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>
103 #include <net/dst_metadata.h>
104 #include <net/pkt_sched.h>
105 #include <net/pkt_cls.h>
106 #include <net/checksum.h>
107 #include <net/xfrm.h>
108 #include <linux/highmem.h>
109 #include <linux/init.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/iw_handler.h>
115 #include <asm/current.h>
116 #include <linux/audit.h>
117 #include <linux/dmaengine.h>
118 #include <linux/err.h>
119 #include <linux/ctype.h>
120 #include <linux/if_arp.h>
121 #include <linux/if_vlan.h>
122 #include <linux/ip.h>
124 #include <net/mpls.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/inetdevice.h>
133 #include <linux/cpu_rmap.h>
134 #include <linux/static_key.h>
135 #include <linux/hashtable.h>
136 #include <linux/vmalloc.h>
137 #include <linux/if_macvlan.h>
138 #include <linux/errqueue.h>
139 #include <linux/hrtimer.h>
140 #include <linux/netfilter_ingress.h>
141 #include <linux/crash_dump.h>
142 #include <linux/sctp.h>
143 #include <net/udp_tunnel.h>
144 #include <linux/net_namespace.h>
145 #include <linux/indirect_call_wrapper.h>
146 #include <net/devlink.h>
147 #include <linux/pm_runtime.h>
148 #include <linux/prandom.h>
150 #include "net-sysfs.h"
152 #define MAX_GRO_SKBS 8
154 /* This should be increased if a protocol with a bigger head is added. */
155 #define GRO_MAX_HEAD (MAX_HEADER + 128)
157 static DEFINE_SPINLOCK(ptype_lock);
158 static DEFINE_SPINLOCK(offload_lock);
159 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
160 struct list_head ptype_all __read_mostly; /* Taps */
161 static struct list_head offload_base __read_mostly;
163 static int netif_rx_internal(struct sk_buff *skb);
164 static int call_netdevice_notifiers_info(unsigned long val,
165 struct netdev_notifier_info *info);
166 static int call_netdevice_notifiers_extack(unsigned long val,
167 struct net_device *dev,
168 struct netlink_ext_ack *extack);
169 static struct napi_struct *napi_by_id(unsigned int napi_id);
172 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
175 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
177 * Writers must hold the rtnl semaphore while they loop through the
178 * dev_base_head list, and hold dev_base_lock for writing when they do the
179 * actual updates. This allows pure readers to access the list even
180 * while a writer is preparing to update it.
182 * To put it another way, dev_base_lock is held for writing only to
183 * protect against pure readers; the rtnl semaphore provides the
184 * protection against other writers.
186 * See, for example usages, register_netdevice() and
187 * unregister_netdevice(), which must be called with the rtnl
190 DEFINE_RWLOCK(dev_base_lock);
191 EXPORT_SYMBOL(dev_base_lock);
193 static DEFINE_MUTEX(ifalias_mutex);
195 /* protects napi_hash addition/deletion and napi_gen_id */
196 static DEFINE_SPINLOCK(napi_hash_lock);
198 static unsigned int napi_gen_id = NR_CPUS;
199 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
201 static DECLARE_RWSEM(devnet_rename_sem);
203 static inline void dev_base_seq_inc(struct net *net)
205 while (++net->dev_base_seq == 0)
209 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
211 unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
213 return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
216 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
218 return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
221 static inline void rps_lock(struct softnet_data *sd)
224 spin_lock(&sd->input_pkt_queue.lock);
228 static inline void rps_unlock(struct softnet_data *sd)
231 spin_unlock(&sd->input_pkt_queue.lock);
235 static struct netdev_name_node *netdev_name_node_alloc(struct net_device *dev,
238 struct netdev_name_node *name_node;
240 name_node = kmalloc(sizeof(*name_node), GFP_KERNEL);
243 INIT_HLIST_NODE(&name_node->hlist);
244 name_node->dev = dev;
245 name_node->name = name;
249 static struct netdev_name_node *
250 netdev_name_node_head_alloc(struct net_device *dev)
252 struct netdev_name_node *name_node;
254 name_node = netdev_name_node_alloc(dev, dev->name);
257 INIT_LIST_HEAD(&name_node->list);
261 static void netdev_name_node_free(struct netdev_name_node *name_node)
266 static void netdev_name_node_add(struct net *net,
267 struct netdev_name_node *name_node)
269 hlist_add_head_rcu(&name_node->hlist,
270 dev_name_hash(net, name_node->name));
273 static void netdev_name_node_del(struct netdev_name_node *name_node)
275 hlist_del_rcu(&name_node->hlist);
278 static struct netdev_name_node *netdev_name_node_lookup(struct net *net,
281 struct hlist_head *head = dev_name_hash(net, name);
282 struct netdev_name_node *name_node;
284 hlist_for_each_entry(name_node, head, hlist)
285 if (!strcmp(name_node->name, name))
290 static struct netdev_name_node *netdev_name_node_lookup_rcu(struct net *net,
293 struct hlist_head *head = dev_name_hash(net, name);
294 struct netdev_name_node *name_node;
296 hlist_for_each_entry_rcu(name_node, head, hlist)
297 if (!strcmp(name_node->name, name))
302 int netdev_name_node_alt_create(struct net_device *dev, const char *name)
304 struct netdev_name_node *name_node;
305 struct net *net = dev_net(dev);
307 name_node = netdev_name_node_lookup(net, name);
310 name_node = netdev_name_node_alloc(dev, name);
313 netdev_name_node_add(net, name_node);
314 /* The node that holds dev->name acts as a head of per-device list. */
315 list_add_tail(&name_node->list, &dev->name_node->list);
319 EXPORT_SYMBOL(netdev_name_node_alt_create);
321 static void __netdev_name_node_alt_destroy(struct netdev_name_node *name_node)
323 list_del(&name_node->list);
324 netdev_name_node_del(name_node);
325 kfree(name_node->name);
326 netdev_name_node_free(name_node);
329 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name)
331 struct netdev_name_node *name_node;
332 struct net *net = dev_net(dev);
334 name_node = netdev_name_node_lookup(net, name);
337 /* lookup might have found our primary name or a name belonging
340 if (name_node == dev->name_node || name_node->dev != dev)
343 __netdev_name_node_alt_destroy(name_node);
347 EXPORT_SYMBOL(netdev_name_node_alt_destroy);
349 static void netdev_name_node_alt_flush(struct net_device *dev)
351 struct netdev_name_node *name_node, *tmp;
353 list_for_each_entry_safe(name_node, tmp, &dev->name_node->list, list)
354 __netdev_name_node_alt_destroy(name_node);
357 /* Device list insertion */
358 static void list_netdevice(struct net_device *dev)
360 struct net *net = dev_net(dev);
364 write_lock_bh(&dev_base_lock);
365 list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
366 netdev_name_node_add(net, dev->name_node);
367 hlist_add_head_rcu(&dev->index_hlist,
368 dev_index_hash(net, dev->ifindex));
369 write_unlock_bh(&dev_base_lock);
371 dev_base_seq_inc(net);
374 /* Device list removal
375 * caller must respect a RCU grace period before freeing/reusing dev
377 static void unlist_netdevice(struct net_device *dev)
381 /* Unlink dev from the device chain */
382 write_lock_bh(&dev_base_lock);
383 list_del_rcu(&dev->dev_list);
384 netdev_name_node_del(dev->name_node);
385 hlist_del_rcu(&dev->index_hlist);
386 write_unlock_bh(&dev_base_lock);
388 dev_base_seq_inc(dev_net(dev));
395 static RAW_NOTIFIER_HEAD(netdev_chain);
398 * Device drivers call our routines to queue packets here. We empty the
399 * queue in the local softnet handler.
402 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
403 EXPORT_PER_CPU_SYMBOL(softnet_data);
405 #ifdef CONFIG_LOCKDEP
407 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
408 * according to dev->type
410 static const unsigned short netdev_lock_type[] = {
411 ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
412 ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
413 ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
414 ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
415 ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
416 ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
417 ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
418 ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
419 ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
420 ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
421 ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
422 ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
423 ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
424 ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
425 ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
427 static const char *const netdev_lock_name[] = {
428 "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
429 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
430 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
431 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
432 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
433 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
434 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
435 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
436 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
437 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
438 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
439 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
440 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
441 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
442 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
444 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
445 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
447 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
451 for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
452 if (netdev_lock_type[i] == dev_type)
454 /* the last key is used by default */
455 return ARRAY_SIZE(netdev_lock_type) - 1;
458 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
459 unsigned short dev_type)
463 i = netdev_lock_pos(dev_type);
464 lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
465 netdev_lock_name[i]);
468 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
472 i = netdev_lock_pos(dev->type);
473 lockdep_set_class_and_name(&dev->addr_list_lock,
474 &netdev_addr_lock_key[i],
475 netdev_lock_name[i]);
478 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
479 unsigned short dev_type)
483 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
488 /*******************************************************************************
490 * Protocol management and registration routines
492 *******************************************************************************/
496 * Add a protocol ID to the list. Now that the input handler is
497 * smarter we can dispense with all the messy stuff that used to be
500 * BEWARE!!! Protocol handlers, mangling input packets,
501 * MUST BE last in hash buckets and checking protocol handlers
502 * MUST start from promiscuous ptype_all chain in net_bh.
503 * It is true now, do not change it.
504 * Explanation follows: if protocol handler, mangling packet, will
505 * be the first on list, it is not able to sense, that packet
506 * is cloned and should be copied-on-write, so that it will
507 * change it and subsequent readers will get broken packet.
511 static inline struct list_head *ptype_head(const struct packet_type *pt)
513 if (pt->type == htons(ETH_P_ALL))
514 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
516 return pt->dev ? &pt->dev->ptype_specific :
517 &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
521 * dev_add_pack - add packet handler
522 * @pt: packet type declaration
524 * Add a protocol handler to the networking stack. The passed &packet_type
525 * is linked into kernel lists and may not be freed until it has been
526 * removed from the kernel lists.
528 * This call does not sleep therefore it can not
529 * guarantee all CPU's that are in middle of receiving packets
530 * will see the new packet type (until the next received packet).
533 void dev_add_pack(struct packet_type *pt)
535 struct list_head *head = ptype_head(pt);
537 spin_lock(&ptype_lock);
538 list_add_rcu(&pt->list, head);
539 spin_unlock(&ptype_lock);
541 EXPORT_SYMBOL(dev_add_pack);
544 * __dev_remove_pack - remove packet handler
545 * @pt: packet type declaration
547 * Remove a protocol handler that was previously added to the kernel
548 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
549 * from the kernel lists and can be freed or reused once this function
552 * The packet type might still be in use by receivers
553 * and must not be freed until after all the CPU's have gone
554 * through a quiescent state.
556 void __dev_remove_pack(struct packet_type *pt)
558 struct list_head *head = ptype_head(pt);
559 struct packet_type *pt1;
561 spin_lock(&ptype_lock);
563 list_for_each_entry(pt1, head, list) {
565 list_del_rcu(&pt->list);
570 pr_warn("dev_remove_pack: %p not found\n", pt);
572 spin_unlock(&ptype_lock);
574 EXPORT_SYMBOL(__dev_remove_pack);
577 * dev_remove_pack - remove packet handler
578 * @pt: packet type declaration
580 * Remove a protocol handler that was previously added to the kernel
581 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
582 * from the kernel lists and can be freed or reused once this function
585 * This call sleeps to guarantee that no CPU is looking at the packet
588 void dev_remove_pack(struct packet_type *pt)
590 __dev_remove_pack(pt);
594 EXPORT_SYMBOL(dev_remove_pack);
598 * dev_add_offload - register offload handlers
599 * @po: protocol offload declaration
601 * Add protocol offload handlers to the networking stack. The passed
602 * &proto_offload is linked into kernel lists and may not be freed until
603 * it has been removed from the kernel lists.
605 * This call does not sleep therefore it can not
606 * guarantee all CPU's that are in middle of receiving packets
607 * will see the new offload handlers (until the next received packet).
609 void dev_add_offload(struct packet_offload *po)
611 struct packet_offload *elem;
613 spin_lock(&offload_lock);
614 list_for_each_entry(elem, &offload_base, list) {
615 if (po->priority < elem->priority)
618 list_add_rcu(&po->list, elem->list.prev);
619 spin_unlock(&offload_lock);
621 EXPORT_SYMBOL(dev_add_offload);
624 * __dev_remove_offload - remove offload handler
625 * @po: packet offload declaration
627 * Remove a protocol offload handler that was previously added to the
628 * kernel offload handlers by dev_add_offload(). The passed &offload_type
629 * is removed from the kernel lists and can be freed or reused once this
632 * The packet type might still be in use by receivers
633 * and must not be freed until after all the CPU's have gone
634 * through a quiescent state.
636 static void __dev_remove_offload(struct packet_offload *po)
638 struct list_head *head = &offload_base;
639 struct packet_offload *po1;
641 spin_lock(&offload_lock);
643 list_for_each_entry(po1, head, list) {
645 list_del_rcu(&po->list);
650 pr_warn("dev_remove_offload: %p not found\n", po);
652 spin_unlock(&offload_lock);
656 * dev_remove_offload - remove packet offload handler
657 * @po: packet offload declaration
659 * Remove a packet offload handler that was previously added to the kernel
660 * offload handlers by dev_add_offload(). The passed &offload_type is
661 * removed from the kernel lists and can be freed or reused once this
664 * This call sleeps to guarantee that no CPU is looking at the packet
667 void dev_remove_offload(struct packet_offload *po)
669 __dev_remove_offload(po);
673 EXPORT_SYMBOL(dev_remove_offload);
675 /******************************************************************************
677 * Device Boot-time Settings Routines
679 ******************************************************************************/
681 /* Boot time configuration table */
682 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
685 * netdev_boot_setup_add - add new setup entry
686 * @name: name of the device
687 * @map: configured settings for the device
689 * Adds new setup entry to the dev_boot_setup list. The function
690 * returns 0 on error and 1 on success. This is a generic routine to
693 static int netdev_boot_setup_add(char *name, struct ifmap *map)
695 struct netdev_boot_setup *s;
699 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
700 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
701 memset(s[i].name, 0, sizeof(s[i].name));
702 strlcpy(s[i].name, name, IFNAMSIZ);
703 memcpy(&s[i].map, map, sizeof(s[i].map));
708 return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
712 * netdev_boot_setup_check - check boot time settings
713 * @dev: the netdevice
715 * Check boot time settings for the device.
716 * The found settings are set for the device to be used
717 * later in the device probing.
718 * Returns 0 if no settings found, 1 if they are.
720 int netdev_boot_setup_check(struct net_device *dev)
722 struct netdev_boot_setup *s = dev_boot_setup;
725 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
726 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
727 !strcmp(dev->name, s[i].name)) {
728 dev->irq = s[i].map.irq;
729 dev->base_addr = s[i].map.base_addr;
730 dev->mem_start = s[i].map.mem_start;
731 dev->mem_end = s[i].map.mem_end;
737 EXPORT_SYMBOL(netdev_boot_setup_check);
741 * netdev_boot_base - get address from boot time settings
742 * @prefix: prefix for network device
743 * @unit: id for network device
745 * Check boot time settings for the base address of device.
746 * The found settings are set for the device to be used
747 * later in the device probing.
748 * Returns 0 if no settings found.
750 unsigned long netdev_boot_base(const char *prefix, int unit)
752 const struct netdev_boot_setup *s = dev_boot_setup;
756 sprintf(name, "%s%d", prefix, unit);
759 * If device already registered then return base of 1
760 * to indicate not to probe for this interface
762 if (__dev_get_by_name(&init_net, name))
765 for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
766 if (!strcmp(name, s[i].name))
767 return s[i].map.base_addr;
772 * Saves at boot time configured settings for any netdevice.
774 int __init netdev_boot_setup(char *str)
779 str = get_options(str, ARRAY_SIZE(ints), ints);
784 memset(&map, 0, sizeof(map));
788 map.base_addr = ints[2];
790 map.mem_start = ints[3];
792 map.mem_end = ints[4];
794 /* Add new entry to the list */
795 return netdev_boot_setup_add(str, &map);
798 __setup("netdev=", netdev_boot_setup);
800 /*******************************************************************************
802 * Device Interface Subroutines
804 *******************************************************************************/
807 * dev_get_iflink - get 'iflink' value of a interface
808 * @dev: targeted interface
810 * Indicates the ifindex the interface is linked to.
811 * Physical interfaces have the same 'ifindex' and 'iflink' values.
814 int dev_get_iflink(const struct net_device *dev)
816 if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
817 return dev->netdev_ops->ndo_get_iflink(dev);
821 EXPORT_SYMBOL(dev_get_iflink);
824 * dev_fill_metadata_dst - Retrieve tunnel egress information.
825 * @dev: targeted interface
828 * For better visibility of tunnel traffic OVS needs to retrieve
829 * egress tunnel information for a packet. Following API allows
830 * user to get this info.
832 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
834 struct ip_tunnel_info *info;
836 if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst)
839 info = skb_tunnel_info_unclone(skb);
842 if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
845 return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
847 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
850 * __dev_get_by_name - find a device by its name
851 * @net: the applicable net namespace
852 * @name: name to find
854 * Find an interface by name. Must be called under RTNL semaphore
855 * or @dev_base_lock. If the name is found a pointer to the device
856 * is returned. If the name is not found then %NULL is returned. The
857 * reference counters are not incremented so the caller must be
858 * careful with locks.
861 struct net_device *__dev_get_by_name(struct net *net, const char *name)
863 struct netdev_name_node *node_name;
865 node_name = netdev_name_node_lookup(net, name);
866 return node_name ? node_name->dev : NULL;
868 EXPORT_SYMBOL(__dev_get_by_name);
871 * dev_get_by_name_rcu - find a device by its name
872 * @net: the applicable net namespace
873 * @name: name to find
875 * Find an interface by name.
876 * If the name is found a pointer to the device is returned.
877 * If the name is not found then %NULL is returned.
878 * The reference counters are not incremented so the caller must be
879 * careful with locks. The caller must hold RCU lock.
882 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
884 struct netdev_name_node *node_name;
886 node_name = netdev_name_node_lookup_rcu(net, name);
887 return node_name ? node_name->dev : NULL;
889 EXPORT_SYMBOL(dev_get_by_name_rcu);
892 * dev_get_by_name - find a device by its name
893 * @net: the applicable net namespace
894 * @name: name to find
896 * Find an interface by name. This can be called from any
897 * context and does its own locking. The returned handle has
898 * the usage count incremented and the caller must use dev_put() to
899 * release it when it is no longer needed. %NULL is returned if no
900 * matching device is found.
903 struct net_device *dev_get_by_name(struct net *net, const char *name)
905 struct net_device *dev;
908 dev = dev_get_by_name_rcu(net, name);
914 EXPORT_SYMBOL(dev_get_by_name);
917 * __dev_get_by_index - find a device by its ifindex
918 * @net: the applicable net namespace
919 * @ifindex: index of device
921 * Search for an interface by index. Returns %NULL if the device
922 * is not found or a pointer to the device. The device has not
923 * had its reference counter increased so the caller must be careful
924 * about locking. The caller must hold either the RTNL semaphore
928 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
930 struct net_device *dev;
931 struct hlist_head *head = dev_index_hash(net, ifindex);
933 hlist_for_each_entry(dev, head, index_hlist)
934 if (dev->ifindex == ifindex)
939 EXPORT_SYMBOL(__dev_get_by_index);
942 * dev_get_by_index_rcu - find a device by its ifindex
943 * @net: the applicable net namespace
944 * @ifindex: index of device
946 * Search for an interface by index. Returns %NULL if the device
947 * is not found or a pointer to the device. The device has not
948 * had its reference counter increased so the caller must be careful
949 * about locking. The caller must hold RCU lock.
952 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
954 struct net_device *dev;
955 struct hlist_head *head = dev_index_hash(net, ifindex);
957 hlist_for_each_entry_rcu(dev, head, index_hlist)
958 if (dev->ifindex == ifindex)
963 EXPORT_SYMBOL(dev_get_by_index_rcu);
967 * dev_get_by_index - find a device by its ifindex
968 * @net: the applicable net namespace
969 * @ifindex: index of device
971 * Search for an interface by index. Returns NULL if the device
972 * is not found or a pointer to the device. The device returned has
973 * had a reference added and the pointer is safe until the user calls
974 * dev_put to indicate they have finished with it.
977 struct net_device *dev_get_by_index(struct net *net, int ifindex)
979 struct net_device *dev;
982 dev = dev_get_by_index_rcu(net, ifindex);
988 EXPORT_SYMBOL(dev_get_by_index);
991 * dev_get_by_napi_id - find a device by napi_id
992 * @napi_id: ID of the NAPI struct
994 * Search for an interface by NAPI ID. Returns %NULL if the device
995 * is not found or a pointer to the device. The device has not had
996 * its reference counter increased so the caller must be careful
997 * about locking. The caller must hold RCU lock.
1000 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
1002 struct napi_struct *napi;
1004 WARN_ON_ONCE(!rcu_read_lock_held());
1006 if (napi_id < MIN_NAPI_ID)
1009 napi = napi_by_id(napi_id);
1011 return napi ? napi->dev : NULL;
1013 EXPORT_SYMBOL(dev_get_by_napi_id);
1016 * netdev_get_name - get a netdevice name, knowing its ifindex.
1017 * @net: network namespace
1018 * @name: a pointer to the buffer where the name will be stored.
1019 * @ifindex: the ifindex of the interface to get the name from.
1021 int netdev_get_name(struct net *net, char *name, int ifindex)
1023 struct net_device *dev;
1026 down_read(&devnet_rename_sem);
1029 dev = dev_get_by_index_rcu(net, ifindex);
1035 strcpy(name, dev->name);
1040 up_read(&devnet_rename_sem);
1045 * dev_getbyhwaddr_rcu - find a device by its hardware address
1046 * @net: the applicable net namespace
1047 * @type: media type of device
1048 * @ha: hardware address
1050 * Search for an interface by MAC address. Returns NULL if the device
1051 * is not found or a pointer to the device.
1052 * The caller must hold RCU or RTNL.
1053 * The returned device has not had its ref count increased
1054 * and the caller must therefore be careful about locking
1058 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
1061 struct net_device *dev;
1063 for_each_netdev_rcu(net, dev)
1064 if (dev->type == type &&
1065 !memcmp(dev->dev_addr, ha, dev->addr_len))
1070 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
1072 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
1074 struct net_device *dev;
1077 for_each_netdev(net, dev)
1078 if (dev->type == type)
1083 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
1085 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
1087 struct net_device *dev, *ret = NULL;
1090 for_each_netdev_rcu(net, dev)
1091 if (dev->type == type) {
1099 EXPORT_SYMBOL(dev_getfirstbyhwtype);
1102 * __dev_get_by_flags - find any device with given flags
1103 * @net: the applicable net namespace
1104 * @if_flags: IFF_* values
1105 * @mask: bitmask of bits in if_flags to check
1107 * Search for any interface with the given flags. Returns NULL if a device
1108 * is not found or a pointer to the device. Must be called inside
1109 * rtnl_lock(), and result refcount is unchanged.
1112 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1113 unsigned short mask)
1115 struct net_device *dev, *ret;
1120 for_each_netdev(net, dev) {
1121 if (((dev->flags ^ if_flags) & mask) == 0) {
1128 EXPORT_SYMBOL(__dev_get_by_flags);
1131 * dev_valid_name - check if name is okay for network device
1132 * @name: name string
1134 * Network device names need to be valid file names to
1135 * allow sysfs to work. We also disallow any kind of
1138 bool dev_valid_name(const char *name)
1142 if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1144 if (!strcmp(name, ".") || !strcmp(name, ".."))
1148 if (*name == '/' || *name == ':' || isspace(*name))
1154 EXPORT_SYMBOL(dev_valid_name);
1157 * __dev_alloc_name - allocate a name for a device
1158 * @net: network namespace to allocate the device name in
1159 * @name: name format string
1160 * @buf: scratch buffer and result name string
1162 * Passed a format string - eg "lt%d" it will try and find a suitable
1163 * id. It scans list of devices to build up a free map, then chooses
1164 * the first empty slot. The caller must hold the dev_base or rtnl lock
1165 * while allocating the name and adding the device in order to avoid
1167 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1168 * Returns the number of the unit assigned or a negative errno code.
1171 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1175 const int max_netdevices = 8*PAGE_SIZE;
1176 unsigned long *inuse;
1177 struct net_device *d;
1179 if (!dev_valid_name(name))
1182 p = strchr(name, '%');
1185 * Verify the string as this thing may have come from
1186 * the user. There must be either one "%d" and no other "%"
1189 if (p[1] != 'd' || strchr(p + 2, '%'))
1192 /* Use one page as a bit array of possible slots */
1193 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1197 for_each_netdev(net, d) {
1198 if (!sscanf(d->name, name, &i))
1200 if (i < 0 || i >= max_netdevices)
1203 /* avoid cases where sscanf is not exact inverse of printf */
1204 snprintf(buf, IFNAMSIZ, name, i);
1205 if (!strncmp(buf, d->name, IFNAMSIZ))
1209 i = find_first_zero_bit(inuse, max_netdevices);
1210 free_page((unsigned long) inuse);
1213 snprintf(buf, IFNAMSIZ, name, i);
1214 if (!__dev_get_by_name(net, buf))
1217 /* It is possible to run out of possible slots
1218 * when the name is long and there isn't enough space left
1219 * for the digits, or if all bits are used.
1224 static int dev_alloc_name_ns(struct net *net,
1225 struct net_device *dev,
1232 ret = __dev_alloc_name(net, name, buf);
1234 strlcpy(dev->name, buf, IFNAMSIZ);
1239 * dev_alloc_name - allocate a name for a device
1241 * @name: name format string
1243 * Passed a format string - eg "lt%d" it will try and find a suitable
1244 * id. It scans list of devices to build up a free map, then chooses
1245 * the first empty slot. The caller must hold the dev_base or rtnl lock
1246 * while allocating the name and adding the device in order to avoid
1248 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1249 * Returns the number of the unit assigned or a negative errno code.
1252 int dev_alloc_name(struct net_device *dev, const char *name)
1254 return dev_alloc_name_ns(dev_net(dev), dev, name);
1256 EXPORT_SYMBOL(dev_alloc_name);
1258 static int dev_get_valid_name(struct net *net, struct net_device *dev,
1263 if (!dev_valid_name(name))
1266 if (strchr(name, '%'))
1267 return dev_alloc_name_ns(net, dev, name);
1268 else if (__dev_get_by_name(net, name))
1270 else if (dev->name != name)
1271 strlcpy(dev->name, name, IFNAMSIZ);
1277 * dev_change_name - change name of a device
1279 * @newname: name (or format string) must be at least IFNAMSIZ
1281 * Change name of a device, can pass format strings "eth%d".
1284 int dev_change_name(struct net_device *dev, const char *newname)
1286 unsigned char old_assign_type;
1287 char oldname[IFNAMSIZ];
1293 BUG_ON(!dev_net(dev));
1297 /* Some auto-enslaved devices e.g. failover slaves are
1298 * special, as userspace might rename the device after
1299 * the interface had been brought up and running since
1300 * the point kernel initiated auto-enslavement. Allow
1301 * live name change even when these slave devices are
1304 * Typically, users of these auto-enslaving devices
1305 * don't actually care about slave name change, as
1306 * they are supposed to operate on master interface
1309 if (dev->flags & IFF_UP &&
1310 likely(!(dev->priv_flags & IFF_LIVE_RENAME_OK)))
1313 down_write(&devnet_rename_sem);
1315 if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1316 up_write(&devnet_rename_sem);
1320 memcpy(oldname, dev->name, IFNAMSIZ);
1322 err = dev_get_valid_name(net, dev, newname);
1324 up_write(&devnet_rename_sem);
1328 if (oldname[0] && !strchr(oldname, '%'))
1329 netdev_info(dev, "renamed from %s\n", oldname);
1331 old_assign_type = dev->name_assign_type;
1332 dev->name_assign_type = NET_NAME_RENAMED;
1335 ret = device_rename(&dev->dev, dev->name);
1337 memcpy(dev->name, oldname, IFNAMSIZ);
1338 dev->name_assign_type = old_assign_type;
1339 up_write(&devnet_rename_sem);
1343 up_write(&devnet_rename_sem);
1345 netdev_adjacent_rename_links(dev, oldname);
1347 write_lock_bh(&dev_base_lock);
1348 netdev_name_node_del(dev->name_node);
1349 write_unlock_bh(&dev_base_lock);
1353 write_lock_bh(&dev_base_lock);
1354 netdev_name_node_add(net, dev->name_node);
1355 write_unlock_bh(&dev_base_lock);
1357 ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1358 ret = notifier_to_errno(ret);
1361 /* err >= 0 after dev_alloc_name() or stores the first errno */
1364 down_write(&devnet_rename_sem);
1365 memcpy(dev->name, oldname, IFNAMSIZ);
1366 memcpy(oldname, newname, IFNAMSIZ);
1367 dev->name_assign_type = old_assign_type;
1368 old_assign_type = NET_NAME_RENAMED;
1371 pr_err("%s: name change rollback failed: %d\n",
1380 * dev_set_alias - change ifalias of a device
1382 * @alias: name up to IFALIASZ
1383 * @len: limit of bytes to copy from info
1385 * Set ifalias for a device,
1387 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1389 struct dev_ifalias *new_alias = NULL;
1391 if (len >= IFALIASZ)
1395 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1399 memcpy(new_alias->ifalias, alias, len);
1400 new_alias->ifalias[len] = 0;
1403 mutex_lock(&ifalias_mutex);
1404 new_alias = rcu_replace_pointer(dev->ifalias, new_alias,
1405 mutex_is_locked(&ifalias_mutex));
1406 mutex_unlock(&ifalias_mutex);
1409 kfree_rcu(new_alias, rcuhead);
1413 EXPORT_SYMBOL(dev_set_alias);
1416 * dev_get_alias - get ifalias of a device
1418 * @name: buffer to store name of ifalias
1419 * @len: size of buffer
1421 * get ifalias for a device. Caller must make sure dev cannot go
1422 * away, e.g. rcu read lock or own a reference count to device.
1424 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1426 const struct dev_ifalias *alias;
1430 alias = rcu_dereference(dev->ifalias);
1432 ret = snprintf(name, len, "%s", alias->ifalias);
1439 * netdev_features_change - device changes features
1440 * @dev: device to cause notification
1442 * Called to indicate a device has changed features.
1444 void netdev_features_change(struct net_device *dev)
1446 call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1448 EXPORT_SYMBOL(netdev_features_change);
1451 * netdev_state_change - device changes state
1452 * @dev: device to cause notification
1454 * Called to indicate a device has changed state. This function calls
1455 * the notifier chains for netdev_chain and sends a NEWLINK message
1456 * to the routing socket.
1458 void netdev_state_change(struct net_device *dev)
1460 if (dev->flags & IFF_UP) {
1461 struct netdev_notifier_change_info change_info = {
1465 call_netdevice_notifiers_info(NETDEV_CHANGE,
1467 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1470 EXPORT_SYMBOL(netdev_state_change);
1473 * netdev_notify_peers - notify network peers about existence of @dev
1474 * @dev: network device
1476 * Generate traffic such that interested network peers are aware of
1477 * @dev, such as by generating a gratuitous ARP. This may be used when
1478 * a device wants to inform the rest of the network about some sort of
1479 * reconfiguration such as a failover event or virtual machine
1482 void netdev_notify_peers(struct net_device *dev)
1485 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1486 call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1489 EXPORT_SYMBOL(netdev_notify_peers);
1491 static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1493 const struct net_device_ops *ops = dev->netdev_ops;
1498 if (!netif_device_present(dev)) {
1499 /* may be detached because parent is runtime-suspended */
1500 if (dev->dev.parent)
1501 pm_runtime_resume(dev->dev.parent);
1502 if (!netif_device_present(dev))
1506 /* Block netpoll from trying to do any rx path servicing.
1507 * If we don't do this there is a chance ndo_poll_controller
1508 * or ndo_poll may be running while we open the device
1510 netpoll_poll_disable(dev);
1512 ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack);
1513 ret = notifier_to_errno(ret);
1517 set_bit(__LINK_STATE_START, &dev->state);
1519 if (ops->ndo_validate_addr)
1520 ret = ops->ndo_validate_addr(dev);
1522 if (!ret && ops->ndo_open)
1523 ret = ops->ndo_open(dev);
1525 netpoll_poll_enable(dev);
1528 clear_bit(__LINK_STATE_START, &dev->state);
1530 dev->flags |= IFF_UP;
1531 dev_set_rx_mode(dev);
1533 add_device_randomness(dev->dev_addr, dev->addr_len);
1540 * dev_open - prepare an interface for use.
1541 * @dev: device to open
1542 * @extack: netlink extended ack
1544 * Takes a device from down to up state. The device's private open
1545 * function is invoked and then the multicast lists are loaded. Finally
1546 * the device is moved into the up state and a %NETDEV_UP message is
1547 * sent to the netdev notifier chain.
1549 * Calling this function on an active interface is a nop. On a failure
1550 * a negative errno code is returned.
1552 int dev_open(struct net_device *dev, struct netlink_ext_ack *extack)
1556 if (dev->flags & IFF_UP)
1559 ret = __dev_open(dev, extack);
1563 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1564 call_netdevice_notifiers(NETDEV_UP, dev);
1568 EXPORT_SYMBOL(dev_open);
1570 static void __dev_close_many(struct list_head *head)
1572 struct net_device *dev;
1577 list_for_each_entry(dev, head, close_list) {
1578 /* Temporarily disable netpoll until the interface is down */
1579 netpoll_poll_disable(dev);
1581 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1583 clear_bit(__LINK_STATE_START, &dev->state);
1585 /* Synchronize to scheduled poll. We cannot touch poll list, it
1586 * can be even on different cpu. So just clear netif_running().
1588 * dev->stop() will invoke napi_disable() on all of it's
1589 * napi_struct instances on this device.
1591 smp_mb__after_atomic(); /* Commit netif_running(). */
1594 dev_deactivate_many(head);
1596 list_for_each_entry(dev, head, close_list) {
1597 const struct net_device_ops *ops = dev->netdev_ops;
1600 * Call the device specific close. This cannot fail.
1601 * Only if device is UP
1603 * We allow it to be called even after a DETACH hot-plug
1609 dev->flags &= ~IFF_UP;
1610 netpoll_poll_enable(dev);
1614 static void __dev_close(struct net_device *dev)
1618 list_add(&dev->close_list, &single);
1619 __dev_close_many(&single);
1623 void dev_close_many(struct list_head *head, bool unlink)
1625 struct net_device *dev, *tmp;
1627 /* Remove the devices that don't need to be closed */
1628 list_for_each_entry_safe(dev, tmp, head, close_list)
1629 if (!(dev->flags & IFF_UP))
1630 list_del_init(&dev->close_list);
1632 __dev_close_many(head);
1634 list_for_each_entry_safe(dev, tmp, head, close_list) {
1635 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1636 call_netdevice_notifiers(NETDEV_DOWN, dev);
1638 list_del_init(&dev->close_list);
1641 EXPORT_SYMBOL(dev_close_many);
1644 * dev_close - shutdown an interface.
1645 * @dev: device to shutdown
1647 * This function moves an active device into down state. A
1648 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1649 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1652 void dev_close(struct net_device *dev)
1654 if (dev->flags & IFF_UP) {
1657 list_add(&dev->close_list, &single);
1658 dev_close_many(&single, true);
1662 EXPORT_SYMBOL(dev_close);
1666 * dev_disable_lro - disable Large Receive Offload on a device
1669 * Disable Large Receive Offload (LRO) on a net device. Must be
1670 * called under RTNL. This is needed if received packets may be
1671 * forwarded to another interface.
1673 void dev_disable_lro(struct net_device *dev)
1675 struct net_device *lower_dev;
1676 struct list_head *iter;
1678 dev->wanted_features &= ~NETIF_F_LRO;
1679 netdev_update_features(dev);
1681 if (unlikely(dev->features & NETIF_F_LRO))
1682 netdev_WARN(dev, "failed to disable LRO!\n");
1684 netdev_for_each_lower_dev(dev, lower_dev, iter)
1685 dev_disable_lro(lower_dev);
1687 EXPORT_SYMBOL(dev_disable_lro);
1690 * dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1693 * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be
1694 * called under RTNL. This is needed if Generic XDP is installed on
1697 static void dev_disable_gro_hw(struct net_device *dev)
1699 dev->wanted_features &= ~NETIF_F_GRO_HW;
1700 netdev_update_features(dev);
1702 if (unlikely(dev->features & NETIF_F_GRO_HW))
1703 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1706 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1709 case NETDEV_##val: \
1710 return "NETDEV_" __stringify(val);
1712 N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1713 N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1714 N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1715 N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1716 N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1717 N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1718 N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1719 N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1720 N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1724 return "UNKNOWN_NETDEV_EVENT";
1726 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1728 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1729 struct net_device *dev)
1731 struct netdev_notifier_info info = {
1735 return nb->notifier_call(nb, val, &info);
1738 static int call_netdevice_register_notifiers(struct notifier_block *nb,
1739 struct net_device *dev)
1743 err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1744 err = notifier_to_errno(err);
1748 if (!(dev->flags & IFF_UP))
1751 call_netdevice_notifier(nb, NETDEV_UP, dev);
1755 static void call_netdevice_unregister_notifiers(struct notifier_block *nb,
1756 struct net_device *dev)
1758 if (dev->flags & IFF_UP) {
1759 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1761 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1763 call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1766 static int call_netdevice_register_net_notifiers(struct notifier_block *nb,
1769 struct net_device *dev;
1772 for_each_netdev(net, dev) {
1773 err = call_netdevice_register_notifiers(nb, dev);
1780 for_each_netdev_continue_reverse(net, dev)
1781 call_netdevice_unregister_notifiers(nb, dev);
1785 static void call_netdevice_unregister_net_notifiers(struct notifier_block *nb,
1788 struct net_device *dev;
1790 for_each_netdev(net, dev)
1791 call_netdevice_unregister_notifiers(nb, dev);
1794 static int dev_boot_phase = 1;
1797 * register_netdevice_notifier - register a network notifier block
1800 * Register a notifier to be called when network device events occur.
1801 * The notifier passed is linked into the kernel structures and must
1802 * not be reused until it has been unregistered. A negative errno code
1803 * is returned on a failure.
1805 * When registered all registration and up events are replayed
1806 * to the new notifier to allow device to have a race free
1807 * view of the network device list.
1810 int register_netdevice_notifier(struct notifier_block *nb)
1815 /* Close race with setup_net() and cleanup_net() */
1816 down_write(&pernet_ops_rwsem);
1818 err = raw_notifier_chain_register(&netdev_chain, nb);
1824 err = call_netdevice_register_net_notifiers(nb, net);
1831 up_write(&pernet_ops_rwsem);
1835 for_each_net_continue_reverse(net)
1836 call_netdevice_unregister_net_notifiers(nb, net);
1838 raw_notifier_chain_unregister(&netdev_chain, nb);
1841 EXPORT_SYMBOL(register_netdevice_notifier);
1844 * unregister_netdevice_notifier - unregister a network notifier block
1847 * Unregister a notifier previously registered by
1848 * register_netdevice_notifier(). The notifier is unlinked into the
1849 * kernel structures and may then be reused. A negative errno code
1850 * is returned on a failure.
1852 * After unregistering unregister and down device events are synthesized
1853 * for all devices on the device list to the removed notifier to remove
1854 * the need for special case cleanup code.
1857 int unregister_netdevice_notifier(struct notifier_block *nb)
1862 /* Close race with setup_net() and cleanup_net() */
1863 down_write(&pernet_ops_rwsem);
1865 err = raw_notifier_chain_unregister(&netdev_chain, nb);
1870 call_netdevice_unregister_net_notifiers(nb, net);
1874 up_write(&pernet_ops_rwsem);
1877 EXPORT_SYMBOL(unregister_netdevice_notifier);
1879 static int __register_netdevice_notifier_net(struct net *net,
1880 struct notifier_block *nb,
1881 bool ignore_call_fail)
1885 err = raw_notifier_chain_register(&net->netdev_chain, nb);
1891 err = call_netdevice_register_net_notifiers(nb, net);
1892 if (err && !ignore_call_fail)
1893 goto chain_unregister;
1898 raw_notifier_chain_unregister(&net->netdev_chain, nb);
1902 static int __unregister_netdevice_notifier_net(struct net *net,
1903 struct notifier_block *nb)
1907 err = raw_notifier_chain_unregister(&net->netdev_chain, nb);
1911 call_netdevice_unregister_net_notifiers(nb, net);
1916 * register_netdevice_notifier_net - register a per-netns network notifier block
1917 * @net: network namespace
1920 * Register a notifier to be called when network device events occur.
1921 * The notifier passed is linked into the kernel structures and must
1922 * not be reused until it has been unregistered. A negative errno code
1923 * is returned on a failure.
1925 * When registered all registration and up events are replayed
1926 * to the new notifier to allow device to have a race free
1927 * view of the network device list.
1930 int register_netdevice_notifier_net(struct net *net, struct notifier_block *nb)
1935 err = __register_netdevice_notifier_net(net, nb, false);
1939 EXPORT_SYMBOL(register_netdevice_notifier_net);
1942 * unregister_netdevice_notifier_net - unregister a per-netns
1943 * network notifier block
1944 * @net: network namespace
1947 * Unregister a notifier previously registered by
1948 * register_netdevice_notifier(). The notifier is unlinked into the
1949 * kernel structures and may then be reused. A negative errno code
1950 * is returned on a failure.
1952 * After unregistering unregister and down device events are synthesized
1953 * for all devices on the device list to the removed notifier to remove
1954 * the need for special case cleanup code.
1957 int unregister_netdevice_notifier_net(struct net *net,
1958 struct notifier_block *nb)
1963 err = __unregister_netdevice_notifier_net(net, nb);
1967 EXPORT_SYMBOL(unregister_netdevice_notifier_net);
1969 int register_netdevice_notifier_dev_net(struct net_device *dev,
1970 struct notifier_block *nb,
1971 struct netdev_net_notifier *nn)
1976 err = __register_netdevice_notifier_net(dev_net(dev), nb, false);
1979 list_add(&nn->list, &dev->net_notifier_list);
1984 EXPORT_SYMBOL(register_netdevice_notifier_dev_net);
1986 int unregister_netdevice_notifier_dev_net(struct net_device *dev,
1987 struct notifier_block *nb,
1988 struct netdev_net_notifier *nn)
1993 list_del(&nn->list);
1994 err = __unregister_netdevice_notifier_net(dev_net(dev), nb);
1998 EXPORT_SYMBOL(unregister_netdevice_notifier_dev_net);
2000 static void move_netdevice_notifiers_dev_net(struct net_device *dev,
2003 struct netdev_net_notifier *nn;
2005 list_for_each_entry(nn, &dev->net_notifier_list, list) {
2006 __unregister_netdevice_notifier_net(dev_net(dev), nn->nb);
2007 __register_netdevice_notifier_net(net, nn->nb, true);
2012 * call_netdevice_notifiers_info - call all network notifier blocks
2013 * @val: value passed unmodified to notifier function
2014 * @info: notifier information data
2016 * Call all network notifier blocks. Parameters and return value
2017 * are as for raw_notifier_call_chain().
2020 static int call_netdevice_notifiers_info(unsigned long val,
2021 struct netdev_notifier_info *info)
2023 struct net *net = dev_net(info->dev);
2028 /* Run per-netns notifier block chain first, then run the global one.
2029 * Hopefully, one day, the global one is going to be removed after
2030 * all notifier block registrators get converted to be per-netns.
2032 ret = raw_notifier_call_chain(&net->netdev_chain, val, info);
2033 if (ret & NOTIFY_STOP_MASK)
2035 return raw_notifier_call_chain(&netdev_chain, val, info);
2038 static int call_netdevice_notifiers_extack(unsigned long val,
2039 struct net_device *dev,
2040 struct netlink_ext_ack *extack)
2042 struct netdev_notifier_info info = {
2047 return call_netdevice_notifiers_info(val, &info);
2051 * call_netdevice_notifiers - call all network notifier blocks
2052 * @val: value passed unmodified to notifier function
2053 * @dev: net_device pointer passed unmodified to notifier function
2055 * Call all network notifier blocks. Parameters and return value
2056 * are as for raw_notifier_call_chain().
2059 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
2061 return call_netdevice_notifiers_extack(val, dev, NULL);
2063 EXPORT_SYMBOL(call_netdevice_notifiers);
2066 * call_netdevice_notifiers_mtu - call all network notifier blocks
2067 * @val: value passed unmodified to notifier function
2068 * @dev: net_device pointer passed unmodified to notifier function
2069 * @arg: additional u32 argument passed to the notifier function
2071 * Call all network notifier blocks. Parameters and return value
2072 * are as for raw_notifier_call_chain().
2074 static int call_netdevice_notifiers_mtu(unsigned long val,
2075 struct net_device *dev, u32 arg)
2077 struct netdev_notifier_info_ext info = {
2082 BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0);
2084 return call_netdevice_notifiers_info(val, &info.info);
2087 #ifdef CONFIG_NET_INGRESS
2088 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
2090 void net_inc_ingress_queue(void)
2092 static_branch_inc(&ingress_needed_key);
2094 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
2096 void net_dec_ingress_queue(void)
2098 static_branch_dec(&ingress_needed_key);
2100 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
2103 #ifdef CONFIG_NET_EGRESS
2104 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
2106 void net_inc_egress_queue(void)
2108 static_branch_inc(&egress_needed_key);
2110 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
2112 void net_dec_egress_queue(void)
2114 static_branch_dec(&egress_needed_key);
2116 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
2119 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
2120 #ifdef CONFIG_JUMP_LABEL
2121 static atomic_t netstamp_needed_deferred;
2122 static atomic_t netstamp_wanted;
2123 static void netstamp_clear(struct work_struct *work)
2125 int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
2128 wanted = atomic_add_return(deferred, &netstamp_wanted);
2130 static_branch_enable(&netstamp_needed_key);
2132 static_branch_disable(&netstamp_needed_key);
2134 static DECLARE_WORK(netstamp_work, netstamp_clear);
2137 void net_enable_timestamp(void)
2139 #ifdef CONFIG_JUMP_LABEL
2143 wanted = atomic_read(&netstamp_wanted);
2146 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
2149 atomic_inc(&netstamp_needed_deferred);
2150 schedule_work(&netstamp_work);
2152 static_branch_inc(&netstamp_needed_key);
2155 EXPORT_SYMBOL(net_enable_timestamp);
2157 void net_disable_timestamp(void)
2159 #ifdef CONFIG_JUMP_LABEL
2163 wanted = atomic_read(&netstamp_wanted);
2166 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
2169 atomic_dec(&netstamp_needed_deferred);
2170 schedule_work(&netstamp_work);
2172 static_branch_dec(&netstamp_needed_key);
2175 EXPORT_SYMBOL(net_disable_timestamp);
2177 static inline void net_timestamp_set(struct sk_buff *skb)
2180 if (static_branch_unlikely(&netstamp_needed_key))
2181 __net_timestamp(skb);
2184 #define net_timestamp_check(COND, SKB) \
2185 if (static_branch_unlikely(&netstamp_needed_key)) { \
2186 if ((COND) && !(SKB)->tstamp) \
2187 __net_timestamp(SKB); \
2190 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
2194 if (!(dev->flags & IFF_UP))
2197 len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
2198 if (skb->len <= len)
2201 /* if TSO is enabled, we don't care about the length as the packet
2202 * could be forwarded without being segmented before
2204 if (skb_is_gso(skb))
2209 EXPORT_SYMBOL_GPL(is_skb_forwardable);
2211 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2213 int ret = ____dev_forward_skb(dev, skb);
2216 skb->protocol = eth_type_trans(skb, dev);
2217 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
2222 EXPORT_SYMBOL_GPL(__dev_forward_skb);
2225 * dev_forward_skb - loopback an skb to another netif
2227 * @dev: destination network device
2228 * @skb: buffer to forward
2231 * NET_RX_SUCCESS (no congestion)
2232 * NET_RX_DROP (packet was dropped, but freed)
2234 * dev_forward_skb can be used for injecting an skb from the
2235 * start_xmit function of one device into the receive queue
2236 * of another device.
2238 * The receiving device may be in another namespace, so
2239 * we have to clear all information in the skb that could
2240 * impact namespace isolation.
2242 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
2244 return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
2246 EXPORT_SYMBOL_GPL(dev_forward_skb);
2248 static inline int deliver_skb(struct sk_buff *skb,
2249 struct packet_type *pt_prev,
2250 struct net_device *orig_dev)
2252 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
2254 refcount_inc(&skb->users);
2255 return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
2258 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
2259 struct packet_type **pt,
2260 struct net_device *orig_dev,
2262 struct list_head *ptype_list)
2264 struct packet_type *ptype, *pt_prev = *pt;
2266 list_for_each_entry_rcu(ptype, ptype_list, list) {
2267 if (ptype->type != type)
2270 deliver_skb(skb, pt_prev, orig_dev);
2276 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
2278 if (!ptype->af_packet_priv || !skb->sk)
2281 if (ptype->id_match)
2282 return ptype->id_match(ptype, skb->sk);
2283 else if ((struct sock *)ptype->af_packet_priv == skb->sk)
2290 * dev_nit_active - return true if any network interface taps are in use
2292 * @dev: network device to check for the presence of taps
2294 bool dev_nit_active(struct net_device *dev)
2296 return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all);
2298 EXPORT_SYMBOL_GPL(dev_nit_active);
2301 * Support routine. Sends outgoing frames to any network
2302 * taps currently in use.
2305 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
2307 struct packet_type *ptype;
2308 struct sk_buff *skb2 = NULL;
2309 struct packet_type *pt_prev = NULL;
2310 struct list_head *ptype_list = &ptype_all;
2314 list_for_each_entry_rcu(ptype, ptype_list, list) {
2315 if (ptype->ignore_outgoing)
2318 /* Never send packets back to the socket
2319 * they originated from - MvS (miquels@drinkel.ow.org)
2321 if (skb_loop_sk(ptype, skb))
2325 deliver_skb(skb2, pt_prev, skb->dev);
2330 /* need to clone skb, done only once */
2331 skb2 = skb_clone(skb, GFP_ATOMIC);
2335 net_timestamp_set(skb2);
2337 /* skb->nh should be correctly
2338 * set by sender, so that the second statement is
2339 * just protection against buggy protocols.
2341 skb_reset_mac_header(skb2);
2343 if (skb_network_header(skb2) < skb2->data ||
2344 skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2345 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2346 ntohs(skb2->protocol),
2348 skb_reset_network_header(skb2);
2351 skb2->transport_header = skb2->network_header;
2352 skb2->pkt_type = PACKET_OUTGOING;
2356 if (ptype_list == &ptype_all) {
2357 ptype_list = &dev->ptype_all;
2362 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2363 pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2369 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2372 * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2373 * @dev: Network device
2374 * @txq: number of queues available
2376 * If real_num_tx_queues is changed the tc mappings may no longer be
2377 * valid. To resolve this verify the tc mapping remains valid and if
2378 * not NULL the mapping. With no priorities mapping to this
2379 * offset/count pair it will no longer be used. In the worst case TC0
2380 * is invalid nothing can be done so disable priority mappings. If is
2381 * expected that drivers will fix this mapping if they can before
2382 * calling netif_set_real_num_tx_queues.
2384 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2387 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2389 /* If TC0 is invalidated disable TC mapping */
2390 if (tc->offset + tc->count > txq) {
2391 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2396 /* Invalidated prio to tc mappings set to TC0 */
2397 for (i = 1; i < TC_BITMASK + 1; i++) {
2398 int q = netdev_get_prio_tc_map(dev, i);
2400 tc = &dev->tc_to_txq[q];
2401 if (tc->offset + tc->count > txq) {
2402 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2404 netdev_set_prio_tc_map(dev, i, 0);
2409 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2412 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2415 /* walk through the TCs and see if it falls into any of them */
2416 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2417 if ((txq - tc->offset) < tc->count)
2421 /* didn't find it, just return -1 to indicate no match */
2427 EXPORT_SYMBOL(netdev_txq_to_tc);
2430 struct static_key xps_needed __read_mostly;
2431 EXPORT_SYMBOL(xps_needed);
2432 struct static_key xps_rxqs_needed __read_mostly;
2433 EXPORT_SYMBOL(xps_rxqs_needed);
2434 static DEFINE_MUTEX(xps_map_mutex);
2435 #define xmap_dereference(P) \
2436 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2438 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2441 struct xps_map *map = NULL;
2445 map = xmap_dereference(dev_maps->attr_map[tci]);
2449 for (pos = map->len; pos--;) {
2450 if (map->queues[pos] != index)
2454 map->queues[pos] = map->queues[--map->len];
2458 RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL);
2459 kfree_rcu(map, rcu);
2466 static bool remove_xps_queue_cpu(struct net_device *dev,
2467 struct xps_dev_maps *dev_maps,
2468 int cpu, u16 offset, u16 count)
2470 int num_tc = dev->num_tc ? : 1;
2471 bool active = false;
2474 for (tci = cpu * num_tc; num_tc--; tci++) {
2477 for (i = count, j = offset; i--; j++) {
2478 if (!remove_xps_queue(dev_maps, tci, j))
2488 static void reset_xps_maps(struct net_device *dev,
2489 struct xps_dev_maps *dev_maps,
2493 static_key_slow_dec_cpuslocked(&xps_rxqs_needed);
2494 RCU_INIT_POINTER(dev->xps_rxqs_map, NULL);
2496 RCU_INIT_POINTER(dev->xps_cpus_map, NULL);
2498 static_key_slow_dec_cpuslocked(&xps_needed);
2499 kfree_rcu(dev_maps, rcu);
2502 static void clean_xps_maps(struct net_device *dev, const unsigned long *mask,
2503 struct xps_dev_maps *dev_maps, unsigned int nr_ids,
2504 u16 offset, u16 count, bool is_rxqs_map)
2506 bool active = false;
2509 for (j = -1; j = netif_attrmask_next(j, mask, nr_ids),
2511 active |= remove_xps_queue_cpu(dev, dev_maps, j, offset,
2514 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2517 for (i = offset + (count - 1); count--; i--) {
2518 netdev_queue_numa_node_write(
2519 netdev_get_tx_queue(dev, i),
2525 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2528 const unsigned long *possible_mask = NULL;
2529 struct xps_dev_maps *dev_maps;
2530 unsigned int nr_ids;
2532 if (!static_key_false(&xps_needed))
2536 mutex_lock(&xps_map_mutex);
2538 if (static_key_false(&xps_rxqs_needed)) {
2539 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2541 nr_ids = dev->num_rx_queues;
2542 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids,
2543 offset, count, true);
2547 dev_maps = xmap_dereference(dev->xps_cpus_map);
2551 if (num_possible_cpus() > 1)
2552 possible_mask = cpumask_bits(cpu_possible_mask);
2553 nr_ids = nr_cpu_ids;
2554 clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count,
2558 mutex_unlock(&xps_map_mutex);
2562 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2564 netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2567 static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index,
2568 u16 index, bool is_rxqs_map)
2570 struct xps_map *new_map;
2571 int alloc_len = XPS_MIN_MAP_ALLOC;
2574 for (pos = 0; map && pos < map->len; pos++) {
2575 if (map->queues[pos] != index)
2580 /* Need to add tx-queue to this CPU's/rx-queue's existing map */
2582 if (pos < map->alloc_len)
2585 alloc_len = map->alloc_len * 2;
2588 /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's
2592 new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL);
2594 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2595 cpu_to_node(attr_index));
2599 for (i = 0; i < pos; i++)
2600 new_map->queues[i] = map->queues[i];
2601 new_map->alloc_len = alloc_len;
2607 /* Must be called under cpus_read_lock */
2608 int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask,
2609 u16 index, bool is_rxqs_map)
2611 const unsigned long *online_mask = NULL, *possible_mask = NULL;
2612 struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2613 int i, j, tci, numa_node_id = -2;
2614 int maps_sz, num_tc = 1, tc = 0;
2615 struct xps_map *map, *new_map;
2616 bool active = false;
2617 unsigned int nr_ids;
2620 /* Do not allow XPS on subordinate device directly */
2621 num_tc = dev->num_tc;
2625 /* If queue belongs to subordinate dev use its map */
2626 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
2628 tc = netdev_txq_to_tc(dev, index);
2633 mutex_lock(&xps_map_mutex);
2635 maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues);
2636 dev_maps = xmap_dereference(dev->xps_rxqs_map);
2637 nr_ids = dev->num_rx_queues;
2639 maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc);
2640 if (num_possible_cpus() > 1) {
2641 online_mask = cpumask_bits(cpu_online_mask);
2642 possible_mask = cpumask_bits(cpu_possible_mask);
2644 dev_maps = xmap_dereference(dev->xps_cpus_map);
2645 nr_ids = nr_cpu_ids;
2648 if (maps_sz < L1_CACHE_BYTES)
2649 maps_sz = L1_CACHE_BYTES;
2651 /* allocate memory for queue storage */
2652 for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids),
2655 new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2656 if (!new_dev_maps) {
2657 mutex_unlock(&xps_map_mutex);
2661 tci = j * num_tc + tc;
2662 map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) :
2665 map = expand_xps_map(map, j, index, is_rxqs_map);
2669 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2673 goto out_no_new_maps;
2676 /* Increment static keys at most once per type */
2677 static_key_slow_inc_cpuslocked(&xps_needed);
2679 static_key_slow_inc_cpuslocked(&xps_rxqs_needed);
2682 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2684 /* copy maps belonging to foreign traffic classes */
2685 for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) {
2686 /* fill in the new device map from the old device map */
2687 map = xmap_dereference(dev_maps->attr_map[tci]);
2688 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2691 /* We need to explicitly update tci as prevous loop
2692 * could break out early if dev_maps is NULL.
2694 tci = j * num_tc + tc;
2696 if (netif_attr_test_mask(j, mask, nr_ids) &&
2697 netif_attr_test_online(j, online_mask, nr_ids)) {
2698 /* add tx-queue to CPU/rx-queue maps */
2701 map = xmap_dereference(new_dev_maps->attr_map[tci]);
2702 while ((pos < map->len) && (map->queues[pos] != index))
2705 if (pos == map->len)
2706 map->queues[map->len++] = index;
2709 if (numa_node_id == -2)
2710 numa_node_id = cpu_to_node(j);
2711 else if (numa_node_id != cpu_to_node(j))
2715 } else if (dev_maps) {
2716 /* fill in the new device map from the old device map */
2717 map = xmap_dereference(dev_maps->attr_map[tci]);
2718 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2721 /* copy maps belonging to foreign traffic classes */
2722 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2723 /* fill in the new device map from the old device map */
2724 map = xmap_dereference(dev_maps->attr_map[tci]);
2725 RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map);
2730 rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps);
2732 rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps);
2734 /* Cleanup old maps */
2736 goto out_no_old_maps;
2738 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2740 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2741 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2742 map = xmap_dereference(dev_maps->attr_map[tci]);
2743 if (map && map != new_map)
2744 kfree_rcu(map, rcu);
2748 kfree_rcu(dev_maps, rcu);
2751 dev_maps = new_dev_maps;
2756 /* update Tx queue numa node */
2757 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2758 (numa_node_id >= 0) ?
2759 numa_node_id : NUMA_NO_NODE);
2765 /* removes tx-queue from unused CPUs/rx-queues */
2766 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2768 for (i = tc, tci = j * num_tc; i--; tci++)
2769 active |= remove_xps_queue(dev_maps, tci, index);
2770 if (!netif_attr_test_mask(j, mask, nr_ids) ||
2771 !netif_attr_test_online(j, online_mask, nr_ids))
2772 active |= remove_xps_queue(dev_maps, tci, index);
2773 for (i = num_tc - tc, tci++; --i; tci++)
2774 active |= remove_xps_queue(dev_maps, tci, index);
2777 /* free map if not active */
2779 reset_xps_maps(dev, dev_maps, is_rxqs_map);
2782 mutex_unlock(&xps_map_mutex);
2786 /* remove any maps that we added */
2787 for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids),
2789 for (i = num_tc, tci = j * num_tc; i--; tci++) {
2790 new_map = xmap_dereference(new_dev_maps->attr_map[tci]);
2792 xmap_dereference(dev_maps->attr_map[tci]) :
2794 if (new_map && new_map != map)
2799 mutex_unlock(&xps_map_mutex);
2801 kfree(new_dev_maps);
2804 EXPORT_SYMBOL_GPL(__netif_set_xps_queue);
2806 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2812 ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false);
2817 EXPORT_SYMBOL(netif_set_xps_queue);
2820 static void netdev_unbind_all_sb_channels(struct net_device *dev)
2822 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2824 /* Unbind any subordinate channels */
2825 while (txq-- != &dev->_tx[0]) {
2827 netdev_unbind_sb_channel(dev, txq->sb_dev);
2831 void netdev_reset_tc(struct net_device *dev)
2834 netif_reset_xps_queues_gt(dev, 0);
2836 netdev_unbind_all_sb_channels(dev);
2838 /* Reset TC configuration of device */
2840 memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2841 memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2843 EXPORT_SYMBOL(netdev_reset_tc);
2845 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2847 if (tc >= dev->num_tc)
2851 netif_reset_xps_queues(dev, offset, count);
2853 dev->tc_to_txq[tc].count = count;
2854 dev->tc_to_txq[tc].offset = offset;
2857 EXPORT_SYMBOL(netdev_set_tc_queue);
2859 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2861 if (num_tc > TC_MAX_QUEUE)
2865 netif_reset_xps_queues_gt(dev, 0);
2867 netdev_unbind_all_sb_channels(dev);
2869 dev->num_tc = num_tc;
2872 EXPORT_SYMBOL(netdev_set_num_tc);
2874 void netdev_unbind_sb_channel(struct net_device *dev,
2875 struct net_device *sb_dev)
2877 struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues];
2880 netif_reset_xps_queues_gt(sb_dev, 0);
2882 memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq));
2883 memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map));
2885 while (txq-- != &dev->_tx[0]) {
2886 if (txq->sb_dev == sb_dev)
2890 EXPORT_SYMBOL(netdev_unbind_sb_channel);
2892 int netdev_bind_sb_channel_queue(struct net_device *dev,
2893 struct net_device *sb_dev,
2894 u8 tc, u16 count, u16 offset)
2896 /* Make certain the sb_dev and dev are already configured */
2897 if (sb_dev->num_tc >= 0 || tc >= dev->num_tc)
2900 /* We cannot hand out queues we don't have */
2901 if ((offset + count) > dev->real_num_tx_queues)
2904 /* Record the mapping */
2905 sb_dev->tc_to_txq[tc].count = count;
2906 sb_dev->tc_to_txq[tc].offset = offset;
2908 /* Provide a way for Tx queue to find the tc_to_txq map or
2909 * XPS map for itself.
2912 netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev;
2916 EXPORT_SYMBOL(netdev_bind_sb_channel_queue);
2918 int netdev_set_sb_channel(struct net_device *dev, u16 channel)
2920 /* Do not use a multiqueue device to represent a subordinate channel */
2921 if (netif_is_multiqueue(dev))
2924 /* We allow channels 1 - 32767 to be used for subordinate channels.
2925 * Channel 0 is meant to be "native" mode and used only to represent
2926 * the main root device. We allow writing 0 to reset the device back
2927 * to normal mode after being used as a subordinate channel.
2929 if (channel > S16_MAX)
2932 dev->num_tc = -channel;
2936 EXPORT_SYMBOL(netdev_set_sb_channel);
2939 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2940 * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2942 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2947 disabling = txq < dev->real_num_tx_queues;
2949 if (txq < 1 || txq > dev->num_tx_queues)
2952 if (dev->reg_state == NETREG_REGISTERED ||
2953 dev->reg_state == NETREG_UNREGISTERING) {
2956 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2962 netif_setup_tc(dev, txq);
2964 dev->real_num_tx_queues = txq;
2968 qdisc_reset_all_tx_gt(dev, txq);
2970 netif_reset_xps_queues_gt(dev, txq);
2974 dev->real_num_tx_queues = txq;
2979 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2983 * netif_set_real_num_rx_queues - set actual number of RX queues used
2984 * @dev: Network device
2985 * @rxq: Actual number of RX queues
2987 * This must be called either with the rtnl_lock held or before
2988 * registration of the net device. Returns 0 on success, or a
2989 * negative error code. If called before registration, it always
2992 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2996 if (rxq < 1 || rxq > dev->num_rx_queues)
2999 if (dev->reg_state == NETREG_REGISTERED) {
3002 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
3008 dev->real_num_rx_queues = rxq;
3011 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
3015 * netif_get_num_default_rss_queues - default number of RSS queues
3017 * This routine should set an upper limit on the number of RSS queues
3018 * used by default by multiqueue devices.
3020 int netif_get_num_default_rss_queues(void)
3022 return is_kdump_kernel() ?
3023 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
3025 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
3027 static void __netif_reschedule(struct Qdisc *q)
3029 struct softnet_data *sd;
3030 unsigned long flags;
3032 local_irq_save(flags);
3033 sd = this_cpu_ptr(&softnet_data);
3034 q->next_sched = NULL;
3035 *sd->output_queue_tailp = q;
3036 sd->output_queue_tailp = &q->next_sched;
3037 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3038 local_irq_restore(flags);
3041 void __netif_schedule(struct Qdisc *q)
3043 if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
3044 __netif_reschedule(q);
3046 EXPORT_SYMBOL(__netif_schedule);
3048 struct dev_kfree_skb_cb {
3049 enum skb_free_reason reason;
3052 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
3054 return (struct dev_kfree_skb_cb *)skb->cb;
3057 void netif_schedule_queue(struct netdev_queue *txq)
3060 if (!netif_xmit_stopped(txq)) {
3061 struct Qdisc *q = rcu_dereference(txq->qdisc);
3063 __netif_schedule(q);
3067 EXPORT_SYMBOL(netif_schedule_queue);
3069 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
3071 if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
3075 q = rcu_dereference(dev_queue->qdisc);
3076 __netif_schedule(q);
3080 EXPORT_SYMBOL(netif_tx_wake_queue);
3082 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
3084 unsigned long flags;
3089 if (likely(refcount_read(&skb->users) == 1)) {
3091 refcount_set(&skb->users, 0);
3092 } else if (likely(!refcount_dec_and_test(&skb->users))) {
3095 get_kfree_skb_cb(skb)->reason = reason;
3096 local_irq_save(flags);
3097 skb->next = __this_cpu_read(softnet_data.completion_queue);
3098 __this_cpu_write(softnet_data.completion_queue, skb);
3099 raise_softirq_irqoff(NET_TX_SOFTIRQ);
3100 local_irq_restore(flags);
3102 EXPORT_SYMBOL(__dev_kfree_skb_irq);
3104 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
3106 if (in_irq() || irqs_disabled())
3107 __dev_kfree_skb_irq(skb, reason);
3111 EXPORT_SYMBOL(__dev_kfree_skb_any);
3115 * netif_device_detach - mark device as removed
3116 * @dev: network device
3118 * Mark device as removed from system and therefore no longer available.
3120 void netif_device_detach(struct net_device *dev)
3122 if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
3123 netif_running(dev)) {
3124 netif_tx_stop_all_queues(dev);
3127 EXPORT_SYMBOL(netif_device_detach);
3130 * netif_device_attach - mark device as attached
3131 * @dev: network device
3133 * Mark device as attached from system and restart if needed.
3135 void netif_device_attach(struct net_device *dev)
3137 if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
3138 netif_running(dev)) {
3139 netif_tx_wake_all_queues(dev);
3140 __netdev_watchdog_up(dev);
3143 EXPORT_SYMBOL(netif_device_attach);
3146 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
3147 * to be used as a distribution range.
3149 static u16 skb_tx_hash(const struct net_device *dev,
3150 const struct net_device *sb_dev,
3151 struct sk_buff *skb)
3155 u16 qcount = dev->real_num_tx_queues;
3158 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
3160 qoffset = sb_dev->tc_to_txq[tc].offset;
3161 qcount = sb_dev->tc_to_txq[tc].count;
3164 if (skb_rx_queue_recorded(skb)) {
3165 hash = skb_get_rx_queue(skb);
3166 if (hash >= qoffset)
3168 while (unlikely(hash >= qcount))
3170 return hash + qoffset;
3173 return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
3176 static void skb_warn_bad_offload(const struct sk_buff *skb)
3178 static const netdev_features_t null_features;
3179 struct net_device *dev = skb->dev;
3180 const char *name = "";
3182 if (!net_ratelimit())
3186 if (dev->dev.parent)
3187 name = dev_driver_string(dev->dev.parent);
3189 name = netdev_name(dev);
3191 skb_dump(KERN_WARNING, skb, false);
3192 WARN(1, "%s: caps=(%pNF, %pNF)\n",
3193 name, dev ? &dev->features : &null_features,
3194 skb->sk ? &skb->sk->sk_route_caps : &null_features);
3198 * Invalidate hardware checksum when packet is to be mangled, and
3199 * complete checksum manually on outgoing path.
3201 int skb_checksum_help(struct sk_buff *skb)
3204 int ret = 0, offset;
3206 if (skb->ip_summed == CHECKSUM_COMPLETE)
3207 goto out_set_summed;
3209 if (unlikely(skb_shinfo(skb)->gso_size)) {
3210 skb_warn_bad_offload(skb);
3214 /* Before computing a checksum, we should make sure no frag could
3215 * be modified by an external entity : checksum could be wrong.
3217 if (skb_has_shared_frag(skb)) {
3218 ret = __skb_linearize(skb);
3223 offset = skb_checksum_start_offset(skb);
3224 BUG_ON(offset >= skb_headlen(skb));
3225 csum = skb_checksum(skb, offset, skb->len - offset, 0);
3227 offset += skb->csum_offset;
3228 BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
3230 ret = skb_ensure_writable(skb, offset + sizeof(__sum16));
3234 *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
3236 skb->ip_summed = CHECKSUM_NONE;
3240 EXPORT_SYMBOL(skb_checksum_help);
3242 int skb_crc32c_csum_help(struct sk_buff *skb)
3245 int ret = 0, offset, start;
3247 if (skb->ip_summed != CHECKSUM_PARTIAL)
3250 if (unlikely(skb_is_gso(skb)))
3253 /* Before computing a checksum, we should make sure no frag could
3254 * be modified by an external entity : checksum could be wrong.
3256 if (unlikely(skb_has_shared_frag(skb))) {
3257 ret = __skb_linearize(skb);
3261 start = skb_checksum_start_offset(skb);
3262 offset = start + offsetof(struct sctphdr, checksum);
3263 if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
3268 ret = skb_ensure_writable(skb, offset + sizeof(__le32));
3272 crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
3273 skb->len - start, ~(__u32)0,
3275 *(__le32 *)(skb->data + offset) = crc32c_csum;
3276 skb->ip_summed = CHECKSUM_NONE;
3277 skb->csum_not_inet = 0;
3282 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
3284 __be16 type = skb->protocol;
3286 /* Tunnel gso handlers can set protocol to ethernet. */
3287 if (type == htons(ETH_P_TEB)) {
3290 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
3293 eth = (struct ethhdr *)skb->data;
3294 type = eth->h_proto;
3297 return __vlan_get_protocol(skb, type, depth);
3301 * skb_mac_gso_segment - mac layer segmentation handler.
3302 * @skb: buffer to segment
3303 * @features: features for the output path (see dev->features)
3305 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
3306 netdev_features_t features)
3308 struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
3309 struct packet_offload *ptype;
3310 int vlan_depth = skb->mac_len;
3311 __be16 type = skb_network_protocol(skb, &vlan_depth);
3313 if (unlikely(!type))
3314 return ERR_PTR(-EINVAL);
3316 __skb_pull(skb, vlan_depth);
3319 list_for_each_entry_rcu(ptype, &offload_base, list) {
3320 if (ptype->type == type && ptype->callbacks.gso_segment) {
3321 segs = ptype->callbacks.gso_segment(skb, features);
3327 __skb_push(skb, skb->data - skb_mac_header(skb));
3331 EXPORT_SYMBOL(skb_mac_gso_segment);
3334 /* openvswitch calls this on rx path, so we need a different check.
3336 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
3339 return skb->ip_summed != CHECKSUM_PARTIAL &&
3340 skb->ip_summed != CHECKSUM_UNNECESSARY;
3342 return skb->ip_summed == CHECKSUM_NONE;
3346 * __skb_gso_segment - Perform segmentation on skb.
3347 * @skb: buffer to segment
3348 * @features: features for the output path (see dev->features)
3349 * @tx_path: whether it is called in TX path
3351 * This function segments the given skb and returns a list of segments.
3353 * It may return NULL if the skb requires no segmentation. This is
3354 * only possible when GSO is used for verifying header integrity.
3356 * Segmentation preserves SKB_GSO_CB_OFFSET bytes of previous skb cb.
3358 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
3359 netdev_features_t features, bool tx_path)
3361 struct sk_buff *segs;
3363 if (unlikely(skb_needs_check(skb, tx_path))) {
3366 /* We're going to init ->check field in TCP or UDP header */
3367 err = skb_cow_head(skb, 0);
3369 return ERR_PTR(err);
3372 /* Only report GSO partial support if it will enable us to
3373 * support segmentation on this frame without needing additional
3376 if (features & NETIF_F_GSO_PARTIAL) {
3377 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
3378 struct net_device *dev = skb->dev;
3380 partial_features |= dev->features & dev->gso_partial_features;
3381 if (!skb_gso_ok(skb, features | partial_features))
3382 features &= ~NETIF_F_GSO_PARTIAL;
3385 BUILD_BUG_ON(SKB_GSO_CB_OFFSET +
3386 sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
3388 SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
3389 SKB_GSO_CB(skb)->encap_level = 0;
3391 skb_reset_mac_header(skb);
3392 skb_reset_mac_len(skb);
3394 segs = skb_mac_gso_segment(skb, features);
3396 if (segs != skb && unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
3397 skb_warn_bad_offload(skb);
3401 EXPORT_SYMBOL(__skb_gso_segment);
3403 /* Take action when hardware reception checksum errors are detected. */
3405 void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb)
3407 if (net_ratelimit()) {
3408 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
3409 skb_dump(KERN_ERR, skb, true);
3413 EXPORT_SYMBOL(netdev_rx_csum_fault);
3416 /* XXX: check that highmem exists at all on the given machine. */
3417 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
3419 #ifdef CONFIG_HIGHMEM
3422 if (!(dev->features & NETIF_F_HIGHDMA)) {
3423 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3424 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3426 if (PageHighMem(skb_frag_page(frag)))
3434 /* If MPLS offload request, verify we are testing hardware MPLS features
3435 * instead of standard features for the netdev.
3437 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
3438 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3439 netdev_features_t features,
3442 if (eth_p_mpls(type))
3443 features &= skb->dev->mpls_features;
3448 static netdev_features_t net_mpls_features(struct sk_buff *skb,
3449 netdev_features_t features,
3456 static netdev_features_t harmonize_features(struct sk_buff *skb,
3457 netdev_features_t features)
3461 type = skb_network_protocol(skb, NULL);
3462 features = net_mpls_features(skb, features, type);
3464 if (skb->ip_summed != CHECKSUM_NONE &&
3465 !can_checksum_protocol(features, type)) {
3466 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3468 if (illegal_highdma(skb->dev, skb))
3469 features &= ~NETIF_F_SG;
3474 netdev_features_t passthru_features_check(struct sk_buff *skb,
3475 struct net_device *dev,
3476 netdev_features_t features)
3480 EXPORT_SYMBOL(passthru_features_check);
3482 static netdev_features_t dflt_features_check(struct sk_buff *skb,
3483 struct net_device *dev,
3484 netdev_features_t features)
3486 return vlan_features_check(skb, features);
3489 static netdev_features_t gso_features_check(const struct sk_buff *skb,
3490 struct net_device *dev,
3491 netdev_features_t features)
3493 u16 gso_segs = skb_shinfo(skb)->gso_segs;
3495 if (gso_segs > dev->gso_max_segs)
3496 return features & ~NETIF_F_GSO_MASK;
3498 /* Support for GSO partial features requires software
3499 * intervention before we can actually process the packets
3500 * so we need to strip support for any partial features now
3501 * and we can pull them back in after we have partially
3502 * segmented the frame.
3504 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
3505 features &= ~dev->gso_partial_features;
3507 /* Make sure to clear the IPv4 ID mangling feature if the
3508 * IPv4 header has the potential to be fragmented.
3510 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
3511 struct iphdr *iph = skb->encapsulation ?
3512 inner_ip_hdr(skb) : ip_hdr(skb);
3514 if (!(iph->frag_off & htons(IP_DF)))
3515 features &= ~NETIF_F_TSO_MANGLEID;
3521 netdev_features_t netif_skb_features(struct sk_buff *skb)
3523 struct net_device *dev = skb->dev;
3524 netdev_features_t features = dev->features;
3526 if (skb_is_gso(skb))
3527 features = gso_features_check(skb, dev, features);
3529 /* If encapsulation offload request, verify we are testing
3530 * hardware encapsulation features instead of standard
3531 * features for the netdev
3533 if (skb->encapsulation)
3534 features &= dev->hw_enc_features;
3536 if (skb_vlan_tagged(skb))
3537 features = netdev_intersect_features(features,
3538 dev->vlan_features |
3539 NETIF_F_HW_VLAN_CTAG_TX |
3540 NETIF_F_HW_VLAN_STAG_TX);
3542 if (dev->netdev_ops->ndo_features_check)
3543 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3546 features &= dflt_features_check(skb, dev, features);
3548 return harmonize_features(skb, features);
3550 EXPORT_SYMBOL(netif_skb_features);
3552 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3553 struct netdev_queue *txq, bool more)
3558 if (dev_nit_active(dev))
3559 dev_queue_xmit_nit(skb, dev);
3562 PRANDOM_ADD_NOISE(skb, dev, txq, len + jiffies);
3563 trace_net_dev_start_xmit(skb, dev);
3564 rc = netdev_start_xmit(skb, dev, txq, more);
3565 trace_net_dev_xmit(skb, rc, dev, len);
3570 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3571 struct netdev_queue *txq, int *ret)
3573 struct sk_buff *skb = first;
3574 int rc = NETDEV_TX_OK;
3577 struct sk_buff *next = skb->next;
3579 skb_mark_not_on_list(skb);
3580 rc = xmit_one(skb, dev, txq, next != NULL);
3581 if (unlikely(!dev_xmit_complete(rc))) {
3587 if (netif_tx_queue_stopped(txq) && skb) {
3588 rc = NETDEV_TX_BUSY;
3598 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3599 netdev_features_t features)
3601 if (skb_vlan_tag_present(skb) &&
3602 !vlan_hw_offload_capable(features, skb->vlan_proto))
3603 skb = __vlan_hwaccel_push_inside(skb);
3607 int skb_csum_hwoffload_help(struct sk_buff *skb,
3608 const netdev_features_t features)
3610 if (unlikely(skb->csum_not_inet))
3611 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3612 skb_crc32c_csum_help(skb);
3614 return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3616 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3618 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3620 netdev_features_t features;
3622 features = netif_skb_features(skb);
3623 skb = validate_xmit_vlan(skb, features);
3627 skb = sk_validate_xmit_skb(skb, dev);
3631 if (netif_needs_gso(skb, features)) {
3632 struct sk_buff *segs;
3634 segs = skb_gso_segment(skb, features);
3642 if (skb_needs_linearize(skb, features) &&
3643 __skb_linearize(skb))
3646 /* If packet is not checksummed and device does not
3647 * support checksumming for this protocol, complete
3648 * checksumming here.
3650 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3651 if (skb->encapsulation)
3652 skb_set_inner_transport_header(skb,
3653 skb_checksum_start_offset(skb));
3655 skb_set_transport_header(skb,
3656 skb_checksum_start_offset(skb));
3657 if (skb_csum_hwoffload_help(skb, features))
3662 skb = validate_xmit_xfrm(skb, features, again);
3669 atomic_long_inc(&dev->tx_dropped);
3673 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3675 struct sk_buff *next, *head = NULL, *tail;
3677 for (; skb != NULL; skb = next) {
3679 skb_mark_not_on_list(skb);
3681 /* in case skb wont be segmented, point to itself */
3684 skb = validate_xmit_skb(skb, dev, again);
3692 /* If skb was segmented, skb->prev points to
3693 * the last segment. If not, it still contains skb.
3699 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3701 static void qdisc_pkt_len_init(struct sk_buff *skb)
3703 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3705 qdisc_skb_cb(skb)->pkt_len = skb->len;
3707 /* To get more precise estimation of bytes sent on wire,
3708 * we add to pkt_len the headers size of all segments
3710 if (shinfo->gso_size && skb_transport_header_was_set(skb)) {
3711 unsigned int hdr_len;
3712 u16 gso_segs = shinfo->gso_segs;
3714 /* mac layer + network layer */
3715 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3717 /* + transport layer */
3718 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3719 const struct tcphdr *th;
3720 struct tcphdr _tcphdr;
3722 th = skb_header_pointer(skb, skb_transport_offset(skb),
3723 sizeof(_tcphdr), &_tcphdr);
3725 hdr_len += __tcp_hdrlen(th);
3727 struct udphdr _udphdr;
3729 if (skb_header_pointer(skb, skb_transport_offset(skb),
3730 sizeof(_udphdr), &_udphdr))
3731 hdr_len += sizeof(struct udphdr);
3734 if (shinfo->gso_type & SKB_GSO_DODGY)
3735 gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3738 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3742 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3743 struct net_device *dev,
3744 struct netdev_queue *txq)
3746 spinlock_t *root_lock = qdisc_lock(q);
3747 struct sk_buff *to_free = NULL;
3751 qdisc_calculate_pkt_len(skb, q);
3753 if (q->flags & TCQ_F_NOLOCK) {
3754 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3757 if (unlikely(to_free))
3758 kfree_skb_list(to_free);
3763 * Heuristic to force contended enqueues to serialize on a
3764 * separate lock before trying to get qdisc main lock.
3765 * This permits qdisc->running owner to get the lock more
3766 * often and dequeue packets faster.
3768 contended = qdisc_is_running(q);
3769 if (unlikely(contended))
3770 spin_lock(&q->busylock);
3772 spin_lock(root_lock);
3773 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3774 __qdisc_drop(skb, &to_free);
3776 } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3777 qdisc_run_begin(q)) {
3779 * This is a work-conserving queue; there are no old skbs
3780 * waiting to be sent out; and the qdisc is not running -
3781 * xmit the skb directly.
3784 qdisc_bstats_update(q, skb);
3786 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3787 if (unlikely(contended)) {
3788 spin_unlock(&q->busylock);
3795 rc = NET_XMIT_SUCCESS;
3797 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3798 if (qdisc_run_begin(q)) {
3799 if (unlikely(contended)) {
3800 spin_unlock(&q->busylock);
3807 spin_unlock(root_lock);
3808 if (unlikely(to_free))
3809 kfree_skb_list(to_free);
3810 if (unlikely(contended))
3811 spin_unlock(&q->busylock);
3815 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3816 static void skb_update_prio(struct sk_buff *skb)
3818 const struct netprio_map *map;
3819 const struct sock *sk;
3820 unsigned int prioidx;
3824 map = rcu_dereference_bh(skb->dev->priomap);
3827 sk = skb_to_full_sk(skb);
3831 prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3833 if (prioidx < map->priomap_len)
3834 skb->priority = map->priomap[prioidx];
3837 #define skb_update_prio(skb)
3841 * dev_loopback_xmit - loop back @skb
3842 * @net: network namespace this loopback is happening in
3843 * @sk: sk needed to be a netfilter okfn
3844 * @skb: buffer to transmit
3846 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3848 skb_reset_mac_header(skb);
3849 __skb_pull(skb, skb_network_offset(skb));
3850 skb->pkt_type = PACKET_LOOPBACK;
3851 skb->ip_summed = CHECKSUM_UNNECESSARY;
3852 WARN_ON(!skb_dst(skb));
3857 EXPORT_SYMBOL(dev_loopback_xmit);
3859 #ifdef CONFIG_NET_EGRESS
3860 static struct sk_buff *
3861 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3863 struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3864 struct tcf_result cl_res;
3869 /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3870 mini_qdisc_bstats_cpu_update(miniq, skb);
3872 switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3874 case TC_ACT_RECLASSIFY:
3875 skb->tc_index = TC_H_MIN(cl_res.classid);
3878 mini_qdisc_qstats_cpu_drop(miniq);
3879 *ret = NET_XMIT_DROP;
3885 *ret = NET_XMIT_SUCCESS;
3888 case TC_ACT_REDIRECT:
3889 /* No need to push/pop skb's mac_header here on egress! */
3890 skb_do_redirect(skb);
3891 *ret = NET_XMIT_SUCCESS;
3899 #endif /* CONFIG_NET_EGRESS */
3902 static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb,
3903 struct xps_dev_maps *dev_maps, unsigned int tci)
3905 struct xps_map *map;
3906 int queue_index = -1;
3910 tci += netdev_get_prio_tc_map(dev, skb->priority);
3913 map = rcu_dereference(dev_maps->attr_map[tci]);
3916 queue_index = map->queues[0];
3918 queue_index = map->queues[reciprocal_scale(
3919 skb_get_hash(skb), map->len)];
3920 if (unlikely(queue_index >= dev->real_num_tx_queues))
3927 static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev,
3928 struct sk_buff *skb)
3931 struct xps_dev_maps *dev_maps;
3932 struct sock *sk = skb->sk;
3933 int queue_index = -1;
3935 if (!static_key_false(&xps_needed))
3939 if (!static_key_false(&xps_rxqs_needed))
3942 dev_maps = rcu_dereference(sb_dev->xps_rxqs_map);
3944 int tci = sk_rx_queue_get(sk);
3946 if (tci >= 0 && tci < dev->num_rx_queues)
3947 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3952 if (queue_index < 0) {
3953 dev_maps = rcu_dereference(sb_dev->xps_cpus_map);
3955 unsigned int tci = skb->sender_cpu - 1;
3957 queue_index = __get_xps_queue_idx(dev, skb, dev_maps,
3969 u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb,
3970 struct net_device *sb_dev)
3974 EXPORT_SYMBOL(dev_pick_tx_zero);
3976 u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb,
3977 struct net_device *sb_dev)
3979 return (u16)raw_smp_processor_id() % dev->real_num_tx_queues;
3981 EXPORT_SYMBOL(dev_pick_tx_cpu_id);
3983 u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb,
3984 struct net_device *sb_dev)
3986 struct sock *sk = skb->sk;
3987 int queue_index = sk_tx_queue_get(sk);
3989 sb_dev = sb_dev ? : dev;
3991 if (queue_index < 0 || skb->ooo_okay ||
3992 queue_index >= dev->real_num_tx_queues) {
3993 int new_index = get_xps_queue(dev, sb_dev, skb);
3996 new_index = skb_tx_hash(dev, sb_dev, skb);
3998 if (queue_index != new_index && sk &&
4000 rcu_access_pointer(sk->sk_dst_cache))
4001 sk_tx_queue_set(sk, new_index);
4003 queue_index = new_index;
4008 EXPORT_SYMBOL(netdev_pick_tx);
4010 struct netdev_queue *netdev_core_pick_tx(struct net_device *dev,
4011 struct sk_buff *skb,
4012 struct net_device *sb_dev)
4014 int queue_index = 0;
4017 u32 sender_cpu = skb->sender_cpu - 1;
4019 if (sender_cpu >= (u32)NR_CPUS)
4020 skb->sender_cpu = raw_smp_processor_id() + 1;
4023 if (dev->real_num_tx_queues != 1) {
4024 const struct net_device_ops *ops = dev->netdev_ops;
4026 if (ops->ndo_select_queue)
4027 queue_index = ops->ndo_select_queue(dev, skb, sb_dev);
4029 queue_index = netdev_pick_tx(dev, skb, sb_dev);
4031 queue_index = netdev_cap_txqueue(dev, queue_index);
4034 skb_set_queue_mapping(skb, queue_index);
4035 return netdev_get_tx_queue(dev, queue_index);
4039 * __dev_queue_xmit - transmit a buffer
4040 * @skb: buffer to transmit
4041 * @sb_dev: suboordinate device used for L2 forwarding offload
4043 * Queue a buffer for transmission to a network device. The caller must
4044 * have set the device and priority and built the buffer before calling
4045 * this function. The function can be called from an interrupt.
4047 * A negative errno code is returned on a failure. A success does not
4048 * guarantee the frame will be transmitted as it may be dropped due
4049 * to congestion or traffic shaping.
4051 * -----------------------------------------------------------------------------------
4052 * I notice this method can also return errors from the queue disciplines,
4053 * including NET_XMIT_DROP, which is a positive value. So, errors can also
4056 * Regardless of the return value, the skb is consumed, so it is currently
4057 * difficult to retry a send to this method. (You can bump the ref count
4058 * before sending to hold a reference for retry if you are careful.)
4060 * When calling this method, interrupts MUST be enabled. This is because
4061 * the BH enable code must have IRQs enabled so that it will not deadlock.
4064 static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev)
4066 struct net_device *dev = skb->dev;
4067 struct netdev_queue *txq;
4072 skb_reset_mac_header(skb);
4074 if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
4075 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
4077 /* Disable soft irqs for various locks below. Also
4078 * stops preemption for RCU.
4082 skb_update_prio(skb);
4084 qdisc_pkt_len_init(skb);
4085 #ifdef CONFIG_NET_CLS_ACT
4086 skb->tc_at_ingress = 0;
4087 # ifdef CONFIG_NET_EGRESS
4088 if (static_branch_unlikely(&egress_needed_key)) {
4089 skb = sch_handle_egress(skb, &rc, dev);
4095 /* If device/qdisc don't need skb->dst, release it right now while
4096 * its hot in this cpu cache.
4098 if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
4103 txq = netdev_core_pick_tx(dev, skb, sb_dev);
4104 q = rcu_dereference_bh(txq->qdisc);
4106 trace_net_dev_queue(skb);
4108 rc = __dev_xmit_skb(skb, q, dev, txq);
4112 /* The device has no queue. Common case for software devices:
4113 * loopback, all the sorts of tunnels...
4115 * Really, it is unlikely that netif_tx_lock protection is necessary
4116 * here. (f.e. loopback and IP tunnels are clean ignoring statistics
4118 * However, it is possible, that they rely on protection
4121 * Check this and shot the lock. It is not prone from deadlocks.
4122 *Either shot noqueue qdisc, it is even simpler 8)
4124 if (dev->flags & IFF_UP) {
4125 int cpu = smp_processor_id(); /* ok because BHs are off */
4127 if (txq->xmit_lock_owner != cpu) {
4128 if (dev_xmit_recursion())
4129 goto recursion_alert;
4131 skb = validate_xmit_skb(skb, dev, &again);
4135 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4136 HARD_TX_LOCK(dev, txq, cpu);
4138 if (!netif_xmit_stopped(txq)) {
4139 dev_xmit_recursion_inc();
4140 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
4141 dev_xmit_recursion_dec();
4142 if (dev_xmit_complete(rc)) {
4143 HARD_TX_UNLOCK(dev, txq);
4147 HARD_TX_UNLOCK(dev, txq);
4148 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
4151 /* Recursion is detected! It is possible,
4155 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
4161 rcu_read_unlock_bh();
4163 atomic_long_inc(&dev->tx_dropped);
4164 kfree_skb_list(skb);
4167 rcu_read_unlock_bh();
4171 int dev_queue_xmit(struct sk_buff *skb)
4173 return __dev_queue_xmit(skb, NULL);
4175 EXPORT_SYMBOL(dev_queue_xmit);
4177 int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev)
4179 return __dev_queue_xmit(skb, sb_dev);
4181 EXPORT_SYMBOL(dev_queue_xmit_accel);
4183 int __dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
4185 struct net_device *dev = skb->dev;
4186 struct sk_buff *orig_skb = skb;
4187 struct netdev_queue *txq;
4188 int ret = NETDEV_TX_BUSY;
4191 if (unlikely(!netif_running(dev) ||
4192 !netif_carrier_ok(dev)))
4195 skb = validate_xmit_skb_list(skb, dev, &again);
4196 if (skb != orig_skb)
4199 skb_set_queue_mapping(skb, queue_id);
4200 txq = skb_get_tx_queue(dev, skb);
4201 PRANDOM_ADD_NOISE(skb, dev, txq, jiffies);
4205 dev_xmit_recursion_inc();
4206 HARD_TX_LOCK(dev, txq, smp_processor_id());
4207 if (!netif_xmit_frozen_or_drv_stopped(txq))
4208 ret = netdev_start_xmit(skb, dev, txq, false);
4209 HARD_TX_UNLOCK(dev, txq);
4210 dev_xmit_recursion_dec();
4215 atomic_long_inc(&dev->tx_dropped);
4216 kfree_skb_list(skb);
4217 return NET_XMIT_DROP;
4219 EXPORT_SYMBOL(__dev_direct_xmit);
4221 /*************************************************************************
4223 *************************************************************************/
4225 int netdev_max_backlog __read_mostly = 1000;
4226 EXPORT_SYMBOL(netdev_max_backlog);
4228 int netdev_tstamp_prequeue __read_mostly = 1;
4229 int netdev_budget __read_mostly = 300;
4230 /* Must be at least 2 jiffes to guarantee 1 jiffy timeout */
4231 unsigned int __read_mostly netdev_budget_usecs = 2 * USEC_PER_SEC / HZ;
4232 int weight_p __read_mostly = 64; /* old backlog weight */
4233 int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */
4234 int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */
4235 int dev_rx_weight __read_mostly = 64;
4236 int dev_tx_weight __read_mostly = 64;
4237 /* Maximum number of GRO_NORMAL skbs to batch up for list-RX */
4238 int gro_normal_batch __read_mostly = 8;
4240 /* Called with irq disabled */
4241 static inline void ____napi_schedule(struct softnet_data *sd,
4242 struct napi_struct *napi)
4244 list_add_tail(&napi->poll_list, &sd->poll_list);
4245 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4250 /* One global table that all flow-based protocols share. */
4251 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
4252 EXPORT_SYMBOL(rps_sock_flow_table);
4253 u32 rps_cpu_mask __read_mostly;
4254 EXPORT_SYMBOL(rps_cpu_mask);
4256 struct static_key_false rps_needed __read_mostly;
4257 EXPORT_SYMBOL(rps_needed);
4258 struct static_key_false rfs_needed __read_mostly;
4259 EXPORT_SYMBOL(rfs_needed);
4261 static struct rps_dev_flow *
4262 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4263 struct rps_dev_flow *rflow, u16 next_cpu)
4265 if (next_cpu < nr_cpu_ids) {
4266 #ifdef CONFIG_RFS_ACCEL
4267 struct netdev_rx_queue *rxqueue;
4268 struct rps_dev_flow_table *flow_table;
4269 struct rps_dev_flow *old_rflow;
4274 /* Should we steer this flow to a different hardware queue? */
4275 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
4276 !(dev->features & NETIF_F_NTUPLE))
4278 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
4279 if (rxq_index == skb_get_rx_queue(skb))
4282 rxqueue = dev->_rx + rxq_index;
4283 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4286 flow_id = skb_get_hash(skb) & flow_table->mask;
4287 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
4288 rxq_index, flow_id);
4292 rflow = &flow_table->flows[flow_id];
4294 if (old_rflow->filter == rflow->filter)
4295 old_rflow->filter = RPS_NO_FILTER;
4299 per_cpu(softnet_data, next_cpu).input_queue_head;
4302 rflow->cpu = next_cpu;
4307 * get_rps_cpu is called from netif_receive_skb and returns the target
4308 * CPU from the RPS map of the receiving queue for a given skb.
4309 * rcu_read_lock must be held on entry.
4311 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
4312 struct rps_dev_flow **rflowp)
4314 const struct rps_sock_flow_table *sock_flow_table;
4315 struct netdev_rx_queue *rxqueue = dev->_rx;
4316 struct rps_dev_flow_table *flow_table;
4317 struct rps_map *map;
4322 if (skb_rx_queue_recorded(skb)) {
4323 u16 index = skb_get_rx_queue(skb);
4325 if (unlikely(index >= dev->real_num_rx_queues)) {
4326 WARN_ONCE(dev->real_num_rx_queues > 1,
4327 "%s received packet on queue %u, but number "
4328 "of RX queues is %u\n",
4329 dev->name, index, dev->real_num_rx_queues);
4335 /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
4337 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4338 map = rcu_dereference(rxqueue->rps_map);
4339 if (!flow_table && !map)
4342 skb_reset_network_header(skb);
4343 hash = skb_get_hash(skb);
4347 sock_flow_table = rcu_dereference(rps_sock_flow_table);
4348 if (flow_table && sock_flow_table) {
4349 struct rps_dev_flow *rflow;
4353 /* First check into global flow table if there is a match */
4354 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
4355 if ((ident ^ hash) & ~rps_cpu_mask)
4358 next_cpu = ident & rps_cpu_mask;
4360 /* OK, now we know there is a match,
4361 * we can look at the local (per receive queue) flow table
4363 rflow = &flow_table->flows[hash & flow_table->mask];
4367 * If the desired CPU (where last recvmsg was done) is
4368 * different from current CPU (one in the rx-queue flow
4369 * table entry), switch if one of the following holds:
4370 * - Current CPU is unset (>= nr_cpu_ids).
4371 * - Current CPU is offline.
4372 * - The current CPU's queue tail has advanced beyond the
4373 * last packet that was enqueued using this table entry.
4374 * This guarantees that all previous packets for the flow
4375 * have been dequeued, thus preserving in order delivery.
4377 if (unlikely(tcpu != next_cpu) &&
4378 (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
4379 ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
4380 rflow->last_qtail)) >= 0)) {
4382 rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
4385 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
4395 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
4396 if (cpu_online(tcpu)) {
4406 #ifdef CONFIG_RFS_ACCEL
4409 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
4410 * @dev: Device on which the filter was set
4411 * @rxq_index: RX queue index
4412 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
4413 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
4415 * Drivers that implement ndo_rx_flow_steer() should periodically call
4416 * this function for each installed filter and remove the filters for
4417 * which it returns %true.
4419 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
4420 u32 flow_id, u16 filter_id)
4422 struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
4423 struct rps_dev_flow_table *flow_table;
4424 struct rps_dev_flow *rflow;
4429 flow_table = rcu_dereference(rxqueue->rps_flow_table);
4430 if (flow_table && flow_id <= flow_table->mask) {
4431 rflow = &flow_table->flows[flow_id];
4432 cpu = READ_ONCE(rflow->cpu);
4433 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
4434 ((int)(per_cpu(softnet_data, cpu).input_queue_head -
4435 rflow->last_qtail) <
4436 (int)(10 * flow_table->mask)))
4442 EXPORT_SYMBOL(rps_may_expire_flow);
4444 #endif /* CONFIG_RFS_ACCEL */
4446 /* Called from hardirq (IPI) context */
4447 static void rps_trigger_softirq(void *data)
4449 struct softnet_data *sd = data;
4451 ____napi_schedule(sd, &sd->backlog);
4455 #endif /* CONFIG_RPS */
4458 * Check if this softnet_data structure is another cpu one
4459 * If yes, queue it to our IPI list and return 1
4462 static int rps_ipi_queued(struct softnet_data *sd)
4465 struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
4468 sd->rps_ipi_next = mysd->rps_ipi_list;
4469 mysd->rps_ipi_list = sd;
4471 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
4474 #endif /* CONFIG_RPS */
4478 #ifdef CONFIG_NET_FLOW_LIMIT
4479 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
4482 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
4484 #ifdef CONFIG_NET_FLOW_LIMIT
4485 struct sd_flow_limit *fl;
4486 struct softnet_data *sd;
4487 unsigned int old_flow, new_flow;
4489 if (qlen < (netdev_max_backlog >> 1))
4492 sd = this_cpu_ptr(&softnet_data);
4495 fl = rcu_dereference(sd->flow_limit);
4497 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
4498 old_flow = fl->history[fl->history_head];
4499 fl->history[fl->history_head] = new_flow;
4502 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
4504 if (likely(fl->buckets[old_flow]))
4505 fl->buckets[old_flow]--;
4507 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
4519 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
4520 * queue (may be a remote CPU queue).
4522 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
4523 unsigned int *qtail)
4525 struct softnet_data *sd;
4526 unsigned long flags;
4529 sd = &per_cpu(softnet_data, cpu);
4531 local_irq_save(flags);
4534 if (!netif_running(skb->dev))
4536 qlen = skb_queue_len(&sd->input_pkt_queue);
4537 if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
4540 __skb_queue_tail(&sd->input_pkt_queue, skb);
4541 input_queue_tail_incr_save(sd, qtail);
4543 local_irq_restore(flags);
4544 return NET_RX_SUCCESS;
4547 /* Schedule NAPI for backlog device
4548 * We can use non atomic operation since we own the queue lock
4550 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
4551 if (!rps_ipi_queued(sd))
4552 ____napi_schedule(sd, &sd->backlog);
4561 local_irq_restore(flags);
4563 atomic_long_inc(&skb->dev->rx_dropped);
4568 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
4570 struct net_device *dev = skb->dev;
4571 struct netdev_rx_queue *rxqueue;
4575 if (skb_rx_queue_recorded(skb)) {
4576 u16 index = skb_get_rx_queue(skb);
4578 if (unlikely(index >= dev->real_num_rx_queues)) {
4579 WARN_ONCE(dev->real_num_rx_queues > 1,
4580 "%s received packet on queue %u, but number "
4581 "of RX queues is %u\n",
4582 dev->name, index, dev->real_num_rx_queues);
4584 return rxqueue; /* Return first rxqueue */
4591 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4592 struct xdp_buff *xdp,
4593 struct bpf_prog *xdp_prog)
4595 struct netdev_rx_queue *rxqueue;
4596 void *orig_data, *orig_data_end;
4597 u32 metalen, act = XDP_DROP;
4598 __be16 orig_eth_type;
4604 /* Reinjected packets coming from act_mirred or similar should
4605 * not get XDP generic processing.
4607 if (skb_is_redirected(skb))
4610 /* XDP packets must be linear and must have sufficient headroom
4611 * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4612 * native XDP provides, thus we need to do it here as well.
4614 if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
4615 skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4616 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4617 int troom = skb->tail + skb->data_len - skb->end;
4619 /* In case we have to go down the path and also linearize,
4620 * then lets do the pskb_expand_head() work just once here.
4622 if (pskb_expand_head(skb,
4623 hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4624 troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4626 if (skb_linearize(skb))
4630 /* The XDP program wants to see the packet starting at the MAC
4633 mac_len = skb->data - skb_mac_header(skb);
4634 hlen = skb_headlen(skb) + mac_len;
4635 xdp->data = skb->data - mac_len;
4636 xdp->data_meta = xdp->data;
4637 xdp->data_end = xdp->data + hlen;
4638 xdp->data_hard_start = skb->data - skb_headroom(skb);
4640 /* SKB "head" area always have tailroom for skb_shared_info */
4641 xdp->frame_sz = (void *)skb_end_pointer(skb) - xdp->data_hard_start;
4642 xdp->frame_sz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
4644 orig_data_end = xdp->data_end;
4645 orig_data = xdp->data;
4646 eth = (struct ethhdr *)xdp->data;
4647 orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest);
4648 orig_eth_type = eth->h_proto;
4650 rxqueue = netif_get_rxqueue(skb);
4651 xdp->rxq = &rxqueue->xdp_rxq;
4653 act = bpf_prog_run_xdp(xdp_prog, xdp);
4655 /* check if bpf_xdp_adjust_head was used */
4656 off = xdp->data - orig_data;
4659 __skb_pull(skb, off);
4661 __skb_push(skb, -off);
4663 skb->mac_header += off;
4664 skb_reset_network_header(skb);
4667 /* check if bpf_xdp_adjust_tail was used */
4668 off = xdp->data_end - orig_data_end;
4670 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4671 skb->len += off; /* positive on grow, negative on shrink */
4674 /* check if XDP changed eth hdr such SKB needs update */
4675 eth = (struct ethhdr *)xdp->data;
4676 if ((orig_eth_type != eth->h_proto) ||
4677 (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) {
4678 __skb_push(skb, ETH_HLEN);
4679 skb->protocol = eth_type_trans(skb, skb->dev);
4685 __skb_push(skb, mac_len);
4688 metalen = xdp->data - xdp->data_meta;
4690 skb_metadata_set(skb, metalen);
4693 bpf_warn_invalid_xdp_action(act);
4696 trace_xdp_exception(skb->dev, xdp_prog, act);
4707 /* When doing generic XDP we have to bypass the qdisc layer and the
4708 * network taps in order to match in-driver-XDP behavior.
4710 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4712 struct net_device *dev = skb->dev;
4713 struct netdev_queue *txq;
4714 bool free_skb = true;
4717 txq = netdev_core_pick_tx(dev, skb, NULL);
4718 cpu = smp_processor_id();
4719 HARD_TX_LOCK(dev, txq, cpu);
4720 if (!netif_xmit_stopped(txq)) {
4721 rc = netdev_start_xmit(skb, dev, txq, 0);
4722 if (dev_xmit_complete(rc))
4725 HARD_TX_UNLOCK(dev, txq);
4727 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4732 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4734 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4737 struct xdp_buff xdp;
4741 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4742 if (act != XDP_PASS) {
4745 err = xdp_do_generic_redirect(skb->dev, skb,
4751 generic_xdp_tx(skb, xdp_prog);
4762 EXPORT_SYMBOL_GPL(do_xdp_generic);
4764 static int netif_rx_internal(struct sk_buff *skb)
4768 net_timestamp_check(netdev_tstamp_prequeue, skb);
4770 trace_netif_rx(skb);
4773 if (static_branch_unlikely(&rps_needed)) {
4774 struct rps_dev_flow voidflow, *rflow = &voidflow;
4780 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4782 cpu = smp_processor_id();
4784 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4793 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4800 * netif_rx - post buffer to the network code
4801 * @skb: buffer to post
4803 * This function receives a packet from a device driver and queues it for
4804 * the upper (protocol) levels to process. It always succeeds. The buffer
4805 * may be dropped during processing for congestion control or by the
4809 * NET_RX_SUCCESS (no congestion)
4810 * NET_RX_DROP (packet was dropped)
4814 int netif_rx(struct sk_buff *skb)
4818 trace_netif_rx_entry(skb);
4820 ret = netif_rx_internal(skb);
4821 trace_netif_rx_exit(ret);
4825 EXPORT_SYMBOL(netif_rx);
4827 int netif_rx_ni(struct sk_buff *skb)
4831 trace_netif_rx_ni_entry(skb);
4834 err = netif_rx_internal(skb);
4835 if (local_softirq_pending())
4838 trace_netif_rx_ni_exit(err);
4842 EXPORT_SYMBOL(netif_rx_ni);
4844 int netif_rx_any_context(struct sk_buff *skb)
4847 * If invoked from contexts which do not invoke bottom half
4848 * processing either at return from interrupt or when softrqs are
4849 * reenabled, use netif_rx_ni() which invokes bottomhalf processing
4853 return netif_rx(skb);
4855 return netif_rx_ni(skb);
4857 EXPORT_SYMBOL(netif_rx_any_context);
4859 static __latent_entropy void net_tx_action(struct softirq_action *h)
4861 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4863 if (sd->completion_queue) {
4864 struct sk_buff *clist;
4866 local_irq_disable();
4867 clist = sd->completion_queue;
4868 sd->completion_queue = NULL;
4872 struct sk_buff *skb = clist;
4874 clist = clist->next;
4876 WARN_ON(refcount_read(&skb->users));
4877 if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4878 trace_consume_skb(skb);
4880 trace_kfree_skb(skb, net_tx_action);
4882 if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4885 __kfree_skb_defer(skb);
4888 __kfree_skb_flush();
4891 if (sd->output_queue) {
4894 local_irq_disable();
4895 head = sd->output_queue;
4896 sd->output_queue = NULL;
4897 sd->output_queue_tailp = &sd->output_queue;
4901 struct Qdisc *q = head;
4902 spinlock_t *root_lock = NULL;
4904 head = head->next_sched;
4906 if (!(q->flags & TCQ_F_NOLOCK)) {
4907 root_lock = qdisc_lock(q);
4908 spin_lock(root_lock);
4910 /* We need to make sure head->next_sched is read
4911 * before clearing __QDISC_STATE_SCHED
4913 smp_mb__before_atomic();
4914 clear_bit(__QDISC_STATE_SCHED, &q->state);
4917 spin_unlock(root_lock);
4921 xfrm_dev_backlog(sd);
4924 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4925 /* This hook is defined here for ATM LANE */
4926 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4927 unsigned char *addr) __read_mostly;
4928 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4931 static inline struct sk_buff *
4932 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4933 struct net_device *orig_dev, bool *another)
4935 #ifdef CONFIG_NET_CLS_ACT
4936 struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4937 struct tcf_result cl_res;
4939 /* If there's at least one ingress present somewhere (so
4940 * we get here via enabled static key), remaining devices
4941 * that are not configured with an ingress qdisc will bail
4948 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4952 qdisc_skb_cb(skb)->pkt_len = skb->len;
4953 skb->tc_at_ingress = 1;
4954 mini_qdisc_bstats_cpu_update(miniq, skb);
4956 switch (tcf_classify_ingress(skb, miniq->block, miniq->filter_list,
4959 case TC_ACT_RECLASSIFY:
4960 skb->tc_index = TC_H_MIN(cl_res.classid);
4963 mini_qdisc_qstats_cpu_drop(miniq);
4971 case TC_ACT_REDIRECT:
4972 /* skb_mac_header check was done by cls/act_bpf, so
4973 * we can safely push the L2 header back before
4974 * redirecting to another netdev
4976 __skb_push(skb, skb->mac_len);
4977 if (skb_do_redirect(skb) == -EAGAIN) {
4978 __skb_pull(skb, skb->mac_len);
4983 case TC_ACT_CONSUMED:
4988 #endif /* CONFIG_NET_CLS_ACT */
4993 * netdev_is_rx_handler_busy - check if receive handler is registered
4994 * @dev: device to check
4996 * Check if a receive handler is already registered for a given device.
4997 * Return true if there one.
4999 * The caller must hold the rtnl_mutex.
5001 bool netdev_is_rx_handler_busy(struct net_device *dev)
5004 return dev && rtnl_dereference(dev->rx_handler);
5006 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
5009 * netdev_rx_handler_register - register receive handler
5010 * @dev: device to register a handler for
5011 * @rx_handler: receive handler to register
5012 * @rx_handler_data: data pointer that is used by rx handler
5014 * Register a receive handler for a device. This handler will then be
5015 * called from __netif_receive_skb. A negative errno code is returned
5018 * The caller must hold the rtnl_mutex.
5020 * For a general description of rx_handler, see enum rx_handler_result.
5022 int netdev_rx_handler_register(struct net_device *dev,
5023 rx_handler_func_t *rx_handler,
5024 void *rx_handler_data)
5026 if (netdev_is_rx_handler_busy(dev))
5029 if (dev->priv_flags & IFF_NO_RX_HANDLER)
5032 /* Note: rx_handler_data must be set before rx_handler */
5033 rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
5034 rcu_assign_pointer(dev->rx_handler, rx_handler);
5038 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
5041 * netdev_rx_handler_unregister - unregister receive handler
5042 * @dev: device to unregister a handler from
5044 * Unregister a receive handler from a device.
5046 * The caller must hold the rtnl_mutex.
5048 void netdev_rx_handler_unregister(struct net_device *dev)
5052 RCU_INIT_POINTER(dev->rx_handler, NULL);
5053 /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
5054 * section has a guarantee to see a non NULL rx_handler_data
5058 RCU_INIT_POINTER(dev->rx_handler_data, NULL);
5060 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
5063 * Limit the use of PFMEMALLOC reserves to those protocols that implement
5064 * the special handling of PFMEMALLOC skbs.
5066 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
5068 switch (skb->protocol) {
5069 case htons(ETH_P_ARP):
5070 case htons(ETH_P_IP):
5071 case htons(ETH_P_IPV6):
5072 case htons(ETH_P_8021Q):
5073 case htons(ETH_P_8021AD):
5080 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
5081 int *ret, struct net_device *orig_dev)
5083 if (nf_hook_ingress_active(skb)) {
5087 *ret = deliver_skb(skb, *pt_prev, orig_dev);
5092 ingress_retval = nf_hook_ingress(skb);
5094 return ingress_retval;
5099 static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
5100 struct packet_type **ppt_prev)
5102 struct packet_type *ptype, *pt_prev;
5103 rx_handler_func_t *rx_handler;
5104 struct sk_buff *skb = *pskb;
5105 struct net_device *orig_dev;
5106 bool deliver_exact = false;
5107 int ret = NET_RX_DROP;
5110 net_timestamp_check(!netdev_tstamp_prequeue, skb);
5112 trace_netif_receive_skb(skb);
5114 orig_dev = skb->dev;
5116 skb_reset_network_header(skb);
5117 if (!skb_transport_header_was_set(skb))
5118 skb_reset_transport_header(skb);
5119 skb_reset_mac_len(skb);
5124 skb->skb_iif = skb->dev->ifindex;
5126 __this_cpu_inc(softnet_data.processed);
5128 if (static_branch_unlikely(&generic_xdp_needed_key)) {
5132 ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
5135 if (ret2 != XDP_PASS) {
5139 skb_reset_mac_len(skb);
5142 if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5143 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5144 skb = skb_vlan_untag(skb);
5149 if (skb_skip_tc_classify(skb))
5155 list_for_each_entry_rcu(ptype, &ptype_all, list) {
5157 ret = deliver_skb(skb, pt_prev, orig_dev);
5161 list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
5163 ret = deliver_skb(skb, pt_prev, orig_dev);
5168 #ifdef CONFIG_NET_INGRESS
5169 if (static_branch_unlikely(&ingress_needed_key)) {
5170 bool another = false;
5172 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev,
5179 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
5183 skb_reset_redirect(skb);
5185 if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
5188 if (skb_vlan_tag_present(skb)) {
5190 ret = deliver_skb(skb, pt_prev, orig_dev);
5193 if (vlan_do_receive(&skb))
5195 else if (unlikely(!skb))
5199 rx_handler = rcu_dereference(skb->dev->rx_handler);
5202 ret = deliver_skb(skb, pt_prev, orig_dev);
5205 switch (rx_handler(&skb)) {
5206 case RX_HANDLER_CONSUMED:
5207 ret = NET_RX_SUCCESS;
5209 case RX_HANDLER_ANOTHER:
5211 case RX_HANDLER_EXACT:
5212 deliver_exact = true;
5213 case RX_HANDLER_PASS:
5220 if (unlikely(skb_vlan_tag_present(skb)) && !netdev_uses_dsa(skb->dev)) {
5222 if (skb_vlan_tag_get_id(skb)) {
5223 /* Vlan id is non 0 and vlan_do_receive() above couldn't
5226 skb->pkt_type = PACKET_OTHERHOST;
5227 } else if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
5228 skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
5229 /* Outer header is 802.1P with vlan 0, inner header is
5230 * 802.1Q or 802.1AD and vlan_do_receive() above could
5231 * not find vlan dev for vlan id 0.
5233 __vlan_hwaccel_clear_tag(skb);
5234 skb = skb_vlan_untag(skb);
5237 if (vlan_do_receive(&skb))
5238 /* After stripping off 802.1P header with vlan 0
5239 * vlan dev is found for inner header.
5242 else if (unlikely(!skb))
5245 /* We have stripped outer 802.1P vlan 0 header.
5246 * But could not find vlan dev.
5247 * check again for vlan id to set OTHERHOST.
5251 /* Note: we might in the future use prio bits
5252 * and set skb->priority like in vlan_do_receive()
5253 * For the time being, just ignore Priority Code Point
5255 __vlan_hwaccel_clear_tag(skb);
5258 type = skb->protocol;
5260 /* deliver only exact match when indicated */
5261 if (likely(!deliver_exact)) {
5262 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5263 &ptype_base[ntohs(type) &
5267 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5268 &orig_dev->ptype_specific);
5270 if (unlikely(skb->dev != orig_dev)) {
5271 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
5272 &skb->dev->ptype_specific);
5276 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
5278 *ppt_prev = pt_prev;
5282 atomic_long_inc(&skb->dev->rx_dropped);
5284 atomic_long_inc(&skb->dev->rx_nohandler);
5286 /* Jamal, now you will not able to escape explaining
5287 * me how you were going to use this. :-)
5293 /* The invariant here is that if *ppt_prev is not NULL
5294 * then skb should also be non-NULL.
5296 * Apparently *ppt_prev assignment above holds this invariant due to
5297 * skb dereferencing near it.
5303 static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
5305 struct net_device *orig_dev = skb->dev;
5306 struct packet_type *pt_prev = NULL;
5309 ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5311 ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
5312 skb->dev, pt_prev, orig_dev);
5317 * netif_receive_skb_core - special purpose version of netif_receive_skb
5318 * @skb: buffer to process
5320 * More direct receive version of netif_receive_skb(). It should
5321 * only be used by callers that have a need to skip RPS and Generic XDP.
5322 * Caller must also take care of handling if ``(page_is_)pfmemalloc``.
5324 * This function may only be called from softirq context and interrupts
5325 * should be enabled.
5327 * Return values (usually ignored):
5328 * NET_RX_SUCCESS: no congestion
5329 * NET_RX_DROP: packet was dropped
5331 int netif_receive_skb_core(struct sk_buff *skb)
5336 ret = __netif_receive_skb_one_core(skb, false);
5341 EXPORT_SYMBOL(netif_receive_skb_core);
5343 static inline void __netif_receive_skb_list_ptype(struct list_head *head,
5344 struct packet_type *pt_prev,
5345 struct net_device *orig_dev)
5347 struct sk_buff *skb, *next;
5351 if (list_empty(head))
5353 if (pt_prev->list_func != NULL)
5354 INDIRECT_CALL_INET(pt_prev->list_func, ipv6_list_rcv,
5355 ip_list_rcv, head, pt_prev, orig_dev);
5357 list_for_each_entry_safe(skb, next, head, list) {
5358 skb_list_del_init(skb);
5359 pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
5363 static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc)
5365 /* Fast-path assumptions:
5366 * - There is no RX handler.
5367 * - Only one packet_type matches.
5368 * If either of these fails, we will end up doing some per-packet
5369 * processing in-line, then handling the 'last ptype' for the whole
5370 * sublist. This can't cause out-of-order delivery to any single ptype,
5371 * because the 'last ptype' must be constant across the sublist, and all
5372 * other ptypes are handled per-packet.
5374 /* Current (common) ptype of sublist */
5375 struct packet_type *pt_curr = NULL;
5376 /* Current (common) orig_dev of sublist */
5377 struct net_device *od_curr = NULL;
5378 struct list_head sublist;
5379 struct sk_buff *skb, *next;
5381 INIT_LIST_HEAD(&sublist);
5382 list_for_each_entry_safe(skb, next, head, list) {
5383 struct net_device *orig_dev = skb->dev;
5384 struct packet_type *pt_prev = NULL;
5386 skb_list_del_init(skb);
5387 __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
5390 if (pt_curr != pt_prev || od_curr != orig_dev) {
5391 /* dispatch old sublist */
5392 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5393 /* start new sublist */
5394 INIT_LIST_HEAD(&sublist);
5398 list_add_tail(&skb->list, &sublist);
5401 /* dispatch final sublist */
5402 __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr);
5405 static int __netif_receive_skb(struct sk_buff *skb)
5409 if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
5410 unsigned int noreclaim_flag;
5413 * PFMEMALLOC skbs are special, they should
5414 * - be delivered to SOCK_MEMALLOC sockets only
5415 * - stay away from userspace
5416 * - have bounded memory usage
5418 * Use PF_MEMALLOC as this saves us from propagating the allocation
5419 * context down to all allocation sites.
5421 noreclaim_flag = memalloc_noreclaim_save();
5422 ret = __netif_receive_skb_one_core(skb, true);
5423 memalloc_noreclaim_restore(noreclaim_flag);
5425 ret = __netif_receive_skb_one_core(skb, false);
5430 static void __netif_receive_skb_list(struct list_head *head)
5432 unsigned long noreclaim_flag = 0;
5433 struct sk_buff *skb, *next;
5434 bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */
5436 list_for_each_entry_safe(skb, next, head, list) {
5437 if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) {
5438 struct list_head sublist;
5440 /* Handle the previous sublist */
5441 list_cut_before(&sublist, head, &skb->list);
5442 if (!list_empty(&sublist))
5443 __netif_receive_skb_list_core(&sublist, pfmemalloc);
5444 pfmemalloc = !pfmemalloc;
5445 /* See comments in __netif_receive_skb */
5447 noreclaim_flag = memalloc_noreclaim_save();
5449 memalloc_noreclaim_restore(noreclaim_flag);
5452 /* Handle the remaining sublist */
5453 if (!list_empty(head))
5454 __netif_receive_skb_list_core(head, pfmemalloc);
5455 /* Restore pflags */
5457 memalloc_noreclaim_restore(noreclaim_flag);
5460 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
5462 struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
5463 struct bpf_prog *new = xdp->prog;
5469 mutex_lock(&new->aux->used_maps_mutex);
5471 /* generic XDP does not work with DEVMAPs that can
5472 * have a bpf_prog installed on an entry
5474 for (i = 0; i < new->aux->used_map_cnt; i++) {
5475 if (dev_map_can_have_prog(new->aux->used_maps[i]) ||
5476 cpu_map_prog_allowed(new->aux->used_maps[i])) {
5477 mutex_unlock(&new->aux->used_maps_mutex);
5482 mutex_unlock(&new->aux->used_maps_mutex);
5485 switch (xdp->command) {
5486 case XDP_SETUP_PROG:
5487 rcu_assign_pointer(dev->xdp_prog, new);
5492 static_branch_dec(&generic_xdp_needed_key);
5493 } else if (new && !old) {
5494 static_branch_inc(&generic_xdp_needed_key);
5495 dev_disable_lro(dev);
5496 dev_disable_gro_hw(dev);
5508 static int netif_receive_skb_internal(struct sk_buff *skb)
5512 net_timestamp_check(netdev_tstamp_prequeue, skb);
5514 if (skb_defer_rx_timestamp(skb))
5515 return NET_RX_SUCCESS;
5519 if (static_branch_unlikely(&rps_needed)) {
5520 struct rps_dev_flow voidflow, *rflow = &voidflow;
5521 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5524 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5530 ret = __netif_receive_skb(skb);
5535 static void netif_receive_skb_list_internal(struct list_head *head)
5537 struct sk_buff *skb, *next;
5538 struct list_head sublist;
5540 INIT_LIST_HEAD(&sublist);
5541 list_for_each_entry_safe(skb, next, head, list) {
5542 net_timestamp_check(netdev_tstamp_prequeue, skb);
5543 skb_list_del_init(skb);
5544 if (!skb_defer_rx_timestamp(skb))
5545 list_add_tail(&skb->list, &sublist);
5547 list_splice_init(&sublist, head);
5551 if (static_branch_unlikely(&rps_needed)) {
5552 list_for_each_entry_safe(skb, next, head, list) {
5553 struct rps_dev_flow voidflow, *rflow = &voidflow;
5554 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
5557 /* Will be handled, remove from list */
5558 skb_list_del_init(skb);
5559 enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
5564 __netif_receive_skb_list(head);
5569 * netif_receive_skb - process receive buffer from network
5570 * @skb: buffer to process
5572 * netif_receive_skb() is the main receive data processing function.
5573 * It always succeeds. The buffer may be dropped during processing
5574 * for congestion control or by the protocol layers.
5576 * This function may only be called from softirq context and interrupts
5577 * should be enabled.
5579 * Return values (usually ignored):
5580 * NET_RX_SUCCESS: no congestion
5581 * NET_RX_DROP: packet was dropped
5583 int netif_receive_skb(struct sk_buff *skb)
5587 trace_netif_receive_skb_entry(skb);
5589 ret = netif_receive_skb_internal(skb);
5590 trace_netif_receive_skb_exit(ret);
5594 EXPORT_SYMBOL(netif_receive_skb);
5597 * netif_receive_skb_list - process many receive buffers from network
5598 * @head: list of skbs to process.
5600 * Since return value of netif_receive_skb() is normally ignored, and
5601 * wouldn't be meaningful for a list, this function returns void.
5603 * This function may only be called from softirq context and interrupts
5604 * should be enabled.
5606 void netif_receive_skb_list(struct list_head *head)
5608 struct sk_buff *skb;
5610 if (list_empty(head))
5612 if (trace_netif_receive_skb_list_entry_enabled()) {
5613 list_for_each_entry(skb, head, list)
5614 trace_netif_receive_skb_list_entry(skb);
5616 netif_receive_skb_list_internal(head);
5617 trace_netif_receive_skb_list_exit(0);
5619 EXPORT_SYMBOL(netif_receive_skb_list);
5621 static DEFINE_PER_CPU(struct work_struct, flush_works);
5623 /* Network device is going away, flush any packets still pending */
5624 static void flush_backlog(struct work_struct *work)
5626 struct sk_buff *skb, *tmp;
5627 struct softnet_data *sd;
5630 sd = this_cpu_ptr(&softnet_data);
5632 local_irq_disable();
5634 skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
5635 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5636 __skb_unlink(skb, &sd->input_pkt_queue);
5637 dev_kfree_skb_irq(skb);
5638 input_queue_head_incr(sd);
5644 skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
5645 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
5646 __skb_unlink(skb, &sd->process_queue);
5648 input_queue_head_incr(sd);
5654 static bool flush_required(int cpu)
5656 #if IS_ENABLED(CONFIG_RPS)
5657 struct softnet_data *sd = &per_cpu(softnet_data, cpu);
5660 local_irq_disable();
5663 /* as insertion into process_queue happens with the rps lock held,
5664 * process_queue access may race only with dequeue
5666 do_flush = !skb_queue_empty(&sd->input_pkt_queue) ||
5667 !skb_queue_empty_lockless(&sd->process_queue);
5673 /* without RPS we can't safely check input_pkt_queue: during a
5674 * concurrent remote skb_queue_splice() we can detect as empty both
5675 * input_pkt_queue and process_queue even if the latter could end-up
5676 * containing a lot of packets.
5681 static void flush_all_backlogs(void)
5683 static cpumask_t flush_cpus;
5686 /* since we are under rtnl lock protection we can use static data
5687 * for the cpumask and avoid allocating on stack the possibly
5694 cpumask_clear(&flush_cpus);
5695 for_each_online_cpu(cpu) {
5696 if (flush_required(cpu)) {
5697 queue_work_on(cpu, system_highpri_wq,
5698 per_cpu_ptr(&flush_works, cpu));
5699 cpumask_set_cpu(cpu, &flush_cpus);
5703 /* we can have in flight packet[s] on the cpus we are not flushing,
5704 * synchronize_net() in rollback_registered_many() will take care of
5707 for_each_cpu(cpu, &flush_cpus)
5708 flush_work(per_cpu_ptr(&flush_works, cpu));
5713 /* Pass the currently batched GRO_NORMAL SKBs up to the stack. */
5714 static void gro_normal_list(struct napi_struct *napi)
5716 if (!napi->rx_count)
5718 netif_receive_skb_list_internal(&napi->rx_list);
5719 INIT_LIST_HEAD(&napi->rx_list);
5723 /* Queue one GRO_NORMAL SKB up for list processing. If batch size exceeded,
5724 * pass the whole batch up to the stack.
5726 static void gro_normal_one(struct napi_struct *napi, struct sk_buff *skb)
5728 list_add_tail(&skb->list, &napi->rx_list);
5729 if (++napi->rx_count >= gro_normal_batch)
5730 gro_normal_list(napi);
5733 INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int));
5734 INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int));
5735 static int napi_gro_complete(struct napi_struct *napi, struct sk_buff *skb)
5737 struct packet_offload *ptype;
5738 __be16 type = skb->protocol;
5739 struct list_head *head = &offload_base;
5742 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
5744 if (NAPI_GRO_CB(skb)->count == 1) {
5745 skb_shinfo(skb)->gso_size = 0;
5750 list_for_each_entry_rcu(ptype, head, list) {
5751 if (ptype->type != type || !ptype->callbacks.gro_complete)
5754 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
5755 ipv6_gro_complete, inet_gro_complete,
5762 WARN_ON(&ptype->list == head);
5764 return NET_RX_SUCCESS;
5768 gro_normal_one(napi, skb);
5769 return NET_RX_SUCCESS;
5772 static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index,
5775 struct list_head *head = &napi->gro_hash[index].list;
5776 struct sk_buff *skb, *p;
5778 list_for_each_entry_safe_reverse(skb, p, head, list) {
5779 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
5781 skb_list_del_init(skb);
5782 napi_gro_complete(napi, skb);
5783 napi->gro_hash[index].count--;
5786 if (!napi->gro_hash[index].count)
5787 __clear_bit(index, &napi->gro_bitmask);
5790 /* napi->gro_hash[].list contains packets ordered by age.
5791 * youngest packets at the head of it.
5792 * Complete skbs in reverse order to reduce latencies.
5794 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
5796 unsigned long bitmask = napi->gro_bitmask;
5797 unsigned int i, base = ~0U;
5799 while ((i = ffs(bitmask)) != 0) {
5802 __napi_gro_flush_chain(napi, base, flush_old);
5805 EXPORT_SYMBOL(napi_gro_flush);
5807 static struct list_head *gro_list_prepare(struct napi_struct *napi,
5808 struct sk_buff *skb)
5810 unsigned int maclen = skb->dev->hard_header_len;
5811 u32 hash = skb_get_hash_raw(skb);
5812 struct list_head *head;
5815 head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list;
5816 list_for_each_entry(p, head, list) {
5817 unsigned long diffs;
5819 NAPI_GRO_CB(p)->flush = 0;
5821 if (hash != skb_get_hash_raw(p)) {
5822 NAPI_GRO_CB(p)->same_flow = 0;
5826 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
5827 diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb);
5828 if (skb_vlan_tag_present(p))
5829 diffs |= skb_vlan_tag_get(p) ^ skb_vlan_tag_get(skb);
5830 diffs |= skb_metadata_dst_cmp(p, skb);
5831 diffs |= skb_metadata_differs(p, skb);
5832 if (maclen == ETH_HLEN)
5833 diffs |= compare_ether_header(skb_mac_header(p),
5834 skb_mac_header(skb));
5836 diffs = memcmp(skb_mac_header(p),
5837 skb_mac_header(skb),
5839 NAPI_GRO_CB(p)->same_flow = !diffs;
5845 static void skb_gro_reset_offset(struct sk_buff *skb)
5847 const struct skb_shared_info *pinfo = skb_shinfo(skb);
5848 const skb_frag_t *frag0 = &pinfo->frags[0];
5850 NAPI_GRO_CB(skb)->data_offset = 0;
5851 NAPI_GRO_CB(skb)->frag0 = NULL;
5852 NAPI_GRO_CB(skb)->frag0_len = 0;
5854 if (!skb_headlen(skb) && pinfo->nr_frags &&
5855 !PageHighMem(skb_frag_page(frag0))) {
5856 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
5857 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
5858 skb_frag_size(frag0),
5859 skb->end - skb->tail);
5863 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
5865 struct skb_shared_info *pinfo = skb_shinfo(skb);
5867 BUG_ON(skb->end - skb->tail < grow);
5869 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
5871 skb->data_len -= grow;
5874 skb_frag_off_add(&pinfo->frags[0], grow);
5875 skb_frag_size_sub(&pinfo->frags[0], grow);
5877 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
5878 skb_frag_unref(skb, 0);
5879 memmove(pinfo->frags, pinfo->frags + 1,
5880 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
5884 static void gro_flush_oldest(struct napi_struct *napi, struct list_head *head)
5886 struct sk_buff *oldest;
5888 oldest = list_last_entry(head, struct sk_buff, list);
5890 /* We are called with head length >= MAX_GRO_SKBS, so this is
5893 if (WARN_ON_ONCE(!oldest))
5896 /* Do not adjust napi->gro_hash[].count, caller is adding a new
5899 skb_list_del_init(oldest);
5900 napi_gro_complete(napi, oldest);
5903 INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *,
5905 INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *,
5907 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5909 u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
5910 struct list_head *head = &offload_base;
5911 struct packet_offload *ptype;
5912 __be16 type = skb->protocol;
5913 struct list_head *gro_head;
5914 struct sk_buff *pp = NULL;
5915 enum gro_result ret;
5919 if (netif_elide_gro(skb->dev))
5922 gro_head = gro_list_prepare(napi, skb);
5925 list_for_each_entry_rcu(ptype, head, list) {
5926 if (ptype->type != type || !ptype->callbacks.gro_receive)
5929 skb_set_network_header(skb, skb_gro_offset(skb));
5930 skb_reset_mac_len(skb);
5931 NAPI_GRO_CB(skb)->same_flow = 0;
5932 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5933 NAPI_GRO_CB(skb)->free = 0;
5934 NAPI_GRO_CB(skb)->encap_mark = 0;
5935 NAPI_GRO_CB(skb)->recursion_counter = 0;
5936 NAPI_GRO_CB(skb)->is_fou = 0;
5937 NAPI_GRO_CB(skb)->is_atomic = 1;
5938 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5940 /* Setup for GRO checksum validation */
5941 switch (skb->ip_summed) {
5942 case CHECKSUM_COMPLETE:
5943 NAPI_GRO_CB(skb)->csum = skb->csum;
5944 NAPI_GRO_CB(skb)->csum_valid = 1;
5945 NAPI_GRO_CB(skb)->csum_cnt = 0;
5947 case CHECKSUM_UNNECESSARY:
5948 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5949 NAPI_GRO_CB(skb)->csum_valid = 0;
5952 NAPI_GRO_CB(skb)->csum_cnt = 0;
5953 NAPI_GRO_CB(skb)->csum_valid = 0;
5956 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
5957 ipv6_gro_receive, inet_gro_receive,
5963 if (&ptype->list == head)
5966 if (PTR_ERR(pp) == -EINPROGRESS) {
5971 same_flow = NAPI_GRO_CB(skb)->same_flow;
5972 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5975 skb_list_del_init(pp);
5976 napi_gro_complete(napi, pp);
5977 napi->gro_hash[hash].count--;
5983 if (NAPI_GRO_CB(skb)->flush)
5986 if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) {
5987 gro_flush_oldest(napi, gro_head);
5989 napi->gro_hash[hash].count++;
5991 NAPI_GRO_CB(skb)->count = 1;
5992 NAPI_GRO_CB(skb)->age = jiffies;
5993 NAPI_GRO_CB(skb)->last = skb;
5994 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5995 list_add(&skb->list, gro_head);
5999 grow = skb_gro_offset(skb) - skb_headlen(skb);
6001 gro_pull_from_frag0(skb, grow);
6003 if (napi->gro_hash[hash].count) {
6004 if (!test_bit(hash, &napi->gro_bitmask))
6005 __set_bit(hash, &napi->gro_bitmask);
6006 } else if (test_bit(hash, &napi->gro_bitmask)) {
6007 __clear_bit(hash, &napi->gro_bitmask);
6017 struct packet_offload *gro_find_receive_by_type(__be16 type)
6019 struct list_head *offload_head = &offload_base;
6020 struct packet_offload *ptype;
6022 list_for_each_entry_rcu(ptype, offload_head, list) {
6023 if (ptype->type != type || !ptype->callbacks.gro_receive)
6029 EXPORT_SYMBOL(gro_find_receive_by_type);
6031 struct packet_offload *gro_find_complete_by_type(__be16 type)
6033 struct list_head *offload_head = &offload_base;
6034 struct packet_offload *ptype;
6036 list_for_each_entry_rcu(ptype, offload_head, list) {
6037 if (ptype->type != type || !ptype->callbacks.gro_complete)
6043 EXPORT_SYMBOL(gro_find_complete_by_type);
6045 static void napi_skb_free_stolen_head(struct sk_buff *skb)
6049 kmem_cache_free(skbuff_head_cache, skb);
6052 static gro_result_t napi_skb_finish(struct napi_struct *napi,
6053 struct sk_buff *skb,
6058 gro_normal_one(napi, skb);
6065 case GRO_MERGED_FREE:
6066 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6067 napi_skb_free_stolen_head(skb);
6081 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
6085 skb_mark_napi_id(skb, napi);
6086 trace_napi_gro_receive_entry(skb);
6088 skb_gro_reset_offset(skb);
6090 ret = napi_skb_finish(napi, skb, dev_gro_receive(napi, skb));
6091 trace_napi_gro_receive_exit(ret);
6095 EXPORT_SYMBOL(napi_gro_receive);
6097 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
6099 if (unlikely(skb->pfmemalloc)) {
6103 __skb_pull(skb, skb_headlen(skb));
6104 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
6105 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
6106 __vlan_hwaccel_clear_tag(skb);
6107 skb->dev = napi->dev;
6110 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
6111 skb->pkt_type = PACKET_HOST;
6113 skb->encapsulation = 0;
6114 skb_shinfo(skb)->gso_type = 0;
6115 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6121 struct sk_buff *napi_get_frags(struct napi_struct *napi)
6123 struct sk_buff *skb = napi->skb;
6126 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
6129 skb_mark_napi_id(skb, napi);
6134 EXPORT_SYMBOL(napi_get_frags);
6136 static gro_result_t napi_frags_finish(struct napi_struct *napi,
6137 struct sk_buff *skb,
6143 __skb_push(skb, ETH_HLEN);
6144 skb->protocol = eth_type_trans(skb, skb->dev);
6145 if (ret == GRO_NORMAL)
6146 gro_normal_one(napi, skb);
6150 napi_reuse_skb(napi, skb);
6153 case GRO_MERGED_FREE:
6154 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
6155 napi_skb_free_stolen_head(skb);
6157 napi_reuse_skb(napi, skb);
6168 /* Upper GRO stack assumes network header starts at gro_offset=0
6169 * Drivers could call both napi_gro_frags() and napi_gro_receive()
6170 * We copy ethernet header into skb->data to have a common layout.
6172 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
6174 struct sk_buff *skb = napi->skb;
6175 const struct ethhdr *eth;
6176 unsigned int hlen = sizeof(*eth);
6180 skb_reset_mac_header(skb);
6181 skb_gro_reset_offset(skb);
6183 if (unlikely(skb_gro_header_hard(skb, hlen))) {
6184 eth = skb_gro_header_slow(skb, hlen, 0);
6185 if (unlikely(!eth)) {
6186 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
6187 __func__, napi->dev->name);
6188 napi_reuse_skb(napi, skb);
6192 eth = (const struct ethhdr *)skb->data;
6193 gro_pull_from_frag0(skb, hlen);
6194 NAPI_GRO_CB(skb)->frag0 += hlen;
6195 NAPI_GRO_CB(skb)->frag0_len -= hlen;
6197 __skb_pull(skb, hlen);
6200 * This works because the only protocols we care about don't require
6202 * We'll fix it up properly in napi_frags_finish()
6204 skb->protocol = eth->h_proto;
6209 gro_result_t napi_gro_frags(struct napi_struct *napi)
6212 struct sk_buff *skb = napi_frags_skb(napi);
6217 trace_napi_gro_frags_entry(skb);
6219 ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
6220 trace_napi_gro_frags_exit(ret);
6224 EXPORT_SYMBOL(napi_gro_frags);
6226 /* Compute the checksum from gro_offset and return the folded value
6227 * after adding in any pseudo checksum.
6229 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
6234 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
6236 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
6237 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
6238 /* See comments in __skb_checksum_complete(). */
6240 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
6241 !skb->csum_complete_sw)
6242 netdev_rx_csum_fault(skb->dev, skb);
6245 NAPI_GRO_CB(skb)->csum = wsum;
6246 NAPI_GRO_CB(skb)->csum_valid = 1;
6250 EXPORT_SYMBOL(__skb_gro_checksum_complete);
6252 static void net_rps_send_ipi(struct softnet_data *remsd)
6256 struct softnet_data *next = remsd->rps_ipi_next;
6258 if (cpu_online(remsd->cpu))
6259 smp_call_function_single_async(remsd->cpu, &remsd->csd);
6266 * net_rps_action_and_irq_enable sends any pending IPI's for rps.
6267 * Note: called with local irq disabled, but exits with local irq enabled.
6269 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
6272 struct softnet_data *remsd = sd->rps_ipi_list;
6275 sd->rps_ipi_list = NULL;
6279 /* Send pending IPI's to kick RPS processing on remote cpus. */
6280 net_rps_send_ipi(remsd);
6286 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
6289 return sd->rps_ipi_list != NULL;
6295 static int process_backlog(struct napi_struct *napi, int quota)
6297 struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
6301 /* Check if we have pending ipi, its better to send them now,
6302 * not waiting net_rx_action() end.
6304 if (sd_has_rps_ipi_waiting(sd)) {
6305 local_irq_disable();
6306 net_rps_action_and_irq_enable(sd);
6309 napi->weight = dev_rx_weight;
6311 struct sk_buff *skb;
6313 while ((skb = __skb_dequeue(&sd->process_queue))) {
6315 __netif_receive_skb(skb);
6317 input_queue_head_incr(sd);
6318 if (++work >= quota)
6323 local_irq_disable();
6325 if (skb_queue_empty(&sd->input_pkt_queue)) {
6327 * Inline a custom version of __napi_complete().
6328 * only current cpu owns and manipulates this napi,
6329 * and NAPI_STATE_SCHED is the only possible flag set
6331 * We can use a plain write instead of clear_bit(),
6332 * and we dont need an smp_mb() memory barrier.
6337 skb_queue_splice_tail_init(&sd->input_pkt_queue,
6338 &sd->process_queue);
6348 * __napi_schedule - schedule for receive
6349 * @n: entry to schedule
6351 * The entry's receive function will be scheduled to run.
6352 * Consider using __napi_schedule_irqoff() if hard irqs are masked.
6354 void __napi_schedule(struct napi_struct *n)
6356 unsigned long flags;
6358 local_irq_save(flags);
6359 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6360 local_irq_restore(flags);
6362 EXPORT_SYMBOL(__napi_schedule);
6365 * napi_schedule_prep - check if napi can be scheduled
6368 * Test if NAPI routine is already running, and if not mark
6369 * it as running. This is used as a condition variable to
6370 * insure only one NAPI poll instance runs. We also make
6371 * sure there is no pending NAPI disable.
6373 bool napi_schedule_prep(struct napi_struct *n)
6375 unsigned long val, new;
6378 val = READ_ONCE(n->state);
6379 if (unlikely(val & NAPIF_STATE_DISABLE))
6381 new = val | NAPIF_STATE_SCHED;
6383 /* Sets STATE_MISSED bit if STATE_SCHED was already set
6384 * This was suggested by Alexander Duyck, as compiler
6385 * emits better code than :
6386 * if (val & NAPIF_STATE_SCHED)
6387 * new |= NAPIF_STATE_MISSED;
6389 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
6391 } while (cmpxchg(&n->state, val, new) != val);
6393 return !(val & NAPIF_STATE_SCHED);
6395 EXPORT_SYMBOL(napi_schedule_prep);
6398 * __napi_schedule_irqoff - schedule for receive
6399 * @n: entry to schedule
6401 * Variant of __napi_schedule() assuming hard irqs are masked
6403 void __napi_schedule_irqoff(struct napi_struct *n)
6405 ____napi_schedule(this_cpu_ptr(&softnet_data), n);
6407 EXPORT_SYMBOL(__napi_schedule_irqoff);
6409 bool napi_complete_done(struct napi_struct *n, int work_done)
6411 unsigned long flags, val, new, timeout = 0;
6415 * 1) Don't let napi dequeue from the cpu poll list
6416 * just in case its running on a different cpu.
6417 * 2) If we are busy polling, do nothing here, we have
6418 * the guarantee we will be called later.
6420 if (unlikely(n->state & (NAPIF_STATE_NPSVC |
6421 NAPIF_STATE_IN_BUSY_POLL)))
6426 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6427 n->defer_hard_irqs_count = READ_ONCE(n->dev->napi_defer_hard_irqs);
6429 if (n->defer_hard_irqs_count > 0) {
6430 n->defer_hard_irqs_count--;
6431 timeout = READ_ONCE(n->dev->gro_flush_timeout);
6435 if (n->gro_bitmask) {
6436 /* When the NAPI instance uses a timeout and keeps postponing
6437 * it, we need to bound somehow the time packets are kept in
6440 napi_gro_flush(n, !!timeout);
6445 if (unlikely(!list_empty(&n->poll_list))) {
6446 /* If n->poll_list is not empty, we need to mask irqs */
6447 local_irq_save(flags);
6448 list_del_init(&n->poll_list);
6449 local_irq_restore(flags);
6453 val = READ_ONCE(n->state);
6455 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
6457 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
6459 /* If STATE_MISSED was set, leave STATE_SCHED set,
6460 * because we will call napi->poll() one more time.
6461 * This C code was suggested by Alexander Duyck to help gcc.
6463 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
6465 } while (cmpxchg(&n->state, val, new) != val);
6467 if (unlikely(val & NAPIF_STATE_MISSED)) {
6473 hrtimer_start(&n->timer, ns_to_ktime(timeout),
6474 HRTIMER_MODE_REL_PINNED);
6477 EXPORT_SYMBOL(napi_complete_done);
6479 /* must be called under rcu_read_lock(), as we dont take a reference */
6480 static struct napi_struct *napi_by_id(unsigned int napi_id)
6482 unsigned int hash = napi_id % HASH_SIZE(napi_hash);
6483 struct napi_struct *napi;
6485 hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
6486 if (napi->napi_id == napi_id)
6492 #if defined(CONFIG_NET_RX_BUSY_POLL)
6494 #define BUSY_POLL_BUDGET 8
6496 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
6500 /* Busy polling means there is a high chance device driver hard irq
6501 * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
6502 * set in napi_schedule_prep().
6503 * Since we are about to call napi->poll() once more, we can safely
6504 * clear NAPI_STATE_MISSED.
6506 * Note: x86 could use a single "lock and ..." instruction
6507 * to perform these two clear_bit()
6509 clear_bit(NAPI_STATE_MISSED, &napi->state);
6510 clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
6514 /* All we really want here is to re-enable device interrupts.
6515 * Ideally, a new ndo_busy_poll_stop() could avoid another round.
6517 rc = napi->poll(napi, BUSY_POLL_BUDGET);
6518 /* We can't gro_normal_list() here, because napi->poll() might have
6519 * rearmed the napi (napi_complete_done()) in which case it could
6520 * already be running on another CPU.
6522 trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
6523 netpoll_poll_unlock(have_poll_lock);
6524 if (rc == BUSY_POLL_BUDGET) {
6525 /* As the whole budget was spent, we still own the napi so can
6526 * safely handle the rx_list.
6528 gro_normal_list(napi);
6529 __napi_schedule(napi);
6534 void napi_busy_loop(unsigned int napi_id,
6535 bool (*loop_end)(void *, unsigned long),
6538 unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
6539 int (*napi_poll)(struct napi_struct *napi, int budget);
6540 void *have_poll_lock = NULL;
6541 struct napi_struct *napi;
6548 napi = napi_by_id(napi_id);
6558 unsigned long val = READ_ONCE(napi->state);
6560 /* If multiple threads are competing for this napi,
6561 * we avoid dirtying napi->state as much as we can.
6563 if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
6564 NAPIF_STATE_IN_BUSY_POLL))
6566 if (cmpxchg(&napi->state, val,
6567 val | NAPIF_STATE_IN_BUSY_POLL |
6568 NAPIF_STATE_SCHED) != val)
6570 have_poll_lock = netpoll_poll_lock(napi);
6571 napi_poll = napi->poll;
6573 work = napi_poll(napi, BUSY_POLL_BUDGET);
6574 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
6575 gro_normal_list(napi);
6578 __NET_ADD_STATS(dev_net(napi->dev),
6579 LINUX_MIB_BUSYPOLLRXPACKETS, work);
6582 if (!loop_end || loop_end(loop_end_arg, start_time))
6585 if (unlikely(need_resched())) {
6587 busy_poll_stop(napi, have_poll_lock);
6591 if (loop_end(loop_end_arg, start_time))
6598 busy_poll_stop(napi, have_poll_lock);
6603 EXPORT_SYMBOL(napi_busy_loop);
6605 #endif /* CONFIG_NET_RX_BUSY_POLL */
6607 static void napi_hash_add(struct napi_struct *napi)
6609 if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state))
6612 spin_lock(&napi_hash_lock);
6614 /* 0..NR_CPUS range is reserved for sender_cpu use */
6616 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
6617 napi_gen_id = MIN_NAPI_ID;
6618 } while (napi_by_id(napi_gen_id));
6619 napi->napi_id = napi_gen_id;
6621 hlist_add_head_rcu(&napi->napi_hash_node,
6622 &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
6624 spin_unlock(&napi_hash_lock);
6627 /* Warning : caller is responsible to make sure rcu grace period
6628 * is respected before freeing memory containing @napi
6630 static void napi_hash_del(struct napi_struct *napi)
6632 spin_lock(&napi_hash_lock);
6634 hlist_del_init_rcu(&napi->napi_hash_node);
6636 spin_unlock(&napi_hash_lock);
6639 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
6641 struct napi_struct *napi;
6643 napi = container_of(timer, struct napi_struct, timer);
6645 /* Note : we use a relaxed variant of napi_schedule_prep() not setting
6646 * NAPI_STATE_MISSED, since we do not react to a device IRQ.
6648 if (!napi_disable_pending(napi) &&
6649 !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
6650 __napi_schedule_irqoff(napi);
6652 return HRTIMER_NORESTART;
6655 static void init_gro_hash(struct napi_struct *napi)
6659 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6660 INIT_LIST_HEAD(&napi->gro_hash[i].list);
6661 napi->gro_hash[i].count = 0;
6663 napi->gro_bitmask = 0;
6666 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
6667 int (*poll)(struct napi_struct *, int), int weight)
6669 if (WARN_ON(test_and_set_bit(NAPI_STATE_LISTED, &napi->state)))
6672 INIT_LIST_HEAD(&napi->poll_list);
6673 INIT_HLIST_NODE(&napi->napi_hash_node);
6674 hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
6675 napi->timer.function = napi_watchdog;
6676 init_gro_hash(napi);
6678 INIT_LIST_HEAD(&napi->rx_list);
6681 if (weight > NAPI_POLL_WEIGHT)
6682 netdev_err_once(dev, "%s() called with weight %d\n", __func__,
6684 napi->weight = weight;
6686 #ifdef CONFIG_NETPOLL
6687 napi->poll_owner = -1;
6689 set_bit(NAPI_STATE_SCHED, &napi->state);
6690 set_bit(NAPI_STATE_NPSVC, &napi->state);
6691 list_add_rcu(&napi->dev_list, &dev->napi_list);
6692 napi_hash_add(napi);
6694 EXPORT_SYMBOL(netif_napi_add);
6696 void napi_disable(struct napi_struct *n)
6699 set_bit(NAPI_STATE_DISABLE, &n->state);
6701 while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
6703 while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
6706 hrtimer_cancel(&n->timer);
6708 clear_bit(NAPI_STATE_DISABLE, &n->state);
6710 EXPORT_SYMBOL(napi_disable);
6712 static void flush_gro_hash(struct napi_struct *napi)
6716 for (i = 0; i < GRO_HASH_BUCKETS; i++) {
6717 struct sk_buff *skb, *n;
6719 list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list)
6721 napi->gro_hash[i].count = 0;
6725 /* Must be called in process context */
6726 void __netif_napi_del(struct napi_struct *napi)
6728 if (!test_and_clear_bit(NAPI_STATE_LISTED, &napi->state))
6731 napi_hash_del(napi);
6732 list_del_rcu(&napi->dev_list);
6733 napi_free_frags(napi);
6735 flush_gro_hash(napi);
6736 napi->gro_bitmask = 0;
6738 EXPORT_SYMBOL(__netif_napi_del);
6740 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
6745 list_del_init(&n->poll_list);
6747 have = netpoll_poll_lock(n);
6751 /* This NAPI_STATE_SCHED test is for avoiding a race
6752 * with netpoll's poll_napi(). Only the entity which
6753 * obtains the lock and sees NAPI_STATE_SCHED set will
6754 * actually make the ->poll() call. Therefore we avoid
6755 * accidentally calling ->poll() when NAPI is not scheduled.
6758 if (test_bit(NAPI_STATE_SCHED, &n->state)) {
6759 work = n->poll(n, weight);
6760 trace_napi_poll(n, work, weight);
6763 if (unlikely(work > weight))
6764 pr_err_once("NAPI poll function %pS returned %d, exceeding its budget of %d.\n",
6765 n->poll, work, weight);
6767 if (likely(work < weight))
6770 /* Drivers must not modify the NAPI state if they
6771 * consume the entire weight. In such cases this code
6772 * still "owns" the NAPI instance and therefore can
6773 * move the instance around on the list at-will.
6775 if (unlikely(napi_disable_pending(n))) {
6780 if (n->gro_bitmask) {
6781 /* flush too old packets
6782 * If HZ < 1000, flush all packets.
6784 napi_gro_flush(n, HZ >= 1000);
6789 /* Some drivers may have called napi_schedule
6790 * prior to exhausting their budget.
6792 if (unlikely(!list_empty(&n->poll_list))) {
6793 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
6794 n->dev ? n->dev->name : "backlog");
6798 list_add_tail(&n->poll_list, repoll);
6801 netpoll_poll_unlock(have);
6806 static __latent_entropy void net_rx_action(struct softirq_action *h)
6808 struct softnet_data *sd = this_cpu_ptr(&softnet_data);
6809 unsigned long time_limit = jiffies +
6810 usecs_to_jiffies(netdev_budget_usecs);
6811 int budget = netdev_budget;
6815 local_irq_disable();
6816 list_splice_init(&sd->poll_list, &list);
6820 struct napi_struct *n;
6822 if (list_empty(&list)) {
6823 if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
6828 n = list_first_entry(&list, struct napi_struct, poll_list);
6829 budget -= napi_poll(n, &repoll);
6831 /* If softirq window is exhausted then punt.
6832 * Allow this to run for 2 jiffies since which will allow
6833 * an average latency of 1.5/HZ.
6835 if (unlikely(budget <= 0 ||
6836 time_after_eq(jiffies, time_limit))) {
6842 local_irq_disable();
6844 list_splice_tail_init(&sd->poll_list, &list);
6845 list_splice_tail(&repoll, &list);
6846 list_splice(&list, &sd->poll_list);
6847 if (!list_empty(&sd->poll_list))
6848 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
6850 net_rps_action_and_irq_enable(sd);
6852 __kfree_skb_flush();
6855 struct netdev_adjacent {
6856 struct net_device *dev;
6858 /* upper master flag, there can only be one master device per list */
6861 /* lookup ignore flag */
6864 /* counter for the number of times this device was added to us */
6867 /* private field for the users */
6870 struct list_head list;
6871 struct rcu_head rcu;
6874 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
6875 struct list_head *adj_list)
6877 struct netdev_adjacent *adj;
6879 list_for_each_entry(adj, adj_list, list) {
6880 if (adj->dev == adj_dev)
6886 static int ____netdev_has_upper_dev(struct net_device *upper_dev,
6887 struct netdev_nested_priv *priv)
6889 struct net_device *dev = (struct net_device *)priv->data;
6891 return upper_dev == dev;
6895 * netdev_has_upper_dev - Check if device is linked to an upper device
6897 * @upper_dev: upper device to check
6899 * Find out if a device is linked to specified upper device and return true
6900 * in case it is. Note that this checks only immediate upper device,
6901 * not through a complete stack of devices. The caller must hold the RTNL lock.
6903 bool netdev_has_upper_dev(struct net_device *dev,
6904 struct net_device *upper_dev)
6906 struct netdev_nested_priv priv = {
6907 .data = (void *)upper_dev,
6912 return netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6915 EXPORT_SYMBOL(netdev_has_upper_dev);
6918 * netdev_has_upper_dev_all - Check if device is linked to an upper device
6920 * @upper_dev: upper device to check
6922 * Find out if a device is linked to specified upper device and return true
6923 * in case it is. Note that this checks the entire upper device chain.
6924 * The caller must hold rcu lock.
6927 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
6928 struct net_device *upper_dev)
6930 struct netdev_nested_priv priv = {
6931 .data = (void *)upper_dev,
6934 return !!netdev_walk_all_upper_dev_rcu(dev, ____netdev_has_upper_dev,
6937 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
6940 * netdev_has_any_upper_dev - Check if device is linked to some device
6943 * Find out if a device is linked to an upper device and return true in case
6944 * it is. The caller must hold the RTNL lock.
6946 bool netdev_has_any_upper_dev(struct net_device *dev)
6950 return !list_empty(&dev->adj_list.upper);
6952 EXPORT_SYMBOL(netdev_has_any_upper_dev);
6955 * netdev_master_upper_dev_get - Get master upper device
6958 * Find a master upper device and return pointer to it or NULL in case
6959 * it's not there. The caller must hold the RTNL lock.
6961 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
6963 struct netdev_adjacent *upper;
6967 if (list_empty(&dev->adj_list.upper))
6970 upper = list_first_entry(&dev->adj_list.upper,
6971 struct netdev_adjacent, list);
6972 if (likely(upper->master))
6976 EXPORT_SYMBOL(netdev_master_upper_dev_get);
6978 static struct net_device *__netdev_master_upper_dev_get(struct net_device *dev)
6980 struct netdev_adjacent *upper;
6984 if (list_empty(&dev->adj_list.upper))
6987 upper = list_first_entry(&dev->adj_list.upper,
6988 struct netdev_adjacent, list);
6989 if (likely(upper->master) && !upper->ignore)
6995 * netdev_has_any_lower_dev - Check if device is linked to some device
6998 * Find out if a device is linked to a lower device and return true in case
6999 * it is. The caller must hold the RTNL lock.
7001 static bool netdev_has_any_lower_dev(struct net_device *dev)
7005 return !list_empty(&dev->adj_list.lower);
7008 void *netdev_adjacent_get_private(struct list_head *adj_list)
7010 struct netdev_adjacent *adj;
7012 adj = list_entry(adj_list, struct netdev_adjacent, list);
7014 return adj->private;
7016 EXPORT_SYMBOL(netdev_adjacent_get_private);
7019 * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
7021 * @iter: list_head ** of the current position
7023 * Gets the next device from the dev's upper list, starting from iter
7024 * position. The caller must hold RCU read lock.
7026 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
7027 struct list_head **iter)
7029 struct netdev_adjacent *upper;
7031 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7033 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7035 if (&upper->list == &dev->adj_list.upper)
7038 *iter = &upper->list;
7042 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
7044 static struct net_device *__netdev_next_upper_dev(struct net_device *dev,
7045 struct list_head **iter,
7048 struct netdev_adjacent *upper;
7050 upper = list_entry((*iter)->next, struct netdev_adjacent, list);
7052 if (&upper->list == &dev->adj_list.upper)
7055 *iter = &upper->list;
7056 *ignore = upper->ignore;
7061 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
7062 struct list_head **iter)
7064 struct netdev_adjacent *upper;
7066 WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
7068 upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7070 if (&upper->list == &dev->adj_list.upper)
7073 *iter = &upper->list;
7078 static int __netdev_walk_all_upper_dev(struct net_device *dev,
7079 int (*fn)(struct net_device *dev,
7080 struct netdev_nested_priv *priv),
7081 struct netdev_nested_priv *priv)
7083 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7084 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7089 iter = &dev->adj_list.upper;
7093 ret = fn(now, priv);
7100 udev = __netdev_next_upper_dev(now, &iter, &ignore);
7107 niter = &udev->adj_list.upper;
7108 dev_stack[cur] = now;
7109 iter_stack[cur++] = iter;
7116 next = dev_stack[--cur];
7117 niter = iter_stack[cur];
7127 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
7128 int (*fn)(struct net_device *dev,
7129 struct netdev_nested_priv *priv),
7130 struct netdev_nested_priv *priv)
7132 struct net_device *udev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7133 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7137 iter = &dev->adj_list.upper;
7141 ret = fn(now, priv);
7148 udev = netdev_next_upper_dev_rcu(now, &iter);
7153 niter = &udev->adj_list.upper;
7154 dev_stack[cur] = now;
7155 iter_stack[cur++] = iter;
7162 next = dev_stack[--cur];
7163 niter = iter_stack[cur];
7172 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
7174 static bool __netdev_has_upper_dev(struct net_device *dev,
7175 struct net_device *upper_dev)
7177 struct netdev_nested_priv priv = {
7179 .data = (void *)upper_dev,
7184 return __netdev_walk_all_upper_dev(dev, ____netdev_has_upper_dev,
7189 * netdev_lower_get_next_private - Get the next ->private from the
7190 * lower neighbour list
7192 * @iter: list_head ** of the current position
7194 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7195 * list, starting from iter position. The caller must hold either hold the
7196 * RTNL lock or its own locking that guarantees that the neighbour lower
7197 * list will remain unchanged.
7199 void *netdev_lower_get_next_private(struct net_device *dev,
7200 struct list_head **iter)
7202 struct netdev_adjacent *lower;
7204 lower = list_entry(*iter, struct netdev_adjacent, list);
7206 if (&lower->list == &dev->adj_list.lower)
7209 *iter = lower->list.next;
7211 return lower->private;
7213 EXPORT_SYMBOL(netdev_lower_get_next_private);
7216 * netdev_lower_get_next_private_rcu - Get the next ->private from the
7217 * lower neighbour list, RCU
7220 * @iter: list_head ** of the current position
7222 * Gets the next netdev_adjacent->private from the dev's lower neighbour
7223 * list, starting from iter position. The caller must hold RCU read lock.
7225 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
7226 struct list_head **iter)
7228 struct netdev_adjacent *lower;
7230 WARN_ON_ONCE(!rcu_read_lock_held());
7232 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7234 if (&lower->list == &dev->adj_list.lower)
7237 *iter = &lower->list;
7239 return lower->private;
7241 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
7244 * netdev_lower_get_next - Get the next device from the lower neighbour
7247 * @iter: list_head ** of the current position
7249 * Gets the next netdev_adjacent from the dev's lower neighbour
7250 * list, starting from iter position. The caller must hold RTNL lock or
7251 * its own locking that guarantees that the neighbour lower
7252 * list will remain unchanged.
7254 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
7256 struct netdev_adjacent *lower;
7258 lower = list_entry(*iter, struct netdev_adjacent, list);
7260 if (&lower->list == &dev->adj_list.lower)
7263 *iter = lower->list.next;
7267 EXPORT_SYMBOL(netdev_lower_get_next);
7269 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
7270 struct list_head **iter)
7272 struct netdev_adjacent *lower;
7274 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7276 if (&lower->list == &dev->adj_list.lower)
7279 *iter = &lower->list;
7284 static struct net_device *__netdev_next_lower_dev(struct net_device *dev,
7285 struct list_head **iter,
7288 struct netdev_adjacent *lower;
7290 lower = list_entry((*iter)->next, struct netdev_adjacent, list);
7292 if (&lower->list == &dev->adj_list.lower)
7295 *iter = &lower->list;
7296 *ignore = lower->ignore;
7301 int netdev_walk_all_lower_dev(struct net_device *dev,
7302 int (*fn)(struct net_device *dev,
7303 struct netdev_nested_priv *priv),
7304 struct netdev_nested_priv *priv)
7306 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7307 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7311 iter = &dev->adj_list.lower;
7315 ret = fn(now, priv);
7322 ldev = netdev_next_lower_dev(now, &iter);
7327 niter = &ldev->adj_list.lower;
7328 dev_stack[cur] = now;
7329 iter_stack[cur++] = iter;
7336 next = dev_stack[--cur];
7337 niter = iter_stack[cur];
7346 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
7348 static int __netdev_walk_all_lower_dev(struct net_device *dev,
7349 int (*fn)(struct net_device *dev,
7350 struct netdev_nested_priv *priv),
7351 struct netdev_nested_priv *priv)
7353 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7354 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7359 iter = &dev->adj_list.lower;
7363 ret = fn(now, priv);
7370 ldev = __netdev_next_lower_dev(now, &iter, &ignore);
7377 niter = &ldev->adj_list.lower;
7378 dev_stack[cur] = now;
7379 iter_stack[cur++] = iter;
7386 next = dev_stack[--cur];
7387 niter = iter_stack[cur];
7397 struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
7398 struct list_head **iter)
7400 struct netdev_adjacent *lower;
7402 lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
7403 if (&lower->list == &dev->adj_list.lower)
7406 *iter = &lower->list;
7410 EXPORT_SYMBOL(netdev_next_lower_dev_rcu);
7412 static u8 __netdev_upper_depth(struct net_device *dev)
7414 struct net_device *udev;
7415 struct list_head *iter;
7419 for (iter = &dev->adj_list.upper,
7420 udev = __netdev_next_upper_dev(dev, &iter, &ignore);
7422 udev = __netdev_next_upper_dev(dev, &iter, &ignore)) {
7425 if (max_depth < udev->upper_level)
7426 max_depth = udev->upper_level;
7432 static u8 __netdev_lower_depth(struct net_device *dev)
7434 struct net_device *ldev;
7435 struct list_head *iter;
7439 for (iter = &dev->adj_list.lower,
7440 ldev = __netdev_next_lower_dev(dev, &iter, &ignore);
7442 ldev = __netdev_next_lower_dev(dev, &iter, &ignore)) {
7445 if (max_depth < ldev->lower_level)
7446 max_depth = ldev->lower_level;
7452 static int __netdev_update_upper_level(struct net_device *dev,
7453 struct netdev_nested_priv *__unused)
7455 dev->upper_level = __netdev_upper_depth(dev) + 1;
7459 static int __netdev_update_lower_level(struct net_device *dev,
7460 struct netdev_nested_priv *priv)
7462 dev->lower_level = __netdev_lower_depth(dev) + 1;
7464 #ifdef CONFIG_LOCKDEP
7468 if (priv->flags & NESTED_SYNC_IMM)
7469 dev->nested_level = dev->lower_level - 1;
7470 if (priv->flags & NESTED_SYNC_TODO)
7471 net_unlink_todo(dev);
7476 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
7477 int (*fn)(struct net_device *dev,
7478 struct netdev_nested_priv *priv),
7479 struct netdev_nested_priv *priv)
7481 struct net_device *ldev, *next, *now, *dev_stack[MAX_NEST_DEV + 1];
7482 struct list_head *niter, *iter, *iter_stack[MAX_NEST_DEV + 1];
7486 iter = &dev->adj_list.lower;
7490 ret = fn(now, priv);
7497 ldev = netdev_next_lower_dev_rcu(now, &iter);
7502 niter = &ldev->adj_list.lower;
7503 dev_stack[cur] = now;
7504 iter_stack[cur++] = iter;
7511 next = dev_stack[--cur];
7512 niter = iter_stack[cur];
7521 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
7524 * netdev_lower_get_first_private_rcu - Get the first ->private from the
7525 * lower neighbour list, RCU
7529 * Gets the first netdev_adjacent->private from the dev's lower neighbour
7530 * list. The caller must hold RCU read lock.
7532 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
7534 struct netdev_adjacent *lower;
7536 lower = list_first_or_null_rcu(&dev->adj_list.lower,
7537 struct netdev_adjacent, list);
7539 return lower->private;
7542 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
7545 * netdev_master_upper_dev_get_rcu - Get master upper device
7548 * Find a master upper device and return pointer to it or NULL in case
7549 * it's not there. The caller must hold the RCU read lock.
7551 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
7553 struct netdev_adjacent *upper;
7555 upper = list_first_or_null_rcu(&dev->adj_list.upper,
7556 struct netdev_adjacent, list);
7557 if (upper && likely(upper->master))
7561 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
7563 static int netdev_adjacent_sysfs_add(struct net_device *dev,
7564 struct net_device *adj_dev,
7565 struct list_head *dev_list)
7567 char linkname[IFNAMSIZ+7];
7569 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7570 "upper_%s" : "lower_%s", adj_dev->name);
7571 return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
7574 static void netdev_adjacent_sysfs_del(struct net_device *dev,
7576 struct list_head *dev_list)
7578 char linkname[IFNAMSIZ+7];
7580 sprintf(linkname, dev_list == &dev->adj_list.upper ?
7581 "upper_%s" : "lower_%s", name);
7582 sysfs_remove_link(&(dev->dev.kobj), linkname);
7585 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
7586 struct net_device *adj_dev,
7587 struct list_head *dev_list)
7589 return (dev_list == &dev->adj_list.upper ||
7590 dev_list == &dev->adj_list.lower) &&
7591 net_eq(dev_net(dev), dev_net(adj_dev));
7594 static int __netdev_adjacent_dev_insert(struct net_device *dev,
7595 struct net_device *adj_dev,
7596 struct list_head *dev_list,
7597 void *private, bool master)
7599 struct netdev_adjacent *adj;
7602 adj = __netdev_find_adj(adj_dev, dev_list);
7606 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
7607 dev->name, adj_dev->name, adj->ref_nr);
7612 adj = kmalloc(sizeof(*adj), GFP_KERNEL);
7617 adj->master = master;
7619 adj->private = private;
7620 adj->ignore = false;
7623 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
7624 dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
7626 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
7627 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
7632 /* Ensure that master link is always the first item in list. */
7634 ret = sysfs_create_link(&(dev->dev.kobj),
7635 &(adj_dev->dev.kobj), "master");
7637 goto remove_symlinks;
7639 list_add_rcu(&adj->list, dev_list);
7641 list_add_tail_rcu(&adj->list, dev_list);
7647 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7648 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7656 static void __netdev_adjacent_dev_remove(struct net_device *dev,
7657 struct net_device *adj_dev,
7659 struct list_head *dev_list)
7661 struct netdev_adjacent *adj;
7663 pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
7664 dev->name, adj_dev->name, ref_nr);
7666 adj = __netdev_find_adj(adj_dev, dev_list);
7669 pr_err("Adjacency does not exist for device %s from %s\n",
7670 dev->name, adj_dev->name);
7675 if (adj->ref_nr > ref_nr) {
7676 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
7677 dev->name, adj_dev->name, ref_nr,
7678 adj->ref_nr - ref_nr);
7679 adj->ref_nr -= ref_nr;
7684 sysfs_remove_link(&(dev->dev.kobj), "master");
7686 if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
7687 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
7689 list_del_rcu(&adj->list);
7690 pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
7691 adj_dev->name, dev->name, adj_dev->name);
7693 kfree_rcu(adj, rcu);
7696 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
7697 struct net_device *upper_dev,
7698 struct list_head *up_list,
7699 struct list_head *down_list,
7700 void *private, bool master)
7704 ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
7709 ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
7712 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
7719 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
7720 struct net_device *upper_dev,
7722 struct list_head *up_list,
7723 struct list_head *down_list)
7725 __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
7726 __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
7729 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
7730 struct net_device *upper_dev,
7731 void *private, bool master)
7733 return __netdev_adjacent_dev_link_lists(dev, upper_dev,
7734 &dev->adj_list.upper,
7735 &upper_dev->adj_list.lower,
7739 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
7740 struct net_device *upper_dev)
7742 __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
7743 &dev->adj_list.upper,
7744 &upper_dev->adj_list.lower);
7747 static int __netdev_upper_dev_link(struct net_device *dev,
7748 struct net_device *upper_dev, bool master,
7749 void *upper_priv, void *upper_info,
7750 struct netdev_nested_priv *priv,
7751 struct netlink_ext_ack *extack)
7753 struct netdev_notifier_changeupper_info changeupper_info = {
7758 .upper_dev = upper_dev,
7761 .upper_info = upper_info,
7763 struct net_device *master_dev;
7768 if (dev == upper_dev)
7771 /* To prevent loops, check if dev is not upper device to upper_dev. */
7772 if (__netdev_has_upper_dev(upper_dev, dev))
7775 if ((dev->lower_level + upper_dev->upper_level) > MAX_NEST_DEV)
7779 if (__netdev_has_upper_dev(dev, upper_dev))
7782 master_dev = __netdev_master_upper_dev_get(dev);
7784 return master_dev == upper_dev ? -EEXIST : -EBUSY;
7787 ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7788 &changeupper_info.info);
7789 ret = notifier_to_errno(ret);
7793 ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
7798 ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7799 &changeupper_info.info);
7800 ret = notifier_to_errno(ret);
7804 __netdev_update_upper_level(dev, NULL);
7805 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7807 __netdev_update_lower_level(upper_dev, priv);
7808 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7814 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7820 * netdev_upper_dev_link - Add a link to the upper device
7822 * @upper_dev: new upper device
7823 * @extack: netlink extended ack
7825 * Adds a link to device which is upper to this one. The caller must hold
7826 * the RTNL lock. On a failure a negative errno code is returned.
7827 * On success the reference counts are adjusted and the function
7830 int netdev_upper_dev_link(struct net_device *dev,
7831 struct net_device *upper_dev,
7832 struct netlink_ext_ack *extack)
7834 struct netdev_nested_priv priv = {
7835 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7839 return __netdev_upper_dev_link(dev, upper_dev, false,
7840 NULL, NULL, &priv, extack);
7842 EXPORT_SYMBOL(netdev_upper_dev_link);
7845 * netdev_master_upper_dev_link - Add a master link to the upper device
7847 * @upper_dev: new upper device
7848 * @upper_priv: upper device private
7849 * @upper_info: upper info to be passed down via notifier
7850 * @extack: netlink extended ack
7852 * Adds a link to device which is upper to this one. In this case, only
7853 * one master upper device can be linked, although other non-master devices
7854 * might be linked as well. The caller must hold the RTNL lock.
7855 * On a failure a negative errno code is returned. On success the reference
7856 * counts are adjusted and the function returns zero.
7858 int netdev_master_upper_dev_link(struct net_device *dev,
7859 struct net_device *upper_dev,
7860 void *upper_priv, void *upper_info,
7861 struct netlink_ext_ack *extack)
7863 struct netdev_nested_priv priv = {
7864 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7868 return __netdev_upper_dev_link(dev, upper_dev, true,
7869 upper_priv, upper_info, &priv, extack);
7871 EXPORT_SYMBOL(netdev_master_upper_dev_link);
7873 static void __netdev_upper_dev_unlink(struct net_device *dev,
7874 struct net_device *upper_dev,
7875 struct netdev_nested_priv *priv)
7877 struct netdev_notifier_changeupper_info changeupper_info = {
7881 .upper_dev = upper_dev,
7887 changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
7889 call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
7890 &changeupper_info.info);
7892 __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
7894 call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
7895 &changeupper_info.info);
7897 __netdev_update_upper_level(dev, NULL);
7898 __netdev_walk_all_lower_dev(dev, __netdev_update_upper_level, NULL);
7900 __netdev_update_lower_level(upper_dev, priv);
7901 __netdev_walk_all_upper_dev(upper_dev, __netdev_update_lower_level,
7906 * netdev_upper_dev_unlink - Removes a link to upper device
7908 * @upper_dev: new upper device
7910 * Removes a link to device which is upper to this one. The caller must hold
7913 void netdev_upper_dev_unlink(struct net_device *dev,
7914 struct net_device *upper_dev)
7916 struct netdev_nested_priv priv = {
7917 .flags = NESTED_SYNC_TODO,
7921 __netdev_upper_dev_unlink(dev, upper_dev, &priv);
7923 EXPORT_SYMBOL(netdev_upper_dev_unlink);
7925 static void __netdev_adjacent_dev_set(struct net_device *upper_dev,
7926 struct net_device *lower_dev,
7929 struct netdev_adjacent *adj;
7931 adj = __netdev_find_adj(lower_dev, &upper_dev->adj_list.lower);
7935 adj = __netdev_find_adj(upper_dev, &lower_dev->adj_list.upper);
7940 static void netdev_adjacent_dev_disable(struct net_device *upper_dev,
7941 struct net_device *lower_dev)
7943 __netdev_adjacent_dev_set(upper_dev, lower_dev, true);
7946 static void netdev_adjacent_dev_enable(struct net_device *upper_dev,
7947 struct net_device *lower_dev)
7949 __netdev_adjacent_dev_set(upper_dev, lower_dev, false);
7952 int netdev_adjacent_change_prepare(struct net_device *old_dev,
7953 struct net_device *new_dev,
7954 struct net_device *dev,
7955 struct netlink_ext_ack *extack)
7957 struct netdev_nested_priv priv = {
7966 if (old_dev && new_dev != old_dev)
7967 netdev_adjacent_dev_disable(dev, old_dev);
7968 err = __netdev_upper_dev_link(new_dev, dev, false, NULL, NULL, &priv,
7971 if (old_dev && new_dev != old_dev)
7972 netdev_adjacent_dev_enable(dev, old_dev);
7978 EXPORT_SYMBOL(netdev_adjacent_change_prepare);
7980 void netdev_adjacent_change_commit(struct net_device *old_dev,
7981 struct net_device *new_dev,
7982 struct net_device *dev)
7984 struct netdev_nested_priv priv = {
7985 .flags = NESTED_SYNC_IMM | NESTED_SYNC_TODO,
7989 if (!new_dev || !old_dev)
7992 if (new_dev == old_dev)
7995 netdev_adjacent_dev_enable(dev, old_dev);
7996 __netdev_upper_dev_unlink(old_dev, dev, &priv);
7998 EXPORT_SYMBOL(netdev_adjacent_change_commit);
8000 void netdev_adjacent_change_abort(struct net_device *old_dev,
8001 struct net_device *new_dev,
8002 struct net_device *dev)
8004 struct netdev_nested_priv priv = {
8012 if (old_dev && new_dev != old_dev)
8013 netdev_adjacent_dev_enable(dev, old_dev);
8015 __netdev_upper_dev_unlink(new_dev, dev, &priv);
8017 EXPORT_SYMBOL(netdev_adjacent_change_abort);
8020 * netdev_bonding_info_change - Dispatch event about slave change
8022 * @bonding_info: info to dispatch
8024 * Send NETDEV_BONDING_INFO to netdev notifiers with info.
8025 * The caller must hold the RTNL lock.
8027 void netdev_bonding_info_change(struct net_device *dev,
8028 struct netdev_bonding_info *bonding_info)
8030 struct netdev_notifier_bonding_info info = {
8034 memcpy(&info.bonding_info, bonding_info,
8035 sizeof(struct netdev_bonding_info));
8036 call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
8039 EXPORT_SYMBOL(netdev_bonding_info_change);
8042 * netdev_get_xmit_slave - Get the xmit slave of master device
8045 * @all_slaves: assume all the slaves are active
8047 * The reference counters are not incremented so the caller must be
8048 * careful with locks. The caller must hold RCU lock.
8049 * %NULL is returned if no slave is found.
8052 struct net_device *netdev_get_xmit_slave(struct net_device *dev,
8053 struct sk_buff *skb,
8056 const struct net_device_ops *ops = dev->netdev_ops;
8058 if (!ops->ndo_get_xmit_slave)
8060 return ops->ndo_get_xmit_slave(dev, skb, all_slaves);
8062 EXPORT_SYMBOL(netdev_get_xmit_slave);
8064 static void netdev_adjacent_add_links(struct net_device *dev)
8066 struct netdev_adjacent *iter;
8068 struct net *net = dev_net(dev);
8070 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8071 if (!net_eq(net, dev_net(iter->dev)))
8073 netdev_adjacent_sysfs_add(iter->dev, dev,
8074 &iter->dev->adj_list.lower);
8075 netdev_adjacent_sysfs_add(dev, iter->dev,
8076 &dev->adj_list.upper);
8079 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8080 if (!net_eq(net, dev_net(iter->dev)))
8082 netdev_adjacent_sysfs_add(iter->dev, dev,
8083 &iter->dev->adj_list.upper);
8084 netdev_adjacent_sysfs_add(dev, iter->dev,
8085 &dev->adj_list.lower);
8089 static void netdev_adjacent_del_links(struct net_device *dev)
8091 struct netdev_adjacent *iter;
8093 struct net *net = dev_net(dev);
8095 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8096 if (!net_eq(net, dev_net(iter->dev)))
8098 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8099 &iter->dev->adj_list.lower);
8100 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8101 &dev->adj_list.upper);
8104 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8105 if (!net_eq(net, dev_net(iter->dev)))
8107 netdev_adjacent_sysfs_del(iter->dev, dev->name,
8108 &iter->dev->adj_list.upper);
8109 netdev_adjacent_sysfs_del(dev, iter->dev->name,
8110 &dev->adj_list.lower);
8114 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
8116 struct netdev_adjacent *iter;
8118 struct net *net = dev_net(dev);
8120 list_for_each_entry(iter, &dev->adj_list.upper, list) {
8121 if (!net_eq(net, dev_net(iter->dev)))
8123 netdev_adjacent_sysfs_del(iter->dev, oldname,
8124 &iter->dev->adj_list.lower);
8125 netdev_adjacent_sysfs_add(iter->dev, dev,
8126 &iter->dev->adj_list.lower);
8129 list_for_each_entry(iter, &dev->adj_list.lower, list) {
8130 if (!net_eq(net, dev_net(iter->dev)))
8132 netdev_adjacent_sysfs_del(iter->dev, oldname,
8133 &iter->dev->adj_list.upper);
8134 netdev_adjacent_sysfs_add(iter->dev, dev,
8135 &iter->dev->adj_list.upper);
8139 void *netdev_lower_dev_get_private(struct net_device *dev,
8140 struct net_device *lower_dev)
8142 struct netdev_adjacent *lower;
8146 lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
8150 return lower->private;
8152 EXPORT_SYMBOL(netdev_lower_dev_get_private);
8156 * netdev_lower_change - Dispatch event about lower device state change
8157 * @lower_dev: device
8158 * @lower_state_info: state to dispatch
8160 * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
8161 * The caller must hold the RTNL lock.
8163 void netdev_lower_state_changed(struct net_device *lower_dev,
8164 void *lower_state_info)
8166 struct netdev_notifier_changelowerstate_info changelowerstate_info = {
8167 .info.dev = lower_dev,
8171 changelowerstate_info.lower_state_info = lower_state_info;
8172 call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
8173 &changelowerstate_info.info);
8175 EXPORT_SYMBOL(netdev_lower_state_changed);
8177 static void dev_change_rx_flags(struct net_device *dev, int flags)
8179 const struct net_device_ops *ops = dev->netdev_ops;
8181 if (ops->ndo_change_rx_flags)
8182 ops->ndo_change_rx_flags(dev, flags);
8185 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
8187 unsigned int old_flags = dev->flags;
8193 dev->flags |= IFF_PROMISC;
8194 dev->promiscuity += inc;
8195 if (dev->promiscuity == 0) {
8198 * If inc causes overflow, untouch promisc and return error.
8201 dev->flags &= ~IFF_PROMISC;
8203 dev->promiscuity -= inc;
8204 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
8209 if (dev->flags != old_flags) {
8210 pr_info("device %s %s promiscuous mode\n",
8212 dev->flags & IFF_PROMISC ? "entered" : "left");
8213 if (audit_enabled) {
8214 current_uid_gid(&uid, &gid);
8215 audit_log(audit_context(), GFP_ATOMIC,
8216 AUDIT_ANOM_PROMISCUOUS,
8217 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
8218 dev->name, (dev->flags & IFF_PROMISC),
8219 (old_flags & IFF_PROMISC),
8220 from_kuid(&init_user_ns, audit_get_loginuid(current)),
8221 from_kuid(&init_user_ns, uid),
8222 from_kgid(&init_user_ns, gid),
8223 audit_get_sessionid(current));
8226 dev_change_rx_flags(dev, IFF_PROMISC);
8229 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
8234 * dev_set_promiscuity - update promiscuity count on a device
8238 * Add or remove promiscuity from a device. While the count in the device
8239 * remains above zero the interface remains promiscuous. Once it hits zero
8240 * the device reverts back to normal filtering operation. A negative inc
8241 * value is used to drop promiscuity on the device.
8242 * Return 0 if successful or a negative errno code on error.
8244 int dev_set_promiscuity(struct net_device *dev, int inc)
8246 unsigned int old_flags = dev->flags;
8249 err = __dev_set_promiscuity(dev, inc, true);
8252 if (dev->flags != old_flags)
8253 dev_set_rx_mode(dev);
8256 EXPORT_SYMBOL(dev_set_promiscuity);
8258 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
8260 unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
8264 dev->flags |= IFF_ALLMULTI;
8265 dev->allmulti += inc;
8266 if (dev->allmulti == 0) {
8269 * If inc causes overflow, untouch allmulti and return error.
8272 dev->flags &= ~IFF_ALLMULTI;
8274 dev->allmulti -= inc;
8275 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
8280 if (dev->flags ^ old_flags) {
8281 dev_change_rx_flags(dev, IFF_ALLMULTI);
8282 dev_set_rx_mode(dev);
8284 __dev_notify_flags(dev, old_flags,
8285 dev->gflags ^ old_gflags);
8291 * dev_set_allmulti - update allmulti count on a device
8295 * Add or remove reception of all multicast frames to a device. While the
8296 * count in the device remains above zero the interface remains listening
8297 * to all interfaces. Once it hits zero the device reverts back to normal
8298 * filtering operation. A negative @inc value is used to drop the counter
8299 * when releasing a resource needing all multicasts.
8300 * Return 0 if successful or a negative errno code on error.
8303 int dev_set_allmulti(struct net_device *dev, int inc)
8305 return __dev_set_allmulti(dev, inc, true);
8307 EXPORT_SYMBOL(dev_set_allmulti);
8310 * Upload unicast and multicast address lists to device and
8311 * configure RX filtering. When the device doesn't support unicast
8312 * filtering it is put in promiscuous mode while unicast addresses
8315 void __dev_set_rx_mode(struct net_device *dev)
8317 const struct net_device_ops *ops = dev->netdev_ops;
8319 /* dev_open will call this function so the list will stay sane. */
8320 if (!(dev->flags&IFF_UP))
8323 if (!netif_device_present(dev))
8326 if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
8327 /* Unicast addresses changes may only happen under the rtnl,
8328 * therefore calling __dev_set_promiscuity here is safe.
8330 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
8331 __dev_set_promiscuity(dev, 1, false);
8332 dev->uc_promisc = true;
8333 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
8334 __dev_set_promiscuity(dev, -1, false);
8335 dev->uc_promisc = false;
8339 if (ops->ndo_set_rx_mode)
8340 ops->ndo_set_rx_mode(dev);
8343 void dev_set_rx_mode(struct net_device *dev)
8345 netif_addr_lock_bh(dev);
8346 __dev_set_rx_mode(dev);
8347 netif_addr_unlock_bh(dev);
8351 * dev_get_flags - get flags reported to userspace
8354 * Get the combination of flag bits exported through APIs to userspace.
8356 unsigned int dev_get_flags(const struct net_device *dev)
8360 flags = (dev->flags & ~(IFF_PROMISC |
8365 (dev->gflags & (IFF_PROMISC |
8368 if (netif_running(dev)) {
8369 if (netif_oper_up(dev))
8370 flags |= IFF_RUNNING;
8371 if (netif_carrier_ok(dev))
8372 flags |= IFF_LOWER_UP;
8373 if (netif_dormant(dev))
8374 flags |= IFF_DORMANT;
8379 EXPORT_SYMBOL(dev_get_flags);
8381 int __dev_change_flags(struct net_device *dev, unsigned int flags,
8382 struct netlink_ext_ack *extack)
8384 unsigned int old_flags = dev->flags;
8390 * Set the flags on our device.
8393 dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
8394 IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
8396 (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
8400 * Load in the correct multicast list now the flags have changed.
8403 if ((old_flags ^ flags) & IFF_MULTICAST)
8404 dev_change_rx_flags(dev, IFF_MULTICAST);
8406 dev_set_rx_mode(dev);
8409 * Have we downed the interface. We handle IFF_UP ourselves
8410 * according to user attempts to set it, rather than blindly
8415 if ((old_flags ^ flags) & IFF_UP) {
8416 if (old_flags & IFF_UP)
8419 ret = __dev_open(dev, extack);
8422 if ((flags ^ dev->gflags) & IFF_PROMISC) {
8423 int inc = (flags & IFF_PROMISC) ? 1 : -1;
8424 unsigned int old_flags = dev->flags;
8426 dev->gflags ^= IFF_PROMISC;
8428 if (__dev_set_promiscuity(dev, inc, false) >= 0)
8429 if (dev->flags != old_flags)
8430 dev_set_rx_mode(dev);
8433 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
8434 * is important. Some (broken) drivers set IFF_PROMISC, when
8435 * IFF_ALLMULTI is requested not asking us and not reporting.
8437 if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
8438 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
8440 dev->gflags ^= IFF_ALLMULTI;
8441 __dev_set_allmulti(dev, inc, false);
8447 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
8448 unsigned int gchanges)
8450 unsigned int changes = dev->flags ^ old_flags;
8453 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
8455 if (changes & IFF_UP) {
8456 if (dev->flags & IFF_UP)
8457 call_netdevice_notifiers(NETDEV_UP, dev);
8459 call_netdevice_notifiers(NETDEV_DOWN, dev);
8462 if (dev->flags & IFF_UP &&
8463 (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
8464 struct netdev_notifier_change_info change_info = {
8468 .flags_changed = changes,
8471 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
8476 * dev_change_flags - change device settings
8478 * @flags: device state flags
8479 * @extack: netlink extended ack
8481 * Change settings on device based state flags. The flags are
8482 * in the userspace exported format.
8484 int dev_change_flags(struct net_device *dev, unsigned int flags,
8485 struct netlink_ext_ack *extack)
8488 unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
8490 ret = __dev_change_flags(dev, flags, extack);
8494 changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
8495 __dev_notify_flags(dev, old_flags, changes);
8498 EXPORT_SYMBOL(dev_change_flags);
8500 int __dev_set_mtu(struct net_device *dev, int new_mtu)
8502 const struct net_device_ops *ops = dev->netdev_ops;
8504 if (ops->ndo_change_mtu)
8505 return ops->ndo_change_mtu(dev, new_mtu);
8507 /* Pairs with all the lockless reads of dev->mtu in the stack */
8508 WRITE_ONCE(dev->mtu, new_mtu);
8511 EXPORT_SYMBOL(__dev_set_mtu);
8513 int dev_validate_mtu(struct net_device *dev, int new_mtu,
8514 struct netlink_ext_ack *extack)
8516 /* MTU must be positive, and in range */
8517 if (new_mtu < 0 || new_mtu < dev->min_mtu) {
8518 NL_SET_ERR_MSG(extack, "mtu less than device minimum");
8522 if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
8523 NL_SET_ERR_MSG(extack, "mtu greater than device maximum");
8530 * dev_set_mtu_ext - Change maximum transfer unit
8532 * @new_mtu: new transfer unit
8533 * @extack: netlink extended ack
8535 * Change the maximum transfer size of the network device.
8537 int dev_set_mtu_ext(struct net_device *dev, int new_mtu,
8538 struct netlink_ext_ack *extack)
8542 if (new_mtu == dev->mtu)
8545 err = dev_validate_mtu(dev, new_mtu, extack);
8549 if (!netif_device_present(dev))
8552 err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
8553 err = notifier_to_errno(err);
8557 orig_mtu = dev->mtu;
8558 err = __dev_set_mtu(dev, new_mtu);
8561 err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8563 err = notifier_to_errno(err);
8565 /* setting mtu back and notifying everyone again,
8566 * so that they have a chance to revert changes.
8568 __dev_set_mtu(dev, orig_mtu);
8569 call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev,
8576 int dev_set_mtu(struct net_device *dev, int new_mtu)
8578 struct netlink_ext_ack extack;
8581 memset(&extack, 0, sizeof(extack));
8582 err = dev_set_mtu_ext(dev, new_mtu, &extack);
8583 if (err && extack._msg)
8584 net_err_ratelimited("%s: %s\n", dev->name, extack._msg);
8587 EXPORT_SYMBOL(dev_set_mtu);
8590 * dev_change_tx_queue_len - Change TX queue length of a netdevice
8592 * @new_len: new tx queue length
8594 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
8596 unsigned int orig_len = dev->tx_queue_len;
8599 if (new_len != (unsigned int)new_len)
8602 if (new_len != orig_len) {
8603 dev->tx_queue_len = new_len;
8604 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
8605 res = notifier_to_errno(res);
8608 res = dev_qdisc_change_tx_queue_len(dev);
8616 netdev_err(dev, "refused to change device tx_queue_len\n");
8617 dev->tx_queue_len = orig_len;
8622 * dev_set_group - Change group this device belongs to
8624 * @new_group: group this device should belong to
8626 void dev_set_group(struct net_device *dev, int new_group)
8628 dev->group = new_group;
8630 EXPORT_SYMBOL(dev_set_group);
8633 * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR.
8635 * @addr: new address
8636 * @extack: netlink extended ack
8638 int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr,
8639 struct netlink_ext_ack *extack)
8641 struct netdev_notifier_pre_changeaddr_info info = {
8643 .info.extack = extack,
8648 rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info);
8649 return notifier_to_errno(rc);
8651 EXPORT_SYMBOL(dev_pre_changeaddr_notify);
8654 * dev_set_mac_address - Change Media Access Control Address
8657 * @extack: netlink extended ack
8659 * Change the hardware (MAC) address of the device
8661 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa,
8662 struct netlink_ext_ack *extack)
8664 const struct net_device_ops *ops = dev->netdev_ops;
8667 if (!ops->ndo_set_mac_address)
8669 if (sa->sa_family != dev->type)
8671 if (!netif_device_present(dev))
8673 err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack);
8676 err = ops->ndo_set_mac_address(dev, sa);
8679 dev->addr_assign_type = NET_ADDR_SET;
8680 call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
8681 add_device_randomness(dev->dev_addr, dev->addr_len);
8684 EXPORT_SYMBOL(dev_set_mac_address);
8687 * dev_change_carrier - Change device carrier
8689 * @new_carrier: new value
8691 * Change device carrier
8693 int dev_change_carrier(struct net_device *dev, bool new_carrier)
8695 const struct net_device_ops *ops = dev->netdev_ops;
8697 if (!ops->ndo_change_carrier)
8699 if (!netif_device_present(dev))
8701 return ops->ndo_change_carrier(dev, new_carrier);
8703 EXPORT_SYMBOL(dev_change_carrier);
8706 * dev_get_phys_port_id - Get device physical port ID
8710 * Get device physical port ID
8712 int dev_get_phys_port_id(struct net_device *dev,
8713 struct netdev_phys_item_id *ppid)
8715 const struct net_device_ops *ops = dev->netdev_ops;
8717 if (!ops->ndo_get_phys_port_id)
8719 return ops->ndo_get_phys_port_id(dev, ppid);
8721 EXPORT_SYMBOL(dev_get_phys_port_id);
8724 * dev_get_phys_port_name - Get device physical port name
8727 * @len: limit of bytes to copy to name
8729 * Get device physical port name
8731 int dev_get_phys_port_name(struct net_device *dev,
8732 char *name, size_t len)
8734 const struct net_device_ops *ops = dev->netdev_ops;
8737 if (ops->ndo_get_phys_port_name) {
8738 err = ops->ndo_get_phys_port_name(dev, name, len);
8739 if (err != -EOPNOTSUPP)
8742 return devlink_compat_phys_port_name_get(dev, name, len);
8744 EXPORT_SYMBOL(dev_get_phys_port_name);
8747 * dev_get_port_parent_id - Get the device's port parent identifier
8748 * @dev: network device
8749 * @ppid: pointer to a storage for the port's parent identifier
8750 * @recurse: allow/disallow recursion to lower devices
8752 * Get the devices's port parent identifier
8754 int dev_get_port_parent_id(struct net_device *dev,
8755 struct netdev_phys_item_id *ppid,
8758 const struct net_device_ops *ops = dev->netdev_ops;
8759 struct netdev_phys_item_id first = { };
8760 struct net_device *lower_dev;
8761 struct list_head *iter;
8764 if (ops->ndo_get_port_parent_id) {
8765 err = ops->ndo_get_port_parent_id(dev, ppid);
8766 if (err != -EOPNOTSUPP)
8770 err = devlink_compat_switch_id_get(dev, ppid);
8771 if (!err || err != -EOPNOTSUPP)
8777 netdev_for_each_lower_dev(dev, lower_dev, iter) {
8778 err = dev_get_port_parent_id(lower_dev, ppid, recurse);
8783 else if (memcmp(&first, ppid, sizeof(*ppid)))
8789 EXPORT_SYMBOL(dev_get_port_parent_id);
8792 * netdev_port_same_parent_id - Indicate if two network devices have
8793 * the same port parent identifier
8794 * @a: first network device
8795 * @b: second network device
8797 bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b)
8799 struct netdev_phys_item_id a_id = { };
8800 struct netdev_phys_item_id b_id = { };
8802 if (dev_get_port_parent_id(a, &a_id, true) ||
8803 dev_get_port_parent_id(b, &b_id, true))
8806 return netdev_phys_item_id_same(&a_id, &b_id);
8808 EXPORT_SYMBOL(netdev_port_same_parent_id);
8811 * dev_change_proto_down - update protocol port state information
8813 * @proto_down: new value
8815 * This info can be used by switch drivers to set the phys state of the
8818 int dev_change_proto_down(struct net_device *dev, bool proto_down)
8820 const struct net_device_ops *ops = dev->netdev_ops;
8822 if (!ops->ndo_change_proto_down)
8824 if (!netif_device_present(dev))
8826 return ops->ndo_change_proto_down(dev, proto_down);
8828 EXPORT_SYMBOL(dev_change_proto_down);
8831 * dev_change_proto_down_generic - generic implementation for
8832 * ndo_change_proto_down that sets carrier according to
8836 * @proto_down: new value
8838 int dev_change_proto_down_generic(struct net_device *dev, bool proto_down)
8841 netif_carrier_off(dev);
8843 netif_carrier_on(dev);
8844 dev->proto_down = proto_down;
8847 EXPORT_SYMBOL(dev_change_proto_down_generic);
8850 * dev_change_proto_down_reason - proto down reason
8853 * @mask: proto down mask
8854 * @value: proto down value
8856 void dev_change_proto_down_reason(struct net_device *dev, unsigned long mask,
8862 dev->proto_down_reason = value;
8864 for_each_set_bit(b, &mask, 32) {
8865 if (value & (1 << b))
8866 dev->proto_down_reason |= BIT(b);
8868 dev->proto_down_reason &= ~BIT(b);
8872 EXPORT_SYMBOL(dev_change_proto_down_reason);
8874 struct bpf_xdp_link {
8875 struct bpf_link link;
8876 struct net_device *dev; /* protected by rtnl_lock, no refcnt held */
8880 static enum bpf_xdp_mode dev_xdp_mode(struct net_device *dev, u32 flags)
8882 if (flags & XDP_FLAGS_HW_MODE)
8884 if (flags & XDP_FLAGS_DRV_MODE)
8885 return XDP_MODE_DRV;
8886 if (flags & XDP_FLAGS_SKB_MODE)
8887 return XDP_MODE_SKB;
8888 return dev->netdev_ops->ndo_bpf ? XDP_MODE_DRV : XDP_MODE_SKB;
8891 static bpf_op_t dev_xdp_bpf_op(struct net_device *dev, enum bpf_xdp_mode mode)
8895 return generic_xdp_install;
8898 return dev->netdev_ops->ndo_bpf;
8904 static struct bpf_xdp_link *dev_xdp_link(struct net_device *dev,
8905 enum bpf_xdp_mode mode)
8907 return dev->xdp_state[mode].link;
8910 static struct bpf_prog *dev_xdp_prog(struct net_device *dev,
8911 enum bpf_xdp_mode mode)
8913 struct bpf_xdp_link *link = dev_xdp_link(dev, mode);
8916 return link->link.prog;
8917 return dev->xdp_state[mode].prog;
8920 u32 dev_xdp_prog_id(struct net_device *dev, enum bpf_xdp_mode mode)
8922 struct bpf_prog *prog = dev_xdp_prog(dev, mode);
8924 return prog ? prog->aux->id : 0;
8927 static void dev_xdp_set_link(struct net_device *dev, enum bpf_xdp_mode mode,
8928 struct bpf_xdp_link *link)
8930 dev->xdp_state[mode].link = link;
8931 dev->xdp_state[mode].prog = NULL;
8934 static void dev_xdp_set_prog(struct net_device *dev, enum bpf_xdp_mode mode,
8935 struct bpf_prog *prog)
8937 dev->xdp_state[mode].link = NULL;
8938 dev->xdp_state[mode].prog = prog;
8941 static int dev_xdp_install(struct net_device *dev, enum bpf_xdp_mode mode,
8942 bpf_op_t bpf_op, struct netlink_ext_ack *extack,
8943 u32 flags, struct bpf_prog *prog)
8945 struct netdev_bpf xdp;
8948 memset(&xdp, 0, sizeof(xdp));
8949 xdp.command = mode == XDP_MODE_HW ? XDP_SETUP_PROG_HW : XDP_SETUP_PROG;
8950 xdp.extack = extack;
8954 /* Drivers assume refcnt is already incremented (i.e, prog pointer is
8955 * "moved" into driver), so they don't increment it on their own, but
8956 * they do decrement refcnt when program is detached or replaced.
8957 * Given net_device also owns link/prog, we need to bump refcnt here
8958 * to prevent drivers from underflowing it.
8962 err = bpf_op(dev, &xdp);
8969 if (mode != XDP_MODE_HW)
8970 bpf_prog_change_xdp(dev_xdp_prog(dev, mode), prog);
8975 static void dev_xdp_uninstall(struct net_device *dev)
8977 struct bpf_xdp_link *link;
8978 struct bpf_prog *prog;
8979 enum bpf_xdp_mode mode;
8984 for (mode = XDP_MODE_SKB; mode < __MAX_XDP_MODE; mode++) {
8985 prog = dev_xdp_prog(dev, mode);
8989 bpf_op = dev_xdp_bpf_op(dev, mode);
8993 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
8995 /* auto-detach link from net device */
8996 link = dev_xdp_link(dev, mode);
9002 dev_xdp_set_link(dev, mode, NULL);
9006 static int dev_xdp_attach(struct net_device *dev, struct netlink_ext_ack *extack,
9007 struct bpf_xdp_link *link, struct bpf_prog *new_prog,
9008 struct bpf_prog *old_prog, u32 flags)
9010 struct bpf_prog *cur_prog;
9011 enum bpf_xdp_mode mode;
9017 /* either link or prog attachment, never both */
9018 if (link && (new_prog || old_prog))
9020 /* link supports only XDP mode flags */
9021 if (link && (flags & ~XDP_FLAGS_MODES)) {
9022 NL_SET_ERR_MSG(extack, "Invalid XDP flags for BPF link attachment");
9025 /* just one XDP mode bit should be set, zero defaults to SKB mode */
9026 if (hweight32(flags & XDP_FLAGS_MODES) > 1) {
9027 NL_SET_ERR_MSG(extack, "Only one XDP mode flag can be set");
9030 /* old_prog != NULL implies XDP_FLAGS_REPLACE is set */
9031 if (old_prog && !(flags & XDP_FLAGS_REPLACE)) {
9032 NL_SET_ERR_MSG(extack, "XDP_FLAGS_REPLACE is not specified");
9036 mode = dev_xdp_mode(dev, flags);
9037 /* can't replace attached link */
9038 if (dev_xdp_link(dev, mode)) {
9039 NL_SET_ERR_MSG(extack, "Can't replace active BPF XDP link");
9043 cur_prog = dev_xdp_prog(dev, mode);
9044 /* can't replace attached prog with link */
9045 if (link && cur_prog) {
9046 NL_SET_ERR_MSG(extack, "Can't replace active XDP program with BPF link");
9049 if ((flags & XDP_FLAGS_REPLACE) && cur_prog != old_prog) {
9050 NL_SET_ERR_MSG(extack, "Active program does not match expected");
9054 /* put effective new program into new_prog */
9056 new_prog = link->link.prog;
9059 bool offload = mode == XDP_MODE_HW;
9060 enum bpf_xdp_mode other_mode = mode == XDP_MODE_SKB
9061 ? XDP_MODE_DRV : XDP_MODE_SKB;
9063 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && cur_prog) {
9064 NL_SET_ERR_MSG(extack, "XDP program already attached");
9067 if (!offload && dev_xdp_prog(dev, other_mode)) {
9068 NL_SET_ERR_MSG(extack, "Native and generic XDP can't be active at the same time");
9071 if (!offload && bpf_prog_is_dev_bound(new_prog->aux)) {
9072 NL_SET_ERR_MSG(extack, "Using device-bound program without HW_MODE flag is not supported");
9075 if (new_prog->expected_attach_type == BPF_XDP_DEVMAP) {
9076 NL_SET_ERR_MSG(extack, "BPF_XDP_DEVMAP programs can not be attached to a device");
9079 if (new_prog->expected_attach_type == BPF_XDP_CPUMAP) {
9080 NL_SET_ERR_MSG(extack, "BPF_XDP_CPUMAP programs can not be attached to a device");
9085 /* don't call drivers if the effective program didn't change */
9086 if (new_prog != cur_prog) {
9087 bpf_op = dev_xdp_bpf_op(dev, mode);
9089 NL_SET_ERR_MSG(extack, "Underlying driver does not support XDP in native mode");
9093 err = dev_xdp_install(dev, mode, bpf_op, extack, flags, new_prog);
9099 dev_xdp_set_link(dev, mode, link);
9101 dev_xdp_set_prog(dev, mode, new_prog);
9103 bpf_prog_put(cur_prog);
9108 static int dev_xdp_attach_link(struct net_device *dev,
9109 struct netlink_ext_ack *extack,
9110 struct bpf_xdp_link *link)
9112 return dev_xdp_attach(dev, extack, link, NULL, NULL, link->flags);
9115 static int dev_xdp_detach_link(struct net_device *dev,
9116 struct netlink_ext_ack *extack,
9117 struct bpf_xdp_link *link)
9119 enum bpf_xdp_mode mode;
9124 mode = dev_xdp_mode(dev, link->flags);
9125 if (dev_xdp_link(dev, mode) != link)
9128 bpf_op = dev_xdp_bpf_op(dev, mode);
9129 WARN_ON(dev_xdp_install(dev, mode, bpf_op, NULL, 0, NULL));
9130 dev_xdp_set_link(dev, mode, NULL);
9134 static void bpf_xdp_link_release(struct bpf_link *link)
9136 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9140 /* if racing with net_device's tear down, xdp_link->dev might be
9141 * already NULL, in which case link was already auto-detached
9143 if (xdp_link->dev) {
9144 WARN_ON(dev_xdp_detach_link(xdp_link->dev, NULL, xdp_link));
9145 xdp_link->dev = NULL;
9151 static int bpf_xdp_link_detach(struct bpf_link *link)
9153 bpf_xdp_link_release(link);
9157 static void bpf_xdp_link_dealloc(struct bpf_link *link)
9159 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9164 static void bpf_xdp_link_show_fdinfo(const struct bpf_link *link,
9165 struct seq_file *seq)
9167 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9172 ifindex = xdp_link->dev->ifindex;
9175 seq_printf(seq, "ifindex:\t%u\n", ifindex);
9178 static int bpf_xdp_link_fill_link_info(const struct bpf_link *link,
9179 struct bpf_link_info *info)
9181 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9186 ifindex = xdp_link->dev->ifindex;
9189 info->xdp.ifindex = ifindex;
9193 static int bpf_xdp_link_update(struct bpf_link *link, struct bpf_prog *new_prog,
9194 struct bpf_prog *old_prog)
9196 struct bpf_xdp_link *xdp_link = container_of(link, struct bpf_xdp_link, link);
9197 enum bpf_xdp_mode mode;
9203 /* link might have been auto-released already, so fail */
9204 if (!xdp_link->dev) {
9209 if (old_prog && link->prog != old_prog) {
9213 old_prog = link->prog;
9214 if (old_prog == new_prog) {
9215 /* no-op, don't disturb drivers */
9216 bpf_prog_put(new_prog);
9220 mode = dev_xdp_mode(xdp_link->dev, xdp_link->flags);
9221 bpf_op = dev_xdp_bpf_op(xdp_link->dev, mode);
9222 err = dev_xdp_install(xdp_link->dev, mode, bpf_op, NULL,
9223 xdp_link->flags, new_prog);
9227 old_prog = xchg(&link->prog, new_prog);
9228 bpf_prog_put(old_prog);
9235 static const struct bpf_link_ops bpf_xdp_link_lops = {
9236 .release = bpf_xdp_link_release,
9237 .dealloc = bpf_xdp_link_dealloc,
9238 .detach = bpf_xdp_link_detach,
9239 .show_fdinfo = bpf_xdp_link_show_fdinfo,
9240 .fill_link_info = bpf_xdp_link_fill_link_info,
9241 .update_prog = bpf_xdp_link_update,
9244 int bpf_xdp_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
9246 struct net *net = current->nsproxy->net_ns;
9247 struct bpf_link_primer link_primer;
9248 struct bpf_xdp_link *link;
9249 struct net_device *dev;
9252 dev = dev_get_by_index(net, attr->link_create.target_ifindex);
9256 link = kzalloc(sizeof(*link), GFP_USER);
9262 bpf_link_init(&link->link, BPF_LINK_TYPE_XDP, &bpf_xdp_link_lops, prog);
9264 link->flags = attr->link_create.flags;
9266 err = bpf_link_prime(&link->link, &link_primer);
9273 err = dev_xdp_attach_link(dev, NULL, link);
9277 bpf_link_cleanup(&link_primer);
9281 fd = bpf_link_settle(&link_primer);
9282 /* link itself doesn't hold dev's refcnt to not complicate shutdown */
9292 * dev_change_xdp_fd - set or clear a bpf program for a device rx path
9294 * @extack: netlink extended ack
9295 * @fd: new program fd or negative value to clear
9296 * @expected_fd: old program fd that userspace expects to replace or clear
9297 * @flags: xdp-related flags
9299 * Set or clear a bpf program for a device
9301 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
9302 int fd, int expected_fd, u32 flags)
9304 enum bpf_xdp_mode mode = dev_xdp_mode(dev, flags);
9305 struct bpf_prog *new_prog = NULL, *old_prog = NULL;
9311 new_prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
9312 mode != XDP_MODE_SKB);
9313 if (IS_ERR(new_prog))
9314 return PTR_ERR(new_prog);
9317 if (expected_fd >= 0) {
9318 old_prog = bpf_prog_get_type_dev(expected_fd, BPF_PROG_TYPE_XDP,
9319 mode != XDP_MODE_SKB);
9320 if (IS_ERR(old_prog)) {
9321 err = PTR_ERR(old_prog);
9327 err = dev_xdp_attach(dev, extack, NULL, new_prog, old_prog, flags);
9330 if (err && new_prog)
9331 bpf_prog_put(new_prog);
9333 bpf_prog_put(old_prog);
9338 * dev_new_index - allocate an ifindex
9339 * @net: the applicable net namespace
9341 * Returns a suitable unique value for a new device interface
9342 * number. The caller must hold the rtnl semaphore or the
9343 * dev_base_lock to be sure it remains unique.
9345 static int dev_new_index(struct net *net)
9347 int ifindex = net->ifindex;
9352 if (!__dev_get_by_index(net, ifindex))
9353 return net->ifindex = ifindex;
9357 /* Delayed registration/unregisteration */
9358 static LIST_HEAD(net_todo_list);
9359 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
9361 static void net_set_todo(struct net_device *dev)
9363 list_add_tail(&dev->todo_list, &net_todo_list);
9364 dev_net(dev)->dev_unreg_count++;
9367 static void rollback_registered_many(struct list_head *head)
9369 struct net_device *dev, *tmp;
9370 LIST_HEAD(close_head);
9372 BUG_ON(dev_boot_phase);
9375 list_for_each_entry_safe(dev, tmp, head, unreg_list) {
9376 /* Some devices call without registering
9377 * for initialization unwind. Remove those
9378 * devices and proceed with the remaining.
9380 if (dev->reg_state == NETREG_UNINITIALIZED) {
9381 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
9385 list_del(&dev->unreg_list);
9388 dev->dismantle = true;
9389 BUG_ON(dev->reg_state != NETREG_REGISTERED);
9392 /* If device is running, close it first. */
9393 list_for_each_entry(dev, head, unreg_list)
9394 list_add_tail(&dev->close_list, &close_head);
9395 dev_close_many(&close_head, true);
9397 list_for_each_entry(dev, head, unreg_list) {
9398 /* And unlink it from device chain. */
9399 unlist_netdevice(dev);
9401 dev->reg_state = NETREG_UNREGISTERING;
9403 flush_all_backlogs();
9407 list_for_each_entry(dev, head, unreg_list) {
9408 struct sk_buff *skb = NULL;
9410 /* Shutdown queueing discipline. */
9413 dev_xdp_uninstall(dev);
9415 /* Notify protocols, that we are about to destroy
9416 * this device. They should clean all the things.
9418 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
9420 if (!dev->rtnl_link_ops ||
9421 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
9422 skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
9423 GFP_KERNEL, NULL, 0);
9426 * Flush the unicast and multicast chains
9431 netdev_name_node_alt_flush(dev);
9432 netdev_name_node_free(dev->name_node);
9434 if (dev->netdev_ops->ndo_uninit)
9435 dev->netdev_ops->ndo_uninit(dev);
9438 rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
9440 /* Notifier chain MUST detach us all upper devices. */
9441 WARN_ON(netdev_has_any_upper_dev(dev));
9442 WARN_ON(netdev_has_any_lower_dev(dev));
9444 /* Remove entries from kobject tree */
9445 netdev_unregister_kobject(dev);
9447 /* Remove XPS queueing entries */
9448 netif_reset_xps_queues_gt(dev, 0);
9454 list_for_each_entry(dev, head, unreg_list)
9458 static void rollback_registered(struct net_device *dev)
9462 list_add(&dev->unreg_list, &single);
9463 rollback_registered_many(&single);
9467 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
9468 struct net_device *upper, netdev_features_t features)
9470 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9471 netdev_features_t feature;
9474 for_each_netdev_feature(upper_disables, feature_bit) {
9475 feature = __NETIF_F_BIT(feature_bit);
9476 if (!(upper->wanted_features & feature)
9477 && (features & feature)) {
9478 netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
9479 &feature, upper->name);
9480 features &= ~feature;
9487 static void netdev_sync_lower_features(struct net_device *upper,
9488 struct net_device *lower, netdev_features_t features)
9490 netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
9491 netdev_features_t feature;
9494 for_each_netdev_feature(upper_disables, feature_bit) {
9495 feature = __NETIF_F_BIT(feature_bit);
9496 if (!(features & feature) && (lower->features & feature)) {
9497 netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
9498 &feature, lower->name);
9499 lower->wanted_features &= ~feature;
9500 __netdev_update_features(lower);
9502 if (unlikely(lower->features & feature))
9503 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
9504 &feature, lower->name);
9506 netdev_features_change(lower);
9511 static netdev_features_t netdev_fix_features(struct net_device *dev,
9512 netdev_features_t features)
9514 /* Fix illegal checksum combinations */
9515 if ((features & NETIF_F_HW_CSUM) &&
9516 (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
9517 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
9518 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
9521 /* TSO requires that SG is present as well. */
9522 if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
9523 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
9524 features &= ~NETIF_F_ALL_TSO;
9527 if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
9528 !(features & NETIF_F_IP_CSUM)) {
9529 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
9530 features &= ~NETIF_F_TSO;
9531 features &= ~NETIF_F_TSO_ECN;
9534 if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
9535 !(features & NETIF_F_IPV6_CSUM)) {
9536 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
9537 features &= ~NETIF_F_TSO6;
9540 /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
9541 if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
9542 features &= ~NETIF_F_TSO_MANGLEID;
9544 /* TSO ECN requires that TSO is present as well. */
9545 if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
9546 features &= ~NETIF_F_TSO_ECN;
9548 /* Software GSO depends on SG. */
9549 if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
9550 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
9551 features &= ~NETIF_F_GSO;
9554 /* GSO partial features require GSO partial be set */
9555 if ((features & dev->gso_partial_features) &&
9556 !(features & NETIF_F_GSO_PARTIAL)) {
9558 "Dropping partially supported GSO features since no GSO partial.\n");
9559 features &= ~dev->gso_partial_features;
9562 if (!(features & NETIF_F_RXCSUM)) {
9563 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
9564 * successfully merged by hardware must also have the
9565 * checksum verified by hardware. If the user does not
9566 * want to enable RXCSUM, logically, we should disable GRO_HW.
9568 if (features & NETIF_F_GRO_HW) {
9569 netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
9570 features &= ~NETIF_F_GRO_HW;
9574 /* LRO/HW-GRO features cannot be combined with RX-FCS */
9575 if (features & NETIF_F_RXFCS) {
9576 if (features & NETIF_F_LRO) {
9577 netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
9578 features &= ~NETIF_F_LRO;
9581 if (features & NETIF_F_GRO_HW) {
9582 netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
9583 features &= ~NETIF_F_GRO_HW;
9590 int __netdev_update_features(struct net_device *dev)
9592 struct net_device *upper, *lower;
9593 netdev_features_t features;
9594 struct list_head *iter;
9599 features = netdev_get_wanted_features(dev);
9601 if (dev->netdev_ops->ndo_fix_features)
9602 features = dev->netdev_ops->ndo_fix_features(dev, features);
9604 /* driver might be less strict about feature dependencies */
9605 features = netdev_fix_features(dev, features);
9607 /* some features can't be enabled if they're off on an upper device */
9608 netdev_for_each_upper_dev_rcu(dev, upper, iter)
9609 features = netdev_sync_upper_features(dev, upper, features);
9611 if (dev->features == features)
9614 netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
9615 &dev->features, &features);
9617 if (dev->netdev_ops->ndo_set_features)
9618 err = dev->netdev_ops->ndo_set_features(dev, features);
9622 if (unlikely(err < 0)) {
9624 "set_features() failed (%d); wanted %pNF, left %pNF\n",
9625 err, &features, &dev->features);
9626 /* return non-0 since some features might have changed and
9627 * it's better to fire a spurious notification than miss it
9633 /* some features must be disabled on lower devices when disabled
9634 * on an upper device (think: bonding master or bridge)
9636 netdev_for_each_lower_dev(dev, lower, iter)
9637 netdev_sync_lower_features(dev, lower, features);
9640 netdev_features_t diff = features ^ dev->features;
9642 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
9643 /* udp_tunnel_{get,drop}_rx_info both need
9644 * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
9645 * device, or they won't do anything.
9646 * Thus we need to update dev->features
9647 * *before* calling udp_tunnel_get_rx_info,
9648 * but *after* calling udp_tunnel_drop_rx_info.
9650 if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
9651 dev->features = features;
9652 udp_tunnel_get_rx_info(dev);
9654 udp_tunnel_drop_rx_info(dev);
9658 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
9659 if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
9660 dev->features = features;
9661 err |= vlan_get_rx_ctag_filter_info(dev);
9663 vlan_drop_rx_ctag_filter_info(dev);
9667 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
9668 if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
9669 dev->features = features;
9670 err |= vlan_get_rx_stag_filter_info(dev);
9672 vlan_drop_rx_stag_filter_info(dev);
9676 dev->features = features;
9679 return err < 0 ? 0 : 1;
9683 * netdev_update_features - recalculate device features
9684 * @dev: the device to check
9686 * Recalculate dev->features set and send notifications if it
9687 * has changed. Should be called after driver or hardware dependent
9688 * conditions might have changed that influence the features.
9690 void netdev_update_features(struct net_device *dev)
9692 if (__netdev_update_features(dev))
9693 netdev_features_change(dev);
9695 EXPORT_SYMBOL(netdev_update_features);
9698 * netdev_change_features - recalculate device features
9699 * @dev: the device to check
9701 * Recalculate dev->features set and send notifications even
9702 * if they have not changed. Should be called instead of
9703 * netdev_update_features() if also dev->vlan_features might
9704 * have changed to allow the changes to be propagated to stacked
9707 void netdev_change_features(struct net_device *dev)
9709 __netdev_update_features(dev);
9710 netdev_features_change(dev);
9712 EXPORT_SYMBOL(netdev_change_features);
9715 * netif_stacked_transfer_operstate - transfer operstate
9716 * @rootdev: the root or lower level device to transfer state from
9717 * @dev: the device to transfer operstate to
9719 * Transfer operational state from root to device. This is normally
9720 * called when a stacking relationship exists between the root
9721 * device and the device(a leaf device).
9723 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
9724 struct net_device *dev)
9726 if (rootdev->operstate == IF_OPER_DORMANT)
9727 netif_dormant_on(dev);
9729 netif_dormant_off(dev);
9731 if (rootdev->operstate == IF_OPER_TESTING)
9732 netif_testing_on(dev);
9734 netif_testing_off(dev);
9736 if (netif_carrier_ok(rootdev))
9737 netif_carrier_on(dev);
9739 netif_carrier_off(dev);
9741 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
9743 static int netif_alloc_rx_queues(struct net_device *dev)
9745 unsigned int i, count = dev->num_rx_queues;
9746 struct netdev_rx_queue *rx;
9747 size_t sz = count * sizeof(*rx);
9752 rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9758 for (i = 0; i < count; i++) {
9761 /* XDP RX-queue setup */
9762 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
9769 /* Rollback successful reg's and free other resources */
9771 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
9777 static void netif_free_rx_queues(struct net_device *dev)
9779 unsigned int i, count = dev->num_rx_queues;
9781 /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
9785 for (i = 0; i < count; i++)
9786 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
9791 static void netdev_init_one_queue(struct net_device *dev,
9792 struct netdev_queue *queue, void *_unused)
9794 /* Initialize queue lock */
9795 spin_lock_init(&queue->_xmit_lock);
9796 netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
9797 queue->xmit_lock_owner = -1;
9798 netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
9801 dql_init(&queue->dql, HZ);
9805 static void netif_free_tx_queues(struct net_device *dev)
9810 static int netif_alloc_netdev_queues(struct net_device *dev)
9812 unsigned int count = dev->num_tx_queues;
9813 struct netdev_queue *tx;
9814 size_t sz = count * sizeof(*tx);
9816 if (count < 1 || count > 0xffff)
9819 tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
9825 netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
9826 spin_lock_init(&dev->tx_global_lock);
9831 void netif_tx_stop_all_queues(struct net_device *dev)
9835 for (i = 0; i < dev->num_tx_queues; i++) {
9836 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
9838 netif_tx_stop_queue(txq);
9841 EXPORT_SYMBOL(netif_tx_stop_all_queues);
9844 * register_netdevice - register a network device
9845 * @dev: device to register
9847 * Take a completed network device structure and add it to the kernel
9848 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
9849 * chain. 0 is returned on success. A negative errno code is returned
9850 * on a failure to set up the device, or if the name is a duplicate.
9852 * Callers must hold the rtnl semaphore. You may want
9853 * register_netdev() instead of this.
9856 * The locking appears insufficient to guarantee two parallel registers
9857 * will not get the same name.
9860 int register_netdevice(struct net_device *dev)
9863 struct net *net = dev_net(dev);
9865 BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
9866 NETDEV_FEATURE_COUNT);
9867 BUG_ON(dev_boot_phase);
9872 /* When net_device's are persistent, this will be fatal. */
9873 BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
9876 ret = ethtool_check_ops(dev->ethtool_ops);
9880 spin_lock_init(&dev->addr_list_lock);
9881 netdev_set_addr_lockdep_class(dev);
9883 ret = dev_get_valid_name(net, dev, dev->name);
9888 dev->name_node = netdev_name_node_head_alloc(dev);
9889 if (!dev->name_node)
9892 /* Init, if this function is available */
9893 if (dev->netdev_ops->ndo_init) {
9894 ret = dev->netdev_ops->ndo_init(dev);
9902 if (((dev->hw_features | dev->features) &
9903 NETIF_F_HW_VLAN_CTAG_FILTER) &&
9904 (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
9905 !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
9906 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
9913 dev->ifindex = dev_new_index(net);
9914 else if (__dev_get_by_index(net, dev->ifindex))
9917 /* Transfer changeable features to wanted_features and enable
9918 * software offloads (GSO and GRO).
9920 dev->hw_features |= (NETIF_F_SOFT_FEATURES | NETIF_F_SOFT_FEATURES_OFF);
9921 dev->features |= NETIF_F_SOFT_FEATURES;
9923 if (dev->netdev_ops->ndo_udp_tunnel_add) {
9924 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9925 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
9928 dev->wanted_features = dev->features & dev->hw_features;
9930 if (!(dev->flags & IFF_LOOPBACK))
9931 dev->hw_features |= NETIF_F_NOCACHE_COPY;
9933 /* If IPv4 TCP segmentation offload is supported we should also
9934 * allow the device to enable segmenting the frame with the option
9935 * of ignoring a static IP ID value. This doesn't enable the
9936 * feature itself but allows the user to enable it later.
9938 if (dev->hw_features & NETIF_F_TSO)
9939 dev->hw_features |= NETIF_F_TSO_MANGLEID;
9940 if (dev->vlan_features & NETIF_F_TSO)
9941 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
9942 if (dev->mpls_features & NETIF_F_TSO)
9943 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
9944 if (dev->hw_enc_features & NETIF_F_TSO)
9945 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
9947 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
9949 dev->vlan_features |= NETIF_F_HIGHDMA;
9951 /* Make NETIF_F_SG inheritable to tunnel devices.
9953 dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
9955 /* Make NETIF_F_SG inheritable to MPLS.
9957 dev->mpls_features |= NETIF_F_SG;
9959 ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
9960 ret = notifier_to_errno(ret);
9964 ret = netdev_register_kobject(dev);
9966 dev->reg_state = NETREG_UNREGISTERED;
9969 dev->reg_state = NETREG_REGISTERED;
9971 __netdev_update_features(dev);
9974 * Default initial state at registry is that the
9975 * device is present.
9978 set_bit(__LINK_STATE_PRESENT, &dev->state);
9980 linkwatch_init_dev(dev);
9982 dev_init_scheduler(dev);
9984 list_netdevice(dev);
9985 add_device_randomness(dev->dev_addr, dev->addr_len);
9987 /* If the device has permanent device address, driver should
9988 * set dev_addr and also addr_assign_type should be set to
9989 * NET_ADDR_PERM (default value).
9991 if (dev->addr_assign_type == NET_ADDR_PERM)
9992 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
9994 /* Notify protocols, that a new device appeared. */
9995 ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
9996 ret = notifier_to_errno(ret);
9998 rollback_registered(dev);
10001 dev->reg_state = NETREG_UNREGISTERED;
10002 /* We should put the kobject that hold in
10003 * netdev_unregister_kobject(), otherwise
10004 * the net device cannot be freed when
10005 * driver calls free_netdev(), because the
10006 * kobject is being hold.
10008 kobject_put(&dev->dev.kobj);
10011 * Prevent userspace races by waiting until the network
10012 * device is fully setup before sending notifications.
10014 if (!dev->rtnl_link_ops ||
10015 dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
10016 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10022 if (dev->netdev_ops->ndo_uninit)
10023 dev->netdev_ops->ndo_uninit(dev);
10024 if (dev->priv_destructor)
10025 dev->priv_destructor(dev);
10027 netdev_name_node_free(dev->name_node);
10030 EXPORT_SYMBOL(register_netdevice);
10033 * init_dummy_netdev - init a dummy network device for NAPI
10034 * @dev: device to init
10036 * This takes a network device structure and initialize the minimum
10037 * amount of fields so it can be used to schedule NAPI polls without
10038 * registering a full blown interface. This is to be used by drivers
10039 * that need to tie several hardware interfaces to a single NAPI
10040 * poll scheduler due to HW limitations.
10042 int init_dummy_netdev(struct net_device *dev)
10044 /* Clear everything. Note we don't initialize spinlocks
10045 * are they aren't supposed to be taken by any of the
10046 * NAPI code and this dummy netdev is supposed to be
10047 * only ever used for NAPI polls
10049 memset(dev, 0, sizeof(struct net_device));
10051 /* make sure we BUG if trying to hit standard
10052 * register/unregister code path
10054 dev->reg_state = NETREG_DUMMY;
10056 /* NAPI wants this */
10057 INIT_LIST_HEAD(&dev->napi_list);
10059 /* a dummy interface is started by default */
10060 set_bit(__LINK_STATE_PRESENT, &dev->state);
10061 set_bit(__LINK_STATE_START, &dev->state);
10063 /* napi_busy_loop stats accounting wants this */
10064 dev_net_set(dev, &init_net);
10066 /* Note : We dont allocate pcpu_refcnt for dummy devices,
10067 * because users of this 'device' dont need to change
10073 EXPORT_SYMBOL_GPL(init_dummy_netdev);
10077 * register_netdev - register a network device
10078 * @dev: device to register
10080 * Take a completed network device structure and add it to the kernel
10081 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
10082 * chain. 0 is returned on success. A negative errno code is returned
10083 * on a failure to set up the device, or if the name is a duplicate.
10085 * This is a wrapper around register_netdevice that takes the rtnl semaphore
10086 * and expands the device name if you passed a format string to
10089 int register_netdev(struct net_device *dev)
10093 if (rtnl_lock_killable())
10095 err = register_netdevice(dev);
10099 EXPORT_SYMBOL(register_netdev);
10101 int netdev_refcnt_read(const struct net_device *dev)
10105 for_each_possible_cpu(i)
10106 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
10109 EXPORT_SYMBOL(netdev_refcnt_read);
10111 #define WAIT_REFS_MIN_MSECS 1
10112 #define WAIT_REFS_MAX_MSECS 250
10114 * netdev_wait_allrefs - wait until all references are gone.
10115 * @dev: target net_device
10117 * This is called when unregistering network devices.
10119 * Any protocol or device that holds a reference should register
10120 * for netdevice notification, and cleanup and put back the
10121 * reference if they receive an UNREGISTER event.
10122 * We can get stuck here if buggy protocols don't correctly
10125 static void netdev_wait_allrefs(struct net_device *dev)
10127 unsigned long rebroadcast_time, warning_time;
10128 int wait = 0, refcnt;
10130 linkwatch_forget_dev(dev);
10132 rebroadcast_time = warning_time = jiffies;
10133 refcnt = netdev_refcnt_read(dev);
10135 while (refcnt != 0) {
10136 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
10139 /* Rebroadcast unregister notification */
10140 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10146 if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
10148 /* We must not have linkwatch events
10149 * pending on unregister. If this
10150 * happens, we simply run the queue
10151 * unscheduled, resulting in a noop
10154 linkwatch_run_queue();
10159 rebroadcast_time = jiffies;
10164 wait = WAIT_REFS_MIN_MSECS;
10167 wait = min(wait << 1, WAIT_REFS_MAX_MSECS);
10170 refcnt = netdev_refcnt_read(dev);
10172 if (refcnt && time_after(jiffies, warning_time + 10 * HZ)) {
10173 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
10174 dev->name, refcnt);
10175 warning_time = jiffies;
10180 /* The sequence is:
10184 * register_netdevice(x1);
10185 * register_netdevice(x2);
10187 * unregister_netdevice(y1);
10188 * unregister_netdevice(y2);
10194 * We are invoked by rtnl_unlock().
10195 * This allows us to deal with problems:
10196 * 1) We can delete sysfs objects which invoke hotplug
10197 * without deadlocking with linkwatch via keventd.
10198 * 2) Since we run with the RTNL semaphore not held, we can sleep
10199 * safely in order to wait for the netdev refcnt to drop to zero.
10201 * We must not return until all unregister events added during
10202 * the interval the lock was held have been completed.
10204 void netdev_run_todo(void)
10206 struct list_head list;
10207 #ifdef CONFIG_LOCKDEP
10208 struct list_head unlink_list;
10210 list_replace_init(&net_unlink_list, &unlink_list);
10212 while (!list_empty(&unlink_list)) {
10213 struct net_device *dev = list_first_entry(&unlink_list,
10216 list_del_init(&dev->unlink_list);
10217 dev->nested_level = dev->lower_level - 1;
10221 /* Snapshot list, allow later requests */
10222 list_replace_init(&net_todo_list, &list);
10227 /* Wait for rcu callbacks to finish before next phase */
10228 if (!list_empty(&list))
10231 while (!list_empty(&list)) {
10232 struct net_device *dev
10233 = list_first_entry(&list, struct net_device, todo_list);
10234 list_del(&dev->todo_list);
10236 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
10237 pr_err("network todo '%s' but state %d\n",
10238 dev->name, dev->reg_state);
10243 dev->reg_state = NETREG_UNREGISTERED;
10245 netdev_wait_allrefs(dev);
10248 BUG_ON(netdev_refcnt_read(dev));
10249 BUG_ON(!list_empty(&dev->ptype_all));
10250 BUG_ON(!list_empty(&dev->ptype_specific));
10251 WARN_ON(rcu_access_pointer(dev->ip_ptr));
10252 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
10253 #if IS_ENABLED(CONFIG_DECNET)
10254 WARN_ON(dev->dn_ptr);
10256 if (dev->priv_destructor)
10257 dev->priv_destructor(dev);
10258 if (dev->needs_free_netdev)
10261 /* Report a network device has been unregistered */
10263 dev_net(dev)->dev_unreg_count--;
10265 wake_up(&netdev_unregistering_wq);
10267 /* Free network device */
10268 kobject_put(&dev->dev.kobj);
10272 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
10273 * all the same fields in the same order as net_device_stats, with only
10274 * the type differing, but rtnl_link_stats64 may have additional fields
10275 * at the end for newer counters.
10277 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
10278 const struct net_device_stats *netdev_stats)
10280 #if BITS_PER_LONG == 64
10281 BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
10282 memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
10283 /* zero out counters that only exist in rtnl_link_stats64 */
10284 memset((char *)stats64 + sizeof(*netdev_stats), 0,
10285 sizeof(*stats64) - sizeof(*netdev_stats));
10287 size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
10288 const unsigned long *src = (const unsigned long *)netdev_stats;
10289 u64 *dst = (u64 *)stats64;
10291 BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
10292 for (i = 0; i < n; i++)
10294 /* zero out counters that only exist in rtnl_link_stats64 */
10295 memset((char *)stats64 + n * sizeof(u64), 0,
10296 sizeof(*stats64) - n * sizeof(u64));
10299 EXPORT_SYMBOL(netdev_stats_to_stats64);
10302 * dev_get_stats - get network device statistics
10303 * @dev: device to get statistics from
10304 * @storage: place to store stats
10306 * Get network statistics from device. Return @storage.
10307 * The device driver may provide its own method by setting
10308 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
10309 * otherwise the internal statistics structure is used.
10311 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
10312 struct rtnl_link_stats64 *storage)
10314 const struct net_device_ops *ops = dev->netdev_ops;
10316 if (ops->ndo_get_stats64) {
10317 memset(storage, 0, sizeof(*storage));
10318 ops->ndo_get_stats64(dev, storage);
10319 } else if (ops->ndo_get_stats) {
10320 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
10322 netdev_stats_to_stats64(storage, &dev->stats);
10324 storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
10325 storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
10326 storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
10329 EXPORT_SYMBOL(dev_get_stats);
10332 * dev_fetch_sw_netstats - get per-cpu network device statistics
10333 * @s: place to store stats
10334 * @netstats: per-cpu network stats to read from
10336 * Read per-cpu network statistics and populate the related fields in @s.
10338 void dev_fetch_sw_netstats(struct rtnl_link_stats64 *s,
10339 const struct pcpu_sw_netstats __percpu *netstats)
10343 for_each_possible_cpu(cpu) {
10344 const struct pcpu_sw_netstats *stats;
10345 struct pcpu_sw_netstats tmp;
10346 unsigned int start;
10348 stats = per_cpu_ptr(netstats, cpu);
10350 start = u64_stats_fetch_begin_irq(&stats->syncp);
10351 tmp.rx_packets = stats->rx_packets;
10352 tmp.rx_bytes = stats->rx_bytes;
10353 tmp.tx_packets = stats->tx_packets;
10354 tmp.tx_bytes = stats->tx_bytes;
10355 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
10357 s->rx_packets += tmp.rx_packets;
10358 s->rx_bytes += tmp.rx_bytes;
10359 s->tx_packets += tmp.tx_packets;
10360 s->tx_bytes += tmp.tx_bytes;
10363 EXPORT_SYMBOL_GPL(dev_fetch_sw_netstats);
10365 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
10367 struct netdev_queue *queue = dev_ingress_queue(dev);
10369 #ifdef CONFIG_NET_CLS_ACT
10372 queue = kzalloc(sizeof(*queue), GFP_KERNEL);
10375 netdev_init_one_queue(dev, queue, NULL);
10376 RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
10377 queue->qdisc_sleeping = &noop_qdisc;
10378 rcu_assign_pointer(dev->ingress_queue, queue);
10383 static const struct ethtool_ops default_ethtool_ops;
10385 void netdev_set_default_ethtool_ops(struct net_device *dev,
10386 const struct ethtool_ops *ops)
10388 if (dev->ethtool_ops == &default_ethtool_ops)
10389 dev->ethtool_ops = ops;
10391 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
10393 void netdev_freemem(struct net_device *dev)
10395 char *addr = (char *)dev - dev->padded;
10401 * alloc_netdev_mqs - allocate network device
10402 * @sizeof_priv: size of private data to allocate space for
10403 * @name: device name format string
10404 * @name_assign_type: origin of device name
10405 * @setup: callback to initialize device
10406 * @txqs: the number of TX subqueues to allocate
10407 * @rxqs: the number of RX subqueues to allocate
10409 * Allocates a struct net_device with private data area for driver use
10410 * and performs basic initialization. Also allocates subqueue structs
10411 * for each queue on the device.
10413 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
10414 unsigned char name_assign_type,
10415 void (*setup)(struct net_device *),
10416 unsigned int txqs, unsigned int rxqs)
10418 struct net_device *dev;
10419 unsigned int alloc_size;
10420 struct net_device *p;
10422 BUG_ON(strlen(name) >= sizeof(dev->name));
10425 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
10430 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
10434 alloc_size = sizeof(struct net_device);
10436 /* ensure 32-byte alignment of private area */
10437 alloc_size = ALIGN(alloc_size, NETDEV_ALIGN);
10438 alloc_size += sizeof_priv;
10440 /* ensure 32-byte alignment of whole construct */
10441 alloc_size += NETDEV_ALIGN - 1;
10443 p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
10447 dev = PTR_ALIGN(p, NETDEV_ALIGN);
10448 dev->padded = (char *)dev - (char *)p;
10450 dev->pcpu_refcnt = alloc_percpu(int);
10451 if (!dev->pcpu_refcnt)
10454 if (dev_addr_init(dev))
10460 dev_net_set(dev, &init_net);
10462 dev->gso_max_size = GSO_MAX_SIZE;
10463 dev->gso_max_segs = GSO_MAX_SEGS;
10464 dev->upper_level = 1;
10465 dev->lower_level = 1;
10466 #ifdef CONFIG_LOCKDEP
10467 dev->nested_level = 0;
10468 INIT_LIST_HEAD(&dev->unlink_list);
10471 INIT_LIST_HEAD(&dev->napi_list);
10472 INIT_LIST_HEAD(&dev->unreg_list);
10473 INIT_LIST_HEAD(&dev->close_list);
10474 INIT_LIST_HEAD(&dev->link_watch_list);
10475 INIT_LIST_HEAD(&dev->adj_list.upper);
10476 INIT_LIST_HEAD(&dev->adj_list.lower);
10477 INIT_LIST_HEAD(&dev->ptype_all);
10478 INIT_LIST_HEAD(&dev->ptype_specific);
10479 INIT_LIST_HEAD(&dev->net_notifier_list);
10480 #ifdef CONFIG_NET_SCHED
10481 hash_init(dev->qdisc_hash);
10483 dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM;
10486 if (!dev->tx_queue_len) {
10487 dev->priv_flags |= IFF_NO_QUEUE;
10488 dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN;
10491 dev->num_tx_queues = txqs;
10492 dev->real_num_tx_queues = txqs;
10493 if (netif_alloc_netdev_queues(dev))
10496 dev->num_rx_queues = rxqs;
10497 dev->real_num_rx_queues = rxqs;
10498 if (netif_alloc_rx_queues(dev))
10501 strcpy(dev->name, name);
10502 dev->name_assign_type = name_assign_type;
10503 dev->group = INIT_NETDEV_GROUP;
10504 if (!dev->ethtool_ops)
10505 dev->ethtool_ops = &default_ethtool_ops;
10507 nf_hook_ingress_init(dev);
10516 free_percpu(dev->pcpu_refcnt);
10518 netdev_freemem(dev);
10521 EXPORT_SYMBOL(alloc_netdev_mqs);
10524 * free_netdev - free network device
10527 * This function does the last stage of destroying an allocated device
10528 * interface. The reference to the device object is released. If this
10529 * is the last reference then it will be freed.Must be called in process
10532 void free_netdev(struct net_device *dev)
10534 struct napi_struct *p, *n;
10537 netif_free_tx_queues(dev);
10538 netif_free_rx_queues(dev);
10540 kfree(rcu_dereference_protected(dev->ingress_queue, 1));
10542 /* Flush device addresses */
10543 dev_addr_flush(dev);
10545 list_for_each_entry_safe(p, n, &dev->napi_list, dev_list)
10548 free_percpu(dev->pcpu_refcnt);
10549 dev->pcpu_refcnt = NULL;
10550 free_percpu(dev->xdp_bulkq);
10551 dev->xdp_bulkq = NULL;
10553 /* Compatibility with error handling in drivers */
10554 if (dev->reg_state == NETREG_UNINITIALIZED) {
10555 netdev_freemem(dev);
10559 BUG_ON(dev->reg_state != NETREG_UNREGISTERED);
10560 dev->reg_state = NETREG_RELEASED;
10562 /* will free via device release */
10563 put_device(&dev->dev);
10565 EXPORT_SYMBOL(free_netdev);
10568 * synchronize_net - Synchronize with packet receive processing
10570 * Wait for packets currently being received to be done.
10571 * Does not block later packets from starting.
10573 void synchronize_net(void)
10576 if (rtnl_is_locked())
10577 synchronize_rcu_expedited();
10581 EXPORT_SYMBOL(synchronize_net);
10584 * unregister_netdevice_queue - remove device from the kernel
10588 * This function shuts down a device interface and removes it
10589 * from the kernel tables.
10590 * If head not NULL, device is queued to be unregistered later.
10592 * Callers must hold the rtnl semaphore. You may want
10593 * unregister_netdev() instead of this.
10596 void unregister_netdevice_queue(struct net_device *dev, struct list_head *head)
10601 list_move_tail(&dev->unreg_list, head);
10603 rollback_registered(dev);
10604 /* Finish processing unregister after unlock */
10608 EXPORT_SYMBOL(unregister_netdevice_queue);
10611 * unregister_netdevice_many - unregister many devices
10612 * @head: list of devices
10614 * Note: As most callers use a stack allocated list_head,
10615 * we force a list_del() to make sure stack wont be corrupted later.
10617 void unregister_netdevice_many(struct list_head *head)
10619 struct net_device *dev;
10621 if (!list_empty(head)) {
10622 rollback_registered_many(head);
10623 list_for_each_entry(dev, head, unreg_list)
10628 EXPORT_SYMBOL(unregister_netdevice_many);
10631 * unregister_netdev - remove device from the kernel
10634 * This function shuts down a device interface and removes it
10635 * from the kernel tables.
10637 * This is just a wrapper for unregister_netdevice that takes
10638 * the rtnl semaphore. In general you want to use this and not
10639 * unregister_netdevice.
10641 void unregister_netdev(struct net_device *dev)
10644 unregister_netdevice(dev);
10647 EXPORT_SYMBOL(unregister_netdev);
10650 * dev_change_net_namespace - move device to different nethost namespace
10652 * @net: network namespace
10653 * @pat: If not NULL name pattern to try if the current device name
10654 * is already taken in the destination network namespace.
10656 * This function shuts down a device interface and moves it
10657 * to a new network namespace. On success 0 is returned, on
10658 * a failure a netagive errno code is returned.
10660 * Callers must hold the rtnl semaphore.
10663 int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat)
10665 struct net *net_old = dev_net(dev);
10666 int err, new_nsid, new_ifindex;
10670 /* Don't allow namespace local devices to be moved. */
10672 if (dev->features & NETIF_F_NETNS_LOCAL)
10675 /* Ensure the device has been registrered */
10676 if (dev->reg_state != NETREG_REGISTERED)
10679 /* Get out if there is nothing todo */
10681 if (net_eq(net_old, net))
10684 /* Pick the destination device name, and ensure
10685 * we can use it in the destination network namespace.
10688 if (__dev_get_by_name(net, dev->name)) {
10689 /* We get here if we can't use the current device name */
10692 err = dev_get_valid_name(net, dev, pat);
10698 * And now a mini version of register_netdevice unregister_netdevice.
10701 /* If device is running close it first. */
10704 /* And unlink it from device chain */
10705 unlist_netdevice(dev);
10709 /* Shutdown queueing discipline. */
10712 /* Notify protocols, that we are about to destroy
10713 * this device. They should clean all the things.
10715 * Note that dev->reg_state stays at NETREG_REGISTERED.
10716 * This is wanted because this way 8021q and macvlan know
10717 * the device is just moving and can keep their slaves up.
10719 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
10722 new_nsid = peernet2id_alloc(dev_net(dev), net, GFP_KERNEL);
10723 /* If there is an ifindex conflict assign a new one */
10724 if (__dev_get_by_index(net, dev->ifindex))
10725 new_ifindex = dev_new_index(net);
10727 new_ifindex = dev->ifindex;
10729 rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid,
10733 * Flush the unicast and multicast chains
10738 /* Send a netdev-removed uevent to the old namespace */
10739 kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE);
10740 netdev_adjacent_del_links(dev);
10742 /* Move per-net netdevice notifiers that are following the netdevice */
10743 move_netdevice_notifiers_dev_net(dev, net);
10745 /* Actually switch the network namespace */
10746 dev_net_set(dev, net);
10747 dev->ifindex = new_ifindex;
10749 /* Send a netdev-add uevent to the new namespace */
10750 kobject_uevent(&dev->dev.kobj, KOBJ_ADD);
10751 netdev_adjacent_add_links(dev);
10753 /* Fixup kobjects */
10754 err = device_rename(&dev->dev, dev->name);
10757 /* Adapt owner in case owning user namespace of target network
10758 * namespace is different from the original one.
10760 err = netdev_change_owner(dev, net_old, net);
10763 /* Add the device back in the hashes */
10764 list_netdevice(dev);
10766 /* Notify protocols, that a new device appeared. */
10767 call_netdevice_notifiers(NETDEV_REGISTER, dev);
10770 * Prevent userspace races by waiting until the network
10771 * device is fully setup before sending notifications.
10773 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
10780 EXPORT_SYMBOL_GPL(dev_change_net_namespace);
10782 static int dev_cpu_dead(unsigned int oldcpu)
10784 struct sk_buff **list_skb;
10785 struct sk_buff *skb;
10787 struct softnet_data *sd, *oldsd, *remsd = NULL;
10789 local_irq_disable();
10790 cpu = smp_processor_id();
10791 sd = &per_cpu(softnet_data, cpu);
10792 oldsd = &per_cpu(softnet_data, oldcpu);
10794 /* Find end of our completion_queue. */
10795 list_skb = &sd->completion_queue;
10797 list_skb = &(*list_skb)->next;
10798 /* Append completion queue from offline CPU. */
10799 *list_skb = oldsd->completion_queue;
10800 oldsd->completion_queue = NULL;
10802 /* Append output queue from offline CPU. */
10803 if (oldsd->output_queue) {
10804 *sd->output_queue_tailp = oldsd->output_queue;
10805 sd->output_queue_tailp = oldsd->output_queue_tailp;
10806 oldsd->output_queue = NULL;
10807 oldsd->output_queue_tailp = &oldsd->output_queue;
10809 /* Append NAPI poll list from offline CPU, with one exception :
10810 * process_backlog() must be called by cpu owning percpu backlog.
10811 * We properly handle process_queue & input_pkt_queue later.
10813 while (!list_empty(&oldsd->poll_list)) {
10814 struct napi_struct *napi = list_first_entry(&oldsd->poll_list,
10815 struct napi_struct,
10818 list_del_init(&napi->poll_list);
10819 if (napi->poll == process_backlog)
10822 ____napi_schedule(sd, napi);
10825 raise_softirq_irqoff(NET_TX_SOFTIRQ);
10826 local_irq_enable();
10829 remsd = oldsd->rps_ipi_list;
10830 oldsd->rps_ipi_list = NULL;
10832 /* send out pending IPI's on offline CPU */
10833 net_rps_send_ipi(remsd);
10835 /* Process offline CPU's input_pkt_queue */
10836 while ((skb = __skb_dequeue(&oldsd->process_queue))) {
10838 input_queue_head_incr(oldsd);
10840 while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) {
10842 input_queue_head_incr(oldsd);
10849 * netdev_increment_features - increment feature set by one
10850 * @all: current feature set
10851 * @one: new feature set
10852 * @mask: mask feature set
10854 * Computes a new feature set after adding a device with feature set
10855 * @one to the master device with current feature set @all. Will not
10856 * enable anything that is off in @mask. Returns the new feature set.
10858 netdev_features_t netdev_increment_features(netdev_features_t all,
10859 netdev_features_t one, netdev_features_t mask)
10861 if (mask & NETIF_F_HW_CSUM)
10862 mask |= NETIF_F_CSUM_MASK;
10863 mask |= NETIF_F_VLAN_CHALLENGED;
10865 all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask;
10866 all &= one | ~NETIF_F_ALL_FOR_ALL;
10868 /* If one device supports hw checksumming, set for all. */
10869 if (all & NETIF_F_HW_CSUM)
10870 all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM);
10874 EXPORT_SYMBOL(netdev_increment_features);
10876 static struct hlist_head * __net_init netdev_create_hash(void)
10879 struct hlist_head *hash;
10881 hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL);
10883 for (i = 0; i < NETDEV_HASHENTRIES; i++)
10884 INIT_HLIST_HEAD(&hash[i]);
10889 /* Initialize per network namespace state */
10890 static int __net_init netdev_init(struct net *net)
10892 BUILD_BUG_ON(GRO_HASH_BUCKETS >
10893 8 * sizeof_field(struct napi_struct, gro_bitmask));
10895 if (net != &init_net)
10896 INIT_LIST_HEAD(&net->dev_base_head);
10898 net->dev_name_head = netdev_create_hash();
10899 if (net->dev_name_head == NULL)
10902 net->dev_index_head = netdev_create_hash();
10903 if (net->dev_index_head == NULL)
10906 RAW_INIT_NOTIFIER_HEAD(&net->netdev_chain);
10911 kfree(net->dev_name_head);
10917 * netdev_drivername - network driver for the device
10918 * @dev: network device
10920 * Determine network driver for device.
10922 const char *netdev_drivername(const struct net_device *dev)
10924 const struct device_driver *driver;
10925 const struct device *parent;
10926 const char *empty = "";
10928 parent = dev->dev.parent;
10932 driver = parent->driver;
10933 if (driver && driver->name)
10934 return driver->name;
10938 static void __netdev_printk(const char *level, const struct net_device *dev,
10939 struct va_format *vaf)
10941 if (dev && dev->dev.parent) {
10942 dev_printk_emit(level[1] - '0',
10945 dev_driver_string(dev->dev.parent),
10946 dev_name(dev->dev.parent),
10947 netdev_name(dev), netdev_reg_state(dev),
10950 printk("%s%s%s: %pV",
10951 level, netdev_name(dev), netdev_reg_state(dev), vaf);
10953 printk("%s(NULL net_device): %pV", level, vaf);
10957 void netdev_printk(const char *level, const struct net_device *dev,
10958 const char *format, ...)
10960 struct va_format vaf;
10963 va_start(args, format);
10968 __netdev_printk(level, dev, &vaf);
10972 EXPORT_SYMBOL(netdev_printk);
10974 #define define_netdev_printk_level(func, level) \
10975 void func(const struct net_device *dev, const char *fmt, ...) \
10977 struct va_format vaf; \
10980 va_start(args, fmt); \
10985 __netdev_printk(level, dev, &vaf); \
10989 EXPORT_SYMBOL(func);
10991 define_netdev_printk_level(netdev_emerg, KERN_EMERG);
10992 define_netdev_printk_level(netdev_alert, KERN_ALERT);
10993 define_netdev_printk_level(netdev_crit, KERN_CRIT);
10994 define_netdev_printk_level(netdev_err, KERN_ERR);
10995 define_netdev_printk_level(netdev_warn, KERN_WARNING);
10996 define_netdev_printk_level(netdev_notice, KERN_NOTICE);
10997 define_netdev_printk_level(netdev_info, KERN_INFO);
10999 static void __net_exit netdev_exit(struct net *net)
11001 kfree(net->dev_name_head);
11002 kfree(net->dev_index_head);
11003 if (net != &init_net)
11004 WARN_ON_ONCE(!list_empty(&net->dev_base_head));
11007 static struct pernet_operations __net_initdata netdev_net_ops = {
11008 .init = netdev_init,
11009 .exit = netdev_exit,
11012 static void __net_exit default_device_exit(struct net *net)
11014 struct net_device *dev, *aux;
11016 * Push all migratable network devices back to the
11017 * initial network namespace
11020 for_each_netdev_safe(net, dev, aux) {
11022 char fb_name[IFNAMSIZ];
11024 /* Ignore unmoveable devices (i.e. loopback) */
11025 if (dev->features & NETIF_F_NETNS_LOCAL)
11028 /* Leave virtual devices for the generic cleanup */
11029 if (dev->rtnl_link_ops)
11032 /* Push remaining network devices to init_net */
11033 snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex);
11034 if (__dev_get_by_name(&init_net, fb_name))
11035 snprintf(fb_name, IFNAMSIZ, "dev%%d");
11036 err = dev_change_net_namespace(dev, &init_net, fb_name);
11038 pr_emerg("%s: failed to move %s to init_net: %d\n",
11039 __func__, dev->name, err);
11046 static void __net_exit rtnl_lock_unregistering(struct list_head *net_list)
11048 /* Return with the rtnl_lock held when there are no network
11049 * devices unregistering in any network namespace in net_list.
11052 bool unregistering;
11053 DEFINE_WAIT_FUNC(wait, woken_wake_function);
11055 add_wait_queue(&netdev_unregistering_wq, &wait);
11057 unregistering = false;
11059 list_for_each_entry(net, net_list, exit_list) {
11060 if (net->dev_unreg_count > 0) {
11061 unregistering = true;
11065 if (!unregistering)
11069 wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
11071 remove_wait_queue(&netdev_unregistering_wq, &wait);
11074 static void __net_exit default_device_exit_batch(struct list_head *net_list)
11076 /* At exit all network devices most be removed from a network
11077 * namespace. Do this in the reverse order of registration.
11078 * Do this across as many network namespaces as possible to
11079 * improve batching efficiency.
11081 struct net_device *dev;
11083 LIST_HEAD(dev_kill_list);
11085 /* To prevent network device cleanup code from dereferencing
11086 * loopback devices or network devices that have been freed
11087 * wait here for all pending unregistrations to complete,
11088 * before unregistring the loopback device and allowing the
11089 * network namespace be freed.
11091 * The netdev todo list containing all network devices
11092 * unregistrations that happen in default_device_exit_batch
11093 * will run in the rtnl_unlock() at the end of
11094 * default_device_exit_batch.
11096 rtnl_lock_unregistering(net_list);
11097 list_for_each_entry(net, net_list, exit_list) {
11098 for_each_netdev_reverse(net, dev) {
11099 if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink)
11100 dev->rtnl_link_ops->dellink(dev, &dev_kill_list);
11102 unregister_netdevice_queue(dev, &dev_kill_list);
11105 unregister_netdevice_many(&dev_kill_list);
11109 static struct pernet_operations __net_initdata default_device_ops = {
11110 .exit = default_device_exit,
11111 .exit_batch = default_device_exit_batch,
11115 * Initialize the DEV module. At boot time this walks the device list and
11116 * unhooks any devices that fail to initialise (normally hardware not
11117 * present) and leaves us with a valid list of present and active devices.
11122 * This is called single threaded during boot, so no need
11123 * to take the rtnl semaphore.
11125 static int __init net_dev_init(void)
11127 int i, rc = -ENOMEM;
11129 BUG_ON(!dev_boot_phase);
11131 if (dev_proc_init())
11134 if (netdev_kobject_init())
11137 INIT_LIST_HEAD(&ptype_all);
11138 for (i = 0; i < PTYPE_HASH_SIZE; i++)
11139 INIT_LIST_HEAD(&ptype_base[i]);
11141 INIT_LIST_HEAD(&offload_base);
11143 if (register_pernet_subsys(&netdev_net_ops))
11147 * Initialise the packet receive queues.
11150 for_each_possible_cpu(i) {
11151 struct work_struct *flush = per_cpu_ptr(&flush_works, i);
11152 struct softnet_data *sd = &per_cpu(softnet_data, i);
11154 INIT_WORK(flush, flush_backlog);
11156 skb_queue_head_init(&sd->input_pkt_queue);
11157 skb_queue_head_init(&sd->process_queue);
11158 #ifdef CONFIG_XFRM_OFFLOAD
11159 skb_queue_head_init(&sd->xfrm_backlog);
11161 INIT_LIST_HEAD(&sd->poll_list);
11162 sd->output_queue_tailp = &sd->output_queue;
11164 sd->csd.func = rps_trigger_softirq;
11169 init_gro_hash(&sd->backlog);
11170 sd->backlog.poll = process_backlog;
11171 sd->backlog.weight = weight_p;
11174 dev_boot_phase = 0;
11176 /* The loopback device is special if any other network devices
11177 * is present in a network namespace the loopback device must
11178 * be present. Since we now dynamically allocate and free the
11179 * loopback device ensure this invariant is maintained by
11180 * keeping the loopback device as the first device on the
11181 * list of network devices. Ensuring the loopback devices
11182 * is the first device that appears and the last network device
11185 if (register_pernet_device(&loopback_net_ops))
11188 if (register_pernet_device(&default_device_ops))
11191 open_softirq(NET_TX_SOFTIRQ, net_tx_action);
11192 open_softirq(NET_RX_SOFTIRQ, net_rx_action);
11194 rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead",
11195 NULL, dev_cpu_dead);
11202 subsys_initcall(net_dev_init);