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/bpf.h>
95 #include <linux/bpf_trace.h>
96 #include <net/net_namespace.h>
98 #include <net/busy_poll.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/stat.h>
102 #include <net/dst_metadata.h>
103 #include <net/pkt_sched.h>
104 #include <net/pkt_cls.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/module.h>
110 #include <linux/netpoll.h>
111 #include <linux/rcupdate.h>
112 #include <linux/delay.h>
113 #include <net/iw_handler.h>
114 #include <asm/current.h>
115 #include <linux/audit.h>
116 #include <linux/dmaengine.h>
117 #include <linux/err.h>
118 #include <linux/ctype.h>
119 #include <linux/if_arp.h>
120 #include <linux/if_vlan.h>
121 #include <linux/ip.h>
123 #include <net/mpls.h>
124 #include <linux/ipv6.h>
125 #include <linux/in.h>
126 #include <linux/jhash.h>
127 #include <linux/random.h>
128 #include <trace/events/napi.h>
129 #include <trace/events/net.h>
130 #include <trace/events/skb.h>
131 #include <linux/inetdevice.h>
132 #include <linux/cpu_rmap.h>
133 #include <linux/static_key.h>
134 #include <linux/hashtable.h>
135 #include <linux/vmalloc.h>
136 #include <linux/if_macvlan.h>
137 #include <linux/errqueue.h>
138 #include <linux/hrtimer.h>
139 #include <linux/netfilter_ingress.h>
140 #include <linux/crash_dump.h>
141 #include <linux/sctp.h>
142 #include <net/udp_tunnel.h>
143 #include <linux/net_namespace.h>
144 #include <linux/indirect_call_wrapper.h>
145 #include <net/devlink.h>
147 #include "net-sysfs.h"
149 #define MAX_GRO_SKBS 8
151 /* This should be increased if a protocol with a bigger head is added. */
152 #define GRO_MAX_HEAD (MAX_HEADER + 128)
154 static DEFINE_SPINLOCK(ptype_lock);
155 static DEFINE_SPINLOCK(offload_lock);
156 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
157 struct list_head ptype_all __read_mostly; /* Taps */
158 static struct list_head offload_base __read_mostly;
160 static int netif_rx_internal(struct sk_buff *skb);
161 static int call_netdevice_notifiers_info(unsigned long val,
162 struct netdev_notifier_info *info);
163 static int call_netdevice_notifiers_extack(unsigned long val,
164 struct net_device *dev,
165 struct netlink_ext_ack *extack);
166 static struct napi_struct *napi_by_id(unsigned int napi_id);
169 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
172 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
174 * Writers must hold the rtnl semaphore while they loop through the
175 * dev_base_head list, and hold dev_base_lock for writing when they do the
176 * actual updates. This allows pure readers to access the list even
177 * while a writer is preparing to update it.
179 * To put it another way, dev_base_lock is held for writing only to
180 * protect against pure readers; the rtnl semaphore provides the
181 * protection against other writers.
183 * See, for example usages, register_netdevice() and
184 * unregister_netdevice(), which must be called with the rtnl
187 DEFINE_RWLOCK(dev_base_lock);
188 EXPORT_SYMBOL(dev_base_lock);
190 static DEFINE_MUTEX(ifalias_mutex);
192 /* protects napi_hash addition/deletion and napi_gen_id */
193 static DEFINE_SPINLOCK(napi_hash_lock);
195 static unsigned int napi_gen_id = NR_CPUS;
196 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
198 static DECLARE_RWSEM(devnet_rename_sem);
200 static inline void dev_base_seq_inc(struct net *net)
202 while (++net->dev_base_seq == 0)
206 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
208 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
210 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
213 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
215 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
218 static inline void rps_lock(struct softnet_data *sd)
221 spin_lock(&sd->input_pkt_queue.lock);
225 static inline void rps_unlock(struct softnet_data *sd)
228 spin_unlock(&sd->input_pkt_queue.lock);
232 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
235 struct netdev_name_node *name_node;
237 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
240 INIT_HLIST_NODE(&name_node->hlist);
241 name_node->dev = dev;
242 name_node->name = name;
246 static struct netdev_name_node *
247 netdev_name_node_head_alloc(struct net_device *dev)
249 struct netdev_name_node *name_node;
251 name_node = netdev_name_node_alloc(dev, dev->name);
254 INIT_LIST_HEAD(&name_node->list);
258 static void netdev_name_node_free(struct netdev_name_node *name_node)
263 static void netdev_name_node_add(struct net *net,
264 struct netdev_name_node *name_node)
266 hlist_add_head_rcu(&name_node->hlist,
267 dev_name_hash(net, name_node->name));
270 static void netdev_name_node_del(struct netdev_name_node *name_node)
272 hlist_del_rcu(&name_node->hlist);
275 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
278 struct hlist_head *head = dev_name_hash(net, name);
279 struct netdev_name_node *name_node;
281 hlist_for_each_entry(name_node, head, hlist)
282 if (!strcmp(name_node->name, name))
287 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
290 struct hlist_head *head = dev_name_hash(net, name);
291 struct netdev_name_node *name_node;
293 hlist_for_each_entry_rcu(name_node, head, hlist)
294 if (!strcmp(name_node->name, name))
299 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
301 struct netdev_name_node *name_node;
302 struct net *net = dev_net(dev);
304 name_node = netdev_name_node_lookup(net, name);
307 name_node = netdev_name_node_alloc(dev, name);
310 netdev_name_node_add(net, name_node);
311 /* The node that holds dev->name acts as a head of per-device list. */
312 list_add_tail(&name_node->list, &dev->name_node->list);
316 EXPORT_SYMBOL(netdev_name_node_alt_create);
318 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
320 list_del(&name_node->list);
321 netdev_name_node_del(name_node);
322 kfree(name_node->name);
323 netdev_name_node_free(name_node);
326 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
328 struct netdev_name_node *name_node;
329 struct net *net = dev_net(dev);
331 name_node = netdev_name_node_lookup(net, name);
334 /* lookup might have found our primary name or a name belonging
337 if (name_node == dev->name_node || name_node->dev != dev)
340 __netdev_name_node_alt_destroy(name_node);
344 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
346 static void netdev_name_node_alt_flush(struct net_device *dev)
348 struct netdev_name_node *name_node, *tmp;
350 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
351 __netdev_name_node_alt_destroy(name_node);
354 /* Device list insertion */
355 static void list_netdevice(struct net_device *dev)
357 struct net *net = dev_net(dev);
361 write_lock_bh(&dev_base_lock);
362 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
363 netdev_name_node_add(net, dev->name_node);
364 hlist_add_head_rcu(&dev->index_hlist,
365 dev_index_hash(net, dev->ifindex));
366 write_unlock_bh(&dev_base_lock);
368 dev_base_seq_inc(net);
371 /* Device list removal
372 * caller must respect a RCU grace period before freeing/reusing dev
374 static void unlist_netdevice(struct net_device *dev)
378 /* Unlink dev from the device chain */
379 write_lock_bh(&dev_base_lock);
380 list_del_rcu(&dev->dev_list);
381 netdev_name_node_del(dev->name_node);
382 hlist_del_rcu(&dev->index_hlist);
383 write_unlock_bh(&dev_base_lock);
385 dev_base_seq_inc(dev_net(dev));
392 static RAW_NOTIFIER_HEAD(netdev_chain);
395 * Device drivers call our routines to queue packets here. We empty the
396 * queue in the local softnet handler.
399 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
400 EXPORT_PER_CPU_SYMBOL(softnet_data);
402 #ifdef CONFIG_LOCKDEP
404 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
405 * according to dev->type
407 static const unsigned short netdev_lock_type[] = {
408 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
409 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
410 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
411 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
412 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
413 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
414 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
415 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
416 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
417 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
418 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
419 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
420 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
421 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
422 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
424 static const char *const netdev_lock_name[] = {
425 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
426 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
427 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
428 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
429 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
430 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
431 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
432 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
433 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
434 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
435 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
436 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
437 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
438 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
439 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
441 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
442 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
444 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
448 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
449 if (netdev_lock_type[i] == dev_type)
451 /* the last key is used by default */
452 return ARRAY_SIZE(netdev_lock_type) - 1;
455 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
456 unsigned short dev_type)
460 i = netdev_lock_pos(dev_type);
461 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
462 netdev_lock_name[i]);
465 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
469 i = netdev_lock_pos(dev->type);
470 lockdep_set_class_and_name(&dev->addr_list_lock,
471 &netdev_addr_lock_key[i],
472 netdev_lock_name[i]);
475 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
476 unsigned short dev_type)
480 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
485 /*******************************************************************************
487 * Protocol management and registration routines
489 *******************************************************************************/
493 * Add a protocol ID to the list. Now that the input handler is
494 * smarter we can dispense with all the messy stuff that used to be
497 * BEWARE!!! Protocol handlers, mangling input packets,
498 * MUST BE last in hash buckets and checking protocol handlers
499 * MUST start from promiscuous ptype_all chain in net_bh.
500 * It is true now, do not change it.
501 * Explanation follows: if protocol handler, mangling packet, will
502 * be the first on list, it is not able to sense, that packet
503 * is cloned and should be copied-on-write, so that it will
504 * change it and subsequent readers will get broken packet.
508 static inline struct list_head *ptype_head(const struct packet_type *pt)
510 if (pt->type == htons(ETH_P_ALL))
511 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
513 return pt->dev ? &pt->dev->ptype_specific :
514 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
518 * dev_add_pack - add packet handler
519 * @pt: packet type declaration
521 * Add a protocol handler to the networking stack. The passed &packet_type
522 * is linked into kernel lists and may not be freed until it has been
523 * removed from the kernel lists.
525 * This call does not sleep therefore it can not
526 * guarantee all CPU's that are in middle of receiving packets
527 * will see the new packet type (until the next received packet).
530 void dev_add_pack(struct packet_type *pt)
532 struct list_head *head = ptype_head(pt);
534 spin_lock(&ptype_lock);
535 list_add_rcu(&pt->list, head);
536 spin_unlock(&ptype_lock);
538 EXPORT_SYMBOL(dev_add_pack);
541 * __dev_remove_pack - remove packet handler
542 * @pt: packet type declaration
544 * Remove a protocol handler that was previously added to the kernel
545 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
546 * from the kernel lists and can be freed or reused once this function
549 * The packet type might still be in use by receivers
550 * and must not be freed until after all the CPU's have gone
551 * through a quiescent state.
553 void __dev_remove_pack(struct packet_type *pt)
555 struct list_head *head = ptype_head(pt);
556 struct packet_type *pt1;
558 spin_lock(&ptype_lock);
560 list_for_each_entry(pt1, head, list) {
562 list_del_rcu(&pt->list);
567 pr_warn("dev_remove_pack: %p not found\n", pt);
569 spin_unlock(&ptype_lock);
571 EXPORT_SYMBOL(__dev_remove_pack);
574 * dev_remove_pack - remove packet handler
575 * @pt: packet type declaration
577 * Remove a protocol handler that was previously added to the kernel
578 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
579 * from the kernel lists and can be freed or reused once this function
582 * This call sleeps to guarantee that no CPU is looking at the packet
585 void dev_remove_pack(struct packet_type *pt)
587 __dev_remove_pack(pt);
591 EXPORT_SYMBOL(dev_remove_pack);
595 * dev_add_offload - register offload handlers
596 * @po: protocol offload declaration
598 * Add protocol offload handlers to the networking stack. The passed
599 * &proto_offload is linked into kernel lists and may not be freed until
600 * it has been removed from the kernel lists.
602 * This call does not sleep therefore it can not
603 * guarantee all CPU's that are in middle of receiving packets
604 * will see the new offload handlers (until the next received packet).
606 void dev_add_offload(struct packet_offload *po)
608 struct packet_offload *elem;
610 spin_lock(&offload_lock);
611 list_for_each_entry(elem, &offload_base, list) {
612 if (po->priority < elem->priority)
615 list_add_rcu(&po->list, elem->list.prev);
616 spin_unlock(&offload_lock);
618 EXPORT_SYMBOL(dev_add_offload);
621 * __dev_remove_offload - remove offload handler
622 * @po: packet offload declaration
624 * Remove a protocol offload handler that was previously added to the
625 * kernel offload handlers by dev_add_offload(). The passed &offload_type
626 * is removed from the kernel lists and can be freed or reused once this
629 * The packet type might still be in use by receivers
630 * and must not be freed until after all the CPU's have gone
631 * through a quiescent state.
633 static void __dev_remove_offload(struct packet_offload *po)
635 struct list_head *head = &offload_base;
636 struct packet_offload *po1;
638 spin_lock(&offload_lock);
640 list_for_each_entry(po1, head, list) {
642 list_del_rcu(&po->list);
647 pr_warn("dev_remove_offload: %p not found\n", po);
649 spin_unlock(&offload_lock);
653 * dev_remove_offload - remove packet offload handler
654 * @po: packet offload declaration
656 * Remove a packet offload handler that was previously added to the kernel
657 * offload handlers by dev_add_offload(). The passed &offload_type is
658 * removed from the kernel lists and can be freed or reused once this
661 * This call sleeps to guarantee that no CPU is looking at the packet
664 void dev_remove_offload(struct packet_offload *po)
666 __dev_remove_offload(po);
670 EXPORT_SYMBOL(dev_remove_offload);
672 /******************************************************************************
674 * Device Boot-time Settings Routines
676 ******************************************************************************/
678 /* Boot time configuration table */
679 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
682 * netdev_boot_setup_add - add new setup entry
683 * @name: name of the device
684 * @map: configured settings for the device
686 * Adds new setup entry to the dev_boot_setup list. The function
687 * returns 0 on error and 1 on success. This is a generic routine to
690 static int netdev_boot_setup_add(char *name, struct ifmap *map)
692 struct netdev_boot_setup *s;
696 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
697 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
698 memset(s[i].name, 0, sizeof(s[i].name));
699 strlcpy(s[i].name, name, IFNAMSIZ);
700 memcpy(&s[i].map, map, sizeof(s[i].map));
705 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
709 * netdev_boot_setup_check - check boot time settings
710 * @dev: the netdevice
712 * Check boot time settings for the device.
713 * The found settings are set for the device to be used
714 * later in the device probing.
715 * Returns 0 if no settings found, 1 if they are.
717 int netdev_boot_setup_check(struct net_device *dev)
719 struct netdev_boot_setup *s = dev_boot_setup;
722 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
723 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
724 !strcmp(dev->name, s[i].name)) {
725 dev->irq = s[i].map.irq;
726 dev->base_addr = s[i].map.base_addr;
727 dev->mem_start = s[i].map.mem_start;
728 dev->mem_end = s[i].map.mem_end;
734 EXPORT_SYMBOL(netdev_boot_setup_check);
738 * netdev_boot_base - get address from boot time settings
739 * @prefix: prefix for network device
740 * @unit: id for network device
742 * Check boot time settings for the base address of device.
743 * The found settings are set for the device to be used
744 * later in the device probing.
745 * Returns 0 if no settings found.
747 unsigned long netdev_boot_base(const char *prefix, int unit)
749 const struct netdev_boot_setup *s = dev_boot_setup;
753 sprintf(name, "%s%d", prefix, unit);
756 * If device already registered then return base of 1
757 * to indicate not to probe for this interface
759 if (__dev_get_by_name(&init_net, name))
762 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
763 if (!strcmp(name, s[i].name))
764 return s[i].map.base_addr;
769 * Saves at boot time configured settings for any netdevice.
771 int __init netdev_boot_setup(char *str)
776 str = get_options(str, ARRAY_SIZE(ints), ints);
781 memset(&map, 0, sizeof(map));
785 map.base_addr = ints[2];
787 map.mem_start = ints[3];
789 map.mem_end = ints[4];
791 /* Add new entry to the list */
792 return netdev_boot_setup_add(str, &map);
795 __setup("netdev=", netdev_boot_setup);
797 /*******************************************************************************
799 * Device Interface Subroutines
801 *******************************************************************************/
804 * dev_get_iflink - get 'iflink' value of a interface
805 * @dev: targeted interface
807 * Indicates the ifindex the interface is linked to.
808 * Physical interfaces have the same 'ifindex' and 'iflink' values.
811 int dev_get_iflink(const struct net_device *dev)
813 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
814 return dev->netdev_ops->ndo_get_iflink(dev);
818 EXPORT_SYMBOL(dev_get_iflink);
821 * dev_fill_metadata_dst - Retrieve tunnel egress information.
822 * @dev: targeted interface
825 * For better visibility of tunnel traffic OVS needs to retrieve
826 * egress tunnel information for a packet. Following API allows
827 * user to get this info.
829 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
831 struct ip_tunnel_info *info;
833 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
836 info = skb_tunnel_info_unclone(skb);
839 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
842 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
844 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
847 * __dev_get_by_name - find a device by its name
848 * @net: the applicable net namespace
849 * @name: name to find
851 * Find an interface by name. Must be called under RTNL semaphore
852 * or @dev_base_lock. If the name is found a pointer to the device
853 * is returned. If the name is not found then %NULL is returned. The
854 * reference counters are not incremented so the caller must be
855 * careful with locks.
858 struct net_device *__dev_get_by_name(struct net *net, const char *name)
860 struct netdev_name_node *node_name;
862 node_name = netdev_name_node_lookup(net, name);
863 return node_name ? node_name->dev : NULL;
865 EXPORT_SYMBOL(__dev_get_by_name);
868 * dev_get_by_name_rcu - find a device by its name
869 * @net: the applicable net namespace
870 * @name: name to find
872 * Find an interface by name.
873 * If the name is found a pointer to the device is returned.
874 * If the name is not found then %NULL is returned.
875 * The reference counters are not incremented so the caller must be
876 * careful with locks. The caller must hold RCU lock.
879 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
881 struct netdev_name_node *node_name;
883 node_name = netdev_name_node_lookup_rcu(net, name);
884 return node_name ? node_name->dev : NULL;
886 EXPORT_SYMBOL(dev_get_by_name_rcu);
889 * dev_get_by_name - find a device by its name
890 * @net: the applicable net namespace
891 * @name: name to find
893 * Find an interface by name. This can be called from any
894 * context and does its own locking. The returned handle has
895 * the usage count incremented and the caller must use dev_put() to
896 * release it when it is no longer needed. %NULL is returned if no
897 * matching device is found.
900 struct net_device *dev_get_by_name(struct net *net, const char *name)
902 struct net_device *dev;
905 dev = dev_get_by_name_rcu(net, name);
911 EXPORT_SYMBOL(dev_get_by_name);
914 * __dev_get_by_index - find a device by its ifindex
915 * @net: the applicable net namespace
916 * @ifindex: index of device
918 * Search for an interface by index. Returns %NULL if the device
919 * is not found or a pointer to the device. The device has not
920 * had its reference counter increased so the caller must be careful
921 * about locking. The caller must hold either the RTNL semaphore
925 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
927 struct net_device *dev;
928 struct hlist_head *head = dev_index_hash(net, ifindex);
930 hlist_for_each_entry(dev, head, index_hlist)
931 if (dev->ifindex == ifindex)
936 EXPORT_SYMBOL(__dev_get_by_index);
939 * dev_get_by_index_rcu - find a device by its ifindex
940 * @net: the applicable net namespace
941 * @ifindex: index of device
943 * Search for an interface by index. Returns %NULL if the device
944 * is not found or a pointer to the device. The device has not
945 * had its reference counter increased so the caller must be careful
946 * about locking. The caller must hold RCU lock.
949 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
951 struct net_device *dev;
952 struct hlist_head *head = dev_index_hash(net, ifindex);
954 hlist_for_each_entry_rcu(dev, head, index_hlist)
955 if (dev->ifindex == ifindex)
960 EXPORT_SYMBOL(dev_get_by_index_rcu);
964 * dev_get_by_index - find a device by its ifindex
965 * @net: the applicable net namespace
966 * @ifindex: index of device
968 * Search for an interface by index. Returns NULL if the device
969 * is not found or a pointer to the device. The device returned has
970 * had a reference added and the pointer is safe until the user calls
971 * dev_put to indicate they have finished with it.
974 struct net_device *dev_get_by_index(struct net *net, int ifindex)
976 struct net_device *dev;
979 dev = dev_get_by_index_rcu(net, ifindex);
985 EXPORT_SYMBOL(dev_get_by_index);
988 * dev_get_by_napi_id - find a device by napi_id
989 * @napi_id: ID of the NAPI struct
991 * Search for an interface by NAPI ID. Returns %NULL if the device
992 * is not found or a pointer to the device. The device has not had
993 * its reference counter increased so the caller must be careful
994 * about locking. The caller must hold RCU lock.
997 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
999 struct napi_struct *napi;
1001 WARN_ON_ONCE(!rcu_read_lock_held());
1003 if (napi_id < MIN_NAPI_ID)
1006 napi = napi_by_id(napi_id);
1008 return napi ? napi->dev : NULL;
1010 EXPORT_SYMBOL(dev_get_by_napi_id);
1013 * netdev_get_name - get a netdevice name, knowing its ifindex.
1014 * @net: network namespace
1015 * @name: a pointer to the buffer where the name will be stored.
1016 * @ifindex: the ifindex of the interface to get the name from.
1018 int netdev_get_name(struct net *net, char *name, int ifindex)
1020 struct net_device *dev;
1023 down_read(&devnet_rename_sem);
1026 dev = dev_get_by_index_rcu(net, ifindex);
1032 strcpy(name, dev->name);
1037 up_read(&devnet_rename_sem);
1042 * dev_getbyhwaddr_rcu - find a device by its hardware address
1043 * @net: the applicable net namespace
1044 * @type: media type of device
1045 * @ha: hardware address
1047 * Search for an interface by MAC address. Returns NULL if the device
1048 * is not found or a pointer to the device.
1049 * The caller must hold RCU or RTNL.
1050 * The returned device has not had its ref count increased
1051 * and the caller must therefore be careful about locking
1055 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
1058 struct net_device *dev;
1060 for_each_netdev_rcu(net, dev)
1061 if (dev->type == type &&
1062 !memcmp(dev->dev_addr, ha, dev->addr_len))
1067 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1069 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
1071 struct net_device *dev;
1074 for_each_netdev(net, dev)
1075 if (dev->type == type)
1080 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
1082 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1084 struct net_device *dev, *ret = NULL;
1087 for_each_netdev_rcu(net, dev)
1088 if (dev->type == type) {
1096 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1099 * __dev_get_by_flags - find any device with given flags
1100 * @net: the applicable net namespace
1101 * @if_flags: IFF_* values
1102 * @mask: bitmask of bits in if_flags to check
1104 * Search for any interface with the given flags. Returns NULL if a device
1105 * is not found or a pointer to the device. Must be called inside
1106 * rtnl_lock(), and result refcount is unchanged.
1109 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1110 unsigned short mask)
1112 struct net_device *dev, *ret;
1117 for_each_netdev(net, dev) {
1118 if (((dev->flags ^ if_flags) & mask) == 0) {
1125 EXPORT_SYMBOL(__dev_get_by_flags);
1128 * dev_valid_name - check if name is okay for network device
1129 * @name: name string
1131 * Network device names need to be valid file names to
1132 * to allow sysfs to work. We also disallow any kind of
1135 bool dev_valid_name(const char *name)
1139 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1141 if (!strcmp(name, ".") || !strcmp(name, ".."))
1145 if (*name == '/' || *name == ':' || isspace(*name))
1151 EXPORT_SYMBOL(dev_valid_name);
1154 * __dev_alloc_name - allocate a name for a device
1155 * @net: network namespace to allocate the device name in
1156 * @name: name format string
1157 * @buf: scratch buffer and result name string
1159 * Passed a format string - eg "lt%d" it will try and find a suitable
1160 * id. It scans list of devices to build up a free map, then chooses
1161 * the first empty slot. The caller must hold the dev_base or rtnl lock
1162 * while allocating the name and adding the device in order to avoid
1164 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1165 * Returns the number of the unit assigned or a negative errno code.
1168 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1172 const int max_netdevices = 8*PAGE_SIZE;
1173 unsigned long *inuse;
1174 struct net_device *d;
1176 if (!dev_valid_name(name))
1179 p = strchr(name, '%');
1182 * Verify the string as this thing may have come from
1183 * the user. There must be either one "%d" and no other "%"
1186 if (p[1] != 'd' || strchr(p + 2, '%'))
1189 /* Use one page as a bit array of possible slots */
1190 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1194 for_each_netdev(net, d) {
1195 if (!sscanf(d->name, name, &i))
1197 if (i < 0 || i >= max_netdevices)
1200 /* avoid cases where sscanf is not exact inverse of printf */
1201 snprintf(buf, IFNAMSIZ, name, i);
1202 if (!strncmp(buf, d->name, IFNAMSIZ))
1206 i = find_first_zero_bit(inuse, max_netdevices);
1207 free_page((unsigned long) inuse);
1210 snprintf(buf, IFNAMSIZ, name, i);
1211 if (!__dev_get_by_name(net, buf))
1214 /* It is possible to run out of possible slots
1215 * when the name is long and there isn't enough space left
1216 * for the digits, or if all bits are used.
1221 static int dev_alloc_name_ns(struct net *net,
1222 struct net_device *dev,
1229 ret = __dev_alloc_name(net, name, buf);
1231 strlcpy(dev->name, buf, IFNAMSIZ);
1236 * dev_alloc_name - allocate a name for a device
1238 * @name: name format string
1240 * Passed a format string - eg "lt%d" it will try and find a suitable
1241 * id. It scans list of devices to build up a free map, then chooses
1242 * the first empty slot. The caller must hold the dev_base or rtnl lock
1243 * while allocating the name and adding the device in order to avoid
1245 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1246 * Returns the number of the unit assigned or a negative errno code.
1249 int dev_alloc_name(struct net_device *dev, const char *name)
1251 return dev_alloc_name_ns(dev_net(dev), dev, name);
1253 EXPORT_SYMBOL(dev_alloc_name);
1255 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1260 if (!dev_valid_name(name))
1263 if (strchr(name, '%'))
1264 return dev_alloc_name_ns(net, dev, name);
1265 else if (__dev_get_by_name(net, name))
1267 else if (dev->name != name)
1268 strlcpy(dev->name, name, IFNAMSIZ);
1274 * dev_change_name - change name of a device
1276 * @newname: name (or format string) must be at least IFNAMSIZ
1278 * Change name of a device, can pass format strings "eth%d".
1281 int dev_change_name(struct net_device *dev, const char *newname)
1283 unsigned char old_assign_type;
1284 char oldname[IFNAMSIZ];
1290 BUG_ON(!dev_net(dev));
1294 /* Some auto-enslaved devices e.g. failover slaves are
1295 * special, as userspace might rename the device after
1296 * the interface had been brought up and running since
1297 * the point kernel initiated auto-enslavement. Allow
1298 * live name change even when these slave devices are
1301 * Typically, users of these auto-enslaving devices
1302 * don't actually care about slave name change, as
1303 * they are supposed to operate on master interface
1306 if (dev->flags & IFF_UP &&
1307 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1310 down_write(&devnet_rename_sem);
1312 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1313 up_write(&devnet_rename_sem);
1317 memcpy(oldname, dev->name, IFNAMSIZ);
1319 err = dev_get_valid_name(net, dev, newname);
1321 up_write(&devnet_rename_sem);
1325 if (oldname[0] && !strchr(oldname, '%'))
1326 netdev_info(dev, "renamed from %s\n", oldname);
1328 old_assign_type = dev->name_assign_type;
1329 dev->name_assign_type = NET_NAME_RENAMED;
1332 ret = device_rename(&dev->dev, dev->name);
1334 memcpy(dev->name, oldname, IFNAMSIZ);
1335 dev->name_assign_type = old_assign_type;
1336 up_write(&devnet_rename_sem);
1340 up_write(&devnet_rename_sem);
1342 netdev_adjacent_rename_links(dev, oldname);
1344 write_lock_bh(&dev_base_lock);
1345 netdev_name_node_del(dev->name_node);
1346 write_unlock_bh(&dev_base_lock);
1350 write_lock_bh(&dev_base_lock);
1351 netdev_name_node_add(net, dev->name_node);
1352 write_unlock_bh(&dev_base_lock);
1354 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1355 ret = notifier_to_errno(ret);
1358 /* err >= 0 after dev_alloc_name() or stores the first errno */
1361 down_write(&devnet_rename_sem);
1362 memcpy(dev->name, oldname, IFNAMSIZ);
1363 memcpy(oldname, newname, IFNAMSIZ);
1364 dev->name_assign_type = old_assign_type;
1365 old_assign_type = NET_NAME_RENAMED;
1368 pr_err("%s: name change rollback failed: %d\n",
1377 * dev_set_alias - change ifalias of a device
1379 * @alias: name up to IFALIASZ
1380 * @len: limit of bytes to copy from info
1382 * Set ifalias for a device,
1384 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1386 struct dev_ifalias *new_alias = NULL;
1388 if (len >= IFALIASZ)
1392 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1396 memcpy(new_alias->ifalias, alias, len);
1397 new_alias->ifalias[len] = 0;
1400 mutex_lock(&ifalias_mutex);
1401 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1402 mutex_is_locked(&ifalias_mutex));
1403 mutex_unlock(&ifalias_mutex);
1406 kfree_rcu(new_alias, rcuhead);
1410 EXPORT_SYMBOL(dev_set_alias);
1413 * dev_get_alias - get ifalias of a device
1415 * @name: buffer to store name of ifalias
1416 * @len: size of buffer
1418 * get ifalias for a device. Caller must make sure dev cannot go
1419 * away, e.g. rcu read lock or own a reference count to device.
1421 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1423 const struct dev_ifalias *alias;
1427 alias = rcu_dereference(dev->ifalias);
1429 ret = snprintf(name, len, "%s", alias->ifalias);
1436 * netdev_features_change - device changes features
1437 * @dev: device to cause notification
1439 * Called to indicate a device has changed features.
1441 void netdev_features_change(struct net_device *dev)
1443 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1445 EXPORT_SYMBOL(netdev_features_change);
1448 * netdev_state_change - device changes state
1449 * @dev: device to cause notification
1451 * Called to indicate a device has changed state. This function calls
1452 * the notifier chains for netdev_chain and sends a NEWLINK message
1453 * to the routing socket.
1455 void netdev_state_change(struct net_device *dev)
1457 if (dev->flags & IFF_UP) {
1458 struct netdev_notifier_change_info change_info = {
1462 call_netdevice_notifiers_info(NETDEV_CHANGE,
1464 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1467 EXPORT_SYMBOL(netdev_state_change);
1470 * netdev_notify_peers - notify network peers about existence of @dev
1471 * @dev: network device
1473 * Generate traffic such that interested network peers are aware of
1474 * @dev, such as by generating a gratuitous ARP. This may be used when
1475 * a device wants to inform the rest of the network about some sort of
1476 * reconfiguration such as a failover event or virtual machine
1479 void netdev_notify_peers(struct net_device *dev)
1482 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1483 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1486 EXPORT_SYMBOL(netdev_notify_peers);
1488 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1490 const struct net_device_ops *ops = dev->netdev_ops;
1495 if (!netif_device_present(dev))
1498 /* Block netpoll from trying to do any rx path servicing.
1499 * If we don't do this there is a chance ndo_poll_controller
1500 * or ndo_poll may be running while we open the device
1502 netpoll_poll_disable(dev);
1504 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1505 ret = notifier_to_errno(ret);
1509 set_bit(__LINK_STATE_START, &dev->state);
1511 if (ops->ndo_validate_addr)
1512 ret = ops->ndo_validate_addr(dev);
1514 if (!ret && ops->ndo_open)
1515 ret = ops->ndo_open(dev);
1517 netpoll_poll_enable(dev);
1520 clear_bit(__LINK_STATE_START, &dev->state);
1522 dev->flags |= IFF_UP;
1523 dev_set_rx_mode(dev);
1525 add_device_randomness(dev->dev_addr, dev->addr_len);
1532 * dev_open - prepare an interface for use.
1533 * @dev: device to open
1534 * @extack: netlink extended ack
1536 * Takes a device from down to up state. The device's private open
1537 * function is invoked and then the multicast lists are loaded. Finally
1538 * the device is moved into the up state and a %NETDEV_UP message is
1539 * sent to the netdev notifier chain.
1541 * Calling this function on an active interface is a nop. On a failure
1542 * a negative errno code is returned.
1544 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1548 if (dev->flags & IFF_UP)
1551 ret = __dev_open(dev, extack);
1555 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1556 call_netdevice_notifiers(NETDEV_UP, dev);
1560 EXPORT_SYMBOL(dev_open);
1562 static void __dev_close_many(struct list_head *head)
1564 struct net_device *dev;
1569 list_for_each_entry(dev, head, close_list) {
1570 /* Temporarily disable netpoll until the interface is down */
1571 netpoll_poll_disable(dev);
1573 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1575 clear_bit(__LINK_STATE_START, &dev->state);
1577 /* Synchronize to scheduled poll. We cannot touch poll list, it
1578 * can be even on different cpu. So just clear netif_running().
1580 * dev->stop() will invoke napi_disable() on all of it's
1581 * napi_struct instances on this device.
1583 smp_mb__after_atomic(); /* Commit netif_running(). */
1586 dev_deactivate_many(head);
1588 list_for_each_entry(dev, head, close_list) {
1589 const struct net_device_ops *ops = dev->netdev_ops;
1592 * Call the device specific close. This cannot fail.
1593 * Only if device is UP
1595 * We allow it to be called even after a DETACH hot-plug
1601 dev->flags &= ~IFF_UP;
1602 netpoll_poll_enable(dev);
1606 static void __dev_close(struct net_device *dev)
1610 list_add(&dev->close_list, &single);
1611 __dev_close_many(&single);
1615 void dev_close_many(struct list_head *head, bool unlink)
1617 struct net_device *dev, *tmp;
1619 /* Remove the devices that don't need to be closed */
1620 list_for_each_entry_safe(dev, tmp, head, close_list)
1621 if (!(dev->flags & IFF_UP))
1622 list_del_init(&dev->close_list);
1624 __dev_close_many(head);
1626 list_for_each_entry_safe(dev, tmp, head, close_list) {
1627 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1628 call_netdevice_notifiers(NETDEV_DOWN, dev);
1630 list_del_init(&dev->close_list);
1633 EXPORT_SYMBOL(dev_close_many);
1636 * dev_close - shutdown an interface.
1637 * @dev: device to shutdown
1639 * This function moves an active device into down state. A
1640 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1641 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1644 void dev_close(struct net_device *dev)
1646 if (dev->flags & IFF_UP) {
1649 list_add(&dev->close_list, &single);
1650 dev_close_many(&single, true);
1654 EXPORT_SYMBOL(dev_close);
1658 * dev_disable_lro - disable Large Receive Offload on a device
1661 * Disable Large Receive Offload (LRO) on a net device. Must be
1662 * called under RTNL. This is needed if received packets may be
1663 * forwarded to another interface.
1665 void dev_disable_lro(struct net_device *dev)
1667 struct net_device *lower_dev;
1668 struct list_head *iter;
1670 dev->wanted_features &= ~NETIF_F_LRO;
1671 netdev_update_features(dev);
1673 if (unlikely(dev->features & NETIF_F_LRO))
1674 netdev_WARN(dev, "failed to disable LRO!\n");
1676 netdev_for_each_lower_dev(dev, lower_dev, iter)
1677 dev_disable_lro(lower_dev);
1679 EXPORT_SYMBOL(dev_disable_lro);
1682 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1685 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1686 * called under RTNL. This is needed if Generic XDP is installed on
1689 static void dev_disable_gro_hw(struct net_device *dev)
1691 dev->wanted_features &= ~NETIF_F_GRO_HW;
1692 netdev_update_features(dev);
1694 if (unlikely(dev->features & NETIF_F_GRO_HW))
1695 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1698 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1701 case NETDEV_##val: \
1702 return "NETDEV_" __stringify(val);
1704 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1705 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1706 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1707 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1708 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1709 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1710 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1711 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1712 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1716 return "UNKNOWN_NETDEV_EVENT";
1718 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1720 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1721 struct net_device *dev)
1723 struct netdev_notifier_info info = {
1727 return nb->notifier_call(nb, val, &info);
1730 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1731 struct net_device *dev)
1735 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1736 err = notifier_to_errno(err);
1740 if (!(dev->flags & IFF_UP))
1743 call_netdevice_notifier(nb, NETDEV_UP, dev);
1747 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1748 struct net_device *dev)
1750 if (dev->flags & IFF_UP) {
1751 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1753 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1755 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1758 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1761 struct net_device *dev;
1764 for_each_netdev(net, dev) {
1765 err = call_netdevice_register_notifiers(nb, dev);
1772 for_each_netdev_continue_reverse(net, dev)
1773 call_netdevice_unregister_notifiers(nb, dev);
1777 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1780 struct net_device *dev;
1782 for_each_netdev(net, dev)
1783 call_netdevice_unregister_notifiers(nb, dev);
1786 static int dev_boot_phase = 1;
1789 * register_netdevice_notifier - register a network notifier block
1792 * Register a notifier to be called when network device events occur.
1793 * The notifier passed is linked into the kernel structures and must
1794 * not be reused until it has been unregistered. A negative errno code
1795 * is returned on a failure.
1797 * When registered all registration and up events are replayed
1798 * to the new notifier to allow device to have a race free
1799 * view of the network device list.
1802 int register_netdevice_notifier(struct notifier_block *nb)
1807 /* Close race with setup_net() and cleanup_net() */
1808 down_write(&pernet_ops_rwsem);
1810 err = raw_notifier_chain_register(&netdev_chain, nb);
1816 err = call_netdevice_register_net_notifiers(nb, net);
1823 up_write(&pernet_ops_rwsem);
1827 for_each_net_continue_reverse(net)
1828 call_netdevice_unregister_net_notifiers(nb, net);
1830 raw_notifier_chain_unregister(&netdev_chain, nb);
1833 EXPORT_SYMBOL(register_netdevice_notifier);
1836 * unregister_netdevice_notifier - unregister a network notifier block
1839 * Unregister a notifier previously registered by
1840 * register_netdevice_notifier(). The notifier is unlinked into the
1841 * kernel structures and may then be reused. A negative errno code
1842 * is returned on a failure.
1844 * After unregistering unregister and down device events are synthesized
1845 * for all devices on the device list to the removed notifier to remove
1846 * the need for special case cleanup code.
1849 int unregister_netdevice_notifier(struct notifier_block *nb)
1854 /* Close race with setup_net() and cleanup_net() */
1855 down_write(&pernet_ops_rwsem);
1857 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1862 call_netdevice_unregister_net_notifiers(nb, net);
1866 up_write(&pernet_ops_rwsem);
1869 EXPORT_SYMBOL(unregister_netdevice_notifier);
1871 static int __register_netdevice_notifier_net(struct net *net,
1872 struct notifier_block *nb,
1873 bool ignore_call_fail)
1877 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1883 err = call_netdevice_register_net_notifiers(nb, net);
1884 if (err && !ignore_call_fail)
1885 goto chain_unregister;
1890 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1894 static int __unregister_netdevice_notifier_net(struct net *net,
1895 struct notifier_block *nb)
1899 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1903 call_netdevice_unregister_net_notifiers(nb, net);
1908 * register_netdevice_notifier_net - register a per-netns network notifier block
1909 * @net: network namespace
1912 * Register a notifier to be called when network device events occur.
1913 * The notifier passed is linked into the kernel structures and must
1914 * not be reused until it has been unregistered. A negative errno code
1915 * is returned on a failure.
1917 * When registered all registration and up events are replayed
1918 * to the new notifier to allow device to have a race free
1919 * view of the network device list.
1922 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1927 err = __register_netdevice_notifier_net(net, nb, false);
1931 EXPORT_SYMBOL(register_netdevice_notifier_net);
1934 * unregister_netdevice_notifier_net - unregister a per-netns
1935 * network notifier block
1936 * @net: network namespace
1939 * Unregister a notifier previously registered by
1940 * register_netdevice_notifier(). The notifier is unlinked into the
1941 * kernel structures and may then be reused. A negative errno code
1942 * is returned on a failure.
1944 * After unregistering unregister and down device events are synthesized
1945 * for all devices on the device list to the removed notifier to remove
1946 * the need for special case cleanup code.
1949 int unregister_netdevice_notifier_net(struct net *net,
1950 struct notifier_block *nb)
1955 err = __unregister_netdevice_notifier_net(net, nb);
1959 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1961 int register_netdevice_notifier_dev_net(struct net_device *dev,
1962 struct notifier_block *nb,
1963 struct netdev_net_notifier *nn)
1968 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1971 list_add(&nn->list, &dev->net_notifier_list);
1976 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1978 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1979 struct notifier_block *nb,
1980 struct netdev_net_notifier *nn)
1985 list_del(&nn->list);
1986 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1990 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
1992 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
1995 struct netdev_net_notifier *nn;
1997 list_for_each_entry(nn, &dev->net_notifier_list, list) {
1998 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
1999 __register_netdevice_notifier_net(net, nn->nb, true);
2004 * call_netdevice_notifiers_info - call all network notifier blocks
2005 * @val: value passed unmodified to notifier function
2006 * @info: notifier information data
2008 * Call all network notifier blocks. Parameters and return value
2009 * are as for raw_notifier_call_chain().
2012 static int call_netdevice_notifiers_info(unsigned long val,
2013 struct netdev_notifier_info *info)
2015 struct net *net = dev_net(info->dev);
2020 /* Run per-netns notifier block chain first, then run the global one.
2021 * Hopefully, one day, the global one is going to be removed after
2022 * all notifier block registrators get converted to be per-netns.
2024 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2025 if (ret & NOTIFY_STOP_MASK)
2027 return raw_notifier_call_chain(&netdev_chain, val, info);
2030 static int call_netdevice_notifiers_extack(unsigned long val,
2031 struct net_device *dev,
2032 struct netlink_ext_ack *extack)
2034 struct netdev_notifier_info info = {
2039 return call_netdevice_notifiers_info(val, &info);
2043 * call_netdevice_notifiers - call all network notifier blocks
2044 * @val: value passed unmodified to notifier function
2045 * @dev: net_device pointer passed unmodified to notifier function
2047 * Call all network notifier blocks. Parameters and return value
2048 * are as for raw_notifier_call_chain().
2051 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2053 return call_netdevice_notifiers_extack(val, dev, NULL);
2055 EXPORT_SYMBOL(call_netdevice_notifiers);
2058 * call_netdevice_notifiers_mtu - call all network notifier blocks
2059 * @val: value passed unmodified to notifier function
2060 * @dev: net_device pointer passed unmodified to notifier function
2061 * @arg: additional u32 argument passed to the notifier function
2063 * Call all network notifier blocks. Parameters and return value
2064 * are as for raw_notifier_call_chain().
2066 static int call_netdevice_notifiers_mtu(unsigned long val,
2067 struct net_device *dev, u32 arg)
2069 struct netdev_notifier_info_ext info = {
2074 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2076 return call_netdevice_notifiers_info(val, &info.info);
2079 #ifdef CONFIG_NET_INGRESS
2080 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2082 void net_inc_ingress_queue(void)
2084 static_branch_inc(&ingress_needed_key);
2086 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2088 void net_dec_ingress_queue(void)
2090 static_branch_dec(&ingress_needed_key);
2092 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2095 #ifdef CONFIG_NET_EGRESS
2096 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2098 void net_inc_egress_queue(void)
2100 static_branch_inc(&egress_needed_key);
2102 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2104 void net_dec_egress_queue(void)
2106 static_branch_dec(&egress_needed_key);
2108 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2111 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2112 #ifdef CONFIG_JUMP_LABEL
2113 static atomic_t netstamp_needed_deferred;
2114 static atomic_t netstamp_wanted;
2115 static void netstamp_clear(struct work_struct *work)
2117 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2120 wanted = atomic_add_return(deferred, &netstamp_wanted);
2122 static_branch_enable(&netstamp_needed_key);
2124 static_branch_disable(&netstamp_needed_key);
2126 static DECLARE_WORK(netstamp_work, netstamp_clear);
2129 void net_enable_timestamp(void)
2131 #ifdef CONFIG_JUMP_LABEL
2135 wanted = atomic_read(&netstamp_wanted);
2138 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2141 atomic_inc(&netstamp_needed_deferred);
2142 schedule_work(&netstamp_work);
2144 static_branch_inc(&netstamp_needed_key);
2147 EXPORT_SYMBOL(net_enable_timestamp);
2149 void net_disable_timestamp(void)
2151 #ifdef CONFIG_JUMP_LABEL
2155 wanted = atomic_read(&netstamp_wanted);
2158 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2161 atomic_dec(&netstamp_needed_deferred);
2162 schedule_work(&netstamp_work);
2164 static_branch_dec(&netstamp_needed_key);
2167 EXPORT_SYMBOL(net_disable_timestamp);
2169 static inline void net_timestamp_set(struct sk_buff *skb)
2172 if (static_branch_unlikely(&netstamp_needed_key))
2173 __net_timestamp(skb);
2176 #define net_timestamp_check(COND, SKB) \
2177 if (static_branch_unlikely(&netstamp_needed_key)) { \
2178 if ((COND) && !(SKB)->tstamp) \
2179 __net_timestamp(SKB); \
2182 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2186 if (!(dev->flags & IFF_UP))
2189 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
2190 if (skb->len <= len)
2193 /* if TSO is enabled, we don't care about the length as the packet
2194 * could be forwarded without being segmented before
2196 if (skb_is_gso(skb))
2201 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2203 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2205 int ret = ____dev_forward_skb(dev, skb);
2208 skb->protocol = eth_type_trans(skb, dev);
2209 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2214 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2217 * dev_forward_skb - loopback an skb to another netif
2219 * @dev: destination network device
2220 * @skb: buffer to forward
2223 * NET_RX_SUCCESS (no congestion)
2224 * NET_RX_DROP (packet was dropped, but freed)
2226 * dev_forward_skb can be used for injecting an skb from the
2227 * start_xmit function of one device into the receive queue
2228 * of another device.
2230 * The receiving device may be in another namespace, so
2231 * we have to clear all information in the skb that could
2232 * impact namespace isolation.
2234 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2236 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2238 EXPORT_SYMBOL_GPL(dev_forward_skb);
2240 static inline int deliver_skb(struct sk_buff *skb,
2241 struct packet_type *pt_prev,
2242 struct net_device *orig_dev)
2244 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2246 refcount_inc(&skb->users);
2247 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2250 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2251 struct packet_type **pt,
2252 struct net_device *orig_dev,
2254 struct list_head *ptype_list)
2256 struct packet_type *ptype, *pt_prev = *pt;
2258 list_for_each_entry_rcu(ptype, ptype_list, list) {
2259 if (ptype->type != type)
2262 deliver_skb(skb, pt_prev, orig_dev);
2268 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2270 if (!ptype->af_packet_priv || !skb->sk)
2273 if (ptype->id_match)
2274 return ptype->id_match(ptype, skb->sk);
2275 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2282 * dev_nit_active - return true if any network interface taps are in use
2284 * @dev: network device to check for the presence of taps
2286 bool dev_nit_active(struct net_device *dev)
2288 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2290 EXPORT_SYMBOL_GPL(dev_nit_active);
2293 * Support routine. Sends outgoing frames to any network
2294 * taps currently in use.
2297 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2299 struct packet_type *ptype;
2300 struct sk_buff *skb2 = NULL;
2301 struct packet_type *pt_prev = NULL;
2302 struct list_head *ptype_list = &ptype_all;
2306 list_for_each_entry_rcu(ptype, ptype_list, list) {
2307 if (ptype->ignore_outgoing)
2310 /* Never send packets back to the socket
2311 * they originated from - MvS (miquels@drinkel.ow.org)
2313 if (skb_loop_sk(ptype, skb))
2317 deliver_skb(skb2, pt_prev, skb->dev);
2322 /* need to clone skb, done only once */
2323 skb2 = skb_clone(skb, GFP_ATOMIC);
2327 net_timestamp_set(skb2);
2329 /* skb->nh should be correctly
2330 * set by sender, so that the second statement is
2331 * just protection against buggy protocols.
2333 skb_reset_mac_header(skb2);
2335 if (skb_network_header(skb2) < skb2->data ||
2336 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2337 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2338 ntohs(skb2->protocol),
2340 skb_reset_network_header(skb2);
2343 skb2->transport_header = skb2->network_header;
2344 skb2->pkt_type = PACKET_OUTGOING;
2348 if (ptype_list == &ptype_all) {
2349 ptype_list = &dev->ptype_all;
2354 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2355 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2361 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2364 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2365 * @dev: Network device
2366 * @txq: number of queues available
2368 * If real_num_tx_queues is changed the tc mappings may no longer be
2369 * valid. To resolve this verify the tc mapping remains valid and if
2370 * not NULL the mapping. With no priorities mapping to this
2371 * offset/count pair it will no longer be used. In the worst case TC0
2372 * is invalid nothing can be done so disable priority mappings. If is
2373 * expected that drivers will fix this mapping if they can before
2374 * calling netif_set_real_num_tx_queues.
2376 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2379 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2381 /* If TC0 is invalidated disable TC mapping */
2382 if (tc->offset + tc->count > txq) {
2383 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2388 /* Invalidated prio to tc mappings set to TC0 */
2389 for (i = 1; i < TC_BITMASK + 1; i++) {
2390 int q = netdev_get_prio_tc_map(dev, i);
2392 tc = &dev->tc_to_txq[q];
2393 if (tc->offset + tc->count > txq) {
2394 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2396 netdev_set_prio_tc_map(dev, i, 0);
2401 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2404 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2407 /* walk through the TCs and see if it falls into any of them */
2408 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2409 if ((txq - tc->offset) < tc->count)
2413 /* didn't find it, just return -1 to indicate no match */
2419 EXPORT_SYMBOL(netdev_txq_to_tc);
2422 struct static_key xps_needed __read_mostly;
2423 EXPORT_SYMBOL(xps_needed);
2424 struct static_key xps_rxqs_needed __read_mostly;
2425 EXPORT_SYMBOL(xps_rxqs_needed);
2426 static DEFINE_MUTEX(xps_map_mutex);
2427 #define xmap_dereference(P) \
2428 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2430 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2433 struct xps_map *map = NULL;
2437 map = xmap_dereference(dev_maps->attr_map[tci]);
2441 for (pos = map->len; pos--;) {
2442 if (map->queues[pos] != index)
2446 map->queues[pos] = map->queues[--map->len];
2450 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2451 kfree_rcu(map, rcu);
2458 static bool remove_xps_queue_cpu(struct net_device *dev,
2459 struct xps_dev_maps *dev_maps,
2460 int cpu, u16 offset, u16 count)
2462 int num_tc = dev->num_tc ? : 1;
2463 bool active = false;
2466 for (tci = cpu * num_tc; num_tc--; tci++) {
2469 for (i = count, j = offset; i--; j++) {
2470 if (!remove_xps_queue(dev_maps, tci, j))
2480 static void reset_xps_maps(struct net_device *dev,
2481 struct xps_dev_maps *dev_maps,
2485 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2486 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2488 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2490 static_key_slow_dec_cpuslocked(&xps_needed);
2491 kfree_rcu(dev_maps, rcu);
2494 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2495 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2496 u16 offset, u16 count, bool is_rxqs_map)
2498 bool active = false;
2501 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2503 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2506 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2509 for (i = offset + (count - 1); count--; i--) {
2510 netdev_queue_numa_node_write(
2511 netdev_get_tx_queue(dev, i),
2517 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2520 const unsigned long *possible_mask = NULL;
2521 struct xps_dev_maps *dev_maps;
2522 unsigned int nr_ids;
2524 if (!static_key_false(&xps_needed))
2528 mutex_lock(&xps_map_mutex);
2530 if (static_key_false(&xps_rxqs_needed)) {
2531 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2533 nr_ids = dev->num_rx_queues;
2534 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2535 offset, count, true);
2539 dev_maps = xmap_dereference(dev->xps_cpus_map);
2543 if (num_possible_cpus() > 1)
2544 possible_mask = cpumask_bits(cpu_possible_mask);
2545 nr_ids = nr_cpu_ids;
2546 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2550 mutex_unlock(&xps_map_mutex);
2554 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2556 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2559 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2560 u16 index, bool is_rxqs_map)
2562 struct xps_map *new_map;
2563 int alloc_len = XPS_MIN_MAP_ALLOC;
2566 for (pos = 0; map && pos < map->len; pos++) {
2567 if (map->queues[pos] != index)
2572 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2574 if (pos < map->alloc_len)
2577 alloc_len = map->alloc_len * 2;
2580 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2584 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2586 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2587 cpu_to_node(attr_index));
2591 for (i = 0; i < pos; i++)
2592 new_map->queues[i] = map->queues[i];
2593 new_map->alloc_len = alloc_len;
2599 /* Must be called under cpus_read_lock */
2600 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2601 u16 index, bool is_rxqs_map)
2603 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2604 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2605 int i, j, tci, numa_node_id = -2;
2606 int maps_sz, num_tc = 1, tc = 0;
2607 struct xps_map *map, *new_map;
2608 bool active = false;
2609 unsigned int nr_ids;
2612 /* Do not allow XPS on subordinate device directly */
2613 num_tc = dev->num_tc;
2617 /* If queue belongs to subordinate dev use its map */
2618 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2620 tc = netdev_txq_to_tc(dev, index);
2625 mutex_lock(&xps_map_mutex);
2627 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2628 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2629 nr_ids = dev->num_rx_queues;
2631 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2632 if (num_possible_cpus() > 1) {
2633 online_mask = cpumask_bits(cpu_online_mask);
2634 possible_mask = cpumask_bits(cpu_possible_mask);
2636 dev_maps = xmap_dereference(dev->xps_cpus_map);
2637 nr_ids = nr_cpu_ids;
2640 if (maps_sz < L1_CACHE_BYTES)
2641 maps_sz = L1_CACHE_BYTES;
2643 /* allocate memory for queue storage */
2644 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2647 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2648 if (!new_dev_maps) {
2649 mutex_unlock(&xps_map_mutex);
2653 tci = j * num_tc + tc;
2654 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2657 map = expand_xps_map(map, j, index, is_rxqs_map);
2661 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2665 goto out_no_new_maps;
2668 /* Increment static keys at most once per type */
2669 static_key_slow_inc_cpuslocked(&xps_needed);
2671 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2674 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2676 /* copy maps belonging to foreign traffic classes */
2677 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2678 /* fill in the new device map from the old device map */
2679 map = xmap_dereference(dev_maps->attr_map[tci]);
2680 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2683 /* We need to explicitly update tci as prevous loop
2684 * could break out early if dev_maps is NULL.
2686 tci = j * num_tc + tc;
2688 if (netif_attr_test_mask(j, mask, nr_ids) &&
2689 netif_attr_test_online(j, online_mask, nr_ids)) {
2690 /* add tx-queue to CPU/rx-queue maps */
2693 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2694 while ((pos < map->len) && (map->queues[pos] != index))
2697 if (pos == map->len)
2698 map->queues[map->len++] = index;
2701 if (numa_node_id == -2)
2702 numa_node_id = cpu_to_node(j);
2703 else if (numa_node_id != cpu_to_node(j))
2707 } else if (dev_maps) {
2708 /* fill in the new device map from the old device map */
2709 map = xmap_dereference(dev_maps->attr_map[tci]);
2710 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2713 /* copy maps belonging to foreign traffic classes */
2714 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2715 /* fill in the new device map from the old device map */
2716 map = xmap_dereference(dev_maps->attr_map[tci]);
2717 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2722 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2724 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2726 /* Cleanup old maps */
2728 goto out_no_old_maps;
2730 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2732 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2733 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2734 map = xmap_dereference(dev_maps->attr_map[tci]);
2735 if (map && map != new_map)
2736 kfree_rcu(map, rcu);
2740 kfree_rcu(dev_maps, rcu);
2743 dev_maps = new_dev_maps;
2748 /* update Tx queue numa node */
2749 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2750 (numa_node_id >= 0) ?
2751 numa_node_id : NUMA_NO_NODE);
2757 /* removes tx-queue from unused CPUs/rx-queues */
2758 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2760 for (i = tc, tci = j * num_tc; i--; tci++)
2761 active |= remove_xps_queue(dev_maps, tci, index);
2762 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2763 !netif_attr_test_online(j, online_mask, nr_ids))
2764 active |= remove_xps_queue(dev_maps, tci, index);
2765 for (i = num_tc - tc, tci++; --i; tci++)
2766 active |= remove_xps_queue(dev_maps, tci, index);
2769 /* free map if not active */
2771 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2774 mutex_unlock(&xps_map_mutex);
2778 /* remove any maps that we added */
2779 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2781 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2782 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2784 xmap_dereference(dev_maps->attr_map[tci]) :
2786 if (new_map && new_map != map)
2791 mutex_unlock(&xps_map_mutex);
2793 kfree(new_dev_maps);
2796 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2798 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2804 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2809 EXPORT_SYMBOL(netif_set_xps_queue);
2812 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2814 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2816 /* Unbind any subordinate channels */
2817 while (txq-- != &dev->_tx[0]) {
2819 netdev_unbind_sb_channel(dev, txq->sb_dev);
2823 void netdev_reset_tc(struct net_device *dev)
2826 netif_reset_xps_queues_gt(dev, 0);
2828 netdev_unbind_all_sb_channels(dev);
2830 /* Reset TC configuration of device */
2832 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2833 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2835 EXPORT_SYMBOL(netdev_reset_tc);
2837 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2839 if (tc >= dev->num_tc)
2843 netif_reset_xps_queues(dev, offset, count);
2845 dev->tc_to_txq[tc].count = count;
2846 dev->tc_to_txq[tc].offset = offset;
2849 EXPORT_SYMBOL(netdev_set_tc_queue);
2851 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2853 if (num_tc > TC_MAX_QUEUE)
2857 netif_reset_xps_queues_gt(dev, 0);
2859 netdev_unbind_all_sb_channels(dev);
2861 dev->num_tc = num_tc;
2864 EXPORT_SYMBOL(netdev_set_num_tc);
2866 void netdev_unbind_sb_channel(struct net_device *dev,
2867 struct net_device *sb_dev)
2869 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2872 netif_reset_xps_queues_gt(sb_dev, 0);
2874 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2875 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2877 while (txq-- != &dev->_tx[0]) {
2878 if (txq->sb_dev == sb_dev)
2882 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2884 int netdev_bind_sb_channel_queue(struct net_device *dev,
2885 struct net_device *sb_dev,
2886 u8 tc, u16 count, u16 offset)
2888 /* Make certain the sb_dev and dev are already configured */
2889 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2892 /* We cannot hand out queues we don't have */
2893 if ((offset + count) > dev->real_num_tx_queues)
2896 /* Record the mapping */
2897 sb_dev->tc_to_txq[tc].count = count;
2898 sb_dev->tc_to_txq[tc].offset = offset;
2900 /* Provide a way for Tx queue to find the tc_to_txq map or
2901 * XPS map for itself.
2904 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2908 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2910 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2912 /* Do not use a multiqueue device to represent a subordinate channel */
2913 if (netif_is_multiqueue(dev))
2916 /* We allow channels 1 - 32767 to be used for subordinate channels.
2917 * Channel 0 is meant to be "native" mode and used only to represent
2918 * the main root device. We allow writing 0 to reset the device back
2919 * to normal mode after being used as a subordinate channel.
2921 if (channel > S16_MAX)
2924 dev->num_tc = -channel;
2928 EXPORT_SYMBOL(netdev_set_sb_channel);
2931 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2932 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2934 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2939 disabling = txq < dev->real_num_tx_queues;
2941 if (txq < 1 || txq > dev->num_tx_queues)
2944 if (dev->reg_state == NETREG_REGISTERED ||
2945 dev->reg_state == NETREG_UNREGISTERING) {
2948 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2954 netif_setup_tc(dev, txq);
2956 dev->real_num_tx_queues = txq;
2960 qdisc_reset_all_tx_gt(dev, txq);
2962 netif_reset_xps_queues_gt(dev, txq);
2966 dev->real_num_tx_queues = txq;
2971 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2975 * netif_set_real_num_rx_queues - set actual number of RX queues used
2976 * @dev: Network device
2977 * @rxq: Actual number of RX queues
2979 * This must be called either with the rtnl_lock held or before
2980 * registration of the net device. Returns 0 on success, or a
2981 * negative error code. If called before registration, it always
2984 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2988 if (rxq < 1 || rxq > dev->num_rx_queues)
2991 if (dev->reg_state == NETREG_REGISTERED) {
2994 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
3000 dev->real_num_rx_queues = rxq;
3003 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3007 * netif_get_num_default_rss_queues - default number of RSS queues
3009 * This routine should set an upper limit on the number of RSS queues
3010 * used by default by multiqueue devices.
3012 int netif_get_num_default_rss_queues(void)
3014 return is_kdump_kernel() ?
3015 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3017 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3019 static void __netif_reschedule(struct Qdisc *q)
3021 struct softnet_data *sd;
3022 unsigned long flags;
3024 local_irq_save(flags);
3025 sd = this_cpu_ptr(&softnet_data);
3026 q->next_sched = NULL;
3027 *sd->output_queue_tailp = q;
3028 sd->output_queue_tailp = &q->next_sched;
3029 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3030 local_irq_restore(flags);
3033 void __netif_schedule(struct Qdisc *q)
3035 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3036 __netif_reschedule(q);
3038 EXPORT_SYMBOL(__netif_schedule);
3040 struct dev_kfree_skb_cb {
3041 enum skb_free_reason reason;
3044 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3046 return (struct dev_kfree_skb_cb *)skb->cb;
3049 void netif_schedule_queue(struct netdev_queue *txq)
3052 if (!netif_xmit_stopped(txq)) {
3053 struct Qdisc *q = rcu_dereference(txq->qdisc);
3055 __netif_schedule(q);
3059 EXPORT_SYMBOL(netif_schedule_queue);
3061 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3063 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3067 q = rcu_dereference(dev_queue->qdisc);
3068 __netif_schedule(q);
3072 EXPORT_SYMBOL(netif_tx_wake_queue);
3074 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3076 unsigned long flags;
3081 if (likely(refcount_read(&skb->users) == 1)) {
3083 refcount_set(&skb->users, 0);
3084 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3087 get_kfree_skb_cb(skb)->reason = reason;
3088 local_irq_save(flags);
3089 skb->next = __this_cpu_read(softnet_data.completion_queue);
3090 __this_cpu_write(softnet_data.completion_queue, skb);
3091 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3092 local_irq_restore(flags);
3094 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3096 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3098 if (in_irq() || irqs_disabled())
3099 __dev_kfree_skb_irq(skb, reason);
3103 EXPORT_SYMBOL(__dev_kfree_skb_any);
3107 * netif_device_detach - mark device as removed
3108 * @dev: network device
3110 * Mark device as removed from system and therefore no longer available.
3112 void netif_device_detach(struct net_device *dev)
3114 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3115 netif_running(dev)) {
3116 netif_tx_stop_all_queues(dev);
3119 EXPORT_SYMBOL(netif_device_detach);
3122 * netif_device_attach - mark device as attached
3123 * @dev: network device
3125 * Mark device as attached from system and restart if needed.
3127 void netif_device_attach(struct net_device *dev)
3129 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3130 netif_running(dev)) {
3131 netif_tx_wake_all_queues(dev);
3132 __netdev_watchdog_up(dev);
3135 EXPORT_SYMBOL(netif_device_attach);
3138 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3139 * to be used as a distribution range.
3141 static u16 skb_tx_hash(const struct net_device *dev,
3142 const struct net_device *sb_dev,
3143 struct sk_buff *skb)
3147 u16 qcount = dev->real_num_tx_queues;
3150 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3152 qoffset = sb_dev->tc_to_txq[tc].offset;
3153 qcount = sb_dev->tc_to_txq[tc].count;
3156 if (skb_rx_queue_recorded(skb)) {
3157 hash = skb_get_rx_queue(skb);
3158 if (hash >= qoffset)
3160 while (unlikely(hash >= qcount))
3162 return hash + qoffset;
3165 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3168 static void skb_warn_bad_offload(const struct sk_buff *skb)
3170 static const netdev_features_t null_features;
3171 struct net_device *dev = skb->dev;
3172 const char *name = "";
3174 if (!net_ratelimit())
3178 if (dev->dev.parent)
3179 name = dev_driver_string(dev->dev.parent);
3181 name = netdev_name(dev);
3183 skb_dump(KERN_WARNING, skb, false);
3184 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3185 name, dev ? &dev->features : &null_features,
3186 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3190 * Invalidate hardware checksum when packet is to be mangled, and
3191 * complete checksum manually on outgoing path.
3193 int skb_checksum_help(struct sk_buff *skb)
3196 int ret = 0, offset;
3198 if (skb->ip_summed == CHECKSUM_COMPLETE)
3199 goto out_set_summed;
3201 if (unlikely(skb_shinfo(skb)->gso_size)) {
3202 skb_warn_bad_offload(skb);
3206 /* Before computing a checksum, we should make sure no frag could
3207 * be modified by an external entity : checksum could be wrong.
3209 if (skb_has_shared_frag(skb)) {
3210 ret = __skb_linearize(skb);
3215 offset = skb_checksum_start_offset(skb);
3216 BUG_ON(offset >= skb_headlen(skb));
3217 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3219 offset += skb->csum_offset;
3220 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3222 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3226 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3228 skb->ip_summed = CHECKSUM_NONE;
3232 EXPORT_SYMBOL(skb_checksum_help);
3234 int skb_crc32c_csum_help(struct sk_buff *skb)
3237 int ret = 0, offset, start;
3239 if (skb->ip_summed != CHECKSUM_PARTIAL)
3242 if (unlikely(skb_is_gso(skb)))
3245 /* Before computing a checksum, we should make sure no frag could
3246 * be modified by an external entity : checksum could be wrong.
3248 if (unlikely(skb_has_shared_frag(skb))) {
3249 ret = __skb_linearize(skb);
3253 start = skb_checksum_start_offset(skb);
3254 offset = start + offsetof(struct sctphdr, checksum);
3255 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3260 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3264 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3265 skb->len - start, ~(__u32)0,
3267 *(__le32 *)(skb->data + offset) = crc32c_csum;
3268 skb->ip_summed = CHECKSUM_NONE;
3269 skb->csum_not_inet = 0;
3274 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3276 __be16 type = skb->protocol;
3278 /* Tunnel gso handlers can set protocol to ethernet. */
3279 if (type == htons(ETH_P_TEB)) {
3282 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3285 eth = (struct ethhdr *)skb->data;
3286 type = eth->h_proto;
3289 return __vlan_get_protocol(skb, type, depth);
3293 * skb_mac_gso_segment - mac layer segmentation handler.
3294 * @skb: buffer to segment
3295 * @features: features for the output path (see dev->features)
3297 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3298 netdev_features_t features)
3300 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3301 struct packet_offload *ptype;
3302 int vlan_depth = skb->mac_len;
3303 __be16 type = skb_network_protocol(skb, &vlan_depth);
3305 if (unlikely(!type))
3306 return ERR_PTR(-EINVAL);
3308 __skb_pull(skb, vlan_depth);
3311 list_for_each_entry_rcu(ptype, &offload_base, list) {
3312 if (ptype->type == type && ptype->callbacks.gso_segment) {
3313 segs = ptype->callbacks.gso_segment(skb, features);
3319 __skb_push(skb, skb->data - skb_mac_header(skb));
3323 EXPORT_SYMBOL(skb_mac_gso_segment);
3326 /* openvswitch calls this on rx path, so we need a different check.
3328 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3331 return skb->ip_summed != CHECKSUM_PARTIAL &&
3332 skb->ip_summed != CHECKSUM_UNNECESSARY;
3334 return skb->ip_summed == CHECKSUM_NONE;
3338 * __skb_gso_segment - Perform segmentation on skb.
3339 * @skb: buffer to segment
3340 * @features: features for the output path (see dev->features)
3341 * @tx_path: whether it is called in TX path
3343 * This function segments the given skb and returns a list of segments.
3345 * It may return NULL if the skb requires no segmentation. This is
3346 * only possible when GSO is used for verifying header integrity.
3348 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3350 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3351 netdev_features_t features, bool tx_path)
3353 struct sk_buff *segs;
3355 if (unlikely(skb_needs_check(skb, tx_path))) {
3358 /* We're going to init ->check field in TCP or UDP header */
3359 err = skb_cow_head(skb, 0);
3361 return ERR_PTR(err);
3364 /* Only report GSO partial support if it will enable us to
3365 * support segmentation on this frame without needing additional
3368 if (features & NETIF_F_GSO_PARTIAL) {
3369 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3370 struct net_device *dev = skb->dev;
3372 partial_features |= dev->features & dev->gso_partial_features;
3373 if (!skb_gso_ok(skb, features | partial_features))
3374 features &= ~NETIF_F_GSO_PARTIAL;
3377 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3378 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3380 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3381 SKB_GSO_CB(skb)->encap_level = 0;
3383 skb_reset_mac_header(skb);
3384 skb_reset_mac_len(skb);
3386 segs = skb_mac_gso_segment(skb, features);
3388 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3389 skb_warn_bad_offload(skb);
3393 EXPORT_SYMBOL(__skb_gso_segment);
3395 /* Take action when hardware reception checksum errors are detected. */
3397 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3399 if (net_ratelimit()) {
3400 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3401 skb_dump(KERN_ERR, skb, true);
3405 EXPORT_SYMBOL(netdev_rx_csum_fault);
3408 /* XXX: check that highmem exists at all on the given machine. */
3409 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3411 #ifdef CONFIG_HIGHMEM
3414 if (!(dev->features & NETIF_F_HIGHDMA)) {
3415 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3416 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3418 if (PageHighMem(skb_frag_page(frag)))
3426 /* If MPLS offload request, verify we are testing hardware MPLS features
3427 * instead of standard features for the netdev.
3429 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3430 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3431 netdev_features_t features,
3434 if (eth_p_mpls(type))
3435 features &= skb->dev->mpls_features;
3440 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3441 netdev_features_t features,
3448 static netdev_features_t harmonize_features(struct sk_buff *skb,
3449 netdev_features_t features)
3454 type = skb_network_protocol(skb, &tmp);
3455 features = net_mpls_features(skb, features, type);
3457 if (skb->ip_summed != CHECKSUM_NONE &&
3458 !can_checksum_protocol(features, type)) {
3459 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3461 if (illegal_highdma(skb->dev, skb))
3462 features &= ~NETIF_F_SG;
3467 netdev_features_t passthru_features_check(struct sk_buff *skb,
3468 struct net_device *dev,
3469 netdev_features_t features)
3473 EXPORT_SYMBOL(passthru_features_check);
3475 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3476 struct net_device *dev,
3477 netdev_features_t features)
3479 return vlan_features_check(skb, features);
3482 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3483 struct net_device *dev,
3484 netdev_features_t features)
3486 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3488 if (gso_segs > dev->gso_max_segs)
3489 return features & ~NETIF_F_GSO_MASK;
3491 /* Support for GSO partial features requires software
3492 * intervention before we can actually process the packets
3493 * so we need to strip support for any partial features now
3494 * and we can pull them back in after we have partially
3495 * segmented the frame.
3497 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3498 features &= ~dev->gso_partial_features;
3500 /* Make sure to clear the IPv4 ID mangling feature if the
3501 * IPv4 header has the potential to be fragmented.
3503 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3504 struct iphdr *iph = skb->encapsulation ?
3505 inner_ip_hdr(skb) : ip_hdr(skb);
3507 if (!(iph->frag_off & htons(IP_DF)))
3508 features &= ~NETIF_F_TSO_MANGLEID;
3514 netdev_features_t netif_skb_features(struct sk_buff *skb)
3516 struct net_device *dev = skb->dev;
3517 netdev_features_t features = dev->features;
3519 if (skb_is_gso(skb))
3520 features = gso_features_check(skb, dev, features);
3522 /* If encapsulation offload request, verify we are testing
3523 * hardware encapsulation features instead of standard
3524 * features for the netdev
3526 if (skb->encapsulation)
3527 features &= dev->hw_enc_features;
3529 if (skb_vlan_tagged(skb))
3530 features = netdev_intersect_features(features,
3531 dev->vlan_features |
3532 NETIF_F_HW_VLAN_CTAG_TX |
3533 NETIF_F_HW_VLAN_STAG_TX);
3535 if (dev->netdev_ops->ndo_features_check)
3536 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3539 features &= dflt_features_check(skb, dev, features);
3541 return harmonize_features(skb, features);
3543 EXPORT_SYMBOL(netif_skb_features);
3545 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3546 struct netdev_queue *txq, bool more)
3551 if (dev_nit_active(dev))
3552 dev_queue_xmit_nit(skb, dev);
3555 trace_net_dev_start_xmit(skb, dev);
3556 rc = netdev_start_xmit(skb, dev, txq, more);
3557 trace_net_dev_xmit(skb, rc, dev, len);
3562 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3563 struct netdev_queue *txq, int *ret)
3565 struct sk_buff *skb = first;
3566 int rc = NETDEV_TX_OK;
3569 struct sk_buff *next = skb->next;
3571 skb_mark_not_on_list(skb);
3572 rc = xmit_one(skb, dev, txq, next != NULL);
3573 if (unlikely(!dev_xmit_complete(rc))) {
3579 if (netif_tx_queue_stopped(txq) && skb) {
3580 rc = NETDEV_TX_BUSY;
3590 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3591 netdev_features_t features)
3593 if (skb_vlan_tag_present(skb) &&
3594 !vlan_hw_offload_capable(features, skb->vlan_proto))
3595 skb = __vlan_hwaccel_push_inside(skb);
3599 int skb_csum_hwoffload_help(struct sk_buff *skb,
3600 const netdev_features_t features)
3602 if (unlikely(skb->csum_not_inet))
3603 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3604 skb_crc32c_csum_help(skb);
3606 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3608 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3610 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3612 netdev_features_t features;
3614 features = netif_skb_features(skb);
3615 skb = validate_xmit_vlan(skb, features);
3619 skb = sk_validate_xmit_skb(skb, dev);
3623 if (netif_needs_gso(skb, features)) {
3624 struct sk_buff *segs;
3626 segs = skb_gso_segment(skb, features);
3634 if (skb_needs_linearize(skb, features) &&
3635 __skb_linearize(skb))
3638 /* If packet is not checksummed and device does not
3639 * support checksumming for this protocol, complete
3640 * checksumming here.
3642 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3643 if (skb->encapsulation)
3644 skb_set_inner_transport_header(skb,
3645 skb_checksum_start_offset(skb));
3647 skb_set_transport_header(skb,
3648 skb_checksum_start_offset(skb));
3649 if (skb_csum_hwoffload_help(skb, features))
3654 skb = validate_xmit_xfrm(skb, features, again);
3661 atomic_long_inc(&dev->tx_dropped);
3665 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3667 struct sk_buff *next, *head = NULL, *tail;
3669 for (; skb != NULL; skb = next) {
3671 skb_mark_not_on_list(skb);
3673 /* in case skb wont be segmented, point to itself */
3676 skb = validate_xmit_skb(skb, dev, again);
3684 /* If skb was segmented, skb->prev points to
3685 * the last segment. If not, it still contains skb.
3691 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3693 static void qdisc_pkt_len_init(struct sk_buff *skb)
3695 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3697 qdisc_skb_cb(skb)->pkt_len = skb->len;
3699 /* To get more precise estimation of bytes sent on wire,
3700 * we add to pkt_len the headers size of all segments
3702 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3703 unsigned int hdr_len;
3704 u16 gso_segs = shinfo->gso_segs;
3706 /* mac layer + network layer */
3707 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3709 /* + transport layer */
3710 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3711 const struct tcphdr *th;
3712 struct tcphdr _tcphdr;
3714 th = skb_header_pointer(skb, skb_transport_offset(skb),
3715 sizeof(_tcphdr), &_tcphdr);
3717 hdr_len += __tcp_hdrlen(th);
3719 struct udphdr _udphdr;
3721 if (skb_header_pointer(skb, skb_transport_offset(skb),
3722 sizeof(_udphdr), &_udphdr))
3723 hdr_len += sizeof(struct udphdr);
3726 if (shinfo->gso_type & SKB_GSO_DODGY)
3727 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3730 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3734 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3735 struct net_device *dev,
3736 struct netdev_queue *txq)
3738 spinlock_t *root_lock = qdisc_lock(q);
3739 struct sk_buff *to_free = NULL;
3743 qdisc_calculate_pkt_len(skb, q);
3745 if (q->flags & TCQ_F_NOLOCK) {
3746 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3749 if (unlikely(to_free))
3750 kfree_skb_list(to_free);
3755 * Heuristic to force contended enqueues to serialize on a
3756 * separate lock before trying to get qdisc main lock.
3757 * This permits qdisc->running owner to get the lock more
3758 * often and dequeue packets faster.
3760 contended = qdisc_is_running(q);
3761 if (unlikely(contended))
3762 spin_lock(&q->busylock);
3764 spin_lock(root_lock);
3765 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3766 __qdisc_drop(skb, &to_free);
3768 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3769 qdisc_run_begin(q)) {
3771 * This is a work-conserving queue; there are no old skbs
3772 * waiting to be sent out; and the qdisc is not running -
3773 * xmit the skb directly.
3776 qdisc_bstats_update(q, skb);
3778 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3779 if (unlikely(contended)) {
3780 spin_unlock(&q->busylock);
3787 rc = NET_XMIT_SUCCESS;
3789 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3790 if (qdisc_run_begin(q)) {
3791 if (unlikely(contended)) {
3792 spin_unlock(&q->busylock);
3799 spin_unlock(root_lock);
3800 if (unlikely(to_free))
3801 kfree_skb_list(to_free);
3802 if (unlikely(contended))
3803 spin_unlock(&q->busylock);
3807 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3808 static void skb_update_prio(struct sk_buff *skb)
3810 const struct netprio_map *map;
3811 const struct sock *sk;
3812 unsigned int prioidx;
3816 map = rcu_dereference_bh(skb->dev->priomap);
3819 sk = skb_to_full_sk(skb);
3823 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3825 if (prioidx < map->priomap_len)
3826 skb->priority = map->priomap[prioidx];
3829 #define skb_update_prio(skb)
3833 * dev_loopback_xmit - loop back @skb
3834 * @net: network namespace this loopback is happening in
3835 * @sk: sk needed to be a netfilter okfn
3836 * @skb: buffer to transmit
3838 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3840 skb_reset_mac_header(skb);
3841 __skb_pull(skb, skb_network_offset(skb));
3842 skb->pkt_type = PACKET_LOOPBACK;
3843 skb->ip_summed = CHECKSUM_UNNECESSARY;
3844 WARN_ON(!skb_dst(skb));
3849 EXPORT_SYMBOL(dev_loopback_xmit);
3851 #ifdef CONFIG_NET_EGRESS
3852 static struct sk_buff *
3853 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3855 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3856 struct tcf_result cl_res;
3861 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3862 mini_qdisc_bstats_cpu_update(miniq, skb);
3864 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3866 case TC_ACT_RECLASSIFY:
3867 skb->tc_index = TC_H_MIN(cl_res.classid);
3870 mini_qdisc_qstats_cpu_drop(miniq);
3871 *ret = NET_XMIT_DROP;
3877 *ret = NET_XMIT_SUCCESS;
3880 case TC_ACT_REDIRECT:
3881 /* No need to push/pop skb's mac_header here on egress! */
3882 skb_do_redirect(skb);
3883 *ret = NET_XMIT_SUCCESS;
3891 #endif /* CONFIG_NET_EGRESS */
3894 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3895 struct xps_dev_maps *dev_maps, unsigned int tci)
3897 struct xps_map *map;
3898 int queue_index = -1;
3902 tci += netdev_get_prio_tc_map(dev, skb->priority);
3905 map = rcu_dereference(dev_maps->attr_map[tci]);
3908 queue_index = map->queues[0];
3910 queue_index = map->queues[reciprocal_scale(
3911 skb_get_hash(skb), map->len)];
3912 if (unlikely(queue_index >= dev->real_num_tx_queues))
3919 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3920 struct sk_buff *skb)
3923 struct xps_dev_maps *dev_maps;
3924 struct sock *sk = skb->sk;
3925 int queue_index = -1;
3927 if (!static_key_false(&xps_needed))
3931 if (!static_key_false(&xps_rxqs_needed))
3934 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3936 int tci = sk_rx_queue_get(sk);
3938 if (tci >= 0 && tci < dev->num_rx_queues)
3939 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3944 if (queue_index < 0) {
3945 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3947 unsigned int tci = skb->sender_cpu - 1;
3949 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3961 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3962 struct net_device *sb_dev)
3966 EXPORT_SYMBOL(dev_pick_tx_zero);
3968 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3969 struct net_device *sb_dev)
3971 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3973 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3975 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3976 struct net_device *sb_dev)
3978 struct sock *sk = skb->sk;
3979 int queue_index = sk_tx_queue_get(sk);
3981 sb_dev = sb_dev ? : dev;
3983 if (queue_index < 0 || skb->ooo_okay ||
3984 queue_index >= dev->real_num_tx_queues) {
3985 int new_index = get_xps_queue(dev, sb_dev, skb);
3988 new_index = skb_tx_hash(dev, sb_dev, skb);
3990 if (queue_index != new_index && sk &&
3992 rcu_access_pointer(sk->sk_dst_cache))
3993 sk_tx_queue_set(sk, new_index);
3995 queue_index = new_index;
4000 EXPORT_SYMBOL(netdev_pick_tx);
4002 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4003 struct sk_buff *skb,
4004 struct net_device *sb_dev)
4006 int queue_index = 0;
4009 u32 sender_cpu = skb->sender_cpu - 1;
4011 if (sender_cpu >= (u32)NR_CPUS)
4012 skb->sender_cpu = raw_smp_processor_id() + 1;
4015 if (dev->real_num_tx_queues != 1) {
4016 const struct net_device_ops *ops = dev->netdev_ops;
4018 if (ops->ndo_select_queue)
4019 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4021 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4023 queue_index = netdev_cap_txqueue(dev, queue_index);
4026 skb_set_queue_mapping(skb, queue_index);
4027 return netdev_get_tx_queue(dev, queue_index);
4031 * __dev_queue_xmit - transmit a buffer
4032 * @skb: buffer to transmit
4033 * @sb_dev: suboordinate device used for L2 forwarding offload
4035 * Queue a buffer for transmission to a network device. The caller must
4036 * have set the device and priority and built the buffer before calling
4037 * this function. The function can be called from an interrupt.
4039 * A negative errno code is returned on a failure. A success does not
4040 * guarantee the frame will be transmitted as it may be dropped due
4041 * to congestion or traffic shaping.
4043 * -----------------------------------------------------------------------------------
4044 * I notice this method can also return errors from the queue disciplines,
4045 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4048 * Regardless of the return value, the skb is consumed, so it is currently
4049 * difficult to retry a send to this method. (You can bump the ref count
4050 * before sending to hold a reference for retry if you are careful.)
4052 * When calling this method, interrupts MUST be enabled. This is because
4053 * the BH enable code must have IRQs enabled so that it will not deadlock.
4056 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4058 struct net_device *dev = skb->dev;
4059 struct netdev_queue *txq;
4064 skb_reset_mac_header(skb);
4066 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4067 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
4069 /* Disable soft irqs for various locks below. Also
4070 * stops preemption for RCU.
4074 skb_update_prio(skb);
4076 qdisc_pkt_len_init(skb);
4077 #ifdef CONFIG_NET_CLS_ACT
4078 skb->tc_at_ingress = 0;
4079 # ifdef CONFIG_NET_EGRESS
4080 if (static_branch_unlikely(&egress_needed_key)) {
4081 skb = sch_handle_egress(skb, &rc, dev);
4087 /* If device/qdisc don't need skb->dst, release it right now while
4088 * its hot in this cpu cache.
4090 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4095 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4096 q = rcu_dereference_bh(txq->qdisc);
4098 trace_net_dev_queue(skb);
4100 rc = __dev_xmit_skb(skb, q, dev, txq);
4104 /* The device has no queue. Common case for software devices:
4105 * loopback, all the sorts of tunnels...
4107 * Really, it is unlikely that netif_tx_lock protection is necessary
4108 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4110 * However, it is possible, that they rely on protection
4113 * Check this and shot the lock. It is not prone from deadlocks.
4114 *Either shot noqueue qdisc, it is even simpler 8)
4116 if (dev->flags & IFF_UP) {
4117 int cpu = smp_processor_id(); /* ok because BHs are off */
4119 if (txq->xmit_lock_owner != cpu) {
4120 if (dev_xmit_recursion())
4121 goto recursion_alert;
4123 skb = validate_xmit_skb(skb, dev, &again);
4127 HARD_TX_LOCK(dev, txq, cpu);
4129 if (!netif_xmit_stopped(txq)) {
4130 dev_xmit_recursion_inc();
4131 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4132 dev_xmit_recursion_dec();
4133 if (dev_xmit_complete(rc)) {
4134 HARD_TX_UNLOCK(dev, txq);
4138 HARD_TX_UNLOCK(dev, txq);
4139 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4142 /* Recursion is detected! It is possible,
4146 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4152 rcu_read_unlock_bh();
4154 atomic_long_inc(&dev->tx_dropped);
4155 kfree_skb_list(skb);
4158 rcu_read_unlock_bh();
4162 int dev_queue_xmit(struct sk_buff *skb)
4164 return __dev_queue_xmit(skb, NULL);
4166 EXPORT_SYMBOL(dev_queue_xmit);
4168 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4170 return __dev_queue_xmit(skb, sb_dev);
4172 EXPORT_SYMBOL(dev_queue_xmit_accel);
4174 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4176 struct net_device *dev = skb->dev;
4177 struct sk_buff *orig_skb = skb;
4178 struct netdev_queue *txq;
4179 int ret = NETDEV_TX_BUSY;
4182 if (unlikely(!netif_running(dev) ||
4183 !netif_carrier_ok(dev)))
4186 skb = validate_xmit_skb_list(skb, dev, &again);
4187 if (skb != orig_skb)
4190 skb_set_queue_mapping(skb, queue_id);
4191 txq = skb_get_tx_queue(dev, skb);
4195 dev_xmit_recursion_inc();
4196 HARD_TX_LOCK(dev, txq, smp_processor_id());
4197 if (!netif_xmit_frozen_or_drv_stopped(txq))
4198 ret = netdev_start_xmit(skb, dev, txq, false);
4199 HARD_TX_UNLOCK(dev, txq);
4200 dev_xmit_recursion_dec();
4204 if (!dev_xmit_complete(ret))
4209 atomic_long_inc(&dev->tx_dropped);
4210 kfree_skb_list(skb);
4211 return NET_XMIT_DROP;
4213 EXPORT_SYMBOL(dev_direct_xmit);
4215 /*************************************************************************
4217 *************************************************************************/
4219 int netdev_max_backlog __read_mostly = 1000;
4220 EXPORT_SYMBOL(netdev_max_backlog);
4222 int netdev_tstamp_prequeue __read_mostly = 1;
4223 int netdev_budget __read_mostly = 300;
4224 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4225 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4226 int weight_p __read_mostly = 64; /* old backlog weight */
4227 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4228 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4229 int dev_rx_weight __read_mostly = 64;
4230 int dev_tx_weight __read_mostly = 64;
4231 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4232 int gro_normal_batch __read_mostly = 8;
4234 /* Called with irq disabled */
4235 static inline void ____napi_schedule(struct softnet_data *sd,
4236 struct napi_struct *napi)
4238 list_add_tail(&napi->poll_list, &sd->poll_list);
4239 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4244 /* One global table that all flow-based protocols share. */
4245 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4246 EXPORT_SYMBOL(rps_sock_flow_table);
4247 u32 rps_cpu_mask __read_mostly;
4248 EXPORT_SYMBOL(rps_cpu_mask);
4250 struct static_key_false rps_needed __read_mostly;
4251 EXPORT_SYMBOL(rps_needed);
4252 struct static_key_false rfs_needed __read_mostly;
4253 EXPORT_SYMBOL(rfs_needed);
4255 static struct rps_dev_flow *
4256 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4257 struct rps_dev_flow *rflow, u16 next_cpu)
4259 if (next_cpu < nr_cpu_ids) {
4260 #ifdef CONFIG_RFS_ACCEL
4261 struct netdev_rx_queue *rxqueue;
4262 struct rps_dev_flow_table *flow_table;
4263 struct rps_dev_flow *old_rflow;
4268 /* Should we steer this flow to a different hardware queue? */
4269 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4270 !(dev->features & NETIF_F_NTUPLE))
4272 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4273 if (rxq_index == skb_get_rx_queue(skb))
4276 rxqueue = dev->_rx + rxq_index;
4277 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4280 flow_id = skb_get_hash(skb) & flow_table->mask;
4281 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4282 rxq_index, flow_id);
4286 rflow = &flow_table->flows[flow_id];
4288 if (old_rflow->filter == rflow->filter)
4289 old_rflow->filter = RPS_NO_FILTER;
4293 per_cpu(softnet_data, next_cpu).input_queue_head;
4296 rflow->cpu = next_cpu;
4301 * get_rps_cpu is called from netif_receive_skb and returns the target
4302 * CPU from the RPS map of the receiving queue for a given skb.
4303 * rcu_read_lock must be held on entry.
4305 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4306 struct rps_dev_flow **rflowp)
4308 const struct rps_sock_flow_table *sock_flow_table;
4309 struct netdev_rx_queue *rxqueue = dev->_rx;
4310 struct rps_dev_flow_table *flow_table;
4311 struct rps_map *map;
4316 if (skb_rx_queue_recorded(skb)) {
4317 u16 index = skb_get_rx_queue(skb);
4319 if (unlikely(index >= dev->real_num_rx_queues)) {
4320 WARN_ONCE(dev->real_num_rx_queues > 1,
4321 "%s received packet on queue %u, but number "
4322 "of RX queues is %u\n",
4323 dev->name, index, dev->real_num_rx_queues);
4329 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4331 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4332 map = rcu_dereference(rxqueue->rps_map);
4333 if (!flow_table && !map)
4336 skb_reset_network_header(skb);
4337 hash = skb_get_hash(skb);
4341 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4342 if (flow_table && sock_flow_table) {
4343 struct rps_dev_flow *rflow;
4347 /* First check into global flow table if there is a match */
4348 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4349 if ((ident ^ hash) & ~rps_cpu_mask)
4352 next_cpu = ident & rps_cpu_mask;
4354 /* OK, now we know there is a match,
4355 * we can look at the local (per receive queue) flow table
4357 rflow = &flow_table->flows[hash & flow_table->mask];
4361 * If the desired CPU (where last recvmsg was done) is
4362 * different from current CPU (one in the rx-queue flow
4363 * table entry), switch if one of the following holds:
4364 * - Current CPU is unset (>= nr_cpu_ids).
4365 * - Current CPU is offline.
4366 * - The current CPU's queue tail has advanced beyond the
4367 * last packet that was enqueued using this table entry.
4368 * This guarantees that all previous packets for the flow
4369 * have been dequeued, thus preserving in order delivery.
4371 if (unlikely(tcpu != next_cpu) &&
4372 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4373 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4374 rflow->last_qtail)) >= 0)) {
4376 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4379 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4389 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4390 if (cpu_online(tcpu)) {
4400 #ifdef CONFIG_RFS_ACCEL
4403 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4404 * @dev: Device on which the filter was set
4405 * @rxq_index: RX queue index
4406 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4407 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4409 * Drivers that implement ndo_rx_flow_steer() should periodically call
4410 * this function for each installed filter and remove the filters for
4411 * which it returns %true.
4413 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4414 u32 flow_id, u16 filter_id)
4416 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4417 struct rps_dev_flow_table *flow_table;
4418 struct rps_dev_flow *rflow;
4423 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4424 if (flow_table && flow_id <= flow_table->mask) {
4425 rflow = &flow_table->flows[flow_id];
4426 cpu = READ_ONCE(rflow->cpu);
4427 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4428 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4429 rflow->last_qtail) <
4430 (int)(10 * flow_table->mask)))
4436 EXPORT_SYMBOL(rps_may_expire_flow);
4438 #endif /* CONFIG_RFS_ACCEL */
4440 /* Called from hardirq (IPI) context */
4441 static void rps_trigger_softirq(void *data)
4443 struct softnet_data *sd = data;
4445 ____napi_schedule(sd, &sd->backlog);
4449 #endif /* CONFIG_RPS */
4452 * Check if this softnet_data structure is another cpu one
4453 * If yes, queue it to our IPI list and return 1
4456 static int rps_ipi_queued(struct softnet_data *sd)
4459 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4462 sd->rps_ipi_next = mysd->rps_ipi_list;
4463 mysd->rps_ipi_list = sd;
4465 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4468 #endif /* CONFIG_RPS */
4472 #ifdef CONFIG_NET_FLOW_LIMIT
4473 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4476 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4478 #ifdef CONFIG_NET_FLOW_LIMIT
4479 struct sd_flow_limit *fl;
4480 struct softnet_data *sd;
4481 unsigned int old_flow, new_flow;
4483 if (qlen < (netdev_max_backlog >> 1))
4486 sd = this_cpu_ptr(&softnet_data);
4489 fl = rcu_dereference(sd->flow_limit);
4491 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4492 old_flow = fl->history[fl->history_head];
4493 fl->history[fl->history_head] = new_flow;
4496 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4498 if (likely(fl->buckets[old_flow]))
4499 fl->buckets[old_flow]--;
4501 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4513 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4514 * queue (may be a remote CPU queue).
4516 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4517 unsigned int *qtail)
4519 struct softnet_data *sd;
4520 unsigned long flags;
4523 sd = &per_cpu(softnet_data, cpu);
4525 local_irq_save(flags);
4528 if (!netif_running(skb->dev))
4530 qlen = skb_queue_len(&sd->input_pkt_queue);
4531 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4534 __skb_queue_tail(&sd->input_pkt_queue, skb);
4535 input_queue_tail_incr_save(sd, qtail);
4537 local_irq_restore(flags);
4538 return NET_RX_SUCCESS;
4541 /* Schedule NAPI for backlog device
4542 * We can use non atomic operation since we own the queue lock
4544 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4545 if (!rps_ipi_queued(sd))
4546 ____napi_schedule(sd, &sd->backlog);
4555 local_irq_restore(flags);
4557 atomic_long_inc(&skb->dev->rx_dropped);
4562 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4564 struct net_device *dev = skb->dev;
4565 struct netdev_rx_queue *rxqueue;
4569 if (skb_rx_queue_recorded(skb)) {
4570 u16 index = skb_get_rx_queue(skb);
4572 if (unlikely(index >= dev->real_num_rx_queues)) {
4573 WARN_ONCE(dev->real_num_rx_queues > 1,
4574 "%s received packet on queue %u, but number "
4575 "of RX queues is %u\n",
4576 dev->name, index, dev->real_num_rx_queues);
4578 return rxqueue; /* Return first rxqueue */
4585 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4586 struct xdp_buff *xdp,
4587 struct bpf_prog *xdp_prog)
4589 struct netdev_rx_queue *rxqueue;
4590 void *orig_data, *orig_data_end;
4591 u32 metalen, act = XDP_DROP;
4592 __be16 orig_eth_type;
4598 /* Reinjected packets coming from act_mirred or similar should
4599 * not get XDP generic processing.
4601 if (skb_is_redirected(skb))
4604 /* XDP packets must be linear and must have sufficient headroom
4605 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4606 * native XDP provides, thus we need to do it here as well.
4608 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4609 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4610 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4611 int troom = skb->tail + skb->data_len - skb->end;
4613 /* In case we have to go down the path and also linearize,
4614 * then lets do the pskb_expand_head() work just once here.
4616 if (pskb_expand_head(skb,
4617 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4618 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4620 if (skb_linearize(skb))
4624 /* The XDP program wants to see the packet starting at the MAC
4627 mac_len = skb->data - skb_mac_header(skb);
4628 hlen = skb_headlen(skb) + mac_len;
4629 xdp->data = skb->data - mac_len;
4630 xdp->data_meta = xdp->data;
4631 xdp->data_end = xdp->data + hlen;
4632 xdp->data_hard_start = skb->data - skb_headroom(skb);
4634 /* SKB "head" area always have tailroom for skb_shared_info */
4635 xdp->frame_sz = (void *)skb_end_pointer(skb) - xdp->data_hard_start;
4636 xdp->frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4638 orig_data_end = xdp->data_end;
4639 orig_data = xdp->data;
4640 eth = (struct ethhdr *)xdp->data;
4641 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4642 orig_eth_type = eth->h_proto;
4644 rxqueue = netif_get_rxqueue(skb);
4645 xdp->rxq = &rxqueue->xdp_rxq;
4647 act = bpf_prog_run_xdp(xdp_prog, xdp);
4649 /* check if bpf_xdp_adjust_head was used */
4650 off = xdp->data - orig_data;
4653 __skb_pull(skb, off);
4655 __skb_push(skb, -off);
4657 skb->mac_header += off;
4658 skb_reset_network_header(skb);
4661 /* check if bpf_xdp_adjust_tail was used */
4662 off = xdp->data_end - orig_data_end;
4664 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4665 skb->len += off; /* positive on grow, negative on shrink */
4668 /* check if XDP changed eth hdr such SKB needs update */
4669 eth = (struct ethhdr *)xdp->data;
4670 if ((orig_eth_type != eth->h_proto) ||
4671 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4672 __skb_push(skb, ETH_HLEN);
4673 skb->protocol = eth_type_trans(skb, skb->dev);
4679 __skb_push(skb, mac_len);
4682 metalen = xdp->data - xdp->data_meta;
4684 skb_metadata_set(skb, metalen);
4687 bpf_warn_invalid_xdp_action(act);
4690 trace_xdp_exception(skb->dev, xdp_prog, act);
4701 /* When doing generic XDP we have to bypass the qdisc layer and the
4702 * network taps in order to match in-driver-XDP behavior.
4704 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4706 struct net_device *dev = skb->dev;
4707 struct netdev_queue *txq;
4708 bool free_skb = true;
4711 txq = netdev_core_pick_tx(dev, skb, NULL);
4712 cpu = smp_processor_id();
4713 HARD_TX_LOCK(dev, txq, cpu);
4714 if (!netif_xmit_stopped(txq)) {
4715 rc = netdev_start_xmit(skb, dev, txq, 0);
4716 if (dev_xmit_complete(rc))
4719 HARD_TX_UNLOCK(dev, txq);
4721 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4726 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4728 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4731 struct xdp_buff xdp;
4735 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4736 if (act != XDP_PASS) {
4739 err = xdp_do_generic_redirect(skb->dev, skb,
4745 generic_xdp_tx(skb, xdp_prog);
4756 EXPORT_SYMBOL_GPL(do_xdp_generic);
4758 static int netif_rx_internal(struct sk_buff *skb)
4762 net_timestamp_check(netdev_tstamp_prequeue, skb);
4764 trace_netif_rx(skb);
4767 if (static_branch_unlikely(&rps_needed)) {
4768 struct rps_dev_flow voidflow, *rflow = &voidflow;
4774 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4776 cpu = smp_processor_id();
4778 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4787 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4794 * netif_rx - post buffer to the network code
4795 * @skb: buffer to post
4797 * This function receives a packet from a device driver and queues it for
4798 * the upper (protocol) levels to process. It always succeeds. The buffer
4799 * may be dropped during processing for congestion control or by the
4803 * NET_RX_SUCCESS (no congestion)
4804 * NET_RX_DROP (packet was dropped)
4808 int netif_rx(struct sk_buff *skb)
4812 trace_netif_rx_entry(skb);
4814 ret = netif_rx_internal(skb);
4815 trace_netif_rx_exit(ret);
4819 EXPORT_SYMBOL(netif_rx);
4821 int netif_rx_ni(struct sk_buff *skb)
4825 trace_netif_rx_ni_entry(skb);
4828 err = netif_rx_internal(skb);
4829 if (local_softirq_pending())
4832 trace_netif_rx_ni_exit(err);
4836 EXPORT_SYMBOL(netif_rx_ni);
4838 static __latent_entropy void net_tx_action(struct softirq_action *h)
4840 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4842 if (sd->completion_queue) {
4843 struct sk_buff *clist;
4845 local_irq_disable();
4846 clist = sd->completion_queue;
4847 sd->completion_queue = NULL;
4851 struct sk_buff *skb = clist;
4853 clist = clist->next;
4855 WARN_ON(refcount_read(&skb->users));
4856 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4857 trace_consume_skb(skb);
4859 trace_kfree_skb(skb, net_tx_action);
4861 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4864 __kfree_skb_defer(skb);
4867 __kfree_skb_flush();
4870 if (sd->output_queue) {
4873 local_irq_disable();
4874 head = sd->output_queue;
4875 sd->output_queue = NULL;
4876 sd->output_queue_tailp = &sd->output_queue;
4880 struct Qdisc *q = head;
4881 spinlock_t *root_lock = NULL;
4883 head = head->next_sched;
4885 if (!(q->flags & TCQ_F_NOLOCK)) {
4886 root_lock = qdisc_lock(q);
4887 spin_lock(root_lock);
4889 /* We need to make sure head->next_sched is read
4890 * before clearing __QDISC_STATE_SCHED
4892 smp_mb__before_atomic();
4893 clear_bit(__QDISC_STATE_SCHED, &q->state);
4896 spin_unlock(root_lock);
4900 xfrm_dev_backlog(sd);
4903 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4904 /* This hook is defined here for ATM LANE */
4905 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4906 unsigned char *addr) __read_mostly;
4907 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4910 static inline struct sk_buff *
4911 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4912 struct net_device *orig_dev)
4914 #ifdef CONFIG_NET_CLS_ACT
4915 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4916 struct tcf_result cl_res;
4918 /* If there's at least one ingress present somewhere (so
4919 * we get here via enabled static key), remaining devices
4920 * that are not configured with an ingress qdisc will bail
4927 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4931 qdisc_skb_cb(skb)->pkt_len = skb->len;
4932 skb->tc_at_ingress = 1;
4933 mini_qdisc_bstats_cpu_update(miniq, skb);
4935 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list,
4938 case TC_ACT_RECLASSIFY:
4939 skb->tc_index = TC_H_MIN(cl_res.classid);
4942 mini_qdisc_qstats_cpu_drop(miniq);
4950 case TC_ACT_REDIRECT:
4951 /* skb_mac_header check was done by cls/act_bpf, so
4952 * we can safely push the L2 header back before
4953 * redirecting to another netdev
4955 __skb_push(skb, skb->mac_len);
4956 skb_do_redirect(skb);
4958 case TC_ACT_CONSUMED:
4963 #endif /* CONFIG_NET_CLS_ACT */
4968 * netdev_is_rx_handler_busy - check if receive handler is registered
4969 * @dev: device to check
4971 * Check if a receive handler is already registered for a given device.
4972 * Return true if there one.
4974 * The caller must hold the rtnl_mutex.
4976 bool netdev_is_rx_handler_busy(struct net_device *dev)
4979 return dev && rtnl_dereference(dev->rx_handler);
4981 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4984 * netdev_rx_handler_register - register receive handler
4985 * @dev: device to register a handler for
4986 * @rx_handler: receive handler to register
4987 * @rx_handler_data: data pointer that is used by rx handler
4989 * Register a receive handler for a device. This handler will then be
4990 * called from __netif_receive_skb. A negative errno code is returned
4993 * The caller must hold the rtnl_mutex.
4995 * For a general description of rx_handler, see enum rx_handler_result.
4997 int netdev_rx_handler_register(struct net_device *dev,
4998 rx_handler_func_t *rx_handler,
4999 void *rx_handler_data)
5001 if (netdev_is_rx_handler_busy(dev))
5004 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5007 /* Note: rx_handler_data must be set before rx_handler */
5008 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5009 rcu_assign_pointer(dev->rx_handler, rx_handler);
5013 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5016 * netdev_rx_handler_unregister - unregister receive handler
5017 * @dev: device to unregister a handler from
5019 * Unregister a receive handler from a device.
5021 * The caller must hold the rtnl_mutex.
5023 void netdev_rx_handler_unregister(struct net_device *dev)
5027 RCU_INIT_POINTER(dev->rx_handler, NULL);
5028 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5029 * section has a guarantee to see a non NULL rx_handler_data
5033 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5035 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5038 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5039 * the special handling of PFMEMALLOC skbs.
5041 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5043 switch (skb->protocol) {
5044 case htons(ETH_P_ARP):
5045 case htons(ETH_P_IP):
5046 case htons(ETH_P_IPV6):
5047 case htons(ETH_P_8021Q):
5048 case htons(ETH_P_8021AD):
5055 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5056 int *ret, struct net_device *orig_dev)
5058 if (nf_hook_ingress_active(skb)) {
5062 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5067 ingress_retval = nf_hook_ingress(skb);
5069 return ingress_retval;
5074 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5075 struct packet_type **ppt_prev)
5077 struct packet_type *ptype, *pt_prev;
5078 rx_handler_func_t *rx_handler;
5079 struct sk_buff *skb = *pskb;
5080 struct net_device *orig_dev;
5081 bool deliver_exact = false;
5082 int ret = NET_RX_DROP;
5085 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5087 trace_netif_receive_skb(skb);
5089 orig_dev = skb->dev;
5091 skb_reset_network_header(skb);
5092 if (!skb_transport_header_was_set(skb))
5093 skb_reset_transport_header(skb);
5094 skb_reset_mac_len(skb);
5099 skb->skb_iif = skb->dev->ifindex;
5101 __this_cpu_inc(softnet_data.processed);
5103 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5107 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5110 if (ret2 != XDP_PASS) {
5114 skb_reset_mac_len(skb);
5117 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5118 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5119 skb = skb_vlan_untag(skb);
5124 if (skb_skip_tc_classify(skb))
5130 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5132 ret = deliver_skb(skb, pt_prev, orig_dev);
5136 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5138 ret = deliver_skb(skb, pt_prev, orig_dev);
5143 #ifdef CONFIG_NET_INGRESS
5144 if (static_branch_unlikely(&ingress_needed_key)) {
5145 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
5149 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5153 skb_reset_redirect(skb);
5155 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5158 if (skb_vlan_tag_present(skb)) {
5160 ret = deliver_skb(skb, pt_prev, orig_dev);
5163 if (vlan_do_receive(&skb))
5165 else if (unlikely(!skb))
5169 rx_handler = rcu_dereference(skb->dev->rx_handler);
5172 ret = deliver_skb(skb, pt_prev, orig_dev);
5175 switch (rx_handler(&skb)) {
5176 case RX_HANDLER_CONSUMED:
5177 ret = NET_RX_SUCCESS;
5179 case RX_HANDLER_ANOTHER:
5181 case RX_HANDLER_EXACT:
5182 deliver_exact = true;
5183 case RX_HANDLER_PASS:
5190 if (unlikely(skb_vlan_tag_present(skb))) {
5192 if (skb_vlan_tag_get_id(skb)) {
5193 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5196 skb->pkt_type = PACKET_OTHERHOST;
5197 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5198 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5199 /* Outer header is 802.1P with vlan 0, inner header is
5200 * 802.1Q or 802.1AD and vlan_do_receive() above could
5201 * not find vlan dev for vlan id 0.
5203 __vlan_hwaccel_clear_tag(skb);
5204 skb = skb_vlan_untag(skb);
5207 if (vlan_do_receive(&skb))
5208 /* After stripping off 802.1P header with vlan 0
5209 * vlan dev is found for inner header.
5212 else if (unlikely(!skb))
5215 /* We have stripped outer 802.1P vlan 0 header.
5216 * But could not find vlan dev.
5217 * check again for vlan id to set OTHERHOST.
5221 /* Note: we might in the future use prio bits
5222 * and set skb->priority like in vlan_do_receive()
5223 * For the time being, just ignore Priority Code Point
5225 __vlan_hwaccel_clear_tag(skb);
5228 type = skb->protocol;
5230 /* deliver only exact match when indicated */
5231 if (likely(!deliver_exact)) {
5232 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5233 &ptype_base[ntohs(type) &
5237 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5238 &orig_dev->ptype_specific);
5240 if (unlikely(skb->dev != orig_dev)) {
5241 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5242 &skb->dev->ptype_specific);
5246 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5248 *ppt_prev = pt_prev;
5252 atomic_long_inc(&skb->dev->rx_dropped);
5254 atomic_long_inc(&skb->dev->rx_nohandler);
5256 /* Jamal, now you will not able to escape explaining
5257 * me how you were going to use this. :-)
5263 /* The invariant here is that if *ppt_prev is not NULL
5264 * then skb should also be non-NULL.
5266 * Apparently *ppt_prev assignment above holds this invariant due to
5267 * skb dereferencing near it.
5273 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5275 struct net_device *orig_dev = skb->dev;
5276 struct packet_type *pt_prev = NULL;
5279 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5281 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5282 skb->dev, pt_prev, orig_dev);
5287 * netif_receive_skb_core - special purpose version of netif_receive_skb
5288 * @skb: buffer to process
5290 * More direct receive version of netif_receive_skb(). It should
5291 * only be used by callers that have a need to skip RPS and Generic XDP.
5292 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5294 * This function may only be called from softirq context and interrupts
5295 * should be enabled.
5297 * Return values (usually ignored):
5298 * NET_RX_SUCCESS: no congestion
5299 * NET_RX_DROP: packet was dropped
5301 int netif_receive_skb_core(struct sk_buff *skb)
5306 ret = __netif_receive_skb_one_core(skb, false);
5311 EXPORT_SYMBOL(netif_receive_skb_core);
5313 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5314 struct packet_type *pt_prev,
5315 struct net_device *orig_dev)
5317 struct sk_buff *skb, *next;
5321 if (list_empty(head))
5323 if (pt_prev->list_func != NULL)
5324 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5325 ip_list_rcv, head, pt_prev, orig_dev);
5327 list_for_each_entry_safe(skb, next, head, list) {
5328 skb_list_del_init(skb);
5329 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5333 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5335 /* Fast-path assumptions:
5336 * - There is no RX handler.
5337 * - Only one packet_type matches.
5338 * If either of these fails, we will end up doing some per-packet
5339 * processing in-line, then handling the 'last ptype' for the whole
5340 * sublist. This can't cause out-of-order delivery to any single ptype,
5341 * because the 'last ptype' must be constant across the sublist, and all
5342 * other ptypes are handled per-packet.
5344 /* Current (common) ptype of sublist */
5345 struct packet_type *pt_curr = NULL;
5346 /* Current (common) orig_dev of sublist */
5347 struct net_device *od_curr = NULL;
5348 struct list_head sublist;
5349 struct sk_buff *skb, *next;
5351 INIT_LIST_HEAD(&sublist);
5352 list_for_each_entry_safe(skb, next, head, list) {
5353 struct net_device *orig_dev = skb->dev;
5354 struct packet_type *pt_prev = NULL;
5356 skb_list_del_init(skb);
5357 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5360 if (pt_curr != pt_prev || od_curr != orig_dev) {
5361 /* dispatch old sublist */
5362 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5363 /* start new sublist */
5364 INIT_LIST_HEAD(&sublist);
5368 list_add_tail(&skb->list, &sublist);
5371 /* dispatch final sublist */
5372 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5375 static int __netif_receive_skb(struct sk_buff *skb)
5379 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5380 unsigned int noreclaim_flag;
5383 * PFMEMALLOC skbs are special, they should
5384 * - be delivered to SOCK_MEMALLOC sockets only
5385 * - stay away from userspace
5386 * - have bounded memory usage
5388 * Use PF_MEMALLOC as this saves us from propagating the allocation
5389 * context down to all allocation sites.
5391 noreclaim_flag = memalloc_noreclaim_save();
5392 ret = __netif_receive_skb_one_core(skb, true);
5393 memalloc_noreclaim_restore(noreclaim_flag);
5395 ret = __netif_receive_skb_one_core(skb, false);
5400 static void __netif_receive_skb_list(struct list_head *head)
5402 unsigned long noreclaim_flag = 0;
5403 struct sk_buff *skb, *next;
5404 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5406 list_for_each_entry_safe(skb, next, head, list) {
5407 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5408 struct list_head sublist;
5410 /* Handle the previous sublist */
5411 list_cut_before(&sublist, head, &skb->list);
5412 if (!list_empty(&sublist))
5413 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5414 pfmemalloc = !pfmemalloc;
5415 /* See comments in __netif_receive_skb */
5417 noreclaim_flag = memalloc_noreclaim_save();
5419 memalloc_noreclaim_restore(noreclaim_flag);
5422 /* Handle the remaining sublist */
5423 if (!list_empty(head))
5424 __netif_receive_skb_list_core(head, pfmemalloc);
5425 /* Restore pflags */
5427 memalloc_noreclaim_restore(noreclaim_flag);
5430 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5432 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5433 struct bpf_prog *new = xdp->prog;
5439 /* generic XDP does not work with DEVMAPs that can
5440 * have a bpf_prog installed on an entry
5442 for (i = 0; i < new->aux->used_map_cnt; i++) {
5443 if (dev_map_can_have_prog(new->aux->used_maps[i]))
5448 switch (xdp->command) {
5449 case XDP_SETUP_PROG:
5450 rcu_assign_pointer(dev->xdp_prog, new);
5455 static_branch_dec(&generic_xdp_needed_key);
5456 } else if (new && !old) {
5457 static_branch_inc(&generic_xdp_needed_key);
5458 dev_disable_lro(dev);
5459 dev_disable_gro_hw(dev);
5463 case XDP_QUERY_PROG:
5464 xdp->prog_id = old ? old->aux->id : 0;
5475 static int netif_receive_skb_internal(struct sk_buff *skb)
5479 net_timestamp_check(netdev_tstamp_prequeue, skb);
5481 if (skb_defer_rx_timestamp(skb))
5482 return NET_RX_SUCCESS;
5486 if (static_branch_unlikely(&rps_needed)) {
5487 struct rps_dev_flow voidflow, *rflow = &voidflow;
5488 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5491 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5497 ret = __netif_receive_skb(skb);
5502 static void netif_receive_skb_list_internal(struct list_head *head)
5504 struct sk_buff *skb, *next;
5505 struct list_head sublist;
5507 INIT_LIST_HEAD(&sublist);
5508 list_for_each_entry_safe(skb, next, head, list) {
5509 net_timestamp_check(netdev_tstamp_prequeue, skb);
5510 skb_list_del_init(skb);
5511 if (!skb_defer_rx_timestamp(skb))
5512 list_add_tail(&skb->list, &sublist);
5514 list_splice_init(&sublist, head);
5518 if (static_branch_unlikely(&rps_needed)) {
5519 list_for_each_entry_safe(skb, next, head, list) {
5520 struct rps_dev_flow voidflow, *rflow = &voidflow;
5521 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5524 /* Will be handled, remove from list */
5525 skb_list_del_init(skb);
5526 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5531 __netif_receive_skb_list(head);
5536 * netif_receive_skb - process receive buffer from network
5537 * @skb: buffer to process
5539 * netif_receive_skb() is the main receive data processing function.
5540 * It always succeeds. The buffer may be dropped during processing
5541 * for congestion control or by the protocol layers.
5543 * This function may only be called from softirq context and interrupts
5544 * should be enabled.
5546 * Return values (usually ignored):
5547 * NET_RX_SUCCESS: no congestion
5548 * NET_RX_DROP: packet was dropped
5550 int netif_receive_skb(struct sk_buff *skb)
5554 trace_netif_receive_skb_entry(skb);
5556 ret = netif_receive_skb_internal(skb);
5557 trace_netif_receive_skb_exit(ret);
5561 EXPORT_SYMBOL(netif_receive_skb);
5564 * netif_receive_skb_list - process many receive buffers from network
5565 * @head: list of skbs to process.
5567 * Since return value of netif_receive_skb() is normally ignored, and
5568 * wouldn't be meaningful for a list, this function returns void.
5570 * This function may only be called from softirq context and interrupts
5571 * should be enabled.
5573 void netif_receive_skb_list(struct list_head *head)
5575 struct sk_buff *skb;
5577 if (list_empty(head))
5579 if (trace_netif_receive_skb_list_entry_enabled()) {
5580 list_for_each_entry(skb, head, list)
5581 trace_netif_receive_skb_list_entry(skb);
5583 netif_receive_skb_list_internal(head);
5584 trace_netif_receive_skb_list_exit(0);
5586 EXPORT_SYMBOL(netif_receive_skb_list);
5588 DEFINE_PER_CPU(struct work_struct, flush_works);
5590 /* Network device is going away, flush any packets still pending */
5591 static void flush_backlog(struct work_struct *work)
5593 struct sk_buff *skb, *tmp;
5594 struct softnet_data *sd;
5597 sd = this_cpu_ptr(&softnet_data);
5599 local_irq_disable();
5601 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5602 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5603 __skb_unlink(skb, &sd->input_pkt_queue);
5604 dev_kfree_skb_irq(skb);
5605 input_queue_head_incr(sd);
5611 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5612 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5613 __skb_unlink(skb, &sd->process_queue);
5615 input_queue_head_incr(sd);
5621 static void flush_all_backlogs(void)
5627 for_each_online_cpu(cpu)
5628 queue_work_on(cpu, system_highpri_wq,
5629 per_cpu_ptr(&flush_works, cpu));
5631 for_each_online_cpu(cpu)
5632 flush_work(per_cpu_ptr(&flush_works, cpu));
5637 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5638 static void gro_normal_list(struct napi_struct *napi)
5640 if (!napi->rx_count)
5642 netif_receive_skb_list_internal(&napi->rx_list);
5643 INIT_LIST_HEAD(&napi->rx_list);
5647 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5648 * pass the whole batch up to the stack.
5650 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb)
5652 list_add_tail(&skb->list, &napi->rx_list);
5653 if (++napi->rx_count >= gro_normal_batch)
5654 gro_normal_list(napi);
5657 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5658 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5659 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5661 struct packet_offload *ptype;
5662 __be16 type = skb->protocol;
5663 struct list_head *head = &offload_base;
5666 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5668 if (NAPI_GRO_CB(skb)->count == 1) {
5669 skb_shinfo(skb)->gso_size = 0;
5674 list_for_each_entry_rcu(ptype, head, list) {
5675 if (ptype->type != type || !ptype->callbacks.gro_complete)
5678 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5679 ipv6_gro_complete, inet_gro_complete,
5686 WARN_ON(&ptype->list == head);
5688 return NET_RX_SUCCESS;
5692 gro_normal_one(napi, skb);
5693 return NET_RX_SUCCESS;
5696 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5699 struct list_head *head = &napi->gro_hash[index].list;
5700 struct sk_buff *skb, *p;
5702 list_for_each_entry_safe_reverse(skb, p, head, list) {
5703 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5705 skb_list_del_init(skb);
5706 napi_gro_complete(napi, skb);
5707 napi->gro_hash[index].count--;
5710 if (!napi->gro_hash[index].count)
5711 __clear_bit(index, &napi->gro_bitmask);
5714 /* napi->gro_hash[].list contains packets ordered by age.
5715 * youngest packets at the head of it.
5716 * Complete skbs in reverse order to reduce latencies.
5718 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5720 unsigned long bitmask = napi->gro_bitmask;
5721 unsigned int i, base = ~0U;
5723 while ((i = ffs(bitmask)) != 0) {
5726 __napi_gro_flush_chain(napi, base, flush_old);
5729 EXPORT_SYMBOL(napi_gro_flush);
5731 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5732 struct sk_buff *skb)
5734 unsigned int maclen = skb->dev->hard_header_len;
5735 u32 hash = skb_get_hash_raw(skb);
5736 struct list_head *head;
5739 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5740 list_for_each_entry(p, head, list) {
5741 unsigned long diffs;
5743 NAPI_GRO_CB(p)->flush = 0;
5745 if (hash != skb_get_hash_raw(p)) {
5746 NAPI_GRO_CB(p)->same_flow = 0;
5750 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5751 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5752 if (skb_vlan_tag_present(p))
5753 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5754 diffs |= skb_metadata_dst_cmp(p, skb);
5755 diffs |= skb_metadata_differs(p, skb);
5756 if (maclen == ETH_HLEN)
5757 diffs |= compare_ether_header(skb_mac_header(p),
5758 skb_mac_header(skb));
5760 diffs = memcmp(skb_mac_header(p),
5761 skb_mac_header(skb),
5763 NAPI_GRO_CB(p)->same_flow = !diffs;
5769 static void skb_gro_reset_offset(struct sk_buff *skb)
5771 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5772 const skb_frag_t *frag0 = &pinfo->frags[0];
5774 NAPI_GRO_CB(skb)->data_offset = 0;
5775 NAPI_GRO_CB(skb)->frag0 = NULL;
5776 NAPI_GRO_CB(skb)->frag0_len = 0;
5778 if (!skb_headlen(skb) && pinfo->nr_frags &&
5779 !PageHighMem(skb_frag_page(frag0))) {
5780 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5781 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5782 skb_frag_size(frag0),
5783 skb->end - skb->tail);
5787 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5789 struct skb_shared_info *pinfo = skb_shinfo(skb);
5791 BUG_ON(skb->end - skb->tail < grow);
5793 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5795 skb->data_len -= grow;
5798 skb_frag_off_add(&pinfo->frags[0], grow);
5799 skb_frag_size_sub(&pinfo->frags[0], grow);
5801 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5802 skb_frag_unref(skb, 0);
5803 memmove(pinfo->frags, pinfo->frags + 1,
5804 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5808 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
5810 struct sk_buff *oldest;
5812 oldest = list_last_entry(head, struct sk_buff, list);
5814 /* We are called with head length >= MAX_GRO_SKBS, so this is
5817 if (WARN_ON_ONCE(!oldest))
5820 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5823 skb_list_del_init(oldest);
5824 napi_gro_complete(napi, oldest);
5827 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5829 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5831 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5833 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5834 struct list_head *head = &offload_base;
5835 struct packet_offload *ptype;
5836 __be16 type = skb->protocol;
5837 struct list_head *gro_head;
5838 struct sk_buff *pp = NULL;
5839 enum gro_result ret;
5843 if (netif_elide_gro(skb->dev))
5846 gro_head = gro_list_prepare(napi, skb);
5849 list_for_each_entry_rcu(ptype, head, list) {
5850 if (ptype->type != type || !ptype->callbacks.gro_receive)
5853 skb_set_network_header(skb, skb_gro_offset(skb));
5854 skb_reset_mac_len(skb);
5855 NAPI_GRO_CB(skb)->same_flow = 0;
5856 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5857 NAPI_GRO_CB(skb)->free = 0;
5858 NAPI_GRO_CB(skb)->encap_mark = 0;
5859 NAPI_GRO_CB(skb)->recursion_counter = 0;
5860 NAPI_GRO_CB(skb)->is_fou = 0;
5861 NAPI_GRO_CB(skb)->is_atomic = 1;
5862 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5864 /* Setup for GRO checksum validation */
5865 switch (skb->ip_summed) {
5866 case CHECKSUM_COMPLETE:
5867 NAPI_GRO_CB(skb)->csum = skb->csum;
5868 NAPI_GRO_CB(skb)->csum_valid = 1;
5869 NAPI_GRO_CB(skb)->csum_cnt = 0;
5871 case CHECKSUM_UNNECESSARY:
5872 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5873 NAPI_GRO_CB(skb)->csum_valid = 0;
5876 NAPI_GRO_CB(skb)->csum_cnt = 0;
5877 NAPI_GRO_CB(skb)->csum_valid = 0;
5880 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5881 ipv6_gro_receive, inet_gro_receive,
5887 if (&ptype->list == head)
5890 if (PTR_ERR(pp) == -EINPROGRESS) {
5895 same_flow = NAPI_GRO_CB(skb)->same_flow;
5896 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5899 skb_list_del_init(pp);
5900 napi_gro_complete(napi, pp);
5901 napi->gro_hash[hash].count--;
5907 if (NAPI_GRO_CB(skb)->flush)
5910 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5911 gro_flush_oldest(napi, gro_head);
5913 napi->gro_hash[hash].count++;
5915 NAPI_GRO_CB(skb)->count = 1;
5916 NAPI_GRO_CB(skb)->age = jiffies;
5917 NAPI_GRO_CB(skb)->last = skb;
5918 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5919 list_add(&skb->list, gro_head);
5923 grow = skb_gro_offset(skb) - skb_headlen(skb);
5925 gro_pull_from_frag0(skb, grow);
5927 if (napi->gro_hash[hash].count) {
5928 if (!test_bit(hash, &napi->gro_bitmask))
5929 __set_bit(hash, &napi->gro_bitmask);
5930 } else if (test_bit(hash, &napi->gro_bitmask)) {
5931 __clear_bit(hash, &napi->gro_bitmask);
5941 struct packet_offload *gro_find_receive_by_type(__be16 type)
5943 struct list_head *offload_head = &offload_base;
5944 struct packet_offload *ptype;
5946 list_for_each_entry_rcu(ptype, offload_head, list) {
5947 if (ptype->type != type || !ptype->callbacks.gro_receive)
5953 EXPORT_SYMBOL(gro_find_receive_by_type);
5955 struct packet_offload *gro_find_complete_by_type(__be16 type)
5957 struct list_head *offload_head = &offload_base;
5958 struct packet_offload *ptype;
5960 list_for_each_entry_rcu(ptype, offload_head, list) {
5961 if (ptype->type != type || !ptype->callbacks.gro_complete)
5967 EXPORT_SYMBOL(gro_find_complete_by_type);
5969 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5973 kmem_cache_free(skbuff_head_cache, skb);
5976 static gro_result_t napi_skb_finish(struct napi_struct *napi,
5977 struct sk_buff *skb,
5982 gro_normal_one(napi, skb);
5989 case GRO_MERGED_FREE:
5990 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5991 napi_skb_free_stolen_head(skb);
6005 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6009 skb_mark_napi_id(skb, napi);
6010 trace_napi_gro_receive_entry(skb);
6012 skb_gro_reset_offset(skb);
6014 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6015 trace_napi_gro_receive_exit(ret);
6019 EXPORT_SYMBOL(napi_gro_receive);
6021 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6023 if (unlikely(skb->pfmemalloc)) {
6027 __skb_pull(skb, skb_headlen(skb));
6028 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6029 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6030 __vlan_hwaccel_clear_tag(skb);
6031 skb->dev = napi->dev;
6034 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6035 skb->pkt_type = PACKET_HOST;
6037 skb->encapsulation = 0;
6038 skb_shinfo(skb)->gso_type = 0;
6039 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6045 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6047 struct sk_buff *skb = napi->skb;
6050 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6053 skb_mark_napi_id(skb, napi);
6058 EXPORT_SYMBOL(napi_get_frags);
6060 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6061 struct sk_buff *skb,
6067 __skb_push(skb, ETH_HLEN);
6068 skb->protocol = eth_type_trans(skb, skb->dev);
6069 if (ret == GRO_NORMAL)
6070 gro_normal_one(napi, skb);
6074 napi_reuse_skb(napi, skb);
6077 case GRO_MERGED_FREE:
6078 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6079 napi_skb_free_stolen_head(skb);
6081 napi_reuse_skb(napi, skb);
6092 /* Upper GRO stack assumes network header starts at gro_offset=0
6093 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6094 * We copy ethernet header into skb->data to have a common layout.
6096 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6098 struct sk_buff *skb = napi->skb;
6099 const struct ethhdr *eth;
6100 unsigned int hlen = sizeof(*eth);
6104 skb_reset_mac_header(skb);
6105 skb_gro_reset_offset(skb);
6107 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6108 eth = skb_gro_header_slow(skb, hlen, 0);
6109 if (unlikely(!eth)) {
6110 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6111 __func__, napi->dev->name);
6112 napi_reuse_skb(napi, skb);
6116 eth = (const struct ethhdr *)skb->data;
6117 gro_pull_from_frag0(skb, hlen);
6118 NAPI_GRO_CB(skb)->frag0 += hlen;
6119 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6121 __skb_pull(skb, hlen);
6124 * This works because the only protocols we care about don't require
6126 * We'll fix it up properly in napi_frags_finish()
6128 skb->protocol = eth->h_proto;
6133 gro_result_t napi_gro_frags(struct napi_struct *napi)
6136 struct sk_buff *skb = napi_frags_skb(napi);
6141 trace_napi_gro_frags_entry(skb);
6143 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6144 trace_napi_gro_frags_exit(ret);
6148 EXPORT_SYMBOL(napi_gro_frags);
6150 /* Compute the checksum from gro_offset and return the folded value
6151 * after adding in any pseudo checksum.
6153 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6158 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6160 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6161 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6162 /* See comments in __skb_checksum_complete(). */
6164 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6165 !skb->csum_complete_sw)
6166 netdev_rx_csum_fault(skb->dev, skb);
6169 NAPI_GRO_CB(skb)->csum = wsum;
6170 NAPI_GRO_CB(skb)->csum_valid = 1;
6174 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6176 static void net_rps_send_ipi(struct softnet_data *remsd)
6180 struct softnet_data *next = remsd->rps_ipi_next;
6182 if (cpu_online(remsd->cpu))
6183 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6190 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6191 * Note: called with local irq disabled, but exits with local irq enabled.
6193 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6196 struct softnet_data *remsd = sd->rps_ipi_list;
6199 sd->rps_ipi_list = NULL;
6203 /* Send pending IPI's to kick RPS processing on remote cpus. */
6204 net_rps_send_ipi(remsd);
6210 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6213 return sd->rps_ipi_list != NULL;
6219 static int process_backlog(struct napi_struct *napi, int quota)
6221 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6225 /* Check if we have pending ipi, its better to send them now,
6226 * not waiting net_rx_action() end.
6228 if (sd_has_rps_ipi_waiting(sd)) {
6229 local_irq_disable();
6230 net_rps_action_and_irq_enable(sd);
6233 napi->weight = dev_rx_weight;
6235 struct sk_buff *skb;
6237 while ((skb = __skb_dequeue(&sd->process_queue))) {
6239 __netif_receive_skb(skb);
6241 input_queue_head_incr(sd);
6242 if (++work >= quota)
6247 local_irq_disable();
6249 if (skb_queue_empty(&sd->input_pkt_queue)) {
6251 * Inline a custom version of __napi_complete().
6252 * only current cpu owns and manipulates this napi,
6253 * and NAPI_STATE_SCHED is the only possible flag set
6255 * We can use a plain write instead of clear_bit(),
6256 * and we dont need an smp_mb() memory barrier.
6261 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6262 &sd->process_queue);
6272 * __napi_schedule - schedule for receive
6273 * @n: entry to schedule
6275 * The entry's receive function will be scheduled to run.
6276 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6278 void __napi_schedule(struct napi_struct *n)
6280 unsigned long flags;
6282 local_irq_save(flags);
6283 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6284 local_irq_restore(flags);
6286 EXPORT_SYMBOL(__napi_schedule);
6289 * napi_schedule_prep - check if napi can be scheduled
6292 * Test if NAPI routine is already running, and if not mark
6293 * it as running. This is used as a condition variable
6294 * insure only one NAPI poll instance runs. We also make
6295 * sure there is no pending NAPI disable.
6297 bool napi_schedule_prep(struct napi_struct *n)
6299 unsigned long val, new;
6302 val = READ_ONCE(n->state);
6303 if (unlikely(val & NAPIF_STATE_DISABLE))
6305 new = val | NAPIF_STATE_SCHED;
6307 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6308 * This was suggested by Alexander Duyck, as compiler
6309 * emits better code than :
6310 * if (val & NAPIF_STATE_SCHED)
6311 * new |= NAPIF_STATE_MISSED;
6313 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6315 } while (cmpxchg(&n->state, val, new) != val);
6317 return !(val & NAPIF_STATE_SCHED);
6319 EXPORT_SYMBOL(napi_schedule_prep);
6322 * __napi_schedule_irqoff - schedule for receive
6323 * @n: entry to schedule
6325 * Variant of __napi_schedule() assuming hard irqs are masked
6327 void __napi_schedule_irqoff(struct napi_struct *n)
6329 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6331 EXPORT_SYMBOL(__napi_schedule_irqoff);
6333 bool napi_complete_done(struct napi_struct *n, int work_done)
6335 unsigned long flags, val, new, timeout = 0;
6339 * 1) Don't let napi dequeue from the cpu poll list
6340 * just in case its running on a different cpu.
6341 * 2) If we are busy polling, do nothing here, we have
6342 * the guarantee we will be called later.
6344 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6345 NAPIF_STATE_IN_BUSY_POLL)))
6350 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6351 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6353 if (n->defer_hard_irqs_count > 0) {
6354 n->defer_hard_irqs_count--;
6355 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6359 if (n->gro_bitmask) {
6360 /* When the NAPI instance uses a timeout and keeps postponing
6361 * it, we need to bound somehow the time packets are kept in
6364 napi_gro_flush(n, !!timeout);
6369 if (unlikely(!list_empty(&n->poll_list))) {
6370 /* If n->poll_list is not empty, we need to mask irqs */
6371 local_irq_save(flags);
6372 list_del_init(&n->poll_list);
6373 local_irq_restore(flags);
6377 val = READ_ONCE(n->state);
6379 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6381 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6383 /* If STATE_MISSED was set, leave STATE_SCHED set,
6384 * because we will call napi->poll() one more time.
6385 * This C code was suggested by Alexander Duyck to help gcc.
6387 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6389 } while (cmpxchg(&n->state, val, new) != val);
6391 if (unlikely(val & NAPIF_STATE_MISSED)) {
6397 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6398 HRTIMER_MODE_REL_PINNED);
6401 EXPORT_SYMBOL(napi_complete_done);
6403 /* must be called under rcu_read_lock(), as we dont take a reference */
6404 static struct napi_struct *napi_by_id(unsigned int napi_id)
6406 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6407 struct napi_struct *napi;
6409 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6410 if (napi->napi_id == napi_id)
6416 #if defined(CONFIG_NET_RX_BUSY_POLL)
6418 #define BUSY_POLL_BUDGET 8
6420 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6424 /* Busy polling means there is a high chance device driver hard irq
6425 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6426 * set in napi_schedule_prep().
6427 * Since we are about to call napi->poll() once more, we can safely
6428 * clear NAPI_STATE_MISSED.
6430 * Note: x86 could use a single "lock and ..." instruction
6431 * to perform these two clear_bit()
6433 clear_bit(NAPI_STATE_MISSED, &napi->state);
6434 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6438 /* All we really want here is to re-enable device interrupts.
6439 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6441 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6442 /* We can't gro_normal_list() here, because napi->poll() might have
6443 * rearmed the napi (napi_complete_done()) in which case it could
6444 * already be running on another CPU.
6446 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6447 netpoll_poll_unlock(have_poll_lock);
6448 if (rc == BUSY_POLL_BUDGET) {
6449 /* As the whole budget was spent, we still own the napi so can
6450 * safely handle the rx_list.
6452 gro_normal_list(napi);
6453 __napi_schedule(napi);
6458 void napi_busy_loop(unsigned int napi_id,
6459 bool (*loop_end)(void *, unsigned long),
6462 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6463 int (*napi_poll)(struct napi_struct *napi, int budget);
6464 void *have_poll_lock = NULL;
6465 struct napi_struct *napi;
6472 napi = napi_by_id(napi_id);
6482 unsigned long val = READ_ONCE(napi->state);
6484 /* If multiple threads are competing for this napi,
6485 * we avoid dirtying napi->state as much as we can.
6487 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6488 NAPIF_STATE_IN_BUSY_POLL))
6490 if (cmpxchg(&napi->state, val,
6491 val | NAPIF_STATE_IN_BUSY_POLL |
6492 NAPIF_STATE_SCHED) != val)
6494 have_poll_lock = netpoll_poll_lock(napi);
6495 napi_poll = napi->poll;
6497 work = napi_poll(napi, BUSY_POLL_BUDGET);
6498 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6499 gro_normal_list(napi);
6502 __NET_ADD_STATS(dev_net(napi->dev),
6503 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6506 if (!loop_end || loop_end(loop_end_arg, start_time))
6509 if (unlikely(need_resched())) {
6511 busy_poll_stop(napi, have_poll_lock);
6515 if (loop_end(loop_end_arg, start_time))
6522 busy_poll_stop(napi, have_poll_lock);
6527 EXPORT_SYMBOL(napi_busy_loop);
6529 #endif /* CONFIG_NET_RX_BUSY_POLL */
6531 static void napi_hash_add(struct napi_struct *napi)
6533 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
6534 test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
6537 spin_lock(&napi_hash_lock);
6539 /* 0..NR_CPUS range is reserved for sender_cpu use */
6541 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6542 napi_gen_id = MIN_NAPI_ID;
6543 } while (napi_by_id(napi_gen_id));
6544 napi->napi_id = napi_gen_id;
6546 hlist_add_head_rcu(&napi->napi_hash_node,
6547 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6549 spin_unlock(&napi_hash_lock);
6552 /* Warning : caller is responsible to make sure rcu grace period
6553 * is respected before freeing memory containing @napi
6555 bool napi_hash_del(struct napi_struct *napi)
6557 bool rcu_sync_needed = false;
6559 spin_lock(&napi_hash_lock);
6561 if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
6562 rcu_sync_needed = true;
6563 hlist_del_rcu(&napi->napi_hash_node);
6565 spin_unlock(&napi_hash_lock);
6566 return rcu_sync_needed;
6568 EXPORT_SYMBOL_GPL(napi_hash_del);
6570 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6572 struct napi_struct *napi;
6574 napi = container_of(timer, struct napi_struct, timer);
6576 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6577 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6579 if (!napi_disable_pending(napi) &&
6580 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6581 __napi_schedule_irqoff(napi);
6583 return HRTIMER_NORESTART;
6586 static void init_gro_hash(struct napi_struct *napi)
6590 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6591 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6592 napi->gro_hash[i].count = 0;
6594 napi->gro_bitmask = 0;
6597 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6598 int (*poll)(struct napi_struct *, int), int weight)
6600 INIT_LIST_HEAD(&napi->poll_list);
6601 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6602 napi->timer.function = napi_watchdog;
6603 init_gro_hash(napi);
6605 INIT_LIST_HEAD(&napi->rx_list);
6608 if (weight > NAPI_POLL_WEIGHT)
6609 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6611 napi->weight = weight;
6612 list_add(&napi->dev_list, &dev->napi_list);
6614 #ifdef CONFIG_NETPOLL
6615 napi->poll_owner = -1;
6617 set_bit(NAPI_STATE_SCHED, &napi->state);
6618 napi_hash_add(napi);
6620 EXPORT_SYMBOL(netif_napi_add);
6622 void napi_disable(struct napi_struct *n)
6625 set_bit(NAPI_STATE_DISABLE, &n->state);
6627 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6629 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6632 hrtimer_cancel(&n->timer);
6634 clear_bit(NAPI_STATE_DISABLE, &n->state);
6636 EXPORT_SYMBOL(napi_disable);
6638 static void flush_gro_hash(struct napi_struct *napi)
6642 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6643 struct sk_buff *skb, *n;
6645 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6647 napi->gro_hash[i].count = 0;
6651 /* Must be called in process context */
6652 void netif_napi_del(struct napi_struct *napi)
6655 if (napi_hash_del(napi))
6657 list_del_init(&napi->dev_list);
6658 napi_free_frags(napi);
6660 flush_gro_hash(napi);
6661 napi->gro_bitmask = 0;
6663 EXPORT_SYMBOL(netif_napi_del);
6665 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6670 list_del_init(&n->poll_list);
6672 have = netpoll_poll_lock(n);
6676 /* This NAPI_STATE_SCHED test is for avoiding a race
6677 * with netpoll's poll_napi(). Only the entity which
6678 * obtains the lock and sees NAPI_STATE_SCHED set will
6679 * actually make the ->poll() call. Therefore we avoid
6680 * accidentally calling ->poll() when NAPI is not scheduled.
6683 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6684 work = n->poll(n, weight);
6685 trace_napi_poll(n, work, weight);
6688 WARN_ON_ONCE(work > weight);
6690 if (likely(work < weight))
6693 /* Drivers must not modify the NAPI state if they
6694 * consume the entire weight. In such cases this code
6695 * still "owns" the NAPI instance and therefore can
6696 * move the instance around on the list at-will.
6698 if (unlikely(napi_disable_pending(n))) {
6703 if (n->gro_bitmask) {
6704 /* flush too old packets
6705 * If HZ < 1000, flush all packets.
6707 napi_gro_flush(n, HZ >= 1000);
6712 /* Some drivers may have called napi_schedule
6713 * prior to exhausting their budget.
6715 if (unlikely(!list_empty(&n->poll_list))) {
6716 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6717 n->dev ? n->dev->name : "backlog");
6721 list_add_tail(&n->poll_list, repoll);
6724 netpoll_poll_unlock(have);
6729 static __latent_entropy void net_rx_action(struct softirq_action *h)
6731 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6732 unsigned long time_limit = jiffies +
6733 usecs_to_jiffies(netdev_budget_usecs);
6734 int budget = netdev_budget;
6738 local_irq_disable();
6739 list_splice_init(&sd->poll_list, &list);
6743 struct napi_struct *n;
6745 if (list_empty(&list)) {
6746 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6751 n = list_first_entry(&list, struct napi_struct, poll_list);
6752 budget -= napi_poll(n, &repoll);
6754 /* If softirq window is exhausted then punt.
6755 * Allow this to run for 2 jiffies since which will allow
6756 * an average latency of 1.5/HZ.
6758 if (unlikely(budget <= 0 ||
6759 time_after_eq(jiffies, time_limit))) {
6765 local_irq_disable();
6767 list_splice_tail_init(&sd->poll_list, &list);
6768 list_splice_tail(&repoll, &list);
6769 list_splice(&list, &sd->poll_list);
6770 if (!list_empty(&sd->poll_list))
6771 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6773 net_rps_action_and_irq_enable(sd);
6775 __kfree_skb_flush();
6778 struct netdev_adjacent {
6779 struct net_device *dev;
6781 /* upper master flag, there can only be one master device per list */
6784 /* lookup ignore flag */
6787 /* counter for the number of times this device was added to us */
6790 /* private field for the users */
6793 struct list_head list;
6794 struct rcu_head rcu;
6797 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6798 struct list_head *adj_list)
6800 struct netdev_adjacent *adj;
6802 list_for_each_entry(adj, adj_list, list) {
6803 if (adj->dev == adj_dev)
6809 static int ____netdev_has_upper_dev(struct net_device *upper_dev, void *data)
6811 struct net_device *dev = data;
6813 return upper_dev == dev;
6817 * netdev_has_upper_dev - Check if device is linked to an upper device
6819 * @upper_dev: upper device to check
6821 * Find out if a device is linked to specified upper device and return true
6822 * in case it is. Note that this checks only immediate upper device,
6823 * not through a complete stack of devices. The caller must hold the RTNL lock.
6825 bool netdev_has_upper_dev(struct net_device *dev,
6826 struct net_device *upper_dev)
6830 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6833 EXPORT_SYMBOL(netdev_has_upper_dev);
6836 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6838 * @upper_dev: upper device to check
6840 * Find out if a device is linked to specified upper device and return true
6841 * in case it is. Note that this checks the entire upper device chain.
6842 * The caller must hold rcu lock.
6845 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6846 struct net_device *upper_dev)
6848 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6851 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6854 * netdev_has_any_upper_dev - Check if device is linked to some device
6857 * Find out if a device is linked to an upper device and return true in case
6858 * it is. The caller must hold the RTNL lock.
6860 bool netdev_has_any_upper_dev(struct net_device *dev)
6864 return !list_empty(&dev->adj_list.upper);
6866 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6869 * netdev_master_upper_dev_get - Get master upper device
6872 * Find a master upper device and return pointer to it or NULL in case
6873 * it's not there. The caller must hold the RTNL lock.
6875 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6877 struct netdev_adjacent *upper;
6881 if (list_empty(&dev->adj_list.upper))
6884 upper = list_first_entry(&dev->adj_list.upper,
6885 struct netdev_adjacent, list);
6886 if (likely(upper->master))
6890 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6892 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6894 struct netdev_adjacent *upper;
6898 if (list_empty(&dev->adj_list.upper))
6901 upper = list_first_entry(&dev->adj_list.upper,
6902 struct netdev_adjacent, list);
6903 if (likely(upper->master) && !upper->ignore)
6909 * netdev_has_any_lower_dev - Check if device is linked to some device
6912 * Find out if a device is linked to a lower device and return true in case
6913 * it is. The caller must hold the RTNL lock.
6915 static bool netdev_has_any_lower_dev(struct net_device *dev)
6919 return !list_empty(&dev->adj_list.lower);
6922 void *netdev_adjacent_get_private(struct list_head *adj_list)
6924 struct netdev_adjacent *adj;
6926 adj = list_entry(adj_list, struct netdev_adjacent, list);
6928 return adj->private;
6930 EXPORT_SYMBOL(netdev_adjacent_get_private);
6933 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6935 * @iter: list_head ** of the current position
6937 * Gets the next device from the dev's upper list, starting from iter
6938 * position. The caller must hold RCU read lock.
6940 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6941 struct list_head **iter)
6943 struct netdev_adjacent *upper;
6945 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6947 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6949 if (&upper->list == &dev->adj_list.upper)
6952 *iter = &upper->list;
6956 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6958 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
6959 struct list_head **iter,
6962 struct netdev_adjacent *upper;
6964 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
6966 if (&upper->list == &dev->adj_list.upper)
6969 *iter = &upper->list;
6970 *ignore = upper->ignore;
6975 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6976 struct list_head **iter)
6978 struct netdev_adjacent *upper;
6980 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6982 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6984 if (&upper->list == &dev->adj_list.upper)
6987 *iter = &upper->list;
6992 static int __netdev_walk_all_upper_dev(struct net_device *dev,
6993 int (*fn)(struct net_device *dev,
6997 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
6998 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7003 iter = &dev->adj_list.upper;
7007 ret = fn(now, data);
7014 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7021 niter = &udev->adj_list.upper;
7022 dev_stack[cur] = now;
7023 iter_stack[cur++] = iter;
7030 next = dev_stack[--cur];
7031 niter = iter_stack[cur];
7041 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7042 int (*fn)(struct net_device *dev,
7046 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7047 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7051 iter = &dev->adj_list.upper;
7055 ret = fn(now, data);
7062 udev = netdev_next_upper_dev_rcu(now, &iter);
7067 niter = &udev->adj_list.upper;
7068 dev_stack[cur] = now;
7069 iter_stack[cur++] = iter;
7076 next = dev_stack[--cur];
7077 niter = iter_stack[cur];
7086 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7088 static bool __netdev_has_upper_dev(struct net_device *dev,
7089 struct net_device *upper_dev)
7093 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7098 * netdev_lower_get_next_private - Get the next ->private from the
7099 * lower neighbour list
7101 * @iter: list_head ** of the current position
7103 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7104 * list, starting from iter position. The caller must hold either hold the
7105 * RTNL lock or its own locking that guarantees that the neighbour lower
7106 * list will remain unchanged.
7108 void *netdev_lower_get_next_private(struct net_device *dev,
7109 struct list_head **iter)
7111 struct netdev_adjacent *lower;
7113 lower = list_entry(*iter, struct netdev_adjacent, list);
7115 if (&lower->list == &dev->adj_list.lower)
7118 *iter = lower->list.next;
7120 return lower->private;
7122 EXPORT_SYMBOL(netdev_lower_get_next_private);
7125 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7126 * lower neighbour list, RCU
7129 * @iter: list_head ** of the current position
7131 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7132 * list, starting from iter position. The caller must hold RCU read lock.
7134 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7135 struct list_head **iter)
7137 struct netdev_adjacent *lower;
7139 WARN_ON_ONCE(!rcu_read_lock_held());
7141 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7143 if (&lower->list == &dev->adj_list.lower)
7146 *iter = &lower->list;
7148 return lower->private;
7150 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7153 * netdev_lower_get_next - Get the next device from the lower neighbour
7156 * @iter: list_head ** of the current position
7158 * Gets the next netdev_adjacent from the dev's lower neighbour
7159 * list, starting from iter position. The caller must hold RTNL lock or
7160 * its own locking that guarantees that the neighbour lower
7161 * list will remain unchanged.
7163 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7165 struct netdev_adjacent *lower;
7167 lower = list_entry(*iter, struct netdev_adjacent, list);
7169 if (&lower->list == &dev->adj_list.lower)
7172 *iter = lower->list.next;
7176 EXPORT_SYMBOL(netdev_lower_get_next);
7178 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7179 struct list_head **iter)
7181 struct netdev_adjacent *lower;
7183 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7185 if (&lower->list == &dev->adj_list.lower)
7188 *iter = &lower->list;
7193 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7194 struct list_head **iter,
7197 struct netdev_adjacent *lower;
7199 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7201 if (&lower->list == &dev->adj_list.lower)
7204 *iter = &lower->list;
7205 *ignore = lower->ignore;
7210 int netdev_walk_all_lower_dev(struct net_device *dev,
7211 int (*fn)(struct net_device *dev,
7215 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7216 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7220 iter = &dev->adj_list.lower;
7224 ret = fn(now, data);
7231 ldev = netdev_next_lower_dev(now, &iter);
7236 niter = &ldev->adj_list.lower;
7237 dev_stack[cur] = now;
7238 iter_stack[cur++] = iter;
7245 next = dev_stack[--cur];
7246 niter = iter_stack[cur];
7255 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7257 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7258 int (*fn)(struct net_device *dev,
7262 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7263 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7268 iter = &dev->adj_list.lower;
7272 ret = fn(now, data);
7279 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7286 niter = &ldev->adj_list.lower;
7287 dev_stack[cur] = now;
7288 iter_stack[cur++] = iter;
7295 next = dev_stack[--cur];
7296 niter = iter_stack[cur];
7306 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7307 struct list_head **iter)
7309 struct netdev_adjacent *lower;
7311 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7312 if (&lower->list == &dev->adj_list.lower)
7315 *iter = &lower->list;
7319 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7321 static u8 __netdev_upper_depth(struct net_device *dev)
7323 struct net_device *udev;
7324 struct list_head *iter;
7328 for (iter = &dev->adj_list.upper,
7329 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7331 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7334 if (max_depth < udev->upper_level)
7335 max_depth = udev->upper_level;
7341 static u8 __netdev_lower_depth(struct net_device *dev)
7343 struct net_device *ldev;
7344 struct list_head *iter;
7348 for (iter = &dev->adj_list.lower,
7349 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7351 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7354 if (max_depth < ldev->lower_level)
7355 max_depth = ldev->lower_level;
7361 static int __netdev_update_upper_level(struct net_device *dev, void *data)
7363 dev->upper_level = __netdev_upper_depth(dev) + 1;
7367 static int __netdev_update_lower_level(struct net_device *dev, void *data)
7369 dev->lower_level = __netdev_lower_depth(dev) + 1;
7373 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7374 int (*fn)(struct net_device *dev,
7378 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7379 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7383 iter = &dev->adj_list.lower;
7387 ret = fn(now, data);
7394 ldev = netdev_next_lower_dev_rcu(now, &iter);
7399 niter = &ldev->adj_list.lower;
7400 dev_stack[cur] = now;
7401 iter_stack[cur++] = iter;
7408 next = dev_stack[--cur];
7409 niter = iter_stack[cur];
7418 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7421 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7422 * lower neighbour list, RCU
7426 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7427 * list. The caller must hold RCU read lock.
7429 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7431 struct netdev_adjacent *lower;
7433 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7434 struct netdev_adjacent, list);
7436 return lower->private;
7439 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7442 * netdev_master_upper_dev_get_rcu - Get master upper device
7445 * Find a master upper device and return pointer to it or NULL in case
7446 * it's not there. The caller must hold the RCU read lock.
7448 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7450 struct netdev_adjacent *upper;
7452 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7453 struct netdev_adjacent, list);
7454 if (upper && likely(upper->master))
7458 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7460 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7461 struct net_device *adj_dev,
7462 struct list_head *dev_list)
7464 char linkname[IFNAMSIZ+7];
7466 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7467 "upper_%s" : "lower_%s", adj_dev->name);
7468 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7471 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7473 struct list_head *dev_list)
7475 char linkname[IFNAMSIZ+7];
7477 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7478 "upper_%s" : "lower_%s", name);
7479 sysfs_remove_link(&(dev->dev.kobj), linkname);
7482 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7483 struct net_device *adj_dev,
7484 struct list_head *dev_list)
7486 return (dev_list == &dev->adj_list.upper ||
7487 dev_list == &dev->adj_list.lower) &&
7488 net_eq(dev_net(dev), dev_net(adj_dev));
7491 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7492 struct net_device *adj_dev,
7493 struct list_head *dev_list,
7494 void *private, bool master)
7496 struct netdev_adjacent *adj;
7499 adj = __netdev_find_adj(adj_dev, dev_list);
7503 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7504 dev->name, adj_dev->name, adj->ref_nr);
7509 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7514 adj->master = master;
7516 adj->private = private;
7517 adj->ignore = false;
7520 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7521 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7523 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7524 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7529 /* Ensure that master link is always the first item in list. */
7531 ret = sysfs_create_link(&(dev->dev.kobj),
7532 &(adj_dev->dev.kobj), "master");
7534 goto remove_symlinks;
7536 list_add_rcu(&adj->list, dev_list);
7538 list_add_tail_rcu(&adj->list, dev_list);
7544 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7545 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7553 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7554 struct net_device *adj_dev,
7556 struct list_head *dev_list)
7558 struct netdev_adjacent *adj;
7560 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7561 dev->name, adj_dev->name, ref_nr);
7563 adj = __netdev_find_adj(adj_dev, dev_list);
7566 pr_err("Adjacency does not exist for device %s from %s\n",
7567 dev->name, adj_dev->name);
7572 if (adj->ref_nr > ref_nr) {
7573 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7574 dev->name, adj_dev->name, ref_nr,
7575 adj->ref_nr - ref_nr);
7576 adj->ref_nr -= ref_nr;
7581 sysfs_remove_link(&(dev->dev.kobj), "master");
7583 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7584 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7586 list_del_rcu(&adj->list);
7587 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7588 adj_dev->name, dev->name, adj_dev->name);
7590 kfree_rcu(adj, rcu);
7593 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7594 struct net_device *upper_dev,
7595 struct list_head *up_list,
7596 struct list_head *down_list,
7597 void *private, bool master)
7601 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7606 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7609 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7616 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7617 struct net_device *upper_dev,
7619 struct list_head *up_list,
7620 struct list_head *down_list)
7622 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7623 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7626 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7627 struct net_device *upper_dev,
7628 void *private, bool master)
7630 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7631 &dev->adj_list.upper,
7632 &upper_dev->adj_list.lower,
7636 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7637 struct net_device *upper_dev)
7639 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7640 &dev->adj_list.upper,
7641 &upper_dev->adj_list.lower);
7644 static int __netdev_upper_dev_link(struct net_device *dev,
7645 struct net_device *upper_dev, bool master,
7646 void *upper_priv, void *upper_info,
7647 struct netlink_ext_ack *extack)
7649 struct netdev_notifier_changeupper_info changeupper_info = {
7654 .upper_dev = upper_dev,
7657 .upper_info = upper_info,
7659 struct net_device *master_dev;
7664 if (dev == upper_dev)
7667 /* To prevent loops, check if dev is not upper device to upper_dev. */
7668 if (__netdev_has_upper_dev(upper_dev, dev))
7671 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7675 if (__netdev_has_upper_dev(dev, upper_dev))
7678 master_dev = __netdev_master_upper_dev_get(dev);
7680 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7683 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7684 &changeupper_info.info);
7685 ret = notifier_to_errno(ret);
7689 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7694 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7695 &changeupper_info.info);
7696 ret = notifier_to_errno(ret);
7700 __netdev_update_upper_level(dev, NULL);
7701 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7703 __netdev_update_lower_level(upper_dev, NULL);
7704 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7710 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7716 * netdev_upper_dev_link - Add a link to the upper device
7718 * @upper_dev: new upper device
7719 * @extack: netlink extended ack
7721 * Adds a link to device which is upper to this one. The caller must hold
7722 * the RTNL lock. On a failure a negative errno code is returned.
7723 * On success the reference counts are adjusted and the function
7726 int netdev_upper_dev_link(struct net_device *dev,
7727 struct net_device *upper_dev,
7728 struct netlink_ext_ack *extack)
7730 return __netdev_upper_dev_link(dev, upper_dev, false,
7731 NULL, NULL, extack);
7733 EXPORT_SYMBOL(netdev_upper_dev_link);
7736 * netdev_master_upper_dev_link - Add a master link to the upper device
7738 * @upper_dev: new upper device
7739 * @upper_priv: upper device private
7740 * @upper_info: upper info to be passed down via notifier
7741 * @extack: netlink extended ack
7743 * Adds a link to device which is upper to this one. In this case, only
7744 * one master upper device can be linked, although other non-master devices
7745 * might be linked as well. The caller must hold the RTNL lock.
7746 * On a failure a negative errno code is returned. On success the reference
7747 * counts are adjusted and the function returns zero.
7749 int netdev_master_upper_dev_link(struct net_device *dev,
7750 struct net_device *upper_dev,
7751 void *upper_priv, void *upper_info,
7752 struct netlink_ext_ack *extack)
7754 return __netdev_upper_dev_link(dev, upper_dev, true,
7755 upper_priv, upper_info, extack);
7757 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7760 * netdev_upper_dev_unlink - Removes a link to upper device
7762 * @upper_dev: new upper device
7764 * Removes a link to device which is upper to this one. The caller must hold
7767 void netdev_upper_dev_unlink(struct net_device *dev,
7768 struct net_device *upper_dev)
7770 struct netdev_notifier_changeupper_info changeupper_info = {
7774 .upper_dev = upper_dev,
7780 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7782 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7783 &changeupper_info.info);
7785 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7787 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7788 &changeupper_info.info);
7790 __netdev_update_upper_level(dev, NULL);
7791 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7793 __netdev_update_lower_level(upper_dev, NULL);
7794 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7797 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7799 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7800 struct net_device *lower_dev,
7803 struct netdev_adjacent *adj;
7805 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7809 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7814 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7815 struct net_device *lower_dev)
7817 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7820 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7821 struct net_device *lower_dev)
7823 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7826 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7827 struct net_device *new_dev,
7828 struct net_device *dev,
7829 struct netlink_ext_ack *extack)
7836 if (old_dev && new_dev != old_dev)
7837 netdev_adjacent_dev_disable(dev, old_dev);
7839 err = netdev_upper_dev_link(new_dev, dev, extack);
7841 if (old_dev && new_dev != old_dev)
7842 netdev_adjacent_dev_enable(dev, old_dev);
7848 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7850 void netdev_adjacent_change_commit(struct net_device *old_dev,
7851 struct net_device *new_dev,
7852 struct net_device *dev)
7854 if (!new_dev || !old_dev)
7857 if (new_dev == old_dev)
7860 netdev_adjacent_dev_enable(dev, old_dev);
7861 netdev_upper_dev_unlink(old_dev, dev);
7863 EXPORT_SYMBOL(netdev_adjacent_change_commit);
7865 void netdev_adjacent_change_abort(struct net_device *old_dev,
7866 struct net_device *new_dev,
7867 struct net_device *dev)
7872 if (old_dev && new_dev != old_dev)
7873 netdev_adjacent_dev_enable(dev, old_dev);
7875 netdev_upper_dev_unlink(new_dev, dev);
7877 EXPORT_SYMBOL(netdev_adjacent_change_abort);
7880 * netdev_bonding_info_change - Dispatch event about slave change
7882 * @bonding_info: info to dispatch
7884 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
7885 * The caller must hold the RTNL lock.
7887 void netdev_bonding_info_change(struct net_device *dev,
7888 struct netdev_bonding_info *bonding_info)
7890 struct netdev_notifier_bonding_info info = {
7894 memcpy(&info.bonding_info, bonding_info,
7895 sizeof(struct netdev_bonding_info));
7896 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
7899 EXPORT_SYMBOL(netdev_bonding_info_change);
7902 * netdev_get_xmit_slave - Get the xmit slave of master device
7904 * @all_slaves: assume all the slaves are active
7906 * The reference counters are not incremented so the caller must be
7907 * careful with locks. The caller must hold RCU lock.
7908 * %NULL is returned if no slave is found.
7911 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
7912 struct sk_buff *skb,
7915 const struct net_device_ops *ops = dev->netdev_ops;
7917 if (!ops->ndo_get_xmit_slave)
7919 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
7921 EXPORT_SYMBOL(netdev_get_xmit_slave);
7923 static void netdev_adjacent_add_links(struct net_device *dev)
7925 struct netdev_adjacent *iter;
7927 struct net *net = dev_net(dev);
7929 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7930 if (!net_eq(net, dev_net(iter->dev)))
7932 netdev_adjacent_sysfs_add(iter->dev, dev,
7933 &iter->dev->adj_list.lower);
7934 netdev_adjacent_sysfs_add(dev, iter->dev,
7935 &dev->adj_list.upper);
7938 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7939 if (!net_eq(net, dev_net(iter->dev)))
7941 netdev_adjacent_sysfs_add(iter->dev, dev,
7942 &iter->dev->adj_list.upper);
7943 netdev_adjacent_sysfs_add(dev, iter->dev,
7944 &dev->adj_list.lower);
7948 static void netdev_adjacent_del_links(struct net_device *dev)
7950 struct netdev_adjacent *iter;
7952 struct net *net = dev_net(dev);
7954 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7955 if (!net_eq(net, dev_net(iter->dev)))
7957 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7958 &iter->dev->adj_list.lower);
7959 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7960 &dev->adj_list.upper);
7963 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7964 if (!net_eq(net, dev_net(iter->dev)))
7966 netdev_adjacent_sysfs_del(iter->dev, dev->name,
7967 &iter->dev->adj_list.upper);
7968 netdev_adjacent_sysfs_del(dev, iter->dev->name,
7969 &dev->adj_list.lower);
7973 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
7975 struct netdev_adjacent *iter;
7977 struct net *net = dev_net(dev);
7979 list_for_each_entry(iter, &dev->adj_list.upper, list) {
7980 if (!net_eq(net, dev_net(iter->dev)))
7982 netdev_adjacent_sysfs_del(iter->dev, oldname,
7983 &iter->dev->adj_list.lower);
7984 netdev_adjacent_sysfs_add(iter->dev, dev,
7985 &iter->dev->adj_list.lower);
7988 list_for_each_entry(iter, &dev->adj_list.lower, list) {
7989 if (!net_eq(net, dev_net(iter->dev)))
7991 netdev_adjacent_sysfs_del(iter->dev, oldname,
7992 &iter->dev->adj_list.upper);
7993 netdev_adjacent_sysfs_add(iter->dev, dev,
7994 &iter->dev->adj_list.upper);
7998 void *netdev_lower_dev_get_private(struct net_device *dev,
7999 struct net_device *lower_dev)
8001 struct netdev_adjacent *lower;
8005 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8009 return lower->private;
8011 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8015 * netdev_lower_change - Dispatch event about lower device state change
8016 * @lower_dev: device
8017 * @lower_state_info: state to dispatch
8019 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8020 * The caller must hold the RTNL lock.
8022 void netdev_lower_state_changed(struct net_device *lower_dev,
8023 void *lower_state_info)
8025 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8026 .info.dev = lower_dev,
8030 changelowerstate_info.lower_state_info = lower_state_info;
8031 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8032 &changelowerstate_info.info);
8034 EXPORT_SYMBOL(netdev_lower_state_changed);
8036 static void dev_change_rx_flags(struct net_device *dev, int flags)
8038 const struct net_device_ops *ops = dev->netdev_ops;
8040 if (ops->ndo_change_rx_flags)
8041 ops->ndo_change_rx_flags(dev, flags);
8044 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8046 unsigned int old_flags = dev->flags;
8052 dev->flags |= IFF_PROMISC;
8053 dev->promiscuity += inc;
8054 if (dev->promiscuity == 0) {
8057 * If inc causes overflow, untouch promisc and return error.
8060 dev->flags &= ~IFF_PROMISC;
8062 dev->promiscuity -= inc;
8063 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8068 if (dev->flags != old_flags) {
8069 pr_info("device %s %s promiscuous mode\n",
8071 dev->flags & IFF_PROMISC ? "entered" : "left");
8072 if (audit_enabled) {
8073 current_uid_gid(&uid, &gid);
8074 audit_log(audit_context(), GFP_ATOMIC,
8075 AUDIT_ANOM_PROMISCUOUS,
8076 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8077 dev->name, (dev->flags & IFF_PROMISC),
8078 (old_flags & IFF_PROMISC),
8079 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8080 from_kuid(&init_user_ns, uid),
8081 from_kgid(&init_user_ns, gid),
8082 audit_get_sessionid(current));
8085 dev_change_rx_flags(dev, IFF_PROMISC);
8088 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8093 * dev_set_promiscuity - update promiscuity count on a device
8097 * Add or remove promiscuity from a device. While the count in the device
8098 * remains above zero the interface remains promiscuous. Once it hits zero
8099 * the device reverts back to normal filtering operation. A negative inc
8100 * value is used to drop promiscuity on the device.
8101 * Return 0 if successful or a negative errno code on error.
8103 int dev_set_promiscuity(struct net_device *dev, int inc)
8105 unsigned int old_flags = dev->flags;
8108 err = __dev_set_promiscuity(dev, inc, true);
8111 if (dev->flags != old_flags)
8112 dev_set_rx_mode(dev);
8115 EXPORT_SYMBOL(dev_set_promiscuity);
8117 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8119 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8123 dev->flags |= IFF_ALLMULTI;
8124 dev->allmulti += inc;
8125 if (dev->allmulti == 0) {
8128 * If inc causes overflow, untouch allmulti and return error.
8131 dev->flags &= ~IFF_ALLMULTI;
8133 dev->allmulti -= inc;
8134 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8139 if (dev->flags ^ old_flags) {
8140 dev_change_rx_flags(dev, IFF_ALLMULTI);
8141 dev_set_rx_mode(dev);
8143 __dev_notify_flags(dev, old_flags,
8144 dev->gflags ^ old_gflags);
8150 * dev_set_allmulti - update allmulti count on a device
8154 * Add or remove reception of all multicast frames to a device. While the
8155 * count in the device remains above zero the interface remains listening
8156 * to all interfaces. Once it hits zero the device reverts back to normal
8157 * filtering operation. A negative @inc value is used to drop the counter
8158 * when releasing a resource needing all multicasts.
8159 * Return 0 if successful or a negative errno code on error.
8162 int dev_set_allmulti(struct net_device *dev, int inc)
8164 return __dev_set_allmulti(dev, inc, true);
8166 EXPORT_SYMBOL(dev_set_allmulti);
8169 * Upload unicast and multicast address lists to device and
8170 * configure RX filtering. When the device doesn't support unicast
8171 * filtering it is put in promiscuous mode while unicast addresses
8174 void __dev_set_rx_mode(struct net_device *dev)
8176 const struct net_device_ops *ops = dev->netdev_ops;
8178 /* dev_open will call this function so the list will stay sane. */
8179 if (!(dev->flags&IFF_UP))
8182 if (!netif_device_present(dev))
8185 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8186 /* Unicast addresses changes may only happen under the rtnl,
8187 * therefore calling __dev_set_promiscuity here is safe.
8189 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8190 __dev_set_promiscuity(dev, 1, false);
8191 dev->uc_promisc = true;
8192 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8193 __dev_set_promiscuity(dev, -1, false);
8194 dev->uc_promisc = false;
8198 if (ops->ndo_set_rx_mode)
8199 ops->ndo_set_rx_mode(dev);
8202 void dev_set_rx_mode(struct net_device *dev)
8204 netif_addr_lock_bh(dev);
8205 __dev_set_rx_mode(dev);
8206 netif_addr_unlock_bh(dev);
8210 * dev_get_flags - get flags reported to userspace
8213 * Get the combination of flag bits exported through APIs to userspace.
8215 unsigned int dev_get_flags(const struct net_device *dev)
8219 flags = (dev->flags & ~(IFF_PROMISC |
8224 (dev->gflags & (IFF_PROMISC |
8227 if (netif_running(dev)) {
8228 if (netif_oper_up(dev))
8229 flags |= IFF_RUNNING;
8230 if (netif_carrier_ok(dev))
8231 flags |= IFF_LOWER_UP;
8232 if (netif_dormant(dev))
8233 flags |= IFF_DORMANT;
8238 EXPORT_SYMBOL(dev_get_flags);
8240 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8241 struct netlink_ext_ack *extack)
8243 unsigned int old_flags = dev->flags;
8249 * Set the flags on our device.
8252 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8253 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8255 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8259 * Load in the correct multicast list now the flags have changed.
8262 if ((old_flags ^ flags) & IFF_MULTICAST)
8263 dev_change_rx_flags(dev, IFF_MULTICAST);
8265 dev_set_rx_mode(dev);
8268 * Have we downed the interface. We handle IFF_UP ourselves
8269 * according to user attempts to set it, rather than blindly
8274 if ((old_flags ^ flags) & IFF_UP) {
8275 if (old_flags & IFF_UP)
8278 ret = __dev_open(dev, extack);
8281 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8282 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8283 unsigned int old_flags = dev->flags;
8285 dev->gflags ^= IFF_PROMISC;
8287 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8288 if (dev->flags != old_flags)
8289 dev_set_rx_mode(dev);
8292 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8293 * is important. Some (broken) drivers set IFF_PROMISC, when
8294 * IFF_ALLMULTI is requested not asking us and not reporting.
8296 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8297 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8299 dev->gflags ^= IFF_ALLMULTI;
8300 __dev_set_allmulti(dev, inc, false);
8306 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8307 unsigned int gchanges)
8309 unsigned int changes = dev->flags ^ old_flags;
8312 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8314 if (changes & IFF_UP) {
8315 if (dev->flags & IFF_UP)
8316 call_netdevice_notifiers(NETDEV_UP, dev);
8318 call_netdevice_notifiers(NETDEV_DOWN, dev);
8321 if (dev->flags & IFF_UP &&
8322 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8323 struct netdev_notifier_change_info change_info = {
8327 .flags_changed = changes,
8330 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8335 * dev_change_flags - change device settings
8337 * @flags: device state flags
8338 * @extack: netlink extended ack
8340 * Change settings on device based state flags. The flags are
8341 * in the userspace exported format.
8343 int dev_change_flags(struct net_device *dev, unsigned int flags,
8344 struct netlink_ext_ack *extack)
8347 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8349 ret = __dev_change_flags(dev, flags, extack);
8353 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8354 __dev_notify_flags(dev, old_flags, changes);
8357 EXPORT_SYMBOL(dev_change_flags);
8359 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8361 const struct net_device_ops *ops = dev->netdev_ops;
8363 if (ops->ndo_change_mtu)
8364 return ops->ndo_change_mtu(dev, new_mtu);
8366 /* Pairs with all the lockless reads of dev->mtu in the stack */
8367 WRITE_ONCE(dev->mtu, new_mtu);
8370 EXPORT_SYMBOL(__dev_set_mtu);
8372 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8373 struct netlink_ext_ack *extack)
8375 /* MTU must be positive, and in range */
8376 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8377 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8381 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8382 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8389 * dev_set_mtu_ext - Change maximum transfer unit
8391 * @new_mtu: new transfer unit
8392 * @extack: netlink extended ack
8394 * Change the maximum transfer size of the network device.
8396 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8397 struct netlink_ext_ack *extack)
8401 if (new_mtu == dev->mtu)
8404 err = dev_validate_mtu(dev, new_mtu, extack);
8408 if (!netif_device_present(dev))
8411 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8412 err = notifier_to_errno(err);
8416 orig_mtu = dev->mtu;
8417 err = __dev_set_mtu(dev, new_mtu);
8420 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8422 err = notifier_to_errno(err);
8424 /* setting mtu back and notifying everyone again,
8425 * so that they have a chance to revert changes.
8427 __dev_set_mtu(dev, orig_mtu);
8428 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8435 int dev_set_mtu(struct net_device *dev, int new_mtu)
8437 struct netlink_ext_ack extack;
8440 memset(&extack, 0, sizeof(extack));
8441 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8442 if (err && extack._msg)
8443 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8446 EXPORT_SYMBOL(dev_set_mtu);
8449 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8451 * @new_len: new tx queue length
8453 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8455 unsigned int orig_len = dev->tx_queue_len;
8458 if (new_len != (unsigned int)new_len)
8461 if (new_len != orig_len) {
8462 dev->tx_queue_len = new_len;
8463 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8464 res = notifier_to_errno(res);
8467 res = dev_qdisc_change_tx_queue_len(dev);
8475 netdev_err(dev, "refused to change device tx_queue_len\n");
8476 dev->tx_queue_len = orig_len;
8481 * dev_set_group - Change group this device belongs to
8483 * @new_group: group this device should belong to
8485 void dev_set_group(struct net_device *dev, int new_group)
8487 dev->group = new_group;
8489 EXPORT_SYMBOL(dev_set_group);
8492 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8494 * @addr: new address
8495 * @extack: netlink extended ack
8497 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8498 struct netlink_ext_ack *extack)
8500 struct netdev_notifier_pre_changeaddr_info info = {
8502 .info.extack = extack,
8507 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8508 return notifier_to_errno(rc);
8510 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8513 * dev_set_mac_address - Change Media Access Control Address
8516 * @extack: netlink extended ack
8518 * Change the hardware (MAC) address of the device
8520 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8521 struct netlink_ext_ack *extack)
8523 const struct net_device_ops *ops = dev->netdev_ops;
8526 if (!ops->ndo_set_mac_address)
8528 if (sa->sa_family != dev->type)
8530 if (!netif_device_present(dev))
8532 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8535 err = ops->ndo_set_mac_address(dev, sa);
8538 dev->addr_assign_type = NET_ADDR_SET;
8539 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8540 add_device_randomness(dev->dev_addr, dev->addr_len);
8543 EXPORT_SYMBOL(dev_set_mac_address);
8546 * dev_change_carrier - Change device carrier
8548 * @new_carrier: new value
8550 * Change device carrier
8552 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8554 const struct net_device_ops *ops = dev->netdev_ops;
8556 if (!ops->ndo_change_carrier)
8558 if (!netif_device_present(dev))
8560 return ops->ndo_change_carrier(dev, new_carrier);
8562 EXPORT_SYMBOL(dev_change_carrier);
8565 * dev_get_phys_port_id - Get device physical port ID
8569 * Get device physical port ID
8571 int dev_get_phys_port_id(struct net_device *dev,
8572 struct netdev_phys_item_id *ppid)
8574 const struct net_device_ops *ops = dev->netdev_ops;
8576 if (!ops->ndo_get_phys_port_id)
8578 return ops->ndo_get_phys_port_id(dev, ppid);
8580 EXPORT_SYMBOL(dev_get_phys_port_id);
8583 * dev_get_phys_port_name - Get device physical port name
8586 * @len: limit of bytes to copy to name
8588 * Get device physical port name
8590 int dev_get_phys_port_name(struct net_device *dev,
8591 char *name, size_t len)
8593 const struct net_device_ops *ops = dev->netdev_ops;
8596 if (ops->ndo_get_phys_port_name) {
8597 err = ops->ndo_get_phys_port_name(dev, name, len);
8598 if (err != -EOPNOTSUPP)
8601 return devlink_compat_phys_port_name_get(dev, name, len);
8603 EXPORT_SYMBOL(dev_get_phys_port_name);
8606 * dev_get_port_parent_id - Get the device's port parent identifier
8607 * @dev: network device
8608 * @ppid: pointer to a storage for the port's parent identifier
8609 * @recurse: allow/disallow recursion to lower devices
8611 * Get the devices's port parent identifier
8613 int dev_get_port_parent_id(struct net_device *dev,
8614 struct netdev_phys_item_id *ppid,
8617 const struct net_device_ops *ops = dev->netdev_ops;
8618 struct netdev_phys_item_id first = { };
8619 struct net_device *lower_dev;
8620 struct list_head *iter;
8623 if (ops->ndo_get_port_parent_id) {
8624 err = ops->ndo_get_port_parent_id(dev, ppid);
8625 if (err != -EOPNOTSUPP)
8629 err = devlink_compat_switch_id_get(dev, ppid);
8630 if (!err || err != -EOPNOTSUPP)
8636 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8637 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
8642 else if (memcmp(&first, ppid, sizeof(*ppid)))
8648 EXPORT_SYMBOL(dev_get_port_parent_id);
8651 * netdev_port_same_parent_id - Indicate if two network devices have
8652 * the same port parent identifier
8653 * @a: first network device
8654 * @b: second network device
8656 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8658 struct netdev_phys_item_id a_id = { };
8659 struct netdev_phys_item_id b_id = { };
8661 if (dev_get_port_parent_id(a, &a_id, true) ||
8662 dev_get_port_parent_id(b, &b_id, true))
8665 return netdev_phys_item_id_same(&a_id, &b_id);
8667 EXPORT_SYMBOL(netdev_port_same_parent_id);
8670 * dev_change_proto_down - update protocol port state information
8672 * @proto_down: new value
8674 * This info can be used by switch drivers to set the phys state of the
8677 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8679 const struct net_device_ops *ops = dev->netdev_ops;
8681 if (!ops->ndo_change_proto_down)
8683 if (!netif_device_present(dev))
8685 return ops->ndo_change_proto_down(dev, proto_down);
8687 EXPORT_SYMBOL(dev_change_proto_down);
8690 * dev_change_proto_down_generic - generic implementation for
8691 * ndo_change_proto_down that sets carrier according to
8695 * @proto_down: new value
8697 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8700 netif_carrier_off(dev);
8702 netif_carrier_on(dev);
8703 dev->proto_down = proto_down;
8706 EXPORT_SYMBOL(dev_change_proto_down_generic);
8708 u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
8709 enum bpf_netdev_command cmd)
8711 struct netdev_bpf xdp;
8716 memset(&xdp, 0, sizeof(xdp));
8719 /* Query must always succeed. */
8720 WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG);
8725 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
8726 struct netlink_ext_ack *extack, u32 flags,
8727 struct bpf_prog *prog)
8729 bool non_hw = !(flags & XDP_FLAGS_HW_MODE);
8730 struct bpf_prog *prev_prog = NULL;
8731 struct netdev_bpf xdp;
8735 prev_prog = bpf_prog_by_id(__dev_xdp_query(dev, bpf_op,
8737 if (IS_ERR(prev_prog))
8741 memset(&xdp, 0, sizeof(xdp));
8742 if (flags & XDP_FLAGS_HW_MODE)
8743 xdp.command = XDP_SETUP_PROG_HW;
8745 xdp.command = XDP_SETUP_PROG;
8746 xdp.extack = extack;
8750 err = bpf_op(dev, &xdp);
8752 bpf_prog_change_xdp(prev_prog, prog);
8755 bpf_prog_put(prev_prog);
8760 static void dev_xdp_uninstall(struct net_device *dev)
8762 struct netdev_bpf xdp;
8765 /* Remove generic XDP */
8766 WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
8768 /* Remove from the driver */
8769 ndo_bpf = dev->netdev_ops->ndo_bpf;
8773 memset(&xdp, 0, sizeof(xdp));
8774 xdp.command = XDP_QUERY_PROG;
8775 WARN_ON(ndo_bpf(dev, &xdp));
8777 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8780 /* Remove HW offload */
8781 memset(&xdp, 0, sizeof(xdp));
8782 xdp.command = XDP_QUERY_PROG_HW;
8783 if (!ndo_bpf(dev, &xdp) && xdp.prog_id)
8784 WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags,
8789 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
8791 * @extack: netlink extended ack
8792 * @fd: new program fd or negative value to clear
8793 * @expected_fd: old program fd that userspace expects to replace or clear
8794 * @flags: xdp-related flags
8796 * Set or clear a bpf program for a device
8798 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
8799 int fd, int expected_fd, u32 flags)
8801 const struct net_device_ops *ops = dev->netdev_ops;
8802 enum bpf_netdev_command query;
8803 u32 prog_id, expected_id = 0;
8804 bpf_op_t bpf_op, bpf_chk;
8805 struct bpf_prog *prog;
8811 offload = flags & XDP_FLAGS_HW_MODE;
8812 query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG;
8814 bpf_op = bpf_chk = ops->ndo_bpf;
8815 if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) {
8816 NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode");
8819 if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
8820 bpf_op = generic_xdp_install;
8821 if (bpf_op == bpf_chk)
8822 bpf_chk = generic_xdp_install;
8824 prog_id = __dev_xdp_query(dev, bpf_op, query);
8825 if (flags & XDP_FLAGS_REPLACE) {
8826 if (expected_fd >= 0) {
8827 prog = bpf_prog_get_type_dev(expected_fd,
8829 bpf_op == ops->ndo_bpf);
8831 return PTR_ERR(prog);
8832 expected_id = prog->aux->id;
8836 if (prog_id != expected_id) {
8837 NL_SET_ERR_MSG(extack, "Active program does not match expected");
8842 if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) {
8843 NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time");
8847 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && prog_id) {
8848 NL_SET_ERR_MSG(extack, "XDP program already attached");
8852 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
8853 bpf_op == ops->ndo_bpf);
8855 return PTR_ERR(prog);
8857 if (!offload && bpf_prog_is_dev_bound(prog->aux)) {
8858 NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
8863 if (prog->expected_attach_type == BPF_XDP_DEVMAP) {
8864 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
8869 /* prog->aux->id may be 0 for orphaned device-bound progs */
8870 if (prog->aux->id && prog->aux->id == prog_id) {
8880 err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
8881 if (err < 0 && prog)
8888 * dev_new_index - allocate an ifindex
8889 * @net: the applicable net namespace
8891 * Returns a suitable unique value for a new device interface
8892 * number. The caller must hold the rtnl semaphore or the
8893 * dev_base_lock to be sure it remains unique.
8895 static int dev_new_index(struct net *net)
8897 int ifindex = net->ifindex;
8902 if (!__dev_get_by_index(net, ifindex))
8903 return net->ifindex = ifindex;
8907 /* Delayed registration/unregisteration */
8908 static LIST_HEAD(net_todo_list);
8909 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
8911 static void net_set_todo(struct net_device *dev)
8913 list_add_tail(&dev->todo_list, &net_todo_list);
8914 dev_net(dev)->dev_unreg_count++;
8917 static void rollback_registered_many(struct list_head *head)
8919 struct net_device *dev, *tmp;
8920 LIST_HEAD(close_head);
8922 BUG_ON(dev_boot_phase);
8925 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
8926 /* Some devices call without registering
8927 * for initialization unwind. Remove those
8928 * devices and proceed with the remaining.
8930 if (dev->reg_state == NETREG_UNINITIALIZED) {
8931 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
8935 list_del(&dev->unreg_list);
8938 dev->dismantle = true;
8939 BUG_ON(dev->reg_state != NETREG_REGISTERED);
8942 /* If device is running, close it first. */
8943 list_for_each_entry(dev, head, unreg_list)
8944 list_add_tail(&dev->close_list, &close_head);
8945 dev_close_many(&close_head, true);
8947 list_for_each_entry(dev, head, unreg_list) {
8948 /* And unlink it from device chain. */
8949 unlist_netdevice(dev);
8951 dev->reg_state = NETREG_UNREGISTERING;
8953 flush_all_backlogs();
8957 list_for_each_entry(dev, head, unreg_list) {
8958 struct sk_buff *skb = NULL;
8960 /* Shutdown queueing discipline. */
8963 dev_xdp_uninstall(dev);
8965 /* Notify protocols, that we are about to destroy
8966 * this device. They should clean all the things.
8968 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8970 if (!dev->rtnl_link_ops ||
8971 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8972 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
8973 GFP_KERNEL, NULL, 0);
8976 * Flush the unicast and multicast chains
8981 netdev_name_node_alt_flush(dev);
8982 netdev_name_node_free(dev->name_node);
8984 if (dev->netdev_ops->ndo_uninit)
8985 dev->netdev_ops->ndo_uninit(dev);
8988 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
8990 /* Notifier chain MUST detach us all upper devices. */
8991 WARN_ON(netdev_has_any_upper_dev(dev));
8992 WARN_ON(netdev_has_any_lower_dev(dev));
8994 /* Remove entries from kobject tree */
8995 netdev_unregister_kobject(dev);
8997 /* Remove XPS queueing entries */
8998 netif_reset_xps_queues_gt(dev, 0);
9004 list_for_each_entry(dev, head, unreg_list)
9008 static void rollback_registered(struct net_device *dev)
9012 list_add(&dev->unreg_list, &single);
9013 rollback_registered_many(&single);
9017 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9018 struct net_device *upper, netdev_features_t features)
9020 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9021 netdev_features_t feature;
9024 for_each_netdev_feature(upper_disables, feature_bit) {
9025 feature = __NETIF_F_BIT(feature_bit);
9026 if (!(upper->wanted_features & feature)
9027 && (features & feature)) {
9028 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9029 &feature, upper->name);
9030 features &= ~feature;
9037 static void netdev_sync_lower_features(struct net_device *upper,
9038 struct net_device *lower, netdev_features_t features)
9040 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9041 netdev_features_t feature;
9044 for_each_netdev_feature(upper_disables, feature_bit) {
9045 feature = __NETIF_F_BIT(feature_bit);
9046 if (!(features & feature) && (lower->features & feature)) {
9047 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9048 &feature, lower->name);
9049 lower->wanted_features &= ~feature;
9050 __netdev_update_features(lower);
9052 if (unlikely(lower->features & feature))
9053 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9054 &feature, lower->name);
9056 netdev_features_change(lower);
9061 static netdev_features_t netdev_fix_features(struct net_device *dev,
9062 netdev_features_t features)
9064 /* Fix illegal checksum combinations */
9065 if ((features & NETIF_F_HW_CSUM) &&
9066 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9067 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9068 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9071 /* TSO requires that SG is present as well. */
9072 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9073 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9074 features &= ~NETIF_F_ALL_TSO;
9077 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9078 !(features & NETIF_F_IP_CSUM)) {
9079 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9080 features &= ~NETIF_F_TSO;
9081 features &= ~NETIF_F_TSO_ECN;
9084 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9085 !(features & NETIF_F_IPV6_CSUM)) {
9086 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9087 features &= ~NETIF_F_TSO6;
9090 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9091 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9092 features &= ~NETIF_F_TSO_MANGLEID;
9094 /* TSO ECN requires that TSO is present as well. */
9095 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9096 features &= ~NETIF_F_TSO_ECN;
9098 /* Software GSO depends on SG. */
9099 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9100 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9101 features &= ~NETIF_F_GSO;
9104 /* GSO partial features require GSO partial be set */
9105 if ((features & dev->gso_partial_features) &&
9106 !(features & NETIF_F_GSO_PARTIAL)) {
9108 "Dropping partially supported GSO features since no GSO partial.\n");
9109 features &= ~dev->gso_partial_features;
9112 if (!(features & NETIF_F_RXCSUM)) {
9113 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9114 * successfully merged by hardware must also have the
9115 * checksum verified by hardware. If the user does not
9116 * want to enable RXCSUM, logically, we should disable GRO_HW.
9118 if (features & NETIF_F_GRO_HW) {
9119 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9120 features &= ~NETIF_F_GRO_HW;
9124 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9125 if (features & NETIF_F_RXFCS) {
9126 if (features & NETIF_F_LRO) {
9127 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9128 features &= ~NETIF_F_LRO;
9131 if (features & NETIF_F_GRO_HW) {
9132 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9133 features &= ~NETIF_F_GRO_HW;
9140 int __netdev_update_features(struct net_device *dev)
9142 struct net_device *upper, *lower;
9143 netdev_features_t features;
9144 struct list_head *iter;
9149 features = netdev_get_wanted_features(dev);
9151 if (dev->netdev_ops->ndo_fix_features)
9152 features = dev->netdev_ops->ndo_fix_features(dev, features);
9154 /* driver might be less strict about feature dependencies */
9155 features = netdev_fix_features(dev, features);
9157 /* some features can't be enabled if they're off an an upper device */
9158 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9159 features = netdev_sync_upper_features(dev, upper, features);
9161 if (dev->features == features)
9164 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9165 &dev->features, &features);
9167 if (dev->netdev_ops->ndo_set_features)
9168 err = dev->netdev_ops->ndo_set_features(dev, features);
9172 if (unlikely(err < 0)) {
9174 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9175 err, &features, &dev->features);
9176 /* return non-0 since some features might have changed and
9177 * it's better to fire a spurious notification than miss it
9183 /* some features must be disabled on lower devices when disabled
9184 * on an upper device (think: bonding master or bridge)
9186 netdev_for_each_lower_dev(dev, lower, iter)
9187 netdev_sync_lower_features(dev, lower, features);
9190 netdev_features_t diff = features ^ dev->features;
9192 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9193 /* udp_tunnel_{get,drop}_rx_info both need
9194 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9195 * device, or they won't do anything.
9196 * Thus we need to update dev->features
9197 * *before* calling udp_tunnel_get_rx_info,
9198 * but *after* calling udp_tunnel_drop_rx_info.
9200 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9201 dev->features = features;
9202 udp_tunnel_get_rx_info(dev);
9204 udp_tunnel_drop_rx_info(dev);
9208 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9209 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9210 dev->features = features;
9211 err |= vlan_get_rx_ctag_filter_info(dev);
9213 vlan_drop_rx_ctag_filter_info(dev);
9217 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9218 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9219 dev->features = features;
9220 err |= vlan_get_rx_stag_filter_info(dev);
9222 vlan_drop_rx_stag_filter_info(dev);
9226 dev->features = features;
9229 return err < 0 ? 0 : 1;
9233 * netdev_update_features - recalculate device features
9234 * @dev: the device to check
9236 * Recalculate dev->features set and send notifications if it
9237 * has changed. Should be called after driver or hardware dependent
9238 * conditions might have changed that influence the features.
9240 void netdev_update_features(struct net_device *dev)
9242 if (__netdev_update_features(dev))
9243 netdev_features_change(dev);
9245 EXPORT_SYMBOL(netdev_update_features);
9248 * netdev_change_features - recalculate device features
9249 * @dev: the device to check
9251 * Recalculate dev->features set and send notifications even
9252 * if they have not changed. Should be called instead of
9253 * netdev_update_features() if also dev->vlan_features might
9254 * have changed to allow the changes to be propagated to stacked
9257 void netdev_change_features(struct net_device *dev)
9259 __netdev_update_features(dev);
9260 netdev_features_change(dev);
9262 EXPORT_SYMBOL(netdev_change_features);
9265 * netif_stacked_transfer_operstate - transfer operstate
9266 * @rootdev: the root or lower level device to transfer state from
9267 * @dev: the device to transfer operstate to
9269 * Transfer operational state from root to device. This is normally
9270 * called when a stacking relationship exists between the root
9271 * device and the device(a leaf device).
9273 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9274 struct net_device *dev)
9276 if (rootdev->operstate == IF_OPER_DORMANT)
9277 netif_dormant_on(dev);
9279 netif_dormant_off(dev);
9281 if (rootdev->operstate == IF_OPER_TESTING)
9282 netif_testing_on(dev);
9284 netif_testing_off(dev);
9286 if (netif_carrier_ok(rootdev))
9287 netif_carrier_on(dev);
9289 netif_carrier_off(dev);
9291 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9293 static int netif_alloc_rx_queues(struct net_device *dev)
9295 unsigned int i, count = dev->num_rx_queues;
9296 struct netdev_rx_queue *rx;
9297 size_t sz = count * sizeof(*rx);
9302 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9308 for (i = 0; i < count; i++) {
9311 /* XDP RX-queue setup */
9312 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
9319 /* Rollback successful reg's and free other resources */
9321 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9327 static void netif_free_rx_queues(struct net_device *dev)
9329 unsigned int i, count = dev->num_rx_queues;
9331 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9335 for (i = 0; i < count; i++)
9336 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9341 static void netdev_init_one_queue(struct net_device *dev,
9342 struct netdev_queue *queue, void *_unused)
9344 /* Initialize queue lock */
9345 spin_lock_init(&queue->_xmit_lock);
9346 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9347 queue->xmit_lock_owner = -1;
9348 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9351 dql_init(&queue->dql, HZ);
9355 static void netif_free_tx_queues(struct net_device *dev)
9360 static int netif_alloc_netdev_queues(struct net_device *dev)
9362 unsigned int count = dev->num_tx_queues;
9363 struct netdev_queue *tx;
9364 size_t sz = count * sizeof(*tx);
9366 if (count < 1 || count > 0xffff)
9369 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9375 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9376 spin_lock_init(&dev->tx_global_lock);
9381 void netif_tx_stop_all_queues(struct net_device *dev)
9385 for (i = 0; i < dev->num_tx_queues; i++) {
9386 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9388 netif_tx_stop_queue(txq);
9391 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9394 * register_netdevice - register a network device
9395 * @dev: device to register
9397 * Take a completed network device structure and add it to the kernel
9398 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9399 * chain. 0 is returned on success. A negative errno code is returned
9400 * on a failure to set up the device, or if the name is a duplicate.
9402 * Callers must hold the rtnl semaphore. You may want
9403 * register_netdev() instead of this.
9406 * The locking appears insufficient to guarantee two parallel registers
9407 * will not get the same name.
9410 int register_netdevice(struct net_device *dev)
9413 struct net *net = dev_net(dev);
9415 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9416 NETDEV_FEATURE_COUNT);
9417 BUG_ON(dev_boot_phase);
9422 /* When net_device's are persistent, this will be fatal. */
9423 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9426 ret = ethtool_check_ops(dev->ethtool_ops);
9430 spin_lock_init(&dev->addr_list_lock);
9431 netdev_set_addr_lockdep_class(dev);
9433 ret = dev_get_valid_name(net, dev, dev->name);
9438 dev->name_node = netdev_name_node_head_alloc(dev);
9439 if (!dev->name_node)
9442 /* Init, if this function is available */
9443 if (dev->netdev_ops->ndo_init) {
9444 ret = dev->netdev_ops->ndo_init(dev);
9452 if (((dev->hw_features | dev->features) &
9453 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9454 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9455 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9456 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9463 dev->ifindex = dev_new_index(net);
9464 else if (__dev_get_by_index(net, dev->ifindex))
9467 /* Transfer changeable features to wanted_features and enable
9468 * software offloads (GSO and GRO).
9470 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9471 dev->features |= NETIF_F_SOFT_FEATURES;
9473 if (dev->netdev_ops->ndo_udp_tunnel_add) {
9474 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9475 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9478 dev->wanted_features = dev->features & dev->hw_features;
9480 if (!(dev->flags & IFF_LOOPBACK))
9481 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9483 /* If IPv4 TCP segmentation offload is supported we should also
9484 * allow the device to enable segmenting the frame with the option
9485 * of ignoring a static IP ID value. This doesn't enable the
9486 * feature itself but allows the user to enable it later.
9488 if (dev->hw_features & NETIF_F_TSO)
9489 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9490 if (dev->vlan_features & NETIF_F_TSO)
9491 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9492 if (dev->mpls_features & NETIF_F_TSO)
9493 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9494 if (dev->hw_enc_features & NETIF_F_TSO)
9495 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9497 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9499 dev->vlan_features |= NETIF_F_HIGHDMA;
9501 /* Make NETIF_F_SG inheritable to tunnel devices.
9503 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9505 /* Make NETIF_F_SG inheritable to MPLS.
9507 dev->mpls_features |= NETIF_F_SG;
9509 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9510 ret = notifier_to_errno(ret);
9514 ret = netdev_register_kobject(dev);
9516 dev->reg_state = NETREG_UNREGISTERED;
9519 dev->reg_state = NETREG_REGISTERED;
9521 __netdev_update_features(dev);
9524 * Default initial state at registry is that the
9525 * device is present.
9528 set_bit(__LINK_STATE_PRESENT, &dev->state);
9530 linkwatch_init_dev(dev);
9532 dev_init_scheduler(dev);
9534 list_netdevice(dev);
9535 add_device_randomness(dev->dev_addr, dev->addr_len);
9537 /* If the device has permanent device address, driver should
9538 * set dev_addr and also addr_assign_type should be set to
9539 * NET_ADDR_PERM (default value).
9541 if (dev->addr_assign_type == NET_ADDR_PERM)
9542 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9544 /* Notify protocols, that a new device appeared. */
9545 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9546 ret = notifier_to_errno(ret);
9548 rollback_registered(dev);
9551 dev->reg_state = NETREG_UNREGISTERED;
9552 /* We should put the kobject that hold in
9553 * netdev_unregister_kobject(), otherwise
9554 * the net device cannot be freed when
9555 * driver calls free_netdev(), because the
9556 * kobject is being hold.
9558 kobject_put(&dev->dev.kobj);
9561 * Prevent userspace races by waiting until the network
9562 * device is fully setup before sending notifications.
9564 if (!dev->rtnl_link_ops ||
9565 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9566 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
9572 if (dev->netdev_ops->ndo_uninit)
9573 dev->netdev_ops->ndo_uninit(dev);
9574 if (dev->priv_destructor)
9575 dev->priv_destructor(dev);
9577 netdev_name_node_free(dev->name_node);
9580 EXPORT_SYMBOL(register_netdevice);
9583 * init_dummy_netdev - init a dummy network device for NAPI
9584 * @dev: device to init
9586 * This takes a network device structure and initialize the minimum
9587 * amount of fields so it can be used to schedule NAPI polls without
9588 * registering a full blown interface. This is to be used by drivers
9589 * that need to tie several hardware interfaces to a single NAPI
9590 * poll scheduler due to HW limitations.
9592 int init_dummy_netdev(struct net_device *dev)
9594 /* Clear everything. Note we don't initialize spinlocks
9595 * are they aren't supposed to be taken by any of the
9596 * NAPI code and this dummy netdev is supposed to be
9597 * only ever used for NAPI polls
9599 memset(dev, 0, sizeof(struct net_device));
9601 /* make sure we BUG if trying to hit standard
9602 * register/unregister code path
9604 dev->reg_state = NETREG_DUMMY;
9606 /* NAPI wants this */
9607 INIT_LIST_HEAD(&dev->napi_list);
9609 /* a dummy interface is started by default */
9610 set_bit(__LINK_STATE_PRESENT, &dev->state);
9611 set_bit(__LINK_STATE_START, &dev->state);
9613 /* napi_busy_loop stats accounting wants this */
9614 dev_net_set(dev, &init_net);
9616 /* Note : We dont allocate pcpu_refcnt for dummy devices,
9617 * because users of this 'device' dont need to change
9623 EXPORT_SYMBOL_GPL(init_dummy_netdev);
9627 * register_netdev - register a network device
9628 * @dev: device to register
9630 * Take a completed network device structure and add it to the kernel
9631 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9632 * chain. 0 is returned on success. A negative errno code is returned
9633 * on a failure to set up the device, or if the name is a duplicate.
9635 * This is a wrapper around register_netdevice that takes the rtnl semaphore
9636 * and expands the device name if you passed a format string to
9639 int register_netdev(struct net_device *dev)
9643 if (rtnl_lock_killable())
9645 err = register_netdevice(dev);
9649 EXPORT_SYMBOL(register_netdev);
9651 int netdev_refcnt_read(const struct net_device *dev)
9655 for_each_possible_cpu(i)
9656 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
9659 EXPORT_SYMBOL(netdev_refcnt_read);
9662 * netdev_wait_allrefs - wait until all references are gone.
9663 * @dev: target net_device
9665 * This is called when unregistering network devices.
9667 * Any protocol or device that holds a reference should register
9668 * for netdevice notification, and cleanup and put back the
9669 * reference if they receive an UNREGISTER event.
9670 * We can get stuck here if buggy protocols don't correctly
9673 static void netdev_wait_allrefs(struct net_device *dev)
9675 unsigned long rebroadcast_time, warning_time;
9678 linkwatch_forget_dev(dev);
9680 rebroadcast_time = warning_time = jiffies;
9681 refcnt = netdev_refcnt_read(dev);
9683 while (refcnt != 0) {
9684 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
9687 /* Rebroadcast unregister notification */
9688 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9694 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
9696 /* We must not have linkwatch events
9697 * pending on unregister. If this
9698 * happens, we simply run the queue
9699 * unscheduled, resulting in a noop
9702 linkwatch_run_queue();
9707 rebroadcast_time = jiffies;
9712 refcnt = netdev_refcnt_read(dev);
9714 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
9715 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
9717 warning_time = jiffies;
9726 * register_netdevice(x1);
9727 * register_netdevice(x2);
9729 * unregister_netdevice(y1);
9730 * unregister_netdevice(y2);
9736 * We are invoked by rtnl_unlock().
9737 * This allows us to deal with problems:
9738 * 1) We can delete sysfs objects which invoke hotplug
9739 * without deadlocking with linkwatch via keventd.
9740 * 2) Since we run with the RTNL semaphore not held, we can sleep
9741 * safely in order to wait for the netdev refcnt to drop to zero.
9743 * We must not return until all unregister events added during
9744 * the interval the lock was held have been completed.
9746 void netdev_run_todo(void)
9748 struct list_head list;
9750 /* Snapshot list, allow later requests */
9751 list_replace_init(&net_todo_list, &list);
9756 /* Wait for rcu callbacks to finish before next phase */
9757 if (!list_empty(&list))
9760 while (!list_empty(&list)) {
9761 struct net_device *dev
9762 = list_first_entry(&list, struct net_device, todo_list);
9763 list_del(&dev->todo_list);
9765 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
9766 pr_err("network todo '%s' but state %d\n",
9767 dev->name, dev->reg_state);
9772 dev->reg_state = NETREG_UNREGISTERED;
9774 netdev_wait_allrefs(dev);
9777 BUG_ON(netdev_refcnt_read(dev));
9778 BUG_ON(!list_empty(&dev->ptype_all));
9779 BUG_ON(!list_empty(&dev->ptype_specific));
9780 WARN_ON(rcu_access_pointer(dev->ip_ptr));
9781 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
9782 #if IS_ENABLED(CONFIG_DECNET)
9783 WARN_ON(dev->dn_ptr);
9785 if (dev->priv_destructor)
9786 dev->priv_destructor(dev);
9787 if (dev->needs_free_netdev)
9790 /* Report a network device has been unregistered */
9792 dev_net(dev)->dev_unreg_count--;
9794 wake_up(&netdev_unregistering_wq);
9796 /* Free network device */
9797 kobject_put(&dev->dev.kobj);
9801 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
9802 * all the same fields in the same order as net_device_stats, with only
9803 * the type differing, but rtnl_link_stats64 may have additional fields
9804 * at the end for newer counters.
9806 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
9807 const struct net_device_stats *netdev_stats)
9809 #if BITS_PER_LONG == 64
9810 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
9811 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
9812 /* zero out counters that only exist in rtnl_link_stats64 */
9813 memset((char *)stats64 + sizeof(*netdev_stats), 0,
9814 sizeof(*stats64) - sizeof(*netdev_stats));
9816 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
9817 const unsigned long *src = (const unsigned long *)netdev_stats;
9818 u64 *dst = (u64 *)stats64;
9820 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
9821 for (i = 0; i < n; i++)
9823 /* zero out counters that only exist in rtnl_link_stats64 */
9824 memset((char *)stats64 + n * sizeof(u64), 0,
9825 sizeof(*stats64) - n * sizeof(u64));
9828 EXPORT_SYMBOL(netdev_stats_to_stats64);
9831 * dev_get_stats - get network device statistics
9832 * @dev: device to get statistics from
9833 * @storage: place to store stats
9835 * Get network statistics from device. Return @storage.
9836 * The device driver may provide its own method by setting
9837 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
9838 * otherwise the internal statistics structure is used.
9840 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
9841 struct rtnl_link_stats64 *storage)
9843 const struct net_device_ops *ops = dev->netdev_ops;
9845 if (ops->ndo_get_stats64) {
9846 memset(storage, 0, sizeof(*storage));
9847 ops->ndo_get_stats64(dev, storage);
9848 } else if (ops->ndo_get_stats) {
9849 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
9851 netdev_stats_to_stats64(storage, &dev->stats);
9853 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
9854 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
9855 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
9858 EXPORT_SYMBOL(dev_get_stats);
9860 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
9862 struct netdev_queue *queue = dev_ingress_queue(dev);
9864 #ifdef CONFIG_NET_CLS_ACT
9867 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
9870 netdev_init_one_queue(dev, queue, NULL);
9871 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
9872 queue->qdisc_sleeping = &noop_qdisc;
9873 rcu_assign_pointer(dev->ingress_queue, queue);
9878 static const struct ethtool_ops default_ethtool_ops;
9880 void netdev_set_default_ethtool_ops(struct net_device *dev,
9881 const struct ethtool_ops *ops)
9883 if (dev->ethtool_ops == &default_ethtool_ops)
9884 dev->ethtool_ops = ops;
9886 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
9888 void netdev_freemem(struct net_device *dev)
9890 char *addr = (char *)dev - dev->padded;
9896 * alloc_netdev_mqs - allocate network device
9897 * @sizeof_priv: size of private data to allocate space for
9898 * @name: device name format string
9899 * @name_assign_type: origin of device name
9900 * @setup: callback to initialize device
9901 * @txqs: the number of TX subqueues to allocate
9902 * @rxqs: the number of RX subqueues to allocate
9904 * Allocates a struct net_device with private data area for driver use
9905 * and performs basic initialization. Also allocates subqueue structs
9906 * for each queue on the device.
9908 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
9909 unsigned char name_assign_type,
9910 void (*setup)(struct net_device *),
9911 unsigned int txqs, unsigned int rxqs)
9913 struct net_device *dev;
9914 unsigned int alloc_size;
9915 struct net_device *p;
9917 BUG_ON(strlen(name) >= sizeof(dev->name));
9920 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
9925 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
9929 alloc_size = sizeof(struct net_device);
9931 /* ensure 32-byte alignment of private area */
9932 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
9933 alloc_size += sizeof_priv;
9935 /* ensure 32-byte alignment of whole construct */
9936 alloc_size += NETDEV_ALIGN - 1;
9938 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9942 dev = PTR_ALIGN(p, NETDEV_ALIGN);
9943 dev->padded = (char *)dev - (char *)p;
9945 dev->pcpu_refcnt = alloc_percpu(int);
9946 if (!dev->pcpu_refcnt)
9949 if (dev_addr_init(dev))
9955 dev_net_set(dev, &init_net);
9957 dev->gso_max_size = GSO_MAX_SIZE;
9958 dev->gso_max_segs = GSO_MAX_SEGS;
9959 dev->upper_level = 1;
9960 dev->lower_level = 1;
9962 INIT_LIST_HEAD(&dev->napi_list);
9963 INIT_LIST_HEAD(&dev->unreg_list);
9964 INIT_LIST_HEAD(&dev->close_list);
9965 INIT_LIST_HEAD(&dev->link_watch_list);
9966 INIT_LIST_HEAD(&dev->adj_list.upper);
9967 INIT_LIST_HEAD(&dev->adj_list.lower);
9968 INIT_LIST_HEAD(&dev->ptype_all);
9969 INIT_LIST_HEAD(&dev->ptype_specific);
9970 INIT_LIST_HEAD(&dev->net_notifier_list);
9971 #ifdef CONFIG_NET_SCHED
9972 hash_init(dev->qdisc_hash);
9974 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
9977 if (!dev->tx_queue_len) {
9978 dev->priv_flags |= IFF_NO_QUEUE;
9979 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
9982 dev->num_tx_queues = txqs;
9983 dev->real_num_tx_queues = txqs;
9984 if (netif_alloc_netdev_queues(dev))
9987 dev->num_rx_queues = rxqs;
9988 dev->real_num_rx_queues = rxqs;
9989 if (netif_alloc_rx_queues(dev))
9992 strcpy(dev->name, name);
9993 dev->name_assign_type = name_assign_type;
9994 dev->group = INIT_NETDEV_GROUP;
9995 if (!dev->ethtool_ops)
9996 dev->ethtool_ops = &default_ethtool_ops;
9998 nf_hook_ingress_init(dev);
10007 free_percpu(dev->pcpu_refcnt);
10009 netdev_freemem(dev);
10012 EXPORT_SYMBOL(alloc_netdev_mqs);
10015 * free_netdev - free network device
10018 * This function does the last stage of destroying an allocated device
10019 * interface. The reference to the device object is released. If this
10020 * is the last reference then it will be freed.Must be called in process
10023 void free_netdev(struct net_device *dev)
10025 struct napi_struct *p, *n;
10028 netif_free_tx_queues(dev);
10029 netif_free_rx_queues(dev);
10031 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10033 /* Flush device addresses */
10034 dev_addr_flush(dev);
10036 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10039 free_percpu(dev->pcpu_refcnt);
10040 dev->pcpu_refcnt = NULL;
10041 free_percpu(dev->xdp_bulkq);
10042 dev->xdp_bulkq = NULL;
10044 /* Compatibility with error handling in drivers */
10045 if (dev->reg_state == NETREG_UNINITIALIZED) {
10046 netdev_freemem(dev);
10050 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10051 dev->reg_state = NETREG_RELEASED;
10053 /* will free via device release */
10054 put_device(&dev->dev);
10056 EXPORT_SYMBOL(free_netdev);
10059 * synchronize_net - Synchronize with packet receive processing
10061 * Wait for packets currently being received to be done.
10062 * Does not block later packets from starting.
10064 void synchronize_net(void)
10067 if (rtnl_is_locked())
10068 synchronize_rcu_expedited();
10072 EXPORT_SYMBOL(synchronize_net);
10075 * unregister_netdevice_queue - remove device from the kernel
10079 * This function shuts down a device interface and removes it
10080 * from the kernel tables.
10081 * If head not NULL, device is queued to be unregistered later.
10083 * Callers must hold the rtnl semaphore. You may want
10084 * unregister_netdev() instead of this.
10087 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10092 list_move_tail(&dev->unreg_list, head);
10094 rollback_registered(dev);
10095 /* Finish processing unregister after unlock */
10099 EXPORT_SYMBOL(unregister_netdevice_queue);
10102 * unregister_netdevice_many - unregister many devices
10103 * @head: list of devices
10105 * Note: As most callers use a stack allocated list_head,
10106 * we force a list_del() to make sure stack wont be corrupted later.
10108 void unregister_netdevice_many(struct list_head *head)
10110 struct net_device *dev;
10112 if (!list_empty(head)) {
10113 rollback_registered_many(head);
10114 list_for_each_entry(dev, head, unreg_list)
10119 EXPORT_SYMBOL(unregister_netdevice_many);
10122 * unregister_netdev - remove device from the kernel
10125 * This function shuts down a device interface and removes it
10126 * from the kernel tables.
10128 * This is just a wrapper for unregister_netdevice that takes
10129 * the rtnl semaphore. In general you want to use this and not
10130 * unregister_netdevice.
10132 void unregister_netdev(struct net_device *dev)
10135 unregister_netdevice(dev);
10138 EXPORT_SYMBOL(unregister_netdev);
10141 * dev_change_net_namespace - move device to different nethost namespace
10143 * @net: network namespace
10144 * @pat: If not NULL name pattern to try if the current device name
10145 * is already taken in the destination network namespace.
10147 * This function shuts down a device interface and moves it
10148 * to a new network namespace. On success 0 is returned, on
10149 * a failure a netagive errno code is returned.
10151 * Callers must hold the rtnl semaphore.
10154 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
10156 struct net *net_old = dev_net(dev);
10157 int err, new_nsid, new_ifindex;
10161 /* Don't allow namespace local devices to be moved. */
10163 if (dev->features & NETIF_F_NETNS_LOCAL)
10166 /* Ensure the device has been registrered */
10167 if (dev->reg_state != NETREG_REGISTERED)
10170 /* Get out if there is nothing todo */
10172 if (net_eq(net_old, net))
10175 /* Pick the destination device name, and ensure
10176 * we can use it in the destination network namespace.
10179 if (__dev_get_by_name(net, dev->name)) {
10180 /* We get here if we can't use the current device name */
10183 err = dev_get_valid_name(net, dev, pat);
10189 * And now a mini version of register_netdevice unregister_netdevice.
10192 /* If device is running close it first. */
10195 /* And unlink it from device chain */
10196 unlist_netdevice(dev);
10200 /* Shutdown queueing discipline. */
10203 /* Notify protocols, that we are about to destroy
10204 * this device. They should clean all the things.
10206 * Note that dev->reg_state stays at NETREG_REGISTERED.
10207 * This is wanted because this way 8021q and macvlan know
10208 * the device is just moving and can keep their slaves up.
10210 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10213 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10214 /* If there is an ifindex conflict assign a new one */
10215 if (__dev_get_by_index(net, dev->ifindex))
10216 new_ifindex = dev_new_index(net);
10218 new_ifindex = dev->ifindex;
10220 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10224 * Flush the unicast and multicast chains
10229 /* Send a netdev-removed uevent to the old namespace */
10230 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10231 netdev_adjacent_del_links(dev);
10233 /* Move per-net netdevice notifiers that are following the netdevice */
10234 move_netdevice_notifiers_dev_net(dev, net);
10236 /* Actually switch the network namespace */
10237 dev_net_set(dev, net);
10238 dev->ifindex = new_ifindex;
10240 /* Send a netdev-add uevent to the new namespace */
10241 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10242 netdev_adjacent_add_links(dev);
10244 /* Fixup kobjects */
10245 err = device_rename(&dev->dev, dev->name);
10248 /* Adapt owner in case owning user namespace of target network
10249 * namespace is different from the original one.
10251 err = netdev_change_owner(dev, net_old, net);
10254 /* Add the device back in the hashes */
10255 list_netdevice(dev);
10257 /* Notify protocols, that a new device appeared. */
10258 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10261 * Prevent userspace races by waiting until the network
10262 * device is fully setup before sending notifications.
10264 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10271 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
10273 static int dev_cpu_dead(unsigned int oldcpu)
10275 struct sk_buff **list_skb;
10276 struct sk_buff *skb;
10278 struct softnet_data *sd, *oldsd, *remsd = NULL;
10280 local_irq_disable();
10281 cpu = smp_processor_id();
10282 sd = &per_cpu(softnet_data, cpu);
10283 oldsd = &per_cpu(softnet_data, oldcpu);
10285 /* Find end of our completion_queue. */
10286 list_skb = &sd->completion_queue;
10288 list_skb = &(*list_skb)->next;
10289 /* Append completion queue from offline CPU. */
10290 *list_skb = oldsd->completion_queue;
10291 oldsd->completion_queue = NULL;
10293 /* Append output queue from offline CPU. */
10294 if (oldsd->output_queue) {
10295 *sd->output_queue_tailp = oldsd->output_queue;
10296 sd->output_queue_tailp = oldsd->output_queue_tailp;
10297 oldsd->output_queue = NULL;
10298 oldsd->output_queue_tailp = &oldsd->output_queue;
10300 /* Append NAPI poll list from offline CPU, with one exception :
10301 * process_backlog() must be called by cpu owning percpu backlog.
10302 * We properly handle process_queue & input_pkt_queue later.
10304 while (!list_empty(&oldsd->poll_list)) {
10305 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10306 struct napi_struct,
10309 list_del_init(&napi->poll_list);
10310 if (napi->poll == process_backlog)
10313 ____napi_schedule(sd, napi);
10316 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10317 local_irq_enable();
10320 remsd = oldsd->rps_ipi_list;
10321 oldsd->rps_ipi_list = NULL;
10323 /* send out pending IPI's on offline CPU */
10324 net_rps_send_ipi(remsd);
10326 /* Process offline CPU's input_pkt_queue */
10327 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10329 input_queue_head_incr(oldsd);
10331 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10333 input_queue_head_incr(oldsd);
10340 * netdev_increment_features - increment feature set by one
10341 * @all: current feature set
10342 * @one: new feature set
10343 * @mask: mask feature set
10345 * Computes a new feature set after adding a device with feature set
10346 * @one to the master device with current feature set @all. Will not
10347 * enable anything that is off in @mask. Returns the new feature set.
10349 netdev_features_t netdev_increment_features(netdev_features_t all,
10350 netdev_features_t one, netdev_features_t mask)
10352 if (mask & NETIF_F_HW_CSUM)
10353 mask |= NETIF_F_CSUM_MASK;
10354 mask |= NETIF_F_VLAN_CHALLENGED;
10356 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10357 all &= one | ~NETIF_F_ALL_FOR_ALL;
10359 /* If one device supports hw checksumming, set for all. */
10360 if (all & NETIF_F_HW_CSUM)
10361 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10365 EXPORT_SYMBOL(netdev_increment_features);
10367 static struct hlist_head * __net_init netdev_create_hash(void)
10370 struct hlist_head *hash;
10372 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10374 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10375 INIT_HLIST_HEAD(&hash[i]);
10380 /* Initialize per network namespace state */
10381 static int __net_init netdev_init(struct net *net)
10383 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10384 8 * sizeof_field(struct napi_struct, gro_bitmask));
10386 if (net != &init_net)
10387 INIT_LIST_HEAD(&net->dev_base_head);
10389 net->dev_name_head = netdev_create_hash();
10390 if (net->dev_name_head == NULL)
10393 net->dev_index_head = netdev_create_hash();
10394 if (net->dev_index_head == NULL)
10397 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10402 kfree(net->dev_name_head);
10408 * netdev_drivername - network driver for the device
10409 * @dev: network device
10411 * Determine network driver for device.
10413 const char *netdev_drivername(const struct net_device *dev)
10415 const struct device_driver *driver;
10416 const struct device *parent;
10417 const char *empty = "";
10419 parent = dev->dev.parent;
10423 driver = parent->driver;
10424 if (driver && driver->name)
10425 return driver->name;
10429 static void __netdev_printk(const char *level, const struct net_device *dev,
10430 struct va_format *vaf)
10432 if (dev && dev->dev.parent) {
10433 dev_printk_emit(level[1] - '0',
10436 dev_driver_string(dev->dev.parent),
10437 dev_name(dev->dev.parent),
10438 netdev_name(dev), netdev_reg_state(dev),
10441 printk("%s%s%s: %pV",
10442 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10444 printk("%s(NULL net_device): %pV", level, vaf);
10448 void netdev_printk(const char *level, const struct net_device *dev,
10449 const char *format, ...)
10451 struct va_format vaf;
10454 va_start(args, format);
10459 __netdev_printk(level, dev, &vaf);
10463 EXPORT_SYMBOL(netdev_printk);
10465 #define define_netdev_printk_level(func, level) \
10466 void func(const struct net_device *dev, const char *fmt, ...) \
10468 struct va_format vaf; \
10471 va_start(args, fmt); \
10476 __netdev_printk(level, dev, &vaf); \
10480 EXPORT_SYMBOL(func);
10482 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10483 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10484 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10485 define_netdev_printk_level(netdev_err, KERN_ERR);
10486 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10487 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10488 define_netdev_printk_level(netdev_info, KERN_INFO);
10490 static void __net_exit netdev_exit(struct net *net)
10492 kfree(net->dev_name_head);
10493 kfree(net->dev_index_head);
10494 if (net != &init_net)
10495 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
10498 static struct pernet_operations __net_initdata netdev_net_ops = {
10499 .init = netdev_init,
10500 .exit = netdev_exit,
10503 static void __net_exit default_device_exit(struct net *net)
10505 struct net_device *dev, *aux;
10507 * Push all migratable network devices back to the
10508 * initial network namespace
10511 for_each_netdev_safe(net, dev, aux) {
10513 char fb_name[IFNAMSIZ];
10515 /* Ignore unmoveable devices (i.e. loopback) */
10516 if (dev->features & NETIF_F_NETNS_LOCAL)
10519 /* Leave virtual devices for the generic cleanup */
10520 if (dev->rtnl_link_ops)
10523 /* Push remaining network devices to init_net */
10524 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
10525 if (__dev_get_by_name(&init_net, fb_name))
10526 snprintf(fb_name, IFNAMSIZ, "dev%%d");
10527 err = dev_change_net_namespace(dev, &init_net, fb_name);
10529 pr_emerg("%s: failed to move %s to init_net: %d\n",
10530 __func__, dev->name, err);
10537 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
10539 /* Return with the rtnl_lock held when there are no network
10540 * devices unregistering in any network namespace in net_list.
10543 bool unregistering;
10544 DEFINE_WAIT_FUNC(wait, woken_wake_function);
10546 add_wait_queue(&netdev_unregistering_wq, &wait);
10548 unregistering = false;
10550 list_for_each_entry(net, net_list, exit_list) {
10551 if (net->dev_unreg_count > 0) {
10552 unregistering = true;
10556 if (!unregistering)
10560 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
10562 remove_wait_queue(&netdev_unregistering_wq, &wait);
10565 static void __net_exit default_device_exit_batch(struct list_head *net_list)
10567 /* At exit all network devices most be removed from a network
10568 * namespace. Do this in the reverse order of registration.
10569 * Do this across as many network namespaces as possible to
10570 * improve batching efficiency.
10572 struct net_device *dev;
10574 LIST_HEAD(dev_kill_list);
10576 /* To prevent network device cleanup code from dereferencing
10577 * loopback devices or network devices that have been freed
10578 * wait here for all pending unregistrations to complete,
10579 * before unregistring the loopback device and allowing the
10580 * network namespace be freed.
10582 * The netdev todo list containing all network devices
10583 * unregistrations that happen in default_device_exit_batch
10584 * will run in the rtnl_unlock() at the end of
10585 * default_device_exit_batch.
10587 rtnl_lock_unregistering(net_list);
10588 list_for_each_entry(net, net_list, exit_list) {
10589 for_each_netdev_reverse(net, dev) {
10590 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
10591 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
10593 unregister_netdevice_queue(dev, &dev_kill_list);
10596 unregister_netdevice_many(&dev_kill_list);
10600 static struct pernet_operations __net_initdata default_device_ops = {
10601 .exit = default_device_exit,
10602 .exit_batch = default_device_exit_batch,
10606 * Initialize the DEV module. At boot time this walks the device list and
10607 * unhooks any devices that fail to initialise (normally hardware not
10608 * present) and leaves us with a valid list of present and active devices.
10613 * This is called single threaded during boot, so no need
10614 * to take the rtnl semaphore.
10616 static int __init net_dev_init(void)
10618 int i, rc = -ENOMEM;
10620 BUG_ON(!dev_boot_phase);
10622 if (dev_proc_init())
10625 if (netdev_kobject_init())
10628 INIT_LIST_HEAD(&ptype_all);
10629 for (i = 0; i < PTYPE_HASH_SIZE; i++)
10630 INIT_LIST_HEAD(&ptype_base[i]);
10632 INIT_LIST_HEAD(&offload_base);
10634 if (register_pernet_subsys(&netdev_net_ops))
10638 * Initialise the packet receive queues.
10641 for_each_possible_cpu(i) {
10642 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
10643 struct softnet_data *sd = &per_cpu(softnet_data, i);
10645 INIT_WORK(flush, flush_backlog);
10647 skb_queue_head_init(&sd->input_pkt_queue);
10648 skb_queue_head_init(&sd->process_queue);
10649 #ifdef CONFIG_XFRM_OFFLOAD
10650 skb_queue_head_init(&sd->xfrm_backlog);
10652 INIT_LIST_HEAD(&sd->poll_list);
10653 sd->output_queue_tailp = &sd->output_queue;
10655 sd->csd.func = rps_trigger_softirq;
10660 init_gro_hash(&sd->backlog);
10661 sd->backlog.poll = process_backlog;
10662 sd->backlog.weight = weight_p;
10665 dev_boot_phase = 0;
10667 /* The loopback device is special if any other network devices
10668 * is present in a network namespace the loopback device must
10669 * be present. Since we now dynamically allocate and free the
10670 * loopback device ensure this invariant is maintained by
10671 * keeping the loopback device as the first device on the
10672 * list of network devices. Ensuring the loopback devices
10673 * is the first device that appears and the last network device
10676 if (register_pernet_device(&loopback_net_ops))
10679 if (register_pernet_device(&default_device_ops))
10682 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
10683 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
10685 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
10686 NULL, dev_cpu_dead);
10693 subsys_initcall(net_dev_init);