYMBOL(netdev_notify_peers);
1556 static int napi_threaded_poll(void *data);
1558 static int napi_kthread_create(struct napi_struct *n)
1562 /* Create and wake up the kthread once to put it in
1563 * TASK_INTERRUPTIBLE mode to avoid the blocked task
1564 * warning and work with loadavg.
1566 n->thread = kthread_run(napi_threaded_poll, n, "napi/%s-%d",
1567 n->dev->name, n->napi_id);
1568 if (IS_ERR(n->thread)) {
1569 err = PTR_ERR(n->thread);
1570 pr_err("kthread_run failed with err %d\n", err);
1577 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1579 const struct net_device_ops *ops = dev->netdev_ops;
1584 if (!netif_device_present(dev)) {
1585 /* may be detached because parent is runtime-suspended */
1586 if (dev->dev.parent)
1587 pm_runtime_resume(dev->dev.parent);
1588 if (!netif_device_present(dev))
1592 /* Block netpoll from trying to do any rx path servicing.
1593 * If we don't do this there is a chance ndo_poll_controller
1594 * or ndo_poll may be running while we open the device
1596 netpoll_poll_disable(dev);
1598 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1599 ret = notifier_to_errno(ret);
1603 set_bit(__LINK_STATE_START, &dev->state);
1605 if (ops->ndo_validate_addr)
1606 ret = ops->ndo_validate_addr(dev);
1608 if (!ret && ops->ndo_open)
1609 ret = ops->ndo_open(dev);
1611 netpoll_poll_enable(dev);
1614 clear_bit(__LINK_STATE_START, &dev->state);
1616 dev->flags |= IFF_UP;
1617 dev_set_rx_mode(dev);
1619 add_device_randomness(dev->dev_addr, dev->addr_len);
1626 * dev_open - prepare an interface for use.
1627 * @dev: device to open
1628 * @extack: netlink extended ack
1630 * Takes a device from down to up state. The device's private open
1631 * function is invoked and then the multicast lists are loaded. Finally
1632 * the device is moved into the up state and a %NETDEV_UP message is
1633 * sent to the netdev notifier chain.
1635 * Calling this function on an active interface is a nop. On a failure
1636 * a negative errno code is returned.
1638 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1642 if (dev->flags & IFF_UP)
1645 ret = __dev_open(dev, extack);
1649 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1650 call_netdevice_notifiers(NETDEV_UP, dev);
1654 EXPORT_SYMBOL(dev_open);
1656 static void __dev_close_many(struct list_head *head)
1658 struct net_device *dev;
1663 list_for_each_entry(dev, head, close_list) {
1664 /* Temporarily disable netpoll until the interface is down */
1665 netpoll_poll_disable(dev);
1667 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1669 clear_bit(__LINK_STATE_START, &dev->state);
1671 /* Synchronize to scheduled poll. We cannot touch poll list, it
1672 * can be even on different cpu. So just clear netif_running().
1674 * dev->stop() will invoke napi_disable() on all of it's
1675 * napi_struct instances on this device.
1677 smp_mb__after_atomic(); /* Commit netif_running(). */
1680 dev_deactivate_many(head);
1682 list_for_each_entry(dev, head, close_list) {
1683 const struct net_device_ops *ops = dev->netdev_ops;
1686 * Call the device specific close. This cannot fail.
1687 * Only if device is UP
1689 * We allow it to be called even after a DETACH hot-plug
1695 dev->flags &= ~IFF_UP;
1696 netpoll_poll_enable(dev);
1700 static void __dev_close(struct net_device *dev)
1704 list_add(&dev->close_list, &single);
1705 __dev_close_many(&single);
1709 void dev_close_many(struct list_head *head, bool unlink)
1711 struct net_device *dev, *tmp;
1713 /* Remove the devices that don't need to be closed */
1714 list_for_each_entry_safe(dev, tmp, head, close_list)
1715 if (!(dev->flags & IFF_UP))
1716 list_del_init(&dev->close_list);
1718 __dev_close_many(head);
1720 list_for_each_entry_safe(dev, tmp, head, close_list) {
1721 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1722 call_netdevice_notifiers(NETDEV_DOWN, dev);
1724 list_del_init(&dev->close_list);
1727 EXPORT_SYMBOL(dev_close_many);
1730 * dev_close - shutdown an interface.
1731 * @dev: device to shutdown
1733 * This function moves an active device into down state. A
1734 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1735 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1738 void dev_close(struct net_device *dev)
1740 if (dev->flags & IFF_UP) {
1743 list_add(&dev->close_list, &single);
1744 dev_close_many(&single, true);
1748 EXPORT_SYMBOL(dev_close);
1752 * dev_disable_lro - disable Large Receive Offload on a device
1755 * Disable Large Receive Offload (LRO) on a net device. Must be
1756 * called under RTNL. This is needed if received packets may be
1757 * forwarded to another interface.
1759 void dev_disable_lro(struct net_device *dev)
1761 struct net_device *lower_dev;
1762 struct list_head *iter;
1764 dev->wanted_features &= ~NETIF_F_LRO;
1765 netdev_update_features(dev);
1767 if (unlikely(dev->features & NETIF_F_LRO))
1768 netdev_WARN(dev, "failed to disable LRO!\n");
1770 netdev_for_each_lower_dev(dev, lower_dev, iter)
1771 dev_disable_lro(lower_dev);
1773 EXPORT_SYMBOL(dev_disable_lro);
1776 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1779 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1780 * called under RTNL. This is needed if Generic XDP is installed on
1783 static void dev_disable_gro_hw(struct net_device *dev)
1785 dev->wanted_features &= ~NETIF_F_GRO_HW;
1786 netdev_update_features(dev);
1788 if (unlikely(dev->features & NETIF_F_GRO_HW))
1789 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1792 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1795 case NETDEV_##val: \
1796 return "NETDEV_" __stringify(val);
1798 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1799 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1800 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1801 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1802 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1803 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1804 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1805 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1806 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1810 return "UNKNOWN_NETDEV_EVENT";
1812 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1814 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1815 struct net_device *dev)
1817 struct netdev_notifier_info info = {
1821 return nb->notifier_call(nb, val, &info);
1824 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1825 struct net_device *dev)
1829 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1830 err = notifier_to_errno(err);
1834 if (!(dev->flags & IFF_UP))
1837 call_netdevice_notifier(nb, NETDEV_UP, dev);
1841 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1842 struct net_device *dev)
1844 if (dev->flags & IFF_UP) {
1845 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1847 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1849 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1852 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1855 struct net_device *dev;
1858 for_each_netdev(net, dev) {
1859 err = call_netdevice_register_notifiers(nb, dev);
1866 for_each_netdev_continue_reverse(net, dev)
1867 call_netdevice_unregister_notifiers(nb, dev);
1871 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1874 struct net_device *dev;
1876 for_each_netdev(net, dev)
1877 call_netdevice_unregister_notifiers(nb, dev);
1880 static int dev_boot_phase = 1;
1883 * register_netdevice_notifier - register a network notifier block
1886 * Register a notifier to be called when network device events occur.
1887 * The notifier passed is linked into the kernel structures and must
1888 * not be reused until it has been unregistered. A negative errno code
1889 * is returned on a failure.
1891 * When registered all registration and up events are replayed
1892 * to the new notifier to allow device to have a race free
1893 * view of the network device list.
1896 int register_netdevice_notifier(struct notifier_block *nb)
1901 /* Close race with setup_net() and cleanup_net() */
1902 down_write(&pernet_ops_rwsem);
1904 err = raw_notifier_chain_register(&netdev_chain, nb);
1910 err = call_netdevice_register_net_notifiers(nb, net);
1917 up_write(&pernet_ops_rwsem);
1921 for_each_net_continue_reverse(net)
1922 call_netdevice_unregister_net_notifiers(nb, net);
1924 raw_notifier_chain_unregister(&netdev_chain, nb);
1927 EXPORT_SYMBOL(register_netdevice_notifier);
1930 * unregister_netdevice_notifier - unregister a network notifier block
1933 * Unregister a notifier previously registered by
1934 * register_netdevice_notifier(). The notifier is unlinked into the
1935 * kernel structures and may then be reused. A negative errno code
1936 * is returned on a failure.
1938 * After unregistering unregister and down device events are synthesized
1939 * for all devices on the device list to the removed notifier to remove
1940 * the need for special case cleanup code.
1943 int unregister_netdevice_notifier(struct notifier_block *nb)
1948 /* Close race with setup_net() and cleanup_net() */
1949 down_write(&pernet_ops_rwsem);
1951 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1956 call_netdevice_unregister_net_notifiers(nb, net);
1960 up_write(&pernet_ops_rwsem);
1963 EXPORT_SYMBOL(unregister_netdevice_notifier);
1965 static int __register_netdevice_notifier_net(struct net *net,
1966 struct notifier_block *nb,
1967 bool ignore_call_fail)
1971 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1977 err = call_netdevice_register_net_notifiers(nb, net);
1978 if (err && !ignore_call_fail)
1979 goto chain_unregister;
1984 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1988 static int __unregister_netdevice_notifier_net(struct net *net,
1989 struct notifier_block *nb)
1993 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1997 call_netdevice_unregister_net_notifiers(nb, net);
2002 * register_netdevice_notifier_net - register a per-netns network notifier block
2003 * @net: network namespace
2006 * Register a notifier to be called when network device events occur.
2007 * The notifier passed is linked into the kernel structures and must
2008 * not be reused until it has been unregistered. A negative errno code
2009 * is returned on a failure.
2011 * When registered all registration and up events are replayed
2012 * to the new notifier to allow device to have a race free
2013 * view of the network device list.
2016 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
2021 err = __register_netdevice_notifier_net(net, nb, false);
2025 EXPORT_SYMBOL(register_netdevice_notifier_net);
2028 * unregister_netdevice_notifier_net - unregister a per-netns
2029 * network notifier block
2030 * @net: network namespace
2033 * Unregister a notifier previously registered by
2034 * register_netdevice_notifier(). The notifier is unlinked into the
2035 * kernel structures and may then be reused. A negative errno code
2036 * is returned on a failure.
2038 * After unregistering unregister and down device events are synthesized
2039 * for all devices on the device list to the removed notifier to remove
2040 * the need for special case cleanup code.
2043 int unregister_netdevice_notifier_net(struct net *net,
2044 struct notifier_block *nb)
2049 err = __unregister_netdevice_notifier_net(net, nb);
2053 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
2055 int register_netdevice_notifier_dev_net(struct net_device *dev,
2056 struct notifier_block *nb,
2057 struct netdev_net_notifier *nn)
2062 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
2065 list_add(&nn->list, &dev->net_notifier_list);
2070 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
2072 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
2073 struct notifier_block *nb,
2074 struct netdev_net_notifier *nn)
2079 list_del(&nn->list);
2080 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
2084 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
2086 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
2089 struct netdev_net_notifier *nn;
2091 list_for_each_entry(nn, &dev->net_notifier_list, list) {
2092 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
2093 __register_netdevice_notifier_net(net, nn->nb, true);
2098 * call_netdevice_notifiers_info - call all network notifier blocks
2099 * @val: value passed unmodified to notifier function
2100 * @info: notifier information data
2102 * Call all network notifier blocks. Parameters and return value
2103 * are as for raw_notifier_call_chain().
2106 static int call_netdevice_notifiers_info(unsigned long val,
2107 struct netdev_notifier_info *info)
2109 struct net *net = dev_net(info->dev);
2114 /* Run per-netns notifier block chain first, then run the global one.
2115 * Hopefully, one day, the global one is going to be removed after
2116 * all notifier block registrators get converted to be per-netns.
2118 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2119 if (ret & NOTIFY_STOP_MASK)
2121 return raw_notifier_call_chain(&netdev_chain, val, info);
2124 static int call_netdevice_notifiers_extack(unsigned long val,
2125 struct net_device *dev,
2126 struct netlink_ext_ack *extack)
2128 struct netdev_notifier_info info = {
2133 return call_netdevice_notifiers_info(val, &info);
2137 * call_netdevice_notifiers - call all network notifier blocks
2138 * @val: value passed unmodified to notifier function
2139 * @dev: net_device pointer passed unmodified to notifier function
2141 * Call all network notifier blocks. Parameters and return value
2142 * are as for raw_notifier_call_chain().
2145 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2147 return call_netdevice_notifiers_extack(val, dev, NULL);
2149 EXPORT_SYMBOL(call_netdevice_notifiers);
2152 * call_netdevice_notifiers_mtu - call all network notifier blocks
2153 * @val: value passed unmodified to notifier function
2154 * @dev: net_device pointer passed unmodified to notifier function
2155 * @arg: additional u32 argument passed to the notifier function
2157 * Call all network notifier blocks. Parameters and return value
2158 * are as for raw_notifier_call_chain().
2160 static int call_netdevice_notifiers_mtu(unsigned long val,
2161 struct net_device *dev, u32 arg)
2163 struct netdev_notifier_info_ext info = {
2168 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2170 return call_netdevice_notifiers_info(val, &info.info);
2173 #ifdef CONFIG_NET_INGRESS
2174 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2176 void net_inc_ingress_queue(void)
2178 static_branch_inc(&ingress_needed_key);
2180 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2182 void net_dec_ingress_queue(void)
2184 static_branch_dec(&ingress_needed_key);
2186 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2189 #ifdef CONFIG_NET_EGRESS
2190 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2192 void net_inc_egress_queue(void)
2194 static_branch_inc(&egress_needed_key);
2196 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2198 void net_dec_egress_queue(void)
2200 static_branch_dec(&egress_needed_key);
2202 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2205 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2206 #ifdef CONFIG_JUMP_LABEL
2207 static atomic_t netstamp_needed_deferred;
2208 static atomic_t netstamp_wanted;
2209 static void netstamp_clear(struct work_struct *work)
2211 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2214 wanted = atomic_add_return(deferred, &netstamp_wanted);
2216 static_branch_enable(&netstamp_needed_key);
2218 static_branch_disable(&netstamp_needed_key);
2220 static DECLARE_WORK(netstamp_work, netstamp_clear);
2223 void net_enable_timestamp(void)
2225 #ifdef CONFIG_JUMP_LABEL
2229 wanted = atomic_read(&netstamp_wanted);
2232 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2235 atomic_inc(&netstamp_needed_deferred);
2236 schedule_work(&netstamp_work);
2238 static_branch_inc(&netstamp_needed_key);
2241 EXPORT_SYMBOL(net_enable_timestamp);
2243 void net_disable_timestamp(void)
2245 #ifdef CONFIG_JUMP_LABEL
2249 wanted = atomic_read(&netstamp_wanted);
2252 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2255 atomic_dec(&netstamp_needed_deferred);
2256 schedule_work(&netstamp_work);
2258 static_branch_dec(&netstamp_needed_key);
2261 EXPORT_SYMBOL(net_disable_timestamp);
2263 static inline void net_timestamp_set(struct sk_buff *skb)
2266 if (static_branch_unlikely(&netstamp_needed_key))
2267 __net_timestamp(skb);
2270 #define net_timestamp_check(COND, SKB) \
2271 if (static_branch_unlikely(&netstamp_needed_key)) { \
2272 if ((COND) && !(SKB)->tstamp) \
2273 __net_timestamp(SKB); \
2276 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2278 return __is_skb_forwardable(dev, skb, true);
2280 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2282 static int __dev_forward_skb2(struct net_device *dev, struct sk_buff *skb,
2285 int ret = ____dev_forward_skb(dev, skb, check_mtu);
2288 skb->protocol = eth_type_trans(skb, dev);
2289 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2295 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2297 return __dev_forward_skb2(dev, skb, true);
2299 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2302 * dev_forward_skb - loopback an skb to another netif
2304 * @dev: destination network device
2305 * @skb: buffer to forward
2308 * NET_RX_SUCCESS (no congestion)
2309 * NET_RX_DROP (packet was dropped, but freed)
2311 * dev_forward_skb can be used for injecting an skb from the
2312 * start_xmit function of one device into the receive queue
2313 * of another device.
2315 * The receiving device may be in another namespace, so
2316 * we have to clear all information in the skb that could
2317 * impact namespace isolation.
2319 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2321 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2323 EXPORT_SYMBOL_GPL(dev_forward_skb);
2325 int dev_forward_skb_nomtu(struct net_device *dev, struct sk_buff *skb)
2327 return __dev_forward_skb2(dev, skb, false) ?: netif_rx_internal(skb);
2330 static inline int deliver_skb(struct sk_buff *skb,
2331 struct packet_type *pt_prev,
2332 struct net_device *orig_dev)
2334 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2336 refcount_inc(&skb->users);
2337 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2340 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2341 struct packet_type **pt,
2342 struct net_device *orig_dev,
2344 struct list_head *ptype_list)
2346 struct packet_type *ptype, *pt_prev = *pt;
2348 list_for_each_entry_rcu(ptype, ptype_list, list) {
2349 if (ptype->type != type)
2352 deliver_skb(skb, pt_prev, orig_dev);
2358 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2360 if (!ptype->af_packet_priv || !skb->sk)
2363 if (ptype->id_match)
2364 return ptype->id_match(ptype, skb->sk);
2365 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2372 * dev_nit_active - return true if any network interface taps are in use
2374 * @dev: network device to check for the presence of taps
2376 bool dev_nit_active(struct net_device *dev)
2378 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2380 EXPORT_SYMBOL_GPL(dev_nit_active);
2383 * Support routine. Sends outgoing frames to any network
2384 * taps currently in use.
2387 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2389 struct packet_type *ptype;
2390 struct sk_buff *skb2 = NULL;
2391 struct packet_type *pt_prev = NULL;
2392 struct list_head *ptype_list = &ptype_all;
2396 list_for_each_entry_rcu(ptype, ptype_list, list) {
2397 if (ptype->ignore_outgoing)
2400 /* Never send packets back to the socket
2401 * they originated from - MvS (miquels@drinkel.ow.org)
2403 if (skb_loop_sk(ptype, skb))
2407 deliver_skb(skb2, pt_prev, skb->dev);
2412 /* need to clone skb, done only once */
2413 skb2 = skb_clone(skb, GFP_ATOMIC);
2417 net_timestamp_set(skb2);
2419 /* skb->nh should be correctly
2420 * set by sender, so that the second statement is
2421 * just protection against buggy protocols.
2423 skb_reset_mac_header(skb2);
2425 if (skb_network_header(skb2) < skb2->data ||
2426 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2427 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2428 ntohs(skb2->protocol),
2430 skb_reset_network_header(skb2);
2433 skb2->transport_header = skb2->network_header;
2434 skb2->pkt_type = PACKET_OUTGOING;
2438 if (ptype_list == &ptype_all) {
2439 ptype_list = &dev->ptype_all;
2444 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2445 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2451 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2454 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2455 * @dev: Network device
2456 * @txq: number of queues available
2458 * If real_num_tx_queues is changed the tc mappings may no longer be
2459 * valid. To resolve this verify the tc mapping remains valid and if
2460 * not NULL the mapping. With no priorities mapping to this
2461 * offset/count pair it will no longer be used. In the worst case TC0
2462 * is invalid nothing can be done so disable priority mappings. If is
2463 * expected that drivers will fix this mapping if they can before
2464 * calling netif_set_real_num_tx_queues.
2466 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2469 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2471 /* If TC0 is invalidated disable TC mapping */
2472 if (tc->offset + tc->count > txq) {
2473 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2478 /* Invalidated prio to tc mappings set to TC0 */
2479 for (i = 1; i < TC_BITMASK + 1; i++) {
2480 int q = netdev_get_prio_tc_map(dev, i);
2482 tc = &dev->tc_to_txq[q];
2483 if (tc->offset + tc->count > txq) {
2484 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2486 netdev_set_prio_tc_map(dev, i, 0);
2491 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2494 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2497 /* walk through the TCs and see if it falls into any of them */
2498 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2499 if ((txq - tc->offset) < tc->count)
2503 /* didn't find it, just return -1 to indicate no match */
2509 EXPORT_SYMBOL(netdev_txq_to_tc);
2512 static struct static_key xps_needed __read_mostly;
2513 static struct static_key xps_rxqs_needed __read_mostly;
2514 static DEFINE_MUTEX(xps_map_mutex);
2515 #define xmap_dereference(P) \
2516 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2518 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2519 struct xps_dev_maps *old_maps, int tci, u16 index)
2521 struct xps_map *map = NULL;
2525 map = xmap_dereference(dev_maps->attr_map[tci]);
2529 for (pos = map->len; pos--;) {
2530 if (map->queues[pos] != index)
2534 map->queues[pos] = map->queues[--map->len];
2539 RCU_INIT_POINTER(old_maps->attr_map[tci], NULL);
2540 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2541 kfree_rcu(map, rcu);
2548 static bool remove_xps_queue_cpu(struct net_device *dev,
2549 struct xps_dev_maps *dev_maps,
2550 int cpu, u16 offset, u16 count)
2552 int num_tc = dev_maps->num_tc;
2553 bool active = false;
2556 for (tci = cpu * num_tc; num_tc--; tci++) {
2559 for (i = count, j = offset; i--; j++) {
2560 if (!remove_xps_queue(dev_maps, NULL, tci, j))
2570 static void reset_xps_maps(struct net_device *dev,
2571 struct xps_dev_maps *dev_maps,
2572 enum xps_map_type type)
2574 static_key_slow_dec_cpuslocked(&xps_needed);
2575 if (type == XPS_RXQS)
2576 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2578 RCU_INIT_POINTER(dev->xps_maps[type], NULL);
2580 kfree_rcu(dev_maps, rcu);
2583 static void clean_xps_maps(struct net_device *dev, enum xps_map_type type,
2584 u16 offset, u16 count)
2586 struct xps_dev_maps *dev_maps;
2587 bool active = false;
2590 dev_maps = xmap_dereference(dev->xps_maps[type]);
2594 for (j = 0; j < dev_maps->nr_ids; j++)
2595 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, count);
2597 reset_xps_maps(dev, dev_maps, type);
2599 if (type == XPS_CPUS) {
2600 for (i = offset + (count - 1); count--; i--)
2601 netdev_queue_numa_node_write(
2602 netdev_get_tx_queue(dev, i), NUMA_NO_NODE);
2606 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2609 if (!static_key_false(&xps_needed))
2613 mutex_lock(&xps_map_mutex);
2615 if (static_key_false(&xps_rxqs_needed))
2616 clean_xps_maps(dev, XPS_RXQS, offset, count);
2618 clean_xps_maps(dev, XPS_CPUS, offset, count);
2620 mutex_unlock(&xps_map_mutex);
2624 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2626 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2629 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2630 u16 index, bool is_rxqs_map)
2632 struct xps_map *new_map;
2633 int alloc_len = XPS_MIN_MAP_ALLOC;
2636 for (pos = 0; map && pos < map->len; pos++) {
2637 if (map->queues[pos] != index)
2642 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2644 if (pos < map->alloc_len)
2647 alloc_len = map->alloc_len * 2;
2650 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2654 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2656 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2657 cpu_to_node(attr_index));
2661 for (i = 0; i < pos; i++)
2662 new_map->queues[i] = map->queues[i];
2663 new_map->alloc_len = alloc_len;
2669 /* Copy xps maps at a given index */
2670 static void xps_copy_dev_maps(struct xps_dev_maps *dev_maps,
2671 struct xps_dev_maps *new_dev_maps, int index,
2672 int tc, bool skip_tc)
2674 int i, tci = index * dev_maps->num_tc;
2675 struct xps_map *map;
2677 /* copy maps belonging to foreign traffic classes */
2678 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2679 if (i == tc && skip_tc)
2682 /* fill in the new device map from the old device map */
2683 map = xmap_dereference(dev_maps->attr_map[tci]);
2684 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2688 /* Must be called under cpus_read_lock */
2689 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2690 u16 index, enum xps_map_type type)
2692 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL, *old_dev_maps = NULL;
2693 const unsigned long *online_mask = NULL;
2694 bool active = false, copy = false;
2695 int i, j, tci, numa_node_id = -2;
2696 int maps_sz, num_tc = 1, tc = 0;
2697 struct xps_map *map, *new_map;
2698 unsigned int nr_ids;
2701 /* Do not allow XPS on subordinate device directly */
2702 num_tc = dev->num_tc;
2706 /* If queue belongs to subordinate dev use its map */
2707 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2709 tc = netdev_txq_to_tc(dev, index);
2714 mutex_lock(&xps_map_mutex);
2716 dev_maps = xmap_dereference(dev->xps_maps[type]);
2717 if (type == XPS_RXQS) {
2718 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2719 nr_ids = dev->num_rx_queues;
2721 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2722 if (num_possible_cpus() > 1)
2723 online_mask = cpumask_bits(cpu_online_mask);
2724 nr_ids = nr_cpu_ids;
2727 if (maps_sz < L1_CACHE_BYTES)
2728 maps_sz = L1_CACHE_BYTES;
2730 /* The old dev_maps could be larger or smaller than the one we're
2731 * setting up now, as dev->num_tc or nr_ids could have been updated in
2732 * between. We could try to be smart, but let's be safe instead and only
2733 * copy foreign traffic classes if the two map sizes match.
2736 dev_maps->num_tc == num_tc && dev_maps->nr_ids == nr_ids)
2739 /* allocate memory for queue storage */
2740 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2742 if (!new_dev_maps) {
2743 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2744 if (!new_dev_maps) {
2745 mutex_unlock(&xps_map_mutex);
2749 new_dev_maps->nr_ids = nr_ids;
2750 new_dev_maps->num_tc = num_tc;
2753 tci = j * num_tc + tc;
2754 map = copy ? xmap_dereference(dev_maps->attr_map[tci]) : NULL;
2756 map = expand_xps_map(map, j, index, type == XPS_RXQS);
2760 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2764 goto out_no_new_maps;
2767 /* Increment static keys at most once per type */
2768 static_key_slow_inc_cpuslocked(&xps_needed);
2769 if (type == XPS_RXQS)
2770 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2773 for (j = 0; j < nr_ids; j++) {
2774 bool skip_tc = false;
2776 tci = j * num_tc + tc;
2777 if (netif_attr_test_mask(j, mask, nr_ids) &&
2778 netif_attr_test_online(j, online_mask, nr_ids)) {
2779 /* add tx-queue to CPU/rx-queue maps */
2784 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2785 while ((pos < map->len) && (map->queues[pos] != index))
2788 if (pos == map->len)
2789 map->queues[map->len++] = index;
2791 if (type == XPS_CPUS) {
2792 if (numa_node_id == -2)
2793 numa_node_id = cpu_to_node(j);
2794 else if (numa_node_id != cpu_to_node(j))
2801 xps_copy_dev_maps(dev_maps, new_dev_maps, j, tc,
2805 rcu_assign_pointer(dev->xps_maps[type], new_dev_maps);
2807 /* Cleanup old maps */
2809 goto out_no_old_maps;
2811 for (j = 0; j < dev_maps->nr_ids; j++) {
2812 for (i = num_tc, tci = j * dev_maps->num_tc; i--; tci++) {
2813 map = xmap_dereference(dev_maps->attr_map[tci]);
2818 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2823 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2824 kfree_rcu(map, rcu);
2828 old_dev_maps = dev_maps;
2831 dev_maps = new_dev_maps;
2835 if (type == XPS_CPUS)
2836 /* update Tx queue numa node */
2837 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2838 (numa_node_id >= 0) ?
2839 numa_node_id : NUMA_NO_NODE);
2844 /* removes tx-queue from unused CPUs/rx-queues */
2845 for (j = 0; j < dev_maps->nr_ids; j++) {
2846 tci = j * dev_maps->num_tc;
2848 for (i = 0; i < dev_maps->num_tc; i++, tci++) {
2850 netif_attr_test_mask(j, mask, dev_maps->nr_ids) &&
2851 netif_attr_test_online(j, online_mask, dev_maps->nr_ids))
2854 active |= remove_xps_queue(dev_maps,
2855 copy ? old_dev_maps : NULL,
2861 kfree_rcu(old_dev_maps, rcu);
2863 /* free map if not active */
2865 reset_xps_maps(dev, dev_maps, type);
2868 mutex_unlock(&xps_map_mutex);
2872 /* remove any maps that we added */
2873 for (j = 0; j < nr_ids; j++) {
2874 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2875 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2877 xmap_dereference(dev_maps->attr_map[tci]) :
2879 if (new_map && new_map != map)
2884 mutex_unlock(&xps_map_mutex);
2886 kfree(new_dev_maps);
2889 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2891 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2897 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, XPS_CPUS);
2902 EXPORT_SYMBOL(netif_set_xps_queue);
2905 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2907 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2909 /* Unbind any subordinate channels */
2910 while (txq-- != &dev->_tx[0]) {
2912 netdev_unbind_sb_channel(dev, txq->sb_dev);
2916 void netdev_reset_tc(struct net_device *dev)
2919 netif_reset_xps_queues_gt(dev, 0);
2921 netdev_unbind_all_sb_channels(dev);
2923 /* Reset TC configuration of device */
2925 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2926 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2928 EXPORT_SYMBOL(netdev_reset_tc);
2930 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2932 if (tc >= dev->num_tc)
2936 netif_reset_xps_queues(dev, offset, count);
2938 dev->tc_to_txq[tc].count = count;
2939 dev->tc_to_txq[tc].offset = offset;
2942 EXPORT_SYMBOL(netdev_set_tc_queue);
2944 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2946 if (num_tc > TC_MAX_QUEUE)
2950 netif_reset_xps_queues_gt(dev, 0);
2952 netdev_unbind_all_sb_channels(dev);
2954 dev->num_tc = num_tc;
2957 EXPORT_SYMBOL(netdev_set_num_tc);
2959 void netdev_unbind_sb_channel(struct net_device *dev,
2960 struct net_device *sb_dev)
2962 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2965 netif_reset_xps_queues_gt(sb_dev, 0);
2967 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2968 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2970 while (txq-- != &dev->_tx[0]) {
2971 if (txq->sb_dev == sb_dev)
2975 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2977 int netdev_bind_sb_channel_queue(struct net_device *dev,
2978 struct net_device *sb_dev,
2979 u8 tc, u16 count, u16 offset)
2981 /* Make certain the sb_dev and dev are already configured */
2982 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2985 /* We cannot hand out queues we don't have */
2986 if ((offset + count) > dev->real_num_tx_queues)
2989 /* Record the mapping */
2990 sb_dev->tc_to_txq[tc].count = count;
2991 sb_dev->tc_to_txq[tc].offset = offset;
2993 /* Provide a way for Tx queue to find the tc_to_txq map or
2994 * XPS map for itself.
2997 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
3001 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
3003 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
3005 /* Do not use a multiqueue device to represent a subordinate channel */
3006 if (netif_is_multiqueue(dev))
3009 /* We allow channels 1 - 32767 to be used for subordinate channels.
3010 * Channel 0 is meant to be "native" mode and used only to represent
3011 * the main root device. We allow writing 0 to reset the device back
3012 * to normal mode after being used as a subordinate channel.
3014 if (channel > S16_MAX)
3017 dev->num_tc = -channel;
3021 EXPORT_SYMBOL(netdev_set_sb_channel);
3024 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
3025 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
3027 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
3032 disabling = txq < dev->real_num_tx_queues;
3034 if (txq < 1 || txq > dev->num_tx_queues)
3037 if (dev->reg_state == NETREG_REGISTERED ||
3038 dev->reg_state == NETREG_UNREGISTERING) {
3041 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
3047 netif_setup_tc(dev, txq);
3049 dev->real_num_tx_queues = txq;
3053 qdisc_reset_all_tx_gt(dev, txq);
3055 netif_reset_xps_queues_gt(dev, txq);
3059 dev->real_num_tx_queues = txq;
3064 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
3068 * netif_set_real_num_rx_queues - set actual number of RX queues used
3069 * @dev: Network device
3070 * @rxq: Actual number of RX queues
3072 * This must be called either with the rtnl_lock held or before
3073 * registration of the net device. Returns 0 on success, or a
3074 * negative error code. If called before registration, it always
3077 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
3081 if (rxq < 1 || rxq > dev->num_rx_queues)
3084 if (dev->reg_state == NETREG_REGISTERED) {
3087 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
3093 dev->real_num_rx_queues = rxq;
3096 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3100 * netif_get_num_default_rss_queues - default number of RSS queues
3102 * This routine should set an upper limit on the number of RSS queues
3103 * used by default by multiqueue devices.
3105 int netif_get_num_default_rss_queues(void)
3107 return is_kdump_kernel() ?
3108 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3110 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3112 static void __netif_reschedule(struct Qdisc *q)
3114 struct softnet_data *sd;
3115 unsigned long flags;
3117 local_irq_save(flags);
3118 sd = this_cpu_ptr(&softnet_data);
3119 q->next_sched = NULL;
3120 *sd->output_queue_tailp = q;
3121 sd->output_queue_tailp = &q->next_sched;
3122 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3123 local_irq_restore(flags);
3126 void __netif_schedule(struct Qdisc *q)
3128 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3129 __netif_reschedule(q);
3131 EXPORT_SYMBOL(__netif_schedule);
3133 struct dev_kfree_skb_cb {
3134 enum skb_free_reason reason;
3137 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3139 return (struct dev_kfree_skb_cb *)skb->cb;
3142 void netif_schedule_queue(struct netdev_queue *txq)
3145 if (!netif_xmit_stopped(txq)) {
3146 struct Qdisc *q = rcu_dereference(txq->qdisc);
3148 __netif_schedule(q);
3152 EXPORT_SYMBOL(netif_schedule_queue);
3154 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3156 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3160 q = rcu_dereference(dev_queue->qdisc);
3161 __netif_schedule(q);
3165 EXPORT_SYMBOL(netif_tx_wake_queue);
3167 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3169 unsigned long flags;
3174 if (likely(refcount_read(&skb->users) == 1)) {
3176 refcount_set(&skb->users, 0);
3177 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3180 get_kfree_skb_cb(skb)->reason = reason;
3181 local_irq_save(flags);
3182 skb->next = __this_cpu_read(softnet_data.completion_queue);
3183 __this_cpu_write(softnet_data.completion_queue, skb);
3184 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3185 local_irq_restore(flags);
3187 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3189 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3191 if (in_irq() || irqs_disabled())
3192 __dev_kfree_skb_irq(skb, reason);
3196 EXPORT_SYMBOL(__dev_kfree_skb_any);
3200 * netif_device_detach - mark device as removed
3201 * @dev: network device
3203 * Mark device as removed from system and therefore no longer available.
3205 void netif_device_detach(struct net_device *dev)
3207 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3208 netif_running(dev)) {
3209 netif_tx_stop_all_queues(dev);
3212 EXPORT_SYMBOL(netif_device_detach);
3215 * netif_device_attach - mark device as attached
3216 * @dev: network device
3218 * Mark device as attached from system and restart if needed.
3220 void netif_device_attach(struct net_device *dev)
3222 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3223 netif_running(dev)) {
3224 netif_tx_wake_all_queues(dev);
3225 __netdev_watchdog_up(dev);
3228 EXPORT_SYMBOL(netif_device_attach);
3231 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3232 * to be used as a distribution range.
3234 static u16 skb_tx_hash(const struct net_device *dev,
3235 const struct net_device *sb_dev,
3236 struct sk_buff *skb)
3240 u16 qcount = dev->real_num_tx_queues;
3243 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3245 qoffset = sb_dev->tc_to_txq[tc].offset;
3246 qcount = sb_dev->tc_to_txq[tc].count;
3249 if (skb_rx_queue_recorded(skb)) {
3250 hash = skb_get_rx_queue(skb);
3251 if (hash >= qoffset)
3253 while (unlikely(hash >= qcount))
3255 return hash + qoffset;
3258 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3261 static void skb_warn_bad_offload(const struct sk_buff *skb)
3263 static const netdev_features_t null_features;
3264 struct net_device *dev = skb->dev;
3265 const char *name = "";
3267 if (!net_ratelimit())
3271 if (dev->dev.parent)
3272 name = dev_driver_string(dev->dev.parent);
3274 name = netdev_name(dev);
3276 skb_dump(KERN_WARNING, skb, false);
3277 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3278 name, dev ? &dev->features : &null_features,
3279 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3283 * Invalidate hardware checksum when packet is to be mangled, and
3284 * complete checksum manually on outgoing path.
3286 int skb_checksum_help(struct sk_buff *skb)
3289 int ret = 0, offset;
3291 if (skb->ip_summed == CHECKSUM_COMPLETE)
3292 goto out_set_summed;
3294 if (unlikely(skb_is_gso(skb))) {
3295 skb_warn_bad_offload(skb);
3299 /* Before computing a checksum, we should make sure no frag could
3300 * be modified by an external entity : checksum could be wrong.
3302 if (skb_has_shared_frag(skb)) {
3303 ret = __skb_linearize(skb);
3308 offset = skb_checksum_start_offset(skb);
3309 BUG_ON(offset >= skb_headlen(skb));
3310 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3312 offset += skb->csum_offset;
3313 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3315 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3319 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3321 skb->ip_summed = CHECKSUM_NONE;
3325 EXPORT_SYMBOL(skb_checksum_help);
3327 int skb_crc32c_csum_help(struct sk_buff *skb)
3330 int ret = 0, offset, start;
3332 if (skb->ip_summed != CHECKSUM_PARTIAL)
3335 if (unlikely(skb_is_gso(skb)))
3338 /* Before computing a checksum, we should make sure no frag could
3339 * be modified by an external entity : checksum could be wrong.
3341 if (unlikely(skb_has_shared_frag(skb))) {
3342 ret = __skb_linearize(skb);
3346 start = skb_checksum_start_offset(skb);
3347 offset = start + offsetof(struct sctphdr, checksum);
3348 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3353 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3357 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3358 skb->len - start, ~(__u32)0,
3360 *(__le32 *)(skb->data + offset) = crc32c_csum;
3361 skb->ip_summed = CHECKSUM_NONE;
3362 skb->csum_not_inet = 0;
3367 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3369 __be16 type = skb->protocol;
3371 /* Tunnel gso handlers can set protocol to ethernet. */
3372 if (type == htons(ETH_P_TEB)) {
3375 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3378 eth = (struct ethhdr *)skb->data;
3379 type = eth->h_proto;
3382 return __vlan_get_protocol(skb, type, depth);
3386 * skb_mac_gso_segment - mac layer segmentation handler.
3387 * @skb: buffer to segment
3388 * @features: features for the output path (see dev->features)
3390 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3391 netdev_features_t features)
3393 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3394 struct packet_offload *ptype;
3395 int vlan_depth = skb->mac_len;
3396 __be16 type = skb_network_protocol(skb, &vlan_depth);
3398 if (unlikely(!type))
3399 return ERR_PTR(-EINVAL);
3401 __skb_pull(skb, vlan_depth);
3404 list_for_each_entry_rcu(ptype, &offload_base, list) {
3405 if (ptype->type == type && ptype->callbacks.gso_segment) {
3406 segs = ptype->callbacks.gso_segment(skb, features);
3412 __skb_push(skb, skb->data - skb_mac_header(skb));
3416 EXPORT_SYMBOL(skb_mac_gso_segment);
3419 /* openvswitch calls this on rx path, so we need a different check.
3421 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3424 return skb->ip_summed != CHECKSUM_PARTIAL &&
3425 skb->ip_summed != CHECKSUM_UNNECESSARY;
3427 return skb->ip_summed == CHECKSUM_NONE;
3431 * __skb_gso_segment - Perform segmentation on skb.
3432 * @skb: buffer to segment
3433 * @features: features for the output path (see dev->features)
3434 * @tx_path: whether it is called in TX path
3436 * This function segments the given skb and returns a list of segments.
3438 * It may return NULL if the skb requires no segmentation. This is
3439 * only possible when GSO is used for verifying header integrity.
3441 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3443 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3444 netdev_features_t features, bool tx_path)
3446 struct sk_buff *segs;
3448 if (unlikely(skb_needs_check(skb, tx_path))) {
3451 /* We're going to init ->check field in TCP or UDP header */
3452 err = skb_cow_head(skb, 0);
3454 return ERR_PTR(err);
3457 /* Only report GSO partial support if it will enable us to
3458 * support segmentation on this frame without needing additional
3461 if (features & NETIF_F_GSO_PARTIAL) {
3462 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3463 struct net_device *dev = skb->dev;
3465 partial_features |= dev->features & dev->gso_partial_features;
3466 if (!skb_gso_ok(skb, features | partial_features))
3467 features &= ~NETIF_F_GSO_PARTIAL;
3470 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3471 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3473 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3474 SKB_GSO_CB(skb)->encap_level = 0;
3476 skb_reset_mac_header(skb);
3477 skb_reset_mac_len(skb);
3479 segs = skb_mac_gso_segment(skb, features);
3481 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3482 skb_warn_bad_offload(skb);
3486 EXPORT_SYMBOL(__skb_gso_segment);
3488 /* Take action when hardware reception checksum errors are detected. */
3490 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3492 if (net_ratelimit()) {
3493 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3494 skb_dump(KERN_ERR, skb, true);
3498 EXPORT_SYMBOL(netdev_rx_csum_fault);
3501 /* XXX: check that highmem exists at all on the given machine. */
3502 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3504 #ifdef CONFIG_HIGHMEM
3507 if (!(dev->features & NETIF_F_HIGHDMA)) {
3508 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3509 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3511 if (PageHighMem(skb_frag_page(frag)))
3519 /* If MPLS offload request, verify we are testing hardware MPLS features
3520 * instead of standard features for the netdev.
3522 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3523 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3524 netdev_features_t features,
3527 if (eth_p_mpls(type))
3528 features &= skb->dev->mpls_features;
3533 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3534 netdev_features_t features,
3541 static netdev_features_t harmonize_features(struct sk_buff *skb,
3542 netdev_features_t features)
3546 type = skb_network_protocol(skb, NULL);
3547 features = net_mpls_features(skb, features, type);
3549 if (skb->ip_summed != CHECKSUM_NONE &&
3550 !can_checksum_protocol(features, type)) {
3551 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3553 if (illegal_highdma(skb->dev, skb))
3554 features &= ~NETIF_F_SG;
3559 netdev_features_t passthru_features_check(struct sk_buff *skb,
3560 struct net_device *dev,
3561 netdev_features_t features)
3565 EXPORT_SYMBOL(passthru_features_check);
3567 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3568 struct net_device *dev,
3569 netdev_features_t features)
3571 return vlan_features_check(skb, features);
3574 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3575 struct net_device *dev,
3576 netdev_features_t features)
3578 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3580 if (gso_segs > dev->gso_max_segs)
3581 return features & ~NETIF_F_GSO_MASK;
3583 if (!skb_shinfo(skb)->gso_type) {
3584 skb_warn_bad_offload(skb);
3585 return features & ~NETIF_F_GSO_MASK;
3588 /* Support for GSO partial features requires software
3589 * intervention before we can actually process the packets
3590 * so we need to strip support for any partial features now
3591 * and we can pull them back in after we have partially
3592 * segmented the frame.
3594 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3595 features &= ~dev->gso_partial_features;
3597 /* Make sure to clear the IPv4 ID mangling feature if the
3598 * IPv4 header has the potential to be fragmented.
3600 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3601 struct iphdr *iph = skb->encapsulation ?
3602 inner_ip_hdr(skb) : ip_hdr(skb);
3604 if (!(iph->frag_off & htons(IP_DF)))
3605 features &= ~NETIF_F_TSO_MANGLEID;
3611 netdev_features_t netif_skb_features(struct sk_buff *skb)
3613 struct net_device *dev = skb->dev;
3614 netdev_features_t features = dev->features;
3616 if (skb_is_gso(skb))
3617 features = gso_features_check(skb, dev, features);
3619 /* If encapsulation offload request, verify we are testing
3620 * hardware encapsulation features instead of standard
3621 * features for the netdev
3623 if (skb->encapsulation)
3624 features &= dev->hw_enc_features;
3626 if (skb_vlan_tagged(skb))
3627 features = netdev_intersect_features(features,
3628 dev->vlan_features |
3629 NETIF_F_HW_VLAN_CTAG_TX |
3630 NETIF_F_HW_VLAN_STAG_TX);
3632 if (dev->netdev_ops->ndo_features_check)
3633 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3636 features &= dflt_features_check(skb, dev, features);
3638 return harmonize_features(skb, features);
3640 EXPORT_SYMBOL(netif_skb_features);
3642 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3643 struct netdev_queue *txq, bool more)
3648 if (dev_nit_active(dev))
3649 dev_queue_xmit_nit(skb, dev);
3652 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3653 trace_net_dev_start_xmit(skb, dev);
3654 rc = netdev_start_xmit(skb, dev, txq, more);
3655 trace_net_dev_xmit(skb, rc, dev, len);
3660 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3661 struct netdev_queue *txq, int *ret)
3663 struct sk_buff *skb = first;
3664 int rc = NETDEV_TX_OK;
3667 struct sk_buff *next = skb->next;
3669 skb_mark_not_on_list(skb);
3670 rc = xmit_one(skb, dev, txq, next != NULL);
3671 if (unlikely(!dev_xmit_complete(rc))) {
3677 if (netif_tx_queue_stopped(txq) && skb) {
3678 rc = NETDEV_TX_BUSY;
3688 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3689 netdev_features_t features)
3691 if (skb_vlan_tag_present(skb) &&
3692 !vlan_hw_offload_capable(features, skb->vlan_proto))
3693 skb = __vlan_hwaccel_push_inside(skb);
3697 int skb_csum_hwoffload_help(struct sk_buff *skb,
3698 const netdev_features_t features)
3700 if (unlikely(skb_csum_is_sctp(skb)))
3701 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3702 skb_crc32c_csum_help(skb);
3704 if (features & NETIF_F_HW_CSUM)
3707 if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3708 switch (skb->csum_offset) {
3709 case offsetof(struct tcphdr, check):
3710 case offsetof(struct udphdr, check):
3715 return skb_checksum_help(skb);
3717 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3719 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3721 netdev_features_t features;
3723 features = netif_skb_features(skb);
3724 skb = validate_xmit_vlan(skb, features);
3728 skb = sk_validate_xmit_skb(skb, dev);
3732 if (netif_needs_gso(skb, features)) {
3733 struct sk_buff *segs;
3735 segs = skb_gso_segment(skb, features);
3743 if (skb_needs_linearize(skb, features) &&
3744 __skb_linearize(skb))
3747 /* If packet is not checksummed and device does not
3748 * support checksumming for this protocol, complete
3749 * checksumming here.
3751 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3752 if (skb->encapsulation)
3753 skb_set_inner_transport_header(skb,
3754 skb_checksum_start_offset(skb));
3756 skb_set_transport_header(skb,
3757 skb_checksum_start_offset(skb));
3758 if (skb_csum_hwoffload_help(skb, features))
3763 skb = validate_xmit_xfrm(skb, features, again);
3770 atomic_long_inc(&dev->tx_dropped);
3774 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3776 struct sk_buff *next, *head = NULL, *tail;
3778 for (; skb != NULL; skb = next) {
3780 skb_mark_not_on_list(skb);
3782 /* in case skb wont be segmented, point to itself */
3785 skb = validate_xmit_skb(skb, dev, again);
3793 /* If skb was segmented, skb->prev points to
3794 * the last segment. If not, it still contains skb.
3800 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3802 static void qdisc_pkt_len_init(struct sk_buff *skb)
3804 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3806 qdisc_skb_cb(skb)->pkt_len = skb->len;
3808 /* To get more precise estimation of bytes sent on wire,
3809 * we add to pkt_len the headers size of all segments
3811 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3812 unsigned int hdr_len;
3813 u16 gso_segs = shinfo->gso_segs;
3815 /* mac layer + network layer */
3816 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3818 /* + transport layer */
3819 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3820 const struct tcphdr *th;
3821 struct tcphdr _tcphdr;
3823 th = skb_header_pointer(skb, skb_transport_offset(skb),
3824 sizeof(_tcphdr), &_tcphdr);
3826 hdr_len += __tcp_hdrlen(th);
3828 struct udphdr _udphdr;
3830 if (skb_header_pointer(skb, skb_transport_offset(skb),
3831 sizeof(_udphdr), &_udphdr))
3832 hdr_len += sizeof(struct udphdr);
3835 if (shinfo->gso_type & SKB_GSO_DODGY)
3836 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3839 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3843 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3844 struct net_device *dev,
3845 struct netdev_queue *txq)
3847 spinlock_t *root_lock = qdisc_lock(q);
3848 struct sk_buff *to_free = NULL;
3852 qdisc_calculate_pkt_len(skb, q);
3854 if (q->flags & TCQ_F_NOLOCK) {
3855 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3858 if (unlikely(to_free))
3859 kfree_skb_list(to_free);
3864 * Heuristic to force contended enqueues to serialize on a
3865 * separate lock before trying to get qdisc main lock.
3866 * This permits qdisc->running owner to get the lock more
3867 * often and dequeue packets faster.
3869 contended = qdisc_is_running(q);
3870 if (unlikely(contended))
3871 spin_lock(&q->busylock);
3873 spin_lock(root_lock);
3874 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3875 __qdisc_drop(skb, &to_free);
3877 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3878 qdisc_run_begin(q)) {
3880 * This is a work-conserving queue; there are no old skbs
3881 * waiting to be sent out; and the qdisc is not running -
3882 * xmit the skb directly.
3885 qdisc_bstats_update(q, skb);
3887 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3888 if (unlikely(contended)) {
3889 spin_unlock(&q->busylock);
3896 rc = NET_XMIT_SUCCESS;
3898 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3899 if (qdisc_run_begin(q)) {
3900 if (unlikely(contended)) {
3901 spin_unlock(&q->busylock);
3908 spin_unlock(root_lock);
3909 if (unlikely(to_free))
3910 kfree_skb_list(to_free);
3911 if (unlikely(contended))
3912 spin_unlock(&q->busylock);
3916 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3917 static void skb_update_prio(struct sk_buff *skb)
3919 const struct netprio_map *map;
3920 const struct sock *sk;
3921 unsigned int prioidx;
3925 map = rcu_dereference_bh(skb->dev->priomap);
3928 sk = skb_to_full_sk(skb);
3932 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3934 if (prioidx < map->priomap_len)
3935 skb->priority = map->priomap[prioidx];
3938 #define skb_update_prio(skb)
3942 * dev_loopback_xmit - loop back @skb
3943 * @net: network namespace this loopback is happening in
3944 * @sk: sk needed to be a netfilter okfn
3945 * @skb: buffer to transmit
3947 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3949 skb_reset_mac_header(skb);
3950 __skb_pull(skb, skb_network_offset(skb));
3951 skb->pkt_type = PACKET_LOOPBACK;
3952 skb->ip_summed = CHECKSUM_UNNECESSARY;
3953 WARN_ON(!skb_dst(skb));
3958 EXPORT_SYMBOL(dev_loopback_xmit);
3960 #ifdef CONFIG_NET_EGRESS
3961 static struct sk_buff *
3962 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3964 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3965 struct tcf_result cl_res;
3970 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3971 qdisc_skb_cb(skb)->mru = 0;
3972 qdisc_skb_cb(skb)->post_ct = false;
3973 mini_qdisc_bstats_cpu_update(miniq, skb);
3975 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3977 case TC_ACT_RECLASSIFY:
3978 skb->tc_index = TC_H_MIN(cl_res.classid);
3981 mini_qdisc_qstats_cpu_drop(miniq);
3982 *ret = NET_XMIT_DROP;
3988 *ret = NET_XMIT_SUCCESS;
3991 case TC_ACT_REDIRECT:
3992 /* No need to push/pop skb's mac_header here on egress! */
3993 skb_do_redirect(skb);
3994 *ret = NET_XMIT_SUCCESS;
4002 #endif /* CONFIG_NET_EGRESS */
4005 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
4006 struct xps_dev_maps *dev_maps, unsigned int tci)
4008 int tc = netdev_get_prio_tc_map(dev, skb->priority);
4009 struct xps_map *map;
4010 int queue_index = -1;
4012 if (tc >= dev_maps->num_tc || tci >= dev_maps->nr_ids)
4015 tci *= dev_maps->num_tc;
4018 map = rcu_dereference(dev_maps->attr_map[tci]);
4021 queue_index = map->queues[0];
4023 queue_index = map->queues[reciprocal_scale(
4024 skb_get_hash(skb), map->len)];
4025 if (unlikely(queue_index >= dev->real_num_tx_queues))
4032 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
4033 struct sk_buff *skb)
4036 struct xps_dev_maps *dev_maps;
4037 struct sock *sk = skb->sk;
4038 int queue_index = -1;
4040 if (!static_key_false(&xps_needed))
4044 if (!static_key_false(&xps_rxqs_needed))
4047 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_RXQS]);
4049 int tci = sk_rx_queue_get(sk);
4052 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4057 if (queue_index < 0) {
4058 dev_maps = rcu_dereference(sb_dev->xps_maps[XPS_CPUS]);
4060 unsigned int tci = skb->sender_cpu - 1;
4062 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
4074 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
4075 struct net_device *sb_dev)
4079 EXPORT_SYMBOL(dev_pick_tx_zero);
4081 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
4082 struct net_device *sb_dev)
4084 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
4086 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
4088 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
4089 struct net_device *sb_dev)
4091 struct sock *sk = skb->sk;
4092 int queue_index = sk_tx_queue_get(sk);
4094 sb_dev = sb_dev ? : dev;
4096 if (queue_index < 0 || skb->ooo_okay ||
4097 queue_index >= dev->real_num_tx_queues) {
4098 int new_index = get_xps_queue(dev, sb_dev, skb);
4101 new_index = skb_tx_hash(dev, sb_dev, skb);
4103 if (queue_index != new_index && sk &&
4105 rcu_access_pointer(sk->sk_dst_cache))
4106 sk_tx_queue_set(sk, new_index);
4108 queue_index = new_index;
4113 EXPORT_SYMBOL(netdev_pick_tx);
4115 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4116 struct sk_buff *skb,
4117 struct net_device *sb_dev)
4119 int queue_index = 0;
4122 u32 sender_cpu = skb->sender_cpu - 1;
4124 if (sender_cpu >= (u32)NR_CPUS)
4125 skb->sender_cpu = raw_smp_processor_id() + 1;
4128 if (dev->real_num_tx_queues != 1) {
4129 const struct net_device_ops *ops = dev->netdev_ops;
4131 if (ops->ndo_select_queue)
4132 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4134 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4136 queue_index = netdev_cap_txqueue(dev, queue_index);
4139 skb_set_queue_mapping(skb, queue_index);
4140 return netdev_get_tx_queue(dev, queue_index);
4144 * __dev_queue_xmit - transmit a buffer
4145 * @skb: buffer to transmit
4146 * @sb_dev: suboordinate device used for L2 forwarding offload
4148 * Queue a buffer for transmission to a network device. The caller must
4149 * have set the device and priority and built the buffer before calling
4150 * this function. The function can be called from an interrupt.
4152 * A negative errno code is returned on a failure. A success does not
4153 * guarantee the frame will be transmitted as it may be dropped due
4154 * to congestion or traffic shaping.
4156 * -----------------------------------------------------------------------------------
4157 * I notice this method can also return errors from the queue disciplines,
4158 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4161 * Regardless of the return value, the skb is consumed, so it is currently
4162 * difficult to retry a send to this method. (You can bump the ref count
4163 * before sending to hold a reference for retry if you are careful.)
4165 * When calling this method, interrupts MUST be enabled. This is because
4166 * the BH enable code must have IRQs enabled so that it will not deadlock.
4169 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4171 struct net_device *dev = skb->dev;
4172 struct netdev_queue *txq;
4177 skb_reset_mac_header(skb);
4179 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4180 __skb_tstamp_tx(skb, NULL, NULL, skb->sk, SCM_TSTAMP_SCHED);
4182 /* Disable soft irqs for various locks below. Also
4183 * stops preemption for RCU.
4187 skb_update_prio(skb);
4189 qdisc_pkt_len_init(skb);
4190 #ifdef CONFIG_NET_CLS_ACT
4191 skb->tc_at_ingress = 0;
4192 # ifdef CONFIG_NET_EGRESS
4193 if (static_branch_unlikely(&egress_needed_key)) {
4194 skb = sch_handle_egress(skb, &rc, dev);
4200 /* If device/qdisc don't need skb->dst, release it right now while
4201 * its hot in this cpu cache.
4203 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4208 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4209 q = rcu_dereference_bh(txq->qdisc);
4211 trace_net_dev_queue(skb);
4213 rc = __dev_xmit_skb(skb, q, dev, txq);
4217 /* The device has no queue. Common case for software devices:
4218 * loopback, all the sorts of tunnels...
4220 * Really, it is unlikely that netif_tx_lock protection is necessary
4221 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4223 * However, it is possible, that they rely on protection
4226 * Check this and shot the lock. It is not prone from deadlocks.
4227 *Either shot noqueue qdisc, it is even simpler 8)
4229 if (dev->flags & IFF_UP) {
4230 int cpu = smp_processor_id(); /* ok because BHs are off */
4232 if (txq->xmit_lock_owner != cpu) {
4233 if (dev_xmit_recursion())
4234 goto recursion_alert;
4236 skb = validate_xmit_skb(skb, dev, &again);
4240 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4241 HARD_TX_LOCK(dev, txq, cpu);
4243 if (!netif_xmit_stopped(txq)) {
4244 dev_xmit_recursion_inc();
4245 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4246 dev_xmit_recursion_dec();
4247 if (dev_xmit_complete(rc)) {
4248 HARD_TX_UNLOCK(dev, txq);
4252 HARD_TX_UNLOCK(dev, txq);
4253 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4256 /* Recursion is detected! It is possible,
4260 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4266 rcu_read_unlock_bh();
4268 atomic_long_inc(&dev->tx_dropped);
4269 kfree_skb_list(skb);
4272 rcu_read_unlock_bh();
4276 int dev_queue_xmit(struct sk_buff *skb)
4278 return __dev_queue_xmit(skb, NULL);
4280 EXPORT_SYMBOL(dev_queue_xmit);
4282 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4284 return __dev_queue_xmit(skb, sb_dev);
4286 EXPORT_SYMBOL(dev_queue_xmit_accel);
4288 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4290 struct net_device *dev = skb->dev;
4291 struct sk_buff *orig_skb = skb;
4292 struct netdev_queue *txq;
4293 int ret = NETDEV_TX_BUSY;
4296 if (unlikely(!netif_running(dev) ||
4297 !netif_carrier_ok(dev)))
4300 skb = validate_xmit_skb_list(skb, dev, &again);
4301 if (skb != orig_skb)
4304 skb_set_queue_mapping(skb, queue_id);
4305 txq = skb_get_tx_queue(dev, skb);
4306 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4310 dev_xmit_recursion_inc();
4311 HARD_TX_LOCK(dev, txq, smp_processor_id());
4312 if (!netif_xmit_frozen_or_drv_stopped(txq))
4313 ret = netdev_start_xmit(skb, dev, txq, false);
4314 HARD_TX_UNLOCK(dev, txq);
4315 dev_xmit_recursion_dec();
4320 atomic_long_inc(&dev->tx_dropped);
4321 kfree_skb_list(skb);
4322 return NET_XMIT_DROP;
4324 EXPORT_SYMBOL(__dev_direct_xmit);
4326 /*************************************************************************
4328 *************************************************************************/
4330 int netdev_max_backlog __read_mostly = 1000;
4331 EXPORT_SYMBOL(netdev_max_backlog);
4333 int netdev_tstamp_prequeue __read_mostly = 1;
4334 int netdev_budget __read_mostly = 300;
4335 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4336 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4337 int weight_p __read_mostly = 64; /* old backlog weight */
4338 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4339 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4340 int dev_rx_weight __read_mostly = 64;
4341 int dev_tx_weight __read_mostly = 64;
4342 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4343 int gro_normal_batch __read_mostly = 8;
4345 /* Called with irq disabled */
4346 static inline void ____napi_schedule(struct softnet_data *sd,
4347 struct napi_struct *napi)
4349 struct task_struct *thread;
4351 if (test_bit(NAPI_STATE_THREADED, &napi->state)) {
4352 /* Paired with smp_mb__before_atomic() in
4353 * napi_enable()/dev_set_threaded().
4354 * Use READ_ONCE() to guarantee a complete
4355 * read on napi->thread. Only call
4356 * wake_up_process() when it's not NULL.
4358 thread = READ_ONCE(napi->thread);
4360 /* Avoid doing set_bit() if the thread is in
4361 * INTERRUPTIBLE state, cause napi_thread_wait()
4362 * makes sure to proceed with napi polling
4363 * if the thread is explicitly woken from here.
4365 if (READ_ONCE(thread->state) != TASK_INTERRUPTIBLE)
4366 set_bit(NAPI_STATE_SCHED_THREADED, &napi->state);
4367 wake_up_process(thread);
4372 list_add_tail(&napi->poll_list, &sd->poll_list);
4373 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4378 /* One global table that all flow-based protocols share. */
4379 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4380 EXPORT_SYMBOL(rps_sock_flow_table);
4381 u32 rps_cpu_mask __read_mostly;
4382 EXPORT_SYMBOL(rps_cpu_mask);
4384 struct static_key_false rps_needed __read_mostly;
4385 EXPORT_SYMBOL(rps_needed);
4386 struct static_key_false rfs_needed __read_mostly;
4387 EXPORT_SYMBOL(rfs_needed);
4389 static struct rps_dev_flow *
4390 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4391 struct rps_dev_flow *rflow, u16 next_cpu)
4393 if (next_cpu < nr_cpu_ids) {
4394 #ifdef CONFIG_RFS_ACCEL
4395 struct netdev_rx_queue *rxqueue;
4396 struct rps_dev_flow_table *flow_table;
4397 struct rps_dev_flow *old_rflow;
4402 /* Should we steer this flow to a different hardware queue? */
4403 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4404 !(dev->features & NETIF_F_NTUPLE))
4406 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4407 if (rxq_index == skb_get_rx_queue(skb))
4410 rxqueue = dev->_rx + rxq_index;
4411 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4414 flow_id = skb_get_hash(skb) & flow_table->mask;
4415 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4416 rxq_index, flow_id);
4420 rflow = &flow_table->flows[flow_id];
4422 if (old_rflow->filter == rflow->filter)
4423 old_rflow->filter = RPS_NO_FILTER;
4427 per_cpu(softnet_data, next_cpu).input_queue_head;
4430 rflow->cpu = next_cpu;
4435 * get_rps_cpu is called from netif_receive_skb and returns the target
4436 * CPU from the RPS map of the receiving queue for a given skb.
4437 * rcu_read_lock must be held on entry.
4439 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4440 struct rps_dev_flow **rflowp)
4442 const struct rps_sock_flow_table *sock_flow_table;
4443 struct netdev_rx_queue *rxqueue = dev->_rx;
4444 struct rps_dev_flow_table *flow_table;
4445 struct rps_map *map;
4450 if (skb_rx_queue_recorded(skb)) {
4451 u16 index = skb_get_rx_queue(skb);
4453 if (unlikely(index >= dev->real_num_rx_queues)) {
4454 WARN_ONCE(dev->real_num_rx_queues > 1,
4455 "%s received packet on queue %u, but number "
4456 "of RX queues is %u\n",
4457 dev->name, index, dev->real_num_rx_queues);
4463 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4465 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4466 map = rcu_dereference(rxqueue->rps_map);
4467 if (!flow_table && !map)
4470 skb_reset_network_header(skb);
4471 hash = skb_get_hash(skb);
4475 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4476 if (flow_table && sock_flow_table) {
4477 struct rps_dev_flow *rflow;
4481 /* First check into global flow table if there is a match */
4482 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4483 if ((ident ^ hash) & ~rps_cpu_mask)
4486 next_cpu = ident & rps_cpu_mask;
4488 /* OK, now we know there is a match,
4489 * we can look at the local (per receive queue) flow table
4491 rflow = &flow_table->flows[hash & flow_table->mask];
4495 * If the desired CPU (where last recvmsg was done) is
4496 * different from current CPU (one in the rx-queue flow
4497 * table entry), switch if one of the following holds:
4498 * - Current CPU is unset (>= nr_cpu_ids).
4499 * - Current CPU is offline.
4500 * - The current CPU's queue tail has advanced beyond the
4501 * last packet that was enqueued using this table entry.
4502 * This guarantees that all previous packets for the flow
4503 * have been dequeued, thus preserving in order delivery.
4505 if (unlikely(tcpu != next_cpu) &&
4506 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4507 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4508 rflow->last_qtail)) >= 0)) {
4510 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4513 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4523 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4524 if (cpu_online(tcpu)) {
4534 #ifdef CONFIG_RFS_ACCEL
4537 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4538 * @dev: Device on which the filter was set
4539 * @rxq_index: RX queue index
4540 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4541 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4543 * Drivers that implement ndo_rx_flow_steer() should periodically call
4544 * this function for each installed filter and remove the filters for
4545 * which it returns %true.
4547 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4548 u32 flow_id, u16 filter_id)
4550 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4551 struct rps_dev_flow_table *flow_table;
4552 struct rps_dev_flow *rflow;
4557 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4558 if (flow_table && flow_id <= flow_table->mask) {
4559 rflow = &flow_table->flows[flow_id];
4560 cpu = READ_ONCE(rflow->cpu);
4561 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4562 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4563 rflow->last_qtail) <
4564 (int)(10 * flow_table->mask)))
4570 EXPORT_SYMBOL(rps_may_expire_flow);
4572 #endif /* CONFIG_RFS_ACCEL */
4574 /* Called from hardirq (IPI) context */
4575 static void rps_trigger_softirq(void *data)
4577 struct softnet_data *sd = data;
4579 ____napi_schedule(sd, &sd->backlog);
4583 #endif /* CONFIG_RPS */
4586 * Check if this softnet_data structure is another cpu one
4587 * If yes, queue it to our IPI list and return 1
4590 static int rps_ipi_queued(struct softnet_data *sd)
4593 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4596 sd->rps_ipi_next = mysd->rps_ipi_list;
4597 mysd->rps_ipi_list = sd;
4599 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4602 #endif /* CONFIG_RPS */
4606 #ifdef CONFIG_NET_FLOW_LIMIT
4607 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4610 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4612 #ifdef CONFIG_NET_FLOW_LIMIT
4613 struct sd_flow_limit *fl;
4614 struct softnet_data *sd;
4615 unsigned int old_flow, new_flow;
4617 if (qlen < (netdev_max_backlog >> 1))
4620 sd = this_cpu_ptr(&softnet_data);
4623 fl = rcu_dereference(sd->flow_limit);
4625 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4626 old_flow = fl->history[fl->history_head];
4627 fl->history[fl->history_head] = new_flow;
4630 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4632 if (likely(fl->buckets[old_flow]))
4633 fl->buckets[old_flow]--;
4635 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4647 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4648 * queue (may be a remote CPU queue).
4650 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4651 unsigned int *qtail)
4653 struct softnet_data *sd;
4654 unsigned long flags;
4657 sd = &per_cpu(softnet_data, cpu);
4659 local_irq_save(flags);
4662 if (!netif_running(skb->dev))
4664 qlen = skb_queue_len(&sd->input_pkt_queue);
4665 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4668 __skb_queue_tail(&sd->input_pkt_queue, skb);
4669 input_queue_tail_incr_save(sd, qtail);
4671 local_irq_restore(flags);
4672 return NET_RX_SUCCESS;
4675 /* Schedule NAPI for backlog device
4676 * We can use non atomic operation since we own the queue lock
4678 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4679 if (!rps_ipi_queued(sd))
4680 ____napi_schedule(sd, &sd->backlog);
4689 local_irq_restore(flags);
4691 atomic_long_inc(&skb->dev->rx_dropped);
4696 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4698 struct net_device *dev = skb->dev;
4699 struct netdev_rx_queue *rxqueue;
4703 if (skb_rx_queue_recorded(skb)) {
4704 u16 index = skb_get_rx_queue(skb);
4706 if (unlikely(index >= dev->real_num_rx_queues)) {
4707 WARN_ONCE(dev->real_num_rx_queues > 1,
4708 "%s received packet on queue %u, but number "
4709 "of RX queues is %u\n",
4710 dev->name, index, dev->real_num_rx_queues);
4712 return rxqueue; /* Return first rxqueue */
4719 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4720 struct xdp_buff *xdp,
4721 struct bpf_prog *xdp_prog)
4723 void *orig_data, *orig_data_end, *hard_start;
4724 struct netdev_rx_queue *rxqueue;
4725 u32 metalen, act = XDP_DROP;
4726 u32 mac_len, frame_sz;
4727 __be16 orig_eth_type;
4732 /* Reinjected packets coming from act_mirred or similar should
4733 * not get XDP generic processing.
4735 if (skb_is_redirected(skb))
4738 /* XDP packets must be linear and must have sufficient headroom
4739 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4740 * native XDP provides, thus we need to do it here as well.
4742 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4743 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4744 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4745 int troom = skb->tail + skb->data_len - skb->end;
4747 /* In case we have to go down the path and also linearize,
4748 * then lets do the pskb_expand_head() work just once here.
4750 if (pskb_expand_head(skb,
4751 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4752 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4754 if (skb_linearize(skb))
4758 /* The XDP program wants to see the packet starting at the MAC
4761 mac_len = skb->data - skb_mac_header(skb);
4762 hard_start = skb->data - skb_headroom(skb);
4764 /* SKB "head" area always have tailroom for skb_shared_info */
4765 frame_sz = (void *)skb_end_pointer(skb) - hard_start;
4766 frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4768 rxqueue = netif_get_rxqueue(skb);
4769 xdp_init_buff(xdp, frame_sz, &rxqueue->xdp_rxq);
4770 xdp_prepare_buff(xdp, hard_start, skb_headroom(skb) - mac_len,
4771 skb_headlen(skb) + mac_len, true);
4773 orig_data_end = xdp->data_end;
4774 orig_data = xdp->data;
4775 eth = (struct ethhdr *)xdp->data;
4776 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4777 orig_eth_type = eth->h_proto;
4779 act = bpf_prog_run_xdp(xdp_prog, xdp);
4781 /* check if bpf_xdp_adjust_head was used */
4782 off = xdp->data - orig_data;
4785 __skb_pull(skb, off);
4787 __skb_push(skb, -off);
4789 skb->mac_header += off;
4790 skb_reset_network_header(skb);
4793 /* check if bpf_xdp_adjust_tail was used */
4794 off = xdp->data_end - orig_data_end;
4796 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4797 skb->len += off; /* positive on grow, negative on shrink */
4800 /* check if XDP changed eth hdr such SKB needs update */
4801 eth = (struct ethhdr *)xdp->data;
4802 if ((orig_eth_type != eth->h_proto) ||
4803 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4804 __skb_push(skb, ETH_HLEN);
4805 skb->protocol = eth_type_trans(skb, skb->dev);
4811 __skb_push(skb, mac_len);
4814 metalen = xdp->data - xdp->data_meta;
4816 skb_metadata_set(skb, metalen);
4819 bpf_warn_invalid_xdp_action(act);
4822 trace_xdp_exception(skb->dev, xdp_prog, act);
4833 /* When doing generic XDP we have to bypass the qdisc layer and the
4834 * network taps in order to match in-driver-XDP behavior.
4836 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4838 struct net_device *dev = skb->dev;
4839 struct netdev_queue *txq;
4840 bool free_skb = true;
4843 txq = netdev_core_pick_tx(dev, skb, NULL);
4844 cpu = smp_processor_id();
4845 HARD_TX_LOCK(dev, txq, cpu);
4846 if (!netif_xmit_stopped(txq)) {
4847 rc = netdev_start_xmit(skb, dev, txq, 0);
4848 if (dev_xmit_complete(rc))
4851 HARD_TX_UNLOCK(dev, txq);
4853 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4858 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4860 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4863 struct xdp_buff xdp;
4867 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4868 if (act != XDP_PASS) {
4871 err = xdp_do_generic_redirect(skb->dev, skb,
4877 generic_xdp_tx(skb, xdp_prog);
4888 EXPORT_SYMBOL_GPL(do_xdp_generic);
4890 static int netif_rx_internal(struct sk_buff *skb)
4894 net_timestamp_check(netdev_tstamp_prequeue, skb);
4896 trace_netif_rx(skb);
4899 if (static_branch_unlikely(&rps_needed)) {
4900 struct rps_dev_flow voidflow, *rflow = &voidflow;
4906 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4908 cpu = smp_processor_id();
4910 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4919 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4926 * netif_rx - post buffer to the network code
4927 * @skb: buffer to post
4929 * This function receives a packet from a device driver and queues it for
4930 * the upper (protocol) levels to process. It always succeeds. The buffer
4931 * may be dropped during processing for congestion control or by the
4935 * NET_RX_SUCCESS (no congestion)
4936 * NET_RX_DROP (packet was dropped)
4940 int netif_rx(struct sk_buff *skb)
4944 trace_netif_rx_entry(skb);
4946 ret = netif_rx_internal(skb);
4947 trace_netif_rx_exit(ret);
4951 EXPORT_SYMBOL(netif_rx);
4953 int netif_rx_ni(struct sk_buff *skb)
4957 trace_netif_rx_ni_entry(skb);
4960 err = netif_rx_internal(skb);
4961 if (local_softirq_pending())
4964 trace_netif_rx_ni_exit(err);
4968 EXPORT_SYMBOL(netif_rx_ni);
4970 int netif_rx_any_context(struct sk_buff *skb)
4973 * If invoked from contexts which do not invoke bottom half
4974 * processing either at return from interrupt or when softrqs are
4975 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
4979 return netif_rx(skb);
4981 return netif_rx_ni(skb);
4983 EXPORT_SYMBOL(netif_rx_any_context);
4985 static __latent_entropy void net_tx_action(struct softirq_action *h)
4987 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4989 if (sd->completion_queue) {
4990 struct sk_buff *clist;
4992 local_irq_disable();
4993 clist = sd->completion_queue;
4994 sd->completion_queue = NULL;
4998 struct sk_buff *skb = clist;
5000 clist = clist->next;
5002 WARN_ON(refcount_read(&skb->users));
5003 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
5004 trace_consume_skb(skb);
5006 trace_kfree_skb(skb, net_tx_action);
5008 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
5011 __kfree_skb_defer(skb);
5015 if (sd->output_queue) {
5018 local_irq_disable();
5019 head = sd->output_queue;
5020 sd->output_queue = NULL;
5021 sd->output_queue_tailp = &sd->output_queue;
5025 struct Qdisc *q = head;
5026 spinlock_t *root_lock = NULL;
5028 head = head->next_sched;
5030 if (!(q->flags & TCQ_F_NOLOCK)) {
5031 root_lock = qdisc_lock(q);
5032 spin_lock(root_lock);
5034 /* We need to make sure head->next_sched is read
5035 * before clearing __QDISC_STATE_SCHED
5037 smp_mb__before_atomic();
5038 clear_bit(__QDISC_STATE_SCHED, &q->state);
5041 spin_unlock(root_lock);
5045 xfrm_dev_backlog(sd);
5048 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
5049 /* This hook is defined here for ATM LANE */
5050 int (*br_fdb_test_addr_hook)(struct net_device *dev,
5051 unsigned char *addr) __read_mostly;
5052 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
5055 static inline struct sk_buff *
5056 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
5057 struct net_device *orig_dev, bool *another)
5059 #ifdef CONFIG_NET_CLS_ACT
5060 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
5061 struct tcf_result cl_res;
5063 /* If there's at least one ingress present somewhere (so
5064 * we get here via enabled static key), remaining devices
5065 * that are not configured with an ingress qdisc will bail
5072 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5076 qdisc_skb_cb(skb)->pkt_len = skb->len;
5077 qdisc_skb_cb(skb)->mru = 0;
5078 qdisc_skb_cb(skb)->post_ct = false;
5079 skb->tc_at_ingress = 1;
5080 mini_qdisc_bstats_cpu_update(miniq, skb);
5082 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list,
5085 case TC_ACT_RECLASSIFY:
5086 skb->tc_index = TC_H_MIN(cl_res.classid);
5089 mini_qdisc_qstats_cpu_drop(miniq);
5097 case TC_ACT_REDIRECT:
5098 /* skb_mac_header check was done by cls/act_bpf, so
5099 * we can safely push the L2 header back before
5100 * redirecting to another netdev
5102 __skb_push(skb, skb->mac_len);
5103 if (skb_do_redirect(skb) == -EAGAIN) {
5104 __skb_pull(skb, skb->mac_len);
5109 case TC_ACT_CONSUMED:
5114 #endif /* CONFIG_NET_CLS_ACT */
5119 * netdev_is_rx_handler_busy - check if receive handler is registered
5120 * @dev: device to check
5122 * Check if a receive handler is already registered for a given device.
5123 * Return true if there one.
5125 * The caller must hold the rtnl_mutex.
5127 bool netdev_is_rx_handler_busy(struct net_device *dev)
5130 return dev && rtnl_dereference(dev->rx_handler);
5132 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5135 * netdev_rx_handler_register - register receive handler
5136 * @dev: device to register a handler for
5137 * @rx_handler: receive handler to register
5138 * @rx_handler_data: data pointer that is used by rx handler
5140 * Register a receive handler for a device. This handler will then be
5141 * called from __netif_receive_skb. A negative errno code is returned
5144 * The caller must hold the rtnl_mutex.
5146 * For a general description of rx_handler, see enum rx_handler_result.
5148 int netdev_rx_handler_register(struct net_device *dev,
5149 rx_handler_func_t *rx_handler,
5150 void *rx_handler_data)
5152 if (netdev_is_rx_handler_busy(dev))
5155 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5158 /* Note: rx_handler_data must be set before rx_handler */
5159 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5160 rcu_assign_pointer(dev->rx_handler, rx_handler);
5164 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5167 * netdev_rx_handler_unregister - unregister receive handler
5168 * @dev: device to unregister a handler from
5170 * Unregister a receive handler from a device.
5172 * The caller must hold the rtnl_mutex.
5174 void netdev_rx_handler_unregister(struct net_device *dev)
5178 RCU_INIT_POINTER(dev->rx_handler, NULL);
5179 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5180 * section has a guarantee to see a non NULL rx_handler_data
5184 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5186 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5189 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5190 * the special handling of PFMEMALLOC skbs.
5192 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5194 switch (skb->protocol) {
5195 case htons(ETH_P_ARP):
5196 case htons(ETH_P_IP):
5197 case htons(ETH_P_IPV6):
5198 case htons(ETH_P_8021Q):
5199 case htons(ETH_P_8021AD):
5206 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5207 int *ret, struct net_device *orig_dev)
5209 if (nf_hook_ingress_active(skb)) {
5213 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5218 ingress_retval = nf_hook_ingress(skb);
5220 return ingress_retval;
5225 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5226 struct packet_type **ppt_prev)
5228 struct packet_type *ptype, *pt_prev;
5229 rx_handler_func_t *rx_handler;
5230 struct sk_buff *skb = *pskb;
5231 struct net_device *orig_dev;
5232 bool deliver_exact = false;
5233 int ret = NET_RX_DROP;
5236 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5238 trace_netif_receive_skb(skb);
5240 orig_dev = skb->dev;
5242 skb_reset_network_header(skb);
5243 if (!skb_transport_header_was_set(skb))
5244 skb_reset_transport_header(skb);
5245 skb_reset_mac_len(skb);
5250 skb->skb_iif = skb->dev->ifindex;
5252 __this_cpu_inc(softnet_data.processed);
5254 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5258 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5261 if (ret2 != XDP_PASS) {
5265 skb_reset_mac_len(skb);
5268 if (eth_type_vlan(skb->protocol)) {
5269 skb = skb_vlan_untag(skb);
5274 if (skb_skip_tc_classify(skb))
5280 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5282 ret = deliver_skb(skb, pt_prev, orig_dev);
5286 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5288 ret = deliver_skb(skb, pt_prev, orig_dev);
5293 #ifdef CONFIG_NET_INGRESS
5294 if (static_branch_unlikely(&ingress_needed_key)) {
5295 bool another = false;
5297 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5304 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5308 skb_reset_redirect(skb);
5310 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5313 if (skb_vlan_tag_present(skb)) {
5315 ret = deliver_skb(skb, pt_prev, orig_dev);
5318 if (vlan_do_receive(&skb))
5320 else if (unlikely(!skb))
5324 rx_handler = rcu_dereference(skb->dev->rx_handler);
5327 ret = deliver_skb(skb, pt_prev, orig_dev);
5330 switch (rx_handler(&skb)) {
5331 case RX_HANDLER_CONSUMED:
5332 ret = NET_RX_SUCCESS;
5334 case RX_HANDLER_ANOTHER:
5336 case RX_HANDLER_EXACT:
5337 deliver_exact = true;
5339 case RX_HANDLER_PASS:
5346 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5348 if (skb_vlan_tag_get_id(skb)) {
5349 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5352 skb->pkt_type = PACKET_OTHERHOST;
5353 } else if (eth_type_vlan(skb->protocol)) {
5354 /* Outer header is 802.1P with vlan 0, inner header is
5355 * 802.1Q or 802.1AD and vlan_do_receive() above could
5356 * not find vlan dev for vlan id 0.
5358 __vlan_hwaccel_clear_tag(skb);
5359 skb = skb_vlan_untag(skb);
5362 if (vlan_do_receive(&skb))
5363 /* After stripping off 802.1P header with vlan 0
5364 * vlan dev is found for inner header.
5367 else if (unlikely(!skb))
5370 /* We have stripped outer 802.1P vlan 0 header.
5371 * But could not find vlan dev.
5372 * check again for vlan id to set OTHERHOST.
5376 /* Note: we might in the future use prio bits
5377 * and set skb->priority like in vlan_do_receive()
5378 * For the time being, just ignore Priority Code Point
5380 __vlan_hwaccel_clear_tag(skb);
5383 type = skb->protocol;
5385 /* deliver only exact match when indicated */
5386 if (likely(!deliver_exact)) {
5387 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5388 &ptype_base[ntohs(type) &
5392 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5393 &orig_dev->ptype_specific);
5395 if (unlikely(skb->dev != orig_dev)) {
5396 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5397 &skb->dev->ptype_specific);
5401 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5403 *ppt_prev = pt_prev;
5407 atomic_long_inc(&skb->dev->rx_dropped);
5409 atomic_long_inc(&skb->dev->rx_nohandler);
5411 /* Jamal, now you will not able to escape explaining
5412 * me how you were going to use this. :-)
5418 /* The invariant here is that if *ppt_prev is not NULL
5419 * then skb should also be non-NULL.
5421 * Apparently *ppt_prev assignment above holds this invariant due to
5422 * skb dereferencing near it.
5428 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5430 struct net_device *orig_dev = skb->dev;
5431 struct packet_type *pt_prev = NULL;
5434 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5436 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5437 skb->dev, pt_prev, orig_dev);
5442 * netif_receive_skb_core - special purpose version of netif_receive_skb
5443 * @skb: buffer to process
5445 * More direct receive version of netif_receive_skb(). It should
5446 * only be used by callers that have a need to skip RPS and Generic XDP.
5447 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5449 * This function may only be called from softirq context and interrupts
5450 * should be enabled.
5452 * Return values (usually ignored):
5453 * NET_RX_SUCCESS: no congestion
5454 * NET_RX_DROP: packet was dropped
5456 int netif_receive_skb_core(struct sk_buff *skb)
5461 ret = __netif_receive_skb_one_core(skb, false);
5466 EXPORT_SYMBOL(netif_receive_skb_core);
5468 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5469 struct packet_type *pt_prev,
5470 struct net_device *orig_dev)
5472 struct sk_buff *skb, *next;
5476 if (list_empty(head))
5478 if (pt_prev->list_func != NULL)
5479 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5480 ip_list_rcv, head, pt_prev, orig_dev);
5482 list_for_each_entry_safe(skb, next, head, list) {
5483 skb_list_del_init(skb);
5484 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5488 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5490 /* Fast-path assumptions:
5491 * - There is no RX handler.
5492 * - Only one packet_type matches.
5493 * If either of these fails, we will end up doing some per-packet
5494 * processing in-line, then handling the 'last ptype' for the whole
5495 * sublist. This can't cause out-of-order delivery to any single ptype,
5496 * because the 'last ptype' must be constant across the sublist, and all
5497 * other ptypes are handled per-packet.
5499 /* Current (common) ptype of sublist */
5500 struct packet_type *pt_curr = NULL;
5501 /* Current (common) orig_dev of sublist */
5502 struct net_device *od_curr = NULL;
5503 struct list_head sublist;
5504 struct sk_buff *skb, *next;
5506 INIT_LIST_HEAD(&sublist);
5507 list_for_each_entry_safe(skb, next, head, list) {
5508 struct net_device *orig_dev = skb->dev;
5509 struct packet_type *pt_prev = NULL;
5511 skb_list_del_init(skb);
5512 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5515 if (pt_curr != pt_prev || od_curr != orig_dev) {
5516 /* dispatch old sublist */
5517 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5518 /* start new sublist */
5519 INIT_LIST_HEAD(&sublist);
5523 list_add_tail(&skb->list, &sublist);
5526 /* dispatch final sublist */
5527 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5530 static int __netif_receive_skb(struct sk_buff *skb)
5534 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5535 unsigned int noreclaim_flag;
5538 * PFMEMALLOC skbs are special, they should
5539 * - be delivered to SOCK_MEMALLOC sockets only
5540 * - stay away from userspace
5541 * - have bounded memory usage
5543 * Use PF_MEMALLOC as this saves us from propagating the allocation
5544 * context down to all allocation sites.
5546 noreclaim_flag = memalloc_noreclaim_save();
5547 ret = __netif_receive_skb_one_core(skb, true);
5548 memalloc_noreclaim_restore(noreclaim_flag);
5550 ret = __netif_receive_skb_one_core(skb, false);
5555 static void __netif_receive_skb_list(struct list_head *head)
5557 unsigned long noreclaim_flag = 0;
5558 struct sk_buff *skb, *next;
5559 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5561 list_for_each_entry_safe(skb, next, head, list) {
5562 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5563 struct list_head sublist;
5565 /* Handle the previous sublist */
5566 list_cut_before(&sublist, head, &skb->list);
5567 if (!list_empty(&sublist))
5568 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5569 pfmemalloc = !pfmemalloc;
5570 /* See comments in __netif_receive_skb */
5572 noreclaim_flag = memalloc_noreclaim_save();
5574 memalloc_noreclaim_restore(noreclaim_flag);
5577 /* Handle the remaining sublist */
5578 if (!list_empty(head))
5579 __netif_receive_skb_list_core(head, pfmemalloc);
5580 /* Restore pflags */
5582 memalloc_noreclaim_restore(noreclaim_flag);
5585 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5587 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5588 struct bpf_prog *new = xdp->prog;
5594 mutex_lock(&new->aux->used_maps_mutex);
5596 /* generic XDP does not work with DEVMAPs that can
5597 * have a bpf_prog installed on an entry
5599 for (i = 0; i < new->aux->used_map_cnt; i++) {
5600 if (dev_map_can_have_prog(new->aux->used_maps[i]) ||
5601 cpu_map_prog_allowed(new->aux->used_maps[i])) {
5602 mutex_unlock(&new->aux->used_maps_mutex);
5607 mutex_unlock(&new->aux->used_maps_mutex);
5610 switch (xdp->command) {
5611 case XDP_SETUP_PROG:
5612 rcu_assign_pointer(dev->xdp_prog, new);
5617 static_branch_dec(&generic_xdp_needed_key);
5618 } else if (new && !old) {
5619 static_branch_inc(&generic_xdp_needed_key);
5620 dev_disable_lro(dev);
5621 dev_disable_gro_hw(dev);
5633 static int netif_receive_skb_internal(struct sk_buff *skb)
5637 net_timestamp_check(netdev_tstamp_prequeue, skb);
5639 if (skb_defer_rx_timestamp(skb))
5640 return NET_RX_SUCCESS;
5644 if (static_branch_unlikely(&rps_needed)) {
5645 struct rps_dev_flow voidflow, *rflow = &voidflow;
5646 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5649 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5655 ret = __netif_receive_skb(skb);
5660 static void netif_receive_skb_list_internal(struct list_head *head)
5662 struct sk_buff *skb, *next;
5663 struct list_head sublist;
5665 INIT_LIST_HEAD(&sublist);
5666 list_for_each_entry_safe(skb, next, head, list) {
5667 net_timestamp_check(netdev_tstamp_prequeue, skb);
5668 skb_list_del_init(skb);
5669 if (!skb_defer_rx_timestamp(skb))
5670 list_add_tail(&skb->list, &sublist);
5672 list_splice_init(&sublist, head);
5676 if (static_branch_unlikely(&rps_needed)) {
5677 list_for_each_entry_safe(skb, next, head, list) {
5678 struct rps_dev_flow voidflow, *rflow = &voidflow;
5679 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5682 /* Will be handled, remove from list */
5683 skb_list_del_init(skb);
5684 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5689 __netif_receive_skb_list(head);
5694 * netif_receive_skb - process receive buffer from network
5695 * @skb: buffer to process
5697 * netif_receive_skb() is the main receive data processing function.
5698 * It always succeeds. The buffer may be dropped during processing
5699 * for congestion control or by the protocol layers.
5701 * This function may only be called from softirq context and interrupts
5702 * should be enabled.
5704 * Return values (usually ignored):
5705 * NET_RX_SUCCESS: no congestion
5706 * NET_RX_DROP: packet was dropped
5708 int netif_receive_skb(struct sk_buff *skb)
5712 trace_netif_receive_skb_entry(skb);
5714 ret = netif_receive_skb_internal(skb);
5715 trace_netif_receive_skb_exit(ret);
5719 EXPORT_SYMBOL(netif_receive_skb);
5722 * netif_receive_skb_list - process many receive buffers from network
5723 * @head: list of skbs to process.
5725 * Since return value of netif_receive_skb() is normally ignored, and
5726 * wouldn't be meaningful for a list, this function returns void.
5728 * This function may only be called from softirq context and interrupts
5729 * should be enabled.
5731 void netif_receive_skb_list(struct list_head *head)
5733 struct sk_buff *skb;
5735 if (list_empty(head))
5737 if (trace_netif_receive_skb_list_entry_enabled()) {
5738 list_for_each_entry(skb, head, list)
5739 trace_netif_receive_skb_list_entry(skb);
5741 netif_receive_skb_list_internal(head);
5742 trace_netif_receive_skb_list_exit(0);
5744 EXPORT_SYMBOL(netif_receive_skb_list);
5746 static DEFINE_PER_CPU(struct work_struct, flush_works);
5748 /* Network device is going away, flush any packets still pending */
5749 static void flush_backlog(struct work_struct *work)
5751 struct sk_buff *skb, *tmp;
5752 struct softnet_data *sd;
5755 sd = this_cpu_ptr(&softnet_data);
5757 local_irq_disable();
5759 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5760 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5761 __skb_unlink(skb, &sd->input_pkt_queue);
5762 dev_kfree_skb_irq(skb);
5763 input_queue_head_incr(sd);
5769 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5770 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5771 __skb_unlink(skb, &sd->process_queue);
5773 input_queue_head_incr(sd);
5779 static bool flush_required(int cpu)
5781 #if IS_ENABLED(CONFIG_RPS)
5782 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5785 local_irq_disable();
5788 /* as insertion into process_queue happens with the rps lock held,
5789 * process_queue access may race only with dequeue
5791 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5792 !skb_queue_empty_lockless(&sd->process_queue);
5798 /* without RPS we can't safely check input_pkt_queue: during a
5799 * concurrent remote skb_queue_splice() we can detect as empty both
5800 * input_pkt_queue and process_queue even if the latter could end-up
5801 * containing a lot of packets.
5806 static void flush_all_backlogs(void)
5808 static cpumask_t flush_cpus;
5811 /* since we are under rtnl lock protection we can use static data
5812 * for the cpumask and avoid allocating on stack the possibly
5819 cpumask_clear(&flush_cpus);
5820 for_each_online_cpu(cpu) {
5821 if (flush_required(cpu)) {
5822 queue_work_on(cpu, system_highpri_wq,
5823 per_cpu_ptr(&flush_works, cpu));
5824 cpumask_set_cpu(cpu, &flush_cpus);
5828 /* we can have in flight packet[s] on the cpus we are not flushing,
5829 * synchronize_net() in unregister_netdevice_many() will take care of
5832 for_each_cpu(cpu, &flush_cpus)
5833 flush_work(per_cpu_ptr(&flush_works, cpu));
5838 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5839 static void gro_normal_list(struct napi_struct *napi)
5841 if (!napi->rx_count)
5843 netif_receive_skb_list_internal(&napi->rx_list);
5844 INIT_LIST_HEAD(&napi->rx_list);
5848 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5849 * pass the whole batch up to the stack.
5851 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb, int segs)
5853 list_add_tail(&skb->list, &napi->rx_list);
5854 napi->rx_count += segs;
5855 if (napi->rx_count >= gro_normal_batch)
5856 gro_normal_list(napi);
5859 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5861 struct packet_offload *ptype;
5862 __be16 type = skb->protocol;
5863 struct list_head *head = &offload_base;
5866 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5868 if (NAPI_GRO_CB(skb)->count == 1) {
5869 skb_shinfo(skb)->gso_size = 0;
5874 list_for_each_entry_rcu(ptype, head, list) {
5875 if (ptype->type != type || !ptype->callbacks.gro_complete)
5878 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5879 ipv6_gro_complete, inet_gro_complete,
5886 WARN_ON(&ptype->list == head);
5888 return NET_RX_SUCCESS;
5892 gro_normal_one(napi, skb, NAPI_GRO_CB(skb)->count);
5893 return NET_RX_SUCCESS;
5896 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5899 struct list_head *head = &napi->gro_hash[index].list;
5900 struct sk_buff *skb, *p;
5902 list_for_each_entry_safe_reverse(skb, p, head, list) {
5903 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5905 skb_list_del_init(skb);
5906 napi_gro_complete(napi, skb);
5907 napi->gro_hash[index].count--;
5910 if (!napi->gro_hash[index].count)
5911 __clear_bit(index, &napi->gro_bitmask);
5914 /* napi->gro_hash[].list contains packets ordered by age.
5915 * youngest packets at the head of it.
5916 * Complete skbs in reverse order to reduce latencies.
5918 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5920 unsigned long bitmask = napi->gro_bitmask;
5921 unsigned int i, base = ~0U;
5923 while ((i = ffs(bitmask)) != 0) {
5926 __napi_gro_flush_chain(napi, base, flush_old);
5929 EXPORT_SYMBOL(napi_gro_flush);
5931 static void gro_list_prepare(const struct list_head *head,
5932 const struct sk_buff *skb)
5934 unsigned int maclen = skb->dev->hard_header_len;
5935 u32 hash = skb_get_hash_raw(skb);
5938 list_for_each_entry(p, head, list) {
5939 unsigned long diffs;
5941 NAPI_GRO_CB(p)->flush = 0;
5943 if (hash != skb_get_hash_raw(p)) {
5944 NAPI_GRO_CB(p)->same_flow = 0;
5948 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5949 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5950 if (skb_vlan_tag_present(p))
5951 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5952 diffs |= skb_metadata_dst_cmp(p, skb);
5953 diffs |= skb_metadata_differs(p, skb);
5954 if (maclen == ETH_HLEN)
5955 diffs |= compare_ether_header(skb_mac_header(p),
5956 skb_mac_header(skb));
5958 diffs = memcmp(skb_mac_header(p),
5959 skb_mac_header(skb),
5961 NAPI_GRO_CB(p)->same_flow = !diffs;
5965 static void skb_gro_reset_offset(struct sk_buff *skb)
5967 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5968 const skb_frag_t *frag0 = &pinfo->frags[0];
5970 NAPI_GRO_CB(skb)->data_offset = 0;
5971 NAPI_GRO_CB(skb)->frag0 = NULL;
5972 NAPI_GRO_CB(skb)->frag0_len = 0;
5974 if (!skb_headlen(skb) && pinfo->nr_frags &&
5975 !PageHighMem(skb_frag_page(frag0))) {
5976 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5977 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5978 skb_frag_size(frag0),
5979 skb->end - skb->tail);
5983 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5985 struct skb_shared_info *pinfo = skb_shinfo(skb);
5987 BUG_ON(skb->end - skb->tail < grow);
5989 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5991 skb->data_len -= grow;
5994 skb_frag_off_add(&pinfo->frags[0], grow);
5995 skb_frag_size_sub(&pinfo->frags[0], grow);
5997 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5998 skb_frag_unref(skb, 0);
5999 memmove(pinfo->frags, pinfo->frags + 1,
6000 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
6004 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
6006 struct sk_buff *oldest;
6008 oldest = list_last_entry(head, struct sk_buff, list);
6010 /* We are called with head length >= MAX_GRO_SKBS, so this is
6013 if (WARN_ON_ONCE(!oldest))
6016 /* Do not adjust napi->gro_hash[].count, caller is adding a new
6019 skb_list_del_init(oldest);
6020 napi_gro_complete(napi, oldest);
6023 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6025 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
6026 struct gro_list *gro_list = &napi->gro_hash[bucket];
6027 struct list_head *head = &offload_base;
6028 struct packet_offload *ptype;
6029 __be16 type = skb->protocol;
6030 struct sk_buff *pp = NULL;
6031 enum gro_result ret;
6035 if (netif_elide_gro(skb->dev))
6038 gro_list_prepare(&gro_list->list, skb);
6041 list_for_each_entry_rcu(ptype, head, list) {
6042 if (ptype->type != type || !ptype->callbacks.gro_receive)
6045 skb_set_network_header(skb, skb_gro_offset(skb));
6046 skb_reset_mac_len(skb);
6047 NAPI_GRO_CB(skb)->same_flow = 0;
6048 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
6049 NAPI_GRO_CB(skb)->free = 0;
6050 NAPI_GRO_CB(skb)->encap_mark = 0;
6051 NAPI_GRO_CB(skb)->recursion_counter = 0;
6052 NAPI_GRO_CB(skb)->is_fou = 0;
6053 NAPI_GRO_CB(skb)->is_atomic = 1;
6054 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
6056 /* Setup for GRO checksum validation */
6057 switch (skb->ip_summed) {
6058 case CHECKSUM_COMPLETE:
6059 NAPI_GRO_CB(skb)->csum = skb->csum;
6060 NAPI_GRO_CB(skb)->csum_valid = 1;
6061 NAPI_GRO_CB(skb)->csum_cnt = 0;
6063 case CHECKSUM_UNNECESSARY:
6064 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
6065 NAPI_GRO_CB(skb)->csum_valid = 0;
6068 NAPI_GRO_CB(skb)->csum_cnt = 0;
6069 NAPI_GRO_CB(skb)->csum_valid = 0;
6072 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
6073 ipv6_gro_receive, inet_gro_receive,
6074 &gro_list->list, skb);
6079 if (&ptype->list == head)
6082 if (PTR_ERR(pp) == -EINPROGRESS) {
6087 same_flow = NAPI_GRO_CB(skb)->same_flow;
6088 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
6091 skb_list_del_init(pp);
6092 napi_gro_complete(napi, pp);
6099 if (NAPI_GRO_CB(skb)->flush)
6102 if (unlikely(gro_list->count >= MAX_GRO_SKBS))
6103 gro_flush_oldest(napi, &gro_list->list);
6107 NAPI_GRO_CB(skb)->count = 1;
6108 NAPI_GRO_CB(skb)->age = jiffies;
6109 NAPI_GRO_CB(skb)->last = skb;
6110 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
6111 list_add(&skb->list, &gro_list->list);
6115 grow = skb_gro_offset(skb) - skb_headlen(skb);
6117 gro_pull_from_frag0(skb, grow);
6119 if (gro_list->count) {
6120 if (!test_bit(bucket, &napi->gro_bitmask))
6121 __set_bit(bucket, &napi->gro_bitmask);
6122 } else if (test_bit(bucket, &napi->gro_bitmask)) {
6123 __clear_bit(bucket, &napi->gro_bitmask);
6133 struct packet_offload *gro_find_receive_by_type(__be16 type)
6135 struct list_head *offload_head = &offload_base;
6136 struct packet_offload *ptype;
6138 list_for_each_entry_rcu(ptype, offload_head, list) {
6139 if (ptype->type != type || !ptype->callbacks.gro_receive)
6145 EXPORT_SYMBOL(gro_find_receive_by_type);
6147 struct packet_offload *gro_find_complete_by_type(__be16 type)
6149 struct list_head *offload_head = &offload_base;
6150 struct packet_offload *ptype;
6152 list_for_each_entry_rcu(ptype, offload_head, list) {
6153 if (ptype->type != type || !ptype->callbacks.gro_complete)
6159 EXPORT_SYMBOL(gro_find_complete_by_type);
6161 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6162 struct sk_buff *skb,
6167 gro_normal_one(napi, skb, 1);
6170 case GRO_MERGED_FREE:
6171 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6172 napi_skb_free_stolen_head(skb);
6174 __kfree_skb_defer(skb);
6186 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6190 skb_mark_napi_id(skb, napi);
6191 trace_napi_gro_receive_entry(skb);
6193 skb_gro_reset_offset(skb);
6195 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6196 trace_napi_gro_receive_exit(ret);
6200 EXPORT_SYMBOL(napi_gro_receive);
6202 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6204 if (unlikely(skb->pfmemalloc)) {
6208 __skb_pull(skb, skb_headlen(skb));
6209 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6210 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6211 __vlan_hwaccel_clear_tag(skb);
6212 skb->dev = napi->dev;
6215 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6216 skb->pkt_type = PACKET_HOST;
6218 skb->encapsulation = 0;
6219 skb_shinfo(skb)->gso_type = 0;
6220 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6226 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6228 struct sk_buff *skb = napi->skb;
6231 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6234 skb_mark_napi_id(skb, napi);
6239 EXPORT_SYMBOL(napi_get_frags);
6241 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6242 struct sk_buff *skb,
6248 __skb_push(skb, ETH_HLEN);
6249 skb->protocol = eth_type_trans(skb, skb->dev);
6250 if (ret == GRO_NORMAL)
6251 gro_normal_one(napi, skb, 1);
6254 case GRO_MERGED_FREE:
6255 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6256 napi_skb_free_stolen_head(skb);
6258 napi_reuse_skb(napi, skb);
6269 /* Upper GRO stack assumes network header starts at gro_offset=0
6270 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6271 * We copy ethernet header into skb->data to have a common layout.
6273 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6275 struct sk_buff *skb = napi->skb;
6276 const struct ethhdr *eth;
6277 unsigned int hlen = sizeof(*eth);
6281 skb_reset_mac_header(skb);
6282 skb_gro_reset_offset(skb);
6284 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6285 eth = skb_gro_header_slow(skb, hlen, 0);
6286 if (unlikely(!eth)) {
6287 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6288 __func__, napi->dev->name);
6289 napi_reuse_skb(napi, skb);
6293 eth = (const struct ethhdr *)skb->data;
6294 gro_pull_from_frag0(skb, hlen);
6295 NAPI_GRO_CB(skb)->frag0 += hlen;
6296 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6298 __skb_pull(skb, hlen);
6301 * This works because the only protocols we care about don't require
6303 * We'll fix it up properly in napi_frags_finish()
6305 skb->protocol = eth->h_proto;
6310 gro_result_t napi_gro_frags(struct napi_struct *napi)
6313 struct sk_buff *skb = napi_frags_skb(napi);
6315 trace_napi_gro_frags_entry(skb);
6317 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6318 trace_napi_gro_frags_exit(ret);
6322 EXPORT_SYMBOL(napi_gro_frags);
6324 /* Compute the checksum from gro_offset and return the folded value
6325 * after adding in any pseudo checksum.
6327 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6332 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6334 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6335 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6336 /* See comments in __skb_checksum_complete(). */
6338 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6339 !skb->csum_complete_sw)
6340 netdev_rx_csum_fault(skb->dev, skb);
6343 NAPI_GRO_CB(skb)->csum = wsum;
6344 NAPI_GRO_CB(skb)->csum_valid = 1;
6348 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6350 static void net_rps_send_ipi(struct softnet_data *remsd)
6354 struct softnet_data *next = remsd->rps_ipi_next;
6356 if (cpu_online(remsd->cpu))
6357 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6364 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6365 * Note: called with local irq disabled, but exits with local irq enabled.
6367 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6370 struct softnet_data *remsd = sd->rps_ipi_list;
6373 sd->rps_ipi_list = NULL;
6377 /* Send pending IPI's to kick RPS processing on remote cpus. */
6378 net_rps_send_ipi(remsd);
6384 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6387 return sd->rps_ipi_list != NULL;
6393 static int process_backlog(struct napi_struct *napi, int quota)
6395 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6399 /* Check if we have pending ipi, its better to send them now,
6400 * not waiting net_rx_action() end.
6402 if (sd_has_rps_ipi_waiting(sd)) {
6403 local_irq_disable();
6404 net_rps_action_and_irq_enable(sd);
6407 napi->weight = dev_rx_weight;
6409 struct sk_buff *skb;
6411 while ((skb = __skb_dequeue(&sd->process_queue))) {
6413 __netif_receive_skb(skb);
6415 input_queue_head_incr(sd);
6416 if (++work >= quota)
6421 local_irq_disable();
6423 if (skb_queue_empty(&sd->input_pkt_queue)) {
6425 * Inline a custom version of __napi_complete().
6426 * only current cpu owns and manipulates this napi,
6427 * and NAPI_STATE_SCHED is the only possible flag set
6429 * We can use a plain write instead of clear_bit(),
6430 * and we dont need an smp_mb() memory barrier.
6435 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6436 &sd->process_queue);
6446 * __napi_schedule - schedule for receive
6447 * @n: entry to schedule
6449 * The entry's receive function will be scheduled to run.
6450 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6452 void __napi_schedule(struct napi_struct *n)
6454 unsigned long flags;
6456 local_irq_save(flags);
6457 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6458 local_irq_restore(flags);
6460 EXPORT_SYMBOL(__napi_schedule);
6463 * napi_schedule_prep - check if napi can be scheduled
6466 * Test if NAPI routine is already running, and if not mark
6467 * it as running. This is used as a condition variable to
6468 * insure only one NAPI poll instance runs. We also make
6469 * sure there is no pending NAPI disable.
6471 bool napi_schedule_prep(struct napi_struct *n)
6473 unsigned long val, new;
6476 val = READ_ONCE(n->state);
6477 if (unlikely(val & NAPIF_STATE_DISABLE))
6479 new = val | NAPIF_STATE_SCHED;
6481 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6482 * This was suggested by Alexander Duyck, as compiler
6483 * emits better code than :
6484 * if (val & NAPIF_STATE_SCHED)
6485 * new |= NAPIF_STATE_MISSED;
6487 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6489 } while (cmpxchg(&n->state, val, new) != val);
6491 return !(val & NAPIF_STATE_SCHED);
6493 EXPORT_SYMBOL(napi_schedule_prep);
6496 * __napi_schedule_irqoff - schedule for receive
6497 * @n: entry to schedule
6499 * Variant of __napi_schedule() assuming hard irqs are masked
6501 void __napi_schedule_irqoff(struct napi_struct *n)
6503 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6505 EXPORT_SYMBOL(__napi_schedule_irqoff);
6507 bool napi_complete_done(struct napi_struct *n, int work_done)
6509 unsigned long flags, val, new, timeout = 0;
6513 * 1) Don't let napi dequeue from the cpu poll list
6514 * just in case its running on a different cpu.
6515 * 2) If we are busy polling, do nothing here, we have
6516 * the guarantee we will be called later.
6518 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6519 NAPIF_STATE_IN_BUSY_POLL)))
6524 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6525 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6527 if (n->defer_hard_irqs_count > 0) {
6528 n->defer_hard_irqs_count--;
6529 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6533 if (n->gro_bitmask) {
6534 /* When the NAPI instance uses a timeout and keeps postponing
6535 * it, we need to bound somehow the time packets are kept in
6538 napi_gro_flush(n, !!timeout);
6543 if (unlikely(!list_empty(&n->poll_list))) {
6544 /* If n->poll_list is not empty, we need to mask irqs */
6545 local_irq_save(flags);
6546 list_del_init(&n->poll_list);
6547 local_irq_restore(flags);
6551 val = READ_ONCE(n->state);
6553 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6555 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED |
6556 NAPIF_STATE_SCHED_THREADED |
6557 NAPIF_STATE_PREFER_BUSY_POLL);
6559 /* If STATE_MISSED was set, leave STATE_SCHED set,
6560 * because we will call napi->poll() one more time.
6561 * This C code was suggested by Alexander Duyck to help gcc.
6563 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6565 } while (cmpxchg(&n->state, val, new) != val);
6567 if (unlikely(val & NAPIF_STATE_MISSED)) {
6573 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6574 HRTIMER_MODE_REL_PINNED);
6577 EXPORT_SYMBOL(napi_complete_done);
6579 /* must be called under rcu_read_lock(), as we dont take a reference */
6580 static struct napi_struct *napi_by_id(unsigned int napi_id)
6582 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6583 struct napi_struct *napi;
6585 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6586 if (napi->napi_id == napi_id)
6592 #if defined(CONFIG_NET_RX_BUSY_POLL)
6594 static void __busy_poll_stop(struct napi_struct *napi, bool skip_schedule)
6596 if (!skip_schedule) {
6597 gro_normal_list(napi);
6598 __napi_schedule(napi);
6602 if (napi->gro_bitmask) {
6603 /* flush too old packets
6604 * If HZ < 1000, flush all packets.
6606 napi_gro_flush(napi, HZ >= 1000);
6609 gro_normal_list(napi);
6610 clear_bit(NAPI_STATE_SCHED, &napi->state);
6613 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock, bool prefer_busy_poll,
6616 bool skip_schedule = false;
6617 unsigned long timeout;
6620 /* Busy polling means there is a high chance device driver hard irq
6621 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6622 * set in napi_schedule_prep().
6623 * Since we are about to call napi->poll() once more, we can safely
6624 * clear NAPI_STATE_MISSED.
6626 * Note: x86 could use a single "lock and ..." instruction
6627 * to perform these two clear_bit()
6629 clear_bit(NAPI_STATE_MISSED, &napi->state);
6630 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6634 if (prefer_busy_poll) {
6635 napi->defer_hard_irqs_count = READ_ONCE(napi->dev->napi_defer_hard_irqs);
6636 timeout = READ_ONCE(napi->dev->gro_flush_timeout);
6637 if (napi->defer_hard_irqs_count && timeout) {
6638 hrtimer_start(&napi->timer, ns_to_ktime(timeout), HRTIMER_MODE_REL_PINNED);
6639 skip_schedule = true;
6643 /* All we really want here is to re-enable device interrupts.
6644 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6646 rc = napi->poll(napi, budget);
6647 /* We can't gro_normal_list() here, because napi->poll() might have
6648 * rearmed the napi (napi_complete_done()) in which case it could
6649 * already be running on another CPU.
6651 trace_napi_poll(napi, rc, budget);
6652 netpoll_poll_unlock(have_poll_lock);
6654 __busy_poll_stop(napi, skip_schedule);
6658 void napi_busy_loop(unsigned int napi_id,
6659 bool (*loop_end)(void *, unsigned long),
6660 void *loop_end_arg, bool prefer_busy_poll, u16 budget)
6662 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6663 int (*napi_poll)(struct napi_struct *napi, int budget);
6664 void *have_poll_lock = NULL;
6665 struct napi_struct *napi;
6672 napi = napi_by_id(napi_id);
6682 unsigned long val = READ_ONCE(napi->state);
6684 /* If multiple threads are competing for this napi,
6685 * we avoid dirtying napi->state as much as we can.
6687 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6688 NAPIF_STATE_IN_BUSY_POLL)) {
6689 if (prefer_busy_poll)
6690 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6693 if (cmpxchg(&napi->state, val,
6694 val | NAPIF_STATE_IN_BUSY_POLL |
6695 NAPIF_STATE_SCHED) != val) {
6696 if (prefer_busy_poll)
6697 set_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6700 have_poll_lock = netpoll_poll_lock(napi);
6701 napi_poll = napi->poll;
6703 work = napi_poll(napi, budget);
6704 trace_napi_poll(napi, work, budget);
6705 gro_normal_list(napi);
6708 __NET_ADD_STATS(dev_net(napi->dev),
6709 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6712 if (!loop_end || loop_end(loop_end_arg, start_time))
6715 if (unlikely(need_resched())) {
6717 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6721 if (loop_end(loop_end_arg, start_time))
6728 busy_poll_stop(napi, have_poll_lock, prefer_busy_poll, budget);
6733 EXPORT_SYMBOL(napi_busy_loop);
6735 #endif /* CONFIG_NET_RX_BUSY_POLL */
6737 static void napi_hash_add(struct napi_struct *napi)
6739 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6742 spin_lock(&napi_hash_lock);
6744 /* 0..NR_CPUS range is reserved for sender_cpu use */
6746 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6747 napi_gen_id = MIN_NAPI_ID;
6748 } while (napi_by_id(napi_gen_id));
6749 napi->napi_id = napi_gen_id;
6751 hlist_add_head_rcu(&napi->napi_hash_node,
6752 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6754 spin_unlock(&napi_hash_lock);
6757 /* Warning : caller is responsible to make sure rcu grace period
6758 * is respected before freeing memory containing @napi
6760 static void napi_hash_del(struct napi_struct *napi)
6762 spin_lock(&napi_hash_lock);
6764 hlist_del_init_rcu(&napi->napi_hash_node);
6766 spin_unlock(&napi_hash_lock);
6769 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6771 struct napi_struct *napi;
6773 napi = container_of(timer, struct napi_struct, timer);
6775 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6776 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6778 if (!napi_disable_pending(napi) &&
6779 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) {
6780 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &napi->state);
6781 __napi_schedule_irqoff(napi);
6784 return HRTIMER_NORESTART;
6787 static void init_gro_hash(struct napi_struct *napi)
6791 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6792 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6793 napi->gro_hash[i].count = 0;
6795 napi->gro_bitmask = 0;
6798 int dev_set_threaded(struct net_device *dev, bool threaded)
6800 struct napi_struct *napi;
6803 if (dev->threaded == threaded)
6807 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6808 if (!napi->thread) {
6809 err = napi_kthread_create(napi);
6818 dev->threaded = threaded;
6820 /* Make sure kthread is created before THREADED bit
6823 smp_mb__before_atomic();
6825 /* Setting/unsetting threaded mode on a napi might not immediately
6826 * take effect, if the current napi instance is actively being
6827 * polled. In this case, the switch between threaded mode and
6828 * softirq mode will happen in the next round of napi_schedule().
6829 * This should not cause hiccups/stalls to the live traffic.
6831 list_for_each_entry(napi, &dev->napi_list, dev_list) {
6833 set_bit(NAPI_STATE_THREADED, &napi->state);
6835 clear_bit(NAPI_STATE_THREADED, &napi->state);
6840 EXPORT_SYMBOL(dev_set_threaded);
6842 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6843 int (*poll)(struct napi_struct *, int), int weight)
6845 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6848 INIT_LIST_HEAD(&napi->poll_list);
6849 INIT_HLIST_NODE(&napi->napi_hash_node);
6850 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6851 napi->timer.function = napi_watchdog;
6852 init_gro_hash(napi);
6854 INIT_LIST_HEAD(&napi->rx_list);
6857 if (weight > NAPI_POLL_WEIGHT)
6858 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6860 napi->weight = weight;
6862 #ifdef CONFIG_NETPOLL
6863 napi->poll_owner = -1;
6865 set_bit(NAPI_STATE_SCHED, &napi->state);
6866 set_bit(NAPI_STATE_NPSVC, &napi->state);
6867 list_add_rcu(&napi->dev_list, &dev->napi_list);
6868 napi_hash_add(napi);
6869 /* Create kthread for this napi if dev->threaded is set.
6870 * Clear dev->threaded if kthread creation failed so that
6871 * threaded mode will not be enabled in napi_enable().
6873 if (dev->threaded && napi_kthread_create(napi))
6876 EXPORT_SYMBOL(netif_napi_add);
6878 void napi_disable(struct napi_struct *n)
6881 set_bit(NAPI_STATE_DISABLE, &n->state);
6883 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6885 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6888 hrtimer_cancel(&n->timer);
6890 clear_bit(NAPI_STATE_PREFER_BUSY_POLL, &n->state);
6891 clear_bit(NAPI_STATE_DISABLE, &n->state);
6892 clear_bit(NAPI_STATE_THREADED, &n->state);
6894 EXPORT_SYMBOL(napi_disable);
6897 * napi_enable - enable NAPI scheduling
6900 * Resume NAPI from being scheduled on this context.
6901 * Must be paired with napi_disable.
6903 void napi_enable(struct napi_struct *n)
6905 BUG_ON(!test_bit(NAPI_STATE_SCHED, &n->state));
6906 smp_mb__before_atomic();
6907 clear_bit(NAPI_STATE_SCHED, &n->state);
6908 clear_bit(NAPI_STATE_NPSVC, &n->state);
6909 if (n->dev->threaded && n->thread)
6910 set_bit(NAPI_STATE_THREADED, &n->state);
6912 EXPORT_SYMBOL(napi_enable);
6914 static void flush_gro_hash(struct napi_struct *napi)
6918 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6919 struct sk_buff *skb, *n;
6921 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6923 napi->gro_hash[i].count = 0;
6927 /* Must be called in process context */
6928 void __netif_napi_del(struct napi_struct *napi)
6930 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6933 napi_hash_del(napi);
6934 list_del_rcu(&napi->dev_list);
6935 napi_free_frags(napi);
6937 flush_gro_hash(napi);
6938 napi->gro_bitmask = 0;
6941 kthread_stop(napi->thread);
6942 napi->thread = NULL;
6945 EXPORT_SYMBOL(__netif_napi_del);
6947 static int __napi_poll(struct napi_struct *n, bool *repoll)
6953 /* This NAPI_STATE_SCHED test is for avoiding a race
6954 * with netpoll's poll_napi(). Only the entity which
6955 * obtains the lock and sees NAPI_STATE_SCHED set will
6956 * actually make the ->poll() call. Therefore we avoid
6957 * accidentally calling ->poll() when NAPI is not scheduled.
6960 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6961 work = n->poll(n, weight);
6962 trace_napi_poll(n, work, weight);
6965 if (unlikely(work > weight))
6966 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6967 n->poll, work, weight);
6969 if (likely(work < weight))
6972 /* Drivers must not modify the NAPI state if they
6973 * consume the entire weight. In such cases this code
6974 * still "owns" the NAPI instance and therefore can
6975 * move the instance around on the list at-will.
6977 if (unlikely(napi_disable_pending(n))) {
6982 /* The NAPI context has more processing work, but busy-polling
6983 * is preferred. Exit early.
6985 if (napi_prefer_busy_poll(n)) {
6986 if (napi_complete_done(n, work)) {
6987 /* If timeout is not set, we need to make sure
6988 * that the NAPI is re-scheduled.
6995 if (n->gro_bitmask) {
6996 /* flush too old packets
6997 * If HZ < 1000, flush all packets.
6999 napi_gro_flush(n, HZ >= 1000);
7004 /* Some drivers may have called napi_schedule
7005 * prior to exhausting their budget.
7007 if (unlikely(!list_empty(&n->poll_list))) {
7008 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
7009 n->dev ? n->dev->name : "backlog");
7018 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
7020 bool do_repoll = false;
7024 list_del_init(&n->poll_list);
7026 have = netpoll_poll_lock(n);
7028 work = __napi_poll(n, &do_repoll);
7031 list_add_tail(&n->poll_list, repoll);
7033 netpoll_poll_unlock(have);
7038 static int napi_thread_wait(struct napi_struct *napi)
7042 set_current_state(TASK_INTERRUPTIBLE);
7044 while (!kthread_should_stop()) {
7045 /* Testing SCHED_THREADED bit here to make sure the current
7046 * kthread owns this napi and could poll on this napi.
7047 * Testing SCHED bit is not enough because SCHED bit might be
7048 * set by some other busy poll thread or by napi_disable().
7050 if (test_bit(NAPI_STATE_SCHED_THREADED, &napi->state) || woken) {
7051 WARN_ON(!list_empty(&napi->poll_list));
7052 __set_current_state(TASK_RUNNING);
7057 /* woken being true indicates this thread owns this napi. */
7059 set_current_state(TASK_INTERRUPTIBLE);
7061 __set_current_state(TASK_RUNNING);
7066 static int napi_threaded_poll(void *data)
7068 struct napi_struct *napi = data;
7071 while (!napi_thread_wait(napi)) {
7073 bool repoll = false;
7077 have = netpoll_poll_lock(napi);
7078 __napi_poll(napi, &repoll);
7079 netpoll_poll_unlock(have);
7092 static __latent_entropy void net_rx_action(struct softirq_action *h)
7094 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
7095 unsigned long time_limit = jiffies +
7096 usecs_to_jiffies(netdev_budget_usecs);
7097 int budget = netdev_budget;
7101 local_irq_disable();
7102 list_splice_init(&sd->poll_list, &list);
7106 struct napi_struct *n;
7108 if (list_empty(&list)) {
7109 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
7114 n = list_first_entry(&list, struct napi_struct, poll_list);
7115 budget -= napi_poll(n, &repoll);
7117 /* If softirq window is exhausted then punt.
7118 * Allow this to run for 2 jiffies since which will allow
7119 * an average latency of 1.5/HZ.
7121 if (unlikely(budget <= 0 ||
7122 time_after_eq(jiffies, time_limit))) {
7128 local_irq_disable();
7130 list_splice_tail_init(&sd->poll_list, &list);
7131 list_splice_tail(&repoll, &list);
7132 list_splice(&list, &sd->poll_list);
7133 if (!list_empty(&sd->poll_list))
7134 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
7136 net_rps_action_and_irq_enable(sd);
7139 struct netdev_adjacent {
7140 struct net_device *dev;
7142 /* upper master flag, there can only be one master device per list */
7145 /* lookup ignore flag */
7148 /* counter for the number of times this device was added to us */
7151 /* private field for the users */
7154 struct list_head list;
7155 struct rcu_head rcu;
7158 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
7159 struct list_head *adj_list)
7161 struct netdev_adjacent *adj;
7163 list_for_each_entry(adj, adj_list, list) {
7164 if (adj->dev == adj_dev)
7170 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
7171 struct netdev_nested_priv *priv)
7173 struct net_device *dev = (struct net_device *)priv->data;
7175 return upper_dev == dev;
7179 * netdev_has_upper_dev - Check if device is linked to an upper device
7181 * @upper_dev: upper device to check
7183 * Find out if a device is linked to specified upper device and return true
7184 * in case it is. Note that this checks only immediate upper device,
7185 * not through a complete stack of devices. The caller must hold the RTNL lock.
7187 bool netdev_has_upper_dev(struct net_device *dev,
7188 struct net_device *upper_dev)
7190 struct netdev_nested_priv priv = {
7191 .data = (void *)upper_dev,
7196 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7199 EXPORT_SYMBOL(netdev_has_upper_dev);
7202 * netdev_has_upper_dev_all_rcu - Check if device is linked to an upper device
7204 * @upper_dev: upper device to check
7206 * Find out if a device is linked to specified upper device and return true
7207 * in case it is. Note that this checks the entire upper device chain.
7208 * The caller must hold rcu lock.
7211 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
7212 struct net_device *upper_dev)
7214 struct netdev_nested_priv priv = {
7215 .data = (void *)upper_dev,
7218 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
7221 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
7224 * netdev_has_any_upper_dev - Check if device is linked to some device
7227 * Find out if a device is linked to an upper device and return true in case
7228 * it is. The caller must hold the RTNL lock.
7230 bool netdev_has_any_upper_dev(struct net_device *dev)
7234 return !list_empty(&dev->adj_list.upper);
7236 EXPORT_SYMBOL(netdev_has_any_upper_dev);
7239 * netdev_master_upper_dev_get - Get master upper device
7242 * Find a master upper device and return pointer to it or NULL in case
7243 * it's not there. The caller must hold the RTNL lock.
7245 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
7247 struct netdev_adjacent *upper;
7251 if (list_empty(&dev->adj_list.upper))
7254 upper = list_first_entry(&dev->adj_list.upper,
7255 struct netdev_adjacent, list);
7256 if (likely(upper->master))
7260 EXPORT_SYMBOL(netdev_master_upper_dev_get);
7262 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
7264 struct netdev_adjacent *upper;
7268 if (list_empty(&dev->adj_list.upper))
7271 upper = list_first_entry(&dev->adj_list.upper,
7272 struct netdev_adjacent, list);
7273 if (likely(upper->master) && !upper->ignore)
7279 * netdev_has_any_lower_dev - Check if device is linked to some device
7282 * Find out if a device is linked to a lower device and return true in case
7283 * it is. The caller must hold the RTNL lock.
7285 static bool netdev_has_any_lower_dev(struct net_device *dev)
7289 return !list_empty(&dev->adj_list.lower);
7292 void *netdev_adjacent_get_private(struct list_head *adj_list)
7294 struct netdev_adjacent *adj;
7296 adj = list_entry(adj_list, struct netdev_adjacent, list);
7298 return adj->private;
7300 EXPORT_SYMBOL(netdev_adjacent_get_private);
7303 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7305 * @iter: list_head ** of the current position
7307 * Gets the next device from the dev's upper list, starting from iter
7308 * position. The caller must hold RCU read lock.
7310 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7311 struct list_head **iter)
7313 struct netdev_adjacent *upper;
7315 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7317 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7319 if (&upper->list == &dev->adj_list.upper)
7322 *iter = &upper->list;
7326 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7328 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7329 struct list_head **iter,
7332 struct netdev_adjacent *upper;
7334 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7336 if (&upper->list == &dev->adj_list.upper)
7339 *iter = &upper->list;
7340 *ignore = upper->ignore;
7345 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7346 struct list_head **iter)
7348 struct netdev_adjacent *upper;
7350 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7352 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7354 if (&upper->list == &dev->adj_list.upper)
7357 *iter = &upper->list;
7362 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7363 int (*fn)(struct net_device *dev,
7364 struct netdev_nested_priv *priv),
7365 struct netdev_nested_priv *priv)
7367 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7368 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7373 iter = &dev->adj_list.upper;
7377 ret = fn(now, priv);
7384 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7391 niter = &udev->adj_list.upper;
7392 dev_stack[cur] = now;
7393 iter_stack[cur++] = iter;
7400 next = dev_stack[--cur];
7401 niter = iter_stack[cur];
7411 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7412 int (*fn)(struct net_device *dev,
7413 struct netdev_nested_priv *priv),
7414 struct netdev_nested_priv *priv)
7416 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7417 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7421 iter = &dev->adj_list.upper;
7425 ret = fn(now, priv);
7432 udev = netdev_next_upper_dev_rcu(now, &iter);
7437 niter = &udev->adj_list.upper;
7438 dev_stack[cur] = now;
7439 iter_stack[cur++] = iter;
7446 next = dev_stack[--cur];
7447 niter = iter_stack[cur];
7456 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7458 static bool __netdev_has_upper_dev(struct net_device *dev,
7459 struct net_device *upper_dev)
7461 struct netdev_nested_priv priv = {
7463 .data = (void *)upper_dev,
7468 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7473 * netdev_lower_get_next_private - Get the next ->private from the
7474 * lower neighbour list
7476 * @iter: list_head ** of the current position
7478 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7479 * list, starting from iter position. The caller must hold either hold the
7480 * RTNL lock or its own locking that guarantees that the neighbour lower
7481 * list will remain unchanged.
7483 void *netdev_lower_get_next_private(struct net_device *dev,
7484 struct list_head **iter)
7486 struct netdev_adjacent *lower;
7488 lower = list_entry(*iter, struct netdev_adjacent, list);
7490 if (&lower->list == &dev->adj_list.lower)
7493 *iter = lower->list.next;
7495 return lower->private;
7497 EXPORT_SYMBOL(netdev_lower_get_next_private);
7500 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7501 * lower neighbour list, RCU
7504 * @iter: list_head ** of the current position
7506 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7507 * list, starting from iter position. The caller must hold RCU read lock.
7509 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7510 struct list_head **iter)
7512 struct netdev_adjacent *lower;
7514 WARN_ON_ONCE(!rcu_read_lock_held());
7516 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7518 if (&lower->list == &dev->adj_list.lower)
7521 *iter = &lower->list;
7523 return lower->private;
7525 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7528 * netdev_lower_get_next - Get the next device from the lower neighbour
7531 * @iter: list_head ** of the current position
7533 * Gets the next netdev_adjacent from the dev's lower neighbour
7534 * list, starting from iter position. The caller must hold RTNL lock or
7535 * its own locking that guarantees that the neighbour lower
7536 * list will remain unchanged.
7538 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7540 struct netdev_adjacent *lower;
7542 lower = list_entry(*iter, struct netdev_adjacent, list);
7544 if (&lower->list == &dev->adj_list.lower)
7547 *iter = lower->list.next;
7551 EXPORT_SYMBOL(netdev_lower_get_next);
7553 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7554 struct list_head **iter)
7556 struct netdev_adjacent *lower;
7558 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7560 if (&lower->list == &dev->adj_list.lower)
7563 *iter = &lower->list;
7568 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7569 struct list_head **iter,
7572 struct netdev_adjacent *lower;
7574 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7576 if (&lower->list == &dev->adj_list.lower)
7579 *iter = &lower->list;
7580 *ignore = lower->ignore;
7585 int netdev_walk_all_lower_dev(struct net_device *dev,
7586 int (*fn)(struct net_device *dev,
7587 struct netdev_nested_priv *priv),
7588 struct netdev_nested_priv *priv)
7590 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7591 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7595 iter = &dev->adj_list.lower;
7599 ret = fn(now, priv);
7606 ldev = netdev_next_lower_dev(now, &iter);
7611 niter = &ldev->adj_list.lower;
7612 dev_stack[cur] = now;
7613 iter_stack[cur++] = iter;
7620 next = dev_stack[--cur];
7621 niter = iter_stack[cur];
7630 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7632 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7633 int (*fn)(struct net_device *dev,
7634 struct netdev_nested_priv *priv),
7635 struct netdev_nested_priv *priv)
7637 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7638 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7643 iter = &dev->adj_list.lower;
7647 ret = fn(now, priv);
7654 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7661 niter = &ldev->adj_list.lower;
7662 dev_stack[cur] = now;
7663 iter_stack[cur++] = iter;
7670 next = dev_stack[--cur];
7671 niter = iter_stack[cur];
7681 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7682 struct list_head **iter)
7684 struct netdev_adjacent *lower;
7686 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7687 if (&lower->list == &dev->adj_list.lower)
7690 *iter = &lower->list;
7694 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7696 static u8 __netdev_upper_depth(struct net_device *dev)
7698 struct net_device *udev;
7699 struct list_head *iter;
7703 for (iter = &dev->adj_list.upper,
7704 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7706 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7709 if (max_depth < udev->upper_level)
7710 max_depth = udev->upper_level;
7716 static u8 __netdev_lower_depth(struct net_device *dev)
7718 struct net_device *ldev;
7719 struct list_head *iter;
7723 for (iter = &dev->adj_list.lower,
7724 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7726 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7729 if (max_depth < ldev->lower_level)
7730 max_depth = ldev->lower_level;
7736 static int __netdev_update_upper_level(struct net_device *dev,
7737 struct netdev_nested_priv *__unused)
7739 dev->upper_level = __netdev_upper_depth(dev) + 1;
7743 static int __netdev_update_lower_level(struct net_device *dev,
7744 struct netdev_nested_priv *priv)
7746 dev->lower_level = __netdev_lower_depth(dev) + 1;
7748 #ifdef CONFIG_LOCKDEP
7752 if (priv->flags & NESTED_SYNC_IMM)
7753 dev->nested_level = dev->lower_level - 1;
7754 if (priv->flags & NESTED_SYNC_TODO)
7755 net_unlink_todo(dev);
7760 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7761 int (*fn)(struct net_device *dev,
7762 struct netdev_nested_priv *priv),
7763 struct netdev_nested_priv *priv)
7765 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7766 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7770 iter = &dev->adj_list.lower;
7774 ret = fn(now, priv);
7781 ldev = netdev_next_lower_dev_rcu(now, &iter);
7786 niter = &ldev->adj_list.lower;
7787 dev_stack[cur] = now;
7788 iter_stack[cur++] = iter;
7795 next = dev_stack[--cur];
7796 niter = iter_stack[cur];
7805 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7808 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7809 * lower neighbour list, RCU
7813 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7814 * list. The caller must hold RCU read lock.
7816 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7818 struct netdev_adjacent *lower;
7820 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7821 struct netdev_adjacent, list);
7823 return lower->private;
7826 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7829 * netdev_master_upper_dev_get_rcu - Get master upper device
7832 * Find a master upper device and return pointer to it or NULL in case
7833 * it's not there. The caller must hold the RCU read lock.
7835 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7837 struct netdev_adjacent *upper;
7839 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7840 struct netdev_adjacent, list);
7841 if (upper && likely(upper->master))
7845 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7847 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7848 struct net_device *adj_dev,
7849 struct list_head *dev_list)
7851 char linkname[IFNAMSIZ+7];
7853 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7854 "upper_%s" : "lower_%s", adj_dev->name);
7855 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7858 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7860 struct list_head *dev_list)
7862 char linkname[IFNAMSIZ+7];
7864 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7865 "upper_%s" : "lower_%s", name);
7866 sysfs_remove_link(&(dev->dev.kobj), linkname);
7869 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7870 struct net_device *adj_dev,
7871 struct list_head *dev_list)
7873 return (dev_list == &dev->adj_list.upper ||
7874 dev_list == &dev->adj_list.lower) &&
7875 net_eq(dev_net(dev), dev_net(adj_dev));
7878 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7879 struct net_device *adj_dev,
7880 struct list_head *dev_list,
7881 void *private, bool master)
7883 struct netdev_adjacent *adj;
7886 adj = __netdev_find_adj(adj_dev, dev_list);
7890 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7891 dev->name, adj_dev->name, adj->ref_nr);
7896 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7901 adj->master = master;
7903 adj->private = private;
7904 adj->ignore = false;
7907 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7908 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7910 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7911 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7916 /* Ensure that master link is always the first item in list. */
7918 ret = sysfs_create_link(&(dev->dev.kobj),
7919 &(adj_dev->dev.kobj), "master");
7921 goto remove_symlinks;
7923 list_add_rcu(&adj->list, dev_list);
7925 list_add_tail_rcu(&adj->list, dev_list);
7931 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7932 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7940 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7941 struct net_device *adj_dev,
7943 struct list_head *dev_list)
7945 struct netdev_adjacent *adj;
7947 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7948 dev->name, adj_dev->name, ref_nr);
7950 adj = __netdev_find_adj(adj_dev, dev_list);
7953 pr_err("Adjacency does not exist for device %s from %s\n",
7954 dev->name, adj_dev->name);
7959 if (adj->ref_nr > ref_nr) {
7960 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7961 dev->name, adj_dev->name, ref_nr,
7962 adj->ref_nr - ref_nr);
7963 adj->ref_nr -= ref_nr;
7968 sysfs_remove_link(&(dev->dev.kobj), "master");
7970 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7971 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7973 list_del_rcu(&adj->list);
7974 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7975 adj_dev->name, dev->name, adj_dev->name);
7977 kfree_rcu(adj, rcu);
7980 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7981 struct net_device *upper_dev,
7982 struct list_head *up_list,
7983 struct list_head *down_list,
7984 void *private, bool master)
7988 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7993 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7996 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
8003 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
8004 struct net_device *upper_dev,
8006 struct list_head *up_list,
8007 struct list_head *down_list)
8009 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
8010 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
8013 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
8014 struct net_device *upper_dev,
8015 void *private, bool master)
8017 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
8018 &dev->adj_list.upper,
8019 &upper_dev->adj_list.lower,
8023 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
8024 struct net_device *upper_dev)
8026 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
8027 &dev->adj_list.upper,
8028 &upper_dev->adj_list.lower);
8031 static int __netdev_upper_dev_link(struct net_device *dev,
8032 struct net_device *upper_dev, bool master,
8033 void *upper_priv, void *upper_info,
8034 struct netdev_nested_priv *priv,
8035 struct netlink_ext_ack *extack)
8037 struct netdev_notifier_changeupper_info changeupper_info = {
8042 .upper_dev = upper_dev,
8045 .upper_info = upper_info,
8047 struct net_device *master_dev;
8052 if (dev == upper_dev)
8055 /* To prevent loops, check if dev is not upper device to upper_dev. */
8056 if (__netdev_has_upper_dev(upper_dev, dev))
8059 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
8063 if (__netdev_has_upper_dev(dev, upper_dev))
8066 master_dev = __netdev_master_upper_dev_get(dev);
8068 return master_dev == upper_dev ? -EEXIST : -EBUSY;
8071 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8072 &changeupper_info.info);
8073 ret = notifier_to_errno(ret);
8077 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
8082 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8083 &changeupper_info.info);
8084 ret = notifier_to_errno(ret);
8088 __netdev_update_upper_level(dev, NULL);
8089 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8091 __netdev_update_lower_level(upper_dev, priv);
8092 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8098 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8104 * netdev_upper_dev_link - Add a link to the upper device
8106 * @upper_dev: new upper device
8107 * @extack: netlink extended ack
8109 * Adds a link to device which is upper to this one. The caller must hold
8110 * the RTNL lock. On a failure a negative errno code is returned.
8111 * On success the reference counts are adjusted and the function
8114 int netdev_upper_dev_link(struct net_device *dev,
8115 struct net_device *upper_dev,
8116 struct netlink_ext_ack *extack)
8118 struct netdev_nested_priv priv = {
8119 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8123 return __netdev_upper_dev_link(dev, upper_dev, false,
8124 NULL, NULL, &priv, extack);
8126 EXPORT_SYMBOL(netdev_upper_dev_link);
8129 * netdev_master_upper_dev_link - Add a master link to the upper device
8131 * @upper_dev: new upper device
8132 * @upper_priv: upper device private
8133 * @upper_info: upper info to be passed down via notifier
8134 * @extack: netlink extended ack
8136 * Adds a link to device which is upper to this one. In this case, only
8137 * one master upper device can be linked, although other non-master devices
8138 * might be linked as well. The caller must hold the RTNL lock.
8139 * On a failure a negative errno code is returned. On success the reference
8140 * counts are adjusted and the function returns zero.
8142 int netdev_master_upper_dev_link(struct net_device *dev,
8143 struct net_device *upper_dev,
8144 void *upper_priv, void *upper_info,
8145 struct netlink_ext_ack *extack)
8147 struct netdev_nested_priv priv = {
8148 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8152 return __netdev_upper_dev_link(dev, upper_dev, true,
8153 upper_priv, upper_info, &priv, extack);
8155 EXPORT_SYMBOL(netdev_master_upper_dev_link);
8157 static void __netdev_upper_dev_unlink(struct net_device *dev,
8158 struct net_device *upper_dev,
8159 struct netdev_nested_priv *priv)
8161 struct netdev_notifier_changeupper_info changeupper_info = {
8165 .upper_dev = upper_dev,
8171 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
8173 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
8174 &changeupper_info.info);
8176 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
8178 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
8179 &changeupper_info.info);
8181 __netdev_update_upper_level(dev, NULL);
8182 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
8184 __netdev_update_lower_level(upper_dev, priv);
8185 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
8190 * netdev_upper_dev_unlink - Removes a link to upper device
8192 * @upper_dev: new upper device
8194 * Removes a link to device which is upper to this one. The caller must hold
8197 void netdev_upper_dev_unlink(struct net_device *dev,
8198 struct net_device *upper_dev)
8200 struct netdev_nested_priv priv = {
8201 .flags = NESTED_SYNC_TODO,
8205 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
8207 EXPORT_SYMBOL(netdev_upper_dev_unlink);
8209 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
8210 struct net_device *lower_dev,
8213 struct netdev_adjacent *adj;
8215 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
8219 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
8224 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
8225 struct net_device *lower_dev)
8227 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
8230 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
8231 struct net_device *lower_dev)
8233 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
8236 int netdev_adjacent_change_prepare(struct net_device *old_dev,
8237 struct net_device *new_dev,
8238 struct net_device *dev,
8239 struct netlink_ext_ack *extack)
8241 struct netdev_nested_priv priv = {
8250 if (old_dev && new_dev != old_dev)
8251 netdev_adjacent_dev_disable(dev, old_dev);
8252 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
8255 if (old_dev && new_dev != old_dev)
8256 netdev_adjacent_dev_enable(dev, old_dev);
8262 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
8264 void netdev_adjacent_change_commit(struct net_device *old_dev,
8265 struct net_device *new_dev,
8266 struct net_device *dev)
8268 struct netdev_nested_priv priv = {
8269 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
8273 if (!new_dev || !old_dev)
8276 if (new_dev == old_dev)
8279 netdev_adjacent_dev_enable(dev, old_dev);
8280 __netdev_upper_dev_unlink(old_dev, dev, &priv);
8282 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8284 void netdev_adjacent_change_abort(struct net_device *old_dev,
8285 struct net_device *new_dev,
8286 struct net_device *dev)
8288 struct netdev_nested_priv priv = {
8296 if (old_dev && new_dev != old_dev)
8297 netdev_adjacent_dev_enable(dev, old_dev);
8299 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8301 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8304 * netdev_bonding_info_change - Dispatch event about slave change
8306 * @bonding_info: info to dispatch
8308 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8309 * The caller must hold the RTNL lock.
8311 void netdev_bonding_info_change(struct net_device *dev,
8312 struct netdev_bonding_info *bonding_info)
8314 struct netdev_notifier_bonding_info info = {
8318 memcpy(&info.bonding_info, bonding_info,
8319 sizeof(struct netdev_bonding_info));
8320 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8323 EXPORT_SYMBOL(netdev_bonding_info_change);
8326 * netdev_get_xmit_slave - Get the xmit slave of master device
8329 * @all_slaves: assume all the slaves are active
8331 * The reference counters are not incremented so the caller must be
8332 * careful with locks. The caller must hold RCU lock.
8333 * %NULL is returned if no slave is found.
8336 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8337 struct sk_buff *skb,
8340 const struct net_device_ops *ops = dev->netdev_ops;
8342 if (!ops->ndo_get_xmit_slave)
8344 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8346 EXPORT_SYMBOL(netdev_get_xmit_slave);
8348 static struct net_device *netdev_sk_get_lower_dev(struct net_device *dev,
8351 const struct net_device_ops *ops = dev->netdev_ops;
8353 if (!ops->ndo_sk_get_lower_dev)
8355 return ops->ndo_sk_get_lower_dev(dev, sk);
8359 * netdev_sk_get_lowest_dev - Get the lowest device in chain given device and socket
8363 * %NULL is returned if no lower device is found.
8366 struct net_device *netdev_sk_get_lowest_dev(struct net_device *dev,
8369 struct net_device *lower;
8371 lower = netdev_sk_get_lower_dev(dev, sk);
8374 lower = netdev_sk_get_lower_dev(dev, sk);
8379 EXPORT_SYMBOL(netdev_sk_get_lowest_dev);
8381 static void netdev_adjacent_add_links(struct net_device *dev)
8383 struct netdev_adjacent *iter;
8385 struct net *net = dev_net(dev);
8387 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8388 if (!net_eq(net, dev_net(iter->dev)))
8390 netdev_adjacent_sysfs_add(iter->dev, dev,
8391 &iter->dev->adj_list.lower);
8392 netdev_adjacent_sysfs_add(dev, iter->dev,
8393 &dev->adj_list.upper);
8396 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8397 if (!net_eq(net, dev_net(iter->dev)))
8399 netdev_adjacent_sysfs_add(iter->dev, dev,
8400 &iter->dev->adj_list.upper);
8401 netdev_adjacent_sysfs_add(dev, iter->dev,
8402 &dev->adj_list.lower);
8406 static void netdev_adjacent_del_links(struct net_device *dev)
8408 struct netdev_adjacent *iter;
8410 struct net *net = dev_net(dev);
8412 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8413 if (!net_eq(net, dev_net(iter->dev)))
8415 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8416 &iter->dev->adj_list.lower);
8417 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8418 &dev->adj_list.upper);
8421 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8422 if (!net_eq(net, dev_net(iter->dev)))
8424 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8425 &iter->dev->adj_list.upper);
8426 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8427 &dev->adj_list.lower);
8431 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8433 struct netdev_adjacent *iter;
8435 struct net *net = dev_net(dev);
8437 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8438 if (!net_eq(net, dev_net(iter->dev)))
8440 netdev_adjacent_sysfs_del(iter->dev, oldname,
8441 &iter->dev->adj_list.lower);
8442 netdev_adjacent_sysfs_add(iter->dev, dev,
8443 &iter->dev->adj_list.lower);
8446 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8447 if (!net_eq(net, dev_net(iter->dev)))
8449 netdev_adjacent_sysfs_del(iter->dev, oldname,
8450 &iter->dev->adj_list.upper);
8451 netdev_adjacent_sysfs_add(iter->dev, dev,
8452 &iter->dev->adj_list.upper);
8456 void *netdev_lower_dev_get_private(struct net_device *dev,
8457 struct net_device *lower_dev)
8459 struct netdev_adjacent *lower;
8463 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8467 return lower->private;
8469 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8473 * netdev_lower_state_changed - Dispatch event about lower device state change
8474 * @lower_dev: device
8475 * @lower_state_info: state to dispatch
8477 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8478 * The caller must hold the RTNL lock.
8480 void netdev_lower_state_changed(struct net_device *lower_dev,
8481 void *lower_state_info)
8483 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8484 .info.dev = lower_dev,
8488 changelowerstate_info.lower_state_info = lower_state_info;
8489 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8490 &changelowerstate_info.info);
8492 EXPORT_SYMBOL(netdev_lower_state_changed);
8494 static void dev_change_rx_flags(struct net_device *dev, int flags)
8496 const struct net_device_ops *ops = dev->netdev_ops;
8498 if (ops->ndo_change_rx_flags)
8499 ops->ndo_change_rx_flags(dev, flags);
8502 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8504 unsigned int old_flags = dev->flags;
8510 dev->flags |= IFF_PROMISC;
8511 dev->promiscuity += inc;
8512 if (dev->promiscuity == 0) {
8515 * If inc causes overflow, untouch promisc and return error.
8518 dev->flags &= ~IFF_PROMISC;
8520 dev->promiscuity -= inc;
8521 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8526 if (dev->flags != old_flags) {
8527 pr_info("device %s %s promiscuous mode\n",
8529 dev->flags & IFF_PROMISC ? "entered" : "left");
8530 if (audit_enabled) {
8531 current_uid_gid(&uid, &gid);
8532 audit_log(audit_context(), GFP_ATOMIC,
8533 AUDIT_ANOM_PROMISCUOUS,
8534 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8535 dev->name, (dev->flags & IFF_PROMISC),
8536 (old_flags & IFF_PROMISC),
8537 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8538 from_kuid(&init_user_ns, uid),
8539 from_kgid(&init_user_ns, gid),
8540 audit_get_sessionid(current));
8543 dev_change_rx_flags(dev, IFF_PROMISC);
8546 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8551 * dev_set_promiscuity - update promiscuity count on a device
8555 * Add or remove promiscuity from a device. While the count in the device
8556 * remains above zero the interface remains promiscuous. Once it hits zero
8557 * the device reverts back to normal filtering operation. A negative inc
8558 * value is used to drop promiscuity on the device.
8559 * Return 0 if successful or a negative errno code on error.
8561 int dev_set_promiscuity(struct net_device *dev, int inc)
8563 unsigned int old_flags = dev->flags;
8566 err = __dev_set_promiscuity(dev, inc, true);
8569 if (dev->flags != old_flags)
8570 dev_set_rx_mode(dev);
8573 EXPORT_SYMBOL(dev_set_promiscuity);
8575 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8577 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8581 dev->flags |= IFF_ALLMULTI;
8582 dev->allmulti += inc;
8583 if (dev->allmulti == 0) {
8586 * If inc causes overflow, untouch allmulti and return error.
8589 dev->flags &= ~IFF_ALLMULTI;
8591 dev->allmulti -= inc;
8592 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8597 if (dev->flags ^ old_flags) {
8598 dev_change_rx_flags(dev, IFF_ALLMULTI);
8599 dev_set_rx_mode(dev);
8601 __dev_notify_flags(dev, old_flags,
8602 dev->gflags ^ old_gflags);
8608 * dev_set_allmulti - update allmulti count on a device
8612 * Add or remove reception of all multicast frames to a device. While the
8613 * count in the device remains above zero the interface remains listening
8614 * to all interfaces. Once it hits zero the device reverts back to normal
8615 * filtering operation. A negative @inc value is used to drop the counter
8616 * when releasing a resource needing all multicasts.
8617 * Return 0 if successful or a negative errno code on error.
8620 int dev_set_allmulti(struct net_device *dev, int inc)
8622 return __dev_set_allmulti(dev, inc, true);
8624 EXPORT_SYMBOL(dev_set_allmulti);
8627 * Upload unicast and multicast address lists to device and
8628 * configure RX filtering. When the device doesn't support unicast
8629 * filtering it is put in promiscuous mode while unicast addresses
8632 void __dev_set_rx_mode(struct net_device *dev)
8634 const struct net_device_ops *ops = dev->netdev_ops;
8636 /* dev_open will call this function so the list will stay sane. */
8637 if (!(dev->flags&IFF_UP))
8640 if (!netif_device_present(dev))
8643 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8644 /* Unicast addresses changes may only happen under the rtnl,
8645 * therefore calling __dev_set_promiscuity here is safe.
8647 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8648 __dev_set_promiscuity(dev, 1, false);
8649 dev->uc_promisc = true;
8650 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8651 __dev_set_promiscuity(dev, -1, false);
8652 dev->uc_promisc = false;
8656 if (ops->ndo_set_rx_mode)
8657 ops->ndo_set_rx_mode(dev);
8660 void dev_set_rx_mode(struct net_device *dev)
8662 netif_addr_lock_bh(dev);
8663 __dev_set_rx_mode(dev);
8664 netif_addr_unlock_bh(dev);
8668 * dev_get_flags - get flags reported to userspace
8671 * Get the combination of flag bits exported through APIs to userspace.
8673 unsigned int dev_get_flags(const struct net_device *dev)
8677 flags = (dev->flags & ~(IFF_PROMISC |
8682 (dev->gflags & (IFF_PROMISC |
8685 if (netif_running(dev)) {
8686 if (netif_oper_up(dev))
8687 flags |= IFF_RUNNING;
8688 if (netif_carrier_ok(dev))
8689 flags |= IFF_LOWER_UP;
8690 if (netif_dormant(dev))
8691 flags |= IFF_DORMANT;
8696 EXPORT_SYMBOL(dev_get_flags);
8698 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8699 struct netlink_ext_ack *extack)
8701 unsigned int old_flags = dev->flags;
8707 * Set the flags on our device.
8710 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8711 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8713 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8717 * Load in the correct multicast list now the flags have changed.
8720 if ((old_flags ^ flags) & IFF_MULTICAST)
8721 dev_change_rx_flags(dev, IFF_MULTICAST);
8723 dev_set_rx_mode(dev);
8726 * Have we downed the interface. We handle IFF_UP ourselves
8727 * according to user attempts to set it, rather than blindly
8732 if ((old_flags ^ flags) & IFF_UP) {
8733 if (old_flags & IFF_UP)
8736 ret = __dev_open(dev, extack);
8739 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8740 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8741 unsigned int old_flags = dev->flags;
8743 dev->gflags ^= IFF_PROMISC;
8745 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8746 if (dev->flags != old_flags)
8747 dev_set_rx_mode(dev);
8750 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8751 * is important. Some (broken) drivers set IFF_PROMISC, when
8752 * IFF_ALLMULTI is requested not asking us and not reporting.
8754 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8755 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8757 dev->gflags ^= IFF_ALLMULTI;
8758 __dev_set_allmulti(dev, inc, false);
8764 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8765 unsigned int gchanges)
8767 unsigned int changes = dev->flags ^ old_flags;
8770 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8772 if (changes & IFF_UP) {
8773 if (dev->flags & IFF_UP)
8774 call_netdevice_notifiers(NETDEV_UP, dev);
8776 call_netdevice_notifiers(NETDEV_DOWN, dev);
8779 if (dev->flags & IFF_UP &&
8780 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8781 struct netdev_notifier_change_info change_info = {
8785 .flags_changed = changes,
8788 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8793 * dev_change_flags - change device settings
8795 * @flags: device state flags
8796 * @extack: netlink extended ack
8798 * Change settings on device based state flags. The flags are
8799 * in the userspace exported format.
8801 int dev_change_flags(struct net_device *dev, unsigned int flags,
8802 struct netlink_ext_ack *extack)
8805 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8807 ret = __dev_change_flags(dev, flags, extack);
8811 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8812 __dev_notify_flags(dev, old_flags, changes);
8815 EXPORT_SYMBOL(dev_change_flags);
8817 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8819 const struct net_device_ops *ops = dev->netdev_ops;
8821 if (ops->ndo_change_mtu)
8822 return ops->ndo_change_mtu(dev, new_mtu);
8824 /* Pairs with all the lockless reads of dev->mtu in the stack */
8825 WRITE_ONCE(dev->mtu, new_mtu);
8828 EXPORT_SYMBOL(__dev_set_mtu);
8830 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8831 struct netlink_ext_ack *extack)
8833 /* MTU must be positive, and in range */
8834 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8835 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8839 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8840 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8847 * dev_set_mtu_ext - Change maximum transfer unit
8849 * @new_mtu: new transfer unit
8850 * @extack: netlink extended ack
8852 * Change the maximum transfer size of the network device.
8854 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8855 struct netlink_ext_ack *extack)
8859 if (new_mtu == dev->mtu)
8862 err = dev_validate_mtu(dev, new_mtu, extack);
8866 if (!netif_device_present(dev))
8869 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8870 err = notifier_to_errno(err);
8874 orig_mtu = dev->mtu;
8875 err = __dev_set_mtu(dev, new_mtu);
8878 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8880 err = notifier_to_errno(err);
8882 /* setting mtu back and notifying everyone again,
8883 * so that they have a chance to revert changes.
8885 __dev_set_mtu(dev, orig_mtu);
8886 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8893 int dev_set_mtu(struct net_device *dev, int new_mtu)
8895 struct netlink_ext_ack extack;
8898 memset(&extack, 0, sizeof(extack));
8899 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8900 if (err && extack._msg)
8901 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8904 EXPORT_SYMBOL(dev_set_mtu);
8907 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8909 * @new_len: new tx queue length
8911 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8913 unsigned int orig_len = dev->tx_queue_len;
8916 if (new_len != (unsigned int)new_len)
8919 if (new_len != orig_len) {
8920 dev->tx_queue_len = new_len;
8921 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8922 res = notifier_to_errno(res);
8925 res = dev_qdisc_change_tx_queue_len(dev);
8933 netdev_err(dev, "refused to change device tx_queue_len\n");
8934 dev->tx_queue_len = orig_len;
8939 * dev_set_group - Change group this device belongs to
8941 * @new_group: group this device should belong to
8943 void dev_set_group(struct net_device *dev, int new_group)
8945 dev->group = new_group;
8947 EXPORT_SYMBOL(dev_set_group);
8950 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8952 * @addr: new address
8953 * @extack: netlink extended ack
8955 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8956 struct netlink_ext_ack *extack)
8958 struct netdev_notifier_pre_changeaddr_info info = {
8960 .info.extack = extack,
8965 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8966 return notifier_to_errno(rc);
8968 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8971 * dev_set_mac_address - Change Media Access Control Address
8974 * @extack: netlink extended ack
8976 * Change the hardware (MAC) address of the device
8978 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8979 struct netlink_ext_ack *extack)
8981 const struct net_device_ops *ops = dev->netdev_ops;
8984 if (!ops->ndo_set_mac_address)
8986 if (sa->sa_family != dev->type)
8988 if (!netif_device_present(dev))
8990 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8993 err = ops->ndo_set_mac_address(dev, sa);
8996 dev->addr_assign_type = NET_ADDR_SET;
8997 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8998 add_device_randomness(dev->dev_addr, dev->addr_len);
9001 EXPORT_SYMBOL(dev_set_mac_address);
9003 static DECLARE_RWSEM(dev_addr_sem);
9005 int dev_set_mac_address_user(struct net_device *dev, struct sockaddr *sa,
9006 struct netlink_ext_ack *extack)
9010 down_write(&dev_addr_sem);
9011 ret = dev_set_mac_address(dev, sa, extack);
9012 up_write(&dev_addr_sem);
9015 EXPORT_SYMBOL(dev_set_mac_address_user);
9017 int dev_get_mac_address(struct sockaddr *sa, struct net *net, char *dev_name)
9019 size_t size = sizeof(sa->sa_data);
9020 struct net_device *dev;
9023 down_read(&dev_addr_sem);
9026 dev = dev_get_by_name_rcu(net, dev_name);
9032 memset(sa->sa_data, 0, size);
9034 memcpy(sa->sa_data, dev->dev_addr,
9035 min_t(size_t, size, dev->addr_len));
9036 sa->sa_family = dev->type;
9040 up_read(&dev_addr_sem);
9043 EXPORT_SYMBOL(dev_get_mac_address);
9046 * dev_change_carrier - Change device carrier
9048 * @new_carrier: new value
9050 * Change device carrier
9052 int dev_change_carrier(struct net_device *dev, bool new_carrier)
9054 const struct net_device_ops *ops = dev->netdev_ops;
9056 if (!ops->ndo_change_carrier)
9058 if (!netif_device_present(dev))
9060 return ops->ndo_change_carrier(dev, new_carrier);
9062 EXPORT_SYMBOL(dev_change_carrier);
9065 * dev_get_phys_port_id - Get device physical port ID
9069 * Get device physical port ID
9071 int dev_get_phys_port_id(struct net_device *dev,
9072 struct netdev_phys_item_id *ppid)
9074 const struct net_device_ops *ops = dev->netdev_ops;
9076 if (!ops->ndo_get_phys_port_id)
9078 return ops->ndo_get_phys_port_id(dev, ppid);
9080 EXPORT_SYMBOL(dev_get_phys_port_id);
9083 * dev_get_phys_port_name - Get device physical port name
9086 * @len: limit of bytes to copy to name
9088 * Get device physical port name
9090 int dev_get_phys_port_name(struct net_device *dev,
9091 char *name, size_t len)
9093 const struct net_device_ops *ops = dev->netdev_ops;
9096 if (ops->ndo_get_phys_port_name) {
9097 err = ops->ndo_get_phys_port_name(dev, name, len);
9098 if (err != -EOPNOTSUPP)
9101 return devlink_compat_phys_port_name_get(dev, name, len);
9103 EXPORT_SYMBOL(dev_get_phys_port_name);
9106 * dev_get_port_parent_id - Get the device's port parent identifier
9107 * @dev: network device
9108 * @ppid: pointer to a storage for the port's parent identifier
9109 * @recurse: allow/disallow recursion to lower devices
9111 * Get the devices's port parent identifier
9113 int dev_get_port_parent_id(struct net_device *dev,
9114 struct netdev_phys_item_id *ppid,
9117 const struct net_device_ops *ops = dev->netdev_ops;
9118 struct netdev_phys_item_id first = { };
9119 struct net_device *lower_dev;
9120 struct list_head *iter;
9123 if (ops->ndo_get_port_parent_id) {
9124 err = ops->ndo_get_port_parent_id(dev, ppid);
9125 if (err != -EOPNOTSUPP)
9129 err = devlink_compat_switch_id_get(dev, ppid);
9130 if (!err || err != -EOPNOTSUPP)
9136 netdev_for_each_lower_dev(dev, lower_dev, iter) {
9137 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
9142 else if (memcmp(&first, ppid, sizeof(*ppid)))
9148 EXPORT_SYMBOL(dev_get_port_parent_id);
9151 * netdev_port_same_parent_id - Indicate if two network devices have
9152 * the same port parent identifier
9153 * @a: first network device
9154 * @b: second network device
9156 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
9158 struct netdev_phys_item_id a_id = { };
9159 struct netdev_phys_item_id b_id = { };
9161 if (dev_get_port_parent_id(a, &a_id, true) ||
9162 dev_get_port_parent_id(b, &b_id, true))
9165 return netdev_phys_item_id_same(&a_id, &b_id);
9167 EXPORT_SYMBOL(netdev_port_same_parent_id);
9170 * dev_change_proto_down - update protocol port state information
9172 * @proto_down: new value
9174 * This info can be used by switch drivers to set the phys state of the
9177 int dev_change_proto_down(struct net_device *dev, bool proto_down)
9179 const struct net_device_ops *ops = dev->netdev_ops;
9181 if (!ops->ndo_change_proto_down)
9183 if (!netif_device_present(dev))
9185 return ops->ndo_change_proto_down(dev, proto_down);
9187 EXPORT_SYMBOL(dev_change_proto_down);
9190 * dev_change_proto_down_generic - generic implementation for
9191 * ndo_change_proto_down that sets carrier according to
9195 * @proto_down: new value
9197 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
9200 netif_carrier_off(dev);
9202 netif_carrier_on(dev);
9203 dev->proto_down = proto_down;
9206 EXPORT_SYMBOL(dev_change_proto_down_generic);
9209 * dev_change_proto_down_reason - proto down reason
9212 * @mask: proto down mask
9213 * @value: proto down value
9215 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
9221 dev->proto_down_reason = value;
9223 for_each_set_bit(b, &mask, 32) {
9224 if (value & (1 << b))
9225 dev->proto_down_reason |= BIT(b);
9227 dev->proto_down_reason &= ~BIT(b);
9231 EXPORT_SYMBOL(dev_change_proto_down_reason);
9233 struct bpf_xdp_link {
9234 struct bpf_link link;
9235 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
9239 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
9241 if (flags & XDP_FLAGS_HW_MODE)
9243 if (flags & XDP_FLAGS_DRV_MODE)
9244 return XDP_MODE_DRV;
9245 if (flags & XDP_FLAGS_SKB_MODE)
9246 return XDP_MODE_SKB;
9247 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
9250 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
9254 return generic_xdp_install;
9257 return dev->netdev_ops->ndo_bpf;
9263 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
9264 enum bpf_xdp_mode mode)
9266 return dev->xdp_state[mode].link;
9269 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
9270 enum bpf_xdp_mode mode)
9272 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
9275 return link->link.prog;
9276 return dev->xdp_state[mode].prog;
9279 static u8 dev_xdp_prog_count(struct net_device *dev)
9284 for (i = 0; i < __MAX_XDP_MODE; i++)
9285 if (dev->xdp_state[i].prog || dev->xdp_state[i].link)
9290 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
9292 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
9294 return prog ? prog->aux->id : 0;
9297 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
9298 struct bpf_xdp_link *link)
9300 dev->xdp_state[mode].link = link;
9301 dev->xdp_state[mode].prog = NULL;
9304 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
9305 struct bpf_prog *prog)
9307 dev->xdp_state[mode].link = NULL;
9308 dev->xdp_state[mode].prog = prog;
9311 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
9312 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
9313 u32 flags, struct bpf_prog *prog)
9315 struct netdev_bpf xdp;
9318 memset(&xdp, 0, sizeof(xdp));
9319 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
9320 xdp.extack = extack;
9324 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
9325 * "moved" into driver), so they don't increment it on their own, but
9326 * they do decrement refcnt when program is detached or replaced.
9327 * Given net_device also owns link/prog, we need to bump refcnt here
9328 * to prevent drivers from underflowing it.
9332 err = bpf_op(dev, &xdp);
9339 if (mode != XDP_MODE_HW)
9340 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
9345 static void dev_xdp_uninstall(struct net_device *dev)
9347 struct bpf_xdp_link *link;
9348 struct bpf_prog *prog;
9349 enum bpf_xdp_mode mode;
9354 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
9355 prog = dev_xdp_prog(dev, mode);
9359 bpf_op = dev_xdp_bpf_op(dev, mode);
9363 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9365 /* auto-detach link from net device */
9366 link = dev_xdp_link(dev, mode);
9372 dev_xdp_set_link(dev, mode, NULL);
9376 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9377 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9378 struct bpf_prog *old_prog, u32 flags)
9380 unsigned int num_modes = hweight32(flags & XDP_FLAGS_MODES);
9381 struct bpf_prog *cur_prog;
9382 enum bpf_xdp_mode mode;
9388 /* either link or prog attachment, never both */
9389 if (link && (new_prog || old_prog))
9391 /* link supports only XDP mode flags */
9392 if (link && (flags & ~XDP_FLAGS_MODES)) {
9393 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9396 /* just one XDP mode bit should be set, zero defaults to drv/skb mode */
9397 if (num_modes > 1) {
9398 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9401 /* avoid ambiguity if offload + drv/skb mode progs are both loaded */
9402 if (!num_modes && dev_xdp_prog_count(dev) > 1) {
9403 NL_SET_ERR_MSG(extack,
9404 "More than one program loaded, unset mode is ambiguous");
9407 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9408 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9409 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9413 mode = dev_xdp_mode(dev, flags);
9414 /* can't replace attached link */
9415 if (dev_xdp_link(dev, mode)) {
9416 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9420 cur_prog = dev_xdp_prog(dev, mode);
9421 /* can't replace attached prog with link */
9422 if (link && cur_prog) {
9423 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9426 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9427 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9431 /* put effective new program into new_prog */
9433 new_prog = link->link.prog;
9436 bool offload = mode == XDP_MODE_HW;
9437 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9438 ? XDP_MODE_DRV : XDP_MODE_SKB;
9440 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9441 NL_SET_ERR_MSG(extack, "XDP program already attached");
9444 if (!offload && dev_xdp_prog(dev, other_mode)) {
9445 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9448 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9449 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9452 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9453 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9456 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9457 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9462 /* don't call drivers if the effective program didn't change */
9463 if (new_prog != cur_prog) {
9464 bpf_op = dev_xdp_bpf_op(dev, mode);
9466 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9470 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9476 dev_xdp_set_link(dev, mode, link);
9478 dev_xdp_set_prog(dev, mode, new_prog);
9480 bpf_prog_put(cur_prog);
9485 static int dev_xdp_attach_link(struct net_device *dev,
9486 struct netlink_ext_ack *extack,
9487 struct bpf_xdp_link *link)
9489 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9492 static int dev_xdp_detach_link(struct net_device *dev,
9493 struct netlink_ext_ack *extack,
9494 struct bpf_xdp_link *link)
9496 enum bpf_xdp_mode mode;
9501 mode = dev_xdp_mode(dev, link->flags);
9502 if (dev_xdp_link(dev, mode) != link)
9505 bpf_op = dev_xdp_bpf_op(dev, mode);
9506 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9507 dev_xdp_set_link(dev, mode, NULL);
9511 static void bpf_xdp_link_release(struct bpf_link *link)
9513 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9517 /* if racing with net_device's tear down, xdp_link->dev might be
9518 * already NULL, in which case link was already auto-detached
9520 if (xdp_link->dev) {
9521 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9522 xdp_link->dev = NULL;
9528 static int bpf_xdp_link_detach(struct bpf_link *link)
9530 bpf_xdp_link_release(link);
9534 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9536 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9541 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9542 struct seq_file *seq)
9544 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9549 ifindex = xdp_link->dev->ifindex;
9552 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9555 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9556 struct bpf_link_info *info)
9558 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9563 ifindex = xdp_link->dev->ifindex;
9566 info->xdp.ifindex = ifindex;
9570 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9571 struct bpf_prog *old_prog)
9573 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9574 enum bpf_xdp_mode mode;
9580 /* link might have been auto-released already, so fail */
9581 if (!xdp_link->dev) {
9586 if (old_prog && link->prog != old_prog) {
9590 old_prog = link->prog;
9591 if (old_prog == new_prog) {
9592 /* no-op, don't disturb drivers */
9593 bpf_prog_put(new_prog);
9597 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9598 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9599 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9600 xdp_link->flags, new_prog);
9604 old_prog = xchg(&link->prog, new_prog);
9605 bpf_prog_put(old_prog);
9612 static const struct bpf_link_ops bpf_xdp_link_lops = {
9613 .release = bpf_xdp_link_release,
9614 .dealloc = bpf_xdp_link_dealloc,
9615 .detach = bpf_xdp_link_detach,
9616 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9617 .fill_link_info = bpf_xdp_link_fill_link_info,
9618 .update_prog = bpf_xdp_link_update,
9621 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9623 struct net *net = current->nsproxy->net_ns;
9624 struct bpf_link_primer link_primer;
9625 struct bpf_xdp_link *link;
9626 struct net_device *dev;
9629 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9633 link = kzalloc(sizeof(*link), GFP_USER);
9639 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9641 link->flags = attr->link_create.flags;
9643 err = bpf_link_prime(&link->link, &link_primer);
9650 err = dev_xdp_attach_link(dev, NULL, link);
9654 bpf_link_cleanup(&link_primer);
9658 fd = bpf_link_settle(&link_primer);
9659 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9669 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9671 * @extack: netlink extended ack
9672 * @fd: new program fd or negative value to clear
9673 * @expected_fd: old program fd that userspace expects to replace or clear
9674 * @flags: xdp-related flags
9676 * Set or clear a bpf program for a device
9678 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9679 int fd, int expected_fd, u32 flags)
9681 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9682 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9688 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9689 mode != XDP_MODE_SKB);
9690 if (IS_ERR(new_prog))
9691 return PTR_ERR(new_prog);
9694 if (expected_fd >= 0) {
9695 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9696 mode != XDP_MODE_SKB);
9697 if (IS_ERR(old_prog)) {
9698 err = PTR_ERR(old_prog);
9704 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9707 if (err && new_prog)
9708 bpf_prog_put(new_prog);
9710 bpf_prog_put(old_prog);
9715 * dev_new_index - allocate an ifindex
9716 * @net: the applicable net namespace
9718 * Returns a suitable unique value for a new device interface
9719 * number. The caller must hold the rtnl semaphore or the
9720 * dev_base_lock to be sure it remains unique.
9722 static int dev_new_index(struct net *net)
9724 int ifindex = net->ifindex;
9729 if (!__dev_get_by_index(net, ifindex))
9730 return net->ifindex = ifindex;
9734 /* Delayed registration/unregisteration */
9735 static LIST_HEAD(net_todo_list);
9736 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9738 static void net_set_todo(struct net_device *dev)
9740 list_add_tail(&dev->todo_list, &net_todo_list);
9741 dev_net(dev)->dev_unreg_count++;
9744 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9745 struct net_device *upper, netdev_features_t features)
9747 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9748 netdev_features_t feature;
9751 for_each_netdev_feature(upper_disables, feature_bit) {
9752 feature = __NETIF_F_BIT(feature_bit);
9753 if (!(upper->wanted_features & feature)
9754 && (features & feature)) {
9755 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9756 &feature, upper->name);
9757 features &= ~feature;
9764 static void netdev_sync_lower_features(struct net_device *upper,
9765 struct net_device *lower, netdev_features_t features)
9767 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9768 netdev_features_t feature;
9771 for_each_netdev_feature(upper_disables, feature_bit) {
9772 feature = __NETIF_F_BIT(feature_bit);
9773 if (!(features & feature) && (lower->features & feature)) {
9774 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9775 &feature, lower->name);
9776 lower->wanted_features &= ~feature;
9777 __netdev_update_features(lower);
9779 if (unlikely(lower->features & feature))
9780 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9781 &feature, lower->name);
9783 netdev_features_change(lower);
9788 static netdev_features_t netdev_fix_features(struct net_device *dev,
9789 netdev_features_t features)
9791 /* Fix illegal checksum combinations */
9792 if ((features & NETIF_F_HW_CSUM) &&
9793 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9794 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9795 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9798 /* TSO requires that SG is present as well. */
9799 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9800 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9801 features &= ~NETIF_F_ALL_TSO;
9804 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9805 !(features & NETIF_F_IP_CSUM)) {
9806 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9807 features &= ~NETIF_F_TSO;
9808 features &= ~NETIF_F_TSO_ECN;
9811 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9812 !(features & NETIF_F_IPV6_CSUM)) {
9813 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9814 features &= ~NETIF_F_TSO6;
9817 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9818 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9819 features &= ~NETIF_F_TSO_MANGLEID;
9821 /* TSO ECN requires that TSO is present as well. */
9822 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9823 features &= ~NETIF_F_TSO_ECN;
9825 /* Software GSO depends on SG. */
9826 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9827 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9828 features &= ~NETIF_F_GSO;
9831 /* GSO partial features require GSO partial be set */
9832 if ((features & dev->gso_partial_features) &&
9833 !(features & NETIF_F_GSO_PARTIAL)) {
9835 "Dropping partially supported GSO features since no GSO partial.\n");
9836 features &= ~dev->gso_partial_features;
9839 if (!(features & NETIF_F_RXCSUM)) {
9840 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9841 * successfully merged by hardware must also have the
9842 * checksum verified by hardware. If the user does not
9843 * want to enable RXCSUM, logically, we should disable GRO_HW.
9845 if (features & NETIF_F_GRO_HW) {
9846 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9847 features &= ~NETIF_F_GRO_HW;
9851 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9852 if (features & NETIF_F_RXFCS) {
9853 if (features & NETIF_F_LRO) {
9854 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9855 features &= ~NETIF_F_LRO;
9858 if (features & NETIF_F_GRO_HW) {
9859 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9860 features &= ~NETIF_F_GRO_HW;
9864 if (features & NETIF_F_HW_TLS_TX) {
9865 bool ip_csum = (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) ==
9866 (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM);
9867 bool hw_csum = features & NETIF_F_HW_CSUM;
9869 if (!ip_csum && !hw_csum) {
9870 netdev_dbg(dev, "Dropping TLS TX HW offload feature since no CSUM feature.\n");
9871 features &= ~NETIF_F_HW_TLS_TX;
9875 if ((features & NETIF_F_HW_TLS_RX) && !(features & NETIF_F_RXCSUM)) {
9876 netdev_dbg(dev, "Dropping TLS RX HW offload feature since no RXCSUM feature.\n");
9877 features &= ~NETIF_F_HW_TLS_RX;
9883 int __netdev_update_features(struct net_device *dev)
9885 struct net_device *upper, *lower;
9886 netdev_features_t features;
9887 struct list_head *iter;
9892 features = netdev_get_wanted_features(dev);
9894 if (dev->netdev_ops->ndo_fix_features)
9895 features = dev->netdev_ops->ndo_fix_features(dev, features);
9897 /* driver might be less strict about feature dependencies */
9898 features = netdev_fix_features(dev, features);
9900 /* some features can't be enabled if they're off on an upper device */
9901 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9902 features = netdev_sync_upper_features(dev, upper, features);
9904 if (dev->features == features)
9907 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9908 &dev->features, &features);
9910 if (dev->netdev_ops->ndo_set_features)
9911 err = dev->netdev_ops->ndo_set_features(dev, features);
9915 if (unlikely(err < 0)) {
9917 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9918 err, &features, &dev->features);
9919 /* return non-0 since some features might have changed and
9920 * it's better to fire a spurious notification than miss it
9926 /* some features must be disabled on lower devices when disabled
9927 * on an upper device (think: bonding master or bridge)
9929 netdev_for_each_lower_dev(dev, lower, iter)
9930 netdev_sync_lower_features(dev, lower, features);
9933 netdev_features_t diff = features ^ dev->features;
9935 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9936 /* udp_tunnel_{get,drop}_rx_info both need
9937 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9938 * device, or they won't do anything.
9939 * Thus we need to update dev->features
9940 * *before* calling udp_tunnel_get_rx_info,
9941 * but *after* calling udp_tunnel_drop_rx_info.
9943 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9944 dev->features = features;
9945 udp_tunnel_get_rx_info(dev);
9947 udp_tunnel_drop_rx_info(dev);
9951 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9952 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9953 dev->features = features;
9954 err |= vlan_get_rx_ctag_filter_info(dev);
9956 vlan_drop_rx_ctag_filter_info(dev);
9960 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9961 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9962 dev->features = features;
9963 err |= vlan_get_rx_stag_filter_info(dev);
9965 vlan_drop_rx_stag_filter_info(dev);
9969 dev->features = features;
9972 return err < 0 ? 0 : 1;
9976 * netdev_update_features - recalculate device features
9977 * @dev: the device to check
9979 * Recalculate dev->features set and send notifications if it
9980 * has changed. Should be called after driver or hardware dependent
9981 * conditions might have changed that influence the features.
9983 void netdev_update_features(struct net_device *dev)
9985 if (__netdev_update_features(dev))
9986 netdev_features_change(dev);
9988 EXPORT_SYMBOL(netdev_update_features);
9991 * netdev_change_features - recalculate device features
9992 * @dev: the device to check
9994 * Recalculate dev->features set and send notifications even
9995 * if they have not changed. Should be called instead of
9996 * netdev_update_features() if also dev->vlan_features might
9997 * have changed to allow the changes to be propagated to stacked
10000 void netdev_change_features(struct net_device *dev)
10002 __netdev_update_features(dev);
10003 netdev_features_change(dev);
10005 EXPORT_SYMBOL(netdev_change_features);
10008 * netif_stacked_transfer_operstate - transfer operstate
10009 * @rootdev: the root or lower level device to transfer state from
10010 * @dev: the device to transfer operstate to
10012 * Transfer operational state from root to device. This is normally
10013 * called when a stacking relationship exists between the root
10014 * device and the device(a leaf device).
10016 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
10017 struct net_device *dev)
10019 if (rootdev->operstate == IF_OPER_DORMANT)
10020 netif_dormant_on(dev);
10022 netif_dormant_off(dev);
10024 if (rootdev->operstate == IF_OPER_TESTING)
10025 netif_testing_on(dev);
10027 netif_testing_off(dev);
10029 if (netif_carrier_ok(rootdev))
10030 netif_carrier_on(dev);
10032 netif_carrier_off(dev);
10034 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
10036 static int netif_alloc_rx_queues(struct net_device *dev)
10038 unsigned int i, count = dev->num_rx_queues;
10039 struct netdev_rx_queue *rx;
10040 size_t sz = count * sizeof(*rx);
10045 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
10051 for (i = 0; i < count; i++) {
10054 /* XDP RX-queue setup */
10055 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i, 0);
10062 /* Rollback successful reg's and free other resources */
10064 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
10070 static void netif_free_rx_queues(struct net_device *dev)
10072 unsigned int i, count = dev->num_rx_queues;
10074 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
10078 for (i = 0; i < count; i++)
10079 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
10084 static void netdev_init_one_queue(struct net_device *dev,
10085 struct netdev_queue *queue, void *_unused)
10087 /* Initialize queue lock */
10088 spin_lock_init(&queue->_xmit_lock);
10089 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
10090 queue->xmit_lock_owner = -1;
10091 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
10094 dql_init(&queue->dql, HZ);
10098 static void netif_free_tx_queues(struct net_device *dev)
10103 static int netif_alloc_netdev_queues(struct net_device *dev)
10105 unsigned int count = dev->num_tx_queues;
10106 struct netdev_queue *tx;
10107 size_t sz = count * sizeof(*tx);
10109 if (count < 1 || count > 0xffff)
10112 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
10118 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
10119 spin_lock_init(&dev->tx_global_lock);
10124 void netif_tx_stop_all_queues(struct net_device *dev)
10128 for (i = 0; i < dev->num_tx_queues; i++) {
10129 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
10131 netif_tx_stop_queue(txq);
10134 EXPORT_SYMBOL(netif_tx_stop_all_queues);
10137 * register_netdevice - register a network device
10138 * @dev: device to register
10140 * Take a completed network device structure and add it to the kernel
10141 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10142 * chain. 0 is returned on success. A negative errno code is returned
10143 * on a failure to set up the device, or if the name is a duplicate.
10145 * Callers must hold the rtnl semaphore. You may want
10146 * register_netdev() instead of this.
10149 * The locking appears insufficient to guarantee two parallel registers
10150 * will not get the same name.
10153 int register_netdevice(struct net_device *dev)
10156 struct net *net = dev_net(dev);
10158 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
10159 NETDEV_FEATURE_COUNT);
10160 BUG_ON(dev_boot_phase);
10165 /* When net_device's are persistent, this will be fatal. */
10166 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
10169 ret = ethtool_check_ops(dev->ethtool_ops);
10173 spin_lock_init(&dev->addr_list_lock);
10174 netdev_set_addr_lockdep_class(dev);
10176 ret = dev_get_valid_name(net, dev, dev->name);
10181 dev->name_node = netdev_name_node_head_alloc(dev);
10182 if (!dev->name_node)
10185 /* Init, if this function is available */
10186 if (dev->netdev_ops->ndo_init) {
10187 ret = dev->netdev_ops->ndo_init(dev);
10191 goto err_free_name;
10195 if (((dev->hw_features | dev->features) &
10196 NETIF_F_HW_VLAN_CTAG_FILTER) &&
10197 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
10198 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
10199 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
10206 dev->ifindex = dev_new_index(net);
10207 else if (__dev_get_by_index(net, dev->ifindex))
10210 /* Transfer changeable features to wanted_features and enable
10211 * software offloads (GSO and GRO).
10213 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
10214 dev->features |= NETIF_F_SOFT_FEATURES;
10216 if (dev->udp_tunnel_nic_info) {
10217 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10218 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
10221 dev->wanted_features = dev->features & dev->hw_features;
10223 if (!(dev->flags & IFF_LOOPBACK))
10224 dev->hw_features |= NETIF_F_NOCACHE_COPY;
10226 /* If IPv4 TCP segmentation offload is supported we should also
10227 * allow the device to enable segmenting the frame with the option
10228 * of ignoring a static IP ID value. This doesn't enable the
10229 * feature itself but allows the user to enable it later.
10231 if (dev->hw_features & NETIF_F_TSO)
10232 dev->hw_features |= NETIF_F_TSO_MANGLEID;
10233 if (dev->vlan_features & NETIF_F_TSO)
10234 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
10235 if (dev->mpls_features & NETIF_F_TSO)
10236 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
10237 if (dev->hw_enc_features & NETIF_F_TSO)
10238 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
10240 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
10242 dev->vlan_features |= NETIF_F_HIGHDMA;
10244 /* Make NETIF_F_SG inheritable to tunnel devices.
10246 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
10248 /* Make NETIF_F_SG inheritable to MPLS.
10250 dev->mpls_features |= NETIF_F_SG;
10252 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
10253 ret = notifier_to_errno(ret);
10257 ret = netdev_register_kobject(dev);
10259 dev->reg_state = NETREG_UNREGISTERED;
10262 dev->reg_state = NETREG_REGISTERED;
10264 __netdev_update_features(dev);
10267 * Default initial state at registry is that the
10268 * device is present.
10271 set_bit(__LINK_STATE_PRESENT, &dev->state);
10273 linkwatch_init_dev(dev);
10275 dev_init_scheduler(dev);
10277 list_netdevice(dev);
10278 add_device_randomness(dev->dev_addr, dev->addr_len);
10280 /* If the device has permanent device address, driver should
10281 * set dev_addr and also addr_assign_type should be set to
10282 * NET_ADDR_PERM (default value).
10284 if (dev->addr_assign_type == NET_ADDR_PERM)
10285 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
10287 /* Notify protocols, that a new device appeared. */
10288 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
10289 ret = notifier_to_errno(ret);
10291 /* Expect explicit free_netdev() on failure */
10292 dev->needs_free_netdev = false;
10293 unregister_netdevice_queue(dev, NULL);
10297 * Prevent userspace races by waiting until the network
10298 * device is fully setup before sending notifications.
10300 if (!dev->rtnl_link_ops ||
10301 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10302 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10308 if (dev->netdev_ops->ndo_uninit)
10309 dev->netdev_ops->ndo_uninit(dev);
10310 if (dev->priv_destructor)
10311 dev->priv_destructor(dev);
10313 netdev_name_node_free(dev->name_node);
10316 EXPORT_SYMBOL(register_netdevice);
10319 * init_dummy_netdev - init a dummy network device for NAPI
10320 * @dev: device to init
10322 * This takes a network device structure and initialize the minimum
10323 * amount of fields so it can be used to schedule NAPI polls without
10324 * registering a full blown interface. This is to be used by drivers
10325 * that need to tie several hardware interfaces to a single NAPI
10326 * poll scheduler due to HW limitations.
10328 int init_dummy_netdev(struct net_device *dev)
10330 /* Clear everything. Note we don't initialize spinlocks
10331 * are they aren't supposed to be taken by any of the
10332 * NAPI code and this dummy netdev is supposed to be
10333 * only ever used for NAPI polls
10335 memset(dev, 0, sizeof(struct net_device));
10337 /* make sure we BUG if trying to hit standard
10338 * register/unregister code path
10340 dev->reg_state = NETREG_DUMMY;
10342 /* NAPI wants this */
10343 INIT_LIST_HEAD(&dev->napi_list);
10345 /* a dummy interface is started by default */
10346 set_bit(__LINK_STATE_PRESENT, &dev->state);
10347 set_bit(__LINK_STATE_START, &dev->state);
10349 /* napi_busy_loop stats accounting wants this */
10350 dev_net_set(dev, &init_net);
10352 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10353 * because users of this 'device' dont need to change
10359 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10363 * register_netdev - register a network device
10364 * @dev: device to register
10366 * Take a completed network device structure and add it to the kernel
10367 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10368 * chain. 0 is returned on success. A negative errno code is returned
10369 * on a failure to set up the device, or if the name is a duplicate.
10371 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10372 * and expands the device name if you passed a format string to
10375 int register_netdev(struct net_device *dev)
10379 if (rtnl_lock_killable())
10381 err = register_netdevice(dev);
10385 EXPORT_SYMBOL(register_netdev);
10387 int netdev_refcnt_read(const struct net_device *dev)
10389 #ifdef CONFIG_PCPU_DEV_REFCNT
10392 for_each_possible_cpu(i)
10393 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10396 return refcount_read(&dev->dev_refcnt);
10399 EXPORT_SYMBOL(netdev_refcnt_read);
10401 int netdev_unregister_timeout_secs __read_mostly = 10;
10403 #define WAIT_REFS_MIN_MSECS 1
10404 #define WAIT_REFS_MAX_MSECS 250
10406 * netdev_wait_allrefs - wait until all references are gone.
10407 * @dev: target net_device
10409 * This is called when unregistering network devices.
10411 * Any protocol or device that holds a reference should register
10412 * for netdevice notification, and cleanup and put back the
10413 * reference if they receive an UNREGISTER event.
10414 * We can get stuck here if buggy protocols don't correctly
10417 static void netdev_wait_allrefs(struct net_device *dev)
10419 unsigned long rebroadcast_time, warning_time;
10420 int wait = 0, refcnt;
10422 linkwatch_forget_dev(dev);
10424 rebroadcast_time = warning_time = jiffies;
10425 refcnt = netdev_refcnt_read(dev);
10427 while (refcnt != 1) {
10428 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10431 /* Rebroadcast unregister notification */
10432 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10438 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10440 /* We must not have linkwatch events
10441 * pending on unregister. If this
10442 * happens, we simply run the queue
10443 * unscheduled, resulting in a noop
10446 linkwatch_run_queue();
10451 rebroadcast_time = jiffies;
10456 wait = WAIT_REFS_MIN_MSECS;
10459 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10462 refcnt = netdev_refcnt_read(dev);
10465 time_after(jiffies, warning_time +
10466 netdev_unregister_timeout_secs * HZ)) {
10467 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10468 dev->name, refcnt);
10469 warning_time = jiffies;
10474 /* The sequence is:
10478 * register_netdevice(x1);
10479 * register_netdevice(x2);
10481 * unregister_netdevice(y1);
10482 * unregister_netdevice(y2);
10488 * We are invoked by rtnl_unlock().
10489 * This allows us to deal with problems:
10490 * 1) We can delete sysfs objects which invoke hotplug
10491 * without deadlocking with linkwatch via keventd.
10492 * 2) Since we run with the RTNL semaphore not held, we can sleep
10493 * safely in order to wait for the netdev refcnt to drop to zero.
10495 * We must not return until all unregister events added during
10496 * the interval the lock was held have been completed.
10498 void netdev_run_todo(void)
10500 struct list_head list;
10501 #ifdef CONFIG_LOCKDEP
10502 struct list_head unlink_list;
10504 list_replace_init(&net_unlink_list, &unlink_list);
10506 while (!list_empty(&unlink_list)) {
10507 struct net_device *dev = list_first_entry(&unlink_list,
10510 list_del_init(&dev->unlink_list);
10511 dev->nested_level = dev->lower_level - 1;
10515 /* Snapshot list, allow later requests */
10516 list_replace_init(&net_todo_list, &list);
10521 /* Wait for rcu callbacks to finish before next phase */
10522 if (!list_empty(&list))
10525 while (!list_empty(&list)) {
10526 struct net_device *dev
10527 = list_first_entry(&list, struct net_device, todo_list);
10528 list_del(&dev->todo_list);
10530 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10531 pr_err("network todo '%s' but state %d\n",
10532 dev->name, dev->reg_state);
10537 dev->reg_state = NETREG_UNREGISTERED;
10539 netdev_wait_allrefs(dev);
10542 BUG_ON(netdev_refcnt_read(dev) != 1);
10543 BUG_ON(!list_empty(&dev->ptype_all));
10544 BUG_ON(!list_empty(&dev->ptype_specific));
10545 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10546 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10547 #if IS_ENABLED(CONFIG_DECNET)
10548 WARN_ON(dev->dn_ptr);
10550 if (dev->priv_destructor)
10551 dev->priv_destructor(dev);
10552 if (dev->needs_free_netdev)
10555 /* Report a network device has been unregistered */
10557 dev_net(dev)->dev_unreg_count--;
10559 wake_up(&netdev_unregistering_wq);
10561 /* Free network device */
10562 kobject_put(&dev->dev.kobj);
10566 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10567 * all the same fields in the same order as net_device_stats, with only
10568 * the type differing, but rtnl_link_stats64 may have additional fields
10569 * at the end for newer counters.
10571 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10572 const struct net_device_stats *netdev_stats)
10574 #if BITS_PER_LONG == 64
10575 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10576 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10577 /* zero out counters that only exist in rtnl_link_stats64 */
10578 memset((char *)stats64 + sizeof(*netdev_stats), 0,
10579 sizeof(*stats64) - sizeof(*netdev_stats));
10581 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10582 const unsigned long *src = (const unsigned long *)netdev_stats;
10583 u64 *dst = (u64 *)stats64;
10585 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10586 for (i = 0; i < n; i++)
10588 /* zero out counters that only exist in rtnl_link_stats64 */
10589 memset((char *)stats64 + n * sizeof(u64), 0,
10590 sizeof(*stats64) - n * sizeof(u64));
10593 EXPORT_SYMBOL(netdev_stats_to_stats64);
10596 * dev_get_stats - get network device statistics
10597 * @dev: device to get statistics from
10598 * @storage: place to store stats
10600 * Get network statistics from device. Return @storage.
10601 * The device driver may provide its own method by setting
10602 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10603 * otherwise the internal statistics structure is used.
10605 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10606 struct rtnl_link_stats64 *storage)
10608 const struct net_device_ops *ops = dev->netdev_ops;
10610 if (ops->ndo_get_stats64) {
10611 memset(storage, 0, sizeof(*storage));
10612 ops->ndo_get_stats64(dev, storage);
10613 } else if (ops->ndo_get_stats) {
10614 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10616 netdev_stats_to_stats64(storage, &dev->stats);
10618 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10619 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10620 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10623 EXPORT_SYMBOL(dev_get_stats);
10626 * dev_fetch_sw_netstats - get per-cpu network device statistics
10627 * @s: place to store stats
10628 * @netstats: per-cpu network stats to read from
10630 * Read per-cpu network statistics and populate the related fields in @s.
10632 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10633 const struct pcpu_sw_netstats __percpu *netstats)
10637 for_each_possible_cpu(cpu) {
10638 const struct pcpu_sw_netstats *stats;
10639 struct pcpu_sw_netstats tmp;
10640 unsigned int start;
10642 stats = per_cpu_ptr(netstats, cpu);
10644 start = u64_stats_fetch_begin_irq(&stats->syncp);
10645 tmp.rx_packets = stats->rx_packets;
10646 tmp.rx_bytes = stats->rx_bytes;
10647 tmp.tx_packets = stats->tx_packets;
10648 tmp.tx_bytes = stats->tx_bytes;
10649 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10651 s->rx_packets += tmp.rx_packets;
10652 s->rx_bytes += tmp.rx_bytes;
10653 s->tx_packets += tmp.tx_packets;
10654 s->tx_bytes += tmp.tx_bytes;
10657 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10660 * dev_get_tstats64 - ndo_get_stats64 implementation
10661 * @dev: device to get statistics from
10662 * @s: place to store stats
10664 * Populate @s from dev->stats and dev->tstats. Can be used as
10665 * ndo_get_stats64() callback.
10667 void dev_get_tstats64(struct net_device *dev, struct rtnl_link_stats64 *s)
10669 netdev_stats_to_stats64(s, &dev->stats);
10670 dev_fetch_sw_netstats(s, dev->tstats);
10672 EXPORT_SYMBOL_GPL(dev_get_tstats64);
10674 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10676 struct netdev_queue *queue = dev_ingress_queue(dev);
10678 #ifdef CONFIG_NET_CLS_ACT
10681 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10684 netdev_init_one_queue(dev, queue, NULL);
10685 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10686 queue->qdisc_sleeping = &noop_qdisc;
10687 rcu_assign_pointer(dev->ingress_queue, queue);
10692 static const struct ethtool_ops default_ethtool_ops;
10694 void netdev_set_default_ethtool_ops(struct net_device *dev,
10695 const struct ethtool_ops *ops)
10697 if (dev->ethtool_ops == &default_ethtool_ops)
10698 dev->ethtool_ops = ops;
10700 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10702 void netdev_freemem(struct net_device *dev)
10704 char *addr = (char *)dev - dev->padded;
10710 * alloc_netdev_mqs - allocate network device
10711 * @sizeof_priv: size of private data to allocate space for
10712 * @name: device name format string
10713 * @name_assign_type: origin of device name
10714 * @setup: callback to initialize device
10715 * @txqs: the number of TX subqueues to allocate
10716 * @rxqs: the number of RX subqueues to allocate
10718 * Allocates a struct net_device with private data area for driver use
10719 * and performs basic initialization. Also allocates subqueue structs
10720 * for each queue on the device.
10722 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10723 unsigned char name_assign_type,
10724 void (*setup)(struct net_device *),
10725 unsigned int txqs, unsigned int rxqs)
10727 struct net_device *dev;
10728 unsigned int alloc_size;
10729 struct net_device *p;
10731 BUG_ON(strlen(name) >= sizeof(dev->name));
10734 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10739 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10743 alloc_size = sizeof(struct net_device);
10745 /* ensure 32-byte alignment of private area */
10746 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10747 alloc_size += sizeof_priv;
10749 /* ensure 32-byte alignment of whole construct */
10750 alloc_size += NETDEV_ALIGN - 1;
10752 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
10756 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10757 dev->padded = (char *)dev - (char *)p;
10759 #ifdef CONFIG_PCPU_DEV_REFCNT
10760 dev->pcpu_refcnt = alloc_percpu(int);
10761 if (!dev->pcpu_refcnt)
10765 refcount_set(&dev->dev_refcnt, 1);
10768 if (dev_addr_init(dev))
10774 dev_net_set(dev, &init_net);
10776 dev->gso_max_size = GSO_MAX_SIZE;
10777 dev->gso_max_segs = GSO_MAX_SEGS;
10778 dev->upper_level = 1;
10779 dev->lower_level = 1;
10780 #ifdef CONFIG_LOCKDEP
10781 dev->nested_level = 0;
10782 INIT_LIST_HEAD(&dev->unlink_list);
10785 INIT_LIST_HEAD(&dev->napi_list);
10786 INIT_LIST_HEAD(&dev->unreg_list);
10787 INIT_LIST_HEAD(&dev->close_list);
10788 INIT_LIST_HEAD(&dev->link_watch_list);
10789 INIT_LIST_HEAD(&dev->adj_list.upper);
10790 INIT_LIST_HEAD(&dev->adj_list.lower);
10791 INIT_LIST_HEAD(&dev->ptype_all);
10792 INIT_LIST_HEAD(&dev->ptype_specific);
10793 INIT_LIST_HEAD(&dev->net_notifier_list);
10794 #ifdef CONFIG_NET_SCHED
10795 hash_init(dev->qdisc_hash);
10797 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10800 if (!dev->tx_queue_len) {
10801 dev->priv_flags |= IFF_NO_QUEUE;
10802 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10805 dev->num_tx_queues = txqs;
10806 dev->real_num_tx_queues = txqs;
10807 if (netif_alloc_netdev_queues(dev))
10810 dev->num_rx_queues = rxqs;
10811 dev->real_num_rx_queues = rxqs;
10812 if (netif_alloc_rx_queues(dev))
10815 strcpy(dev->name, name);
10816 dev->name_assign_type = name_assign_type;
10817 dev->group = INIT_NETDEV_GROUP;
10818 if (!dev->ethtool_ops)
10819 dev->ethtool_ops = &default_ethtool_ops;
10821 nf_hook_ingress_init(dev);
10830 #ifdef CONFIG_PCPU_DEV_REFCNT
10831 free_percpu(dev->pcpu_refcnt);
10834 netdev_freemem(dev);
10837 EXPORT_SYMBOL(alloc_netdev_mqs);
10840 * free_netdev - free network device
10843 * This function does the last stage of destroying an allocated device
10844 * interface. The reference to the device object is released. If this
10845 * is the last reference then it will be freed.Must be called in process
10848 void free_netdev(struct net_device *dev)
10850 struct napi_struct *p, *n;
10854 /* When called immediately after register_netdevice() failed the unwind
10855 * handling may still be dismantling the device. Handle that case by
10856 * deferring the free.
10858 if (dev->reg_state == NETREG_UNREGISTERING) {
10860 dev->needs_free_netdev = true;
10864 netif_free_tx_queues(dev);
10865 netif_free_rx_queues(dev);
10867 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10869 /* Flush device addresses */
10870 dev_addr_flush(dev);
10872 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10875 #ifdef CONFIG_PCPU_DEV_REFCNT
10876 free_percpu(dev->pcpu_refcnt);
10877 dev->pcpu_refcnt = NULL;
10879 free_percpu(dev->xdp_bulkq);
10880 dev->xdp_bulkq = NULL;
10882 /* Compatibility with error handling in drivers */
10883 if (dev->reg_state == NETREG_UNINITIALIZED) {
10884 netdev_freemem(dev);
10888 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10889 dev->reg_state = NETREG_RELEASED;
10891 /* will free via device release */
10892 put_device(&dev->dev);
10894 EXPORT_SYMBOL(free_netdev);
10897 * synchronize_net - Synchronize with packet receive processing
10899 * Wait for packets currently being received to be done.
10900 * Does not block later packets from starting.
10902 void synchronize_net(void)
10905 if (rtnl_is_locked())
10906 synchronize_rcu_expedited();
10910 EXPORT_SYMBOL(synchronize_net);
10913 * unregister_netdevice_queue - remove device from the kernel
10917 * This function shuts down a device interface and removes it
10918 * from the kernel tables.
10919 * If head not NULL, device is queued to be unregistered later.
10921 * Callers must hold the rtnl semaphore. You may want
10922 * unregister_netdev() instead of this.
10925 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10930 list_move_tail(&dev->unreg_list, head);
10934 list_add(&dev->unreg_list, &single);
10935 unregister_netdevice_many(&single);
10938 EXPORT_SYMBOL(unregister_netdevice_queue);
10941 * unregister_netdevice_many - unregister many devices
10942 * @head: list of devices
10944 * Note: As most callers use a stack allocated list_head,
10945 * we force a list_del() to make sure stack wont be corrupted later.
10947 void unregister_netdevice_many(struct list_head *head)
10949 struct net_device *dev, *tmp;
10950 LIST_HEAD(close_head);
10952 BUG_ON(dev_boot_phase);
10955 if (list_empty(head))
10958 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
10959 /* Some devices call without registering
10960 * for initialization unwind. Remove those
10961 * devices and proceed with the remaining.
10963 if (dev->reg_state == NETREG_UNINITIALIZED) {
10964 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
10968 list_del(&dev->unreg_list);
10971 dev->dismantle = true;
10972 BUG_ON(dev->reg_state != NETREG_REGISTERED);
10975 /* If device is running, close it first. */
10976 list_for_each_entry(dev, head, unreg_list)
10977 list_add_tail(&dev->close_list, &close_head);
10978 dev_close_many(&close_head, true);
10980 list_for_each_entry(dev, head, unreg_list) {
10981 /* And unlink it from device chain. */
10982 unlist_netdevice(dev);
10984 dev->reg_state = NETREG_UNREGISTERING;
10986 flush_all_backlogs();
10990 list_for_each_entry(dev, head, unreg_list) {
10991 struct sk_buff *skb = NULL;
10993 /* Shutdown queueing discipline. */
10996 dev_xdp_uninstall(dev);
10998 /* Notify protocols, that we are about to destroy
10999 * this device. They should clean all the things.
11001 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11003 if (!dev->rtnl_link_ops ||
11004 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
11005 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
11006 GFP_KERNEL, NULL, 0);
11009 * Flush the unicast and multicast chains
11014 netdev_name_node_alt_flush(dev);
11015 netdev_name_node_free(dev->name_node);
11017 if (dev->netdev_ops->ndo_uninit)
11018 dev->netdev_ops->ndo_uninit(dev);
11021 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
11023 /* Notifier chain MUST detach us all upper devices. */
11024 WARN_ON(netdev_has_any_upper_dev(dev));
11025 WARN_ON(netdev_has_any_lower_dev(dev));
11027 /* Remove entries from kobject tree */
11028 netdev_unregister_kobject(dev);
11030 /* Remove XPS queueing entries */
11031 netif_reset_xps_queues_gt(dev, 0);
11037 list_for_each_entry(dev, head, unreg_list) {
11044 EXPORT_SYMBOL(unregister_netdevice_many);
11047 * unregister_netdev - remove device from the kernel
11050 * This function shuts down a device interface and removes it
11051 * from the kernel tables.
11053 * This is just a wrapper for unregister_netdevice that takes
11054 * the rtnl semaphore. In general you want to use this and not
11055 * unregister_netdevice.
11057 void unregister_netdev(struct net_device *dev)
11060 unregister_netdevice(dev);
11063 EXPORT_SYMBOL(unregister_netdev);
11066 * __dev_change_net_namespace - move device to different nethost namespace
11068 * @net: network namespace
11069 * @pat: If not NULL name pattern to try if the current device name
11070 * is already taken in the destination network namespace.
11071 * @new_ifindex: If not zero, specifies device index in the target
11074 * This function shuts down a device interface and moves it
11075 * to a new network namespace. On success 0 is returned, on
11076 * a failure a netagive errno code is returned.
11078 * Callers must hold the rtnl semaphore.
11081 int __dev_change_net_namespace(struct net_device *dev, struct net *net,
11082 const char *pat, int new_ifindex)
11084 struct net *net_old = dev_net(dev);
11089 /* Don't allow namespace local devices to be moved. */
11091 if (dev->features & NETIF_F_NETNS_LOCAL)
11094 /* Ensure the device has been registrered */
11095 if (dev->reg_state != NETREG_REGISTERED)
11098 /* Get out if there is nothing todo */
11100 if (net_eq(net_old, net))
11103 /* Pick the destination device name, and ensure
11104 * we can use it in the destination network namespace.
11107 if (__dev_get_by_name(net, dev->name)) {
11108 /* We get here if we can't use the current device name */
11111 err = dev_get_valid_name(net, dev, pat);
11116 /* Check that new_ifindex isn't used yet. */
11118 if (new_ifindex && __dev_get_by_index(net, new_ifindex))
11122 * And now a mini version of register_netdevice unregister_netdevice.
11125 /* If device is running close it first. */
11128 /* And unlink it from device chain */
11129 unlist_netdevice(dev);
11133 /* Shutdown queueing discipline. */
11136 /* Notify protocols, that we are about to destroy
11137 * this device. They should clean all the things.
11139 * Note that dev->reg_state stays at NETREG_REGISTERED.
11140 * This is wanted because this way 8021q and macvlan know
11141 * the device is just moving and can keep their slaves up.
11143 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
11146 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
11147 /* If there is an ifindex conflict assign a new one */
11148 if (!new_ifindex) {
11149 if (__dev_get_by_index(net, dev->ifindex))
11150 new_ifindex = dev_new_index(net);
11152 new_ifindex = dev->ifindex;
11155 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
11159 * Flush the unicast and multicast chains
11164 /* Send a netdev-removed uevent to the old namespace */
11165 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
11166 netdev_adjacent_del_links(dev);
11168 /* Move per-net netdevice notifiers that are following the netdevice */
11169 move_netdevice_notifiers_dev_net(dev, net);
11171 /* Actually switch the network namespace */
11172 dev_net_set(dev, net);
11173 dev->ifindex = new_ifindex;
11175 /* Send a netdev-add uevent to the new namespace */
11176 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
11177 netdev_adjacent_add_links(dev);
11179 /* Fixup kobjects */
11180 err = device_rename(&dev->dev, dev->name);
11183 /* Adapt owner in case owning user namespace of target network
11184 * namespace is different from the original one.
11186 err = netdev_change_owner(dev, net_old, net);
11189 /* Add the device back in the hashes */
11190 list_netdevice(dev);
11192 /* Notify protocols, that a new device appeared. */
11193 call_netdevice_notifiers(NETDEV_REGISTER, dev);
11196 * Prevent userspace races by waiting until the network
11197 * device is fully setup before sending notifications.
11199 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
11206 EXPORT_SYMBOL_GPL(__dev_change_net_namespace);
11208 static int dev_cpu_dead(unsigned int oldcpu)
11210 struct sk_buff **list_skb;
11211 struct sk_buff *skb;
11213 struct softnet_data *sd, *oldsd, *remsd = NULL;
11215 local_irq_disable();
11216 cpu = smp_processor_id();
11217 sd = &per_cpu(softnet_data, cpu);
11218 oldsd = &per_cpu(softnet_data, oldcpu);
11220 /* Find end of our completion_queue. */
11221 list_skb = &sd->completion_queue;
11223 list_skb = &(*list_skb)->next;
11224 /* Append completion queue from offline CPU. */
11225 *list_skb = oldsd->completion_queue;
11226 oldsd->completion_queue = NULL;
11228 /* Append output queue from offline CPU. */
11229 if (oldsd->output_queue) {
11230 *sd->output_queue_tailp = oldsd->output_queue;
11231 sd->output_queue_tailp = oldsd->output_queue_tailp;
11232 oldsd->output_queue = NULL;
11233 oldsd->output_queue_tailp = &oldsd->output_queue;
11235 /* Append NAPI poll list from offline CPU, with one exception :
11236 * process_backlog() must be called by cpu owning percpu backlog.
11237 * We properly handle process_queue & input_pkt_queue later.
11239 while (!list_empty(&oldsd->poll_list)) {
11240 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
11241 struct napi_struct,
11244 list_del_init(&napi->poll_list);
11245 if (napi->poll == process_backlog)
11248 ____napi_schedule(sd, napi);
11251 raise_softirq_irqoff(NET_TX_SOFTIRQ);
11252 local_irq_enable();
11255 remsd = oldsd->rps_ipi_list;
11256 oldsd->rps_ipi_list = NULL;
11258 /* send out pending IPI's on offline CPU */
11259 net_rps_send_ipi(remsd);
11261 /* Process offline CPU's input_pkt_queue */
11262 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
11264 input_queue_head_incr(oldsd);
11266 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
11268 input_queue_head_incr(oldsd);
11275 * netdev_increment_features - increment feature set by one
11276 * @all: current feature set
11277 * @one: new feature set
11278 * @mask: mask feature set
11280 * Computes a new feature set after adding a device with feature set
11281 * @one to the master device with current feature set @all. Will not
11282 * enable anything that is off in @mask. Returns the new feature set.
11284 netdev_features_t netdev_increment_features(netdev_features_t all,
11285 netdev_features_t one, netdev_features_t mask)
11287 if (mask & NETIF_F_HW_CSUM)
11288 mask |= NETIF_F_CSUM_MASK;
11289 mask |= NETIF_F_VLAN_CHALLENGED;
11291 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
11292 all &= one | ~NETIF_F_ALL_FOR_ALL;
11294 /* If one device supports hw checksumming, set for all. */
11295 if (all & NETIF_F_HW_CSUM)
11296 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
11300 EXPORT_SYMBOL(netdev_increment_features);
11302 static struct hlist_head * __net_init netdev_create_hash(void)
11305 struct hlist_head *hash;
11307 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
11309 for (i = 0; i < NETDEV_HASHENTRIES; i++)
11310 INIT_HLIST_HEAD(&hash[i]);
11315 /* Initialize per network namespace state */
11316 static int __net_init netdev_init(struct net *net)
11318 BUILD_BUG_ON(GRO_HASH_BUCKETS >
11319 8 * sizeof_field(struct napi_struct, gro_bitmask));
11321 if (net != &init_net)
11322 INIT_LIST_HEAD(&net->dev_base_head);
11324 net->dev_name_head = netdev_create_hash();
11325 if (net->dev_name_head == NULL)
11328 net->dev_index_head = netdev_create_hash();
11329 if (net->dev_index_head == NULL)
11332 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
11337 kfree(net->dev_name_head);
11343 * netdev_drivername - network driver for the device
11344 * @dev: network device
11346 * Determine network driver for device.
11348 const char *netdev_drivername(const struct net_device *dev)
11350 const struct device_driver *driver;
11351 const struct device *parent;
11352 const char *empty = "";
11354 parent = dev->dev.parent;
11358 driver = parent->driver;
11359 if (driver && driver->name)
11360 return driver->name;
11364 static void __netdev_printk(const char *level, const struct net_device *dev,
11365 struct va_format *vaf)
11367 if (dev && dev->dev.parent) {
11368 dev_printk_emit(level[1] - '0',
11371 dev_driver_string(dev->dev.parent),
11372 dev_name(dev->dev.parent),
11373 netdev_name(dev), netdev_reg_state(dev),
11376 printk("%s%s%s: %pV",
11377 level, netdev_name(dev), netdev_reg_state(dev), vaf);
11379 printk("%s(NULL net_device): %pV", level, vaf);
11383 void netdev_printk(const char *level, const struct net_device *dev,
11384 const char *format, ...)
11386 struct va_format vaf;
11389 va_start(args, format);
11394 __netdev_printk(level, dev, &vaf);
11398 EXPORT_SYMBOL(netdev_printk);
11400 #define define_netdev_printk_level(func, level) \
11401 void func(const struct net_device *dev, const char *fmt, ...) \
11403 struct va_format vaf; \
11406 va_start(args, fmt); \
11411 __netdev_printk(level, dev, &vaf); \
11415 EXPORT_SYMBOL(func);
11417 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
11418 define_netdev_printk_level(netdev_alert, KERN_ALERT);
11419 define_netdev_printk_level(netdev_crit, KERN_CRIT);
11420 define_netdev_printk_level(netdev_err, KERN_ERR);
11421 define_netdev_printk_level(netdev_warn, KERN_WARNING);
11422 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
11423 define_netdev_printk_level(netdev_info, KERN_INFO);
11425 static void __net_exit netdev_exit(struct net *net)
11427 kfree(net->dev_name_head);
11428 kfree(net->dev_index_head);
11429 if (net != &init_net)
11430 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11433 static struct pernet_operations __net_initdata netdev_net_ops = {
11434 .init = netdev_init,
11435 .exit = netdev_exit,
11438 static void __net_exit default_device_exit(struct net *net)
11440 struct net_device *dev, *aux;
11442 * Push all migratable network devices back to the
11443 * initial network namespace
11446 for_each_netdev_safe(net, dev, aux) {
11448 char fb_name[IFNAMSIZ];
11450 /* Ignore unmoveable devices (i.e. loopback) */
11451 if (dev->features & NETIF_F_NETNS_LOCAL)
11454 /* Leave virtual devices for the generic cleanup */
11455 if (dev->rtnl_link_ops && !dev->rtnl_link_ops->netns_refund)
11458 /* Push remaining network devices to init_net */
11459 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11460 if (__dev_get_by_name(&init_net, fb_name))
11461 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11462 err = dev_change_net_namespace(dev, &init_net, fb_name);
11464 pr_emerg("%s: failed to move %s to init_net: %d\n",
11465 __func__, dev->name, err);
11472 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11474 /* Return with the rtnl_lock held when there are no network
11475 * devices unregistering in any network namespace in net_list.
11478 bool unregistering;
11479 DEFINE_WAIT_FUNC(wait, woken_wake_function);
11481 add_wait_queue(&netdev_unregistering_wq, &wait);
11483 unregistering = false;
11485 list_for_each_entry(net, net_list, exit_list) {
11486 if (net->dev_unreg_count > 0) {
11487 unregistering = true;
11491 if (!unregistering)
11495 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11497 remove_wait_queue(&netdev_unregistering_wq, &wait);
11500 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11502 /* At exit all network devices most be removed from a network
11503 * namespace. Do this in the reverse order of registration.
11504 * Do this across as many network namespaces as possible to
11505 * improve batching efficiency.
11507 struct net_device *dev;
11509 LIST_HEAD(dev_kill_list);
11511 /* To prevent network device cleanup code from dereferencing
11512 * loopback devices or network devices that have been freed
11513 * wait here for all pending unregistrations to complete,
11514 * before unregistring the loopback device and allowing the
11515 * network namespace be freed.
11517 * The netdev todo list containing all network devices
11518 * unregistrations that happen in default_device_exit_batch
11519 * will run in the rtnl_unlock() at the end of
11520 * default_device_exit_batch.
11522 rtnl_lock_unregistering(net_list);
11523 list_for_each_entry(net, net_list, exit_list) {
11524 for_each_netdev_reverse(net, dev) {
11525 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11526 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11528 unregister_netdevice_queue(dev, &dev_kill_list);
11531 unregister_netdevice_many(&dev_kill_list);
11535 static struct pernet_operations __net_initdata default_device_ops = {
11536 .exit = default_device_exit,
11537 .exit_batch = default_device_exit_batch,
11541 * Initialize the DEV module. At boot time this walks the device list and
11542 * unhooks any devices that fail to initialise (normally hardware not
11543 * present) and leaves us with a valid list of present and active devices.
11548 * This is called single threaded during boot, so no need
11549 * to take the rtnl semaphore.
11551 static int __init net_dev_init(void)
11553 int i, rc = -ENOMEM;
11555 BUG_ON(!dev_boot_phase);
11557 if (dev_proc_init())
11560 if (netdev_kobject_init())
11563 INIT_LIST_HEAD(&ptype_all);
11564 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11565 INIT_LIST_HEAD(&ptype_base[i]);
11567 INIT_LIST_HEAD(&offload_base);
11569 if (register_pernet_subsys(&netdev_net_ops))
11573 * Initialise the packet receive queues.
11576 for_each_possible_cpu(i) {
11577 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11578 struct softnet_data *sd = &per_cpu(softnet_data, i);
11580 INIT_WORK(flush, flush_backlog);
11582 skb_queue_head_init(&sd->input_pkt_queue);
11583 skb_queue_head_init(&sd->process_queue);
11584 #ifdef CONFIG_XFRM_OFFLOAD
11585 skb_queue_head_init(&sd->xfrm_backlog);
11587 INIT_LIST_HEAD(&sd->poll_list);
11588 sd->output_queue_tailp = &sd->output_queue;
11590 INIT_CSD(&sd->csd, rps_trigger_softirq, sd);
11594 init_gro_hash(&sd->backlog);
11595 sd->backlog.poll = process_backlog;
11596 sd->backlog.weight = weight_p;
11599 dev_boot_phase = 0;
11601 /* The loopback device is special if any other network devices
11602 * is present in a network namespace the loopback device must
11603 * be present. Since we now dynamically allocate and free the
11604 * loopback device ensure this invariant is maintained by
11605 * keeping the loopback device as the first device on the
11606 * list of network devices. Ensuring the loopback devices
11607 * is the first device that appears and the last network device
11610 if (register_pernet_device(&loopback_net_ops))
11613 if (register_pernet_device(&default_device_ops))
11616 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11617 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11619 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11620 NULL, dev_cpu_dead);
11627 subsys_initcall(net_dev_init);