~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/core/dev.c

Version: ~ [ linux-4.18-rc5 ] ~ [ linux-4.17.6 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.55 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.112 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.140 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.115 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.57 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *      NET3    Protocol independent device support routines.
  3  *
  4  *              This program is free software; you can redistribute it and/or
  5  *              modify it under the terms of the GNU General Public License
  6  *              as published by the Free Software Foundation; either version
  7  *              2 of the License, or (at your option) any later version.
  8  *
  9  *      Derived from the non IP parts of dev.c 1.0.19
 10  *              Authors:        Ross Biro
 11  *                              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 12  *                              Mark Evans, <evansmp@uhura.aston.ac.uk>
 13  *
 14  *      Additional Authors:
 15  *              Florian la Roche <rzsfl@rz.uni-sb.de>
 16  *              Alan Cox <gw4pts@gw4pts.ampr.org>
 17  *              David Hinds <dahinds@users.sourceforge.net>
 18  *              Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
 19  *              Adam Sulmicki <adam@cfar.umd.edu>
 20  *              Pekka Riikonen <priikone@poesidon.pspt.fi>
 21  *
 22  *      Changes:
 23  *              D.J. Barrow     :       Fixed bug where dev->refcnt gets set
 24  *                                      to 2 if register_netdev gets called
 25  *                                      before net_dev_init & also removed a
 26  *                                      few lines of code in the process.
 27  *              Alan Cox        :       device private ioctl copies fields back.
 28  *              Alan Cox        :       Transmit queue code does relevant
 29  *                                      stunts to keep the queue safe.
 30  *              Alan Cox        :       Fixed double lock.
 31  *              Alan Cox        :       Fixed promisc NULL pointer trap
 32  *              ????????        :       Support the full private ioctl range
 33  *              Alan Cox        :       Moved ioctl permission check into
 34  *                                      drivers
 35  *              Tim Kordas      :       SIOCADDMULTI/SIOCDELMULTI
 36  *              Alan Cox        :       100 backlog just doesn't cut it when
 37  *                                      you start doing multicast video 8)
 38  *              Alan Cox        :       Rewrote net_bh and list manager.
 39  *              Alan Cox        :       Fix ETH_P_ALL echoback lengths.
 40  *              Alan Cox        :       Took out transmit every packet pass
 41  *                                      Saved a few bytes in the ioctl handler
 42  *              Alan Cox        :       Network driver sets packet type before
 43  *                                      calling netif_rx. Saves a function
 44  *                                      call a packet.
 45  *              Alan Cox        :       Hashed net_bh()
 46  *              Richard Kooijman:       Timestamp fixes.
 47  *              Alan Cox        :       Wrong field in SIOCGIFDSTADDR
 48  *              Alan Cox        :       Device lock protection.
 49  *              Alan Cox        :       Fixed nasty side effect of device close
 50  *                                      changes.
 51  *              Rudi Cilibrasi  :       Pass the right thing to
 52  *                                      set_mac_address()
 53  *              Dave Miller     :       32bit quantity for the device lock to
 54  *                                      make it work out on a Sparc.
 55  *              Bjorn Ekwall    :       Added KERNELD hack.
 56  *              Alan Cox        :       Cleaned up the backlog initialise.
 57  *              Craig Metz      :       SIOCGIFCONF fix if space for under
 58  *                                      1 device.
 59  *          Thomas Bogendoerfer :       Return ENODEV for dev_open, if there
 60  *                                      is no device open function.
 61  *              Andi Kleen      :       Fix error reporting for SIOCGIFCONF
 62  *          Michael Chastain    :       Fix signed/unsigned for SIOCGIFCONF
 63  *              Cyrus Durgin    :       Cleaned for KMOD
 64  *              Adam Sulmicki   :       Bug Fix : Network Device Unload
 65  *                                      A network device unload needs to purge
 66  *                                      the backlog queue.
 67  *      Paul Rusty Russell      :       SIOCSIFNAME
 68  *              Pekka Riikonen  :       Netdev boot-time settings code
 69  *              Andrew Morton   :       Make unregister_netdevice wait
 70  *                                      indefinitely on dev->refcnt
 71  *              J Hadi Salim    :       - Backlog queue sampling
 72  *                                      - netif_rx() feedback
 73  */
 74 
 75 #include <linux/uaccess.h>
 76 #include <linux/bitops.h>
 77 #include <linux/capability.h>
 78 #include <linux/cpu.h>
 79 #include <linux/types.h>
 80 #include <linux/kernel.h>
 81 #include <linux/hash.h>
 82 #include <linux/slab.h>
 83 #include <linux/sched.h>
 84 #include <linux/sched/mm.h>
 85 #include <linux/mutex.h>
 86 #include <linux/string.h>
 87 #include <linux/mm.h>
 88 #include <linux/socket.h>
 89 #include <linux/sockios.h>
 90 #include <linux/errno.h>
 91 #include <linux/interrupt.h>
 92 #include <linux/if_ether.h>
 93 #include <linux/netdevice.h>
 94 #include <linux/etherdevice.h>
 95 #include <linux/ethtool.h>
 96 #include <linux/notifier.h>
 97 #include <linux/skbuff.h>
 98 #include <linux/bpf.h>
 99 #include <linux/bpf_trace.h>
100 #include <net/net_namespace.h>
101 #include <net/sock.h>
102 #include <net/busy_poll.h>
103 #include <linux/rtnetlink.h>
104 #include <linux/stat.h>
105 #include <net/dst.h>
106 #include <net/dst_metadata.h>
107 #include <net/pkt_sched.h>
108 #include <net/pkt_cls.h>
109 #include <net/checksum.h>
110 #include <net/xfrm.h>
111 #include <linux/highmem.h>
112 #include <linux/init.h>
113 #include <linux/module.h>
114 #include <linux/netpoll.h>
115 #include <linux/rcupdate.h>
116 #include <linux/delay.h>
117 #include <net/iw_handler.h>
118 #include <asm/current.h>
119 #include <linux/audit.h>
120 #include <linux/dmaengine.h>
121 #include <linux/err.h>
122 #include <linux/ctype.h>
123 #include <linux/if_arp.h>
124 #include <linux/if_vlan.h>
125 #include <linux/ip.h>
126 #include <net/ip.h>
127 #include <net/mpls.h>
128 #include <linux/ipv6.h>
129 #include <linux/in.h>
130 #include <linux/jhash.h>
131 #include <linux/random.h>
132 #include <trace/events/napi.h>
133 #include <trace/events/net.h>
134 #include <trace/events/skb.h>
135 #include <linux/pci.h>
136 #include <linux/inetdevice.h>
137 #include <linux/cpu_rmap.h>
138 #include <linux/static_key.h>
139 #include <linux/hashtable.h>
140 #include <linux/vmalloc.h>
141 #include <linux/if_macvlan.h>
142 #include <linux/errqueue.h>
143 #include <linux/hrtimer.h>
144 #include <linux/netfilter_ingress.h>
145 #include <linux/crash_dump.h>
146 #include <linux/sctp.h>
147 #include <net/udp_tunnel.h>
148 #include <linux/net_namespace.h>
149 
150 #include "net-sysfs.h"
151 
152 /* Instead of increasing this, you should create a hash table. */
153 #define MAX_GRO_SKBS 8
154 
155 /* This should be increased if a protocol with a bigger head is added. */
156 #define GRO_MAX_HEAD (MAX_HEADER + 128)
157 
158 static DEFINE_SPINLOCK(ptype_lock);
159 static DEFINE_SPINLOCK(offload_lock);
160 struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly;
161 struct list_head ptype_all __read_mostly;       /* Taps */
162 static struct list_head offload_base __read_mostly;
163 
164 static int netif_rx_internal(struct sk_buff *skb);
165 static int call_netdevice_notifiers_info(unsigned long val,
166                                          struct netdev_notifier_info *info);
167 static struct napi_struct *napi_by_id(unsigned int napi_id);
168 
169 /*
170  * The @dev_base_head list is protected by @dev_base_lock and the rtnl
171  * semaphore.
172  *
173  * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
174  *
175  * Writers must hold the rtnl semaphore while they loop through the
176  * dev_base_head list, and hold dev_base_lock for writing when they do the
177  * actual updates.  This allows pure readers to access the list even
178  * while a writer is preparing to update it.
179  *
180  * To put it another way, dev_base_lock is held for writing only to
181  * protect against pure readers; the rtnl semaphore provides the
182  * protection against other writers.
183  *
184  * See, for example usages, register_netdevice() and
185  * unregister_netdevice(), which must be called with the rtnl
186  * semaphore held.
187  */
188 DEFINE_RWLOCK(dev_base_lock);
189 EXPORT_SYMBOL(dev_base_lock);
190 
191 static DEFINE_MUTEX(ifalias_mutex);
192 
193 /* protects napi_hash addition/deletion and napi_gen_id */
194 static DEFINE_SPINLOCK(napi_hash_lock);
195 
196 static unsigned int napi_gen_id = NR_CPUS;
197 static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8);
198 
199 static seqcount_t devnet_rename_seq;
200 
201 static inline void dev_base_seq_inc(struct net *net)
202 {
203         while (++net->dev_base_seq == 0)
204                 ;
205 }
206 
207 static inline struct hlist_head *dev_name_hash(struct net *net, const char *name)
208 {
209         unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ));
210 
211         return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)];
212 }
213 
214 static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex)
215 {
216         return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)];
217 }
218 
219 static inline void rps_lock(struct softnet_data *sd)
220 {
221 #ifdef CONFIG_RPS
222         spin_lock(&sd->input_pkt_queue.lock);
223 #endif
224 }
225 
226 static inline void rps_unlock(struct softnet_data *sd)
227 {
228 #ifdef CONFIG_RPS
229         spin_unlock(&sd->input_pkt_queue.lock);
230 #endif
231 }
232 
233 /* Device list insertion */
234 static void list_netdevice(struct net_device *dev)
235 {
236         struct net *net = dev_net(dev);
237 
238         ASSERT_RTNL();
239 
240         write_lock_bh(&dev_base_lock);
241         list_add_tail_rcu(&dev->dev_list, &net->dev_base_head);
242         hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
243         hlist_add_head_rcu(&dev->index_hlist,
244                            dev_index_hash(net, dev->ifindex));
245         write_unlock_bh(&dev_base_lock);
246 
247         dev_base_seq_inc(net);
248 }
249 
250 /* Device list removal
251  * caller must respect a RCU grace period before freeing/reusing dev
252  */
253 static void unlist_netdevice(struct net_device *dev)
254 {
255         ASSERT_RTNL();
256 
257         /* Unlink dev from the device chain */
258         write_lock_bh(&dev_base_lock);
259         list_del_rcu(&dev->dev_list);
260         hlist_del_rcu(&dev->name_hlist);
261         hlist_del_rcu(&dev->index_hlist);
262         write_unlock_bh(&dev_base_lock);
263 
264         dev_base_seq_inc(dev_net(dev));
265 }
266 
267 /*
268  *      Our notifier list
269  */
270 
271 static RAW_NOTIFIER_HEAD(netdev_chain);
272 
273 /*
274  *      Device drivers call our routines to queue packets here. We empty the
275  *      queue in the local softnet handler.
276  */
277 
278 DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
279 EXPORT_PER_CPU_SYMBOL(softnet_data);
280 
281 #ifdef CONFIG_LOCKDEP
282 /*
283  * register_netdevice() inits txq->_xmit_lock and sets lockdep class
284  * according to dev->type
285  */
286 static const unsigned short netdev_lock_type[] = {
287          ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25,
288          ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET,
289          ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM,
290          ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP,
291          ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD,
292          ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25,
293          ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP,
294          ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD,
295          ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI,
296          ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE,
297          ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET,
298          ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL,
299          ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM,
300          ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE,
301          ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE};
302 
303 static const char *const netdev_lock_name[] = {
304         "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
305         "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
306         "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
307         "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
308         "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
309         "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
310         "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
311         "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
312         "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
313         "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
314         "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
315         "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
316         "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
317         "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
318         "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
319 
320 static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
321 static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)];
322 
323 static inline unsigned short netdev_lock_pos(unsigned short dev_type)
324 {
325         int i;
326 
327         for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++)
328                 if (netdev_lock_type[i] == dev_type)
329                         return i;
330         /* the last key is used by default */
331         return ARRAY_SIZE(netdev_lock_type) - 1;
332 }
333 
334 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
335                                                  unsigned short dev_type)
336 {
337         int i;
338 
339         i = netdev_lock_pos(dev_type);
340         lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i],
341                                    netdev_lock_name[i]);
342 }
343 
344 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
345 {
346         int i;
347 
348         i = netdev_lock_pos(dev->type);
349         lockdep_set_class_and_name(&dev->addr_list_lock,
350                                    &netdev_addr_lock_key[i],
351                                    netdev_lock_name[i]);
352 }
353 #else
354 static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock,
355                                                  unsigned short dev_type)
356 {
357 }
358 static inline void netdev_set_addr_lockdep_class(struct net_device *dev)
359 {
360 }
361 #endif
362 
363 /*******************************************************************************
364  *
365  *              Protocol management and registration routines
366  *
367  *******************************************************************************/
368 
369 
370 /*
371  *      Add a protocol ID to the list. Now that the input handler is
372  *      smarter we can dispense with all the messy stuff that used to be
373  *      here.
374  *
375  *      BEWARE!!! Protocol handlers, mangling input packets,
376  *      MUST BE last in hash buckets and checking protocol handlers
377  *      MUST start from promiscuous ptype_all chain in net_bh.
378  *      It is true now, do not change it.
379  *      Explanation follows: if protocol handler, mangling packet, will
380  *      be the first on list, it is not able to sense, that packet
381  *      is cloned and should be copied-on-write, so that it will
382  *      change it and subsequent readers will get broken packet.
383  *                                                      --ANK (980803)
384  */
385 
386 static inline struct list_head *ptype_head(const struct packet_type *pt)
387 {
388         if (pt->type == htons(ETH_P_ALL))
389                 return pt->dev ? &pt->dev->ptype_all : &ptype_all;
390         else
391                 return pt->dev ? &pt->dev->ptype_specific :
392                                  &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK];
393 }
394 
395 /**
396  *      dev_add_pack - add packet handler
397  *      @pt: packet type declaration
398  *
399  *      Add a protocol handler to the networking stack. The passed &packet_type
400  *      is linked into kernel lists and may not be freed until it has been
401  *      removed from the kernel lists.
402  *
403  *      This call does not sleep therefore it can not
404  *      guarantee all CPU's that are in middle of receiving packets
405  *      will see the new packet type (until the next received packet).
406  */
407 
408 void dev_add_pack(struct packet_type *pt)
409 {
410         struct list_head *head = ptype_head(pt);
411 
412         spin_lock(&ptype_lock);
413         list_add_rcu(&pt->list, head);
414         spin_unlock(&ptype_lock);
415 }
416 EXPORT_SYMBOL(dev_add_pack);
417 
418 /**
419  *      __dev_remove_pack        - remove packet handler
420  *      @pt: packet type declaration
421  *
422  *      Remove a protocol handler that was previously added to the kernel
423  *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
424  *      from the kernel lists and can be freed or reused once this function
425  *      returns.
426  *
427  *      The packet type might still be in use by receivers
428  *      and must not be freed until after all the CPU's have gone
429  *      through a quiescent state.
430  */
431 void __dev_remove_pack(struct packet_type *pt)
432 {
433         struct list_head *head = ptype_head(pt);
434         struct packet_type *pt1;
435 
436         spin_lock(&ptype_lock);
437 
438         list_for_each_entry(pt1, head, list) {
439                 if (pt == pt1) {
440                         list_del_rcu(&pt->list);
441                         goto out;
442                 }
443         }
444 
445         pr_warn("dev_remove_pack: %p not found\n", pt);
446 out:
447         spin_unlock(&ptype_lock);
448 }
449 EXPORT_SYMBOL(__dev_remove_pack);
450 
451 /**
452  *      dev_remove_pack  - remove packet handler
453  *      @pt: packet type declaration
454  *
455  *      Remove a protocol handler that was previously added to the kernel
456  *      protocol handlers by dev_add_pack(). The passed &packet_type is removed
457  *      from the kernel lists and can be freed or reused once this function
458  *      returns.
459  *
460  *      This call sleeps to guarantee that no CPU is looking at the packet
461  *      type after return.
462  */
463 void dev_remove_pack(struct packet_type *pt)
464 {
465         __dev_remove_pack(pt);
466 
467         synchronize_net();
468 }
469 EXPORT_SYMBOL(dev_remove_pack);
470 
471 
472 /**
473  *      dev_add_offload - register offload handlers
474  *      @po: protocol offload declaration
475  *
476  *      Add protocol offload handlers to the networking stack. The passed
477  *      &proto_offload is linked into kernel lists and may not be freed until
478  *      it has been removed from the kernel lists.
479  *
480  *      This call does not sleep therefore it can not
481  *      guarantee all CPU's that are in middle of receiving packets
482  *      will see the new offload handlers (until the next received packet).
483  */
484 void dev_add_offload(struct packet_offload *po)
485 {
486         struct packet_offload *elem;
487 
488         spin_lock(&offload_lock);
489         list_for_each_entry(elem, &offload_base, list) {
490                 if (po->priority < elem->priority)
491                         break;
492         }
493         list_add_rcu(&po->list, elem->list.prev);
494         spin_unlock(&offload_lock);
495 }
496 EXPORT_SYMBOL(dev_add_offload);
497 
498 /**
499  *      __dev_remove_offload     - remove offload handler
500  *      @po: packet offload declaration
501  *
502  *      Remove a protocol offload handler that was previously added to the
503  *      kernel offload handlers by dev_add_offload(). The passed &offload_type
504  *      is removed from the kernel lists and can be freed or reused once this
505  *      function returns.
506  *
507  *      The packet type might still be in use by receivers
508  *      and must not be freed until after all the CPU's have gone
509  *      through a quiescent state.
510  */
511 static void __dev_remove_offload(struct packet_offload *po)
512 {
513         struct list_head *head = &offload_base;
514         struct packet_offload *po1;
515 
516         spin_lock(&offload_lock);
517 
518         list_for_each_entry(po1, head, list) {
519                 if (po == po1) {
520                         list_del_rcu(&po->list);
521                         goto out;
522                 }
523         }
524 
525         pr_warn("dev_remove_offload: %p not found\n", po);
526 out:
527         spin_unlock(&offload_lock);
528 }
529 
530 /**
531  *      dev_remove_offload       - remove packet offload handler
532  *      @po: packet offload declaration
533  *
534  *      Remove a packet offload handler that was previously added to the kernel
535  *      offload handlers by dev_add_offload(). The passed &offload_type is
536  *      removed from the kernel lists and can be freed or reused once this
537  *      function returns.
538  *
539  *      This call sleeps to guarantee that no CPU is looking at the packet
540  *      type after return.
541  */
542 void dev_remove_offload(struct packet_offload *po)
543 {
544         __dev_remove_offload(po);
545 
546         synchronize_net();
547 }
548 EXPORT_SYMBOL(dev_remove_offload);
549 
550 /******************************************************************************
551  *
552  *                    Device Boot-time Settings Routines
553  *
554  ******************************************************************************/
555 
556 /* Boot time configuration table */
557 static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX];
558 
559 /**
560  *      netdev_boot_setup_add   - add new setup entry
561  *      @name: name of the device
562  *      @map: configured settings for the device
563  *
564  *      Adds new setup entry to the dev_boot_setup list.  The function
565  *      returns 0 on error and 1 on success.  This is a generic routine to
566  *      all netdevices.
567  */
568 static int netdev_boot_setup_add(char *name, struct ifmap *map)
569 {
570         struct netdev_boot_setup *s;
571         int i;
572 
573         s = dev_boot_setup;
574         for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
575                 if (s[i].name[0] == '\0' || s[i].name[0] == ' ') {
576                         memset(s[i].name, 0, sizeof(s[i].name));
577                         strlcpy(s[i].name, name, IFNAMSIZ);
578                         memcpy(&s[i].map, map, sizeof(s[i].map));
579                         break;
580                 }
581         }
582 
583         return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1;
584 }
585 
586 /**
587  * netdev_boot_setup_check      - check boot time settings
588  * @dev: the netdevice
589  *
590  * Check boot time settings for the device.
591  * The found settings are set for the device to be used
592  * later in the device probing.
593  * Returns 0 if no settings found, 1 if they are.
594  */
595 int netdev_boot_setup_check(struct net_device *dev)
596 {
597         struct netdev_boot_setup *s = dev_boot_setup;
598         int i;
599 
600         for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) {
601                 if (s[i].name[0] != '\0' && s[i].name[0] != ' ' &&
602                     !strcmp(dev->name, s[i].name)) {
603                         dev->irq = s[i].map.irq;
604                         dev->base_addr = s[i].map.base_addr;
605                         dev->mem_start = s[i].map.mem_start;
606                         dev->mem_end = s[i].map.mem_end;
607                         return 1;
608                 }
609         }
610         return 0;
611 }
612 EXPORT_SYMBOL(netdev_boot_setup_check);
613 
614 
615 /**
616  * netdev_boot_base     - get address from boot time settings
617  * @prefix: prefix for network device
618  * @unit: id for network device
619  *
620  * Check boot time settings for the base address of device.
621  * The found settings are set for the device to be used
622  * later in the device probing.
623  * Returns 0 if no settings found.
624  */
625 unsigned long netdev_boot_base(const char *prefix, int unit)
626 {
627         const struct netdev_boot_setup *s = dev_boot_setup;
628         char name[IFNAMSIZ];
629         int i;
630 
631         sprintf(name, "%s%d", prefix, unit);
632 
633         /*
634          * If device already registered then return base of 1
635          * to indicate not to probe for this interface
636          */
637         if (__dev_get_by_name(&init_net, name))
638                 return 1;
639 
640         for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++)
641                 if (!strcmp(name, s[i].name))
642                         return s[i].map.base_addr;
643         return 0;
644 }
645 
646 /*
647  * Saves at boot time configured settings for any netdevice.
648  */
649 int __init netdev_boot_setup(char *str)
650 {
651         int ints[5];
652         struct ifmap map;
653 
654         str = get_options(str, ARRAY_SIZE(ints), ints);
655         if (!str || !*str)
656                 return 0;
657 
658         /* Save settings */
659         memset(&map, 0, sizeof(map));
660         if (ints[0] > 0)
661                 map.irq = ints[1];
662         if (ints[0] > 1)
663                 map.base_addr = ints[2];
664         if (ints[0] > 2)
665                 map.mem_start = ints[3];
666         if (ints[0] > 3)
667                 map.mem_end = ints[4];
668 
669         /* Add new entry to the list */
670         return netdev_boot_setup_add(str, &map);
671 }
672 
673 __setup("netdev=", netdev_boot_setup);
674 
675 /*******************************************************************************
676  *
677  *                          Device Interface Subroutines
678  *
679  *******************************************************************************/
680 
681 /**
682  *      dev_get_iflink  - get 'iflink' value of a interface
683  *      @dev: targeted interface
684  *
685  *      Indicates the ifindex the interface is linked to.
686  *      Physical interfaces have the same 'ifindex' and 'iflink' values.
687  */
688 
689 int dev_get_iflink(const struct net_device *dev)
690 {
691         if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink)
692                 return dev->netdev_ops->ndo_get_iflink(dev);
693 
694         return dev->ifindex;
695 }
696 EXPORT_SYMBOL(dev_get_iflink);
697 
698 /**
699  *      dev_fill_metadata_dst - Retrieve tunnel egress information.
700  *      @dev: targeted interface
701  *      @skb: The packet.
702  *
703  *      For better visibility of tunnel traffic OVS needs to retrieve
704  *      egress tunnel information for a packet. Following API allows
705  *      user to get this info.
706  */
707 int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb)
708 {
709         struct ip_tunnel_info *info;
710 
711         if (!dev->netdev_ops  || !dev->netdev_ops->ndo_fill_metadata_dst)
712                 return -EINVAL;
713 
714         info = skb_tunnel_info_unclone(skb);
715         if (!info)
716                 return -ENOMEM;
717         if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX)))
718                 return -EINVAL;
719 
720         return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb);
721 }
722 EXPORT_SYMBOL_GPL(dev_fill_metadata_dst);
723 
724 /**
725  *      __dev_get_by_name       - find a device by its name
726  *      @net: the applicable net namespace
727  *      @name: name to find
728  *
729  *      Find an interface by name. Must be called under RTNL semaphore
730  *      or @dev_base_lock. If the name is found a pointer to the device
731  *      is returned. If the name is not found then %NULL is returned. The
732  *      reference counters are not incremented so the caller must be
733  *      careful with locks.
734  */
735 
736 struct net_device *__dev_get_by_name(struct net *net, const char *name)
737 {
738         struct net_device *dev;
739         struct hlist_head *head = dev_name_hash(net, name);
740 
741         hlist_for_each_entry(dev, head, name_hlist)
742                 if (!strncmp(dev->name, name, IFNAMSIZ))
743                         return dev;
744 
745         return NULL;
746 }
747 EXPORT_SYMBOL(__dev_get_by_name);
748 
749 /**
750  * dev_get_by_name_rcu  - find a device by its name
751  * @net: the applicable net namespace
752  * @name: name to find
753  *
754  * Find an interface by name.
755  * If the name is found a pointer to the device is returned.
756  * If the name is not found then %NULL is returned.
757  * The reference counters are not incremented so the caller must be
758  * careful with locks. The caller must hold RCU lock.
759  */
760 
761 struct net_device *dev_get_by_name_rcu(struct net *net, const char *name)
762 {
763         struct net_device *dev;
764         struct hlist_head *head = dev_name_hash(net, name);
765 
766         hlist_for_each_entry_rcu(dev, head, name_hlist)
767                 if (!strncmp(dev->name, name, IFNAMSIZ))
768                         return dev;
769 
770         return NULL;
771 }
772 EXPORT_SYMBOL(dev_get_by_name_rcu);
773 
774 /**
775  *      dev_get_by_name         - find a device by its name
776  *      @net: the applicable net namespace
777  *      @name: name to find
778  *
779  *      Find an interface by name. This can be called from any
780  *      context and does its own locking. The returned handle has
781  *      the usage count incremented and the caller must use dev_put() to
782  *      release it when it is no longer needed. %NULL is returned if no
783  *      matching device is found.
784  */
785 
786 struct net_device *dev_get_by_name(struct net *net, const char *name)
787 {
788         struct net_device *dev;
789 
790         rcu_read_lock();
791         dev = dev_get_by_name_rcu(net, name);
792         if (dev)
793                 dev_hold(dev);
794         rcu_read_unlock();
795         return dev;
796 }
797 EXPORT_SYMBOL(dev_get_by_name);
798 
799 /**
800  *      __dev_get_by_index - find a device by its ifindex
801  *      @net: the applicable net namespace
802  *      @ifindex: index of device
803  *
804  *      Search for an interface by index. Returns %NULL if the device
805  *      is not found or a pointer to the device. The device has not
806  *      had its reference counter increased so the caller must be careful
807  *      about locking. The caller must hold either the RTNL semaphore
808  *      or @dev_base_lock.
809  */
810 
811 struct net_device *__dev_get_by_index(struct net *net, int ifindex)
812 {
813         struct net_device *dev;
814         struct hlist_head *head = dev_index_hash(net, ifindex);
815 
816         hlist_for_each_entry(dev, head, index_hlist)
817                 if (dev->ifindex == ifindex)
818                         return dev;
819 
820         return NULL;
821 }
822 EXPORT_SYMBOL(__dev_get_by_index);
823 
824 /**
825  *      dev_get_by_index_rcu - find a device by its ifindex
826  *      @net: the applicable net namespace
827  *      @ifindex: index of device
828  *
829  *      Search for an interface by index. Returns %NULL if the device
830  *      is not found or a pointer to the device. The device has not
831  *      had its reference counter increased so the caller must be careful
832  *      about locking. The caller must hold RCU lock.
833  */
834 
835 struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex)
836 {
837         struct net_device *dev;
838         struct hlist_head *head = dev_index_hash(net, ifindex);
839 
840         hlist_for_each_entry_rcu(dev, head, index_hlist)
841                 if (dev->ifindex == ifindex)
842                         return dev;
843 
844         return NULL;
845 }
846 EXPORT_SYMBOL(dev_get_by_index_rcu);
847 
848 
849 /**
850  *      dev_get_by_index - find a device by its ifindex
851  *      @net: the applicable net namespace
852  *      @ifindex: index of device
853  *
854  *      Search for an interface by index. Returns NULL if the device
855  *      is not found or a pointer to the device. The device returned has
856  *      had a reference added and the pointer is safe until the user calls
857  *      dev_put to indicate they have finished with it.
858  */
859 
860 struct net_device *dev_get_by_index(struct net *net, int ifindex)
861 {
862         struct net_device *dev;
863 
864         rcu_read_lock();
865         dev = dev_get_by_index_rcu(net, ifindex);
866         if (dev)
867                 dev_hold(dev);
868         rcu_read_unlock();
869         return dev;
870 }
871 EXPORT_SYMBOL(dev_get_by_index);
872 
873 /**
874  *      dev_get_by_napi_id - find a device by napi_id
875  *      @napi_id: ID of the NAPI struct
876  *
877  *      Search for an interface by NAPI ID. Returns %NULL if the device
878  *      is not found or a pointer to the device. The device has not had
879  *      its reference counter increased so the caller must be careful
880  *      about locking. The caller must hold RCU lock.
881  */
882 
883 struct net_device *dev_get_by_napi_id(unsigned int napi_id)
884 {
885         struct napi_struct *napi;
886 
887         WARN_ON_ONCE(!rcu_read_lock_held());
888 
889         if (napi_id < MIN_NAPI_ID)
890                 return NULL;
891 
892         napi = napi_by_id(napi_id);
893 
894         return napi ? napi->dev : NULL;
895 }
896 EXPORT_SYMBOL(dev_get_by_napi_id);
897 
898 /**
899  *      netdev_get_name - get a netdevice name, knowing its ifindex.
900  *      @net: network namespace
901  *      @name: a pointer to the buffer where the name will be stored.
902  *      @ifindex: the ifindex of the interface to get the name from.
903  *
904  *      The use of raw_seqcount_begin() and cond_resched() before
905  *      retrying is required as we want to give the writers a chance
906  *      to complete when CONFIG_PREEMPT is not set.
907  */
908 int netdev_get_name(struct net *net, char *name, int ifindex)
909 {
910         struct net_device *dev;
911         unsigned int seq;
912 
913 retry:
914         seq = raw_seqcount_begin(&devnet_rename_seq);
915         rcu_read_lock();
916         dev = dev_get_by_index_rcu(net, ifindex);
917         if (!dev) {
918                 rcu_read_unlock();
919                 return -ENODEV;
920         }
921 
922         strcpy(name, dev->name);
923         rcu_read_unlock();
924         if (read_seqcount_retry(&devnet_rename_seq, seq)) {
925                 cond_resched();
926                 goto retry;
927         }
928 
929         return 0;
930 }
931 
932 /**
933  *      dev_getbyhwaddr_rcu - find a device by its hardware address
934  *      @net: the applicable net namespace
935  *      @type: media type of device
936  *      @ha: hardware address
937  *
938  *      Search for an interface by MAC address. Returns NULL if the device
939  *      is not found or a pointer to the device.
940  *      The caller must hold RCU or RTNL.
941  *      The returned device has not had its ref count increased
942  *      and the caller must therefore be careful about locking
943  *
944  */
945 
946 struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type,
947                                        const char *ha)
948 {
949         struct net_device *dev;
950 
951         for_each_netdev_rcu(net, dev)
952                 if (dev->type == type &&
953                     !memcmp(dev->dev_addr, ha, dev->addr_len))
954                         return dev;
955 
956         return NULL;
957 }
958 EXPORT_SYMBOL(dev_getbyhwaddr_rcu);
959 
960 struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type)
961 {
962         struct net_device *dev;
963 
964         ASSERT_RTNL();
965         for_each_netdev(net, dev)
966                 if (dev->type == type)
967                         return dev;
968 
969         return NULL;
970 }
971 EXPORT_SYMBOL(__dev_getfirstbyhwtype);
972 
973 struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type)
974 {
975         struct net_device *dev, *ret = NULL;
976 
977         rcu_read_lock();
978         for_each_netdev_rcu(net, dev)
979                 if (dev->type == type) {
980                         dev_hold(dev);
981                         ret = dev;
982                         break;
983                 }
984         rcu_read_unlock();
985         return ret;
986 }
987 EXPORT_SYMBOL(dev_getfirstbyhwtype);
988 
989 /**
990  *      __dev_get_by_flags - find any device with given flags
991  *      @net: the applicable net namespace
992  *      @if_flags: IFF_* values
993  *      @mask: bitmask of bits in if_flags to check
994  *
995  *      Search for any interface with the given flags. Returns NULL if a device
996  *      is not found or a pointer to the device. Must be called inside
997  *      rtnl_lock(), and result refcount is unchanged.
998  */
999 
1000 struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags,
1001                                       unsigned short mask)
1002 {
1003         struct net_device *dev, *ret;
1004 
1005         ASSERT_RTNL();
1006 
1007         ret = NULL;
1008         for_each_netdev(net, dev) {
1009                 if (((dev->flags ^ if_flags) & mask) == 0) {
1010                         ret = dev;
1011                         break;
1012                 }
1013         }
1014         return ret;
1015 }
1016 EXPORT_SYMBOL(__dev_get_by_flags);
1017 
1018 /**
1019  *      dev_valid_name - check if name is okay for network device
1020  *      @name: name string
1021  *
1022  *      Network device names need to be valid file names to
1023  *      to allow sysfs to work.  We also disallow any kind of
1024  *      whitespace.
1025  */
1026 bool dev_valid_name(const char *name)
1027 {
1028         if (*name == '\0')
1029                 return false;
1030         if (strnlen(name, IFNAMSIZ) == IFNAMSIZ)
1031                 return false;
1032         if (!strcmp(name, ".") || !strcmp(name, ".."))
1033                 return false;
1034 
1035         while (*name) {
1036                 if (*name == '/' || *name == ':' || isspace(*name))
1037                         return false;
1038                 name++;
1039         }
1040         return true;
1041 }
1042 EXPORT_SYMBOL(dev_valid_name);
1043 
1044 /**
1045  *      __dev_alloc_name - allocate a name for a device
1046  *      @net: network namespace to allocate the device name in
1047  *      @name: name format string
1048  *      @buf:  scratch buffer and result name string
1049  *
1050  *      Passed a format string - eg "lt%d" it will try and find a suitable
1051  *      id. It scans list of devices to build up a free map, then chooses
1052  *      the first empty slot. The caller must hold the dev_base or rtnl lock
1053  *      while allocating the name and adding the device in order to avoid
1054  *      duplicates.
1055  *      Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1056  *      Returns the number of the unit assigned or a negative errno code.
1057  */
1058 
1059 static int __dev_alloc_name(struct net *net, const char *name, char *buf)
1060 {
1061         int i = 0;
1062         const char *p;
1063         const int max_netdevices = 8*PAGE_SIZE;
1064         unsigned long *inuse;
1065         struct net_device *d;
1066 
1067         if (!dev_valid_name(name))
1068                 return -EINVAL;
1069 
1070         p = strchr(name, '%');
1071         if (p) {
1072                 /*
1073                  * Verify the string as this thing may have come from
1074                  * the user.  There must be either one "%d" and no other "%"
1075                  * characters.
1076                  */
1077                 if (p[1] != 'd' || strchr(p + 2, '%'))
1078                         return -EINVAL;
1079 
1080                 /* Use one page as a bit array of possible slots */
1081                 inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC);
1082                 if (!inuse)
1083                         return -ENOMEM;
1084 
1085                 for_each_netdev(net, d) {
1086                         if (!sscanf(d->name, name, &i))
1087                                 continue;
1088                         if (i < 0 || i >= max_netdevices)
1089                                 continue;
1090 
1091                         /*  avoid cases where sscanf is not exact inverse of printf */
1092                         snprintf(buf, IFNAMSIZ, name, i);
1093                         if (!strncmp(buf, d->name, IFNAMSIZ))
1094                                 set_bit(i, inuse);
1095                 }
1096 
1097                 i = find_first_zero_bit(inuse, max_netdevices);
1098                 free_page((unsigned long) inuse);
1099         }
1100 
1101         snprintf(buf, IFNAMSIZ, name, i);
1102         if (!__dev_get_by_name(net, buf))
1103                 return i;
1104 
1105         /* It is possible to run out of possible slots
1106          * when the name is long and there isn't enough space left
1107          * for the digits, or if all bits are used.
1108          */
1109         return -ENFILE;
1110 }
1111 
1112 static int dev_alloc_name_ns(struct net *net,
1113                              struct net_device *dev,
1114                              const char *name)
1115 {
1116         char buf[IFNAMSIZ];
1117         int ret;
1118 
1119         BUG_ON(!net);
1120         ret = __dev_alloc_name(net, name, buf);
1121         if (ret >= 0)
1122                 strlcpy(dev->name, buf, IFNAMSIZ);
1123         return ret;
1124 }
1125 
1126 /**
1127  *      dev_alloc_name - allocate a name for a device
1128  *      @dev: device
1129  *      @name: name format string
1130  *
1131  *      Passed a format string - eg "lt%d" it will try and find a suitable
1132  *      id. It scans list of devices to build up a free map, then chooses
1133  *      the first empty slot. The caller must hold the dev_base or rtnl lock
1134  *      while allocating the name and adding the device in order to avoid
1135  *      duplicates.
1136  *      Limited to bits_per_byte * page size devices (ie 32K on most platforms).
1137  *      Returns the number of the unit assigned or a negative errno code.
1138  */
1139 
1140 int dev_alloc_name(struct net_device *dev, const char *name)
1141 {
1142         return dev_alloc_name_ns(dev_net(dev), dev, name);
1143 }
1144 EXPORT_SYMBOL(dev_alloc_name);
1145 
1146 int dev_get_valid_name(struct net *net, struct net_device *dev,
1147                        const char *name)
1148 {
1149         BUG_ON(!net);
1150 
1151         if (!dev_valid_name(name))
1152                 return -EINVAL;
1153 
1154         if (strchr(name, '%'))
1155                 return dev_alloc_name_ns(net, dev, name);
1156         else if (__dev_get_by_name(net, name))
1157                 return -EEXIST;
1158         else if (dev->name != name)
1159                 strlcpy(dev->name, name, IFNAMSIZ);
1160 
1161         return 0;
1162 }
1163 EXPORT_SYMBOL(dev_get_valid_name);
1164 
1165 /**
1166  *      dev_change_name - change name of a device
1167  *      @dev: device
1168  *      @newname: name (or format string) must be at least IFNAMSIZ
1169  *
1170  *      Change name of a device, can pass format strings "eth%d".
1171  *      for wildcarding.
1172  */
1173 int dev_change_name(struct net_device *dev, const char *newname)
1174 {
1175         unsigned char old_assign_type;
1176         char oldname[IFNAMSIZ];
1177         int err = 0;
1178         int ret;
1179         struct net *net;
1180 
1181         ASSERT_RTNL();
1182         BUG_ON(!dev_net(dev));
1183 
1184         net = dev_net(dev);
1185         if (dev->flags & IFF_UP)
1186                 return -EBUSY;
1187 
1188         write_seqcount_begin(&devnet_rename_seq);
1189 
1190         if (strncmp(newname, dev->name, IFNAMSIZ) == 0) {
1191                 write_seqcount_end(&devnet_rename_seq);
1192                 return 0;
1193         }
1194 
1195         memcpy(oldname, dev->name, IFNAMSIZ);
1196 
1197         err = dev_get_valid_name(net, dev, newname);
1198         if (err < 0) {
1199                 write_seqcount_end(&devnet_rename_seq);
1200                 return err;
1201         }
1202 
1203         if (oldname[0] && !strchr(oldname, '%'))
1204                 netdev_info(dev, "renamed from %s\n", oldname);
1205 
1206         old_assign_type = dev->name_assign_type;
1207         dev->name_assign_type = NET_NAME_RENAMED;
1208 
1209 rollback:
1210         ret = device_rename(&dev->dev, dev->name);
1211         if (ret) {
1212                 memcpy(dev->name, oldname, IFNAMSIZ);
1213                 dev->name_assign_type = old_assign_type;
1214                 write_seqcount_end(&devnet_rename_seq);
1215                 return ret;
1216         }
1217 
1218         write_seqcount_end(&devnet_rename_seq);
1219 
1220         netdev_adjacent_rename_links(dev, oldname);
1221 
1222         write_lock_bh(&dev_base_lock);
1223         hlist_del_rcu(&dev->name_hlist);
1224         write_unlock_bh(&dev_base_lock);
1225 
1226         synchronize_rcu();
1227 
1228         write_lock_bh(&dev_base_lock);
1229         hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name));
1230         write_unlock_bh(&dev_base_lock);
1231 
1232         ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev);
1233         ret = notifier_to_errno(ret);
1234 
1235         if (ret) {
1236                 /* err >= 0 after dev_alloc_name() or stores the first errno */
1237                 if (err >= 0) {
1238                         err = ret;
1239                         write_seqcount_begin(&devnet_rename_seq);
1240                         memcpy(dev->name, oldname, IFNAMSIZ);
1241                         memcpy(oldname, newname, IFNAMSIZ);
1242                         dev->name_assign_type = old_assign_type;
1243                         old_assign_type = NET_NAME_RENAMED;
1244                         goto rollback;
1245                 } else {
1246                         pr_err("%s: name change rollback failed: %d\n",
1247                                dev->name, ret);
1248                 }
1249         }
1250 
1251         return err;
1252 }
1253 
1254 /**
1255  *      dev_set_alias - change ifalias of a device
1256  *      @dev: device
1257  *      @alias: name up to IFALIASZ
1258  *      @len: limit of bytes to copy from info
1259  *
1260  *      Set ifalias for a device,
1261  */
1262 int dev_set_alias(struct net_device *dev, const char *alias, size_t len)
1263 {
1264         struct dev_ifalias *new_alias = NULL;
1265 
1266         if (len >= IFALIASZ)
1267                 return -EINVAL;
1268 
1269         if (len) {
1270                 new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL);
1271                 if (!new_alias)
1272                         return -ENOMEM;
1273 
1274                 memcpy(new_alias->ifalias, alias, len);
1275                 new_alias->ifalias[len] = 0;
1276         }
1277 
1278         mutex_lock(&ifalias_mutex);
1279         rcu_swap_protected(dev->ifalias, new_alias,
1280                            mutex_is_locked(&ifalias_mutex));
1281         mutex_unlock(&ifalias_mutex);
1282 
1283         if (new_alias)
1284                 kfree_rcu(new_alias, rcuhead);
1285 
1286         return len;
1287 }
1288 EXPORT_SYMBOL(dev_set_alias);
1289 
1290 /**
1291  *      dev_get_alias - get ifalias of a device
1292  *      @dev: device
1293  *      @name: buffer to store name of ifalias
1294  *      @len: size of buffer
1295  *
1296  *      get ifalias for a device.  Caller must make sure dev cannot go
1297  *      away,  e.g. rcu read lock or own a reference count to device.
1298  */
1299 int dev_get_alias(const struct net_device *dev, char *name, size_t len)
1300 {
1301         const struct dev_ifalias *alias;
1302         int ret = 0;
1303 
1304         rcu_read_lock();
1305         alias = rcu_dereference(dev->ifalias);
1306         if (alias)
1307                 ret = snprintf(name, len, "%s", alias->ifalias);
1308         rcu_read_unlock();
1309 
1310         return ret;
1311 }
1312 
1313 /**
1314  *      netdev_features_change - device changes features
1315  *      @dev: device to cause notification
1316  *
1317  *      Called to indicate a device has changed features.
1318  */
1319 void netdev_features_change(struct net_device *dev)
1320 {
1321         call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev);
1322 }
1323 EXPORT_SYMBOL(netdev_features_change);
1324 
1325 /**
1326  *      netdev_state_change - device changes state
1327  *      @dev: device to cause notification
1328  *
1329  *      Called to indicate a device has changed state. This function calls
1330  *      the notifier chains for netdev_chain and sends a NEWLINK message
1331  *      to the routing socket.
1332  */
1333 void netdev_state_change(struct net_device *dev)
1334 {
1335         if (dev->flags & IFF_UP) {
1336                 struct netdev_notifier_change_info change_info = {
1337                         .info.dev = dev,
1338                 };
1339 
1340                 call_netdevice_notifiers_info(NETDEV_CHANGE,
1341                                               &change_info.info);
1342                 rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL);
1343         }
1344 }
1345 EXPORT_SYMBOL(netdev_state_change);
1346 
1347 /**
1348  * netdev_notify_peers - notify network peers about existence of @dev
1349  * @dev: network device
1350  *
1351  * Generate traffic such that interested network peers are aware of
1352  * @dev, such as by generating a gratuitous ARP. This may be used when
1353  * a device wants to inform the rest of the network about some sort of
1354  * reconfiguration such as a failover event or virtual machine
1355  * migration.
1356  */
1357 void netdev_notify_peers(struct net_device *dev)
1358 {
1359         rtnl_lock();
1360         call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev);
1361         call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev);
1362         rtnl_unlock();
1363 }
1364 EXPORT_SYMBOL(netdev_notify_peers);
1365 
1366 static int __dev_open(struct net_device *dev)
1367 {
1368         const struct net_device_ops *ops = dev->netdev_ops;
1369         int ret;
1370 
1371         ASSERT_RTNL();
1372 
1373         if (!netif_device_present(dev))
1374                 return -ENODEV;
1375 
1376         /* Block netpoll from trying to do any rx path servicing.
1377          * If we don't do this there is a chance ndo_poll_controller
1378          * or ndo_poll may be running while we open the device
1379          */
1380         netpoll_poll_disable(dev);
1381 
1382         ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev);
1383         ret = notifier_to_errno(ret);
1384         if (ret)
1385                 return ret;
1386 
1387         set_bit(__LINK_STATE_START, &dev->state);
1388 
1389         if (ops->ndo_validate_addr)
1390                 ret = ops->ndo_validate_addr(dev);
1391 
1392         if (!ret && ops->ndo_open)
1393                 ret = ops->ndo_open(dev);
1394 
1395         netpoll_poll_enable(dev);
1396 
1397         if (ret)
1398                 clear_bit(__LINK_STATE_START, &dev->state);
1399         else {
1400                 dev->flags |= IFF_UP;
1401                 dev_set_rx_mode(dev);
1402                 dev_activate(dev);
1403                 add_device_randomness(dev->dev_addr, dev->addr_len);
1404         }
1405 
1406         return ret;
1407 }
1408 
1409 /**
1410  *      dev_open        - prepare an interface for use.
1411  *      @dev:   device to open
1412  *
1413  *      Takes a device from down to up state. The device's private open
1414  *      function is invoked and then the multicast lists are loaded. Finally
1415  *      the device is moved into the up state and a %NETDEV_UP message is
1416  *      sent to the netdev notifier chain.
1417  *
1418  *      Calling this function on an active interface is a nop. On a failure
1419  *      a negative errno code is returned.
1420  */
1421 int dev_open(struct net_device *dev)
1422 {
1423         int ret;
1424 
1425         if (dev->flags & IFF_UP)
1426                 return 0;
1427 
1428         ret = __dev_open(dev);
1429         if (ret < 0)
1430                 return ret;
1431 
1432         rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1433         call_netdevice_notifiers(NETDEV_UP, dev);
1434 
1435         return ret;
1436 }
1437 EXPORT_SYMBOL(dev_open);
1438 
1439 static void __dev_close_many(struct list_head *head)
1440 {
1441         struct net_device *dev;
1442 
1443         ASSERT_RTNL();
1444         might_sleep();
1445 
1446         list_for_each_entry(dev, head, close_list) {
1447                 /* Temporarily disable netpoll until the interface is down */
1448                 netpoll_poll_disable(dev);
1449 
1450                 call_netdevice_notifiers(NETDEV_GOING_DOWN, dev);
1451 
1452                 clear_bit(__LINK_STATE_START, &dev->state);
1453 
1454                 /* Synchronize to scheduled poll. We cannot touch poll list, it
1455                  * can be even on different cpu. So just clear netif_running().
1456                  *
1457                  * dev->stop() will invoke napi_disable() on all of it's
1458                  * napi_struct instances on this device.
1459                  */
1460                 smp_mb__after_atomic(); /* Commit netif_running(). */
1461         }
1462 
1463         dev_deactivate_many(head);
1464 
1465         list_for_each_entry(dev, head, close_list) {
1466                 const struct net_device_ops *ops = dev->netdev_ops;
1467 
1468                 /*
1469                  *      Call the device specific close. This cannot fail.
1470                  *      Only if device is UP
1471                  *
1472                  *      We allow it to be called even after a DETACH hot-plug
1473                  *      event.
1474                  */
1475                 if (ops->ndo_stop)
1476                         ops->ndo_stop(dev);
1477 
1478                 dev->flags &= ~IFF_UP;
1479                 netpoll_poll_enable(dev);
1480         }
1481 }
1482 
1483 static void __dev_close(struct net_device *dev)
1484 {
1485         LIST_HEAD(single);
1486 
1487         list_add(&dev->close_list, &single);
1488         __dev_close_many(&single);
1489         list_del(&single);
1490 }
1491 
1492 void dev_close_many(struct list_head *head, bool unlink)
1493 {
1494         struct net_device *dev, *tmp;
1495 
1496         /* Remove the devices that don't need to be closed */
1497         list_for_each_entry_safe(dev, tmp, head, close_list)
1498                 if (!(dev->flags & IFF_UP))
1499                         list_del_init(&dev->close_list);
1500 
1501         __dev_close_many(head);
1502 
1503         list_for_each_entry_safe(dev, tmp, head, close_list) {
1504                 rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL);
1505                 call_netdevice_notifiers(NETDEV_DOWN, dev);
1506                 if (unlink)
1507                         list_del_init(&dev->close_list);
1508         }
1509 }
1510 EXPORT_SYMBOL(dev_close_many);
1511 
1512 /**
1513  *      dev_close - shutdown an interface.
1514  *      @dev: device to shutdown
1515  *
1516  *      This function moves an active device into down state. A
1517  *      %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1518  *      is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1519  *      chain.
1520  */
1521 void dev_close(struct net_device *dev)
1522 {
1523         if (dev->flags & IFF_UP) {
1524                 LIST_HEAD(single);
1525 
1526                 list_add(&dev->close_list, &single);
1527                 dev_close_many(&single, true);
1528                 list_del(&single);
1529         }
1530 }
1531 EXPORT_SYMBOL(dev_close);
1532 
1533 
1534 /**
1535  *      dev_disable_lro - disable Large Receive Offload on a device
1536  *      @dev: device
1537  *
1538  *      Disable Large Receive Offload (LRO) on a net device.  Must be
1539  *      called under RTNL.  This is needed if received packets may be
1540  *      forwarded to another interface.
1541  */
1542 void dev_disable_lro(struct net_device *dev)
1543 {
1544         struct net_device *lower_dev;
1545         struct list_head *iter;
1546 
1547         dev->wanted_features &= ~NETIF_F_LRO;
1548         netdev_update_features(dev);
1549 
1550         if (unlikely(dev->features & NETIF_F_LRO))
1551                 netdev_WARN(dev, "failed to disable LRO!\n");
1552 
1553         netdev_for_each_lower_dev(dev, lower_dev, iter)
1554                 dev_disable_lro(lower_dev);
1555 }
1556 EXPORT_SYMBOL(dev_disable_lro);
1557 
1558 /**
1559  *      dev_disable_gro_hw - disable HW Generic Receive Offload on a device
1560  *      @dev: device
1561  *
1562  *      Disable HW Generic Receive Offload (GRO_HW) on a net device.  Must be
1563  *      called under RTNL.  This is needed if Generic XDP is installed on
1564  *      the device.
1565  */
1566 static void dev_disable_gro_hw(struct net_device *dev)
1567 {
1568         dev->wanted_features &= ~NETIF_F_GRO_HW;
1569         netdev_update_features(dev);
1570 
1571         if (unlikely(dev->features & NETIF_F_GRO_HW))
1572                 netdev_WARN(dev, "failed to disable GRO_HW!\n");
1573 }
1574 
1575 const char *netdev_cmd_to_name(enum netdev_cmd cmd)
1576 {
1577 #define N(val)                                          \
1578         case NETDEV_##val:                              \
1579                 return "NETDEV_" __stringify(val);
1580         switch (cmd) {
1581         N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER)
1582         N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE)
1583         N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE)
1584         N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER)
1585         N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO)
1586         N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO)
1587         N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN)
1588         N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO)
1589         N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO)
1590         }
1591 #undef N
1592         return "UNKNOWN_NETDEV_EVENT";
1593 }
1594 EXPORT_SYMBOL_GPL(netdev_cmd_to_name);
1595 
1596 static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val,
1597                                    struct net_device *dev)
1598 {
1599         struct netdev_notifier_info info = {
1600                 .dev = dev,
1601         };
1602 
1603         return nb->notifier_call(nb, val, &info);
1604 }
1605 
1606 static int dev_boot_phase = 1;
1607 
1608 /**
1609  * register_netdevice_notifier - register a network notifier block
1610  * @nb: notifier
1611  *
1612  * Register a notifier to be called when network device events occur.
1613  * The notifier passed is linked into the kernel structures and must
1614  * not be reused until it has been unregistered. A negative errno code
1615  * is returned on a failure.
1616  *
1617  * When registered all registration and up events are replayed
1618  * to the new notifier to allow device to have a race free
1619  * view of the network device list.
1620  */
1621 
1622 int register_netdevice_notifier(struct notifier_block *nb)
1623 {
1624         struct net_device *dev;
1625         struct net_device *last;
1626         struct net *net;
1627         int err;
1628 
1629         /* Close race with setup_net() and cleanup_net() */
1630         down_write(&pernet_ops_rwsem);
1631         rtnl_lock();
1632         err = raw_notifier_chain_register(&netdev_chain, nb);
1633         if (err)
1634                 goto unlock;
1635         if (dev_boot_phase)
1636                 goto unlock;
1637         for_each_net(net) {
1638                 for_each_netdev(net, dev) {
1639                         err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev);
1640                         err = notifier_to_errno(err);
1641                         if (err)
1642                                 goto rollback;
1643 
1644                         if (!(dev->flags & IFF_UP))
1645                                 continue;
1646 
1647                         call_netdevice_notifier(nb, NETDEV_UP, dev);
1648                 }
1649         }
1650 
1651 unlock:
1652         rtnl_unlock();
1653         up_write(&pernet_ops_rwsem);
1654         return err;
1655 
1656 rollback:
1657         last = dev;
1658         for_each_net(net) {
1659                 for_each_netdev(net, dev) {
1660                         if (dev == last)
1661                                 goto outroll;
1662 
1663                         if (dev->flags & IFF_UP) {
1664                                 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1665                                                         dev);
1666                                 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1667                         }
1668                         call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1669                 }
1670         }
1671 
1672 outroll:
1673         raw_notifier_chain_unregister(&netdev_chain, nb);
1674         goto unlock;
1675 }
1676 EXPORT_SYMBOL(register_netdevice_notifier);
1677 
1678 /**
1679  * unregister_netdevice_notifier - unregister a network notifier block
1680  * @nb: notifier
1681  *
1682  * Unregister a notifier previously registered by
1683  * register_netdevice_notifier(). The notifier is unlinked into the
1684  * kernel structures and may then be reused. A negative errno code
1685  * is returned on a failure.
1686  *
1687  * After unregistering unregister and down device events are synthesized
1688  * for all devices on the device list to the removed notifier to remove
1689  * the need for special case cleanup code.
1690  */
1691 
1692 int unregister_netdevice_notifier(struct notifier_block *nb)
1693 {
1694         struct net_device *dev;
1695         struct net *net;
1696         int err;
1697 
1698         /* Close race with setup_net() and cleanup_net() */
1699         down_write(&pernet_ops_rwsem);
1700         rtnl_lock();
1701         err = raw_notifier_chain_unregister(&netdev_chain, nb);
1702         if (err)
1703                 goto unlock;
1704 
1705         for_each_net(net) {
1706                 for_each_netdev(net, dev) {
1707                         if (dev->flags & IFF_UP) {
1708                                 call_netdevice_notifier(nb, NETDEV_GOING_DOWN,
1709                                                         dev);
1710                                 call_netdevice_notifier(nb, NETDEV_DOWN, dev);
1711                         }
1712                         call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev);
1713                 }
1714         }
1715 unlock:
1716         rtnl_unlock();
1717         up_write(&pernet_ops_rwsem);
1718         return err;
1719 }
1720 EXPORT_SYMBOL(unregister_netdevice_notifier);
1721 
1722 /**
1723  *      call_netdevice_notifiers_info - call all network notifier blocks
1724  *      @val: value passed unmodified to notifier function
1725  *      @info: notifier information data
1726  *
1727  *      Call all network notifier blocks.  Parameters and return value
1728  *      are as for raw_notifier_call_chain().
1729  */
1730 
1731 static int call_netdevice_notifiers_info(unsigned long val,
1732                                          struct netdev_notifier_info *info)
1733 {
1734         ASSERT_RTNL();
1735         return raw_notifier_call_chain(&netdev_chain, val, info);
1736 }
1737 
1738 /**
1739  *      call_netdevice_notifiers - call all network notifier blocks
1740  *      @val: value passed unmodified to notifier function
1741  *      @dev: net_device pointer passed unmodified to notifier function
1742  *
1743  *      Call all network notifier blocks.  Parameters and return value
1744  *      are as for raw_notifier_call_chain().
1745  */
1746 
1747 int call_netdevice_notifiers(unsigned long val, struct net_device *dev)
1748 {
1749         struct netdev_notifier_info info = {
1750                 .dev = dev,
1751         };
1752 
1753         return call_netdevice_notifiers_info(val, &info);
1754 }
1755 EXPORT_SYMBOL(call_netdevice_notifiers);
1756 
1757 #ifdef CONFIG_NET_INGRESS
1758 static DEFINE_STATIC_KEY_FALSE(ingress_needed_key);
1759 
1760 void net_inc_ingress_queue(void)
1761 {
1762         static_branch_inc(&ingress_needed_key);
1763 }
1764 EXPORT_SYMBOL_GPL(net_inc_ingress_queue);
1765 
1766 void net_dec_ingress_queue(void)
1767 {
1768         static_branch_dec(&ingress_needed_key);
1769 }
1770 EXPORT_SYMBOL_GPL(net_dec_ingress_queue);
1771 #endif
1772 
1773 #ifdef CONFIG_NET_EGRESS
1774 static DEFINE_STATIC_KEY_FALSE(egress_needed_key);
1775 
1776 void net_inc_egress_queue(void)
1777 {
1778         static_branch_inc(&egress_needed_key);
1779 }
1780 EXPORT_SYMBOL_GPL(net_inc_egress_queue);
1781 
1782 void net_dec_egress_queue(void)
1783 {
1784         static_branch_dec(&egress_needed_key);
1785 }
1786 EXPORT_SYMBOL_GPL(net_dec_egress_queue);
1787 #endif
1788 
1789 static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key);
1790 #ifdef HAVE_JUMP_LABEL
1791 static atomic_t netstamp_needed_deferred;
1792 static atomic_t netstamp_wanted;
1793 static void netstamp_clear(struct work_struct *work)
1794 {
1795         int deferred = atomic_xchg(&netstamp_needed_deferred, 0);
1796         int wanted;
1797 
1798         wanted = atomic_add_return(deferred, &netstamp_wanted);
1799         if (wanted > 0)
1800                 static_branch_enable(&netstamp_needed_key);
1801         else
1802                 static_branch_disable(&netstamp_needed_key);
1803 }
1804 static DECLARE_WORK(netstamp_work, netstamp_clear);
1805 #endif
1806 
1807 void net_enable_timestamp(void)
1808 {
1809 #ifdef HAVE_JUMP_LABEL
1810         int wanted;
1811 
1812         while (1) {
1813                 wanted = atomic_read(&netstamp_wanted);
1814                 if (wanted <= 0)
1815                         break;
1816                 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted)
1817                         return;
1818         }
1819         atomic_inc(&netstamp_needed_deferred);
1820         schedule_work(&netstamp_work);
1821 #else
1822         static_branch_inc(&netstamp_needed_key);
1823 #endif
1824 }
1825 EXPORT_SYMBOL(net_enable_timestamp);
1826 
1827 void net_disable_timestamp(void)
1828 {
1829 #ifdef HAVE_JUMP_LABEL
1830         int wanted;
1831 
1832         while (1) {
1833                 wanted = atomic_read(&netstamp_wanted);
1834                 if (wanted <= 1)
1835                         break;
1836                 if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted)
1837                         return;
1838         }
1839         atomic_dec(&netstamp_needed_deferred);
1840         schedule_work(&netstamp_work);
1841 #else
1842         static_branch_dec(&netstamp_needed_key);
1843 #endif
1844 }
1845 EXPORT_SYMBOL(net_disable_timestamp);
1846 
1847 static inline void net_timestamp_set(struct sk_buff *skb)
1848 {
1849         skb->tstamp = 0;
1850         if (static_branch_unlikely(&netstamp_needed_key))
1851                 __net_timestamp(skb);
1852 }
1853 
1854 #define net_timestamp_check(COND, SKB)                          \
1855         if (static_branch_unlikely(&netstamp_needed_key)) {     \
1856                 if ((COND) && !(SKB)->tstamp)                   \
1857                         __net_timestamp(SKB);                   \
1858         }                                                       \
1859 
1860 bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb)
1861 {
1862         unsigned int len;
1863 
1864         if (!(dev->flags & IFF_UP))
1865                 return false;
1866 
1867         len = dev->mtu + dev->hard_header_len + VLAN_HLEN;
1868         if (skb->len <= len)
1869                 return true;
1870 
1871         /* if TSO is enabled, we don't care about the length as the packet
1872          * could be forwarded without being segmented before
1873          */
1874         if (skb_is_gso(skb))
1875                 return true;
1876 
1877         return false;
1878 }
1879 EXPORT_SYMBOL_GPL(is_skb_forwardable);
1880 
1881 int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1882 {
1883         int ret = ____dev_forward_skb(dev, skb);
1884 
1885         if (likely(!ret)) {
1886                 skb->protocol = eth_type_trans(skb, dev);
1887                 skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN);
1888         }
1889 
1890         return ret;
1891 }
1892 EXPORT_SYMBOL_GPL(__dev_forward_skb);
1893 
1894 /**
1895  * dev_forward_skb - loopback an skb to another netif
1896  *
1897  * @dev: destination network device
1898  * @skb: buffer to forward
1899  *
1900  * return values:
1901  *      NET_RX_SUCCESS  (no congestion)
1902  *      NET_RX_DROP     (packet was dropped, but freed)
1903  *
1904  * dev_forward_skb can be used for injecting an skb from the
1905  * start_xmit function of one device into the receive queue
1906  * of another device.
1907  *
1908  * The receiving device may be in another namespace, so
1909  * we have to clear all information in the skb that could
1910  * impact namespace isolation.
1911  */
1912 int dev_forward_skb(struct net_device *dev, struct sk_buff *skb)
1913 {
1914         return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb);
1915 }
1916 EXPORT_SYMBOL_GPL(dev_forward_skb);
1917 
1918 static inline int deliver_skb(struct sk_buff *skb,
1919                               struct packet_type *pt_prev,
1920                               struct net_device *orig_dev)
1921 {
1922         if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
1923                 return -ENOMEM;
1924         refcount_inc(&skb->users);
1925         return pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
1926 }
1927 
1928 static inline void deliver_ptype_list_skb(struct sk_buff *skb,
1929                                           struct packet_type **pt,
1930                                           struct net_device *orig_dev,
1931                                           __be16 type,
1932                                           struct list_head *ptype_list)
1933 {
1934         struct packet_type *ptype, *pt_prev = *pt;
1935 
1936         list_for_each_entry_rcu(ptype, ptype_list, list) {
1937                 if (ptype->type != type)
1938                         continue;
1939                 if (pt_prev)
1940                         deliver_skb(skb, pt_prev, orig_dev);
1941                 pt_prev = ptype;
1942         }
1943         *pt = pt_prev;
1944 }
1945 
1946 static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb)
1947 {
1948         if (!ptype->af_packet_priv || !skb->sk)
1949                 return false;
1950 
1951         if (ptype->id_match)
1952                 return ptype->id_match(ptype, skb->sk);
1953         else if ((struct sock *)ptype->af_packet_priv == skb->sk)
1954                 return true;
1955 
1956         return false;
1957 }
1958 
1959 /*
1960  *      Support routine. Sends outgoing frames to any network
1961  *      taps currently in use.
1962  */
1963 
1964 void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev)
1965 {
1966         struct packet_type *ptype;
1967         struct sk_buff *skb2 = NULL;
1968         struct packet_type *pt_prev = NULL;
1969         struct list_head *ptype_list = &ptype_all;
1970 
1971         rcu_read_lock();
1972 again:
1973         list_for_each_entry_rcu(ptype, ptype_list, list) {
1974                 /* Never send packets back to the socket
1975                  * they originated from - MvS (miquels@drinkel.ow.org)
1976                  */
1977                 if (skb_loop_sk(ptype, skb))
1978                         continue;
1979 
1980                 if (pt_prev) {
1981                         deliver_skb(skb2, pt_prev, skb->dev);
1982                         pt_prev = ptype;
1983                         continue;
1984                 }
1985 
1986                 /* need to clone skb, done only once */
1987                 skb2 = skb_clone(skb, GFP_ATOMIC);
1988                 if (!skb2)
1989                         goto out_unlock;
1990 
1991                 net_timestamp_set(skb2);
1992 
1993                 /* skb->nh should be correctly
1994                  * set by sender, so that the second statement is
1995                  * just protection against buggy protocols.
1996                  */
1997                 skb_reset_mac_header(skb2);
1998 
1999                 if (skb_network_header(skb2) < skb2->data ||
2000                     skb_network_header(skb2) > skb_tail_pointer(skb2)) {
2001                         net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
2002                                              ntohs(skb2->protocol),
2003                                              dev->name);
2004                         skb_reset_network_header(skb2);
2005                 }
2006 
2007                 skb2->transport_header = skb2->network_header;
2008                 skb2->pkt_type = PACKET_OUTGOING;
2009                 pt_prev = ptype;
2010         }
2011 
2012         if (ptype_list == &ptype_all) {
2013                 ptype_list = &dev->ptype_all;
2014                 goto again;
2015         }
2016 out_unlock:
2017         if (pt_prev) {
2018                 if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC))
2019                         pt_prev->func(skb2, skb->dev, pt_prev, skb->dev);
2020                 else
2021                         kfree_skb(skb2);
2022         }
2023         rcu_read_unlock();
2024 }
2025 EXPORT_SYMBOL_GPL(dev_queue_xmit_nit);
2026 
2027 /**
2028  * netif_setup_tc - Handle tc mappings on real_num_tx_queues change
2029  * @dev: Network device
2030  * @txq: number of queues available
2031  *
2032  * If real_num_tx_queues is changed the tc mappings may no longer be
2033  * valid. To resolve this verify the tc mapping remains valid and if
2034  * not NULL the mapping. With no priorities mapping to this
2035  * offset/count pair it will no longer be used. In the worst case TC0
2036  * is invalid nothing can be done so disable priority mappings. If is
2037  * expected that drivers will fix this mapping if they can before
2038  * calling netif_set_real_num_tx_queues.
2039  */
2040 static void netif_setup_tc(struct net_device *dev, unsigned int txq)
2041 {
2042         int i;
2043         struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2044 
2045         /* If TC0 is invalidated disable TC mapping */
2046         if (tc->offset + tc->count > txq) {
2047                 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
2048                 dev->num_tc = 0;
2049                 return;
2050         }
2051 
2052         /* Invalidated prio to tc mappings set to TC0 */
2053         for (i = 1; i < TC_BITMASK + 1; i++) {
2054                 int q = netdev_get_prio_tc_map(dev, i);
2055 
2056                 tc = &dev->tc_to_txq[q];
2057                 if (tc->offset + tc->count > txq) {
2058                         pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
2059                                 i, q);
2060                         netdev_set_prio_tc_map(dev, i, 0);
2061                 }
2062         }
2063 }
2064 
2065 int netdev_txq_to_tc(struct net_device *dev, unsigned int txq)
2066 {
2067         if (dev->num_tc) {
2068                 struct netdev_tc_txq *tc = &dev->tc_to_txq[0];
2069                 int i;
2070 
2071                 for (i = 0; i < TC_MAX_QUEUE; i++, tc++) {
2072                         if ((txq - tc->offset) < tc->count)
2073                                 return i;
2074                 }
2075 
2076                 return -1;
2077         }
2078 
2079         return 0;
2080 }
2081 EXPORT_SYMBOL(netdev_txq_to_tc);
2082 
2083 #ifdef CONFIG_XPS
2084 static DEFINE_MUTEX(xps_map_mutex);
2085 #define xmap_dereference(P)             \
2086         rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
2087 
2088 static bool remove_xps_queue(struct xps_dev_maps *dev_maps,
2089                              int tci, u16 index)
2090 {
2091         struct xps_map *map = NULL;
2092         int pos;
2093 
2094         if (dev_maps)
2095                 map = xmap_dereference(dev_maps->cpu_map[tci]);
2096         if (!map)
2097                 return false;
2098 
2099         for (pos = map->len; pos--;) {
2100                 if (map->queues[pos] != index)
2101                         continue;
2102 
2103                 if (map->len > 1) {
2104                         map->queues[pos] = map->queues[--map->len];
2105                         break;
2106                 }
2107 
2108                 RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL);
2109                 kfree_rcu(map, rcu);
2110                 return false;
2111         }
2112 
2113         return true;
2114 }
2115 
2116 static bool remove_xps_queue_cpu(struct net_device *dev,
2117                                  struct xps_dev_maps *dev_maps,
2118                                  int cpu, u16 offset, u16 count)
2119 {
2120         int num_tc = dev->num_tc ? : 1;
2121         bool active = false;
2122         int tci;
2123 
2124         for (tci = cpu * num_tc; num_tc--; tci++) {
2125                 int i, j;
2126 
2127                 for (i = count, j = offset; i--; j++) {
2128                         if (!remove_xps_queue(dev_maps, tci, j))
2129                                 break;
2130                 }
2131 
2132                 active |= i < 0;
2133         }
2134 
2135         return active;
2136 }
2137 
2138 static void netif_reset_xps_queues(struct net_device *dev, u16 offset,
2139                                    u16 count)
2140 {
2141         struct xps_dev_maps *dev_maps;
2142         int cpu, i;
2143         bool active = false;
2144 
2145         mutex_lock(&xps_map_mutex);
2146         dev_maps = xmap_dereference(dev->xps_maps);
2147 
2148         if (!dev_maps)
2149                 goto out_no_maps;
2150 
2151         for_each_possible_cpu(cpu)
2152                 active |= remove_xps_queue_cpu(dev, dev_maps, cpu,
2153                                                offset, count);
2154 
2155         if (!active) {
2156                 RCU_INIT_POINTER(dev->xps_maps, NULL);
2157                 kfree_rcu(dev_maps, rcu);
2158         }
2159 
2160         for (i = offset + (count - 1); count--; i--)
2161                 netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i),
2162                                              NUMA_NO_NODE);
2163 
2164 out_no_maps:
2165         mutex_unlock(&xps_map_mutex);
2166 }
2167 
2168 static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index)
2169 {
2170         netif_reset_xps_queues(dev, index, dev->num_tx_queues - index);
2171 }
2172 
2173 static struct xps_map *expand_xps_map(struct xps_map *map,
2174                                       int cpu, u16 index)
2175 {
2176         struct xps_map *new_map;
2177         int alloc_len = XPS_MIN_MAP_ALLOC;
2178         int i, pos;
2179 
2180         for (pos = 0; map && pos < map->len; pos++) {
2181                 if (map->queues[pos] != index)
2182                         continue;
2183                 return map;
2184         }
2185 
2186         /* Need to add queue to this CPU's existing map */
2187         if (map) {
2188                 if (pos < map->alloc_len)
2189                         return map;
2190 
2191                 alloc_len = map->alloc_len * 2;
2192         }
2193 
2194         /* Need to allocate new map to store queue on this CPU's map */
2195         new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL,
2196                                cpu_to_node(cpu));
2197         if (!new_map)
2198                 return NULL;
2199 
2200         for (i = 0; i < pos; i++)
2201                 new_map->queues[i] = map->queues[i];
2202         new_map->alloc_len = alloc_len;
2203         new_map->len = pos;
2204 
2205         return new_map;
2206 }
2207 
2208 int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask,
2209                         u16 index)
2210 {
2211         struct xps_dev_maps *dev_maps, *new_dev_maps = NULL;
2212         int i, cpu, tci, numa_node_id = -2;
2213         int maps_sz, num_tc = 1, tc = 0;
2214         struct xps_map *map, *new_map;
2215         bool active = false;
2216 
2217         if (dev->num_tc) {
2218                 num_tc = dev->num_tc;
2219                 tc = netdev_txq_to_tc(dev, index);
2220                 if (tc < 0)
2221                         return -EINVAL;
2222         }
2223 
2224         maps_sz = XPS_DEV_MAPS_SIZE(num_tc);
2225         if (maps_sz < L1_CACHE_BYTES)
2226                 maps_sz = L1_CACHE_BYTES;
2227 
2228         mutex_lock(&xps_map_mutex);
2229 
2230         dev_maps = xmap_dereference(dev->xps_maps);
2231 
2232         /* allocate memory for queue storage */
2233         for_each_cpu_and(cpu, cpu_online_mask, mask) {
2234                 if (!new_dev_maps)
2235                         new_dev_maps = kzalloc(maps_sz, GFP_KERNEL);
2236                 if (!new_dev_maps) {
2237                         mutex_unlock(&xps_map_mutex);
2238                         return -ENOMEM;
2239                 }
2240 
2241                 tci = cpu * num_tc + tc;
2242                 map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) :
2243                                  NULL;
2244 
2245                 map = expand_xps_map(map, cpu, index);
2246                 if (!map)
2247                         goto error;
2248 
2249                 RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2250         }
2251 
2252         if (!new_dev_maps)
2253                 goto out_no_new_maps;
2254 
2255         for_each_possible_cpu(cpu) {
2256                 /* copy maps belonging to foreign traffic classes */
2257                 for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) {
2258                         /* fill in the new device map from the old device map */
2259                         map = xmap_dereference(dev_maps->cpu_map[tci]);
2260                         RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2261                 }
2262 
2263                 /* We need to explicitly update tci as prevous loop
2264                  * could break out early if dev_maps is NULL.
2265                  */
2266                 tci = cpu * num_tc + tc;
2267 
2268                 if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) {
2269                         /* add queue to CPU maps */
2270                         int pos = 0;
2271 
2272                         map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2273                         while ((pos < map->len) && (map->queues[pos] != index))
2274                                 pos++;
2275 
2276                         if (pos == map->len)
2277                                 map->queues[map->len++] = index;
2278 #ifdef CONFIG_NUMA
2279                         if (numa_node_id == -2)
2280                                 numa_node_id = cpu_to_node(cpu);
2281                         else if (numa_node_id != cpu_to_node(cpu))
2282                                 numa_node_id = -1;
2283 #endif
2284                 } else if (dev_maps) {
2285                         /* fill in the new device map from the old device map */
2286                         map = xmap_dereference(dev_maps->cpu_map[tci]);
2287                         RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2288                 }
2289 
2290                 /* copy maps belonging to foreign traffic classes */
2291                 for (i = num_tc - tc, tci++; dev_maps && --i; tci++) {
2292                         /* fill in the new device map from the old device map */
2293                         map = xmap_dereference(dev_maps->cpu_map[tci]);
2294                         RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map);
2295                 }
2296         }
2297 
2298         rcu_assign_pointer(dev->xps_maps, new_dev_maps);
2299 
2300         /* Cleanup old maps */
2301         if (!dev_maps)
2302                 goto out_no_old_maps;
2303 
2304         for_each_possible_cpu(cpu) {
2305                 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2306                         new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2307                         map = xmap_dereference(dev_maps->cpu_map[tci]);
2308                         if (map && map != new_map)
2309                                 kfree_rcu(map, rcu);
2310                 }
2311         }
2312 
2313         kfree_rcu(dev_maps, rcu);
2314 
2315 out_no_old_maps:
2316         dev_maps = new_dev_maps;
2317         active = true;
2318 
2319 out_no_new_maps:
2320         /* update Tx queue numa node */
2321         netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index),
2322                                      (numa_node_id >= 0) ? numa_node_id :
2323                                      NUMA_NO_NODE);
2324 
2325         if (!dev_maps)
2326                 goto out_no_maps;
2327 
2328         /* removes queue from unused CPUs */
2329         for_each_possible_cpu(cpu) {
2330                 for (i = tc, tci = cpu * num_tc; i--; tci++)
2331                         active |= remove_xps_queue(dev_maps, tci, index);
2332                 if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu))
2333                         active |= remove_xps_queue(dev_maps, tci, index);
2334                 for (i = num_tc - tc, tci++; --i; tci++)
2335                         active |= remove_xps_queue(dev_maps, tci, index);
2336         }
2337 
2338         /* free map if not active */
2339         if (!active) {
2340                 RCU_INIT_POINTER(dev->xps_maps, NULL);
2341                 kfree_rcu(dev_maps, rcu);
2342         }
2343 
2344 out_no_maps:
2345         mutex_unlock(&xps_map_mutex);
2346 
2347         return 0;
2348 error:
2349         /* remove any maps that we added */
2350         for_each_possible_cpu(cpu) {
2351                 for (i = num_tc, tci = cpu * num_tc; i--; tci++) {
2352                         new_map = xmap_dereference(new_dev_maps->cpu_map[tci]);
2353                         map = dev_maps ?
2354                               xmap_dereference(dev_maps->cpu_map[tci]) :
2355                               NULL;
2356                         if (new_map && new_map != map)
2357                                 kfree(new_map);
2358                 }
2359         }
2360 
2361         mutex_unlock(&xps_map_mutex);
2362 
2363         kfree(new_dev_maps);
2364         return -ENOMEM;
2365 }
2366 EXPORT_SYMBOL(netif_set_xps_queue);
2367 
2368 #endif
2369 void netdev_reset_tc(struct net_device *dev)
2370 {
2371 #ifdef CONFIG_XPS
2372         netif_reset_xps_queues_gt(dev, 0);
2373 #endif
2374         dev->num_tc = 0;
2375         memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq));
2376         memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map));
2377 }
2378 EXPORT_SYMBOL(netdev_reset_tc);
2379 
2380 int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset)
2381 {
2382         if (tc >= dev->num_tc)
2383                 return -EINVAL;
2384 
2385 #ifdef CONFIG_XPS
2386         netif_reset_xps_queues(dev, offset, count);
2387 #endif
2388         dev->tc_to_txq[tc].count = count;
2389         dev->tc_to_txq[tc].offset = offset;
2390         return 0;
2391 }
2392 EXPORT_SYMBOL(netdev_set_tc_queue);
2393 
2394 int netdev_set_num_tc(struct net_device *dev, u8 num_tc)
2395 {
2396         if (num_tc > TC_MAX_QUEUE)
2397                 return -EINVAL;
2398 
2399 #ifdef CONFIG_XPS
2400         netif_reset_xps_queues_gt(dev, 0);
2401 #endif
2402         dev->num_tc = num_tc;
2403         return 0;
2404 }
2405 EXPORT_SYMBOL(netdev_set_num_tc);
2406 
2407 /*
2408  * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
2409  * greater than real_num_tx_queues stale skbs on the qdisc must be flushed.
2410  */
2411 int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq)
2412 {
2413         bool disabling;
2414         int rc;
2415 
2416         disabling = txq < dev->real_num_tx_queues;
2417 
2418         if (txq < 1 || txq > dev->num_tx_queues)
2419                 return -EINVAL;
2420 
2421         if (dev->reg_state == NETREG_REGISTERED ||
2422             dev->reg_state == NETREG_UNREGISTERING) {
2423                 ASSERT_RTNL();
2424 
2425                 rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues,
2426                                                   txq);
2427                 if (rc)
2428                         return rc;
2429 
2430                 if (dev->num_tc)
2431                         netif_setup_tc(dev, txq);
2432 
2433                 dev->real_num_tx_queues = txq;
2434 
2435                 if (disabling) {
2436                         synchronize_net();
2437                         qdisc_reset_all_tx_gt(dev, txq);
2438 #ifdef CONFIG_XPS
2439                         netif_reset_xps_queues_gt(dev, txq);
2440 #endif
2441                 }
2442         } else {
2443                 dev->real_num_tx_queues = txq;
2444         }
2445 
2446         return 0;
2447 }
2448 EXPORT_SYMBOL(netif_set_real_num_tx_queues);
2449 
2450 #ifdef CONFIG_SYSFS
2451 /**
2452  *      netif_set_real_num_rx_queues - set actual number of RX queues used
2453  *      @dev: Network device
2454  *      @rxq: Actual number of RX queues
2455  *
2456  *      This must be called either with the rtnl_lock held or before
2457  *      registration of the net device.  Returns 0 on success, or a
2458  *      negative error code.  If called before registration, it always
2459  *      succeeds.
2460  */
2461 int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq)
2462 {
2463         int rc;
2464 
2465         if (rxq < 1 || rxq > dev->num_rx_queues)
2466                 return -EINVAL;
2467 
2468         if (dev->reg_state == NETREG_REGISTERED) {
2469                 ASSERT_RTNL();
2470 
2471                 rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues,
2472                                                   rxq);
2473                 if (rc)
2474                         return rc;
2475         }
2476 
2477         dev->real_num_rx_queues = rxq;
2478         return 0;
2479 }
2480 EXPORT_SYMBOL(netif_set_real_num_rx_queues);
2481 #endif
2482 
2483 /**
2484  * netif_get_num_default_rss_queues - default number of RSS queues
2485  *
2486  * This routine should set an upper limit on the number of RSS queues
2487  * used by default by multiqueue devices.
2488  */
2489 int netif_get_num_default_rss_queues(void)
2490 {
2491         return is_kdump_kernel() ?
2492                 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus());
2493 }
2494 EXPORT_SYMBOL(netif_get_num_default_rss_queues);
2495 
2496 static void __netif_reschedule(struct Qdisc *q)
2497 {
2498         struct softnet_data *sd;
2499         unsigned long flags;
2500 
2501         local_irq_save(flags);
2502         sd = this_cpu_ptr(&softnet_data);
2503         q->next_sched = NULL;
2504         *sd->output_queue_tailp = q;
2505         sd->output_queue_tailp = &q->next_sched;
2506         raise_softirq_irqoff(NET_TX_SOFTIRQ);
2507         local_irq_restore(flags);
2508 }
2509 
2510 void __netif_schedule(struct Qdisc *q)
2511 {
2512         if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state))
2513                 __netif_reschedule(q);
2514 }
2515 EXPORT_SYMBOL(__netif_schedule);
2516 
2517 struct dev_kfree_skb_cb {
2518         enum skb_free_reason reason;
2519 };
2520 
2521 static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb)
2522 {
2523         return (struct dev_kfree_skb_cb *)skb->cb;
2524 }
2525 
2526 void netif_schedule_queue(struct netdev_queue *txq)
2527 {
2528         rcu_read_lock();
2529         if (!(txq->state & QUEUE_STATE_ANY_XOFF)) {
2530                 struct Qdisc *q = rcu_dereference(txq->qdisc);
2531 
2532                 __netif_schedule(q);
2533         }
2534         rcu_read_unlock();
2535 }
2536 EXPORT_SYMBOL(netif_schedule_queue);
2537 
2538 void netif_tx_wake_queue(struct netdev_queue *dev_queue)
2539 {
2540         if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) {
2541                 struct Qdisc *q;
2542 
2543                 rcu_read_lock();
2544                 q = rcu_dereference(dev_queue->qdisc);
2545                 __netif_schedule(q);
2546                 rcu_read_unlock();
2547         }
2548 }
2549 EXPORT_SYMBOL(netif_tx_wake_queue);
2550 
2551 void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason)
2552 {
2553         unsigned long flags;
2554 
2555         if (unlikely(!skb))
2556                 return;
2557 
2558         if (likely(refcount_read(&skb->users) == 1)) {
2559                 smp_rmb();
2560                 refcount_set(&skb->users, 0);
2561         } else if (likely(!refcount_dec_and_test(&skb->users))) {
2562                 return;
2563         }
2564         get_kfree_skb_cb(skb)->reason = reason;
2565         local_irq_save(flags);
2566         skb->next = __this_cpu_read(softnet_data.completion_queue);
2567         __this_cpu_write(softnet_data.completion_queue, skb);
2568         raise_softirq_irqoff(NET_TX_SOFTIRQ);
2569         local_irq_restore(flags);
2570 }
2571 EXPORT_SYMBOL(__dev_kfree_skb_irq);
2572 
2573 void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason)
2574 {
2575         if (in_irq() || irqs_disabled())
2576                 __dev_kfree_skb_irq(skb, reason);
2577         else
2578                 dev_kfree_skb(skb);
2579 }
2580 EXPORT_SYMBOL(__dev_kfree_skb_any);
2581 
2582 
2583 /**
2584  * netif_device_detach - mark device as removed
2585  * @dev: network device
2586  *
2587  * Mark device as removed from system and therefore no longer available.
2588  */
2589 void netif_device_detach(struct net_device *dev)
2590 {
2591         if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) &&
2592             netif_running(dev)) {
2593                 netif_tx_stop_all_queues(dev);
2594         }
2595 }
2596 EXPORT_SYMBOL(netif_device_detach);
2597 
2598 /**
2599  * netif_device_attach - mark device as attached
2600  * @dev: network device
2601  *
2602  * Mark device as attached from system and restart if needed.
2603  */
2604 void netif_device_attach(struct net_device *dev)
2605 {
2606         if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) &&
2607             netif_running(dev)) {
2608                 netif_tx_wake_all_queues(dev);
2609                 __netdev_watchdog_up(dev);
2610         }
2611 }
2612 EXPORT_SYMBOL(netif_device_attach);
2613 
2614 /*
2615  * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2616  * to be used as a distribution range.
2617  */
2618 static u16 skb_tx_hash(const struct net_device *dev, struct sk_buff *skb)
2619 {
2620         u32 hash;
2621         u16 qoffset = 0;
2622         u16 qcount = dev->real_num_tx_queues;
2623 
2624         if (skb_rx_queue_recorded(skb)) {
2625                 hash = skb_get_rx_queue(skb);
2626                 while (unlikely(hash >= qcount))
2627                         hash -= qcount;
2628                 return hash;
2629         }
2630 
2631         if (dev->num_tc) {
2632                 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
2633 
2634                 qoffset = dev->tc_to_txq[tc].offset;
2635                 qcount = dev->tc_to_txq[tc].count;
2636         }
2637 
2638         return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset;
2639 }
2640 
2641 static void skb_warn_bad_offload(const struct sk_buff *skb)
2642 {
2643         static const netdev_features_t null_features;
2644         struct net_device *dev = skb->dev;
2645         const char *name = "";
2646 
2647         if (!net_ratelimit())
2648                 return;
2649 
2650         if (dev) {
2651                 if (dev->dev.parent)
2652                         name = dev_driver_string(dev->dev.parent);
2653                 else
2654                         name = netdev_name(dev);
2655         }
2656         WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
2657              "gso_type=%d ip_summed=%d\n",
2658              name, dev ? &dev->features : &null_features,
2659              skb->sk ? &skb->sk->sk_route_caps : &null_features,
2660              skb->len, skb->data_len, skb_shinfo(skb)->gso_size,
2661              skb_shinfo(skb)->gso_type, skb->ip_summed);
2662 }
2663 
2664 /*
2665  * Invalidate hardware checksum when packet is to be mangled, and
2666  * complete checksum manually on outgoing path.
2667  */
2668 int skb_checksum_help(struct sk_buff *skb)
2669 {
2670         __wsum csum;
2671         int ret = 0, offset;
2672 
2673         if (skb->ip_summed == CHECKSUM_COMPLETE)
2674                 goto out_set_summed;
2675 
2676         if (unlikely(skb_shinfo(skb)->gso_size)) {
2677                 skb_warn_bad_offload(skb);
2678                 return -EINVAL;
2679         }
2680 
2681         /* Before computing a checksum, we should make sure no frag could
2682          * be modified by an external entity : checksum could be wrong.
2683          */
2684         if (skb_has_shared_frag(skb)) {
2685                 ret = __skb_linearize(skb);
2686                 if (ret)
2687                         goto out;
2688         }
2689 
2690         offset = skb_checksum_start_offset(skb);
2691         BUG_ON(offset >= skb_headlen(skb));
2692         csum = skb_checksum(skb, offset, skb->len - offset, 0);
2693 
2694         offset += skb->csum_offset;
2695         BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb));
2696 
2697         if (skb_cloned(skb) &&
2698             !skb_clone_writable(skb, offset + sizeof(__sum16))) {
2699                 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2700                 if (ret)
2701                         goto out;
2702         }
2703 
2704         *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0;
2705 out_set_summed:
2706         skb->ip_summed = CHECKSUM_NONE;
2707 out:
2708         return ret;
2709 }
2710 EXPORT_SYMBOL(skb_checksum_help);
2711 
2712 int skb_crc32c_csum_help(struct sk_buff *skb)
2713 {
2714         __le32 crc32c_csum;
2715         int ret = 0, offset, start;
2716 
2717         if (skb->ip_summed != CHECKSUM_PARTIAL)
2718                 goto out;
2719 
2720         if (unlikely(skb_is_gso(skb)))
2721                 goto out;
2722 
2723         /* Before computing a checksum, we should make sure no frag could
2724          * be modified by an external entity : checksum could be wrong.
2725          */
2726         if (unlikely(skb_has_shared_frag(skb))) {
2727                 ret = __skb_linearize(skb);
2728                 if (ret)
2729                         goto out;
2730         }
2731         start = skb_checksum_start_offset(skb);
2732         offset = start + offsetof(struct sctphdr, checksum);
2733         if (WARN_ON_ONCE(offset >= skb_headlen(skb))) {
2734                 ret = -EINVAL;
2735                 goto out;
2736         }
2737         if (skb_cloned(skb) &&
2738             !skb_clone_writable(skb, offset + sizeof(__le32))) {
2739                 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2740                 if (ret)
2741                         goto out;
2742         }
2743         crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start,
2744                                                   skb->len - start, ~(__u32)0,
2745                                                   crc32c_csum_stub));
2746         *(__le32 *)(skb->data + offset) = crc32c_csum;
2747         skb->ip_summed = CHECKSUM_NONE;
2748         skb->csum_not_inet = 0;
2749 out:
2750         return ret;
2751 }
2752 
2753 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
2754 {
2755         __be16 type = skb->protocol;
2756 
2757         /* Tunnel gso handlers can set protocol to ethernet. */
2758         if (type == htons(ETH_P_TEB)) {
2759                 struct ethhdr *eth;
2760 
2761                 if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr))))
2762                         return 0;
2763 
2764                 eth = (struct ethhdr *)skb->data;
2765                 type = eth->h_proto;
2766         }
2767 
2768         return __vlan_get_protocol(skb, type, depth);
2769 }
2770 
2771 /**
2772  *      skb_mac_gso_segment - mac layer segmentation handler.
2773  *      @skb: buffer to segment
2774  *      @features: features for the output path (see dev->features)
2775  */
2776 struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb,
2777                                     netdev_features_t features)
2778 {
2779         struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT);
2780         struct packet_offload *ptype;
2781         int vlan_depth = skb->mac_len;
2782         __be16 type = skb_network_protocol(skb, &vlan_depth);
2783 
2784         if (unlikely(!type))
2785                 return ERR_PTR(-EINVAL);
2786 
2787         __skb_pull(skb, vlan_depth);
2788 
2789         rcu_read_lock();
2790         list_for_each_entry_rcu(ptype, &offload_base, list) {
2791                 if (ptype->type == type && ptype->callbacks.gso_segment) {
2792                         segs = ptype->callbacks.gso_segment(skb, features);
2793                         break;
2794                 }
2795         }
2796         rcu_read_unlock();
2797 
2798         __skb_push(skb, skb->data - skb_mac_header(skb));
2799 
2800         return segs;
2801 }
2802 EXPORT_SYMBOL(skb_mac_gso_segment);
2803 
2804 
2805 /* openvswitch calls this on rx path, so we need a different check.
2806  */
2807 static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path)
2808 {
2809         if (tx_path)
2810                 return skb->ip_summed != CHECKSUM_PARTIAL &&
2811                        skb->ip_summed != CHECKSUM_UNNECESSARY;
2812 
2813         return skb->ip_summed == CHECKSUM_NONE;
2814 }
2815 
2816 /**
2817  *      __skb_gso_segment - Perform segmentation on skb.
2818  *      @skb: buffer to segment
2819  *      @features: features for the output path (see dev->features)
2820  *      @tx_path: whether it is called in TX path
2821  *
2822  *      This function segments the given skb and returns a list of segments.
2823  *
2824  *      It may return NULL if the skb requires no segmentation.  This is
2825  *      only possible when GSO is used for verifying header integrity.
2826  *
2827  *      Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb.
2828  */
2829 struct sk_buff *__skb_gso_segment(struct sk_buff *skb,
2830                                   netdev_features_t features, bool tx_path)
2831 {
2832         struct sk_buff *segs;
2833 
2834         if (unlikely(skb_needs_check(skb, tx_path))) {
2835                 int err;
2836 
2837                 /* We're going to init ->check field in TCP or UDP header */
2838                 err = skb_cow_head(skb, 0);
2839                 if (err < 0)
2840                         return ERR_PTR(err);
2841         }
2842 
2843         /* Only report GSO partial support if it will enable us to
2844          * support segmentation on this frame without needing additional
2845          * work.
2846          */
2847         if (features & NETIF_F_GSO_PARTIAL) {
2848                 netdev_features_t partial_features = NETIF_F_GSO_ROBUST;
2849                 struct net_device *dev = skb->dev;
2850 
2851                 partial_features |= dev->features & dev->gso_partial_features;
2852                 if (!skb_gso_ok(skb, features | partial_features))
2853                         features &= ~NETIF_F_GSO_PARTIAL;
2854         }
2855 
2856         BUILD_BUG_ON(SKB_SGO_CB_OFFSET +
2857                      sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb));
2858 
2859         SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb);
2860         SKB_GSO_CB(skb)->encap_level = 0;
2861 
2862         skb_reset_mac_header(skb);
2863         skb_reset_mac_len(skb);
2864 
2865         segs = skb_mac_gso_segment(skb, features);
2866 
2867         if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs)))
2868                 skb_warn_bad_offload(skb);
2869 
2870         return segs;
2871 }
2872 EXPORT_SYMBOL(__skb_gso_segment);
2873 
2874 /* Take action when hardware reception checksum errors are detected. */
2875 #ifdef CONFIG_BUG
2876 void netdev_rx_csum_fault(struct net_device *dev)
2877 {
2878         if (net_ratelimit()) {
2879                 pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>");
2880                 dump_stack();
2881         }
2882 }
2883 EXPORT_SYMBOL(netdev_rx_csum_fault);
2884 #endif
2885 
2886 /* XXX: check that highmem exists at all on the given machine. */
2887 static int illegal_highdma(struct net_device *dev, struct sk_buff *skb)
2888 {
2889 #ifdef CONFIG_HIGHMEM
2890         int i;
2891 
2892         if (!(dev->features & NETIF_F_HIGHDMA)) {
2893                 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2894                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2895 
2896                         if (PageHighMem(skb_frag_page(frag)))
2897                                 return 1;
2898                 }
2899         }
2900 #endif
2901         return 0;
2902 }
2903 
2904 /* If MPLS offload request, verify we are testing hardware MPLS features
2905  * instead of standard features for the netdev.
2906  */
2907 #if IS_ENABLED(CONFIG_NET_MPLS_GSO)
2908 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2909                                            netdev_features_t features,
2910                                            __be16 type)
2911 {
2912         if (eth_p_mpls(type))
2913                 features &= skb->dev->mpls_features;
2914 
2915         return features;
2916 }
2917 #else
2918 static netdev_features_t net_mpls_features(struct sk_buff *skb,
2919                                            netdev_features_t features,
2920                                            __be16 type)
2921 {
2922         return features;
2923 }
2924 #endif
2925 
2926 static netdev_features_t harmonize_features(struct sk_buff *skb,
2927         netdev_features_t features)
2928 {
2929         int tmp;
2930         __be16 type;
2931 
2932         type = skb_network_protocol(skb, &tmp);
2933         features = net_mpls_features(skb, features, type);
2934 
2935         if (skb->ip_summed != CHECKSUM_NONE &&
2936             !can_checksum_protocol(features, type)) {
2937                 features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2938         }
2939         if (illegal_highdma(skb->dev, skb))
2940                 features &= ~NETIF_F_SG;
2941 
2942         return features;
2943 }
2944 
2945 netdev_features_t passthru_features_check(struct sk_buff *skb,
2946                                           struct net_device *dev,
2947                                           netdev_features_t features)
2948 {
2949         return features;
2950 }
2951 EXPORT_SYMBOL(passthru_features_check);
2952 
2953 static netdev_features_t dflt_features_check(struct sk_buff *skb,
2954                                              struct net_device *dev,
2955                                              netdev_features_t features)
2956 {
2957         return vlan_features_check(skb, features);
2958 }
2959 
2960 static netdev_features_t gso_features_check(const struct sk_buff *skb,
2961                                             struct net_device *dev,
2962                                             netdev_features_t features)
2963 {
2964         u16 gso_segs = skb_shinfo(skb)->gso_segs;
2965 
2966         if (gso_segs > dev->gso_max_segs)
2967                 return features & ~NETIF_F_GSO_MASK;
2968 
2969         /* Support for GSO partial features requires software
2970          * intervention before we can actually process the packets
2971          * so we need to strip support for any partial features now
2972          * and we can pull them back in after we have partially
2973          * segmented the frame.
2974          */
2975         if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL))
2976                 features &= ~dev->gso_partial_features;
2977 
2978         /* Make sure to clear the IPv4 ID mangling feature if the
2979          * IPv4 header has the potential to be fragmented.
2980          */
2981         if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) {
2982                 struct iphdr *iph = skb->encapsulation ?
2983                                     inner_ip_hdr(skb) : ip_hdr(skb);
2984 
2985                 if (!(iph->frag_off & htons(IP_DF)))
2986                         features &= ~NETIF_F_TSO_MANGLEID;
2987         }
2988 
2989         return features;
2990 }
2991 
2992 netdev_features_t netif_skb_features(struct sk_buff *skb)
2993 {
2994         struct net_device *dev = skb->dev;
2995         netdev_features_t features = dev->features;
2996 
2997         if (skb_is_gso(skb))
2998                 features = gso_features_check(skb, dev, features);
2999 
3000         /* If encapsulation offload request, verify we are testing
3001          * hardware encapsulation features instead of standard
3002          * features for the netdev
3003          */
3004         if (skb->encapsulation)
3005                 features &= dev->hw_enc_features;
3006 
3007         if (skb_vlan_tagged(skb))
3008                 features = netdev_intersect_features(features,
3009                                                      dev->vlan_features |
3010                                                      NETIF_F_HW_VLAN_CTAG_TX |
3011                                                      NETIF_F_HW_VLAN_STAG_TX);
3012 
3013         if (dev->netdev_ops->ndo_features_check)
3014                 features &= dev->netdev_ops->ndo_features_check(skb, dev,
3015                                                                 features);
3016         else
3017                 features &= dflt_features_check(skb, dev, features);
3018 
3019         return harmonize_features(skb, features);
3020 }
3021 EXPORT_SYMBOL(netif_skb_features);
3022 
3023 static int xmit_one(struct sk_buff *skb, struct net_device *dev,
3024                     struct netdev_queue *txq, bool more)
3025 {
3026         unsigned int len;
3027         int rc;
3028 
3029         if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all))
3030                 dev_queue_xmit_nit(skb, dev);
3031 
3032         len = skb->len;
3033         trace_net_dev_start_xmit(skb, dev);
3034         rc = netdev_start_xmit(skb, dev, txq, more);
3035         trace_net_dev_xmit(skb, rc, dev, len);
3036 
3037         return rc;
3038 }
3039 
3040 struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev,
3041                                     struct netdev_queue *txq, int *ret)
3042 {
3043         struct sk_buff *skb = first;
3044         int rc = NETDEV_TX_OK;
3045 
3046         while (skb) {
3047                 struct sk_buff *next = skb->next;
3048 
3049                 skb->next = NULL;
3050                 rc = xmit_one(skb, dev, txq, next != NULL);
3051                 if (unlikely(!dev_xmit_complete(rc))) {
3052                         skb->next = next;
3053                         goto out;
3054                 }
3055 
3056                 skb = next;
3057                 if (netif_xmit_stopped(txq) && skb) {
3058                         rc = NETDEV_TX_BUSY;
3059                         break;
3060                 }
3061         }
3062 
3063 out:
3064         *ret = rc;
3065         return skb;
3066 }
3067 
3068 static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb,
3069                                           netdev_features_t features)
3070 {
3071         if (skb_vlan_tag_present(skb) &&
3072             !vlan_hw_offload_capable(features, skb->vlan_proto))
3073                 skb = __vlan_hwaccel_push_inside(skb);
3074         return skb;
3075 }
3076 
3077 int skb_csum_hwoffload_help(struct sk_buff *skb,
3078                             const netdev_features_t features)
3079 {
3080         if (unlikely(skb->csum_not_inet))
3081                 return !!(features & NETIF_F_SCTP_CRC) ? 0 :
3082                         skb_crc32c_csum_help(skb);
3083 
3084         return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb);
3085 }
3086 EXPORT_SYMBOL(skb_csum_hwoffload_help);
3087 
3088 static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again)
3089 {
3090         netdev_features_t features;
3091 
3092         features = netif_skb_features(skb);
3093         skb = validate_xmit_vlan(skb, features);
3094         if (unlikely(!skb))
3095                 goto out_null;
3096 
3097         skb = sk_validate_xmit_skb(skb, dev);
3098         if (unlikely(!skb))
3099                 goto out_null;
3100 
3101         if (netif_needs_gso(skb, features)) {
3102                 struct sk_buff *segs;
3103 
3104                 segs = skb_gso_segment(skb, features);
3105                 if (IS_ERR(segs)) {
3106                         goto out_kfree_skb;
3107                 } else if (segs) {
3108                         consume_skb(skb);
3109                         skb = segs;
3110                 }
3111         } else {
3112                 if (skb_needs_linearize(skb, features) &&
3113                     __skb_linearize(skb))
3114                         goto out_kfree_skb;
3115 
3116                 /* If packet is not checksummed and device does not
3117                  * support checksumming for this protocol, complete
3118                  * checksumming here.
3119                  */
3120                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3121                         if (skb->encapsulation)
3122                                 skb_set_inner_transport_header(skb,
3123                                                                skb_checksum_start_offset(skb));
3124                         else
3125                                 skb_set_transport_header(skb,
3126                                                          skb_checksum_start_offset(skb));
3127                         if (skb_csum_hwoffload_help(skb, features))
3128                                 goto out_kfree_skb;
3129                 }
3130         }
3131 
3132         skb = validate_xmit_xfrm(skb, features, again);
3133 
3134         return skb;
3135 
3136 out_kfree_skb:
3137         kfree_skb(skb);
3138 out_null:
3139         atomic_long_inc(&dev->tx_dropped);
3140         return NULL;
3141 }
3142 
3143 struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again)
3144 {
3145         struct sk_buff *next, *head = NULL, *tail;
3146 
3147         for (; skb != NULL; skb = next) {
3148                 next = skb->next;
3149                 skb->next = NULL;
3150 
3151                 /* in case skb wont be segmented, point to itself */
3152                 skb->prev = skb;
3153 
3154                 skb = validate_xmit_skb(skb, dev, again);
3155                 if (!skb)
3156                         continue;
3157 
3158                 if (!head)
3159                         head = skb;
3160                 else
3161                         tail->next = skb;
3162                 /* If skb was segmented, skb->prev points to
3163                  * the last segment. If not, it still contains skb.
3164                  */
3165                 tail = skb->prev;
3166         }
3167         return head;
3168 }
3169 EXPORT_SYMBOL_GPL(validate_xmit_skb_list);
3170 
3171 static void qdisc_pkt_len_init(struct sk_buff *skb)
3172 {
3173         const struct skb_shared_info *shinfo = skb_shinfo(skb);
3174 
3175         qdisc_skb_cb(skb)->pkt_len = skb->len;
3176 
3177         /* To get more precise estimation of bytes sent on wire,
3178          * we add to pkt_len the headers size of all segments
3179          */
3180         if (shinfo->gso_size)  {
3181                 unsigned int hdr_len;
3182                 u16 gso_segs = shinfo->gso_segs;
3183 
3184                 /* mac layer + network layer */
3185                 hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
3186 
3187                 /* + transport layer */
3188                 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
3189                         const struct tcphdr *th;
3190                         struct tcphdr _tcphdr;
3191 
3192                         th = skb_header_pointer(skb, skb_transport_offset(skb),
3193                                                 sizeof(_tcphdr), &_tcphdr);
3194                         if (likely(th))
3195                                 hdr_len += __tcp_hdrlen(th);
3196                 } else {
3197                         struct udphdr _udphdr;
3198 
3199                         if (skb_header_pointer(skb, skb_transport_offset(skb),
3200                                                sizeof(_udphdr), &_udphdr))
3201                                 hdr_len += sizeof(struct udphdr);
3202                 }
3203 
3204                 if (shinfo->gso_type & SKB_GSO_DODGY)
3205                         gso_segs = DIV_ROUND_UP(skb->len - hdr_len,
3206                                                 shinfo->gso_size);
3207 
3208                 qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len;
3209         }
3210 }
3211 
3212 static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
3213                                  struct net_device *dev,
3214                                  struct netdev_queue *txq)
3215 {
3216         spinlock_t *root_lock = qdisc_lock(q);
3217         struct sk_buff *to_free = NULL;
3218         bool contended;
3219         int rc;
3220 
3221         qdisc_calculate_pkt_len(skb, q);
3222 
3223         if (q->flags & TCQ_F_NOLOCK) {
3224                 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3225                         __qdisc_drop(skb, &to_free);
3226                         rc = NET_XMIT_DROP;
3227                 } else {
3228                         rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3229                         qdisc_run(q);
3230                 }
3231 
3232                 if (unlikely(to_free))
3233                         kfree_skb_list(to_free);
3234                 return rc;
3235         }
3236 
3237         /*
3238          * Heuristic to force contended enqueues to serialize on a
3239          * separate lock before trying to get qdisc main lock.
3240          * This permits qdisc->running owner to get the lock more
3241          * often and dequeue packets faster.
3242          */
3243         contended = qdisc_is_running(q);
3244         if (unlikely(contended))
3245                 spin_lock(&q->busylock);
3246 
3247         spin_lock(root_lock);
3248         if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) {
3249                 __qdisc_drop(skb, &to_free);
3250                 rc = NET_XMIT_DROP;
3251         } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) &&
3252                    qdisc_run_begin(q)) {
3253                 /*
3254                  * This is a work-conserving queue; there are no old skbs
3255                  * waiting to be sent out; and the qdisc is not running -
3256                  * xmit the skb directly.
3257                  */
3258 
3259                 qdisc_bstats_update(q, skb);
3260 
3261                 if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) {
3262                         if (unlikely(contended)) {
3263                                 spin_unlock(&q->busylock);
3264                                 contended = false;
3265                         }
3266                         __qdisc_run(q);
3267                 }
3268 
3269                 qdisc_run_end(q);
3270                 rc = NET_XMIT_SUCCESS;
3271         } else {
3272                 rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK;
3273                 if (qdisc_run_begin(q)) {
3274                         if (unlikely(contended)) {
3275                                 spin_unlock(&q->busylock);
3276                                 contended = false;
3277                         }
3278                         __qdisc_run(q);
3279                         qdisc_run_end(q);
3280                 }
3281         }
3282         spin_unlock(root_lock);
3283         if (unlikely(to_free))
3284                 kfree_skb_list(to_free);
3285         if (unlikely(contended))
3286                 spin_unlock(&q->busylock);
3287         return rc;
3288 }
3289 
3290 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
3291 static void skb_update_prio(struct sk_buff *skb)
3292 {
3293         const struct netprio_map *map;
3294         const struct sock *sk;
3295         unsigned int prioidx;
3296 
3297         if (skb->priority)
3298                 return;
3299         map = rcu_dereference_bh(skb->dev->priomap);
3300         if (!map)
3301                 return;
3302         sk = skb_to_full_sk(skb);
3303         if (!sk)
3304                 return;
3305 
3306         prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data);
3307 
3308         if (prioidx < map->priomap_len)
3309                 skb->priority = map->priomap[prioidx];
3310 }
3311 #else
3312 #define skb_update_prio(skb)
3313 #endif
3314 
3315 DEFINE_PER_CPU(int, xmit_recursion);
3316 EXPORT_SYMBOL(xmit_recursion);
3317 
3318 /**
3319  *      dev_loopback_xmit - loop back @skb
3320  *      @net: network namespace this loopback is happening in
3321  *      @sk:  sk needed to be a netfilter okfn
3322  *      @skb: buffer to transmit
3323  */
3324 int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
3325 {
3326         skb_reset_mac_header(skb);
3327         __skb_pull(skb, skb_network_offset(skb));
3328         skb->pkt_type = PACKET_LOOPBACK;
3329         skb->ip_summed = CHECKSUM_UNNECESSARY;
3330         WARN_ON(!skb_dst(skb));
3331         skb_dst_force(skb);
3332         netif_rx_ni(skb);
3333         return 0;
3334 }
3335 EXPORT_SYMBOL(dev_loopback_xmit);
3336 
3337 #ifdef CONFIG_NET_EGRESS
3338 static struct sk_buff *
3339 sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
3340 {
3341         struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress);
3342         struct tcf_result cl_res;
3343 
3344         if (!miniq)
3345                 return skb;
3346 
3347         /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */
3348         mini_qdisc_bstats_cpu_update(miniq, skb);
3349 
3350         switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
3351         case TC_ACT_OK:
3352         case TC_ACT_RECLASSIFY:
3353                 skb->tc_index = TC_H_MIN(cl_res.classid);
3354                 break;
3355         case TC_ACT_SHOT:
3356                 mini_qdisc_qstats_cpu_drop(miniq);
3357                 *ret = NET_XMIT_DROP;
3358                 kfree_skb(skb);
3359                 return NULL;
3360         case TC_ACT_STOLEN:
3361         case TC_ACT_QUEUED:
3362         case TC_ACT_TRAP:
3363                 *ret = NET_XMIT_SUCCESS;
3364                 consume_skb(skb);
3365                 return NULL;
3366         case TC_ACT_REDIRECT:
3367                 /* No need to push/pop skb's mac_header here on egress! */
3368                 skb_do_redirect(skb);
3369                 *ret = NET_XMIT_SUCCESS;
3370                 return NULL;
3371         default:
3372                 break;
3373         }
3374 
3375         return skb;
3376 }
3377 #endif /* CONFIG_NET_EGRESS */
3378 
3379 static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb)
3380 {
3381 #ifdef CONFIG_XPS
3382         struct xps_dev_maps *dev_maps;
3383         struct xps_map *map;
3384         int queue_index = -1;
3385 
3386         rcu_read_lock();
3387         dev_maps = rcu_dereference(dev->xps_maps);
3388         if (dev_maps) {
3389                 unsigned int tci = skb->sender_cpu - 1;
3390 
3391                 if (dev->num_tc) {
3392                         tci *= dev->num_tc;
3393                         tci += netdev_get_prio_tc_map(dev, skb->priority);
3394                 }
3395 
3396                 map = rcu_dereference(dev_maps->cpu_map[tci]);
3397                 if (map) {
3398                         if (map->len == 1)
3399                                 queue_index = map->queues[0];
3400                         else
3401                                 queue_index = map->queues[reciprocal_scale(skb_get_hash(skb),
3402                                                                            map->len)];
3403                         if (unlikely(queue_index >= dev->real_num_tx_queues))
3404                                 queue_index = -1;
3405                 }
3406         }
3407         rcu_read_unlock();
3408 
3409         return queue_index;
3410 #else
3411         return -1;
3412 #endif
3413 }
3414 
3415 static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb)
3416 {
3417         struct sock *sk = skb->sk;
3418         int queue_index = sk_tx_queue_get(sk);
3419 
3420         if (queue_index < 0 || skb->ooo_okay ||
3421             queue_index >= dev->real_num_tx_queues) {
3422                 int new_index = get_xps_queue(dev, skb);
3423 
3424                 if (new_index < 0)
3425                         new_index = skb_tx_hash(dev, skb);
3426 
3427                 if (queue_index != new_index && sk &&
3428                     sk_fullsock(sk) &&
3429                     rcu_access_pointer(sk->sk_dst_cache))
3430                         sk_tx_queue_set(sk, new_index);
3431 
3432                 queue_index = new_index;
3433         }
3434 
3435         return queue_index;
3436 }
3437 
3438 struct netdev_queue *netdev_pick_tx(struct net_device *dev,
3439                                     struct sk_buff *skb,
3440                                     void *accel_priv)
3441 {
3442         int queue_index = 0;
3443 
3444 #ifdef CONFIG_XPS
3445         u32 sender_cpu = skb->sender_cpu - 1;
3446 
3447         if (sender_cpu >= (u32)NR_CPUS)
3448                 skb->sender_cpu = raw_smp_processor_id() + 1;
3449 #endif
3450 
3451         if (dev->real_num_tx_queues != 1) {
3452                 const struct net_device_ops *ops = dev->netdev_ops;
3453 
3454                 if (ops->ndo_select_queue)
3455                         queue_index = ops->ndo_select_queue(dev, skb, accel_priv,
3456                                                             __netdev_pick_tx);
3457                 else
3458                         queue_index = __netdev_pick_tx(dev, skb);
3459 
3460                 queue_index = netdev_cap_txqueue(dev, queue_index);
3461         }
3462 
3463         skb_set_queue_mapping(skb, queue_index);
3464         return netdev_get_tx_queue(dev, queue_index);
3465 }
3466 
3467 /**
3468  *      __dev_queue_xmit - transmit a buffer
3469  *      @skb: buffer to transmit
3470  *      @accel_priv: private data used for L2 forwarding offload
3471  *
3472  *      Queue a buffer for transmission to a network device. The caller must
3473  *      have set the device and priority and built the buffer before calling
3474  *      this function. The function can be called from an interrupt.
3475  *
3476  *      A negative errno code is returned on a failure. A success does not
3477  *      guarantee the frame will be transmitted as it may be dropped due
3478  *      to congestion or traffic shaping.
3479  *
3480  * -----------------------------------------------------------------------------------
3481  *      I notice this method can also return errors from the queue disciplines,
3482  *      including NET_XMIT_DROP, which is a positive value.  So, errors can also
3483  *      be positive.
3484  *
3485  *      Regardless of the return value, the skb is consumed, so it is currently
3486  *      difficult to retry a send to this method.  (You can bump the ref count
3487  *      before sending to hold a reference for retry if you are careful.)
3488  *
3489  *      When calling this method, interrupts MUST be enabled.  This is because
3490  *      the BH enable code must have IRQs enabled so that it will not deadlock.
3491  *          --BLG
3492  */
3493 static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv)
3494 {
3495         struct net_device *dev = skb->dev;
3496         struct netdev_queue *txq;
3497         struct Qdisc *q;
3498         int rc = -ENOMEM;
3499         bool again = false;
3500 
3501         skb_reset_mac_header(skb);
3502 
3503         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP))
3504                 __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED);
3505 
3506         /* Disable soft irqs for various locks below. Also
3507          * stops preemption for RCU.
3508          */
3509         rcu_read_lock_bh();
3510 
3511         skb_update_prio(skb);
3512 
3513         qdisc_pkt_len_init(skb);
3514 #ifdef CONFIG_NET_CLS_ACT
3515         skb->tc_at_ingress = 0;
3516 # ifdef CONFIG_NET_EGRESS
3517         if (static_branch_unlikely(&egress_needed_key)) {
3518                 skb = sch_handle_egress(skb, &rc, dev);
3519                 if (!skb)
3520                         goto out;
3521         }
3522 # endif
3523 #endif
3524         /* If device/qdisc don't need skb->dst, release it right now while
3525          * its hot in this cpu cache.
3526          */
3527         if (dev->priv_flags & IFF_XMIT_DST_RELEASE)
3528                 skb_dst_drop(skb);
3529         else
3530                 skb_dst_force(skb);
3531 
3532         txq = netdev_pick_tx(dev, skb, accel_priv);
3533         q = rcu_dereference_bh(txq->qdisc);
3534 
3535         trace_net_dev_queue(skb);
3536         if (q->enqueue) {
3537                 rc = __dev_xmit_skb(skb, q, dev, txq);
3538                 goto out;
3539         }
3540 
3541         /* The device has no queue. Common case for software devices:
3542          * loopback, all the sorts of tunnels...
3543 
3544          * Really, it is unlikely that netif_tx_lock protection is necessary
3545          * here.  (f.e. loopback and IP tunnels are clean ignoring statistics
3546          * counters.)
3547          * However, it is possible, that they rely on protection
3548          * made by us here.
3549 
3550          * Check this and shot the lock. It is not prone from deadlocks.
3551          *Either shot noqueue qdisc, it is even simpler 8)
3552          */
3553         if (dev->flags & IFF_UP) {
3554                 int cpu = smp_processor_id(); /* ok because BHs are off */
3555 
3556                 if (txq->xmit_lock_owner != cpu) {
3557                         if (unlikely(__this_cpu_read(xmit_recursion) >
3558                                      XMIT_RECURSION_LIMIT))
3559                                 goto recursion_alert;
3560 
3561                         skb = validate_xmit_skb(skb, dev, &again);
3562                         if (!skb)
3563                                 goto out;
3564 
3565                         HARD_TX_LOCK(dev, txq, cpu);
3566 
3567                         if (!netif_xmit_stopped(txq)) {
3568                                 __this_cpu_inc(xmit_recursion);
3569                                 skb = dev_hard_start_xmit(skb, dev, txq, &rc);
3570                                 __this_cpu_dec(xmit_recursion);
3571                                 if (dev_xmit_complete(rc)) {
3572                                         HARD_TX_UNLOCK(dev, txq);
3573                                         goto out;
3574                                 }
3575                         }
3576                         HARD_TX_UNLOCK(dev, txq);
3577                         net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
3578                                              dev->name);
3579                 } else {
3580                         /* Recursion is detected! It is possible,
3581                          * unfortunately
3582                          */
3583 recursion_alert:
3584                         net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
3585                                              dev->name);
3586                 }
3587         }
3588 
3589         rc = -ENETDOWN;
3590         rcu_read_unlock_bh();
3591 
3592         atomic_long_inc(&dev->tx_dropped);
3593         kfree_skb_list(skb);
3594         return rc;
3595 out:
3596         rcu_read_unlock_bh();
3597         return rc;
3598 }
3599 
3600 int dev_queue_xmit(struct sk_buff *skb)
3601 {
3602         return __dev_queue_xmit(skb, NULL);
3603 }
3604 EXPORT_SYMBOL(dev_queue_xmit);
3605 
3606 int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv)
3607 {
3608         return __dev_queue_xmit(skb, accel_priv);
3609 }
3610 EXPORT_SYMBOL(dev_queue_xmit_accel);
3611 
3612 int dev_direct_xmit(struct sk_buff *skb, u16 queue_id)
3613 {
3614         struct net_device *dev = skb->dev;
3615         struct sk_buff *orig_skb = skb;
3616         struct netdev_queue *txq;
3617         int ret = NETDEV_TX_BUSY;
3618         bool again = false;
3619 
3620         if (unlikely(!netif_running(dev) ||
3621                      !netif_carrier_ok(dev)))
3622                 goto drop;
3623 
3624         skb = validate_xmit_skb_list(skb, dev, &again);
3625         if (skb != orig_skb)
3626                 goto drop;
3627 
3628         skb_set_queue_mapping(skb, queue_id);
3629         txq = skb_get_tx_queue(dev, skb);
3630 
3631         local_bh_disable();
3632 
3633         HARD_TX_LOCK(dev, txq, smp_processor_id());
3634         if (!netif_xmit_frozen_or_drv_stopped(txq))
3635                 ret = netdev_start_xmit(skb, dev, txq, false);
3636         HARD_TX_UNLOCK(dev, txq);
3637 
3638         local_bh_enable();
3639 
3640         if (!dev_xmit_complete(ret))
3641                 kfree_skb(skb);
3642 
3643         return ret;
3644 drop:
3645         atomic_long_inc(&dev->tx_dropped);
3646         kfree_skb_list(skb);
3647         return NET_XMIT_DROP;
3648 }
3649 EXPORT_SYMBOL(dev_direct_xmit);
3650 
3651 /*************************************************************************
3652  *                      Receiver routines
3653  *************************************************************************/
3654 
3655 int netdev_max_backlog __read_mostly = 1000;
3656 EXPORT_SYMBOL(netdev_max_backlog);
3657 
3658 int netdev_tstamp_prequeue __read_mostly = 1;
3659 int netdev_budget __read_mostly = 300;
3660 unsigned int __read_mostly netdev_budget_usecs = 2000;
3661 int weight_p __read_mostly = 64;           /* old backlog weight */
3662 int dev_weight_rx_bias __read_mostly = 1;  /* bias for backlog weight */
3663 int dev_weight_tx_bias __read_mostly = 1;  /* bias for output_queue quota */
3664 int dev_rx_weight __read_mostly = 64;
3665 int dev_tx_weight __read_mostly = 64;
3666 
3667 /* Called with irq disabled */
3668 static inline void ____napi_schedule(struct softnet_data *sd,
3669                                      struct napi_struct *napi)
3670 {
3671         list_add_tail(&napi->poll_list, &sd->poll_list);
3672         __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3673 }
3674 
3675 #ifdef CONFIG_RPS
3676 
3677 /* One global table that all flow-based protocols share. */
3678 struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly;
3679 EXPORT_SYMBOL(rps_sock_flow_table);
3680 u32 rps_cpu_mask __read_mostly;
3681 EXPORT_SYMBOL(rps_cpu_mask);
3682 
3683 struct static_key rps_needed __read_mostly;
3684 EXPORT_SYMBOL(rps_needed);
3685 struct static_key rfs_needed __read_mostly;
3686 EXPORT_SYMBOL(rfs_needed);
3687 
3688 static struct rps_dev_flow *
3689 set_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3690             struct rps_dev_flow *rflow, u16 next_cpu)
3691 {
3692         if (next_cpu < nr_cpu_ids) {
3693 #ifdef CONFIG_RFS_ACCEL
3694                 struct netdev_rx_queue *rxqueue;
3695                 struct rps_dev_flow_table *flow_table;
3696                 struct rps_dev_flow *old_rflow;
3697                 u32 flow_id;
3698                 u16 rxq_index;
3699                 int rc;
3700 
3701                 /* Should we steer this flow to a different hardware queue? */
3702                 if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap ||
3703                     !(dev->features & NETIF_F_NTUPLE))
3704                         goto out;
3705                 rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu);
3706                 if (rxq_index == skb_get_rx_queue(skb))
3707                         goto out;
3708 
3709                 rxqueue = dev->_rx + rxq_index;
3710                 flow_table = rcu_dereference(rxqueue->rps_flow_table);
3711                 if (!flow_table)
3712                         goto out;
3713                 flow_id = skb_get_hash(skb) & flow_table->mask;
3714                 rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb,
3715                                                         rxq_index, flow_id);
3716                 if (rc < 0)
3717                         goto out;
3718                 old_rflow = rflow;
3719                 rflow = &flow_table->flows[flow_id];
3720                 rflow->filter = rc;
3721                 if (old_rflow->filter == rflow->filter)
3722                         old_rflow->filter = RPS_NO_FILTER;
3723         out:
3724 #endif
3725                 rflow->last_qtail =
3726                         per_cpu(softnet_data, next_cpu).input_queue_head;
3727         }
3728 
3729         rflow->cpu = next_cpu;
3730         return rflow;
3731 }
3732 
3733 /*
3734  * get_rps_cpu is called from netif_receive_skb and returns the target
3735  * CPU from the RPS map of the receiving queue for a given skb.
3736  * rcu_read_lock must be held on entry.
3737  */
3738 static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb,
3739                        struct rps_dev_flow **rflowp)
3740 {
3741         const struct rps_sock_flow_table *sock_flow_table;
3742         struct netdev_rx_queue *rxqueue = dev->_rx;
3743         struct rps_dev_flow_table *flow_table;
3744         struct rps_map *map;
3745         int cpu = -1;
3746         u32 tcpu;
3747         u32 hash;
3748 
3749         if (skb_rx_queue_recorded(skb)) {
3750                 u16 index = skb_get_rx_queue(skb);
3751 
3752                 if (unlikely(index >= dev->real_num_rx_queues)) {
3753                         WARN_ONCE(dev->real_num_rx_queues > 1,
3754                                   "%s received packet on queue %u, but number "
3755                                   "of RX queues is %u\n",
3756                                   dev->name, index, dev->real_num_rx_queues);
3757                         goto done;
3758                 }
3759                 rxqueue += index;
3760         }
3761 
3762         /* Avoid computing hash if RFS/RPS is not active for this rxqueue */
3763 
3764         flow_table = rcu_dereference(rxqueue->rps_flow_table);
3765         map = rcu_dereference(rxqueue->rps_map);
3766         if (!flow_table && !map)
3767                 goto done;
3768 
3769         skb_reset_network_header(skb);
3770         hash = skb_get_hash(skb);
3771         if (!hash)
3772                 goto done;
3773 
3774         sock_flow_table = rcu_dereference(rps_sock_flow_table);
3775         if (flow_table && sock_flow_table) {
3776                 struct rps_dev_flow *rflow;
3777                 u32 next_cpu;
3778                 u32 ident;
3779 
3780                 /* First check into global flow table if there is a match */
3781                 ident = sock_flow_table->ents[hash & sock_flow_table->mask];
3782                 if ((ident ^ hash) & ~rps_cpu_mask)
3783                         goto try_rps;
3784 
3785                 next_cpu = ident & rps_cpu_mask;
3786 
3787                 /* OK, now we know there is a match,
3788                  * we can look at the local (per receive queue) flow table
3789                  */
3790                 rflow = &flow_table->flows[hash & flow_table->mask];
3791                 tcpu = rflow->cpu;
3792 
3793                 /*
3794                  * If the desired CPU (where last recvmsg was done) is
3795                  * different from current CPU (one in the rx-queue flow
3796                  * table entry), switch if one of the following holds:
3797                  *   - Current CPU is unset (>= nr_cpu_ids).
3798                  *   - Current CPU is offline.
3799                  *   - The current CPU's queue tail has advanced beyond the
3800                  *     last packet that was enqueued using this table entry.
3801                  *     This guarantees that all previous packets for the flow
3802                  *     have been dequeued, thus preserving in order delivery.
3803                  */
3804                 if (unlikely(tcpu != next_cpu) &&
3805                     (tcpu >= nr_cpu_ids || !cpu_online(tcpu) ||
3806                      ((int)(per_cpu(softnet_data, tcpu).input_queue_head -
3807                       rflow->last_qtail)) >= 0)) {
3808                         tcpu = next_cpu;
3809                         rflow = set_rps_cpu(dev, skb, rflow, next_cpu);
3810                 }
3811 
3812                 if (tcpu < nr_cpu_ids && cpu_online(tcpu)) {
3813                         *rflowp = rflow;
3814                         cpu = tcpu;
3815                         goto done;
3816                 }
3817         }
3818 
3819 try_rps:
3820 
3821         if (map) {
3822                 tcpu = map->cpus[reciprocal_scale(hash, map->len)];
3823                 if (cpu_online(tcpu)) {
3824                         cpu = tcpu;
3825                         goto done;
3826                 }
3827         }
3828 
3829 done:
3830         return cpu;
3831 }
3832 
3833 #ifdef CONFIG_RFS_ACCEL
3834 
3835 /**
3836  * rps_may_expire_flow - check whether an RFS hardware filter may be removed
3837  * @dev: Device on which the filter was set
3838  * @rxq_index: RX queue index
3839  * @flow_id: Flow ID passed to ndo_rx_flow_steer()
3840  * @filter_id: Filter ID returned by ndo_rx_flow_steer()
3841  *
3842  * Drivers that implement ndo_rx_flow_steer() should periodically call
3843  * this function for each installed filter and remove the filters for
3844  * which it returns %true.
3845  */
3846 bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index,
3847                          u32 flow_id, u16 filter_id)
3848 {
3849         struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index;
3850         struct rps_dev_flow_table *flow_table;
3851         struct rps_dev_flow *rflow;
3852         bool expire = true;
3853         unsigned int cpu;
3854 
3855         rcu_read_lock();
3856         flow_table = rcu_dereference(rxqueue->rps_flow_table);
3857         if (flow_table && flow_id <= flow_table->mask) {
3858                 rflow = &flow_table->flows[flow_id];
3859                 cpu = READ_ONCE(rflow->cpu);
3860                 if (rflow->filter == filter_id && cpu < nr_cpu_ids &&
3861                     ((int)(per_cpu(softnet_data, cpu).input_queue_head -
3862                            rflow->last_qtail) <
3863                      (int)(10 * flow_table->mask)))
3864                         expire = false;
3865         }
3866         rcu_read_unlock();
3867         return expire;
3868 }
3869 EXPORT_SYMBOL(rps_may_expire_flow);
3870 
3871 #endif /* CONFIG_RFS_ACCEL */
3872 
3873 /* Called from hardirq (IPI) context */
3874 static void rps_trigger_softirq(void *data)
3875 {
3876         struct softnet_data *sd = data;
3877 
3878         ____napi_schedule(sd, &sd->backlog);
3879         sd->received_rps++;
3880 }
3881 
3882 #endif /* CONFIG_RPS */
3883 
3884 /*
3885  * Check if this softnet_data structure is another cpu one
3886  * If yes, queue it to our IPI list and return 1
3887  * If no, return 0
3888  */
3889 static int rps_ipi_queued(struct softnet_data *sd)
3890 {
3891 #ifdef CONFIG_RPS
3892         struct softnet_data *mysd = this_cpu_ptr(&softnet_data);
3893 
3894         if (sd != mysd) {
3895                 sd->rps_ipi_next = mysd->rps_ipi_list;
3896                 mysd->rps_ipi_list = sd;
3897 
3898                 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
3899                 return 1;
3900         }
3901 #endif /* CONFIG_RPS */
3902         return 0;
3903 }
3904 
3905 #ifdef CONFIG_NET_FLOW_LIMIT
3906 int netdev_flow_limit_table_len __read_mostly = (1 << 12);
3907 #endif
3908 
3909 static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen)
3910 {
3911 #ifdef CONFIG_NET_FLOW_LIMIT
3912         struct sd_flow_limit *fl;
3913         struct softnet_data *sd;
3914         unsigned int old_flow, new_flow;
3915 
3916         if (qlen < (netdev_max_backlog >> 1))
3917                 return false;
3918 
3919         sd = this_cpu_ptr(&softnet_data);
3920 
3921         rcu_read_lock();
3922         fl = rcu_dereference(sd->flow_limit);
3923         if (fl) {
3924                 new_flow = skb_get_hash(skb) & (fl->num_buckets - 1);
3925                 old_flow = fl->history[fl->history_head];
3926                 fl->history[fl->history_head] = new_flow;
3927 
3928                 fl->history_head++;
3929                 fl->history_head &= FLOW_LIMIT_HISTORY - 1;
3930 
3931                 if (likely(fl->buckets[old_flow]))
3932                         fl->buckets[old_flow]--;
3933 
3934                 if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) {
3935                         fl->count++;
3936                         rcu_read_unlock();
3937                         return true;
3938                 }
3939         }
3940         rcu_read_unlock();
3941 #endif
3942         return false;
3943 }
3944 
3945 /*
3946  * enqueue_to_backlog is called to queue an skb to a per CPU backlog
3947  * queue (may be a remote CPU queue).
3948  */
3949 static int enqueue_to_backlog(struct sk_buff *skb, int cpu,
3950                               unsigned int *qtail)
3951 {
3952         struct softnet_data *sd;
3953         unsigned long flags;
3954         unsigned int qlen;
3955 
3956         sd = &per_cpu(softnet_data, cpu);
3957 
3958         local_irq_save(flags);
3959 
3960         rps_lock(sd);
3961         if (!netif_running(skb->dev))
3962                 goto drop;
3963         qlen = skb_queue_len(&sd->input_pkt_queue);
3964         if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) {
3965                 if (qlen) {
3966 enqueue:
3967                         __skb_queue_tail(&sd->input_pkt_queue, skb);
3968                         input_queue_tail_incr_save(sd, qtail);
3969                         rps_unlock(sd);
3970                         local_irq_restore(flags);
3971                         return NET_RX_SUCCESS;
3972                 }
3973 
3974                 /* Schedule NAPI for backlog device
3975                  * We can use non atomic operation since we own the queue lock
3976                  */
3977                 if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) {
3978                         if (!rps_ipi_queued(sd))
3979                                 ____napi_schedule(sd, &sd->backlog);
3980                 }
3981                 goto enqueue;
3982         }
3983 
3984 drop:
3985         sd->dropped++;
3986         rps_unlock(sd);
3987 
3988         local_irq_restore(flags);
3989 
3990         atomic_long_inc(&skb->dev->rx_dropped);
3991         kfree_skb(skb);
3992         return NET_RX_DROP;
3993 }
3994 
3995 static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb)
3996 {
3997         struct net_device *dev = skb->dev;
3998         struct netdev_rx_queue *rxqueue;
3999 
4000         rxqueue = dev->_rx;
4001 
4002         if (skb_rx_queue_recorded(skb)) {
4003                 u16 index = skb_get_rx_queue(skb);
4004 
4005                 if (unlikely(index >= dev->real_num_rx_queues)) {
4006                         WARN_ONCE(dev->real_num_rx_queues > 1,
4007                                   "%s received packet on queue %u, but number "
4008                                   "of RX queues is %u\n",
4009                                   dev->name, index, dev->real_num_rx_queues);
4010 
4011                         return rxqueue; /* Return first rxqueue */
4012                 }
4013                 rxqueue += index;
4014         }
4015         return rxqueue;
4016 }
4017 
4018 static u32 netif_receive_generic_xdp(struct sk_buff *skb,
4019                                      struct xdp_buff *xdp,
4020                                      struct bpf_prog *xdp_prog)
4021 {
4022         struct netdev_rx_queue *rxqueue;
4023         void *orig_data, *orig_data_end;
4024         u32 metalen, act = XDP_DROP;
4025         int hlen, off;
4026         u32 mac_len;
4027 
4028         /* Reinjected packets coming from act_mirred or similar should
4029          * not get XDP generic processing.
4030          */
4031         if (skb_cloned(skb))
4032                 return XDP_PASS;
4033 
4034         /* XDP packets must be linear and must have sufficient headroom
4035          * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also
4036          * native XDP provides, thus we need to do it here as well.
4037          */
4038         if (skb_is_nonlinear(skb) ||
4039             skb_headroom(skb) < XDP_PACKET_HEADROOM) {
4040                 int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
4041                 int troom = skb->tail + skb->data_len - skb->end;
4042 
4043                 /* In case we have to go down the path and also linearize,
4044                  * then lets do the pskb_expand_head() work just once here.
4045                  */
4046                 if (pskb_expand_head(skb,
4047                                      hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0,
4048                                      troom > 0 ? troom + 128 : 0, GFP_ATOMIC))
4049                         goto do_drop;
4050                 if (skb_linearize(skb))
4051                         goto do_drop;
4052         }
4053 
4054         /* The XDP program wants to see the packet starting at the MAC
4055          * header.
4056          */
4057         mac_len = skb->data - skb_mac_header(skb);
4058         hlen = skb_headlen(skb) + mac_len;
4059         xdp->data = skb->data - mac_len;
4060         xdp->data_meta = xdp->data;
4061         xdp->data_end = xdp->data + hlen;
4062         xdp->data_hard_start = skb->data - skb_headroom(skb);
4063         orig_data_end = xdp->data_end;
4064         orig_data = xdp->data;
4065 
4066         rxqueue = netif_get_rxqueue(skb);
4067         xdp->rxq = &rxqueue->xdp_rxq;
4068 
4069         act = bpf_prog_run_xdp(xdp_prog, xdp);
4070 
4071         off = xdp->data - orig_data;
4072         if (off > 0)
4073                 __skb_pull(skb, off);
4074         else if (off < 0)
4075                 __skb_push(skb, -off);
4076         skb->mac_header += off;
4077 
4078         /* check if bpf_xdp_adjust_tail was used. it can only "shrink"
4079          * pckt.
4080          */
4081         off = orig_data_end - xdp->data_end;
4082         if (off != 0) {
4083                 skb_set_tail_pointer(skb, xdp->data_end - xdp->data);
4084                 skb->len -= off;
4085 
4086         }
4087 
4088         switch (act) {
4089         case XDP_REDIRECT:
4090         case XDP_TX:
4091                 __skb_push(skb, mac_len);
4092                 break;
4093         case XDP_PASS:
4094                 metalen = xdp->data - xdp->data_meta;
4095                 if (metalen)
4096                         skb_metadata_set(skb, metalen);
4097                 break;
4098         default:
4099                 bpf_warn_invalid_xdp_action(act);
4100                 /* fall through */
4101         case XDP_ABORTED:
4102                 trace_xdp_exception(skb->dev, xdp_prog, act);
4103                 /* fall through */
4104         case XDP_DROP:
4105         do_drop:
4106                 kfree_skb(skb);
4107                 break;
4108         }
4109 
4110         return act;
4111 }
4112 
4113 /* When doing generic XDP we have to bypass the qdisc layer and the
4114  * network taps in order to match in-driver-XDP behavior.
4115  */
4116 void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog)
4117 {
4118         struct net_device *dev = skb->dev;
4119         struct netdev_queue *txq;
4120         bool free_skb = true;
4121         int cpu, rc;
4122 
4123         txq = netdev_pick_tx(dev, skb, NULL);
4124         cpu = smp_processor_id();
4125         HARD_TX_LOCK(dev, txq, cpu);
4126         if (!netif_xmit_stopped(txq)) {
4127                 rc = netdev_start_xmit(skb, dev, txq, 0);
4128                 if (dev_xmit_complete(rc))
4129                         free_skb = false;
4130         }
4131         HARD_TX_UNLOCK(dev, txq);
4132         if (free_skb) {
4133                 trace_xdp_exception(dev, xdp_prog, XDP_TX);
4134                 kfree_skb(skb);
4135         }
4136 }
4137 EXPORT_SYMBOL_GPL(generic_xdp_tx);
4138 
4139 static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key);
4140 
4141 int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb)
4142 {
4143         if (xdp_prog) {
4144                 struct xdp_buff xdp;
4145                 u32 act;
4146                 int err;
4147 
4148                 act = netif_receive_generic_xdp(skb, &xdp, xdp_prog);
4149                 if (act != XDP_PASS) {
4150                         switch (act) {
4151                         case XDP_REDIRECT:
4152                                 err = xdp_do_generic_redirect(skb->dev, skb,
4153                                                               &xdp, xdp_prog);
4154                                 if (err)
4155                                         goto out_redir;
4156                                 break;
4157                         case XDP_TX:
4158                                 generic_xdp_tx(skb, xdp_prog);
4159                                 break;
4160                         }
4161                         return XDP_DROP;
4162                 }
4163         }
4164         return XDP_PASS;
4165 out_redir:
4166         kfree_skb(skb);
4167         return XDP_DROP;
4168 }
4169 EXPORT_SYMBOL_GPL(do_xdp_generic);
4170 
4171 static int netif_rx_internal(struct sk_buff *skb)
4172 {
4173         int ret;
4174 
4175         net_timestamp_check(netdev_tstamp_prequeue, skb);
4176 
4177         trace_netif_rx(skb);
4178 
4179         if (static_branch_unlikely(&generic_xdp_needed_key)) {
4180                 int ret;
4181 
4182                 preempt_disable();
4183                 rcu_read_lock();
4184                 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4185                 rcu_read_unlock();
4186                 preempt_enable();
4187 
4188                 /* Consider XDP consuming the packet a success from
4189                  * the netdev point of view we do not want to count
4190                  * this as an error.
4191                  */
4192                 if (ret != XDP_PASS)
4193                         return NET_RX_SUCCESS;
4194         }
4195 
4196 #ifdef CONFIG_RPS
4197         if (static_key_false(&rps_needed)) {
4198                 struct rps_dev_flow voidflow, *rflow = &voidflow;
4199                 int cpu;
4200 
4201                 preempt_disable();
4202                 rcu_read_lock();
4203 
4204                 cpu = get_rps_cpu(skb->dev, skb, &rflow);
4205                 if (cpu < 0)
4206                         cpu = smp_processor_id();
4207 
4208                 ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4209 
4210                 rcu_read_unlock();
4211                 preempt_enable();
4212         } else
4213 #endif
4214         {
4215                 unsigned int qtail;
4216 
4217                 ret = enqueue_to_backlog(skb, get_cpu(), &qtail);
4218                 put_cpu();
4219         }
4220         return ret;
4221 }
4222 
4223 /**
4224  *      netif_rx        -       post buffer to the network code
4225  *      @skb: buffer to post
4226  *
4227  *      This function receives a packet from a device driver and queues it for
4228  *      the upper (protocol) levels to process.  It always succeeds. The buffer
4229  *      may be dropped during processing for congestion control or by the
4230  *      protocol layers.
4231  *
4232  *      return values:
4233  *      NET_RX_SUCCESS  (no congestion)
4234  *      NET_RX_DROP     (packet was dropped)
4235  *
4236  */
4237 
4238 int netif_rx(struct sk_buff *skb)
4239 {
4240         trace_netif_rx_entry(skb);
4241 
4242         return netif_rx_internal(skb);
4243 }
4244 EXPORT_SYMBOL(netif_rx);
4245 
4246 int netif_rx_ni(struct sk_buff *skb)
4247 {
4248         int err;
4249 
4250         trace_netif_rx_ni_entry(skb);
4251 
4252         preempt_disable();
4253         err = netif_rx_internal(skb);
4254         if (local_softirq_pending())
4255                 do_softirq();
4256         preempt_enable();
4257 
4258         return err;
4259 }
4260 EXPORT_SYMBOL(netif_rx_ni);
4261 
4262 static __latent_entropy void net_tx_action(struct softirq_action *h)
4263 {
4264         struct softnet_data *sd = this_cpu_ptr(&softnet_data);
4265 
4266         if (sd->completion_queue) {
4267                 struct sk_buff *clist;
4268 
4269                 local_irq_disable();
4270                 clist = sd->completion_queue;
4271                 sd->completion_queue = NULL;
4272                 local_irq_enable();
4273 
4274                 while (clist) {
4275                         struct sk_buff *skb = clist;
4276 
4277                         clist = clist->next;
4278 
4279                         WARN_ON(refcount_read(&skb->users));
4280                         if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED))
4281                                 trace_consume_skb(skb);
4282                         else
4283                                 trace_kfree_skb(skb, net_tx_action);
4284 
4285                         if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
4286                                 __kfree_skb(skb);
4287                         else
4288                                 __kfree_skb_defer(skb);
4289                 }
4290 
4291                 __kfree_skb_flush();
4292         }
4293 
4294         if (sd->output_queue) {
4295                 struct Qdisc *head;
4296 
4297                 local_irq_disable();
4298                 head = sd->output_queue;
4299                 sd->output_queue = NULL;
4300                 sd->output_queue_tailp = &sd->output_queue;
4301                 local_irq_enable();
4302 
4303                 while (head) {
4304                         struct Qdisc *q = head;
4305                         spinlock_t *root_lock = NULL;
4306 
4307                         head = head->next_sched;
4308 
4309                         if (!(q->flags & TCQ_F_NOLOCK)) {
4310                                 root_lock = qdisc_lock(q);
4311                                 spin_lock(root_lock);
4312                         }
4313                         /* We need to make sure head->next_sched is read
4314                          * before clearing __QDISC_STATE_SCHED
4315                          */
4316                         smp_mb__before_atomic();
4317                         clear_bit(__QDISC_STATE_SCHED, &q->state);
4318                         qdisc_run(q);
4319                         if (root_lock)
4320                                 spin_unlock(root_lock);
4321                 }
4322         }
4323 
4324         xfrm_dev_backlog(sd);
4325 }
4326 
4327 #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE)
4328 /* This hook is defined here for ATM LANE */
4329 int (*br_fdb_test_addr_hook)(struct net_device *dev,
4330                              unsigned char *addr) __read_mostly;
4331 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook);
4332 #endif
4333 
4334 static inline struct sk_buff *
4335 sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
4336                    struct net_device *orig_dev)
4337 {
4338 #ifdef CONFIG_NET_CLS_ACT
4339         struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress);
4340         struct tcf_result cl_res;
4341 
4342         /* If there's at least one ingress present somewhere (so
4343          * we get here via enabled static key), remaining devices
4344          * that are not configured with an ingress qdisc will bail
4345          * out here.
4346          */
4347         if (!miniq)
4348                 return skb;
4349 
4350         if (*pt_prev) {
4351                 *ret = deliver_skb(skb, *pt_prev, orig_dev);
4352                 *pt_prev = NULL;
4353         }
4354 
4355         qdisc_skb_cb(skb)->pkt_len = skb->len;
4356         skb->tc_at_ingress = 1;
4357         mini_qdisc_bstats_cpu_update(miniq, skb);
4358 
4359         switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) {
4360         case TC_ACT_OK:
4361         case TC_ACT_RECLASSIFY:
4362                 skb->tc_index = TC_H_MIN(cl_res.classid);
4363                 break;
4364         case TC_ACT_SHOT:
4365                 mini_qdisc_qstats_cpu_drop(miniq);
4366                 kfree_skb(skb);
4367                 return NULL;
4368         case TC_ACT_STOLEN:
4369         case TC_ACT_QUEUED:
4370         case TC_ACT_TRAP:
4371                 consume_skb(skb);
4372                 return NULL;
4373         case TC_ACT_REDIRECT:
4374                 /* skb_mac_header check was done by cls/act_bpf, so
4375                  * we can safely push the L2 header back before
4376                  * redirecting to another netdev
4377                  */
4378                 __skb_push(skb, skb->mac_len);
4379                 skb_do_redirect(skb);
4380                 return NULL;
4381         default:
4382                 break;
4383         }
4384 #endif /* CONFIG_NET_CLS_ACT */
4385         return skb;
4386 }
4387 
4388 /**
4389  *      netdev_is_rx_handler_busy - check if receive handler is registered
4390  *      @dev: device to check
4391  *
4392  *      Check if a receive handler is already registered for a given device.
4393  *      Return true if there one.
4394  *
4395  *      The caller must hold the rtnl_mutex.
4396  */
4397 bool netdev_is_rx_handler_busy(struct net_device *dev)
4398 {
4399         ASSERT_RTNL();
4400         return dev && rtnl_dereference(dev->rx_handler);
4401 }
4402 EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy);
4403 
4404 /**
4405  *      netdev_rx_handler_register - register receive handler
4406  *      @dev: device to register a handler for
4407  *      @rx_handler: receive handler to register
4408  *      @rx_handler_data: data pointer that is used by rx handler
4409  *
4410  *      Register a receive handler for a device. This handler will then be
4411  *      called from __netif_receive_skb. A negative errno code is returned
4412  *      on a failure.
4413  *
4414  *      The caller must hold the rtnl_mutex.
4415  *
4416  *      For a general description of rx_handler, see enum rx_handler_result.
4417  */
4418 int netdev_rx_handler_register(struct net_device *dev,
4419                                rx_handler_func_t *rx_handler,
4420                                void *rx_handler_data)
4421 {
4422         if (netdev_is_rx_handler_busy(dev))
4423                 return -EBUSY;
4424 
4425         if (dev->priv_flags & IFF_NO_RX_HANDLER)
4426                 return -EINVAL;
4427 
4428         /* Note: rx_handler_data must be set before rx_handler */
4429         rcu_assign_pointer(dev->rx_handler_data, rx_handler_data);
4430         rcu_assign_pointer(dev->rx_handler, rx_handler);
4431 
4432         return 0;
4433 }
4434 EXPORT_SYMBOL_GPL(netdev_rx_handler_register);
4435 
4436 /**
4437  *      netdev_rx_handler_unregister - unregister receive handler
4438  *      @dev: device to unregister a handler from
4439  *
4440  *      Unregister a receive handler from a device.
4441  *
4442  *      The caller must hold the rtnl_mutex.
4443  */
4444 void netdev_rx_handler_unregister(struct net_device *dev)
4445 {
4446 
4447         ASSERT_RTNL();
4448         RCU_INIT_POINTER(dev->rx_handler, NULL);
4449         /* a reader seeing a non NULL rx_handler in a rcu_read_lock()
4450          * section has a guarantee to see a non NULL rx_handler_data
4451          * as well.
4452          */
4453         synchronize_net();
4454         RCU_INIT_POINTER(dev->rx_handler_data, NULL);
4455 }
4456 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister);
4457 
4458 /*
4459  * Limit the use of PFMEMALLOC reserves to those protocols that implement
4460  * the special handling of PFMEMALLOC skbs.
4461  */
4462 static bool skb_pfmemalloc_protocol(struct sk_buff *skb)
4463 {
4464         switch (skb->protocol) {
4465         case htons(ETH_P_ARP):
4466         case htons(ETH_P_IP):
4467         case htons(ETH_P_IPV6):
4468         case htons(ETH_P_8021Q):
4469         case htons(ETH_P_8021AD):
4470                 return true;
4471         default:
4472                 return false;
4473         }
4474 }
4475 
4476 static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
4477                              int *ret, struct net_device *orig_dev)
4478 {
4479 #ifdef CONFIG_NETFILTER_INGRESS
4480         if (nf_hook_ingress_active(skb)) {
4481                 int ingress_retval;
4482 
4483                 if (*pt_prev) {
4484                         *ret = deliver_skb(skb, *pt_prev, orig_dev);
4485                         *pt_prev = NULL;
4486                 }
4487 
4488                 rcu_read_lock();
4489                 ingress_retval = nf_hook_ingress(skb);
4490                 rcu_read_unlock();
4491                 return ingress_retval;
4492         }
4493 #endif /* CONFIG_NETFILTER_INGRESS */
4494         return 0;
4495 }
4496 
4497 static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc)
4498 {
4499         struct packet_type *ptype, *pt_prev;
4500         rx_handler_func_t *rx_handler;
4501         struct net_device *orig_dev;
4502         bool deliver_exact = false;
4503         int ret = NET_RX_DROP;
4504         __be16 type;
4505 
4506         net_timestamp_check(!netdev_tstamp_prequeue, skb);
4507 
4508         trace_netif_receive_skb(skb);
4509 
4510         orig_dev = skb->dev;
4511 
4512         skb_reset_network_header(skb);
4513         if (!skb_transport_header_was_set(skb))
4514                 skb_reset_transport_header(skb);
4515         skb_reset_mac_len(skb);
4516 
4517         pt_prev = NULL;
4518 
4519 another_round:
4520         skb->skb_iif = skb->dev->ifindex;
4521 
4522         __this_cpu_inc(softnet_data.processed);
4523 
4524         if (skb->protocol == cpu_to_be16(ETH_P_8021Q) ||
4525             skb->protocol == cpu_to_be16(ETH_P_8021AD)) {
4526                 skb = skb_vlan_untag(skb);
4527                 if (unlikely(!skb))
4528                         goto out;
4529         }
4530 
4531         if (skb_skip_tc_classify(skb))
4532                 goto skip_classify;
4533 
4534         if (pfmemalloc)
4535                 goto skip_taps;
4536 
4537         list_for_each_entry_rcu(ptype, &ptype_all, list) {
4538                 if (pt_prev)
4539                         ret = deliver_skb(skb, pt_prev, orig_dev);
4540                 pt_prev = ptype;
4541         }
4542 
4543         list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) {
4544                 if (pt_prev)
4545                         ret = deliver_skb(skb, pt_prev, orig_dev);
4546                 pt_prev = ptype;
4547         }
4548 
4549 skip_taps:
4550 #ifdef CONFIG_NET_INGRESS
4551         if (static_branch_unlikely(&ingress_needed_key)) {
4552                 skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev);
4553                 if (!skb)
4554                         goto out;
4555 
4556                 if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0)
4557                         goto out;
4558         }
4559 #endif
4560         skb_reset_tc(skb);
4561 skip_classify:
4562         if (pfmemalloc && !skb_pfmemalloc_protocol(skb))
4563                 goto drop;
4564 
4565         if (skb_vlan_tag_present(skb)) {
4566                 if (pt_prev) {
4567                         ret = deliver_skb(skb, pt_prev, orig_dev);
4568                         pt_prev = NULL;
4569                 }
4570                 if (vlan_do_receive(&skb))
4571                         goto another_round;
4572                 else if (unlikely(!skb))
4573                         goto out;
4574         }
4575 
4576         rx_handler = rcu_dereference(skb->dev->rx_handler);
4577         if (rx_handler) {
4578                 if (pt_prev) {
4579                         ret = deliver_skb(skb, pt_prev, orig_dev);
4580                         pt_prev = NULL;
4581                 }
4582                 switch (rx_handler(&skb)) {
4583                 case RX_HANDLER_CONSUMED:
4584                         ret = NET_RX_SUCCESS;
4585                         goto out;
4586                 case RX_HANDLER_ANOTHER:
4587                         goto another_round;
4588                 case RX_HANDLER_EXACT:
4589                         deliver_exact = true;
4590                 case RX_HANDLER_PASS:
4591                         break;
4592                 default:
4593                         BUG();
4594                 }
4595         }
4596 
4597         if (unlikely(skb_vlan_tag_present(skb))) {
4598                 if (skb_vlan_tag_get_id(skb))
4599                         skb->pkt_type = PACKET_OTHERHOST;
4600                 /* Note: we might in the future use prio bits
4601                  * and set skb->priority like in vlan_do_receive()
4602                  * For the time being, just ignore Priority Code Point
4603                  */
4604                 skb->vlan_tci = 0;
4605         }
4606 
4607         type = skb->protocol;
4608 
4609         /* deliver only exact match when indicated */
4610         if (likely(!deliver_exact)) {
4611                 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4612                                        &ptype_base[ntohs(type) &
4613                                                    PTYPE_HASH_MASK]);
4614         }
4615 
4616         deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4617                                &orig_dev->ptype_specific);
4618 
4619         if (unlikely(skb->dev != orig_dev)) {
4620                 deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type,
4621                                        &skb->dev->ptype_specific);
4622         }
4623 
4624         if (pt_prev) {
4625                 if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC)))
4626                         goto drop;
4627                 else
4628                         ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev);
4629         } else {
4630 drop:
4631                 if (!deliver_exact)
4632                         atomic_long_inc(&skb->dev->rx_dropped);
4633                 else
4634                         atomic_long_inc(&skb->dev->rx_nohandler);
4635                 kfree_skb(skb);
4636                 /* Jamal, now you will not able to escape explaining
4637                  * me how you were going to use this. :-)
4638                  */
4639                 ret = NET_RX_DROP;
4640         }
4641 
4642 out:
4643         return ret;
4644 }
4645 
4646 /**
4647  *      netif_receive_skb_core - special purpose version of netif_receive_skb
4648  *      @skb: buffer to process
4649  *
4650  *      More direct receive version of netif_receive_skb().  It should
4651  *      only be used by callers that have a need to skip RPS and Generic XDP.
4652  *      Caller must also take care of handling if (page_is_)pfmemalloc.
4653  *
4654  *      This function may only be called from softirq context and interrupts
4655  *      should be enabled.
4656  *
4657  *      Return values (usually ignored):
4658  *      NET_RX_SUCCESS: no congestion
4659  *      NET_RX_DROP: packet was dropped
4660  */
4661 int netif_receive_skb_core(struct sk_buff *skb)
4662 {
4663         int ret;
4664 
4665         rcu_read_lock();
4666         ret = __netif_receive_skb_core(skb, false);
4667         rcu_read_unlock();
4668 
4669         return ret;
4670 }
4671 EXPORT_SYMBOL(netif_receive_skb_core);
4672 
4673 static int __netif_receive_skb(struct sk_buff *skb)
4674 {
4675         int ret;
4676 
4677         if (sk_memalloc_socks() && skb_pfmemalloc(skb)) {
4678                 unsigned int noreclaim_flag;
4679 
4680                 /*
4681                  * PFMEMALLOC skbs are special, they should
4682                  * - be delivered to SOCK_MEMALLOC sockets only
4683                  * - stay away from userspace
4684                  * - have bounded memory usage
4685                  *
4686                  * Use PF_MEMALLOC as this saves us from propagating the allocation
4687                  * context down to all allocation sites.
4688                  */
4689                 noreclaim_flag = memalloc_noreclaim_save();
4690                 ret = __netif_receive_skb_core(skb, true);
4691                 memalloc_noreclaim_restore(noreclaim_flag);
4692         } else
4693                 ret = __netif_receive_skb_core(skb, false);
4694 
4695         return ret;
4696 }
4697 
4698 static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp)
4699 {
4700         struct bpf_prog *old = rtnl_dereference(dev->xdp_prog);
4701         struct bpf_prog *new = xdp->prog;
4702         int ret = 0;
4703 
4704         switch (xdp->command) {
4705         case XDP_SETUP_PROG:
4706                 rcu_assign_pointer(dev->xdp_prog, new);
4707                 if (old)
4708                         bpf_prog_put(old);
4709 
4710                 if (old && !new) {
4711                         static_branch_dec(&generic_xdp_needed_key);
4712                 } else if (new && !old) {
4713                         static_branch_inc(&generic_xdp_needed_key);
4714                         dev_disable_lro(dev);
4715                         dev_disable_gro_hw(dev);
4716                 }
4717                 break;
4718 
4719         case XDP_QUERY_PROG:
4720                 xdp->prog_attached = !!old;
4721                 xdp->prog_id = old ? old->aux->id : 0;
4722                 break;
4723 
4724         default:
4725                 ret = -EINVAL;
4726                 break;
4727         }
4728 
4729         return ret;
4730 }
4731 
4732 static int netif_receive_skb_internal(struct sk_buff *skb)
4733 {
4734         int ret;
4735 
4736         net_timestamp_check(netdev_tstamp_prequeue, skb);
4737 
4738         if (skb_defer_rx_timestamp(skb))
4739                 return NET_RX_SUCCESS;
4740 
4741         if (static_branch_unlikely(&generic_xdp_needed_key)) {
4742                 int ret;
4743 
4744                 preempt_disable();
4745                 rcu_read_lock();
4746                 ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
4747                 rcu_read_unlock();
4748                 preempt_enable();
4749 
4750                 if (ret != XDP_PASS)
4751                         return NET_RX_DROP;
4752         }
4753 
4754         rcu_read_lock();
4755 #ifdef CONFIG_RPS
4756         if (static_key_false(&rps_needed)) {
4757                 struct rps_dev_flow voidflow, *rflow = &voidflow;
4758                 int cpu = get_rps_cpu(skb->dev, skb, &rflow);
4759 
4760                 if (cpu >= 0) {
4761                         ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail);
4762                         rcu_read_unlock();
4763                         return ret;
4764                 }
4765         }
4766 #endif
4767         ret = __netif_receive_skb(skb);
4768         rcu_read_unlock();
4769         return ret;
4770 }
4771 
4772 /**
4773  *      netif_receive_skb - process receive buffer from network
4774  *      @skb: buffer to process
4775  *
4776  *      netif_receive_skb() is the main receive data processing function.
4777  *      It always succeeds. The buffer may be dropped during processing
4778  *      for congestion control or by the protocol layers.
4779  *
4780  *      This function may only be called from softirq context and interrupts
4781  *      should be enabled.
4782  *
4783  *      Return values (usually ignored):
4784  *      NET_RX_SUCCESS: no congestion
4785  *      NET_RX_DROP: packet was dropped
4786  */
4787 int netif_receive_skb(struct sk_buff *skb)
4788 {
4789         trace_netif_receive_skb_entry(skb);
4790 
4791         return netif_receive_skb_internal(skb);
4792 }
4793 EXPORT_SYMBOL(netif_receive_skb);
4794 
4795 DEFINE_PER_CPU(struct work_struct, flush_works);
4796 
4797 /* Network device is going away, flush any packets still pending */
4798 static void flush_backlog(struct work_struct *work)
4799 {
4800         struct sk_buff *skb, *tmp;
4801         struct softnet_data *sd;
4802 
4803         local_bh_disable();
4804         sd = this_cpu_ptr(&softnet_data);
4805 
4806         local_irq_disable();
4807         rps_lock(sd);
4808         skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) {
4809                 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4810                         __skb_unlink(skb, &sd->input_pkt_queue);
4811                         kfree_skb(skb);
4812                         input_queue_head_incr(sd);
4813                 }
4814         }
4815         rps_unlock(sd);
4816         local_irq_enable();
4817 
4818         skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
4819                 if (skb->dev->reg_state == NETREG_UNREGISTERING) {
4820                         __skb_unlink(skb, &sd->process_queue);
4821                         kfree_skb(skb);
4822                         input_queue_head_incr(sd);
4823                 }
4824         }
4825         local_bh_enable();
4826 }
4827 
4828 static void flush_all_backlogs(void)
4829 {
4830         unsigned int cpu;
4831 
4832         get_online_cpus();
4833 
4834         for_each_online_cpu(cpu)
4835                 queue_work_on(cpu, system_highpri_wq,
4836                               per_cpu_ptr(&flush_works, cpu));
4837 
4838         for_each_online_cpu(cpu)
4839                 flush_work(per_cpu_ptr(&flush_works, cpu));
4840 
4841         put_online_cpus();
4842 }
4843 
4844 static int napi_gro_complete(struct sk_buff *skb)
4845 {
4846         struct packet_offload *ptype;
4847         __be16 type = skb->protocol;
4848         struct list_head *head = &offload_base;
4849         int err = -ENOENT;
4850 
4851         BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
4852 
4853         if (NAPI_GRO_CB(skb)->count == 1) {
4854                 skb_shinfo(skb)->gso_size = 0;
4855                 goto out;
4856         }
4857 
4858         rcu_read_lock();
4859         list_for_each_entry_rcu(ptype, head, list) {
4860                 if (ptype->type != type || !ptype->callbacks.gro_complete)
4861                         continue;
4862 
4863                 err = ptype->callbacks.gro_complete(skb, 0);
4864                 break;
4865         }
4866         rcu_read_unlock();
4867 
4868         if (err) {
4869                 WARN_ON(&ptype->list == head);
4870                 kfree_skb(skb);
4871                 return NET_RX_SUCCESS;
4872         }
4873 
4874 out:
4875         return netif_receive_skb_internal(skb);
4876 }
4877 
4878 /* napi->gro_list contains packets ordered by age.
4879  * youngest packets at the head of it.
4880  * Complete skbs in reverse order to reduce latencies.
4881  */
4882 void napi_gro_flush(struct napi_struct *napi, bool flush_old)
4883 {
4884         struct sk_buff *skb, *prev = NULL;
4885 
4886         /* scan list and build reverse chain */
4887         for (skb = napi->gro_list; skb != NULL; skb = skb->next) {
4888                 skb->prev = prev;
4889                 prev = skb;
4890         }
4891 
4892         for (skb = prev; skb; skb = prev) {
4893                 skb->next = NULL;
4894 
4895                 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
4896                         return;
4897 
4898                 prev = skb->prev;
4899                 napi_gro_complete(skb);
4900                 napi->gro_count--;
4901         }
4902 
4903         napi->gro_list = NULL;
4904 }
4905 EXPORT_SYMBOL(napi_gro_flush);
4906 
4907 static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb)
4908 {
4909         struct sk_buff *p;
4910         unsigned int maclen = skb->dev->hard_header_len;
4911         u32 hash = skb_get_hash_raw(skb);
4912 
4913         for (p = napi->gro_list; p; p = p->next) {
4914                 unsigned long diffs;
4915 
4916                 NAPI_GRO_CB(p)->flush = 0;
4917 
4918                 if (hash != skb_get_hash_raw(p)) {
4919                         NAPI_GRO_CB(p)->same_flow = 0;
4920                         continue;
4921                 }
4922 
4923                 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
4924                 diffs |= p->vlan_tci ^ skb->vlan_tci;
4925                 diffs |= skb_metadata_dst_cmp(p, skb);
4926                 diffs |= skb_metadata_differs(p, skb);
4927                 if (maclen == ETH_HLEN)
4928                         diffs |= compare_ether_header(skb_mac_header(p),
4929                                                       skb_mac_header(skb));
4930                 else if (!diffs)
4931                         diffs = memcmp(skb_mac_header(p),
4932                                        skb_mac_header(skb),
4933                                        maclen);
4934                 NAPI_GRO_CB(p)->same_flow = !diffs;
4935         }
4936 }
4937 
4938 static void skb_gro_reset_offset(struct sk_buff *skb)
4939 {
4940         const struct skb_shared_info *pinfo = skb_shinfo(skb);
4941         const skb_frag_t *frag0 = &pinfo->frags[0];
4942 
4943         NAPI_GRO_CB(skb)->data_offset = 0;
4944         NAPI_GRO_CB(skb)->frag0 = NULL;
4945         NAPI_GRO_CB(skb)->frag0_len = 0;
4946 
4947         if (skb_mac_header(skb) == skb_tail_pointer(skb) &&
4948             pinfo->nr_frags &&
4949             !PageHighMem(skb_frag_page(frag0))) {
4950                 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
4951                 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
4952                                                     skb_frag_size(frag0),
4953                                                     skb->end - skb->tail);
4954         }
4955 }
4956 
4957 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
4958 {
4959         struct skb_shared_info *pinfo = skb_shinfo(skb);
4960 
4961         BUG_ON(skb->end - skb->tail < grow);
4962 
4963         memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
4964 
4965         skb->data_len -= grow;
4966         skb->tail += grow;
4967 
4968         pinfo->frags[0].page_offset += grow;
4969         skb_frag_size_sub(&pinfo->frags[0], grow);
4970 
4971         if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
4972                 skb_frag_unref(skb, 0);
4973                 memmove(pinfo->frags, pinfo->frags + 1,
4974                         --pinfo->nr_frags * sizeof(pinfo->frags[0]));
4975         }
4976 }
4977 
4978 static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
4979 {
4980         struct sk_buff **pp = NULL;
4981         struct packet_offload *ptype;
4982         __be16 type = skb->protocol;
4983         struct list_head *head = &offload_base;
4984         int same_flow;
4985         enum gro_result ret;
4986         int grow;
4987 
4988         if (netif_elide_gro(skb->dev))
4989                 goto normal;
4990 
4991         gro_list_prepare(napi, skb);
4992 
4993         rcu_read_lock();
4994         list_for_each_entry_rcu(ptype, head, list) {
4995                 if (ptype->type != type || !ptype->callbacks.gro_receive)
4996                         continue;
4997 
4998                 skb_set_network_header(skb, skb_gro_offset(skb));
4999                 skb_reset_mac_len(skb);
5000                 NAPI_GRO_CB(skb)->same_flow = 0;
5001                 NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb);
5002                 NAPI_GRO_CB(skb)->free = 0;
5003                 NAPI_GRO_CB(skb)->encap_mark = 0;
5004                 NAPI_GRO_CB(skb)->recursion_counter = 0;
5005                 NAPI_GRO_CB(skb)->is_fou = 0;
5006                 NAPI_GRO_CB(skb)->is_atomic = 1;
5007                 NAPI_GRO_CB(skb)->gro_remcsum_start = 0;
5008 
5009                 /* Setup for GRO checksum validation */
5010                 switch (skb->ip_summed) {
5011                 case CHECKSUM_COMPLETE:
5012                         NAPI_GRO_CB(skb)->csum = skb->csum;
5013                         NAPI_GRO_CB(skb)->csum_valid = 1;
5014                         NAPI_GRO_CB(skb)->csum_cnt = 0;
5015                         break;
5016                 case CHECKSUM_UNNECESSARY:
5017                         NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
5018                         NAPI_GRO_CB(skb)->csum_valid = 0;
5019                         break;
5020                 default:
5021                         NAPI_GRO_CB(skb)->csum_cnt = 0;
5022                         NAPI_GRO_CB(skb)->csum_valid = 0;
5023                 }
5024 
5025                 pp = ptype->callbacks.gro_receive(&napi->gro_list, skb);
5026                 break;
5027         }
5028         rcu_read_unlock();
5029 
5030         if (&ptype->list == head)
5031                 goto normal;
5032 
5033         if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
5034                 ret = GRO_CONSUMED;
5035                 goto ok;
5036         }
5037 
5038         same_flow = NAPI_GRO_CB(skb)->same_flow;
5039         ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
5040 
5041         if (pp) {
5042                 struct sk_buff *nskb = *pp;
5043 
5044                 *pp = nskb->next;
5045                 nskb->next = NULL;
5046                 napi_gro_complete(nskb);
5047                 napi->gro_count--;
5048         }
5049 
5050         if (same_flow)
5051                 goto ok;
5052 
5053         if (NAPI_GRO_CB(skb)->flush)
5054                 goto normal;
5055 
5056         if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) {
5057                 struct sk_buff *nskb = napi->gro_list;
5058 
5059                 /* locate the end of the list to select the 'oldest' flow */
5060                 while (nskb->next) {
5061                         pp = &nskb->next;
5062                         nskb = *pp;
5063                 }
5064                 *pp = NULL;
5065                 nskb->next = NULL;
5066                 napi_gro_complete(nskb);
5067         } else {
5068                 napi->gro_count++;
5069         }
5070         NAPI_GRO_CB(skb)->count = 1;
5071         NAPI_GRO_CB(skb)->age = jiffies;
5072         NAPI_GRO_CB(skb)->last = skb;
5073         skb_shinfo(skb)->gso_size = skb_gro_len(skb);
5074         skb->next = napi->gro_list;
5075         napi->gro_list = skb;
5076         ret = GRO_HELD;
5077 
5078 pull:
5079         grow = skb_gro_offset(skb) - skb_headlen(skb);
5080         if (grow > 0)
5081                 gro_pull_from_frag0(skb, grow);
5082 ok:
5083         return ret;
5084 
5085 normal:
5086         ret = GRO_NORMAL;
5087         goto pull;
5088 }
5089 
5090 struct packet_offload *gro_find_receive_by_type(__be16 type)
5091 {
5092         struct list_head *offload_head = &offload_base;
5093         struct packet_offload *ptype;
5094 
5095         list_for_each_entry_rcu(ptype, offload_head, list) {
5096                 if (ptype->type != type || !ptype->callbacks.gro_receive)
5097                         continue;
5098                 return ptype;
5099         }
5100         return NULL;
5101 }
5102 EXPORT_SYMBOL(gro_find_receive_by_type);
5103 
5104 struct packet_offload *gro_find_complete_by_type(__be16 type)
5105 {
5106         struct list_head *offload_head = &offload_base;
5107         struct packet_offload *ptype;
5108 
5109         list_for_each_entry_rcu(ptype, offload_head, list) {
5110                 if (ptype->type != type || !ptype->callbacks.gro_complete)
5111                         continue;
5112                 return ptype;
5113         }
5114         return NULL;
5115 }
5116 EXPORT_SYMBOL(gro_find_complete_by_type);
5117 
5118 static void napi_skb_free_stolen_head(struct sk_buff *skb)
5119 {
5120         skb_dst_drop(skb);
5121         secpath_reset(skb);
5122         kmem_cache_free(skbuff_head_cache, skb);
5123 }
5124 
5125 static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
5126 {
5127         switch (ret) {
5128         case GRO_NORMAL:
5129                 if (netif_receive_skb_internal(skb))
5130                         ret = GRO_DROP;
5131                 break;
5132 
5133         case GRO_DROP:
5134                 kfree_skb(skb);
5135                 break;
5136 
5137         case GRO_MERGED_FREE:
5138                 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5139                         napi_skb_free_stolen_head(skb);
5140                 else
5141                         __kfree_skb(skb);
5142                 break;
5143 
5144         case GRO_HELD:
5145         case GRO_MERGED:
5146         case GRO_CONSUMED:
5147                 break;
5148         }
5149 
5150         return ret;
5151 }
5152 
5153 gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb)
5154 {
5155         skb_mark_napi_id(skb, napi);
5156         trace_napi_gro_receive_entry(skb);
5157 
5158         skb_gro_reset_offset(skb);
5159 
5160         return napi_skb_finish(dev_gro_receive(napi, skb), skb);
5161 }
5162 EXPORT_SYMBOL(napi_gro_receive);
5163 
5164 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
5165 {
5166         if (unlikely(skb->pfmemalloc)) {
5167                 consume_skb(skb);
5168                 return;
5169         }
5170         __skb_pull(skb, skb_headlen(skb));
5171         /* restore the reserve we had after netdev_alloc_skb_ip_align() */
5172         skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
5173         skb->vlan_tci = 0;
5174         skb->dev = napi->dev;
5175         skb->skb_iif = 0;
5176         skb->encapsulation = 0;
5177         skb_shinfo(skb)->gso_type = 0;
5178         skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5179         secpath_reset(skb);
5180 
5181         napi->skb = skb;
5182 }
5183 
5184 struct sk_buff *napi_get_frags(struct napi_struct *napi)
5185 {
5186         struct sk_buff *skb = napi->skb;
5187 
5188         if (!skb) {
5189                 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
5190                 if (skb) {
5191                         napi->skb = skb;
5192                         skb_mark_napi_id(skb, napi);
5193                 }
5194         }
5195         return skb;
5196 }
5197 EXPORT_SYMBOL(napi_get_frags);
5198 
5199 static gro_result_t napi_frags_finish(struct napi_struct *napi,
5200                                       struct sk_buff *skb,
5201                                       gro_result_t ret)
5202 {
5203         switch (ret) {
5204         case GRO_NORMAL:
5205         case GRO_HELD:
5206                 __skb_push(skb, ETH_HLEN);
5207                 skb->protocol = eth_type_trans(skb, skb->dev);
5208                 if (ret == GRO_NORMAL && netif_receive_skb_internal(skb))
5209                         ret = GRO_DROP;
5210                 break;
5211 
5212         case GRO_DROP:
5213                 napi_reuse_skb(napi, skb);
5214                 break;
5215 
5216         case GRO_MERGED_FREE:
5217                 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
5218                         napi_skb_free_stolen_head(skb);
5219                 else
5220                         napi_reuse_skb(napi, skb);
5221                 break;
5222 
5223         case GRO_MERGED:
5224         case GRO_CONSUMED:
5225                 break;
5226         }
5227 
5228         return ret;
5229 }
5230 
5231 /* Upper GRO stack assumes network header starts at gro_offset=0
5232  * Drivers could call both napi_gro_frags() and napi_gro_receive()
5233  * We copy ethernet header into skb->data to have a common layout.
5234  */
5235 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
5236 {
5237         struct sk_buff *skb = napi->skb;
5238         const struct ethhdr *eth;
5239         unsigned int hlen = sizeof(*eth);
5240 
5241         napi->skb = NULL;
5242 
5243         skb_reset_mac_header(skb);
5244         skb_gro_reset_offset(skb);
5245 
5246         eth = skb_gro_header_fast(skb, 0);
5247         if (unlikely(skb_gro_header_hard(skb, hlen))) {
5248                 eth = skb_gro_header_slow(skb, hlen, 0);
5249                 if (unlikely(!eth)) {
5250                         net_warn_ratelimited("%s: dropping impossible skb from %s\n",
5251                                              __func__, napi->dev->name);
5252                         napi_reuse_skb(napi, skb);
5253                         return NULL;
5254                 }
5255         } else {
5256                 gro_pull_from_frag0(skb, hlen);
5257                 NAPI_GRO_CB(skb)->frag0 += hlen;
5258                 NAPI_GRO_CB(skb)->frag0_len -= hlen;
5259         }
5260         __skb_pull(skb, hlen);
5261 
5262         /*
5263          * This works because the only protocols we care about don't require
5264          * special handling.
5265          * We'll fix it up properly in napi_frags_finish()
5266          */
5267         skb->protocol = eth->h_proto;
5268 
5269         return skb;
5270 }
5271 
5272 gro_result_t napi_gro_frags(struct napi_struct *napi)
5273 {
5274         struct sk_buff *skb = napi_frags_skb(napi);
5275 
5276         if (!skb)
5277                 return GRO_DROP;
5278 
5279         trace_napi_gro_frags_entry(skb);
5280 
5281         return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb));
5282 }
5283 EXPORT_SYMBOL(napi_gro_frags);
5284 
5285 /* Compute the checksum from gro_offset and return the folded value
5286  * after adding in any pseudo checksum.
5287  */
5288 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
5289 {
5290         __wsum wsum;
5291         __sum16 sum;
5292 
5293         wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
5294 
5295         /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
5296         sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
5297         if (likely(!sum)) {
5298                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
5299                     !skb->csum_complete_sw)
5300                         netdev_rx_csum_fault(skb->dev);
5301         }
5302 
5303         NAPI_GRO_CB(skb)->csum = wsum;
5304         NAPI_GRO_CB(skb)->csum_valid = 1;
5305 
5306         return sum;
5307 }
5308 EXPORT_SYMBOL(__skb_gro_checksum_complete);
5309 
5310 static void net_rps_send_ipi(struct softnet_data *remsd)
5311 {
5312 #ifdef CONFIG_RPS
5313         while (remsd) {
5314                 struct softnet_data *next = remsd->rps_ipi_next;
5315 
5316                 if (cpu_online(remsd->cpu))
5317                         smp_call_function_single_async(remsd->cpu, &remsd->csd);
5318                 remsd = next;
5319         }
5320 #endif
5321 }
5322 
5323 /*
5324  * net_rps_action_and_irq_enable sends any pending IPI's for rps.
5325  * Note: called with local irq disabled, but exits with local irq enabled.
5326  */
5327 static void net_rps_action_and_irq_enable(struct softnet_data *sd)
5328 {
5329 #ifdef CONFIG_RPS
5330         struct softnet_data *remsd = sd->rps_ipi_list;
5331 
5332         if (remsd) {
5333                 sd->rps_ipi_list = NULL;
5334 
5335                 local_irq_enable();
5336 
5337                 /* Send pending IPI's to kick RPS processing on remote cpus. */
5338                 net_rps_send_ipi(remsd);
5339         } else
5340 #endif
5341                 local_irq_enable();
5342 }
5343 
5344 static bool sd_has_rps_ipi_waiting(struct softnet_data *sd)
5345 {
5346 #ifdef CONFIG_RPS
5347         return sd->rps_ipi_list != NULL;
5348 #else
5349         return false;
5350 #endif
5351 }
5352 
5353 static int process_backlog(struct napi_struct *napi, int quota)
5354 {
5355         struct softnet_data *sd = container_of(napi, struct softnet_data, backlog);
5356         bool again = true;
5357         int work = 0;
5358 
5359         /* Check if we have pending ipi, its better to send them now,
5360          * not waiting net_rx_action() end.
5361          */
5362         if (sd_has_rps_ipi_waiting(sd)) {
5363                 local_irq_disable();
5364                 net_rps_action_and_irq_enable(sd);
5365         }
5366 
5367         napi->weight = dev_rx_weight;
5368         while (again) {
5369                 struct sk_buff *skb;
5370 
5371                 while ((skb = __skb_dequeue(&sd->process_queue))) {
5372                         rcu_read_lock();
5373                         __netif_receive_skb(skb);
5374                         rcu_read_unlock();
5375                         input_queue_head_incr(sd);
5376                         if (++work >= quota)
5377                                 return work;
5378 
5379                 }
5380 
5381                 local_irq_disable();
5382                 rps_lock(sd);
5383                 if (skb_queue_empty(&sd->input_pkt_queue)) {
5384                         /*
5385                          * Inline a custom version of __napi_complete().
5386                          * only current cpu owns and manipulates this napi,
5387                          * and NAPI_STATE_SCHED is the only possible flag set
5388                          * on backlog.
5389                          * We can use a plain write instead of clear_bit(),
5390                          * and we dont need an smp_mb() memory barrier.
5391                          */
5392                         napi->state = 0;
5393                         again = false;
5394                 } else {
5395                         skb_queue_splice_tail_init(&sd->input_pkt_queue,
5396                                                    &sd->process_queue);
5397                 }
5398                 rps_unlock(sd);
5399                 local_irq_enable();
5400         }
5401 
5402         return work;
5403 }
5404 
5405 /**
5406  * __napi_schedule - schedule for receive
5407  * @n: entry to schedule
5408  *
5409  * The entry's receive function will be scheduled to run.
5410  * Consider using __napi_schedule_irqoff() if hard irqs are masked.
5411  */
5412 void __napi_schedule(struct napi_struct *n)
5413 {
5414         unsigned long flags;
5415 
5416         local_irq_save(flags);
5417         ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5418         local_irq_restore(flags);
5419 }
5420 EXPORT_SYMBOL(__napi_schedule);
5421 
5422 /**
5423  *      napi_schedule_prep - check if napi can be scheduled
5424  *      @n: napi context
5425  *
5426  * Test if NAPI routine is already running, and if not mark
5427  * it as running.  This is used as a condition variable
5428  * insure only one NAPI poll instance runs.  We also make
5429  * sure there is no pending NAPI disable.
5430  */
5431 bool napi_schedule_prep(struct napi_struct *n)
5432 {
5433         unsigned long val, new;
5434 
5435         do {
5436                 val = READ_ONCE(n->state);
5437                 if (unlikely(val & NAPIF_STATE_DISABLE))
5438                         return false;
5439                 new = val | NAPIF_STATE_SCHED;
5440 
5441                 /* Sets STATE_MISSED bit if STATE_SCHED was already set
5442                  * This was suggested by Alexander Duyck, as compiler
5443                  * emits better code than :
5444                  * if (val & NAPIF_STATE_SCHED)
5445                  *     new |= NAPIF_STATE_MISSED;
5446                  */
5447                 new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED *
5448                                                    NAPIF_STATE_MISSED;
5449         } while (cmpxchg(&n->state, val, new) != val);
5450 
5451         return !(val & NAPIF_STATE_SCHED);
5452 }
5453 EXPORT_SYMBOL(napi_schedule_prep);
5454 
5455 /**
5456  * __napi_schedule_irqoff - schedule for receive
5457  * @n: entry to schedule
5458  *
5459  * Variant of __napi_schedule() assuming hard irqs are masked
5460  */
5461 void __napi_schedule_irqoff(struct napi_struct *n)
5462 {
5463         ____napi_schedule(this_cpu_ptr(&softnet_data), n);
5464 }
5465 EXPORT_SYMBOL(__napi_schedule_irqoff);
5466 
5467 bool napi_complete_done(struct napi_struct *n, int work_done)
5468 {
5469         unsigned long flags, val, new;
5470 
5471         /*
5472          * 1) Don't let napi dequeue from the cpu poll list
5473          *    just in case its running on a different cpu.
5474          * 2) If we are busy polling, do nothing here, we have
5475          *    the guarantee we will be called later.
5476          */
5477         if (unlikely(n->state & (NAPIF_STATE_NPSVC |
5478                                  NAPIF_STATE_IN_BUSY_POLL)))
5479                 return false;
5480 
5481         if (n->gro_list) {
5482                 unsigned long timeout = 0;
5483 
5484                 if (work_done)
5485                         timeout = n->dev->gro_flush_timeout;
5486 
5487                 if (timeout)
5488                         hrtimer_start(&n->timer, ns_to_ktime(timeout),
5489                                       HRTIMER_MODE_REL_PINNED);
5490                 else
5491                         napi_gro_flush(n, false);
5492         }
5493         if (unlikely(!list_empty(&n->poll_list))) {
5494                 /* If n->poll_list is not empty, we need to mask irqs */
5495                 local_irq_save(flags);
5496                 list_del_init(&n->poll_list);
5497                 local_irq_restore(flags);
5498         }
5499 
5500         do {
5501                 val = READ_ONCE(n->state);
5502 
5503                 WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED));
5504 
5505                 new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED);
5506 
5507                 /* If STATE_MISSED was set, leave STATE_SCHED set,
5508                  * because we will call napi->poll() one more time.
5509                  * This C code was suggested by Alexander Duyck to help gcc.
5510                  */
5511                 new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED *
5512                                                     NAPIF_STATE_SCHED;
5513         } while (cmpxchg(&n->state, val, new) != val);
5514 
5515         if (unlikely(val & NAPIF_STATE_MISSED)) {
5516                 __napi_schedule(n);
5517                 return false;
5518         }
5519 
5520         return true;
5521 }
5522 EXPORT_SYMBOL(napi_complete_done);
5523 
5524 /* must be called under rcu_read_lock(), as we dont take a reference */
5525 static struct napi_struct *napi_by_id(unsigned int napi_id)
5526 {
5527         unsigned int hash = napi_id % HASH_SIZE(napi_hash);
5528         struct napi_struct *napi;
5529 
5530         hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node)
5531                 if (napi->napi_id == napi_id)
5532                         return napi;
5533 
5534         return NULL;
5535 }
5536 
5537 #if defined(CONFIG_NET_RX_BUSY_POLL)
5538 
5539 #define BUSY_POLL_BUDGET 8
5540 
5541 static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock)
5542 {
5543         int rc;
5544 
5545         /* Busy polling means there is a high chance device driver hard irq
5546          * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was
5547          * set in napi_schedule_prep().
5548          * Since we are about to call napi->poll() once more, we can safely
5549          * clear NAPI_STATE_MISSED.
5550          *
5551          * Note: x86 could use a single "lock and ..." instruction
5552          * to perform these two clear_bit()
5553          */
5554         clear_bit(NAPI_STATE_MISSED, &napi->state);
5555         clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state);
5556 
5557         local_bh_disable();
5558 
5559         /* All we really want here is to re-enable device interrupts.
5560          * Ideally, a new ndo_busy_poll_stop() could avoid another round.
5561          */
5562         rc = napi->poll(napi, BUSY_POLL_BUDGET);
5563         trace_napi_poll(napi, rc, BUSY_POLL_BUDGET);
5564         netpoll_poll_unlock(have_poll_lock);
5565         if (rc == BUSY_POLL_BUDGET)
5566                 __napi_schedule(napi);
5567         local_bh_enable();
5568 }
5569 
5570 void napi_busy_loop(unsigned int napi_id,
5571                     bool (*loop_end)(void *, unsigned long),
5572                     void *loop_end_arg)
5573 {
5574         unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
5575         int (*napi_poll)(struct napi_struct *napi, int budget);
5576         void *have_poll_lock = NULL;
5577         struct napi_struct *napi;
5578 
5579 restart:
5580         napi_poll = NULL;
5581 
5582         rcu_read_lock();
5583 
5584         napi = napi_by_id(napi_id);
5585         if (!napi)
5586                 goto out;
5587 
5588         preempt_disable();
5589         for (;;) {
5590                 int work = 0;
5591 
5592                 local_bh_disable();
5593                 if (!napi_poll) {
5594                         unsigned long val = READ_ONCE(napi->state);
5595 
5596                         /* If multiple threads are competing for this napi,
5597                          * we avoid dirtying napi->state as much as we can.
5598                          */
5599                         if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED |
5600                                    NAPIF_STATE_IN_BUSY_POLL))
5601                                 goto count;
5602                         if (cmpxchg(&napi->state, val,
5603                                     val | NAPIF_STATE_IN_BUSY_POLL |
5604                                           NAPIF_STATE_SCHED) != val)
5605                                 goto count;
5606                         have_poll_lock = netpoll_poll_lock(napi);
5607                         napi_poll = napi->poll;
5608                 }
5609                 work = napi_poll(napi, BUSY_POLL_BUDGET);
5610                 trace_napi_poll(napi, work, BUSY_POLL_BUDGET);
5611 count:
5612                 if (work > 0)
5613                         __NET_ADD_STATS(dev_net(napi->dev),
5614                                         LINUX_MIB_BUSYPOLLRXPACKETS, work);
5615                 local_bh_enable();
5616 
5617                 if (!loop_end || loop_end(loop_end_arg, start_time))
5618                         break;
5619 
5620                 if (unlikely(need_resched())) {
5621                         if (napi_poll)
5622                                 busy_poll_stop(napi, have_poll_lock);
5623                         preempt_enable();
5624                         rcu_read_unlock();
5625                         cond_resched();
5626                         if (loop_end(loop_end_arg, start_time))
5627                                 return;
5628                         goto restart;
5629                 }
5630                 cpu_relax();
5631         }
5632         if (napi_poll)
5633                 busy_poll_stop(napi, have_poll_lock);
5634         preempt_enable();
5635 out:
5636         rcu_read_unlock();
5637 }
5638 EXPORT_SYMBOL(napi_busy_loop);
5639 
5640 #endif /* CONFIG_NET_RX_BUSY_POLL */
5641 
5642 static void napi_hash_add(struct napi_struct *napi)
5643 {
5644         if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) ||
5645             test_and_set_bit(NAPI_STATE_HASHED, &napi->state))
5646                 return;
5647 
5648         spin_lock(&napi_hash_lock);
5649 
5650         /* 0..NR_CPUS range is reserved for sender_cpu use */
5651         do {
5652                 if (unlikely(++napi_gen_id < MIN_NAPI_ID))
5653                         napi_gen_id = MIN_NAPI_ID;
5654         } while (napi_by_id(napi_gen_id));
5655         napi->napi_id = napi_gen_id;
5656 
5657         hlist_add_head_rcu(&napi->napi_hash_node,
5658                            &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]);
5659 
5660         spin_unlock(&napi_hash_lock);
5661 }
5662 
5663 /* Warning : caller is responsible to make sure rcu grace period
5664  * is respected before freeing memory containing @napi
5665  */
5666 bool napi_hash_del(struct napi_struct *napi)
5667 {
5668         bool rcu_sync_needed = false;
5669 
5670         spin_lock(&napi_hash_lock);
5671 
5672         if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) {
5673                 rcu_sync_needed = true;
5674                 hlist_del_rcu(&napi->napi_hash_node);
5675         }
5676         spin_unlock(&napi_hash_lock);
5677         return rcu_sync_needed;
5678 }
5679 EXPORT_SYMBOL_GPL(napi_hash_del);
5680 
5681 static enum hrtimer_restart napi_watchdog(struct hrtimer *timer)
5682 {
5683         struct napi_struct *napi;
5684 
5685         napi = container_of(timer, struct napi_struct, timer);
5686 
5687         /* Note : we use a relaxed variant of napi_schedule_prep() not setting
5688          * NAPI_STATE_MISSED, since we do not react to a device IRQ.
5689          */
5690         if (napi->gro_list && !napi_disable_pending(napi) &&
5691             !test_and_set_bit(NAPI_STATE_SCHED, &napi->state))
5692                 __napi_schedule_irqoff(napi);
5693 
5694         return HRTIMER_NORESTART;
5695 }
5696 
5697 void netif_napi_add(struct net_device *dev, struct napi_struct *napi,
5698                     int (*poll)(struct napi_struct *, int), int weight)
5699 {
5700         INIT_LIST_HEAD(&napi->poll_list);
5701         hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED);
5702         napi->timer.function = napi_watchdog;
5703         napi->gro_count = 0;
5704         napi->gro_list = NULL;
5705         napi->skb = NULL;
5706         napi->poll = poll;
5707         if (weight > NAPI_POLL_WEIGHT)
5708                 pr_err_once("netif_napi_add() called with weight %d on device %s\n",
5709                             weight, dev->name);
5710         napi->weight = weight;
5711         list_add(&napi->dev_list, &dev->napi_list);
5712         napi->dev = dev;
5713 #ifdef CONFIG_NETPOLL
5714         napi->poll_owner = -1;
5715 #endif
5716         set_bit(NAPI_STATE_SCHED, &napi->state);
5717         napi_hash_add(napi);
5718 }
5719 EXPORT_SYMBOL(netif_napi_add);
5720 
5721 void napi_disable(struct napi_struct *n)
5722 {
5723         might_sleep();
5724         set_bit(NAPI_STATE_DISABLE, &n->state);
5725 
5726         while (test_and_set_bit(NAPI_STATE_SCHED, &n->state))
5727                 msleep(1);
5728         while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state))
5729                 msleep(1);
5730 
5731         hrtimer_cancel(&n->timer);
5732 
5733         clear_bit(NAPI_STATE_DISABLE, &n->state);
5734 }
5735 EXPORT_SYMBOL(napi_disable);
5736 
5737 /* Must be called in process context */
5738 void netif_napi_del(struct napi_struct *napi)
5739 {
5740         might_sleep();
5741         if (napi_hash_del(napi))
5742                 synchronize_net();
5743         list_del_init(&napi->dev_list);
5744         napi_free_frags(napi);
5745 
5746         kfree_skb_list(napi->gro_list);
5747         napi->gro_list = NULL;
5748         napi->gro_count = 0;
5749 }
5750 EXPORT_SYMBOL(netif_napi_del);
5751 
5752 static int napi_poll(struct napi_struct *n, struct list_head *repoll)
5753 {
5754         void *have;
5755         int work, weight;
5756 
5757         list_del_init(&n->poll_list);
5758 
5759         have = netpoll_poll_lock(n);
5760 
5761         weight = n->weight;
5762 
5763         /* This NAPI_STATE_SCHED test is for avoiding a race
5764          * with netpoll's poll_napi().  Only the entity which
5765          * obtains the lock and sees NAPI_STATE_SCHED set will
5766          * actually make the ->poll() call.  Therefore we avoid
5767          * accidentally calling ->poll() when NAPI is not scheduled.
5768          */
5769         work = 0;
5770         if (test_bit(NAPI_STATE_SCHED, &n->state)) {
5771                 work = n->poll(n, weight);
5772                 trace_napi_poll(n, work, weight);
5773         }
5774 
5775         WARN_ON_ONCE(work > weight);
5776 
5777         if (likely(work < weight))
5778                 goto out_unlock;
5779 
5780         /* Drivers must not modify the NAPI state if they
5781          * consume the entire weight.  In such cases this code
5782          * still "owns" the NAPI instance and therefore can
5783          * move the instance around on the list at-will.
5784          */
5785         if (unlikely(napi_disable_pending(n))) {
5786                 napi_complete(n);
5787                 goto out_unlock;
5788         }
5789 
5790         if (n->gro_list) {
5791                 /* flush too old packets
5792                  * If HZ < 1000, flush all packets.
5793                  */
5794                 napi_gro_flush(n, HZ >= 1000);
5795         }
5796 
5797         /* Some drivers may have called napi_schedule
5798          * prior to exhausting their budget.
5799          */
5800         if (unlikely(!list_empty(&n->poll_list))) {
5801                 pr_warn_once("%s: Budget exhausted after napi rescheduled\n",
5802                              n->dev ? n->dev->name : "backlog");
5803                 goto out_unlock;
5804         }
5805 
5806         list_add_tail(&n->poll_list, repoll);
5807 
5808 out_unlock:
5809         netpoll_poll_unlock(have);
5810 
5811         return work;
5812 }
5813 
5814 static __latent_entropy void net_rx_action(struct softirq_action *h)
5815 {
5816         struct softnet_data *sd = this_cpu_ptr(&softnet_data);
5817         unsigned long time_limit = jiffies +
5818                 usecs_to_jiffies(netdev_budget_usecs);
5819         int budget = netdev_budget;
5820         LIST_HEAD(list);
5821         LIST_HEAD(repoll);
5822 
5823         local_irq_disable();
5824         list_splice_init(&sd->poll_list, &list);
5825         local_irq_enable();
5826 
5827         for (;;) {
5828                 struct napi_struct *n;
5829 
5830                 if (list_empty(&list)) {
5831                         if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll))
5832                                 goto out;
5833                         break;
5834                 }
5835 
5836                 n = list_first_entry(&list, struct napi_struct, poll_list);
5837                 budget -= napi_poll(n, &repoll);
5838 
5839                 /* If softirq window is exhausted then punt.
5840                  * Allow this to run for 2 jiffies since which will allow
5841                  * an average latency of 1.5/HZ.
5842                  */
5843                 if (unlikely(budget <= 0 ||
5844                              time_after_eq(jiffies, time_limit))) {
5845                         sd->time_squeeze++;
5846                         break;
5847                 }
5848         }
5849 
5850         local_irq_disable();
5851 
5852         list_splice_tail_init(&sd->poll_list, &list);
5853         list_splice_tail(&repoll, &list);
5854         list_splice(&list, &sd->poll_list);
5855         if (!list_empty(&sd->poll_list))
5856                 __raise_softirq_irqoff(NET_RX_SOFTIRQ);
5857 
5858         net_rps_action_and_irq_enable(sd);
5859 out:
5860         __kfree_skb_flush();
5861 }
5862 
5863 struct netdev_adjacent {
5864         struct net_device *dev;
5865 
5866         /* upper master flag, there can only be one master device per list */
5867         bool master;
5868 
5869         /* counter for the number of times this device was added to us */
5870         u16 ref_nr;
5871 
5872         /* private field for the users */
5873         void *private;
5874 
5875         struct list_head list;
5876         struct rcu_head rcu;
5877 };
5878 
5879 static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev,
5880                                                  struct list_head *adj_list)
5881 {
5882         struct netdev_adjacent *adj;
5883 
5884         list_for_each_entry(adj, adj_list, list) {
5885                 if (adj->dev == adj_dev)
5886                         return adj;
5887         }
5888         return NULL;
5889 }
5890 
5891 static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data)
5892 {
5893         struct net_device *dev = data;
5894 
5895         return upper_dev == dev;
5896 }
5897 
5898 /**
5899  * netdev_has_upper_dev - Check if device is linked to an upper device
5900  * @dev: device
5901  * @upper_dev: upper device to check
5902  *
5903  * Find out if a device is linked to specified upper device and return true
5904  * in case it is. Note that this checks only immediate upper device,
5905  * not through a complete stack of devices. The caller must hold the RTNL lock.
5906  */
5907 bool netdev_has_upper_dev(struct net_device *dev,
5908                           struct net_device *upper_dev)
5909 {
5910         ASSERT_RTNL();
5911 
5912         return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5913                                              upper_dev);
5914 }
5915 EXPORT_SYMBOL(netdev_has_upper_dev);
5916 
5917 /**
5918  * netdev_has_upper_dev_all - Check if device is linked to an upper device
5919  * @dev: device
5920  * @upper_dev: upper device to check
5921  *
5922  * Find out if a device is linked to specified upper device and return true
5923  * in case it is. Note that this checks the entire upper device chain.
5924  * The caller must hold rcu lock.
5925  */
5926 
5927 bool netdev_has_upper_dev_all_rcu(struct net_device *dev,
5928                                   struct net_device *upper_dev)
5929 {
5930         return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev,
5931                                                upper_dev);
5932 }
5933 EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu);
5934 
5935 /**
5936  * netdev_has_any_upper_dev - Check if device is linked to some device
5937  * @dev: device
5938  *
5939  * Find out if a device is linked to an upper device and return true in case
5940  * it is. The caller must hold the RTNL lock.
5941  */
5942 bool netdev_has_any_upper_dev(struct net_device *dev)
5943 {
5944         ASSERT_RTNL();
5945 
5946         return !list_empty(&dev->adj_list.upper);
5947 }
5948 EXPORT_SYMBOL(netdev_has_any_upper_dev);
5949 
5950 /**
5951  * netdev_master_upper_dev_get - Get master upper device
5952  * @dev: device
5953  *
5954  * Find a master upper device and return pointer to it or NULL in case
5955  * it's not there. The caller must hold the RTNL lock.
5956  */
5957 struct net_device *netdev_master_upper_dev_get(struct net_device *dev)
5958 {
5959         struct netdev_adjacent *upper;
5960 
5961         ASSERT_RTNL();
5962 
5963         if (list_empty(&dev->adj_list.upper))
5964                 return NULL;
5965 
5966         upper = list_first_entry(&dev->adj_list.upper,
5967                                  struct netdev_adjacent, list);
5968         if (likely(upper->master))
5969                 return upper->dev;
5970         return NULL;
5971 }
5972 EXPORT_SYMBOL(netdev_master_upper_dev_get);
5973 
5974 /**
5975  * netdev_has_any_lower_dev - Check if device is linked to some device
5976  * @dev: device
5977  *
5978  * Find out if a device is linked to a lower device and return true in case
5979  * it is. The caller must hold the RTNL lock.
5980  */
5981 static bool netdev_has_any_lower_dev(struct net_device *dev)
5982 {
5983         ASSERT_RTNL();
5984 
5985         return !list_empty(&dev->adj_list.lower);
5986 }
5987 
5988 void *netdev_adjacent_get_private(struct list_head *adj_list)
5989 {
5990         struct netdev_adjacent *adj;
5991 
5992         adj = list_entry(adj_list, struct netdev_adjacent, list);
5993 
5994         return adj->private;
5995 }
5996 EXPORT_SYMBOL(netdev_adjacent_get_private);
5997 
5998 /**
5999  * netdev_upper_get_next_dev_rcu - Get the next dev from upper list
6000  * @dev: device
6001  * @iter: list_head ** of the current position
6002  *
6003  * Gets the next device from the dev's upper list, starting from iter
6004  * position. The caller must hold RCU read lock.
6005  */
6006 struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev,
6007                                                  struct list_head **iter)
6008 {
6009         struct netdev_adjacent *upper;
6010 
6011         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6012 
6013         upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6014 
6015         if (&upper->list == &dev->adj_list.upper)
6016                 return NULL;
6017 
6018         *iter = &upper->list;
6019 
6020         return upper->dev;
6021 }
6022 EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu);
6023 
6024 static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev,
6025                                                     struct list_head **iter)
6026 {
6027         struct netdev_adjacent *upper;
6028 
6029         WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held());
6030 
6031         upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6032 
6033         if (&upper->list == &dev->adj_list.upper)
6034                 return NULL;
6035 
6036         *iter = &upper->list;
6037 
6038         return upper->dev;
6039 }
6040 
6041 int netdev_walk_all_upper_dev_rcu(struct net_device *dev,
6042                                   int (*fn)(struct net_device *dev,
6043                                             void *data),
6044                                   void *data)
6045 {
6046         struct net_device *udev;
6047         struct list_head *iter;
6048         int ret;
6049 
6050         for (iter = &dev->adj_list.upper,
6051              udev = netdev_next_upper_dev_rcu(dev, &iter);
6052              udev;
6053              udev = netdev_next_upper_dev_rcu(dev, &iter)) {
6054                 /* first is the upper device itself */
6055                 ret = fn(udev, data);
6056                 if (ret)
6057                         return ret;
6058 
6059                 /* then look at all of its upper devices */
6060                 ret = netdev_walk_all_upper_dev_rcu(udev, fn, data);
6061                 if (ret)
6062                         return ret;
6063         }
6064 
6065         return 0;
6066 }
6067 EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu);
6068 
6069 /**
6070  * netdev_lower_get_next_private - Get the next ->private from the
6071  *                                 lower neighbour list
6072  * @dev: device
6073  * @iter: list_head ** of the current position
6074  *
6075  * Gets the next netdev_adjacent->private from the dev's lower neighbour
6076  * list, starting from iter position. The caller must hold either hold the
6077  * RTNL lock or its own locking that guarantees that the neighbour lower
6078  * list will remain unchanged.
6079  */
6080 void *netdev_lower_get_next_private(struct net_device *dev,
6081                                     struct list_head **iter)
6082 {
6083         struct netdev_adjacent *lower;
6084 
6085         lower = list_entry(*iter, struct netdev_adjacent, list);
6086 
6087         if (&lower->list == &dev->adj_list.lower)
6088                 return NULL;
6089 
6090         *iter = lower->list.next;
6091 
6092         return lower->private;
6093 }
6094 EXPORT_SYMBOL(netdev_lower_get_next_private);
6095 
6096 /**
6097  * netdev_lower_get_next_private_rcu - Get the next ->private from the
6098  *                                     lower neighbour list, RCU
6099  *                                     variant
6100  * @dev: device
6101  * @iter: list_head ** of the current position
6102  *
6103  * Gets the next netdev_adjacent->private from the dev's lower neighbour
6104  * list, starting from iter position. The caller must hold RCU read lock.
6105  */
6106 void *netdev_lower_get_next_private_rcu(struct net_device *dev,
6107                                         struct list_head **iter)
6108 {
6109         struct netdev_adjacent *lower;
6110 
6111         WARN_ON_ONCE(!rcu_read_lock_held());
6112 
6113         lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6114 
6115         if (&lower->list == &dev->adj_list.lower)
6116                 return NULL;
6117 
6118         *iter = &lower->list;
6119 
6120         return lower->private;
6121 }
6122 EXPORT_SYMBOL(netdev_lower_get_next_private_rcu);
6123 
6124 /**
6125  * netdev_lower_get_next - Get the next device from the lower neighbour
6126  *                         list
6127  * @dev: device
6128  * @iter: list_head ** of the current position
6129  *
6130  * Gets the next netdev_adjacent from the dev's lower neighbour
6131  * list, starting from iter position. The caller must hold RTNL lock or
6132  * its own locking that guarantees that the neighbour lower
6133  * list will remain unchanged.
6134  */
6135 void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter)
6136 {
6137         struct netdev_adjacent *lower;
6138 
6139         lower = list_entry(*iter, struct netdev_adjacent, list);
6140 
6141         if (&lower->list == &dev->adj_list.lower)
6142                 return NULL;
6143 
6144         *iter = lower->list.next;
6145 
6146         return lower->dev;
6147 }
6148 EXPORT_SYMBOL(netdev_lower_get_next);
6149 
6150 static struct net_device *netdev_next_lower_dev(struct net_device *dev,
6151                                                 struct list_head **iter)
6152 {
6153         struct netdev_adjacent *lower;
6154 
6155         lower = list_entry((*iter)->next, struct netdev_adjacent, list);
6156 
6157         if (&lower->list == &dev->adj_list.lower)
6158                 return NULL;
6159 
6160         *iter = &lower->list;
6161 
6162         return lower->dev;
6163 }
6164 
6165 int netdev_walk_all_lower_dev(struct net_device *dev,
6166                               int (*fn)(struct net_device *dev,
6167                                         void *data),
6168                               void *data)
6169 {
6170         struct net_device *ldev;
6171         struct list_head *iter;
6172         int ret;
6173 
6174         for (iter = &dev->adj_list.lower,
6175              ldev = netdev_next_lower_dev(dev, &iter);
6176              ldev;
6177              ldev = netdev_next_lower_dev(dev, &iter)) {
6178                 /* first is the lower device itself */
6179                 ret = fn(ldev, data);
6180                 if (ret)
6181                         return ret;
6182 
6183                 /* then look at all of its lower devices */
6184                 ret = netdev_walk_all_lower_dev(ldev, fn, data);
6185                 if (ret)
6186                         return ret;
6187         }
6188 
6189         return 0;
6190 }
6191 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev);
6192 
6193 static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev,
6194                                                     struct list_head **iter)
6195 {
6196         struct netdev_adjacent *lower;
6197 
6198         lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list);
6199         if (&lower->list == &dev->adj_list.lower)
6200                 return NULL;
6201 
6202         *iter = &lower->list;
6203 
6204         return lower->dev;
6205 }
6206 
6207 int netdev_walk_all_lower_dev_rcu(struct net_device *dev,
6208                                   int (*fn)(struct net_device *dev,
6209                                             void *data),
6210                                   void *data)
6211 {
6212         struct net_device *ldev;
6213         struct list_head *iter;
6214         int ret;
6215 
6216         for (iter = &dev->adj_list.lower,
6217              ldev = netdev_next_lower_dev_rcu(dev, &iter);
6218              ldev;
6219              ldev = netdev_next_lower_dev_rcu(dev, &iter)) {
6220                 /* first is the lower device itself */
6221                 ret = fn(ldev, data);
6222                 if (ret)
6223                         return ret;
6224 
6225                 /* then look at all of its lower devices */
6226                 ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data);
6227                 if (ret)
6228                         return ret;
6229         }
6230 
6231         return 0;
6232 }
6233 EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu);
6234 
6235 /**
6236  * netdev_lower_get_first_private_rcu - Get the first ->private from the
6237  *                                     lower neighbour list, RCU
6238  *                                     variant
6239  * @dev: device
6240  *
6241  * Gets the first netdev_adjacent->private from the dev's lower neighbour
6242  * list. The caller must hold RCU read lock.
6243  */
6244 void *netdev_lower_get_first_private_rcu(struct net_device *dev)
6245 {
6246         struct netdev_adjacent *lower;
6247 
6248         lower = list_first_or_null_rcu(&dev->adj_list.lower,
6249                         struct netdev_adjacent, list);
6250         if (lower)
6251                 return lower->private;
6252         return NULL;
6253 }
6254 EXPORT_SYMBOL(netdev_lower_get_first_private_rcu);
6255 
6256 /**
6257  * netdev_master_upper_dev_get_rcu - Get master upper device
6258  * @dev: device
6259  *
6260  * Find a master upper device and return pointer to it or NULL in case
6261  * it's not there. The caller must hold the RCU read lock.
6262  */
6263 struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev)
6264 {
6265         struct netdev_adjacent *upper;
6266 
6267         upper = list_first_or_null_rcu(&dev->adj_list.upper,
6268                                        struct netdev_adjacent, list);
6269         if (upper && likely(upper->master))
6270                 return upper->dev;
6271         return NULL;
6272 }
6273 EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu);
6274 
6275 static int netdev_adjacent_sysfs_add(struct net_device *dev,
6276                               struct net_device *adj_dev,
6277                               struct list_head *dev_list)
6278 {
6279         char linkname[IFNAMSIZ+7];
6280 
6281         sprintf(linkname, dev_list == &dev->adj_list.upper ?
6282                 "upper_%s" : "lower_%s", adj_dev->name);
6283         return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj),
6284                                  linkname);
6285 }
6286 static void netdev_adjacent_sysfs_del(struct net_device *dev,
6287                                char *name,
6288                                struct list_head *dev_list)
6289 {
6290         char linkname[IFNAMSIZ+7];
6291 
6292         sprintf(linkname, dev_list == &dev->adj_list.upper ?
6293                 "upper_%s" : "lower_%s", name);
6294         sysfs_remove_link(&(dev->dev.kobj), linkname);
6295 }
6296 
6297 static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev,
6298                                                  struct net_device *adj_dev,
6299                                                  struct list_head *dev_list)
6300 {
6301         return (dev_list == &dev->adj_list.upper ||
6302                 dev_list == &dev->adj_list.lower) &&
6303                 net_eq(dev_net(dev), dev_net(adj_dev));
6304 }
6305 
6306 static int __netdev_adjacent_dev_insert(struct net_device *dev,
6307                                         struct net_device *adj_dev,
6308                                         struct list_head *dev_list,
6309                                         void *private, bool master)
6310 {
6311         struct netdev_adjacent *adj;
6312         int ret;
6313 
6314         adj = __netdev_find_adj(adj_dev, dev_list);
6315 
6316         if (adj) {
6317                 adj->ref_nr += 1;
6318                 pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n",
6319                          dev->name, adj_dev->name, adj->ref_nr);
6320 
6321                 return 0;
6322         }
6323 
6324         adj = kmalloc(sizeof(*adj), GFP_KERNEL);
6325         if (!adj)
6326                 return -ENOMEM;
6327 
6328         adj->dev = adj_dev;
6329         adj->master = master;
6330         adj->ref_nr = 1;
6331         adj->private = private;
6332         dev_hold(adj_dev);
6333 
6334         pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n",
6335                  dev->name, adj_dev->name, adj->ref_nr, adj_dev->name);
6336 
6337         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) {
6338                 ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list);
6339                 if (ret)
6340                         goto free_adj;
6341         }
6342 
6343         /* Ensure that master link is always the first item in list. */
6344         if (master) {
6345                 ret = sysfs_create_link(&(dev->dev.kobj),
6346                                         &(adj_dev->dev.kobj), "master");
6347                 if (ret)
6348                         goto remove_symlinks;
6349 
6350                 list_add_rcu(&adj->list, dev_list);
6351         } else {
6352                 list_add_tail_rcu(&adj->list, dev_list);
6353         }
6354 
6355         return 0;
6356 
6357 remove_symlinks:
6358         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6359                 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6360 free_adj:
6361         kfree(adj);
6362         dev_put(adj_dev);
6363 
6364         return ret;
6365 }
6366 
6367 static void __netdev_adjacent_dev_remove(struct net_device *dev,
6368                                          struct net_device *adj_dev,
6369                                          u16 ref_nr,
6370                                          struct list_head *dev_list)
6371 {
6372         struct netdev_adjacent *adj;
6373 
6374         pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n",
6375                  dev->name, adj_dev->name, ref_nr);
6376 
6377         adj = __netdev_find_adj(adj_dev, dev_list);
6378 
6379         if (!adj) {
6380                 pr_err("Adjacency does not exist for device %s from %s\n",
6381                        dev->name, adj_dev->name);
6382                 WARN_ON(1);
6383                 return;
6384         }
6385 
6386         if (adj->ref_nr > ref_nr) {
6387                 pr_debug("adjacency: %s to %s ref_nr - %d = %d\n",
6388                          dev->name, adj_dev->name, ref_nr,
6389                          adj->ref_nr - ref_nr);
6390                 adj->ref_nr -= ref_nr;
6391                 return;
6392         }
6393 
6394         if (adj->master)
6395                 sysfs_remove_link(&(dev->dev.kobj), "master");
6396 
6397         if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list))
6398                 netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list);
6399 
6400         list_del_rcu(&adj->list);
6401         pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n",
6402                  adj_dev->name, dev->name, adj_dev->name);
6403         dev_put(adj_dev);
6404         kfree_rcu(adj, rcu);
6405 }
6406 
6407 static int __netdev_adjacent_dev_link_lists(struct net_device *dev,
6408                                             struct net_device *upper_dev,
6409                                             struct list_head *up_list,
6410                                             struct list_head *down_list,
6411                                             void *private, bool master)
6412 {
6413         int ret;
6414 
6415         ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list,
6416                                            private, master);
6417         if (ret)
6418                 return ret;
6419 
6420         ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list,
6421                                            private, false);
6422         if (ret) {
6423                 __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list);
6424                 return ret;
6425         }
6426 
6427         return 0;
6428 }
6429 
6430 static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev,
6431                                                struct net_device *upper_dev,
6432                                                u16 ref_nr,
6433                                                struct list_head *up_list,
6434                                                struct list_head *down_list)
6435 {
6436         __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list);
6437         __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list);
6438 }
6439 
6440 static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev,
6441                                                 struct net_device *upper_dev,
6442                                                 void *private, bool master)
6443 {
6444         return __netdev_adjacent_dev_link_lists(dev, upper_dev,
6445                                                 &dev->adj_list.upper,
6446                                                 &upper_dev->adj_list.lower,
6447                                                 private, master);
6448 }
6449 
6450 static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev,
6451                                                    struct net_device *upper_dev)
6452 {
6453         __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1,
6454                                            &dev->adj_list.upper,
6455                                            &upper_dev->adj_list.lower);
6456 }
6457 
6458 static int __netdev_upper_dev_link(struct net_device *dev,
6459                                    struct net_device *upper_dev, bool master,
6460                                    void *upper_priv, void *upper_info,
6461                                    struct netlink_ext_ack *extack)
6462 {
6463         struct netdev_notifier_changeupper_info changeupper_info = {
6464                 .info = {
6465                         .dev = dev,
6466                         .extack = extack,
6467                 },
6468                 .upper_dev = upper_dev,
6469                 .master = master,
6470                 .linking = true,
6471                 .upper_info = upper_info,
6472         };
6473         struct net_device *master_dev;
6474         int ret = 0;
6475 
6476         ASSERT_RTNL();
6477 
6478         if (dev == upper_dev)
6479                 return -EBUSY;
6480 
6481         /* To prevent loops, check if dev is not upper device to upper_dev. */
6482         if (netdev_has_upper_dev(upper_dev, dev))
6483                 return -EBUSY;
6484 
6485         if (!master) {
6486                 if (netdev_has_upper_dev(dev, upper_dev))
6487                         return -EEXIST;
6488         } else {
6489                 master_dev = netdev_master_upper_dev_get(dev);
6490                 if (master_dev)
6491                         return master_dev == upper_dev ? -EEXIST : -EBUSY;
6492         }
6493 
6494         ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
6495                                             &changeupper_info.info);
6496         ret = notifier_to_errno(ret);
6497         if (ret)
6498                 return ret;
6499 
6500         ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv,
6501                                                    master);
6502         if (ret)
6503                 return ret;
6504 
6505         ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
6506                                             &changeupper_info.info);
6507         ret = notifier_to_errno(ret);
6508         if (ret)
6509                 goto rollback;
6510 
6511         return 0;
6512 
6513 rollback:
6514         __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6515 
6516         return ret;
6517 }
6518 
6519 /**
6520  * netdev_upper_dev_link - Add a link to the upper device
6521  * @dev: device
6522  * @upper_dev: new upper device
6523  * @extack: netlink extended ack
6524  *
6525  * Adds a link to device which is upper to this one. The caller must hold
6526  * the RTNL lock. On a failure a negative errno code is returned.
6527  * On success the reference counts are adjusted and the function
6528  * returns zero.
6529  */
6530 int netdev_upper_dev_link(struct net_device *dev,
6531                           struct net_device *upper_dev,
6532                           struct netlink_ext_ack *extack)
6533 {
6534         return __netdev_upper_dev_link(dev, upper_dev, false,
6535                                        NULL, NULL, extack);
6536 }
6537 EXPORT_SYMBOL(netdev_upper_dev_link);
6538 
6539 /**
6540  * netdev_master_upper_dev_link - Add a master link to the upper device
6541  * @dev: device
6542  * @upper_dev: new upper device
6543  * @upper_priv: upper device private
6544  * @upper_info: upper info to be passed down via notifier
6545  * @extack: netlink extended ack
6546  *
6547  * Adds a link to device which is upper to this one. In this case, only
6548  * one master upper device can be linked, although other non-master devices
6549  * might be linked as well. The caller must hold the RTNL lock.
6550  * On a failure a negative errno code is returned. On success the reference
6551  * counts are adjusted and the function returns zero.
6552  */
6553 int netdev_master_upper_dev_link(struct net_device *dev,
6554                                  struct net_device *upper_dev,
6555                                  void *upper_priv, void *upper_info,
6556                                  struct netlink_ext_ack *extack)
6557 {
6558         return __netdev_upper_dev_link(dev, upper_dev, true,
6559                                        upper_priv, upper_info, extack);
6560 }
6561 EXPORT_SYMBOL(netdev_master_upper_dev_link);
6562 
6563 /**
6564  * netdev_upper_dev_unlink - Removes a link to upper device
6565  * @dev: device
6566  * @upper_dev: new upper device
6567  *
6568  * Removes a link to device which is upper to this one. The caller must hold
6569  * the RTNL lock.
6570  */
6571 void netdev_upper_dev_unlink(struct net_device *dev,
6572                              struct net_device *upper_dev)
6573 {
6574         struct netdev_notifier_changeupper_info changeupper_info = {
6575                 .info = {
6576                         .dev = dev,
6577                 },
6578                 .upper_dev = upper_dev,
6579                 .linking = false,
6580         };
6581 
6582         ASSERT_RTNL();
6583 
6584         changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev;
6585 
6586         call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER,
6587                                       &changeupper_info.info);
6588 
6589         __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev);
6590 
6591         call_netdevice_notifiers_info(NETDEV_CHANGEUPPER,
6592                                       &changeupper_info.info);
6593 }
6594 EXPORT_SYMBOL(netdev_upper_dev_unlink);
6595 
6596 /**
6597  * netdev_bonding_info_change - Dispatch event about slave change
6598  * @dev: device
6599  * @bonding_info: info to dispatch
6600  *
6601  * Send NETDEV_BONDING_INFO to netdev notifiers with info.
6602  * The caller must hold the RTNL lock.
6603  */
6604 void netdev_bonding_info_change(struct net_device *dev,
6605                                 struct netdev_bonding_info *bonding_info)
6606 {
6607         struct netdev_notifier_bonding_info info = {
6608                 .info.dev = dev,
6609         };
6610 
6611         memcpy(&info.bonding_info, bonding_info,
6612                sizeof(struct netdev_bonding_info));
6613         call_netdevice_notifiers_info(NETDEV_BONDING_INFO,
6614                                       &info.info);
6615 }
6616 EXPORT_SYMBOL(netdev_bonding_info_change);
6617 
6618 static void netdev_adjacent_add_links(struct net_device *dev)
6619 {
6620         struct netdev_adjacent *iter;
6621 
6622         struct net *net = dev_net(dev);
6623 
6624         list_for_each_entry(iter, &dev->adj_list.upper, list) {
6625                 if (!net_eq(net, dev_net(iter->dev)))
6626                         continue;
6627                 netdev_adjacent_sysfs_add(iter->dev, dev,
6628                                           &iter->dev->adj_list.lower);
6629                 netdev_adjacent_sysfs_add(dev, iter->dev,
6630                                           &dev->adj_list.upper);
6631         }
6632 
6633         list_for_each_entry(iter, &dev->adj_list.lower, list) {
6634                 if (!net_eq(net, dev_net(iter->dev)))
6635                         continue;
6636                 netdev_adjacent_sysfs_add(iter->dev, dev,
6637                                           &iter->dev->adj_list.upper);
6638                 netdev_adjacent_sysfs_add(dev, iter->dev,
6639                                           &dev->adj_list.lower);
6640         }
6641 }
6642 
6643 static void netdev_adjacent_del_links(struct net_device *dev)
6644 {
6645         struct netdev_adjacent *iter;
6646 
6647         struct net *net = dev_net(dev);
6648 
6649         list_for_each_entry(iter, &dev->adj_list.upper, list) {
6650                 if (!net_eq(net, dev_net(iter->dev)))
6651                         continue;
6652                 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6653                                           &iter->dev->adj_list.lower);
6654                 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6655                                           &dev->adj_list.upper);
6656         }
6657 
6658         list_for_each_entry(iter, &dev->adj_list.lower, list) {
6659                 if (!net_eq(net, dev_net(iter->dev)))
6660                         continue;
6661                 netdev_adjacent_sysfs_del(iter->dev, dev->name,
6662                                           &iter->dev->adj_list.upper);
6663                 netdev_adjacent_sysfs_del(dev, iter->dev->name,
6664                                           &dev->adj_list.lower);
6665         }
6666 }
6667 
6668 void netdev_adjacent_rename_links(struct net_device *dev, char *oldname)
6669 {
6670         struct netdev_adjacent *iter;
6671 
6672         struct net *net = dev_net(dev);
6673 
6674         list_for_each_entry(iter, &dev->adj_list.upper, list) {
6675                 if (!net_eq(net, dev_net(iter->dev)))
6676                         continue;
6677                 netdev_adjacent_sysfs_del(iter->dev, oldname,
6678                                           &iter->dev->adj_list.lower);
6679                 netdev_adjacent_sysfs_add(iter->dev, dev,
6680                                           &iter->dev->adj_list.lower);
6681         }
6682 
6683         list_for_each_entry(iter, &dev->adj_list.lower, list) {
6684                 if (!net_eq(net, dev_net(iter->dev)))
6685                         continue;
6686                 netdev_adjacent_sysfs_del(iter->dev, oldname,
6687                                           &iter->dev->adj_list.upper);
6688                 netdev_adjacent_sysfs_add(iter->dev, dev,
6689                                           &iter->dev->adj_list.upper);
6690         }
6691 }
6692 
6693 void *netdev_lower_dev_get_private(struct net_device *dev,
6694                                    struct net_device *lower_dev)
6695 {
6696         struct netdev_adjacent *lower;
6697 
6698         if (!lower_dev)
6699                 return NULL;
6700         lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower);
6701         if (!lower)
6702                 return NULL;
6703 
6704         return lower->private;
6705 }
6706 EXPORT_SYMBOL(netdev_lower_dev_get_private);
6707 
6708 
6709 int dev_get_nest_level(struct net_device *dev)
6710 {
6711         struct net_device *lower = NULL;
6712         struct list_head *iter;
6713         int max_nest = -1;
6714         int nest;
6715 
6716         ASSERT_RTNL();
6717 
6718         netdev_for_each_lower_dev(dev, lower, iter) {
6719                 nest = dev_get_nest_level(lower);
6720                 if (max_nest < nest)
6721                         max_nest = nest;
6722         }
6723 
6724         return max_nest + 1;
6725 }
6726 EXPORT_SYMBOL(dev_get_nest_level);
6727 
6728 /**
6729  * netdev_lower_change - Dispatch event about lower device state change
6730  * @lower_dev: device
6731  * @lower_state_info: state to dispatch
6732  *
6733  * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info.
6734  * The caller must hold the RTNL lock.
6735  */
6736 void netdev_lower_state_changed(struct net_device *lower_dev,
6737                                 void *lower_state_info)
6738 {
6739         struct netdev_notifier_changelowerstate_info changelowerstate_info = {
6740                 .info.dev = lower_dev,
6741         };
6742 
6743         ASSERT_RTNL();
6744         changelowerstate_info.lower_state_info = lower_state_info;
6745         call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE,
6746                                       &changelowerstate_info.info);
6747 }
6748 EXPORT_SYMBOL(netdev_lower_state_changed);
6749 
6750 static void dev_change_rx_flags(struct net_device *dev, int flags)
6751 {
6752         const struct net_device_ops *ops = dev->netdev_ops;
6753 
6754         if (ops->ndo_change_rx_flags)
6755                 ops->ndo_change_rx_flags(dev, flags);
6756 }
6757 
6758 static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify)
6759 {
6760         unsigned int old_flags = dev->flags;
6761         kuid_t uid;
6762         kgid_t gid;
6763 
6764         ASSERT_RTNL();
6765 
6766         dev->flags |= IFF_PROMISC;
6767         dev->promiscuity += inc;
6768         if (dev->promiscuity == 0) {
6769                 /*
6770                  * Avoid overflow.
6771                  * If inc causes overflow, untouch promisc and return error.
6772                  */
6773                 if (inc < 0)
6774                         dev->flags &= ~IFF_PROMISC;
6775                 else {
6776                         dev->promiscuity -= inc;
6777                         pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
6778                                 dev->name);
6779                         return -EOVERFLOW;
6780                 }
6781         }
6782         if (dev->flags != old_flags) {
6783                 pr_info("device %s %s promiscuous mode\n",
6784                         dev->name,
6785                         dev->flags & IFF_PROMISC ? "entered" : "left");
6786                 if (audit_enabled) {
6787                         current_uid_gid(&uid, &gid);
6788                         audit_log(audit_context(), GFP_ATOMIC,
6789                                   AUDIT_ANOM_PROMISCUOUS,
6790                                   "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
6791                                   dev->name, (dev->flags & IFF_PROMISC),
6792                                   (old_flags & IFF_PROMISC),
6793                                   from_kuid(&init_user_ns, audit_get_loginuid(current)),
6794                                   from_kuid(&init_user_ns, uid),
6795                                   from_kgid(&init_user_ns, gid),
6796                                   audit_get_sessionid(current));
6797                 }
6798 
6799                 dev_change_rx_flags(dev, IFF_PROMISC);
6800         }
6801         if (notify)
6802                 __dev_notify_flags(dev, old_flags, IFF_PROMISC);
6803         return 0;
6804 }
6805 
6806 /**
6807  *      dev_set_promiscuity     - update promiscuity count on a device
6808  *      @dev: device
6809  *      @inc: modifier
6810  *
6811  *      Add or remove promiscuity from a device. While the count in the device
6812  *      remains above zero the interface remains promiscuous. Once it hits zero
6813  *      the device reverts back to normal filtering operation. A negative inc
6814  *      value is used to drop promiscuity on the device.
6815  *      Return 0 if successful or a negative errno code on error.
6816  */
6817 int dev_set_promiscuity(struct net_device *dev, int inc)
6818 {
6819         unsigned int old_flags = dev->flags;
6820         int err;
6821 
6822         err = __dev_set_promiscuity(dev, inc, true);
6823         if (err < 0)
6824                 return err;
6825         if (dev->flags != old_flags)
6826                 dev_set_rx_mode(dev);
6827         return err;
6828 }
6829 EXPORT_SYMBOL(dev_set_promiscuity);
6830 
6831 static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify)
6832 {
6833         unsigned int old_flags = dev->flags, old_gflags = dev->gflags;
6834 
6835         ASSERT_RTNL();
6836 
6837         dev->flags |= IFF_ALLMULTI;
6838         dev->allmulti += inc;
6839         if (dev->allmulti == 0) {
6840                 /*
6841                  * Avoid overflow.
6842                  * If inc causes overflow, untouch allmulti and return error.
6843                  */
6844                 if (inc < 0)
6845                         dev->flags &= ~IFF_ALLMULTI;
6846                 else {
6847                         dev->allmulti -= inc;
6848                         pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
6849                                 dev->name);
6850                         return -EOVERFLOW;
6851                 }
6852         }
6853         if (dev->flags ^ old_flags) {
6854                 dev_change_rx_flags(dev, IFF_ALLMULTI);
6855                 dev_set_rx_mode(dev);
6856                 if (notify)
6857                         __dev_notify_flags(dev, old_flags,
6858                                            dev->gflags ^ old_gflags);
6859         }
6860         return 0;
6861 }
6862 
6863 /**
6864  *      dev_set_allmulti        - update allmulti count on a device
6865  *      @dev: device
6866  *      @inc: modifier
6867  *
6868  *      Add or remove reception of all multicast frames to a device. While the
6869  *      count in the device remains above zero the interface remains listening
6870  *      to all interfaces. Once it hits zero the device reverts back to normal
6871  *      filtering operation. A negative @inc value is used to drop the counter
6872  *      when releasing a resource needing all multicasts.
6873  *      Return 0 if successful or a negative errno code on error.
6874  */
6875 
6876 int dev_set_allmulti(struct net_device *dev, int inc)
6877 {
6878         return __dev_set_allmulti(dev, inc, true);
6879 }
6880 EXPORT_SYMBOL(dev_set_allmulti);
6881 
6882 /*
6883  *      Upload unicast and multicast address lists to device and
6884  *      configure RX filtering. When the device doesn't support unicast
6885  *      filtering it is put in promiscuous mode while unicast addresses
6886  *      are present.
6887  */
6888 void __dev_set_rx_mode(struct net_device *dev)
6889 {
6890         const struct net_device_ops *ops = dev->netdev_ops;
6891 
6892         /* dev_open will call this function so the list will stay sane. */
6893         if (!(dev->flags&IFF_UP))
6894                 return;
6895 
6896         if (!netif_device_present(dev))
6897                 return;
6898 
6899         if (!(dev->priv_flags & IFF_UNICAST_FLT)) {
6900                 /* Unicast addresses changes may only happen under the rtnl,
6901                  * therefore calling __dev_set_promiscuity here is safe.
6902                  */
6903                 if (!netdev_uc_empty(dev) && !dev->uc_promisc) {
6904                         __dev_set_promiscuity(dev, 1, false);
6905                         dev->uc_promisc = true;
6906                 } else if (netdev_uc_empty(dev) && dev->uc_promisc) {
6907                         __dev_set_promiscuity(dev, -1, false);
6908                         dev->uc_promisc = false;
6909                 }
6910         }
6911 
6912         if (ops->ndo_set_rx_mode)
6913                 ops->ndo_set_rx_mode(dev);
6914 }
6915 
6916 void dev_set_rx_mode(struct net_device *dev)
6917 {
6918         netif_addr_lock_bh(dev);
6919         __dev_set_rx_mode(dev);
6920         netif_addr_unlock_bh(dev);
6921 }
6922 
6923 /**
6924  *      dev_get_flags - get flags reported to userspace
6925  *      @dev: device
6926  *
6927  *      Get the combination of flag bits exported through APIs to userspace.
6928  */
6929 unsigned int dev_get_flags(const struct net_device *dev)
6930 {
6931         unsigned int flags;
6932 
6933         flags = (dev->flags & ~(IFF_PROMISC |
6934                                 IFF_ALLMULTI |
6935                                 IFF_RUNNING |
6936                                 IFF_LOWER_UP |
6937                                 IFF_DORMANT)) |
6938                 (dev->gflags & (IFF_PROMISC |
6939                                 IFF_ALLMULTI));
6940 
6941         if (netif_running(dev)) {
6942                 if (netif_oper_up(dev))
6943                         flags |= IFF_RUNNING;
6944                 if (netif_carrier_ok(dev))
6945                         flags |= IFF_LOWER_UP;
6946                 if (netif_dormant(dev))
6947                         flags |= IFF_DORMANT;
6948         }
6949 
6950         return flags;
6951 }
6952 EXPORT_SYMBOL(dev_get_flags);
6953 
6954 int __dev_change_flags(struct net_device *dev, unsigned int flags)
6955 {
6956         unsigned int old_flags = dev->flags;
6957         int ret;
6958 
6959         ASSERT_RTNL();
6960 
6961         /*
6962          *      Set the flags on our device.
6963          */
6964 
6965         dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP |
6966                                IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL |
6967                                IFF_AUTOMEDIA)) |
6968                      (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC |
6969                                     IFF_ALLMULTI));
6970 
6971         /*
6972          *      Load in the correct multicast list now the flags have changed.
6973          */
6974 
6975         if ((old_flags ^ flags) & IFF_MULTICAST)
6976                 dev_change_rx_flags(dev, IFF_MULTICAST);
6977 
6978         dev_set_rx_mode(dev);
6979 
6980         /*
6981          *      Have we downed the interface. We handle IFF_UP ourselves
6982          *      according to user attempts to set it, rather than blindly
6983          *      setting it.
6984          */
6985 
6986         ret = 0;
6987         if ((old_flags ^ flags) & IFF_UP) {
6988                 if (old_flags & IFF_UP)
6989                         __dev_close(dev);
6990                 else
6991                         ret = __dev_open(dev);
6992         }
6993 
6994         if ((flags ^ dev->gflags) & IFF_PROMISC) {
6995                 int inc = (flags & IFF_PROMISC) ? 1 : -1;
6996                 unsigned int old_flags = dev->flags;
6997 
6998                 dev->gflags ^= IFF_PROMISC;
6999 
7000                 if (__dev_set_promiscuity(dev, inc, false) >= 0)
7001                         if (dev->flags != old_flags)
7002                                 dev_set_rx_mode(dev);
7003         }
7004 
7005         /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
7006          * is important. Some (broken) drivers set IFF_PROMISC, when
7007          * IFF_ALLMULTI is requested not asking us and not reporting.
7008          */
7009         if ((flags ^ dev->gflags) & IFF_ALLMULTI) {
7010                 int inc = (flags & IFF_ALLMULTI) ? 1 : -1;
7011 
7012                 dev->gflags ^= IFF_ALLMULTI;
7013                 __dev_set_allmulti(dev, inc, false);
7014         }
7015 
7016         return ret;
7017 }
7018 
7019 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
7020                         unsigned int gchanges)
7021 {
7022         unsigned int changes = dev->flags ^ old_flags;
7023 
7024         if (gchanges)
7025                 rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC);
7026 
7027         if (changes & IFF_UP) {
7028                 if (dev->flags & IFF_UP)
7029                         call_netdevice_notifiers(NETDEV_UP, dev);
7030                 else
7031                         call_netdevice_notifiers(NETDEV_DOWN, dev);
7032         }
7033 
7034         if (dev->flags & IFF_UP &&
7035             (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) {
7036                 struct netdev_notifier_change_info change_info = {
7037                         .info = {
7038                                 .dev = dev,
7039                         },
7040                         .flags_changed = changes,
7041                 };
7042 
7043                 call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info);
7044         }
7045 }
7046 
7047 /**
7048  *      dev_change_flags - change device settings
7049  *      @dev: device
7050  *      @flags: device state flags
7051  *
7052  *      Change settings on device based state flags. The flags are
7053  *      in the userspace exported format.
7054  */
7055 int dev_change_flags(struct net_device *dev, unsigned int flags)
7056 {
7057         int ret;
7058         unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags;
7059 
7060         ret = __dev_change_flags(dev, flags);
7061         if (ret < 0)
7062                 return ret;
7063 
7064         changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags);
7065         __dev_notify_flags(dev, old_flags, changes);
7066         return ret;
7067 }
7068 EXPORT_SYMBOL(dev_change_flags);
7069 
7070 int __dev_set_mtu(struct net_device *dev, int new_mtu)
7071 {
7072         const struct net_device_ops *ops = dev->netdev_ops;
7073 
7074         if (ops->ndo_change_mtu)
7075                 return ops->ndo_change_mtu(dev, new_mtu);
7076 
7077         dev->mtu = new_mtu;
7078         return 0;
7079 }
7080 EXPORT_SYMBOL(__dev_set_mtu);
7081 
7082 /**
7083  *      dev_set_mtu - Change maximum transfer unit
7084  *      @dev: device
7085  *      @new_mtu: new transfer unit
7086  *
7087  *      Change the maximum transfer size of the network device.
7088  */
7089 int dev_set_mtu(struct net_device *dev, int new_mtu)
7090 {
7091         int err, orig_mtu;
7092 
7093         if (new_mtu == dev->mtu)
7094                 return 0;
7095 
7096         /* MTU must be positive, and in range */
7097         if (new_mtu < 0 || new_mtu < dev->min_mtu) {
7098                 net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n",
7099                                     dev->name, new_mtu, dev->min_mtu);
7100                 return -EINVAL;
7101         }
7102 
7103         if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) {
7104                 net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n",
7105                                     dev->name, new_mtu, dev->max_mtu);
7106                 return -EINVAL;
7107         }
7108 
7109         if (!netif_device_present(dev))
7110                 return -ENODEV;
7111 
7112         err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev);
7113         err = notifier_to_errno(err);
7114         if (err)
7115                 return err;
7116 
7117         orig_mtu = dev->mtu;
7118         err = __dev_set_mtu(dev, new_mtu);
7119 
7120         if (!err) {
7121                 err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
7122                 err = notifier_to_errno(err);
7123                 if (err) {
7124                         /* setting mtu back and notifying everyone again,
7125                          * so that they have a chance to revert changes.
7126                          */
7127                         __dev_set_mtu(dev, orig_mtu);
7128                         call_netdevice_notifiers(NETDEV_CHANGEMTU, dev);
7129                 }
7130         }
7131         return err;
7132 }
7133 EXPORT_SYMBOL(dev_set_mtu);
7134 
7135 /**
7136  *      dev_change_tx_queue_len - Change TX queue length of a netdevice
7137  *      @dev: device
7138  *      @new_len: new tx queue length
7139  */
7140 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len)
7141 {
7142         unsigned int orig_len = dev->tx_queue_len;
7143         int res;
7144 
7145         if (new_len != (unsigned int)new_len)
7146                 return -ERANGE;
7147 
7148         if (new_len != orig_len) {
7149                 dev->tx_queue_len = new_len;
7150                 res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev);
7151                 res = notifier_to_errno(res);
7152                 if (res) {
7153                         netdev_err(dev,
7154                                    "refused to change device tx_queue_len\n");
7155                         dev->tx_queue_len = orig_len;
7156                         return res;
7157                 }
7158                 return dev_qdisc_change_tx_queue_len(dev);
7159         }
7160 
7161         return 0;
7162 }
7163 
7164 /**
7165  *      dev_set_group - Change group this device belongs to
7166  *      @dev: device
7167  *      @new_group: group this device should belong to
7168  */
7169 void dev_set_group(struct net_device *dev, int new_group)
7170 {
7171         dev->group = new_group;
7172 }
7173 EXPORT_SYMBOL(dev_set_group);
7174 
7175 /**
7176  *      dev_set_mac_address - Change Media Access Control Address
7177  *      @dev: device
7178  *      @sa: new address
7179  *
7180  *      Change the hardware (MAC) address of the device
7181  */
7182 int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa)
7183 {
7184         const struct net_device_ops *ops = dev->netdev_ops;
7185         int err;
7186 
7187         if (!ops->ndo_set_mac_address)
7188                 return -EOPNOTSUPP;
7189         if (sa->sa_family != dev->type)
7190                 return -EINVAL;
7191         if (!netif_device_present(dev))
7192                 return -ENODEV;
7193         err = ops->ndo_set_mac_address(dev, sa);
7194         if (err)
7195                 return err;
7196         dev->addr_assign_type = NET_ADDR_SET;
7197         call_netdevice_notifiers(NETDEV_CHANGEADDR, dev);
7198         add_device_randomness(dev->dev_addr, dev->addr_len);
7199         return 0;
7200 }
7201 EXPORT_SYMBOL(dev_set_mac_address);
7202 
7203 /**
7204  *      dev_change_carrier - Change device carrier
7205  *      @dev: device
7206  *      @new_carrier: new value
7207  *
7208  *      Change device carrier
7209  */
7210 int dev_change_carrier(struct net_device *dev, bool new_carrier)
7211 {
7212         const struct net_device_ops *ops = dev->netdev_ops;
7213 
7214         if (!ops->ndo_change_carrier)
7215                 return -EOPNOTSUPP;
7216         if (!netif_device_present(dev))
7217                 return -ENODEV;
7218         return ops->ndo_change_carrier(dev, new_carrier);
7219 }
7220 EXPORT_SYMBOL(dev_change_carrier);
7221 
7222 /**
7223  *      dev_get_phys_port_id - Get device physical port ID
7224  *      @dev: device
7225  *      @ppid: port ID
7226  *
7227  *      Get device physical port ID
7228  */
7229 int dev_get_phys_port_id(struct net_device *dev,
7230                          struct netdev_phys_item_id *ppid)
7231 {
7232         const struct net_device_ops *ops = dev->netdev_ops;
7233 
7234         if (!ops->ndo_get_phys_port_id)
7235                 return -EOPNOTSUPP;
7236         return ops->ndo_get_phys_port_id(dev, ppid);
7237 }
7238 EXPORT_SYMBOL(dev_get_phys_port_id);
7239 
7240 /**
7241  *      dev_get_phys_port_name - Get device physical port name
7242  *      @dev: device
7243  *      @name: port name
7244  *      @len: limit of bytes to copy to name
7245  *
7246  *      Get device physical port name
7247  */
7248 int dev_get_phys_port_name(struct net_device *dev,
7249                            char *name, size_t len)
7250 {
7251         const struct net_device_ops *ops = dev->netdev_ops;
7252 
7253         if (!ops->ndo_get_phys_port_name)
7254                 return -EOPNOTSUPP;
7255         return ops->ndo_get_phys_port_name(dev, name, len);
7256 }
7257 EXPORT_SYMBOL(dev_get_phys_port_name);
7258 
7259 /**
7260  *      dev_change_proto_down - update protocol port state information
7261  *      @dev: device
7262  *      @proto_down: new value
7263  *
7264  *      This info can be used by switch drivers to set the phys state of the
7265  *      port.
7266  */
7267 int dev_change_proto_down(struct net_device *dev, bool proto_down)
7268 {
7269         const struct net_device_ops *ops = dev->netdev_ops;
7270 
7271         if (!ops->ndo_change_proto_down)
7272                 return -EOPNOTSUPP;
7273         if (!netif_device_present(dev))
7274                 return -ENODEV;
7275         return ops->ndo_change_proto_down(dev, proto_down);
7276 }
7277 EXPORT_SYMBOL(dev_change_proto_down);
7278 
7279 void __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op,
7280                      struct netdev_bpf *xdp)
7281 {
7282         memset(xdp, 0, sizeof(*xdp));
7283         xdp->command = XDP_QUERY_PROG;
7284 
7285         /* Query must always succeed. */
7286         WARN_ON(bpf_op(dev, xdp) < 0);
7287 }
7288 
7289 static u8 __dev_xdp_attached(struct net_device *dev, bpf_op_t bpf_op)
7290 {
7291         struct netdev_bpf xdp;
7292 
7293         __dev_xdp_query(dev, bpf_op, &xdp);
7294 
7295         return xdp.prog_attached;
7296 }
7297 
7298 static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op,
7299                            struct netlink_ext_ack *extack, u32 flags,
7300                            struct bpf_prog *prog)
7301 {
7302         struct netdev_bpf xdp;
7303 
7304         memset(&xdp, 0, sizeof(xdp));
7305         if (flags & XDP_FLAGS_HW_MODE)
7306                 xdp.command = XDP_SETUP_PROG_HW;
7307         else
7308                 xdp.command = XDP_SETUP_PROG;
7309         xdp.extack = extack;
7310         xdp.flags = flags;
7311         xdp.prog = prog;
7312 
7313         return bpf_op(dev, &xdp);
7314 }
7315 
7316 static void dev_xdp_uninstall(struct net_device *dev)
7317 {
7318         struct netdev_bpf xdp;
7319         bpf_op_t ndo_bpf;
7320 
7321         /* Remove generic XDP */
7322         WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL));
7323 
7324         /* Remove from the driver */
7325         ndo_bpf = dev->netdev_ops->ndo_bpf;
7326         if (!ndo_bpf)
7327                 return;
7328 
7329         __dev_xdp_query(dev, ndo_bpf, &xdp);
7330         if (xdp.prog_attached == XDP_ATTACHED_NONE)
7331                 return;
7332 
7333         /* Program removal should always succeed */
7334         WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, NULL));
7335 }
7336 
7337 /**
7338  *      dev_change_xdp_fd - set or clear a bpf program for a device rx path
7339  *      @dev: device
7340  *      @extack: netlink extended ack
7341  *      @fd: new program fd or negative value to clear
7342  *      @flags: xdp-related flags
7343  *
7344  *      Set or clear a bpf program for a device
7345  */
7346 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
7347                       int fd, u32 flags)
7348 {
7349         const struct net_device_ops *ops = dev->netdev_ops;
7350         struct bpf_prog *prog = NULL;
7351         bpf_op_t bpf_op, bpf_chk;
7352         int err;
7353 
7354         ASSERT_RTNL();
7355 
7356         bpf_op = bpf_chk = ops->ndo_bpf;
7357         if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE)))
7358                 return -EOPNOTSUPP;
7359         if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE))
7360                 bpf_op = generic_xdp_install;
7361         if (bpf_op == bpf_chk)
7362                 bpf_chk = generic_xdp_install;
7363 
7364         if (fd >= 0) {
7365                 if (bpf_chk && __dev_xdp_attached(dev, bpf_chk))
7366                         return -EEXIST;
7367                 if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) &&
7368                     __dev_xdp_attached(dev, bpf_op))
7369                         return -EBUSY;
7370 
7371                 prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP,
7372                                              bpf_op == ops->ndo_bpf);
7373                 if (IS_ERR(prog))
7374                         return PTR_ERR(prog);
7375 
7376                 if (!(flags & XDP_FLAGS_HW_MODE) &&
7377                     bpf_prog_is_dev_bound(prog->aux)) {
7378                         NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported");
7379                         bpf_prog_put(prog);
7380                         return -EINVAL;
7381                 }
7382         }
7383 
7384         err = dev_xdp_install(dev, bpf_op, extack, flags, prog);
7385         if (err < 0 && prog)
7386                 bpf_prog_put(prog);
7387 
7388         return err;
7389 }
7390 
7391 /**
7392  *      dev_new_index   -       allocate an ifindex
7393  *      @net: the applicable net namespace
7394  *
7395  *      Returns a suitable unique value for a new device interface
7396  *      number.  The caller must hold the rtnl semaphore or the
7397  *      dev_base_lock to be sure it remains unique.
7398  */
7399 static int dev_new_index(struct net *net)
7400 {
7401         int ifindex = net->ifindex;
7402 
7403         for (;;) {
7404                 if (++ifindex <= 0)
7405                         ifindex = 1;
7406                 if (!__dev_get_by_index(net, ifindex))
7407                         return net->ifindex = ifindex;
7408         }
7409 }
7410 
7411 /* Delayed registration/unregisteration */
7412 static LIST_HEAD(net_todo_list);
7413 DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq);
7414 
7415 static void net_set_todo(struct net_device *dev)
7416 {
7417         list_add_tail(&dev->todo_list, &net_todo_list);
7418         dev_net(dev)->dev_unreg_count++;
7419 }
7420 
7421 static void rollback_registered_many(struct list_head *head)
7422 {
7423         struct net_device *dev, *tmp;
7424         LIST_HEAD(close_head);
7425 
7426         BUG_ON(dev_boot_phase);
7427         ASSERT_RTNL();
7428 
7429         list_for_each_entry_safe(dev, tmp, head, unreg_list) {
7430                 /* Some devices call without registering
7431                  * for initialization unwind. Remove those
7432                  * devices and proceed with the remaining.
7433                  */
7434                 if (dev->reg_state == NETREG_UNINITIALIZED) {
7435                         pr_debug("unregister_netdevice: device %s/%p never was registered\n",
7436                                  dev->name, dev);
7437 
7438                         WARN_ON(1);
7439                         list_del(&dev->unreg_list);
7440                         continue;
7441                 }
7442                 dev->dismantle = true;
7443                 BUG_ON(dev->reg_state != NETREG_REGISTERED);
7444         }
7445 
7446         /* If device is running, close it first. */
7447         list_for_each_entry(dev, head, unreg_list)
7448                 list_add_tail(&dev->close_list, &close_head);
7449         dev_close_many(&close_head, true);
7450 
7451         list_for_each_entry(dev, head, unreg_list) {
7452                 /* And unlink it from device chain. */
7453                 unlist_netdevice(dev);
7454 
7455                 dev->reg_state = NETREG_UNREGISTERING;
7456         }
7457         flush_all_backlogs();
7458 
7459         synchronize_net();
7460 
7461         list_for_each_entry(dev, head, unreg_list) {
7462                 struct sk_buff *skb = NULL;
7463 
7464                 /* Shutdown queueing discipline. */
7465                 dev_shutdown(dev);
7466 
7467                 dev_xdp_uninstall(dev);
7468 
7469                 /* Notify protocols, that we are about to destroy
7470                  * this device. They should clean all the things.
7471                  */
7472                 call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
7473 
7474                 if (!dev->rtnl_link_ops ||
7475                     dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
7476                         skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0,
7477                                                      GFP_KERNEL, NULL, 0);
7478 
7479                 /*
7480                  *      Flush the unicast and multicast chains
7481                  */
7482                 dev_uc_flush(dev);
7483                 dev_mc_flush(dev);
7484 
7485                 if (dev->netdev_ops->ndo_uninit)
7486                         dev->netdev_ops->ndo_uninit(dev);
7487 
7488                 if (skb)
7489                         rtmsg_ifinfo_send(skb, dev, GFP_KERNEL);
7490 
7491                 /* Notifier chain MUST detach us all upper devices. */
7492                 WARN_ON(netdev_has_any_upper_dev(dev));
7493                 WARN_ON(netdev_has_any_lower_dev(dev));
7494 
7495                 /* Remove entries from kobject tree */
7496                 netdev_unregister_kobject(dev);
7497 #ifdef CONFIG_XPS
7498                 /* Remove XPS queueing entries */
7499                 netif_reset_xps_queues_gt(dev, 0);
7500 #endif
7501         }
7502 
7503         synchronize_net();
7504 
7505         list_for_each_entry(dev, head, unreg_list)
7506                 dev_put(dev);
7507 }
7508 
7509 static void rollback_registered(struct net_device *dev)
7510 {
7511         LIST_HEAD(single);
7512 
7513         list_add(&dev->unreg_list, &single);
7514         rollback_registered_many(&single);
7515         list_del(&single);
7516 }
7517 
7518 static netdev_features_t netdev_sync_upper_features(struct net_device *lower,
7519         struct net_device *upper, netdev_features_t features)
7520 {
7521         netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
7522         netdev_features_t feature;
7523         int feature_bit;
7524 
7525         for_each_netdev_feature(&upper_disables, feature_bit) {
7526                 feature = __NETIF_F_BIT(feature_bit);
7527                 if (!(upper->wanted_features & feature)
7528                     && (features & feature)) {
7529                         netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n",
7530                                    &feature, upper->name);
7531                         features &= ~feature;
7532                 }
7533         }
7534 
7535         return features;
7536 }
7537 
7538 static void netdev_sync_lower_features(struct net_device *upper,
7539         struct net_device *lower, netdev_features_t features)
7540 {
7541         netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES;
7542         netdev_features_t feature;
7543         int feature_bit;
7544 
7545         for_each_netdev_feature(&upper_disables, feature_bit) {
7546                 feature = __NETIF_F_BIT(feature_bit);
7547                 if (!(features & feature) && (lower->features & feature)) {
7548                         netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n",
7549                                    &feature, lower->name);
7550                         lower->wanted_features &= ~feature;
7551                         netdev_update_features(lower);
7552 
7553                         if (unlikely(lower->features & feature))
7554                                 netdev_WARN(upper, "failed to disable %pNF on %s!\n",
7555                                             &feature, lower->name);
7556                 }
7557         }
7558 }
7559 
7560 static netdev_features_t netdev_fix_features(struct net_device *dev,
7561         netdev_features_t features)
7562 {
7563         /* Fix illegal checksum combinations */
7564         if ((features & NETIF_F_HW_CSUM) &&
7565             (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) {
7566                 netdev_warn(dev, "mixed HW and IP checksum settings.\n");
7567                 features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM);
7568         }
7569 
7570         /* TSO requires that SG is present as well. */
7571         if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) {
7572                 netdev_dbg(dev, "Dropping TSO features since no SG feature.\n");
7573                 features &= ~NETIF_F_ALL_TSO;
7574         }
7575 
7576         if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) &&
7577                                         !(features & NETIF_F_IP_CSUM)) {
7578                 netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n");
7579                 features &= ~NETIF_F_TSO;
7580                 features &= ~NETIF_F_TSO_ECN;
7581         }
7582 
7583         if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) &&
7584                                          !(features & NETIF_F_IPV6_CSUM)) {
7585                 netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n");
7586                 features &= ~NETIF_F_TSO6;
7587         }
7588 
7589         /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */
7590         if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO))
7591                 features &= ~NETIF_F_TSO_MANGLEID;
7592 
7593         /* TSO ECN requires that TSO is present as well. */
7594         if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN)
7595                 features &= ~NETIF_F_TSO_ECN;
7596 
7597         /* Software GSO depends on SG. */
7598         if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) {
7599                 netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n");
7600                 features &= ~NETIF_F_GSO;
7601         }
7602 
7603         /* GSO partial features require GSO partial be set */
7604         if ((features & dev->gso_partial_features) &&
7605             !(features & NETIF_F_GSO_PARTIAL)) {
7606                 netdev_dbg(dev,
7607                            "Dropping partially supported GSO features since no GSO partial.\n");
7608                 features &= ~dev->gso_partial_features;
7609         }
7610 
7611         if (!(features & NETIF_F_RXCSUM)) {
7612                 /* NETIF_F_GRO_HW implies doing RXCSUM since every packet
7613                  * successfully merged by hardware must also have the
7614                  * checksum verified by hardware.  If the user does not
7615                  * want to enable RXCSUM, logically, we should disable GRO_HW.
7616                  */
7617                 if (features & NETIF_F_GRO_HW) {
7618                         netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n");
7619                         features &= ~NETIF_F_GRO_HW;
7620                 }
7621         }
7622 
7623         /* LRO/HW-GRO features cannot be combined with RX-FCS */
7624         if (features & NETIF_F_RXFCS) {
7625                 if (features & NETIF_F_LRO) {
7626                         netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n");
7627                         features &= ~NETIF_F_LRO;
7628                 }
7629 
7630                 if (features & NETIF_F_GRO_HW) {
7631                         netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n");
7632                         features &= ~NETIF_F_GRO_HW;
7633                 }
7634         }
7635 
7636         return features;
7637 }
7638 
7639 int __netdev_update_features(struct net_device *dev)
7640 {
7641         struct net_device *upper, *lower;
7642         netdev_features_t features;
7643         struct list_head *iter;
7644         int err = -1;
7645 
7646         ASSERT_RTNL();
7647 
7648         features = netdev_get_wanted_features(dev);
7649 
7650         if (dev->netdev_ops->ndo_fix_features)
7651                 features = dev->netdev_ops->ndo_fix_features(dev, features);
7652 
7653         /* driver might be less strict about feature dependencies */
7654         features = netdev_fix_features(dev, features);
7655 
7656         /* some features can't be enabled if they're off an an upper device */
7657         netdev_for_each_upper_dev_rcu(dev, upper, iter)
7658                 features = netdev_sync_upper_features(dev, upper, features);
7659 
7660         if (dev->features == features)
7661                 goto sync_lower;
7662 
7663         netdev_dbg(dev, "Features changed: %pNF -> %pNF\n",
7664                 &dev->features, &features);
7665 
7666         if (dev->netdev_ops->ndo_set_features)
7667                 err = dev->netdev_ops->ndo_set_features(dev, features);
7668         else
7669                 err = 0;
7670 
7671         if (unlikely(err < 0)) {
7672                 netdev_err(dev,
7673                         "set_features() failed (%d); wanted %pNF, left %pNF\n",
7674                         err, &features, &dev->features);
7675                 /* return non-0 since some features might have changed and
7676                  * it's better to fire a spurious notification than miss it
7677                  */
7678                 return -1;
7679         }
7680 
7681 sync_lower:
7682         /* some features must be disabled on lower devices when disabled
7683          * on an upper device (think: bonding master or bridge)
7684          */
7685         netdev_for_each_lower_dev(dev, lower, iter)
7686                 netdev_sync_lower_features(dev, lower, features);
7687 
7688         if (!err) {
7689                 netdev_features_t diff = features ^ dev->features;
7690 
7691                 if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) {
7692                         /* udp_tunnel_{get,drop}_rx_info both need
7693                          * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the
7694                          * device, or they won't do anything.
7695                          * Thus we need to update dev->features
7696                          * *before* calling udp_tunnel_get_rx_info,
7697                          * but *after* calling udp_tunnel_drop_rx_info.
7698                          */
7699                         if (features & NETIF_F_RX_UDP_TUNNEL_PORT) {
7700                                 dev->features = features;
7701                                 udp_tunnel_get_rx_info(dev);
7702                         } else {
7703                                 udp_tunnel_drop_rx_info(dev);
7704                         }
7705                 }
7706 
7707                 if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) {
7708                         if (features & NETIF_F_HW_VLAN_CTAG_FILTER) {
7709                                 dev->features = features;
7710                                 err |= vlan_get_rx_ctag_filter_info(dev);
7711                         } else {
7712                                 vlan_drop_rx_ctag_filter_info(dev);
7713                         }
7714                 }
7715 
7716                 if (diff & NETIF_F_HW_VLAN_STAG_FILTER) {
7717                         if (features & NETIF_F_HW_VLAN_STAG_FILTER) {
7718                                 dev->features = features;
7719                                 err |= vlan_get_rx_stag_filter_info(dev);
7720                         } else {
7721                                 vlan_drop_rx_stag_filter_info(dev);
7722                         }
7723                 }
7724 
7725                 dev->features = features;
7726         }
7727 
7728         return err < 0 ? 0 : 1;
7729 }
7730 
7731 /**
7732  *      netdev_update_features - recalculate device features
7733  *      @dev: the device to check
7734  *
7735  *      Recalculate dev->features set and send notifications if it
7736  *      has changed. Should be called after driver or hardware dependent
7737  *      conditions might have changed that influence the features.
7738  */
7739 void netdev_update_features(struct net_device *dev)
7740 {
7741         if (__netdev_update_features(dev))
7742                 netdev_features_change(dev);
7743 }
7744 EXPORT_SYMBOL(netdev_update_features);
7745 
7746 /**
7747  *      netdev_change_features - recalculate device features
7748  *      @dev: the device to check
7749  *
7750  *      Recalculate dev->features set and send notifications even
7751  *      if they have not changed. Should be called instead of
7752  *      netdev_update_features() if also dev->vlan_features might
7753  *      have changed to allow the changes to be propagated to stacked
7754  *      VLAN devices.
7755  */
7756 void netdev_change_features(struct net_device *dev)
7757 {
7758         __netdev_update_features(dev);
7759         netdev_features_change(dev);
7760 }
7761 EXPORT_SYMBOL(netdev_change_features);
7762 
7763 /**
7764  *      netif_stacked_transfer_operstate -      transfer operstate
7765  *      @rootdev: the root or lower level device to transfer state from
7766  *      @dev: the device to transfer operstate to
7767  *
7768  *      Transfer operational state from root to device. This is normally
7769  *      called when a stacking relationship exists between the root
7770  *      device and the device(a leaf device).
7771  */
7772 void netif_stacked_transfer_operstate(const struct net_device *rootdev,
7773                                         struct net_device *dev)
7774 {
7775         if (rootdev->operstate == IF_OPER_DORMANT)
7776                 netif_dormant_on(dev);
7777         else
7778                 netif_dormant_off(dev);
7779 
7780         if (netif_carrier_ok(rootdev))
7781                 netif_carrier_on(dev);
7782         else
7783                 netif_carrier_off(dev);
7784 }
7785 EXPORT_SYMBOL(netif_stacked_transfer_operstate);
7786 
7787 static int netif_alloc_rx_queues(struct net_device *dev)
7788 {
7789         unsigned int i, count = dev->num_rx_queues;
7790         struct netdev_rx_queue *rx;
7791         size_t sz = count * sizeof(*rx);
7792         int err = 0;
7793 
7794         BUG_ON(count < 1);
7795 
7796         rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7797         if (!rx)
7798                 return -ENOMEM;
7799 
7800         dev->_rx = rx;
7801 
7802         for (i = 0; i < count; i++) {
7803                 rx[i].dev = dev;
7804 
7805                 /* XDP RX-queue setup */
7806                 err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i);
7807                 if (err < 0)
7808                         goto err_rxq_info;
7809         }
7810         return 0;
7811 
7812 err_rxq_info:
7813         /* Rollback successful reg's and free other resources */
7814         while (i--)
7815                 xdp_rxq_info_unreg(&rx[i].xdp_rxq);
7816         kvfree(dev->_rx);
7817         dev->_rx = NULL;
7818         return err;
7819 }
7820 
7821 static void netif_free_rx_queues(struct net_device *dev)
7822 {
7823         unsigned int i, count = dev->num_rx_queues;
7824 
7825         /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */
7826         if (!dev->_rx)
7827                 return;
7828 
7829         for (i = 0; i < count; i++)
7830                 xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq);
7831 
7832         kvfree(dev->_rx);
7833 }
7834 
7835 static void netdev_init_one_queue(struct net_device *dev,
7836                                   struct netdev_queue *queue, void *_unused)
7837 {
7838         /* Initialize queue lock */
7839         spin_lock_init(&queue->_xmit_lock);
7840         netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type);
7841         queue->xmit_lock_owner = -1;
7842         netdev_queue_numa_node_write(queue, NUMA_NO_NODE);
7843         queue->dev = dev;
7844 #ifdef CONFIG_BQL
7845         dql_init(&queue->dql, HZ);
7846 #endif
7847 }
7848 
7849 static void netif_free_tx_queues(struct net_device *dev)
7850 {
7851         kvfree(dev->_tx);
7852 }
7853 
7854 static int netif_alloc_netdev_queues(struct net_device *dev)
7855 {
7856         unsigned int count = dev->num_tx_queues;
7857         struct netdev_queue *tx;
7858         size_t sz = count * sizeof(*tx);
7859 
7860         if (count < 1 || count > 0xffff)
7861                 return -EINVAL;
7862 
7863         tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL);
7864         if (!tx)
7865                 return -ENOMEM;
7866 
7867         dev->_tx = tx;
7868 
7869         netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
7870         spin_lock_init(&dev->tx_global_lock);
7871 
7872         return 0;
7873 }
7874 
7875 void netif_tx_stop_all_queues(struct net_device *dev)
7876 {
7877         unsigned int i;
7878 
7879         for (i = 0; i < dev->num_tx_queues; i++) {
7880                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
7881 
7882                 netif_tx_stop_queue(txq);
7883         }
7884 }
7885 EXPORT_SYMBOL(netif_tx_stop_all_queues);
7886 
7887 /**
7888  *      register_netdevice      - register a network device
7889  *      @dev: device to register
7890  *
7891  *      Take a completed network device structure and add it to the kernel
7892  *      interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
7893  *      chain. 0 is returned on success. A negative errno code is returned
7894  *      on a failure to set up the device, or if the name is a duplicate.
7895  *
7896  *      Callers must hold the rtnl semaphore. You may want
7897  *      register_netdev() instead of this.
7898  *
7899  *      BUGS:
7900  *      The locking appears insufficient to guarantee two parallel registers
7901  *      will not get the same name.
7902  */
7903 
7904 int register_netdevice(struct net_device *dev)
7905 {
7906         int ret;
7907         struct net *net = dev_net(dev);
7908 
7909         BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE <
7910                      NETDEV_FEATURE_COUNT);
7911         BUG_ON(dev_boot_phase);
7912         ASSERT_RTNL();
7913 
7914         might_sleep();
7915 
7916         /* When net_device's are persistent, this will be fatal. */
7917         BUG_ON(dev->reg_state != NETREG_UNINITIALIZED);
7918         BUG_ON(!net);
7919 
7920         spin_lock_init(&dev->addr_list_lock);
7921         netdev_set_addr_lockdep_class(dev);
7922 
7923         ret = dev_get_valid_name(net, dev, dev->name);
7924         if (ret < 0)
7925                 goto out;
7926 
7927         /* Init, if this function is available */
7928         if (dev->netdev_ops->ndo_init) {
7929                 ret = dev->netdev_ops->ndo_init(dev);
7930                 if (ret) {
7931                         if (ret > 0)
7932                                 ret = -EIO;
7933                         goto out;
7934                 }
7935         }
7936 
7937         if (((dev->hw_features | dev->features) &
7938              NETIF_F_HW_VLAN_CTAG_FILTER) &&
7939             (!dev->netdev_ops->ndo_vlan_rx_add_vid ||
7940              !dev->netdev_ops->ndo_vlan_rx_kill_vid)) {
7941                 netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n");
7942                 ret = -EINVAL;
7943                 goto err_uninit;
7944         }
7945 
7946         ret = -EBUSY;
7947         if (!dev->ifindex)
7948                 dev->ifindex = dev_new_index(net);
7949         else if (__dev_get_by_index(net, dev->ifindex))
7950                 goto err_uninit;
7951 
7952         /* Transfer changeable features to wanted_features and enable
7953          * software offloads (GSO and GRO).
7954          */
7955         dev->hw_features |= NETIF_F_SOFT_FEATURES;
7956         dev->features |= NETIF_F_SOFT_FEATURES;
7957 
7958         if (dev->netdev_ops->ndo_udp_tunnel_add) {
7959                 dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT;
7960                 dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT;
7961         }
7962 
7963         dev->wanted_features = dev->features & dev->hw_features;
7964 
7965         if (!(dev->flags & IFF_LOOPBACK))
7966                 dev->hw_features |= NETIF_F_NOCACHE_COPY;
7967 
7968         /* If IPv4 TCP segmentation offload is supported we should also
7969          * allow the device to enable segmenting the frame with the option
7970          * of ignoring a static IP ID value.  This doesn't enable the
7971          * feature itself but allows the user to enable it later.
7972          */
7973         if (dev->hw_features & NETIF_F_TSO)
7974                 dev->hw_features |= NETIF_F_TSO_MANGLEID;
7975         if (dev->vlan_features & NETIF_F_TSO)
7976                 dev->vlan_features |= NETIF_F_TSO_MANGLEID;
7977         if (dev->mpls_features & NETIF_F_TSO)
7978                 dev->mpls_features |= NETIF_F_TSO_MANGLEID;
7979         if (dev->hw_enc_features & NETIF_F_TSO)
7980                 dev->hw_enc_features |= NETIF_F_TSO_MANGLEID;
7981 
7982         /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
7983          */
7984         dev->vlan_features |= NETIF_F_HIGHDMA;
7985 
7986         /* Make NETIF_F_SG inheritable to tunnel devices.
7987          */
7988         dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL;
7989 
7990         /* Make NETIF_F_SG inheritable to MPLS.
7991          */
7992         dev->mpls_features |= NETIF_F_SG;
7993 
7994         ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev);
7995         ret = notifier_to_errno(ret);
7996         if (ret)
7997                 goto err_uninit;
7998 
7999         ret = netdev_register_kobject(dev);
8000         if (ret)
8001                 goto err_uninit;
8002         dev->reg_state = NETREG_REGISTERED;
8003 
8004         __netdev_update_features(dev);
8005 
8006         /*
8007          *      Default initial state at registry is that the
8008          *      device is present.
8009          */
8010 
8011         set_bit(__LINK_STATE_PRESENT, &dev->state);
8012 
8013         linkwatch_init_dev(dev);
8014 
8015         dev_init_scheduler(dev);
8016         dev_hold(dev);
8017         list_netdevice(dev);
8018         add_device_randomness(dev->dev_addr, dev->addr_len);
8019 
8020         /* If the device has permanent device address, driver should
8021          * set dev_addr and also addr_assign_type should be set to
8022          * NET_ADDR_PERM (default value).
8023          */
8024         if (dev->addr_assign_type == NET_ADDR_PERM)
8025                 memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
8026 
8027         /* Notify protocols, that a new device appeared. */
8028         ret = call_netdevice_notifiers(NETDEV_REGISTER, dev);
8029         ret = notifier_to_errno(ret);
8030         if (ret) {
8031                 rollback_registered(dev);
8032                 dev->reg_state = NETREG_UNREGISTERED;
8033         }
8034         /*
8035          *      Prevent userspace races by waiting until the network
8036          *      device is fully setup before sending notifications.
8037          */
8038         if (!dev->rtnl_link_ops ||
8039             dev->rtnl_link_state == RTNL_LINK_INITIALIZED)
8040                 rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL);
8041 
8042 out:
8043         return ret;
8044 
8045 err_uninit:
8046         if (dev->netdev_ops->ndo_uninit)
8047                 dev->netdev_ops->ndo_uninit(dev);
8048         if (dev->priv_destructor)
8049                 dev->priv_destructor(dev);
8050         goto out;
8051 }
8052 EXPORT_SYMBOL(register_netdevice);
8053 
8054 /**
8055  *      init_dummy_netdev       - init a dummy network device for NAPI
8056  *      @dev: device to init
8057  *
8058  *      This takes a network device structure and initialize the minimum
8059  *      amount of fields so it can be used to schedule NAPI polls without
8060  *      registering a full blown interface. This is to be used by drivers
8061  *      that need to tie several hardware interfaces to a single NAPI
8062  *      poll scheduler due to HW limitations.
8063  */
8064 int init_dummy_netdev(struct net_device *dev)
8065 {
8066         /* Clear everything. Note we don't initialize spinlocks
8067          * are they aren't supposed to be taken by any of the
8068          * NAPI code and this dummy netdev is supposed to be
8069          * only ever used for NAPI polls
8070          */
8071         memset(dev, 0, sizeof(struct net_device));
8072 
8073         /* make sure we BUG if trying to hit standard
8074          * register/unregister code path
8075          */
8076         dev->reg_state = NETREG_DUMMY;
8077 
8078         /* NAPI wants this */
8079         INIT_LIST_HEAD(&dev->napi_list);
8080 
8081         /* a dummy interface is started by default */
8082         set_bit(__LINK_STATE_PRESENT, &dev->state);
8083         set_bit(__LINK_STATE_START, &dev->state);
8084 
8085         /* Note : We dont allocate pcpu_refcnt for dummy devices,
8086          * because users of this 'device' dont need to change
8087          * its refcount.
8088          */
8089 
8090         return 0;
8091 }
8092 EXPORT_SYMBOL_GPL(init_dummy_netdev);
8093 
8094 
8095 /**
8096  *      register_netdev - register a network device
8097  *      @dev: device to register
8098  *
8099  *      Take a completed network device structure and add it to the kernel
8100  *      interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
8101  *      chain. 0 is returned on success. A negative errno code is returned
8102  *      on a failure to set up the device, or if the name is a duplicate.
8103  *
8104  *      This is a wrapper around register_netdevice that takes the rtnl semaphore
8105  *      and expands the device name if you passed a format string to
8106  *      alloc_netdev.
8107  */
8108 int register_netdev(struct net_device *dev)
8109 {
8110         int err;
8111 
8112         if (rtnl_lock_killable())
8113                 return -EINTR;
8114         err = register_netdevice(dev);
8115         rtnl_unlock();
8116         return err;
8117 }
8118 EXPORT_SYMBOL(register_netdev);
8119 
8120 int netdev_refcnt_read(const struct net_device *dev)
8121 {
8122         int i, refcnt = 0;
8123 
8124         for_each_possible_cpu(i)
8125                 refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i);
8126         return refcnt;
8127 }
8128 EXPORT_SYMBOL(netdev_refcnt_read);
8129 
8130 /**
8131  * netdev_wait_allrefs - wait until all references are gone.
8132  * @dev: target net_device
8133  *
8134  * This is called when unregistering network devices.
8135  *
8136  * Any protocol or device that holds a reference should register
8137  * for netdevice notification, and cleanup and put back the
8138  * reference if they receive an UNREGISTER event.
8139  * We can get stuck here if buggy protocols don't correctly
8140  * call dev_put.
8141  */
8142 static void netdev_wait_allrefs(struct net_device *dev)
8143 {
8144         unsigned long rebroadcast_time, warning_time;
8145         int refcnt;
8146 
8147         linkwatch_forget_dev(dev);
8148 
8149         rebroadcast_time = warning_time = jiffies;
8150         refcnt = netdev_refcnt_read(dev);
8151 
8152         while (refcnt != 0) {
8153                 if (time_after(jiffies, rebroadcast_time + 1 * HZ)) {
8154                         rtnl_lock();
8155 
8156                         /* Rebroadcast unregister notification */
8157                         call_netdevice_notifiers(NETDEV_UNREGISTER, dev);
8158 
8159                         __rtnl_unlock();
8160                         rcu_barrier();
8161                         rtnl_lock();
8162 
8163                         if (test_bit(__LINK_STATE_LINKWATCH_PENDING,
8164                                      &dev->state)) {
8165                                 /* We must not have linkwatch events
8166                                  * pending on unregister. If this
8167                                  * happens, we simply run the queue
8168                                  * unscheduled, resulting in a noop
8169                                  * for this device.
8170                                  */
8171                                 linkwatch_run_queue();
8172                         }
8173 
8174                         __rtnl_unlock();
8175 
8176                         rebroadcast_time = jiffies;
8177                 }
8178 
8179                 msleep(250);
8180 
8181                 refcnt = netdev_refcnt_read(dev);
8182 
8183                 if (time_after(jiffies, warning_time + 10 * HZ)) {
8184                         pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
8185                                  dev->name, refcnt);
8186                         warning_time = jiffies;
8187                 }
8188         }
8189 }
8190 
8191 /* The sequence is:
8192  *
8193  *      rtnl_lock();
8194  *      ...
8195  *      register_netdevice(x1);
8196  *      register_netdevice(x2);
8197  *      ...
8198  *      unregister_netdevice(y1);
8199  *      unregister_netdevice(y2);
8200  *      ...
8201  *      rtnl_unlock();
8202  *      free_netdev(y1);
8203  *      free_netdev(y2);
8204  *
8205  * We are invoked by rtnl_unlock().
8206  * This allows us to deal with problems:
8207  * 1) We can delete sysfs objects which invoke hotplug
8208  *    without deadlocking with linkwatch via keventd.
8209  * 2) Since we run with the RTNL semaphore not held, we can sleep
8210  *    safely in order to wait for the netdev refcnt to drop to zero.
8211  *
8212  * We must not return until all unregister events added during
8213  * the interval the lock was held have been completed.
8214  */
8215 void netdev_run_todo(void)
8216 {
8217         struct list_head list;
8218 
8219         /* Snapshot list, allow later requests */
8220         list_replace_init(&net_todo_list, &list);
8221 
8222         __rtnl_unlock();
8223 
8224 
8225         /* Wait for rcu callbacks to finish before next phase */
8226         if (!list_empty(&list))
8227                 rcu_barrier();
8228 
8229         while (!list_empty(&list)) {
8230                 struct net_device *dev
8231                         = list_first_entry(&list, struct net_device, todo_list);
8232                 list_del(&dev->todo_list);
8233 
8234                 if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) {
8235                         pr_err("network todo '%s' but state %d\n",
8236                                dev->name, dev->reg_state);
8237                         dump_stack();
8238                         continue;
8239                 }
8240 
8241                 dev->reg_state = NETREG_UNREGISTERED;
8242 
8243                 netdev_wait_allrefs(dev);
8244 
8245                 /* paranoia */
8246                 BUG_ON(netdev_refcnt_read(dev));
8247                 BUG_ON(!list_empty(&dev->ptype_all));
8248                 BUG_ON(!list_empty(&dev->ptype_specific));
8249                 WARN_ON(rcu_access_pointer(dev->ip_ptr));
8250                 WARN_ON(rcu_access_pointer(dev->ip6_ptr));
8251 #if IS_ENABLED(CONFIG_DECNET)
8252                 WARN_ON(dev->dn_ptr);
8253 #endif
8254                 if (dev->priv_destructor)
8255                         dev->priv_destructor(dev);
8256                 if (dev->needs_free_netdev)
8257                         free_netdev(dev);
8258 
8259                 /* Report a network device has been unregistered */
8260                 rtnl_lock();
8261                 dev_net(dev)->dev_unreg_count--;
8262                 __rtnl_unlock();
8263                 wake_up(&netdev_unregistering_wq);
8264 
8265                 /* Free network device */
8266                 kobject_put(&dev->dev.kobj);
8267         }
8268 }
8269 
8270 /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has
8271  * all the same fields in the same order as net_device_stats, with only
8272  * the type differing, but rtnl_link_stats64 may have additional fields
8273  * at the end for newer counters.
8274  */
8275 void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64,
8276                              const struct net_device_stats *netdev_stats)
8277 {
8278 #if BITS_PER_LONG == 64
8279         BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats));
8280         memcpy(stats64, netdev_stats, sizeof(*netdev_stats));
8281         /* zero out counters that only exist in rtnl_link_stats64 */
8282         memset((char *)stats64 + sizeof(*netdev_stats), 0,
8283                sizeof(*stats64) - sizeof(*netdev_stats));
8284 #else
8285         size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long);
8286         const unsigned long *src = (const unsigned long *)netdev_stats;
8287         u64 *dst = (u64 *)stats64;
8288 
8289         BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64));
8290         for (i = 0; i < n; i++)
8291                 dst[i] = src[i];
8292         /* zero out counters that only exist in rtnl_link_stats64 */
8293         memset((char *)stats64 + n * sizeof(u64), 0,
8294                sizeof(*stats64) - n * sizeof(u64));
8295 #endif
8296 }
8297 EXPORT_SYMBOL(netdev_stats_to_stats64);
8298 
8299 /**
8300  *      dev_get_stats   - get network device statistics
8301  *      @dev: device to get statistics from
8302  *      @storage: place to store stats
8303  *
8304  *      Get network statistics from device. Return @storage.
8305  *      The device driver may provide its own method by setting
8306  *      dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
8307  *      otherwise the internal statistics structure is used.
8308  */
8309 struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev,
8310                                         struct rtnl_link_stats64 *storage)
8311 {
8312         const struct net_device_ops *ops = dev->netdev_ops;
8313 
8314         if (ops->ndo_get_stats64) {
8315                 memset(storage, 0, sizeof(*storage));
8316                 ops->ndo_get_stats64(dev, storage);
8317         } else if (ops->ndo_get_stats) {
8318                 netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev));
8319         } else {
8320                 netdev_stats_to_stats64(storage, &dev->stats);
8321         }
8322         storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped);
8323         storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped);
8324         storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler);
8325         return storage;
8326 }
8327 EXPORT_SYMBOL(dev_get_stats);
8328 
8329 struct netdev_queue *dev_ingress_queue_create(struct net_device *dev)
8330 {
8331         struct netdev_queue *queue = dev_ingress_queue(dev);
8332 
8333 #ifdef CONFIG_NET_CLS_ACT
8334         if (queue)
8335                 return queue;
8336         queue = kzalloc(sizeof(*queue), GFP_KERNEL);
8337         if (!queue)
8338                 return NULL;
8339         netdev_init_one_queue(dev, queue, NULL);
8340         RCU_INIT_POINTER(queue->qdisc, &noop_qdisc);
8341         queue->qdisc_sleeping = &noop_qdisc;
8342         rcu_assign_pointer(dev->ingress_queue, queue);
8343 #endif
8344         return queue;
8345 }
8346 
8347 static const struct ethtool_ops default_ethtool_ops;
8348 
8349 void netdev_set_default_ethtool_ops(struct net_device *dev,
8350                                     const struct ethtool_ops *ops)
8351 {
8352         if (dev->ethtool_ops == &default_ethtool_ops)
8353                 dev->ethtool_ops = ops;
8354 }
8355 EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops);
8356 
8357 void netdev_freemem(struct net_device *dev)
8358 {
8359         char *addr = (char *)dev - dev->padded;
8360 
8361         kvfree(addr);
8362 }
8363 
8364 /**
8365  * alloc_netdev_mqs - allocate network device
8366  * @sizeof_priv: size of private data to allocate space for
8367  * @name: device name format string
8368  * @name_assign_type: origin of device name
8369  * @setup: callback to initialize device
8370  * @txqs: the number of TX subqueues to allocate
8371  * @rxqs: the number of RX subqueues to allocate
8372  *
8373  * Allocates a struct net_device with private data area for driver use
8374  * and performs basic initialization.  Also allocates subqueue structs
8375  * for each queue on the device.
8376  */
8377 struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name,
8378                 unsigned char name_assign_type,
8379                 void (*setup)(struct net_device *),
8380                 unsigned int txqs, unsigned int rxqs)
8381 {
8382         struct net_device *dev;
8383         unsigned int alloc_size;
8384         struct net_device *p;
8385 
8386         BUG_ON(strlen(name) >= sizeof(dev->name));
8387 
8388         if (txqs < 1) {
8389                 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
8390                 ret