]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
722c9a0c | 2 | * NET3 Protocol independent device support routines. |
1da177e4 LT |
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 | |
722c9a0c | 10 | * Authors: Ross Biro |
1da177e4 LT |
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 | |
722c9a0c | 24 | * to 2 if register_netdev gets called |
25 | * before net_dev_init & also removed a | |
26 | * few lines of code in the process. | |
1da177e4 LT |
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. | |
722c9a0c | 39 | * Alan Cox : Fix ETH_P_ALL echoback lengths. |
1da177e4 LT |
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. | |
722c9a0c | 49 | * Alan Cox : Fixed nasty side effect of device close |
1da177e4 LT |
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 | |
722c9a0c | 70 | * indefinitely on dev->refcnt |
71 | * J Hadi Salim : - Backlog queue sampling | |
1da177e4 LT |
72 | * - netif_rx() feedback |
73 | */ | |
74 | ||
7c0f6ba6 | 75 | #include <linux/uaccess.h> |
1da177e4 | 76 | #include <linux/bitops.h> |
4fc268d2 | 77 | #include <linux/capability.h> |
1da177e4 LT |
78 | #include <linux/cpu.h> |
79 | #include <linux/types.h> | |
80 | #include <linux/kernel.h> | |
08e9897d | 81 | #include <linux/hash.h> |
5a0e3ad6 | 82 | #include <linux/slab.h> |
1da177e4 | 83 | #include <linux/sched.h> |
f1083048 | 84 | #include <linux/sched/mm.h> |
4a3e2f71 | 85 | #include <linux/mutex.h> |
1da177e4 LT |
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> | |
0187bdfb | 95 | #include <linux/ethtool.h> |
1da177e4 LT |
96 | #include <linux/notifier.h> |
97 | #include <linux/skbuff.h> | |
a7862b45 | 98 | #include <linux/bpf.h> |
b5cdae32 | 99 | #include <linux/bpf_trace.h> |
457c4cbc | 100 | #include <net/net_namespace.h> |
1da177e4 | 101 | #include <net/sock.h> |
02d62e86 | 102 | #include <net/busy_poll.h> |
1da177e4 | 103 | #include <linux/rtnetlink.h> |
1da177e4 | 104 | #include <linux/stat.h> |
1da177e4 | 105 | #include <net/dst.h> |
fc4099f1 | 106 | #include <net/dst_metadata.h> |
1da177e4 | 107 | #include <net/pkt_sched.h> |
87d83093 | 108 | #include <net/pkt_cls.h> |
1da177e4 | 109 | #include <net/checksum.h> |
44540960 | 110 | #include <net/xfrm.h> |
1da177e4 LT |
111 | #include <linux/highmem.h> |
112 | #include <linux/init.h> | |
1da177e4 | 113 | #include <linux/module.h> |
1da177e4 LT |
114 | #include <linux/netpoll.h> |
115 | #include <linux/rcupdate.h> | |
116 | #include <linux/delay.h> | |
1da177e4 | 117 | #include <net/iw_handler.h> |
1da177e4 | 118 | #include <asm/current.h> |
5bdb9886 | 119 | #include <linux/audit.h> |
db217334 | 120 | #include <linux/dmaengine.h> |
f6a78bfc | 121 | #include <linux/err.h> |
c7fa9d18 | 122 | #include <linux/ctype.h> |
723e98b7 | 123 | #include <linux/if_arp.h> |
6de329e2 | 124 | #include <linux/if_vlan.h> |
8f0f2223 | 125 | #include <linux/ip.h> |
ad55dcaf | 126 | #include <net/ip.h> |
25cd9ba0 | 127 | #include <net/mpls.h> |
8f0f2223 DM |
128 | #include <linux/ipv6.h> |
129 | #include <linux/in.h> | |
b6b2fed1 DM |
130 | #include <linux/jhash.h> |
131 | #include <linux/random.h> | |
9cbc1cb8 | 132 | #include <trace/events/napi.h> |
cf66ba58 | 133 | #include <trace/events/net.h> |
07dc22e7 | 134 | #include <trace/events/skb.h> |
5acbbd42 | 135 | #include <linux/pci.h> |
caeda9b9 | 136 | #include <linux/inetdevice.h> |
c445477d | 137 | #include <linux/cpu_rmap.h> |
c5905afb | 138 | #include <linux/static_key.h> |
af12fa6e | 139 | #include <linux/hashtable.h> |
60877a32 | 140 | #include <linux/vmalloc.h> |
529d0489 | 141 | #include <linux/if_macvlan.h> |
e7fd2885 | 142 | #include <linux/errqueue.h> |
3b47d303 | 143 | #include <linux/hrtimer.h> |
e687ad60 | 144 | #include <linux/netfilter_ingress.h> |
40e4e713 | 145 | #include <linux/crash_dump.h> |
b72b5bf6 | 146 | #include <linux/sctp.h> |
ae847f40 | 147 | #include <net/udp_tunnel.h> |
1da177e4 | 148 | |
342709ef PE |
149 | #include "net-sysfs.h" |
150 | ||
d565b0a1 HX |
151 | /* Instead of increasing this, you should create a hash table. */ |
152 | #define MAX_GRO_SKBS 8 | |
153 | ||
5d38a079 HX |
154 | /* This should be increased if a protocol with a bigger head is added. */ |
155 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
156 | ||
1da177e4 | 157 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 158 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
159 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
160 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 161 | static struct list_head offload_base __read_mostly; |
1da177e4 | 162 | |
ae78dbfa | 163 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 LP |
164 | static int call_netdevice_notifiers_info(unsigned long val, |
165 | struct net_device *dev, | |
166 | struct netdev_notifier_info *info); | |
90b602f8 | 167 | static struct napi_struct *napi_by_id(unsigned int napi_id); |
ae78dbfa | 168 | |
1da177e4 | 169 | /* |
7562f876 | 170 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
171 | * semaphore. |
172 | * | |
c6d14c84 | 173 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
174 | * |
175 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 176 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
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 | */ | |
1da177e4 | 188 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
189 | EXPORT_SYMBOL(dev_base_lock); |
190 | ||
6c557001 FW |
191 | static DEFINE_MUTEX(ifalias_mutex); |
192 | ||
af12fa6e ET |
193 | /* protects napi_hash addition/deletion and napi_gen_id */ |
194 | static DEFINE_SPINLOCK(napi_hash_lock); | |
195 | ||
52bd2d62 | 196 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 197 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 198 | |
18afa4b0 | 199 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 200 | |
4e985ada TG |
201 | static inline void dev_base_seq_inc(struct net *net) |
202 | { | |
643aa9cb | 203 | while (++net->dev_base_seq == 0) |
204 | ; | |
4e985ada TG |
205 | } |
206 | ||
881d966b | 207 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 208 | { |
8387ff25 | 209 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 210 | |
08e9897d | 211 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
212 | } |
213 | ||
881d966b | 214 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 215 | { |
7c28bd0b | 216 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
217 | } |
218 | ||
e36fa2f7 | 219 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
220 | { |
221 | #ifdef CONFIG_RPS | |
e36fa2f7 | 222 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
223 | #endif |
224 | } | |
225 | ||
e36fa2f7 | 226 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
227 | { |
228 | #ifdef CONFIG_RPS | |
e36fa2f7 | 229 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
230 | #endif |
231 | } | |
232 | ||
ce286d32 | 233 | /* Device list insertion */ |
53759be9 | 234 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 235 | { |
c346dca1 | 236 | struct net *net = dev_net(dev); |
ce286d32 EB |
237 | |
238 | ASSERT_RTNL(); | |
239 | ||
240 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 241 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 242 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
243 | hlist_add_head_rcu(&dev->index_hlist, |
244 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 245 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
246 | |
247 | dev_base_seq_inc(net); | |
ce286d32 EB |
248 | } |
249 | ||
fb699dfd ED |
250 | /* Device list removal |
251 | * caller must respect a RCU grace period before freeing/reusing dev | |
252 | */ | |
ce286d32 EB |
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); | |
c6d14c84 | 259 | list_del_rcu(&dev->dev_list); |
72c9528b | 260 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 261 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 262 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
263 | |
264 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
265 | } |
266 | ||
1da177e4 LT |
267 | /* |
268 | * Our notifier list | |
269 | */ | |
270 | ||
f07d5b94 | 271 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
272 | |
273 | /* | |
274 | * Device drivers call our routines to queue packets here. We empty the | |
275 | * queue in the local softnet handler. | |
276 | */ | |
bea3348e | 277 | |
9958da05 | 278 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 279 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 280 | |
cf508b12 | 281 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 282 | /* |
c773e847 | 283 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
284 | * according to dev->type |
285 | */ | |
643aa9cb | 286 | static const unsigned short netdev_lock_type[] = { |
287 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
723e98b7 JP |
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, | |
211ed865 PG |
299 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
300 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
301 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 302 | |
643aa9cb | 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"}; | |
723e98b7 JP |
319 | |
320 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 321 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
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 | ||
cf508b12 DM |
334 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
335 | unsigned short dev_type) | |
723e98b7 JP |
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 | } | |
cf508b12 DM |
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 | } | |
723e98b7 | 353 | #else |
cf508b12 DM |
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) | |
723e98b7 JP |
359 | { |
360 | } | |
361 | #endif | |
1da177e4 LT |
362 | |
363 | /******************************************************************************* | |
eb13da1a | 364 | * |
365 | * Protocol management and registration routines | |
366 | * | |
367 | *******************************************************************************/ | |
1da177e4 | 368 | |
1da177e4 | 369 | |
1da177e4 LT |
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 | ||
c07b68e8 ED |
386 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
387 | { | |
388 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 389 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 390 | else |
7866a621 SN |
391 | return pt->dev ? &pt->dev->ptype_specific : |
392 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
393 | } |
394 | ||
1da177e4 LT |
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 | * | |
4ec93edb | 403 | * This call does not sleep therefore it can not |
1da177e4 LT |
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 | { | |
c07b68e8 | 410 | struct list_head *head = ptype_head(pt); |
1da177e4 | 411 | |
c07b68e8 ED |
412 | spin_lock(&ptype_lock); |
413 | list_add_rcu(&pt->list, head); | |
414 | spin_unlock(&ptype_lock); | |
1da177e4 | 415 | } |
d1b19dff | 416 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 417 | |
1da177e4 LT |
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 | |
4ec93edb | 425 | * returns. |
1da177e4 LT |
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 | { | |
c07b68e8 | 433 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
434 | struct packet_type *pt1; |
435 | ||
c07b68e8 | 436 | spin_lock(&ptype_lock); |
1da177e4 LT |
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 | ||
7b6cd1ce | 445 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 446 | out: |
c07b68e8 | 447 | spin_unlock(&ptype_lock); |
1da177e4 | 448 | } |
d1b19dff ED |
449 | EXPORT_SYMBOL(__dev_remove_pack); |
450 | ||
1da177e4 LT |
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); | |
4ec93edb | 466 | |
1da177e4 LT |
467 | synchronize_net(); |
468 | } | |
d1b19dff | 469 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 470 | |
62532da9 VY |
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 | { | |
bdef7de4 | 486 | struct packet_offload *elem; |
62532da9 VY |
487 | |
488 | spin_lock(&offload_lock); | |
bdef7de4 DM |
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); | |
62532da9 VY |
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 | */ | |
1d143d9f | 511 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
512 | { |
513 | struct list_head *head = &offload_base; | |
514 | struct packet_offload *po1; | |
515 | ||
c53aa505 | 516 | spin_lock(&offload_lock); |
62532da9 VY |
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: | |
c53aa505 | 527 | spin_unlock(&offload_lock); |
62532da9 | 528 | } |
62532da9 VY |
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 | ||
1da177e4 | 550 | /****************************************************************************** |
eb13da1a | 551 | * |
552 | * Device Boot-time Settings Routines | |
553 | * | |
554 | ******************************************************************************/ | |
1da177e4 LT |
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)); | |
93b3cff9 | 577 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
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 | /** | |
722c9a0c | 587 | * netdev_boot_setup_check - check boot time settings |
588 | * @dev: the netdevice | |
1da177e4 | 589 | * |
722c9a0c | 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. | |
1da177e4 LT |
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] != ' ' && | |
93b3cff9 | 602 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 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; | |
1da177e4 LT |
607 | return 1; |
608 | } | |
609 | } | |
610 | return 0; | |
611 | } | |
d1b19dff | 612 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
613 | |
614 | ||
615 | /** | |
722c9a0c | 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. | |
1da177e4 LT |
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 | */ | |
881d966b | 637 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
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 | /******************************************************************************* | |
eb13da1a | 676 | * |
677 | * Device Interface Subroutines | |
678 | * | |
679 | *******************************************************************************/ | |
1da177e4 | 680 | |
a54acb3a ND |
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 | ||
7a66bbc9 | 694 | return dev->ifindex; |
a54acb3a ND |
695 | } |
696 | EXPORT_SYMBOL(dev_get_iflink); | |
697 | ||
fc4099f1 PS |
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 | ||
1da177e4 LT |
724 | /** |
725 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 726 | * @net: the applicable net namespace |
1da177e4 LT |
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 | ||
881d966b | 736 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 737 | { |
0bd8d536 ED |
738 | struct net_device *dev; |
739 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 740 | |
b67bfe0d | 741 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
742 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
743 | return dev; | |
0bd8d536 | 744 | |
1da177e4 LT |
745 | return NULL; |
746 | } | |
d1b19dff | 747 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 748 | |
72c9528b | 749 | /** |
722c9a0c | 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. | |
72c9528b ED |
759 | */ |
760 | ||
761 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
762 | { | |
72c9528b ED |
763 | struct net_device *dev; |
764 | struct hlist_head *head = dev_name_hash(net, name); | |
765 | ||
b67bfe0d | 766 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
767 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
768 | return dev; | |
769 | ||
770 | return NULL; | |
771 | } | |
772 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
773 | ||
1da177e4 LT |
774 | /** |
775 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 776 | * @net: the applicable net namespace |
1da177e4 LT |
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 | ||
881d966b | 786 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
787 | { |
788 | struct net_device *dev; | |
789 | ||
72c9528b ED |
790 | rcu_read_lock(); |
791 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
792 | if (dev) |
793 | dev_hold(dev); | |
72c9528b | 794 | rcu_read_unlock(); |
1da177e4 LT |
795 | return dev; |
796 | } | |
d1b19dff | 797 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
798 | |
799 | /** | |
800 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 801 | * @net: the applicable net namespace |
1da177e4 LT |
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 | ||
881d966b | 811 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 812 | { |
0bd8d536 ED |
813 | struct net_device *dev; |
814 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 815 | |
b67bfe0d | 816 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
817 | if (dev->ifindex == ifindex) |
818 | return dev; | |
0bd8d536 | 819 | |
1da177e4 LT |
820 | return NULL; |
821 | } | |
d1b19dff | 822 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 823 | |
fb699dfd ED |
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 | { | |
fb699dfd ED |
837 | struct net_device *dev; |
838 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
839 | ||
b67bfe0d | 840 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
841 | if (dev->ifindex == ifindex) |
842 | return dev; | |
843 | ||
844 | return NULL; | |
845 | } | |
846 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
847 | ||
1da177e4 LT |
848 | |
849 | /** | |
850 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 851 | * @net: the applicable net namespace |
1da177e4 LT |
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 | ||
881d966b | 860 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
861 | { |
862 | struct net_device *dev; | |
863 | ||
fb699dfd ED |
864 | rcu_read_lock(); |
865 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
866 | if (dev) |
867 | dev_hold(dev); | |
fb699dfd | 868 | rcu_read_unlock(); |
1da177e4 LT |
869 | return dev; |
870 | } | |
d1b19dff | 871 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 872 | |
90b602f8 ML |
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 | ||
5dbe7c17 NS |
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 | ||
1da177e4 | 932 | /** |
941666c2 | 933 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 934 | * @net: the applicable net namespace |
1da177e4 LT |
935 | * @type: media type of device |
936 | * @ha: hardware address | |
937 | * | |
938 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
939 | * is not found or a pointer to the device. |
940 | * The caller must hold RCU or RTNL. | |
941666c2 | 941 | * The returned device has not had its ref count increased |
1da177e4 LT |
942 | * and the caller must therefore be careful about locking |
943 | * | |
1da177e4 LT |
944 | */ |
945 | ||
941666c2 ED |
946 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
947 | const char *ha) | |
1da177e4 LT |
948 | { |
949 | struct net_device *dev; | |
950 | ||
941666c2 | 951 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
952 | if (dev->type == type && |
953 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
954 | return dev; |
955 | ||
956 | return NULL; | |
1da177e4 | 957 | } |
941666c2 | 958 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 959 | |
881d966b | 960 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
961 | { |
962 | struct net_device *dev; | |
963 | ||
4e9cac2b | 964 | ASSERT_RTNL(); |
881d966b | 965 | for_each_netdev(net, dev) |
4e9cac2b | 966 | if (dev->type == type) |
7562f876 PE |
967 | return dev; |
968 | ||
969 | return NULL; | |
4e9cac2b | 970 | } |
4e9cac2b PM |
971 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
972 | ||
881d966b | 973 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 974 | { |
99fe3c39 | 975 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 976 | |
99fe3c39 ED |
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; | |
1da177e4 | 986 | } |
1da177e4 LT |
987 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
988 | ||
989 | /** | |
6c555490 | 990 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 991 | * @net: the applicable net namespace |
1da177e4 LT |
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 | |
bb69ae04 | 996 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 997 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
998 | */ |
999 | ||
6c555490 WC |
1000 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1001 | unsigned short mask) | |
1da177e4 | 1002 | { |
7562f876 | 1003 | struct net_device *dev, *ret; |
1da177e4 | 1004 | |
6c555490 WC |
1005 | ASSERT_RTNL(); |
1006 | ||
7562f876 | 1007 | ret = NULL; |
6c555490 | 1008 | for_each_netdev(net, dev) { |
1da177e4 | 1009 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1010 | ret = dev; |
1da177e4 LT |
1011 | break; |
1012 | } | |
1013 | } | |
7562f876 | 1014 | return ret; |
1da177e4 | 1015 | } |
6c555490 | 1016 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
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 | |
c7fa9d18 DM |
1023 | * to allow sysfs to work. We also disallow any kind of |
1024 | * whitespace. | |
1da177e4 | 1025 | */ |
95f050bf | 1026 | bool dev_valid_name(const char *name) |
1da177e4 | 1027 | { |
c7fa9d18 | 1028 | if (*name == '\0') |
95f050bf | 1029 | return false; |
b6fe17d6 | 1030 | if (strlen(name) >= IFNAMSIZ) |
95f050bf | 1031 | return false; |
c7fa9d18 | 1032 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1033 | return false; |
c7fa9d18 DM |
1034 | |
1035 | while (*name) { | |
a4176a93 | 1036 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1037 | return false; |
c7fa9d18 DM |
1038 | name++; |
1039 | } | |
95f050bf | 1040 | return true; |
1da177e4 | 1041 | } |
d1b19dff | 1042 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1043 | |
1044 | /** | |
b267b179 EB |
1045 | * __dev_alloc_name - allocate a name for a device |
1046 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1047 | * @name: name format string |
b267b179 | 1048 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1049 | * |
1050 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
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. | |
1da177e4 LT |
1057 | */ |
1058 | ||
b267b179 | 1059 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1060 | { |
1061 | int i = 0; | |
1da177e4 LT |
1062 | const char *p; |
1063 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1064 | unsigned long *inuse; |
1da177e4 LT |
1065 | struct net_device *d; |
1066 | ||
1067 | p = strnchr(name, IFNAMSIZ-1, '%'); | |
1068 | if (p) { | |
1069 | /* | |
1070 | * Verify the string as this thing may have come from | |
1071 | * the user. There must be either one "%d" and no other "%" | |
1072 | * characters. | |
1073 | */ | |
1074 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1075 | return -EINVAL; | |
1076 | ||
1077 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1078 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1079 | if (!inuse) |
1080 | return -ENOMEM; | |
1081 | ||
881d966b | 1082 | for_each_netdev(net, d) { |
1da177e4 LT |
1083 | if (!sscanf(d->name, name, &i)) |
1084 | continue; | |
1085 | if (i < 0 || i >= max_netdevices) | |
1086 | continue; | |
1087 | ||
1088 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1089 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1090 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1091 | set_bit(i, inuse); | |
1092 | } | |
1093 | ||
1094 | i = find_first_zero_bit(inuse, max_netdevices); | |
1095 | free_page((unsigned long) inuse); | |
1096 | } | |
1097 | ||
d9031024 OP |
1098 | if (buf != name) |
1099 | snprintf(buf, IFNAMSIZ, name, i); | |
b267b179 | 1100 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1101 | return i; |
1da177e4 LT |
1102 | |
1103 | /* It is possible to run out of possible slots | |
1104 | * when the name is long and there isn't enough space left | |
1105 | * for the digits, or if all bits are used. | |
1106 | */ | |
1107 | return -ENFILE; | |
1108 | } | |
1109 | ||
b267b179 EB |
1110 | /** |
1111 | * dev_alloc_name - allocate a name for a device | |
1112 | * @dev: device | |
1113 | * @name: name format string | |
1114 | * | |
1115 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1116 | * id. It scans list of devices to build up a free map, then chooses | |
1117 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1118 | * while allocating the name and adding the device in order to avoid | |
1119 | * duplicates. | |
1120 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1121 | * Returns the number of the unit assigned or a negative errno code. | |
1122 | */ | |
1123 | ||
1124 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1125 | { | |
1126 | char buf[IFNAMSIZ]; | |
1127 | struct net *net; | |
1128 | int ret; | |
1129 | ||
c346dca1 YH |
1130 | BUG_ON(!dev_net(dev)); |
1131 | net = dev_net(dev); | |
b267b179 EB |
1132 | ret = __dev_alloc_name(net, name, buf); |
1133 | if (ret >= 0) | |
1134 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1135 | return ret; | |
1136 | } | |
d1b19dff | 1137 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1138 | |
828de4f6 G |
1139 | static int dev_alloc_name_ns(struct net *net, |
1140 | struct net_device *dev, | |
1141 | const char *name) | |
d9031024 | 1142 | { |
828de4f6 G |
1143 | char buf[IFNAMSIZ]; |
1144 | int ret; | |
8ce6cebc | 1145 | |
828de4f6 G |
1146 | ret = __dev_alloc_name(net, name, buf); |
1147 | if (ret >= 0) | |
1148 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1149 | return ret; | |
1150 | } | |
1151 | ||
1152 | static int dev_get_valid_name(struct net *net, | |
1153 | struct net_device *dev, | |
1154 | const char *name) | |
1155 | { | |
1156 | BUG_ON(!net); | |
8ce6cebc | 1157 | |
d9031024 OP |
1158 | if (!dev_valid_name(name)) |
1159 | return -EINVAL; | |
1160 | ||
1c5cae81 | 1161 | if (strchr(name, '%')) |
828de4f6 | 1162 | return dev_alloc_name_ns(net, dev, name); |
d9031024 OP |
1163 | else if (__dev_get_by_name(net, name)) |
1164 | return -EEXIST; | |
8ce6cebc DL |
1165 | else if (dev->name != name) |
1166 | strlcpy(dev->name, name, IFNAMSIZ); | |
d9031024 OP |
1167 | |
1168 | return 0; | |
1169 | } | |
1da177e4 LT |
1170 | |
1171 | /** | |
1172 | * dev_change_name - change name of a device | |
1173 | * @dev: device | |
1174 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1175 | * | |
1176 | * Change name of a device, can pass format strings "eth%d". | |
1177 | * for wildcarding. | |
1178 | */ | |
cf04a4c7 | 1179 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1180 | { |
238fa362 | 1181 | unsigned char old_assign_type; |
fcc5a03a | 1182 | char oldname[IFNAMSIZ]; |
1da177e4 | 1183 | int err = 0; |
fcc5a03a | 1184 | int ret; |
881d966b | 1185 | struct net *net; |
1da177e4 LT |
1186 | |
1187 | ASSERT_RTNL(); | |
c346dca1 | 1188 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1189 | |
c346dca1 | 1190 | net = dev_net(dev); |
1da177e4 LT |
1191 | if (dev->flags & IFF_UP) |
1192 | return -EBUSY; | |
1193 | ||
30e6c9fa | 1194 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1195 | |
1196 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1197 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1198 | return 0; |
c91f6df2 | 1199 | } |
c8d90dca | 1200 | |
fcc5a03a HX |
1201 | memcpy(oldname, dev->name, IFNAMSIZ); |
1202 | ||
828de4f6 | 1203 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1204 | if (err < 0) { |
30e6c9fa | 1205 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1206 | return err; |
c91f6df2 | 1207 | } |
1da177e4 | 1208 | |
6fe82a39 VF |
1209 | if (oldname[0] && !strchr(oldname, '%')) |
1210 | netdev_info(dev, "renamed from %s\n", oldname); | |
1211 | ||
238fa362 TG |
1212 | old_assign_type = dev->name_assign_type; |
1213 | dev->name_assign_type = NET_NAME_RENAMED; | |
1214 | ||
fcc5a03a | 1215 | rollback: |
a1b3f594 EB |
1216 | ret = device_rename(&dev->dev, dev->name); |
1217 | if (ret) { | |
1218 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1219 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1220 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1221 | return ret; |
dcc99773 | 1222 | } |
7f988eab | 1223 | |
30e6c9fa | 1224 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1225 | |
5bb025fa VF |
1226 | netdev_adjacent_rename_links(dev, oldname); |
1227 | ||
7f988eab | 1228 | write_lock_bh(&dev_base_lock); |
372b2312 | 1229 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1230 | write_unlock_bh(&dev_base_lock); |
1231 | ||
1232 | synchronize_rcu(); | |
1233 | ||
1234 | write_lock_bh(&dev_base_lock); | |
1235 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1236 | write_unlock_bh(&dev_base_lock); |
1237 | ||
056925ab | 1238 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1239 | ret = notifier_to_errno(ret); |
1240 | ||
1241 | if (ret) { | |
91e9c07b ED |
1242 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1243 | if (err >= 0) { | |
fcc5a03a | 1244 | err = ret; |
30e6c9fa | 1245 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1246 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1247 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1248 | dev->name_assign_type = old_assign_type; |
1249 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1250 | goto rollback; |
91e9c07b | 1251 | } else { |
7b6cd1ce | 1252 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1253 | dev->name, ret); |
fcc5a03a HX |
1254 | } |
1255 | } | |
1da177e4 LT |
1256 | |
1257 | return err; | |
1258 | } | |
1259 | ||
0b815a1a SH |
1260 | /** |
1261 | * dev_set_alias - change ifalias of a device | |
1262 | * @dev: device | |
1263 | * @alias: name up to IFALIASZ | |
f0db275a | 1264 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1265 | * |
1266 | * Set ifalias for a device, | |
1267 | */ | |
1268 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1269 | { | |
6c557001 | 1270 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1271 | |
1272 | if (len >= IFALIASZ) | |
1273 | return -EINVAL; | |
1274 | ||
6c557001 FW |
1275 | if (len) { |
1276 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1277 | if (!new_alias) | |
1278 | return -ENOMEM; | |
1279 | ||
1280 | memcpy(new_alias->ifalias, alias, len); | |
1281 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1282 | } |
1283 | ||
6c557001 FW |
1284 | mutex_lock(&ifalias_mutex); |
1285 | rcu_swap_protected(dev->ifalias, new_alias, | |
1286 | mutex_is_locked(&ifalias_mutex)); | |
1287 | mutex_unlock(&ifalias_mutex); | |
1288 | ||
1289 | if (new_alias) | |
1290 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1291 | |
0b815a1a SH |
1292 | return len; |
1293 | } | |
1294 | ||
6c557001 FW |
1295 | /** |
1296 | * dev_get_alias - get ifalias of a device | |
1297 | * @dev: device | |
1298 | * @alias: buffer to store name of ifalias | |
1299 | * @len: size of buffer | |
1300 | * | |
1301 | * get ifalias for a device. Caller must make sure dev cannot go | |
1302 | * away, e.g. rcu read lock or own a reference count to device. | |
1303 | */ | |
1304 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1305 | { | |
1306 | const struct dev_ifalias *alias; | |
1307 | int ret = 0; | |
1308 | ||
1309 | rcu_read_lock(); | |
1310 | alias = rcu_dereference(dev->ifalias); | |
1311 | if (alias) | |
1312 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1313 | rcu_read_unlock(); | |
1314 | ||
1315 | return ret; | |
1316 | } | |
0b815a1a | 1317 | |
d8a33ac4 | 1318 | /** |
3041a069 | 1319 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1320 | * @dev: device to cause notification |
1321 | * | |
1322 | * Called to indicate a device has changed features. | |
1323 | */ | |
1324 | void netdev_features_change(struct net_device *dev) | |
1325 | { | |
056925ab | 1326 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1327 | } |
1328 | EXPORT_SYMBOL(netdev_features_change); | |
1329 | ||
1da177e4 LT |
1330 | /** |
1331 | * netdev_state_change - device changes state | |
1332 | * @dev: device to cause notification | |
1333 | * | |
1334 | * Called to indicate a device has changed state. This function calls | |
1335 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1336 | * to the routing socket. | |
1337 | */ | |
1338 | void netdev_state_change(struct net_device *dev) | |
1339 | { | |
1340 | if (dev->flags & IFF_UP) { | |
54951194 LP |
1341 | struct netdev_notifier_change_info change_info; |
1342 | ||
1343 | change_info.flags_changed = 0; | |
1344 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
1345 | &change_info.info); | |
7f294054 | 1346 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1347 | } |
1348 | } | |
d1b19dff | 1349 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1350 | |
ee89bab1 | 1351 | /** |
722c9a0c | 1352 | * netdev_notify_peers - notify network peers about existence of @dev |
1353 | * @dev: network device | |
ee89bab1 AW |
1354 | * |
1355 | * Generate traffic such that interested network peers are aware of | |
1356 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1357 | * a device wants to inform the rest of the network about some sort of | |
1358 | * reconfiguration such as a failover event or virtual machine | |
1359 | * migration. | |
1360 | */ | |
1361 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1362 | { |
ee89bab1 AW |
1363 | rtnl_lock(); |
1364 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1365 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1366 | rtnl_unlock(); |
c1da4ac7 | 1367 | } |
ee89bab1 | 1368 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1369 | |
bd380811 | 1370 | static int __dev_open(struct net_device *dev) |
1da177e4 | 1371 | { |
d314774c | 1372 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1373 | int ret; |
1da177e4 | 1374 | |
e46b66bc BH |
1375 | ASSERT_RTNL(); |
1376 | ||
1da177e4 LT |
1377 | if (!netif_device_present(dev)) |
1378 | return -ENODEV; | |
1379 | ||
ca99ca14 NH |
1380 | /* Block netpoll from trying to do any rx path servicing. |
1381 | * If we don't do this there is a chance ndo_poll_controller | |
1382 | * or ndo_poll may be running while we open the device | |
1383 | */ | |
66b5552f | 1384 | netpoll_poll_disable(dev); |
ca99ca14 | 1385 | |
3b8bcfd5 JB |
1386 | ret = call_netdevice_notifiers(NETDEV_PRE_UP, dev); |
1387 | ret = notifier_to_errno(ret); | |
1388 | if (ret) | |
1389 | return ret; | |
1390 | ||
1da177e4 | 1391 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1392 | |
d314774c SH |
1393 | if (ops->ndo_validate_addr) |
1394 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1395 | |
d314774c SH |
1396 | if (!ret && ops->ndo_open) |
1397 | ret = ops->ndo_open(dev); | |
1da177e4 | 1398 | |
66b5552f | 1399 | netpoll_poll_enable(dev); |
ca99ca14 | 1400 | |
bada339b JG |
1401 | if (ret) |
1402 | clear_bit(__LINK_STATE_START, &dev->state); | |
1403 | else { | |
1da177e4 | 1404 | dev->flags |= IFF_UP; |
4417da66 | 1405 | dev_set_rx_mode(dev); |
1da177e4 | 1406 | dev_activate(dev); |
7bf23575 | 1407 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1408 | } |
bada339b | 1409 | |
1da177e4 LT |
1410 | return ret; |
1411 | } | |
1412 | ||
1413 | /** | |
bd380811 PM |
1414 | * dev_open - prepare an interface for use. |
1415 | * @dev: device to open | |
1da177e4 | 1416 | * |
bd380811 PM |
1417 | * Takes a device from down to up state. The device's private open |
1418 | * function is invoked and then the multicast lists are loaded. Finally | |
1419 | * the device is moved into the up state and a %NETDEV_UP message is | |
1420 | * sent to the netdev notifier chain. | |
1421 | * | |
1422 | * Calling this function on an active interface is a nop. On a failure | |
1423 | * a negative errno code is returned. | |
1da177e4 | 1424 | */ |
bd380811 PM |
1425 | int dev_open(struct net_device *dev) |
1426 | { | |
1427 | int ret; | |
1428 | ||
bd380811 PM |
1429 | if (dev->flags & IFF_UP) |
1430 | return 0; | |
1431 | ||
bd380811 PM |
1432 | ret = __dev_open(dev); |
1433 | if (ret < 0) | |
1434 | return ret; | |
1435 | ||
7f294054 | 1436 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
bd380811 PM |
1437 | call_netdevice_notifiers(NETDEV_UP, dev); |
1438 | ||
1439 | return ret; | |
1440 | } | |
1441 | EXPORT_SYMBOL(dev_open); | |
1442 | ||
7051b88a | 1443 | static void __dev_close_many(struct list_head *head) |
1da177e4 | 1444 | { |
44345724 | 1445 | struct net_device *dev; |
e46b66bc | 1446 | |
bd380811 | 1447 | ASSERT_RTNL(); |
9d5010db DM |
1448 | might_sleep(); |
1449 | ||
5cde2829 | 1450 | list_for_each_entry(dev, head, close_list) { |
3f4df206 | 1451 | /* Temporarily disable netpoll until the interface is down */ |
66b5552f | 1452 | netpoll_poll_disable(dev); |
3f4df206 | 1453 | |
44345724 | 1454 | call_netdevice_notifiers(NETDEV_GOING_DOWN, dev); |
1da177e4 | 1455 | |
44345724 | 1456 | clear_bit(__LINK_STATE_START, &dev->state); |
1da177e4 | 1457 | |
44345724 OP |
1458 | /* Synchronize to scheduled poll. We cannot touch poll list, it |
1459 | * can be even on different cpu. So just clear netif_running(). | |
1460 | * | |
1461 | * dev->stop() will invoke napi_disable() on all of it's | |
1462 | * napi_struct instances on this device. | |
1463 | */ | |
4e857c58 | 1464 | smp_mb__after_atomic(); /* Commit netif_running(). */ |
44345724 | 1465 | } |
1da177e4 | 1466 | |
44345724 | 1467 | dev_deactivate_many(head); |
d8b2a4d2 | 1468 | |
5cde2829 | 1469 | list_for_each_entry(dev, head, close_list) { |
44345724 | 1470 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 | 1471 | |
44345724 OP |
1472 | /* |
1473 | * Call the device specific close. This cannot fail. | |
1474 | * Only if device is UP | |
1475 | * | |
1476 | * We allow it to be called even after a DETACH hot-plug | |
1477 | * event. | |
1478 | */ | |
1479 | if (ops->ndo_stop) | |
1480 | ops->ndo_stop(dev); | |
1481 | ||
44345724 | 1482 | dev->flags &= ~IFF_UP; |
66b5552f | 1483 | netpoll_poll_enable(dev); |
44345724 | 1484 | } |
44345724 OP |
1485 | } |
1486 | ||
7051b88a | 1487 | static void __dev_close(struct net_device *dev) |
44345724 OP |
1488 | { |
1489 | LIST_HEAD(single); | |
1490 | ||
5cde2829 | 1491 | list_add(&dev->close_list, &single); |
7051b88a | 1492 | __dev_close_many(&single); |
f87e6f47 | 1493 | list_del(&single); |
44345724 OP |
1494 | } |
1495 | ||
7051b88a | 1496 | void dev_close_many(struct list_head *head, bool unlink) |
44345724 OP |
1497 | { |
1498 | struct net_device *dev, *tmp; | |
1da177e4 | 1499 | |
5cde2829 EB |
1500 | /* Remove the devices that don't need to be closed */ |
1501 | list_for_each_entry_safe(dev, tmp, head, close_list) | |
44345724 | 1502 | if (!(dev->flags & IFF_UP)) |
5cde2829 | 1503 | list_del_init(&dev->close_list); |
44345724 OP |
1504 | |
1505 | __dev_close_many(head); | |
1da177e4 | 1506 | |
5cde2829 | 1507 | list_for_each_entry_safe(dev, tmp, head, close_list) { |
7f294054 | 1508 | rtmsg_ifinfo(RTM_NEWLINK, dev, IFF_UP|IFF_RUNNING, GFP_KERNEL); |
44345724 | 1509 | call_netdevice_notifiers(NETDEV_DOWN, dev); |
99c4a26a DM |
1510 | if (unlink) |
1511 | list_del_init(&dev->close_list); | |
44345724 | 1512 | } |
bd380811 | 1513 | } |
99c4a26a | 1514 | EXPORT_SYMBOL(dev_close_many); |
bd380811 PM |
1515 | |
1516 | /** | |
1517 | * dev_close - shutdown an interface. | |
1518 | * @dev: device to shutdown | |
1519 | * | |
1520 | * This function moves an active device into down state. A | |
1521 | * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device | |
1522 | * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier | |
1523 | * chain. | |
1524 | */ | |
7051b88a | 1525 | void dev_close(struct net_device *dev) |
bd380811 | 1526 | { |
e14a5993 ED |
1527 | if (dev->flags & IFF_UP) { |
1528 | LIST_HEAD(single); | |
1da177e4 | 1529 | |
5cde2829 | 1530 | list_add(&dev->close_list, &single); |
99c4a26a | 1531 | dev_close_many(&single, true); |
e14a5993 ED |
1532 | list_del(&single); |
1533 | } | |
1da177e4 | 1534 | } |
d1b19dff | 1535 | EXPORT_SYMBOL(dev_close); |
1da177e4 LT |
1536 | |
1537 | ||
0187bdfb BH |
1538 | /** |
1539 | * dev_disable_lro - disable Large Receive Offload on a device | |
1540 | * @dev: device | |
1541 | * | |
1542 | * Disable Large Receive Offload (LRO) on a net device. Must be | |
1543 | * called under RTNL. This is needed if received packets may be | |
1544 | * forwarded to another interface. | |
1545 | */ | |
1546 | void dev_disable_lro(struct net_device *dev) | |
1547 | { | |
fbe168ba MK |
1548 | struct net_device *lower_dev; |
1549 | struct list_head *iter; | |
529d0489 | 1550 | |
bc5787c6 MM |
1551 | dev->wanted_features &= ~NETIF_F_LRO; |
1552 | netdev_update_features(dev); | |
27660515 | 1553 | |
22d5969f MM |
1554 | if (unlikely(dev->features & NETIF_F_LRO)) |
1555 | netdev_WARN(dev, "failed to disable LRO!\n"); | |
fbe168ba MK |
1556 | |
1557 | netdev_for_each_lower_dev(dev, lower_dev, iter) | |
1558 | dev_disable_lro(lower_dev); | |
0187bdfb BH |
1559 | } |
1560 | EXPORT_SYMBOL(dev_disable_lro); | |
1561 | ||
351638e7 JP |
1562 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1563 | struct net_device *dev) | |
1564 | { | |
1565 | struct netdev_notifier_info info; | |
1566 | ||
1567 | netdev_notifier_info_init(&info, dev); | |
1568 | return nb->notifier_call(nb, val, &info); | |
1569 | } | |
0187bdfb | 1570 | |
881d966b EB |
1571 | static int dev_boot_phase = 1; |
1572 | ||
1da177e4 | 1573 | /** |
722c9a0c | 1574 | * register_netdevice_notifier - register a network notifier block |
1575 | * @nb: notifier | |
1da177e4 | 1576 | * |
722c9a0c | 1577 | * Register a notifier to be called when network device events occur. |
1578 | * The notifier passed is linked into the kernel structures and must | |
1579 | * not be reused until it has been unregistered. A negative errno code | |
1580 | * is returned on a failure. | |
1da177e4 | 1581 | * |
722c9a0c | 1582 | * When registered all registration and up events are replayed |
1583 | * to the new notifier to allow device to have a race free | |
1584 | * view of the network device list. | |
1da177e4 LT |
1585 | */ |
1586 | ||
1587 | int register_netdevice_notifier(struct notifier_block *nb) | |
1588 | { | |
1589 | struct net_device *dev; | |
fcc5a03a | 1590 | struct net_device *last; |
881d966b | 1591 | struct net *net; |
1da177e4 LT |
1592 | int err; |
1593 | ||
1594 | rtnl_lock(); | |
f07d5b94 | 1595 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1596 | if (err) |
1597 | goto unlock; | |
881d966b EB |
1598 | if (dev_boot_phase) |
1599 | goto unlock; | |
1600 | for_each_net(net) { | |
1601 | for_each_netdev(net, dev) { | |
351638e7 | 1602 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1603 | err = notifier_to_errno(err); |
1604 | if (err) | |
1605 | goto rollback; | |
1606 | ||
1607 | if (!(dev->flags & IFF_UP)) | |
1608 | continue; | |
1da177e4 | 1609 | |
351638e7 | 1610 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1611 | } |
1da177e4 | 1612 | } |
fcc5a03a HX |
1613 | |
1614 | unlock: | |
1da177e4 LT |
1615 | rtnl_unlock(); |
1616 | return err; | |
fcc5a03a HX |
1617 | |
1618 | rollback: | |
1619 | last = dev; | |
881d966b EB |
1620 | for_each_net(net) { |
1621 | for_each_netdev(net, dev) { | |
1622 | if (dev == last) | |
8f891489 | 1623 | goto outroll; |
fcc5a03a | 1624 | |
881d966b | 1625 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1626 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1627 | dev); | |
1628 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1629 | } |
351638e7 | 1630 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1631 | } |
fcc5a03a | 1632 | } |
c67625a1 | 1633 | |
8f891489 | 1634 | outroll: |
c67625a1 | 1635 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1636 | goto unlock; |
1da177e4 | 1637 | } |
d1b19dff | 1638 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1639 | |
1640 | /** | |
722c9a0c | 1641 | * unregister_netdevice_notifier - unregister a network notifier block |
1642 | * @nb: notifier | |
1da177e4 | 1643 | * |
722c9a0c | 1644 | * Unregister a notifier previously registered by |
1645 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1646 | * kernel structures and may then be reused. A negative errno code | |
1647 | * is returned on a failure. | |
7d3d43da | 1648 | * |
722c9a0c | 1649 | * After unregistering unregister and down device events are synthesized |
1650 | * for all devices on the device list to the removed notifier to remove | |
1651 | * the need for special case cleanup code. | |
1da177e4 LT |
1652 | */ |
1653 | ||
1654 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1655 | { | |
7d3d43da EB |
1656 | struct net_device *dev; |
1657 | struct net *net; | |
9f514950 HX |
1658 | int err; |
1659 | ||
1660 | rtnl_lock(); | |
f07d5b94 | 1661 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1662 | if (err) |
1663 | goto unlock; | |
1664 | ||
1665 | for_each_net(net) { | |
1666 | for_each_netdev(net, dev) { | |
1667 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1668 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1669 | dev); | |
1670 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1671 | } |
351638e7 | 1672 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1673 | } |
1674 | } | |
1675 | unlock: | |
9f514950 HX |
1676 | rtnl_unlock(); |
1677 | return err; | |
1da177e4 | 1678 | } |
d1b19dff | 1679 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1680 | |
351638e7 JP |
1681 | /** |
1682 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1683 | * @val: value passed unmodified to notifier function | |
1684 | * @dev: net_device pointer passed unmodified to notifier function | |
1685 | * @info: notifier information data | |
1686 | * | |
1687 | * Call all network notifier blocks. Parameters and return value | |
1688 | * are as for raw_notifier_call_chain(). | |
1689 | */ | |
1690 | ||
1d143d9f | 1691 | static int call_netdevice_notifiers_info(unsigned long val, |
1692 | struct net_device *dev, | |
1693 | struct netdev_notifier_info *info) | |
351638e7 JP |
1694 | { |
1695 | ASSERT_RTNL(); | |
1696 | netdev_notifier_info_init(info, dev); | |
1697 | return raw_notifier_call_chain(&netdev_chain, val, info); | |
1698 | } | |
351638e7 | 1699 | |
1da177e4 LT |
1700 | /** |
1701 | * call_netdevice_notifiers - call all network notifier blocks | |
1702 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1703 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1704 | * |
1705 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1706 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1707 | */ |
1708 | ||
ad7379d4 | 1709 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1710 | { |
351638e7 JP |
1711 | struct netdev_notifier_info info; |
1712 | ||
1713 | return call_netdevice_notifiers_info(val, dev, &info); | |
1da177e4 | 1714 | } |
edf947f1 | 1715 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1716 | |
1cf51900 | 1717 | #ifdef CONFIG_NET_INGRESS |
4577139b DB |
1718 | static struct static_key ingress_needed __read_mostly; |
1719 | ||
1720 | void net_inc_ingress_queue(void) | |
1721 | { | |
1722 | static_key_slow_inc(&ingress_needed); | |
1723 | } | |
1724 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1725 | ||
1726 | void net_dec_ingress_queue(void) | |
1727 | { | |
1728 | static_key_slow_dec(&ingress_needed); | |
1729 | } | |
1730 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1731 | #endif | |
1732 | ||
1f211a1b DB |
1733 | #ifdef CONFIG_NET_EGRESS |
1734 | static struct static_key egress_needed __read_mostly; | |
1735 | ||
1736 | void net_inc_egress_queue(void) | |
1737 | { | |
1738 | static_key_slow_inc(&egress_needed); | |
1739 | } | |
1740 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1741 | ||
1742 | void net_dec_egress_queue(void) | |
1743 | { | |
1744 | static_key_slow_dec(&egress_needed); | |
1745 | } | |
1746 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1747 | #endif | |
1748 | ||
c5905afb | 1749 | static struct static_key netstamp_needed __read_mostly; |
b90e5794 | 1750 | #ifdef HAVE_JUMP_LABEL |
b90e5794 | 1751 | static atomic_t netstamp_needed_deferred; |
13baa00a | 1752 | static atomic_t netstamp_wanted; |
5fa8bbda | 1753 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 1754 | { |
b90e5794 | 1755 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 1756 | int wanted; |
b90e5794 | 1757 | |
13baa00a ED |
1758 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
1759 | if (wanted > 0) | |
1760 | static_key_enable(&netstamp_needed); | |
1761 | else | |
1762 | static_key_disable(&netstamp_needed); | |
5fa8bbda ED |
1763 | } |
1764 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 1765 | #endif |
5fa8bbda ED |
1766 | |
1767 | void net_enable_timestamp(void) | |
1768 | { | |
13baa00a ED |
1769 | #ifdef HAVE_JUMP_LABEL |
1770 | int wanted; | |
1771 | ||
1772 | while (1) { | |
1773 | wanted = atomic_read(&netstamp_wanted); | |
1774 | if (wanted <= 0) | |
1775 | break; | |
1776 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
1777 | return; | |
1778 | } | |
1779 | atomic_inc(&netstamp_needed_deferred); | |
1780 | schedule_work(&netstamp_work); | |
1781 | #else | |
c5905afb | 1782 | static_key_slow_inc(&netstamp_needed); |
13baa00a | 1783 | #endif |
1da177e4 | 1784 | } |
d1b19dff | 1785 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1786 | |
1787 | void net_disable_timestamp(void) | |
1788 | { | |
b90e5794 | 1789 | #ifdef HAVE_JUMP_LABEL |
13baa00a ED |
1790 | int wanted; |
1791 | ||
1792 | while (1) { | |
1793 | wanted = atomic_read(&netstamp_wanted); | |
1794 | if (wanted <= 1) | |
1795 | break; | |
1796 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
1797 | return; | |
1798 | } | |
1799 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
1800 | schedule_work(&netstamp_work); |
1801 | #else | |
c5905afb | 1802 | static_key_slow_dec(&netstamp_needed); |
5fa8bbda | 1803 | #endif |
1da177e4 | 1804 | } |
d1b19dff | 1805 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1806 | |
3b098e2d | 1807 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1808 | { |
2456e855 | 1809 | skb->tstamp = 0; |
c5905afb | 1810 | if (static_key_false(&netstamp_needed)) |
a61bbcf2 | 1811 | __net_timestamp(skb); |
1da177e4 LT |
1812 | } |
1813 | ||
588f0330 | 1814 | #define net_timestamp_check(COND, SKB) \ |
c5905afb | 1815 | if (static_key_false(&netstamp_needed)) { \ |
2456e855 | 1816 | if ((COND) && !(SKB)->tstamp) \ |
588f0330 ED |
1817 | __net_timestamp(SKB); \ |
1818 | } \ | |
3b098e2d | 1819 | |
f4b05d27 | 1820 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
1821 | { |
1822 | unsigned int len; | |
1823 | ||
1824 | if (!(dev->flags & IFF_UP)) | |
1825 | return false; | |
1826 | ||
1827 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1828 | if (skb->len <= len) | |
1829 | return true; | |
1830 | ||
1831 | /* if TSO is enabled, we don't care about the length as the packet | |
1832 | * could be forwarded without being segmented before | |
1833 | */ | |
1834 | if (skb_is_gso(skb)) | |
1835 | return true; | |
1836 | ||
1837 | return false; | |
1838 | } | |
1ee481fb | 1839 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1840 | |
a0265d28 HX |
1841 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1842 | { | |
4e3264d2 | 1843 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 1844 | |
4e3264d2 MKL |
1845 | if (likely(!ret)) { |
1846 | skb->protocol = eth_type_trans(skb, dev); | |
1847 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
1848 | } | |
a0265d28 | 1849 | |
4e3264d2 | 1850 | return ret; |
a0265d28 HX |
1851 | } |
1852 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1853 | ||
44540960 AB |
1854 | /** |
1855 | * dev_forward_skb - loopback an skb to another netif | |
1856 | * | |
1857 | * @dev: destination network device | |
1858 | * @skb: buffer to forward | |
1859 | * | |
1860 | * return values: | |
1861 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1862 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1863 | * |
1864 | * dev_forward_skb can be used for injecting an skb from the | |
1865 | * start_xmit function of one device into the receive queue | |
1866 | * of another device. | |
1867 | * | |
1868 | * The receiving device may be in another namespace, so | |
1869 | * we have to clear all information in the skb that could | |
1870 | * impact namespace isolation. | |
1871 | */ | |
1872 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1873 | { | |
a0265d28 | 1874 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1875 | } |
1876 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1877 | ||
71d9dec2 CG |
1878 | static inline int deliver_skb(struct sk_buff *skb, |
1879 | struct packet_type *pt_prev, | |
1880 | struct net_device *orig_dev) | |
1881 | { | |
1f8b977a | 1882 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 1883 | return -ENOMEM; |
63354797 | 1884 | refcount_inc(&skb->users); |
71d9dec2 CG |
1885 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1886 | } | |
1887 | ||
7866a621 SN |
1888 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1889 | struct packet_type **pt, | |
fbcb2170 JP |
1890 | struct net_device *orig_dev, |
1891 | __be16 type, | |
7866a621 SN |
1892 | struct list_head *ptype_list) |
1893 | { | |
1894 | struct packet_type *ptype, *pt_prev = *pt; | |
1895 | ||
1896 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1897 | if (ptype->type != type) | |
1898 | continue; | |
1899 | if (pt_prev) | |
fbcb2170 | 1900 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1901 | pt_prev = ptype; |
1902 | } | |
1903 | *pt = pt_prev; | |
1904 | } | |
1905 | ||
c0de08d0 EL |
1906 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1907 | { | |
a3d744e9 | 1908 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1909 | return false; |
1910 | ||
1911 | if (ptype->id_match) | |
1912 | return ptype->id_match(ptype, skb->sk); | |
1913 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1914 | return true; | |
1915 | ||
1916 | return false; | |
1917 | } | |
1918 | ||
1da177e4 LT |
1919 | /* |
1920 | * Support routine. Sends outgoing frames to any network | |
1921 | * taps currently in use. | |
1922 | */ | |
1923 | ||
74b20582 | 1924 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
1925 | { |
1926 | struct packet_type *ptype; | |
71d9dec2 CG |
1927 | struct sk_buff *skb2 = NULL; |
1928 | struct packet_type *pt_prev = NULL; | |
7866a621 | 1929 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 1930 | |
1da177e4 | 1931 | rcu_read_lock(); |
7866a621 SN |
1932 | again: |
1933 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1da177e4 LT |
1934 | /* Never send packets back to the socket |
1935 | * they originated from - MvS (miquels@drinkel.ow.org) | |
1936 | */ | |
7866a621 SN |
1937 | if (skb_loop_sk(ptype, skb)) |
1938 | continue; | |
71d9dec2 | 1939 | |
7866a621 SN |
1940 | if (pt_prev) { |
1941 | deliver_skb(skb2, pt_prev, skb->dev); | |
1942 | pt_prev = ptype; | |
1943 | continue; | |
1944 | } | |
1da177e4 | 1945 | |
7866a621 SN |
1946 | /* need to clone skb, done only once */ |
1947 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
1948 | if (!skb2) | |
1949 | goto out_unlock; | |
70978182 | 1950 | |
7866a621 | 1951 | net_timestamp_set(skb2); |
1da177e4 | 1952 | |
7866a621 SN |
1953 | /* skb->nh should be correctly |
1954 | * set by sender, so that the second statement is | |
1955 | * just protection against buggy protocols. | |
1956 | */ | |
1957 | skb_reset_mac_header(skb2); | |
1958 | ||
1959 | if (skb_network_header(skb2) < skb2->data || | |
1960 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
1961 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
1962 | ntohs(skb2->protocol), | |
1963 | dev->name); | |
1964 | skb_reset_network_header(skb2); | |
1da177e4 | 1965 | } |
7866a621 SN |
1966 | |
1967 | skb2->transport_header = skb2->network_header; | |
1968 | skb2->pkt_type = PACKET_OUTGOING; | |
1969 | pt_prev = ptype; | |
1970 | } | |
1971 | ||
1972 | if (ptype_list == &ptype_all) { | |
1973 | ptype_list = &dev->ptype_all; | |
1974 | goto again; | |
1da177e4 | 1975 | } |
7866a621 | 1976 | out_unlock: |
581fe0ea WB |
1977 | if (pt_prev) { |
1978 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
1979 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
1980 | else | |
1981 | kfree_skb(skb2); | |
1982 | } | |
1da177e4 LT |
1983 | rcu_read_unlock(); |
1984 | } | |
74b20582 | 1985 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 1986 | |
2c53040f BH |
1987 | /** |
1988 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
1989 | * @dev: Network device |
1990 | * @txq: number of queues available | |
1991 | * | |
1992 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
1993 | * valid. To resolve this verify the tc mapping remains valid and if | |
1994 | * not NULL the mapping. With no priorities mapping to this | |
1995 | * offset/count pair it will no longer be used. In the worst case TC0 | |
1996 | * is invalid nothing can be done so disable priority mappings. If is | |
1997 | * expected that drivers will fix this mapping if they can before | |
1998 | * calling netif_set_real_num_tx_queues. | |
1999 | */ | |
bb134d22 | 2000 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2001 | { |
2002 | int i; | |
2003 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2004 | ||
2005 | /* If TC0 is invalidated disable TC mapping */ | |
2006 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 2007 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2008 | dev->num_tc = 0; |
2009 | return; | |
2010 | } | |
2011 | ||
2012 | /* Invalidated prio to tc mappings set to TC0 */ | |
2013 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2014 | int q = netdev_get_prio_tc_map(dev, i); | |
2015 | ||
2016 | tc = &dev->tc_to_txq[q]; | |
2017 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
2018 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2019 | i, q); | |
4f57c087 JF |
2020 | netdev_set_prio_tc_map(dev, i, 0); |
2021 | } | |
2022 | } | |
2023 | } | |
2024 | ||
8d059b0f AD |
2025 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2026 | { | |
2027 | if (dev->num_tc) { | |
2028 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2029 | int i; | |
2030 | ||
2031 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { | |
2032 | if ((txq - tc->offset) < tc->count) | |
2033 | return i; | |
2034 | } | |
2035 | ||
2036 | return -1; | |
2037 | } | |
2038 | ||
2039 | return 0; | |
2040 | } | |
2041 | ||
537c00de AD |
2042 | #ifdef CONFIG_XPS |
2043 | static DEFINE_MUTEX(xps_map_mutex); | |
2044 | #define xmap_dereference(P) \ | |
2045 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2046 | ||
6234f874 AD |
2047 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2048 | int tci, u16 index) | |
537c00de | 2049 | { |
10cdc3f3 AD |
2050 | struct xps_map *map = NULL; |
2051 | int pos; | |
537c00de | 2052 | |
10cdc3f3 | 2053 | if (dev_maps) |
6234f874 AD |
2054 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2055 | if (!map) | |
2056 | return false; | |
537c00de | 2057 | |
6234f874 AD |
2058 | for (pos = map->len; pos--;) { |
2059 | if (map->queues[pos] != index) | |
2060 | continue; | |
2061 | ||
2062 | if (map->len > 1) { | |
2063 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2064 | break; |
537c00de | 2065 | } |
6234f874 AD |
2066 | |
2067 | RCU_INIT_POINTER(dev_maps->cpu_map[tci], NULL); | |
2068 | kfree_rcu(map, rcu); | |
2069 | return false; | |
537c00de AD |
2070 | } |
2071 | ||
6234f874 | 2072 | return true; |
10cdc3f3 AD |
2073 | } |
2074 | ||
6234f874 AD |
2075 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2076 | struct xps_dev_maps *dev_maps, | |
2077 | int cpu, u16 offset, u16 count) | |
2078 | { | |
184c449f AD |
2079 | int num_tc = dev->num_tc ? : 1; |
2080 | bool active = false; | |
2081 | int tci; | |
6234f874 | 2082 | |
184c449f AD |
2083 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2084 | int i, j; | |
2085 | ||
2086 | for (i = count, j = offset; i--; j++) { | |
2087 | if (!remove_xps_queue(dev_maps, cpu, j)) | |
2088 | break; | |
2089 | } | |
2090 | ||
2091 | active |= i < 0; | |
6234f874 AD |
2092 | } |
2093 | ||
184c449f | 2094 | return active; |
6234f874 AD |
2095 | } |
2096 | ||
2097 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, | |
2098 | u16 count) | |
10cdc3f3 AD |
2099 | { |
2100 | struct xps_dev_maps *dev_maps; | |
024e9679 | 2101 | int cpu, i; |
10cdc3f3 AD |
2102 | bool active = false; |
2103 | ||
2104 | mutex_lock(&xps_map_mutex); | |
2105 | dev_maps = xmap_dereference(dev->xps_maps); | |
2106 | ||
2107 | if (!dev_maps) | |
2108 | goto out_no_maps; | |
2109 | ||
6234f874 AD |
2110 | for_each_possible_cpu(cpu) |
2111 | active |= remove_xps_queue_cpu(dev, dev_maps, cpu, | |
2112 | offset, count); | |
10cdc3f3 AD |
2113 | |
2114 | if (!active) { | |
537c00de AD |
2115 | RCU_INIT_POINTER(dev->xps_maps, NULL); |
2116 | kfree_rcu(dev_maps, rcu); | |
2117 | } | |
2118 | ||
6234f874 | 2119 | for (i = offset + (count - 1); count--; i--) |
024e9679 AD |
2120 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, i), |
2121 | NUMA_NO_NODE); | |
2122 | ||
537c00de AD |
2123 | out_no_maps: |
2124 | mutex_unlock(&xps_map_mutex); | |
2125 | } | |
2126 | ||
6234f874 AD |
2127 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2128 | { | |
2129 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2130 | } | |
2131 | ||
01c5f864 AD |
2132 | static struct xps_map *expand_xps_map(struct xps_map *map, |
2133 | int cpu, u16 index) | |
2134 | { | |
2135 | struct xps_map *new_map; | |
2136 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2137 | int i, pos; | |
2138 | ||
2139 | for (pos = 0; map && pos < map->len; pos++) { | |
2140 | if (map->queues[pos] != index) | |
2141 | continue; | |
2142 | return map; | |
2143 | } | |
2144 | ||
2145 | /* Need to add queue to this CPU's existing map */ | |
2146 | if (map) { | |
2147 | if (pos < map->alloc_len) | |
2148 | return map; | |
2149 | ||
2150 | alloc_len = map->alloc_len * 2; | |
2151 | } | |
2152 | ||
2153 | /* Need to allocate new map to store queue on this CPU's map */ | |
2154 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2155 | cpu_to_node(cpu)); | |
2156 | if (!new_map) | |
2157 | return NULL; | |
2158 | ||
2159 | for (i = 0; i < pos; i++) | |
2160 | new_map->queues[i] = map->queues[i]; | |
2161 | new_map->alloc_len = alloc_len; | |
2162 | new_map->len = pos; | |
2163 | ||
2164 | return new_map; | |
2165 | } | |
2166 | ||
3573540c MT |
2167 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, |
2168 | u16 index) | |
537c00de | 2169 | { |
01c5f864 | 2170 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
184c449f AD |
2171 | int i, cpu, tci, numa_node_id = -2; |
2172 | int maps_sz, num_tc = 1, tc = 0; | |
537c00de | 2173 | struct xps_map *map, *new_map; |
01c5f864 | 2174 | bool active = false; |
537c00de | 2175 | |
184c449f AD |
2176 | if (dev->num_tc) { |
2177 | num_tc = dev->num_tc; | |
2178 | tc = netdev_txq_to_tc(dev, index); | |
2179 | if (tc < 0) | |
2180 | return -EINVAL; | |
2181 | } | |
2182 | ||
2183 | maps_sz = XPS_DEV_MAPS_SIZE(num_tc); | |
2184 | if (maps_sz < L1_CACHE_BYTES) | |
2185 | maps_sz = L1_CACHE_BYTES; | |
2186 | ||
537c00de AD |
2187 | mutex_lock(&xps_map_mutex); |
2188 | ||
2189 | dev_maps = xmap_dereference(dev->xps_maps); | |
2190 | ||
01c5f864 | 2191 | /* allocate memory for queue storage */ |
184c449f | 2192 | for_each_cpu_and(cpu, cpu_online_mask, mask) { |
01c5f864 AD |
2193 | if (!new_dev_maps) |
2194 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2195 | if (!new_dev_maps) { |
2196 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2197 | return -ENOMEM; |
2bb60cb9 | 2198 | } |
01c5f864 | 2199 | |
184c449f AD |
2200 | tci = cpu * num_tc + tc; |
2201 | map = dev_maps ? xmap_dereference(dev_maps->cpu_map[tci]) : | |
01c5f864 AD |
2202 | NULL; |
2203 | ||
2204 | map = expand_xps_map(map, cpu, index); | |
2205 | if (!map) | |
2206 | goto error; | |
2207 | ||
184c449f | 2208 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); |
01c5f864 AD |
2209 | } |
2210 | ||
2211 | if (!new_dev_maps) | |
2212 | goto out_no_new_maps; | |
2213 | ||
537c00de | 2214 | for_each_possible_cpu(cpu) { |
184c449f AD |
2215 | /* copy maps belonging to foreign traffic classes */ |
2216 | for (i = tc, tci = cpu * num_tc; dev_maps && i--; tci++) { | |
2217 | /* fill in the new device map from the old device map */ | |
2218 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2219 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2220 | } | |
2221 | ||
2222 | /* We need to explicitly update tci as prevous loop | |
2223 | * could break out early if dev_maps is NULL. | |
2224 | */ | |
2225 | tci = cpu * num_tc + tc; | |
2226 | ||
01c5f864 AD |
2227 | if (cpumask_test_cpu(cpu, mask) && cpu_online(cpu)) { |
2228 | /* add queue to CPU maps */ | |
2229 | int pos = 0; | |
2230 | ||
184c449f | 2231 | map = xmap_dereference(new_dev_maps->cpu_map[tci]); |
01c5f864 AD |
2232 | while ((pos < map->len) && (map->queues[pos] != index)) |
2233 | pos++; | |
2234 | ||
2235 | if (pos == map->len) | |
2236 | map->queues[map->len++] = index; | |
537c00de | 2237 | #ifdef CONFIG_NUMA |
537c00de AD |
2238 | if (numa_node_id == -2) |
2239 | numa_node_id = cpu_to_node(cpu); | |
2240 | else if (numa_node_id != cpu_to_node(cpu)) | |
2241 | numa_node_id = -1; | |
537c00de | 2242 | #endif |
01c5f864 AD |
2243 | } else if (dev_maps) { |
2244 | /* fill in the new device map from the old device map */ | |
184c449f AD |
2245 | map = xmap_dereference(dev_maps->cpu_map[tci]); |
2246 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
537c00de | 2247 | } |
01c5f864 | 2248 | |
184c449f AD |
2249 | /* copy maps belonging to foreign traffic classes */ |
2250 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2251 | /* fill in the new device map from the old device map */ | |
2252 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
2253 | RCU_INIT_POINTER(new_dev_maps->cpu_map[tci], map); | |
2254 | } | |
537c00de AD |
2255 | } |
2256 | ||
01c5f864 AD |
2257 | rcu_assign_pointer(dev->xps_maps, new_dev_maps); |
2258 | ||
537c00de | 2259 | /* Cleanup old maps */ |
184c449f AD |
2260 | if (!dev_maps) |
2261 | goto out_no_old_maps; | |
2262 | ||
2263 | for_each_possible_cpu(cpu) { | |
2264 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { | |
2265 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2266 | map = xmap_dereference(dev_maps->cpu_map[tci]); | |
01c5f864 AD |
2267 | if (map && map != new_map) |
2268 | kfree_rcu(map, rcu); | |
2269 | } | |
537c00de AD |
2270 | } |
2271 | ||
184c449f AD |
2272 | kfree_rcu(dev_maps, rcu); |
2273 | ||
2274 | out_no_old_maps: | |
01c5f864 AD |
2275 | dev_maps = new_dev_maps; |
2276 | active = true; | |
537c00de | 2277 | |
01c5f864 AD |
2278 | out_no_new_maps: |
2279 | /* update Tx queue numa node */ | |
537c00de AD |
2280 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), |
2281 | (numa_node_id >= 0) ? numa_node_id : | |
2282 | NUMA_NO_NODE); | |
2283 | ||
01c5f864 AD |
2284 | if (!dev_maps) |
2285 | goto out_no_maps; | |
2286 | ||
2287 | /* removes queue from unused CPUs */ | |
2288 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2289 | for (i = tc, tci = cpu * num_tc; i--; tci++) |
2290 | active |= remove_xps_queue(dev_maps, tci, index); | |
2291 | if (!cpumask_test_cpu(cpu, mask) || !cpu_online(cpu)) | |
2292 | active |= remove_xps_queue(dev_maps, tci, index); | |
2293 | for (i = num_tc - tc, tci++; --i; tci++) | |
2294 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2295 | } |
2296 | ||
2297 | /* free map if not active */ | |
2298 | if (!active) { | |
2299 | RCU_INIT_POINTER(dev->xps_maps, NULL); | |
2300 | kfree_rcu(dev_maps, rcu); | |
2301 | } | |
2302 | ||
2303 | out_no_maps: | |
537c00de AD |
2304 | mutex_unlock(&xps_map_mutex); |
2305 | ||
2306 | return 0; | |
2307 | error: | |
01c5f864 AD |
2308 | /* remove any maps that we added */ |
2309 | for_each_possible_cpu(cpu) { | |
184c449f AD |
2310 | for (i = num_tc, tci = cpu * num_tc; i--; tci++) { |
2311 | new_map = xmap_dereference(new_dev_maps->cpu_map[tci]); | |
2312 | map = dev_maps ? | |
2313 | xmap_dereference(dev_maps->cpu_map[tci]) : | |
2314 | NULL; | |
2315 | if (new_map && new_map != map) | |
2316 | kfree(new_map); | |
2317 | } | |
01c5f864 AD |
2318 | } |
2319 | ||
537c00de AD |
2320 | mutex_unlock(&xps_map_mutex); |
2321 | ||
537c00de AD |
2322 | kfree(new_dev_maps); |
2323 | return -ENOMEM; | |
2324 | } | |
2325 | EXPORT_SYMBOL(netif_set_xps_queue); | |
2326 | ||
2327 | #endif | |
9cf1f6a8 AD |
2328 | void netdev_reset_tc(struct net_device *dev) |
2329 | { | |
6234f874 AD |
2330 | #ifdef CONFIG_XPS |
2331 | netif_reset_xps_queues_gt(dev, 0); | |
2332 | #endif | |
9cf1f6a8 AD |
2333 | dev->num_tc = 0; |
2334 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2335 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2336 | } | |
2337 | EXPORT_SYMBOL(netdev_reset_tc); | |
2338 | ||
2339 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2340 | { | |
2341 | if (tc >= dev->num_tc) | |
2342 | return -EINVAL; | |
2343 | ||
6234f874 AD |
2344 | #ifdef CONFIG_XPS |
2345 | netif_reset_xps_queues(dev, offset, count); | |
2346 | #endif | |
9cf1f6a8 AD |
2347 | dev->tc_to_txq[tc].count = count; |
2348 | dev->tc_to_txq[tc].offset = offset; | |
2349 | return 0; | |
2350 | } | |
2351 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2352 | ||
2353 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2354 | { | |
2355 | if (num_tc > TC_MAX_QUEUE) | |
2356 | return -EINVAL; | |
2357 | ||
6234f874 AD |
2358 | #ifdef CONFIG_XPS |
2359 | netif_reset_xps_queues_gt(dev, 0); | |
2360 | #endif | |
9cf1f6a8 AD |
2361 | dev->num_tc = num_tc; |
2362 | return 0; | |
2363 | } | |
2364 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2365 | ||
f0796d5c JF |
2366 | /* |
2367 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
2368 | * greater then real_num_tx_queues stale skbs on the qdisc must be flushed. | |
2369 | */ | |
e6484930 | 2370 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2371 | { |
1d24eb48 TH |
2372 | int rc; |
2373 | ||
e6484930 TH |
2374 | if (txq < 1 || txq > dev->num_tx_queues) |
2375 | return -EINVAL; | |
f0796d5c | 2376 | |
5c56580b BH |
2377 | if (dev->reg_state == NETREG_REGISTERED || |
2378 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2379 | ASSERT_RTNL(); |
2380 | ||
1d24eb48 TH |
2381 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2382 | txq); | |
bf264145 TH |
2383 | if (rc) |
2384 | return rc; | |
2385 | ||
4f57c087 JF |
2386 | if (dev->num_tc) |
2387 | netif_setup_tc(dev, txq); | |
2388 | ||
024e9679 | 2389 | if (txq < dev->real_num_tx_queues) { |
e6484930 | 2390 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2391 | #ifdef CONFIG_XPS |
2392 | netif_reset_xps_queues_gt(dev, txq); | |
2393 | #endif | |
2394 | } | |
f0796d5c | 2395 | } |
e6484930 TH |
2396 | |
2397 | dev->real_num_tx_queues = txq; | |
2398 | return 0; | |
f0796d5c JF |
2399 | } |
2400 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2401 | |
a953be53 | 2402 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2403 | /** |
2404 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2405 | * @dev: Network device | |
2406 | * @rxq: Actual number of RX queues | |
2407 | * | |
2408 | * This must be called either with the rtnl_lock held or before | |
2409 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2410 | * negative error code. If called before registration, it always |
2411 | * succeeds. | |
62fe0b40 BH |
2412 | */ |
2413 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2414 | { | |
2415 | int rc; | |
2416 | ||
bd25fa7b TH |
2417 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2418 | return -EINVAL; | |
2419 | ||
62fe0b40 BH |
2420 | if (dev->reg_state == NETREG_REGISTERED) { |
2421 | ASSERT_RTNL(); | |
2422 | ||
62fe0b40 BH |
2423 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2424 | rxq); | |
2425 | if (rc) | |
2426 | return rc; | |
62fe0b40 BH |
2427 | } |
2428 | ||
2429 | dev->real_num_rx_queues = rxq; | |
2430 | return 0; | |
2431 | } | |
2432 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2433 | #endif | |
2434 | ||
2c53040f BH |
2435 | /** |
2436 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2437 | * |
2438 | * This routine should set an upper limit on the number of RSS queues | |
2439 | * used by default by multiqueue devices. | |
2440 | */ | |
a55b138b | 2441 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2442 | { |
40e4e713 HS |
2443 | return is_kdump_kernel() ? |
2444 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2445 | } |
2446 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2447 | ||
3bcb846c | 2448 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2449 | { |
def82a1d JP |
2450 | struct softnet_data *sd; |
2451 | unsigned long flags; | |
56079431 | 2452 | |
def82a1d | 2453 | local_irq_save(flags); |
903ceff7 | 2454 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2455 | q->next_sched = NULL; |
2456 | *sd->output_queue_tailp = q; | |
2457 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2458 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2459 | local_irq_restore(flags); | |
2460 | } | |
2461 | ||
2462 | void __netif_schedule(struct Qdisc *q) | |
2463 | { | |
2464 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2465 | __netif_reschedule(q); | |
56079431 DV |
2466 | } |
2467 | EXPORT_SYMBOL(__netif_schedule); | |
2468 | ||
e6247027 ED |
2469 | struct dev_kfree_skb_cb { |
2470 | enum skb_free_reason reason; | |
2471 | }; | |
2472 | ||
2473 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2474 | { |
e6247027 ED |
2475 | return (struct dev_kfree_skb_cb *)skb->cb; |
2476 | } | |
2477 | ||
46e5da40 JF |
2478 | void netif_schedule_queue(struct netdev_queue *txq) |
2479 | { | |
2480 | rcu_read_lock(); | |
2481 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2482 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2483 | ||
2484 | __netif_schedule(q); | |
2485 | } | |
2486 | rcu_read_unlock(); | |
2487 | } | |
2488 | EXPORT_SYMBOL(netif_schedule_queue); | |
2489 | ||
46e5da40 JF |
2490 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2491 | { | |
2492 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2493 | struct Qdisc *q; | |
2494 | ||
2495 | rcu_read_lock(); | |
2496 | q = rcu_dereference(dev_queue->qdisc); | |
2497 | __netif_schedule(q); | |
2498 | rcu_read_unlock(); | |
2499 | } | |
2500 | } | |
2501 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2502 | ||
e6247027 | 2503 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2504 | { |
e6247027 | 2505 | unsigned long flags; |
56079431 | 2506 | |
9899886d MJ |
2507 | if (unlikely(!skb)) |
2508 | return; | |
2509 | ||
63354797 | 2510 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 2511 | smp_rmb(); |
63354797 RE |
2512 | refcount_set(&skb->users, 0); |
2513 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 2514 | return; |
bea3348e | 2515 | } |
e6247027 ED |
2516 | get_kfree_skb_cb(skb)->reason = reason; |
2517 | local_irq_save(flags); | |
2518 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2519 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2520 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2521 | local_irq_restore(flags); | |
56079431 | 2522 | } |
e6247027 | 2523 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2524 | |
e6247027 | 2525 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2526 | { |
2527 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2528 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2529 | else |
2530 | dev_kfree_skb(skb); | |
2531 | } | |
e6247027 | 2532 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2533 | |
2534 | ||
bea3348e SH |
2535 | /** |
2536 | * netif_device_detach - mark device as removed | |
2537 | * @dev: network device | |
2538 | * | |
2539 | * Mark device as removed from system and therefore no longer available. | |
2540 | */ | |
56079431 DV |
2541 | void netif_device_detach(struct net_device *dev) |
2542 | { | |
2543 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2544 | netif_running(dev)) { | |
d543103a | 2545 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2546 | } |
2547 | } | |
2548 | EXPORT_SYMBOL(netif_device_detach); | |
2549 | ||
bea3348e SH |
2550 | /** |
2551 | * netif_device_attach - mark device as attached | |
2552 | * @dev: network device | |
2553 | * | |
2554 | * Mark device as attached from system and restart if needed. | |
2555 | */ | |
56079431 DV |
2556 | void netif_device_attach(struct net_device *dev) |
2557 | { | |
2558 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2559 | netif_running(dev)) { | |
d543103a | 2560 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2561 | __netdev_watchdog_up(dev); |
56079431 DV |
2562 | } |
2563 | } | |
2564 | EXPORT_SYMBOL(netif_device_attach); | |
2565 | ||
5605c762 JP |
2566 | /* |
2567 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2568 | * to be used as a distribution range. | |
2569 | */ | |
2570 | u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, | |
2571 | unsigned int num_tx_queues) | |
2572 | { | |
2573 | u32 hash; | |
2574 | u16 qoffset = 0; | |
2575 | u16 qcount = num_tx_queues; | |
2576 | ||
2577 | if (skb_rx_queue_recorded(skb)) { | |
2578 | hash = skb_get_rx_queue(skb); | |
2579 | while (unlikely(hash >= num_tx_queues)) | |
2580 | hash -= num_tx_queues; | |
2581 | return hash; | |
2582 | } | |
2583 | ||
2584 | if (dev->num_tc) { | |
2585 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
f4563a75 | 2586 | |
5605c762 JP |
2587 | qoffset = dev->tc_to_txq[tc].offset; |
2588 | qcount = dev->tc_to_txq[tc].count; | |
2589 | } | |
2590 | ||
2591 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2592 | } | |
2593 | EXPORT_SYMBOL(__skb_tx_hash); | |
2594 | ||
36c92474 BH |
2595 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2596 | { | |
84d15ae5 | 2597 | static const netdev_features_t null_features; |
36c92474 | 2598 | struct net_device *dev = skb->dev; |
88ad4175 | 2599 | const char *name = ""; |
36c92474 | 2600 | |
c846ad9b BG |
2601 | if (!net_ratelimit()) |
2602 | return; | |
2603 | ||
88ad4175 BM |
2604 | if (dev) { |
2605 | if (dev->dev.parent) | |
2606 | name = dev_driver_string(dev->dev.parent); | |
2607 | else | |
2608 | name = netdev_name(dev); | |
2609 | } | |
36c92474 BH |
2610 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2611 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2612 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2613 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2614 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2615 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2616 | } | |
2617 | ||
1da177e4 LT |
2618 | /* |
2619 | * Invalidate hardware checksum when packet is to be mangled, and | |
2620 | * complete checksum manually on outgoing path. | |
2621 | */ | |
84fa7933 | 2622 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2623 | { |
d3bc23e7 | 2624 | __wsum csum; |
663ead3b | 2625 | int ret = 0, offset; |
1da177e4 | 2626 | |
84fa7933 | 2627 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2628 | goto out_set_summed; |
2629 | ||
2630 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2631 | skb_warn_bad_offload(skb); |
2632 | return -EINVAL; | |
1da177e4 LT |
2633 | } |
2634 | ||
cef401de ED |
2635 | /* Before computing a checksum, we should make sure no frag could |
2636 | * be modified by an external entity : checksum could be wrong. | |
2637 | */ | |
2638 | if (skb_has_shared_frag(skb)) { | |
2639 | ret = __skb_linearize(skb); | |
2640 | if (ret) | |
2641 | goto out; | |
2642 | } | |
2643 | ||
55508d60 | 2644 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2645 | BUG_ON(offset >= skb_headlen(skb)); |
2646 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2647 | ||
2648 | offset += skb->csum_offset; | |
2649 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2650 | ||
2651 | if (skb_cloned(skb) && | |
2652 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2653 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2654 | if (ret) | |
2655 | goto out; | |
2656 | } | |
2657 | ||
4f2e4ad5 | 2658 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 2659 | out_set_summed: |
1da177e4 | 2660 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2661 | out: |
1da177e4 LT |
2662 | return ret; |
2663 | } | |
d1b19dff | 2664 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2665 | |
b72b5bf6 DC |
2666 | int skb_crc32c_csum_help(struct sk_buff *skb) |
2667 | { | |
2668 | __le32 crc32c_csum; | |
2669 | int ret = 0, offset, start; | |
2670 | ||
2671 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
2672 | goto out; | |
2673 | ||
2674 | if (unlikely(skb_is_gso(skb))) | |
2675 | goto out; | |
2676 | ||
2677 | /* Before computing a checksum, we should make sure no frag could | |
2678 | * be modified by an external entity : checksum could be wrong. | |
2679 | */ | |
2680 | if (unlikely(skb_has_shared_frag(skb))) { | |
2681 | ret = __skb_linearize(skb); | |
2682 | if (ret) | |
2683 | goto out; | |
2684 | } | |
2685 | start = skb_checksum_start_offset(skb); | |
2686 | offset = start + offsetof(struct sctphdr, checksum); | |
2687 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
2688 | ret = -EINVAL; | |
2689 | goto out; | |
2690 | } | |
2691 | if (skb_cloned(skb) && | |
2692 | !skb_clone_writable(skb, offset + sizeof(__le32))) { | |
2693 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
2694 | if (ret) | |
2695 | goto out; | |
2696 | } | |
2697 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, | |
2698 | skb->len - start, ~(__u32)0, | |
2699 | crc32c_csum_stub)); | |
2700 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
2701 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 2702 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
2703 | out: |
2704 | return ret; | |
2705 | } | |
2706 | ||
53d6471c | 2707 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2708 | { |
252e3346 | 2709 | __be16 type = skb->protocol; |
f6a78bfc | 2710 | |
19acc327 PS |
2711 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2712 | if (type == htons(ETH_P_TEB)) { | |
2713 | struct ethhdr *eth; | |
2714 | ||
2715 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
2716 | return 0; | |
2717 | ||
2718 | eth = (struct ethhdr *)skb_mac_header(skb); | |
2719 | type = eth->h_proto; | |
2720 | } | |
2721 | ||
d4bcef3f | 2722 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
2723 | } |
2724 | ||
2725 | /** | |
2726 | * skb_mac_gso_segment - mac layer segmentation handler. | |
2727 | * @skb: buffer to segment | |
2728 | * @features: features for the output path (see dev->features) | |
2729 | */ | |
2730 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
2731 | netdev_features_t features) | |
2732 | { | |
2733 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
2734 | struct packet_offload *ptype; | |
53d6471c VY |
2735 | int vlan_depth = skb->mac_len; |
2736 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
2737 | |
2738 | if (unlikely(!type)) | |
2739 | return ERR_PTR(-EINVAL); | |
2740 | ||
53d6471c | 2741 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
2742 | |
2743 | rcu_read_lock(); | |
22061d80 | 2744 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 2745 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 2746 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
2747 | break; |
2748 | } | |
2749 | } | |
2750 | rcu_read_unlock(); | |
2751 | ||
98e399f8 | 2752 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 2753 | |
f6a78bfc HX |
2754 | return segs; |
2755 | } | |
05e8ef4a PS |
2756 | EXPORT_SYMBOL(skb_mac_gso_segment); |
2757 | ||
2758 | ||
2759 | /* openvswitch calls this on rx path, so we need a different check. | |
2760 | */ | |
2761 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
2762 | { | |
2763 | if (tx_path) | |
93991221 | 2764 | return skb->ip_summed != CHECKSUM_PARTIAL; |
6e7bc478 ED |
2765 | |
2766 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
2767 | } |
2768 | ||
2769 | /** | |
2770 | * __skb_gso_segment - Perform segmentation on skb. | |
2771 | * @skb: buffer to segment | |
2772 | * @features: features for the output path (see dev->features) | |
2773 | * @tx_path: whether it is called in TX path | |
2774 | * | |
2775 | * This function segments the given skb and returns a list of segments. | |
2776 | * | |
2777 | * It may return NULL if the skb requires no segmentation. This is | |
2778 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
2779 | * |
2780 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
2781 | */ |
2782 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
2783 | netdev_features_t features, bool tx_path) | |
2784 | { | |
b2504a5d ED |
2785 | struct sk_buff *segs; |
2786 | ||
05e8ef4a PS |
2787 | if (unlikely(skb_needs_check(skb, tx_path))) { |
2788 | int err; | |
2789 | ||
b2504a5d | 2790 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 2791 | err = skb_cow_head(skb, 0); |
2792 | if (err < 0) | |
05e8ef4a PS |
2793 | return ERR_PTR(err); |
2794 | } | |
2795 | ||
802ab55a AD |
2796 | /* Only report GSO partial support if it will enable us to |
2797 | * support segmentation on this frame without needing additional | |
2798 | * work. | |
2799 | */ | |
2800 | if (features & NETIF_F_GSO_PARTIAL) { | |
2801 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
2802 | struct net_device *dev = skb->dev; | |
2803 | ||
2804 | partial_features |= dev->features & dev->gso_partial_features; | |
2805 | if (!skb_gso_ok(skb, features | partial_features)) | |
2806 | features &= ~NETIF_F_GSO_PARTIAL; | |
2807 | } | |
2808 | ||
9207f9d4 KK |
2809 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
2810 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
2811 | ||
68c33163 | 2812 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
2813 | SKB_GSO_CB(skb)->encap_level = 0; |
2814 | ||
05e8ef4a PS |
2815 | skb_reset_mac_header(skb); |
2816 | skb_reset_mac_len(skb); | |
2817 | ||
b2504a5d ED |
2818 | segs = skb_mac_gso_segment(skb, features); |
2819 | ||
2820 | if (unlikely(skb_needs_check(skb, tx_path))) | |
2821 | skb_warn_bad_offload(skb); | |
2822 | ||
2823 | return segs; | |
05e8ef4a | 2824 | } |
12b0004d | 2825 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 2826 | |
fb286bb2 HX |
2827 | /* Take action when hardware reception checksum errors are detected. */ |
2828 | #ifdef CONFIG_BUG | |
2829 | void netdev_rx_csum_fault(struct net_device *dev) | |
2830 | { | |
2831 | if (net_ratelimit()) { | |
7b6cd1ce | 2832 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
fb286bb2 HX |
2833 | dump_stack(); |
2834 | } | |
2835 | } | |
2836 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
2837 | #endif | |
2838 | ||
1da177e4 LT |
2839 | /* Actually, we should eliminate this check as soon as we know, that: |
2840 | * 1. IOMMU is present and allows to map all the memory. | |
2841 | * 2. No high memory really exists on this machine. | |
2842 | */ | |
2843 | ||
c1e756bf | 2844 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 2845 | { |
3d3a8533 | 2846 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 2847 | int i; |
f4563a75 | 2848 | |
5acbbd42 | 2849 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
2850 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
2851 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 2852 | |
ea2ab693 | 2853 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 2854 | return 1; |
ea2ab693 | 2855 | } |
5acbbd42 | 2856 | } |
1da177e4 | 2857 | |
5acbbd42 FT |
2858 | if (PCI_DMA_BUS_IS_PHYS) { |
2859 | struct device *pdev = dev->dev.parent; | |
1da177e4 | 2860 | |
9092c658 ED |
2861 | if (!pdev) |
2862 | return 0; | |
5acbbd42 | 2863 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
ea2ab693 IC |
2864 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; |
2865 | dma_addr_t addr = page_to_phys(skb_frag_page(frag)); | |
f4563a75 | 2866 | |
5acbbd42 FT |
2867 | if (!pdev->dma_mask || addr + PAGE_SIZE - 1 > *pdev->dma_mask) |
2868 | return 1; | |
2869 | } | |
2870 | } | |
3d3a8533 | 2871 | #endif |
1da177e4 LT |
2872 | return 0; |
2873 | } | |
1da177e4 | 2874 | |
3b392ddb SH |
2875 | /* If MPLS offload request, verify we are testing hardware MPLS features |
2876 | * instead of standard features for the netdev. | |
2877 | */ | |
d0edc7bf | 2878 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
2879 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
2880 | netdev_features_t features, | |
2881 | __be16 type) | |
2882 | { | |
25cd9ba0 | 2883 | if (eth_p_mpls(type)) |
3b392ddb SH |
2884 | features &= skb->dev->mpls_features; |
2885 | ||
2886 | return features; | |
2887 | } | |
2888 | #else | |
2889 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
2890 | netdev_features_t features, | |
2891 | __be16 type) | |
2892 | { | |
2893 | return features; | |
2894 | } | |
2895 | #endif | |
2896 | ||
c8f44aff | 2897 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 2898 | netdev_features_t features) |
f01a5236 | 2899 | { |
53d6471c | 2900 | int tmp; |
3b392ddb SH |
2901 | __be16 type; |
2902 | ||
2903 | type = skb_network_protocol(skb, &tmp); | |
2904 | features = net_mpls_features(skb, features, type); | |
53d6471c | 2905 | |
c0d680e5 | 2906 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 2907 | !can_checksum_protocol(features, type)) { |
996e8021 | 2908 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 2909 | } |
7be2c82c ED |
2910 | if (illegal_highdma(skb->dev, skb)) |
2911 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
2912 | |
2913 | return features; | |
2914 | } | |
2915 | ||
e38f3025 TM |
2916 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
2917 | struct net_device *dev, | |
2918 | netdev_features_t features) | |
2919 | { | |
2920 | return features; | |
2921 | } | |
2922 | EXPORT_SYMBOL(passthru_features_check); | |
2923 | ||
8cb65d00 TM |
2924 | static netdev_features_t dflt_features_check(const struct sk_buff *skb, |
2925 | struct net_device *dev, | |
2926 | netdev_features_t features) | |
2927 | { | |
2928 | return vlan_features_check(skb, features); | |
2929 | } | |
2930 | ||
cbc53e08 AD |
2931 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
2932 | struct net_device *dev, | |
2933 | netdev_features_t features) | |
2934 | { | |
2935 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
2936 | ||
2937 | if (gso_segs > dev->gso_max_segs) | |
2938 | return features & ~NETIF_F_GSO_MASK; | |
2939 | ||
802ab55a AD |
2940 | /* Support for GSO partial features requires software |
2941 | * intervention before we can actually process the packets | |
2942 | * so we need to strip support for any partial features now | |
2943 | * and we can pull them back in after we have partially | |
2944 | * segmented the frame. | |
2945 | */ | |
2946 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
2947 | features &= ~dev->gso_partial_features; | |
2948 | ||
2949 | /* Make sure to clear the IPv4 ID mangling feature if the | |
2950 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
2951 | */ |
2952 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
2953 | struct iphdr *iph = skb->encapsulation ? | |
2954 | inner_ip_hdr(skb) : ip_hdr(skb); | |
2955 | ||
2956 | if (!(iph->frag_off & htons(IP_DF))) | |
2957 | features &= ~NETIF_F_TSO_MANGLEID; | |
2958 | } | |
2959 | ||
2960 | return features; | |
2961 | } | |
2962 | ||
c1e756bf | 2963 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 2964 | { |
5f35227e | 2965 | struct net_device *dev = skb->dev; |
fcbeb976 | 2966 | netdev_features_t features = dev->features; |
58e998c6 | 2967 | |
cbc53e08 AD |
2968 | if (skb_is_gso(skb)) |
2969 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 2970 | |
5f35227e JG |
2971 | /* If encapsulation offload request, verify we are testing |
2972 | * hardware encapsulation features instead of standard | |
2973 | * features for the netdev | |
2974 | */ | |
2975 | if (skb->encapsulation) | |
2976 | features &= dev->hw_enc_features; | |
2977 | ||
f5a7fb88 TM |
2978 | if (skb_vlan_tagged(skb)) |
2979 | features = netdev_intersect_features(features, | |
2980 | dev->vlan_features | | |
2981 | NETIF_F_HW_VLAN_CTAG_TX | | |
2982 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 2983 | |
5f35227e JG |
2984 | if (dev->netdev_ops->ndo_features_check) |
2985 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
2986 | features); | |
8cb65d00 TM |
2987 | else |
2988 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 2989 | |
c1e756bf | 2990 | return harmonize_features(skb, features); |
58e998c6 | 2991 | } |
c1e756bf | 2992 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 2993 | |
2ea25513 | 2994 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 2995 | struct netdev_queue *txq, bool more) |
f6a78bfc | 2996 | { |
2ea25513 DM |
2997 | unsigned int len; |
2998 | int rc; | |
00829823 | 2999 | |
7866a621 | 3000 | if (!list_empty(&ptype_all) || !list_empty(&dev->ptype_all)) |
2ea25513 | 3001 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3002 | |
2ea25513 DM |
3003 | len = skb->len; |
3004 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 3005 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3006 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3007 | |
2ea25513 DM |
3008 | return rc; |
3009 | } | |
7b9c6090 | 3010 | |
8dcda22a DM |
3011 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3012 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3013 | { |
3014 | struct sk_buff *skb = first; | |
3015 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3016 | |
7f2e870f DM |
3017 | while (skb) { |
3018 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3019 | |
7f2e870f | 3020 | skb->next = NULL; |
95f6b3dd | 3021 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3022 | if (unlikely(!dev_xmit_complete(rc))) { |
3023 | skb->next = next; | |
3024 | goto out; | |
3025 | } | |
6afff0ca | 3026 | |
7f2e870f DM |
3027 | skb = next; |
3028 | if (netif_xmit_stopped(txq) && skb) { | |
3029 | rc = NETDEV_TX_BUSY; | |
3030 | break; | |
9ccb8975 | 3031 | } |
7f2e870f | 3032 | } |
9ccb8975 | 3033 | |
7f2e870f DM |
3034 | out: |
3035 | *ret = rc; | |
3036 | return skb; | |
3037 | } | |
b40863c6 | 3038 | |
1ff0dc94 ED |
3039 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3040 | netdev_features_t features) | |
f6a78bfc | 3041 | { |
df8a39de | 3042 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3043 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3044 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3045 | return skb; |
3046 | } | |
f6a78bfc | 3047 | |
43c26a1a DC |
3048 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3049 | const netdev_features_t features) | |
3050 | { | |
3051 | if (unlikely(skb->csum_not_inet)) | |
3052 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : | |
3053 | skb_crc32c_csum_help(skb); | |
3054 | ||
3055 | return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); | |
3056 | } | |
3057 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3058 | ||
55a93b3e | 3059 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev) |
eae3f88e DM |
3060 | { |
3061 | netdev_features_t features; | |
f6a78bfc | 3062 | |
eae3f88e DM |
3063 | features = netif_skb_features(skb); |
3064 | skb = validate_xmit_vlan(skb, features); | |
3065 | if (unlikely(!skb)) | |
3066 | goto out_null; | |
7b9c6090 | 3067 | |
8b86a61d | 3068 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3069 | struct sk_buff *segs; |
3070 | ||
3071 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3072 | if (IS_ERR(segs)) { |
af6dabc9 | 3073 | goto out_kfree_skb; |
cecda693 JW |
3074 | } else if (segs) { |
3075 | consume_skb(skb); | |
3076 | skb = segs; | |
f6a78bfc | 3077 | } |
eae3f88e DM |
3078 | } else { |
3079 | if (skb_needs_linearize(skb, features) && | |
3080 | __skb_linearize(skb)) | |
3081 | goto out_kfree_skb; | |
4ec93edb | 3082 | |
f6e27114 SK |
3083 | if (validate_xmit_xfrm(skb, features)) |
3084 | goto out_kfree_skb; | |
3085 | ||
eae3f88e DM |
3086 | /* If packet is not checksummed and device does not |
3087 | * support checksumming for this protocol, complete | |
3088 | * checksumming here. | |
3089 | */ | |
3090 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3091 | if (skb->encapsulation) | |
3092 | skb_set_inner_transport_header(skb, | |
3093 | skb_checksum_start_offset(skb)); | |
3094 | else | |
3095 | skb_set_transport_header(skb, | |
3096 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3097 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3098 | goto out_kfree_skb; |
7b9c6090 | 3099 | } |
0c772159 | 3100 | } |
7b9c6090 | 3101 | |
eae3f88e | 3102 | return skb; |
fc70fb64 | 3103 | |
f6a78bfc HX |
3104 | out_kfree_skb: |
3105 | kfree_skb(skb); | |
eae3f88e | 3106 | out_null: |
d21fd63e | 3107 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3108 | return NULL; |
3109 | } | |
6afff0ca | 3110 | |
55a93b3e ED |
3111 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev) |
3112 | { | |
3113 | struct sk_buff *next, *head = NULL, *tail; | |
3114 | ||
bec3cfdc | 3115 | for (; skb != NULL; skb = next) { |
55a93b3e ED |
3116 | next = skb->next; |
3117 | skb->next = NULL; | |
bec3cfdc ED |
3118 | |
3119 | /* in case skb wont be segmented, point to itself */ | |
3120 | skb->prev = skb; | |
3121 | ||
55a93b3e | 3122 | skb = validate_xmit_skb(skb, dev); |
bec3cfdc ED |
3123 | if (!skb) |
3124 | continue; | |
55a93b3e | 3125 | |
bec3cfdc ED |
3126 | if (!head) |
3127 | head = skb; | |
3128 | else | |
3129 | tail->next = skb; | |
3130 | /* If skb was segmented, skb->prev points to | |
3131 | * the last segment. If not, it still contains skb. | |
3132 | */ | |
3133 | tail = skb->prev; | |
55a93b3e ED |
3134 | } |
3135 | return head; | |
f6a78bfc | 3136 | } |
104ba78c | 3137 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3138 | |
1def9238 ED |
3139 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3140 | { | |
3141 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3142 | ||
3143 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3144 | ||
3145 | /* To get more precise estimation of bytes sent on wire, | |
3146 | * we add to pkt_len the headers size of all segments | |
3147 | */ | |
3148 | if (shinfo->gso_size) { | |
757b8b1d | 3149 | unsigned int hdr_len; |
15e5a030 | 3150 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3151 | |
757b8b1d ED |
3152 | /* mac layer + network layer */ |
3153 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3154 | ||
3155 | /* + transport layer */ | |
1def9238 ED |
3156 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) |
3157 | hdr_len += tcp_hdrlen(skb); | |
3158 | else | |
3159 | hdr_len += sizeof(struct udphdr); | |
15e5a030 JW |
3160 | |
3161 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3162 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3163 | shinfo->gso_size); | |
3164 | ||
3165 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3166 | } |
3167 | } | |
3168 | ||
bbd8a0d3 KK |
3169 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3170 | struct net_device *dev, | |
3171 | struct netdev_queue *txq) | |
3172 | { | |
3173 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3174 | struct sk_buff *to_free = NULL; |
a2da570d | 3175 | bool contended; |
bbd8a0d3 KK |
3176 | int rc; |
3177 | ||
a2da570d | 3178 | qdisc_calculate_pkt_len(skb, q); |
79640a4c ED |
3179 | /* |
3180 | * Heuristic to force contended enqueues to serialize on a | |
3181 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3182 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3183 | * often and dequeue packets faster. |
79640a4c | 3184 | */ |
a2da570d | 3185 | contended = qdisc_is_running(q); |
79640a4c ED |
3186 | if (unlikely(contended)) |
3187 | spin_lock(&q->busylock); | |
3188 | ||
bbd8a0d3 KK |
3189 | spin_lock(root_lock); |
3190 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3191 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3192 | rc = NET_XMIT_DROP; |
3193 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3194 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3195 | /* |
3196 | * This is a work-conserving queue; there are no old skbs | |
3197 | * waiting to be sent out; and the qdisc is not running - | |
3198 | * xmit the skb directly. | |
3199 | */ | |
bfe0d029 | 3200 | |
bfe0d029 ED |
3201 | qdisc_bstats_update(q, skb); |
3202 | ||
55a93b3e | 3203 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3204 | if (unlikely(contended)) { |
3205 | spin_unlock(&q->busylock); | |
3206 | contended = false; | |
3207 | } | |
bbd8a0d3 | 3208 | __qdisc_run(q); |
79640a4c | 3209 | } else |
bc135b23 | 3210 | qdisc_run_end(q); |
bbd8a0d3 KK |
3211 | |
3212 | rc = NET_XMIT_SUCCESS; | |
3213 | } else { | |
520ac30f | 3214 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3215 | if (qdisc_run_begin(q)) { |
3216 | if (unlikely(contended)) { | |
3217 | spin_unlock(&q->busylock); | |
3218 | contended = false; | |
3219 | } | |
3220 | __qdisc_run(q); | |
3221 | } | |
bbd8a0d3 KK |
3222 | } |
3223 | spin_unlock(root_lock); | |
520ac30f ED |
3224 | if (unlikely(to_free)) |
3225 | kfree_skb_list(to_free); | |
79640a4c ED |
3226 | if (unlikely(contended)) |
3227 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3228 | return rc; |
3229 | } | |
3230 | ||
86f8515f | 3231 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3232 | static void skb_update_prio(struct sk_buff *skb) |
3233 | { | |
6977a79d | 3234 | struct netprio_map *map = rcu_dereference_bh(skb->dev->priomap); |
5bc1421e | 3235 | |
91c68ce2 | 3236 | if (!skb->priority && skb->sk && map) { |
2a56a1fe TH |
3237 | unsigned int prioidx = |
3238 | sock_cgroup_prioidx(&skb->sk->sk_cgrp_data); | |
91c68ce2 ED |
3239 | |
3240 | if (prioidx < map->priomap_len) | |
3241 | skb->priority = map->priomap[prioidx]; | |
3242 | } | |
5bc1421e NH |
3243 | } |
3244 | #else | |
3245 | #define skb_update_prio(skb) | |
3246 | #endif | |
3247 | ||
f60e5990 | 3248 | DEFINE_PER_CPU(int, xmit_recursion); |
3249 | EXPORT_SYMBOL(xmit_recursion); | |
3250 | ||
95603e22 MM |
3251 | /** |
3252 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3253 | * @net: network namespace this loopback is happening in |
3254 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3255 | * @skb: buffer to transmit |
3256 | */ | |
0c4b51f0 | 3257 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3258 | { |
3259 | skb_reset_mac_header(skb); | |
3260 | __skb_pull(skb, skb_network_offset(skb)); | |
3261 | skb->pkt_type = PACKET_LOOPBACK; | |
3262 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3263 | WARN_ON(!skb_dst(skb)); | |
3264 | skb_dst_force(skb); | |
3265 | netif_rx_ni(skb); | |
3266 | return 0; | |
3267 | } | |
3268 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3269 | ||
1f211a1b DB |
3270 | #ifdef CONFIG_NET_EGRESS |
3271 | static struct sk_buff * | |
3272 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3273 | { | |
3274 | struct tcf_proto *cl = rcu_dereference_bh(dev->egress_cl_list); | |
3275 | struct tcf_result cl_res; | |
3276 | ||
3277 | if (!cl) | |
3278 | return skb; | |
3279 | ||
8dc07fdb | 3280 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
1f211a1b DB |
3281 | qdisc_bstats_cpu_update(cl->q, skb); |
3282 | ||
87d83093 | 3283 | switch (tcf_classify(skb, cl, &cl_res, false)) { |
1f211a1b DB |
3284 | case TC_ACT_OK: |
3285 | case TC_ACT_RECLASSIFY: | |
3286 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3287 | break; | |
3288 | case TC_ACT_SHOT: | |
3289 | qdisc_qstats_cpu_drop(cl->q); | |
3290 | *ret = NET_XMIT_DROP; | |
7e2c3aea DB |
3291 | kfree_skb(skb); |
3292 | return NULL; | |
1f211a1b DB |
3293 | case TC_ACT_STOLEN: |
3294 | case TC_ACT_QUEUED: | |
e25ea21f | 3295 | case TC_ACT_TRAP: |
1f211a1b | 3296 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3297 | consume_skb(skb); |
1f211a1b DB |
3298 | return NULL; |
3299 | case TC_ACT_REDIRECT: | |
3300 | /* No need to push/pop skb's mac_header here on egress! */ | |
3301 | skb_do_redirect(skb); | |
3302 | *ret = NET_XMIT_SUCCESS; | |
3303 | return NULL; | |
3304 | default: | |
3305 | break; | |
3306 | } | |
3307 | ||
3308 | return skb; | |
3309 | } | |
3310 | #endif /* CONFIG_NET_EGRESS */ | |
3311 | ||
638b2a69 JP |
3312 | static inline int get_xps_queue(struct net_device *dev, struct sk_buff *skb) |
3313 | { | |
3314 | #ifdef CONFIG_XPS | |
3315 | struct xps_dev_maps *dev_maps; | |
3316 | struct xps_map *map; | |
3317 | int queue_index = -1; | |
3318 | ||
3319 | rcu_read_lock(); | |
3320 | dev_maps = rcu_dereference(dev->xps_maps); | |
3321 | if (dev_maps) { | |
184c449f AD |
3322 | unsigned int tci = skb->sender_cpu - 1; |
3323 | ||
3324 | if (dev->num_tc) { | |
3325 | tci *= dev->num_tc; | |
3326 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3327 | } | |
3328 | ||
3329 | map = rcu_dereference(dev_maps->cpu_map[tci]); | |
638b2a69 JP |
3330 | if (map) { |
3331 | if (map->len == 1) | |
3332 | queue_index = map->queues[0]; | |
3333 | else | |
3334 | queue_index = map->queues[reciprocal_scale(skb_get_hash(skb), | |
3335 | map->len)]; | |
3336 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3337 | queue_index = -1; | |
3338 | } | |
3339 | } | |
3340 | rcu_read_unlock(); | |
3341 | ||
3342 | return queue_index; | |
3343 | #else | |
3344 | return -1; | |
3345 | #endif | |
3346 | } | |
3347 | ||
3348 | static u16 __netdev_pick_tx(struct net_device *dev, struct sk_buff *skb) | |
3349 | { | |
3350 | struct sock *sk = skb->sk; | |
3351 | int queue_index = sk_tx_queue_get(sk); | |
3352 | ||
3353 | if (queue_index < 0 || skb->ooo_okay || | |
3354 | queue_index >= dev->real_num_tx_queues) { | |
3355 | int new_index = get_xps_queue(dev, skb); | |
f4563a75 | 3356 | |
638b2a69 JP |
3357 | if (new_index < 0) |
3358 | new_index = skb_tx_hash(dev, skb); | |
3359 | ||
3360 | if (queue_index != new_index && sk && | |
004a5d01 | 3361 | sk_fullsock(sk) && |
638b2a69 JP |
3362 | rcu_access_pointer(sk->sk_dst_cache)) |
3363 | sk_tx_queue_set(sk, new_index); | |
3364 | ||
3365 | queue_index = new_index; | |
3366 | } | |
3367 | ||
3368 | return queue_index; | |
3369 | } | |
3370 | ||
3371 | struct netdev_queue *netdev_pick_tx(struct net_device *dev, | |
3372 | struct sk_buff *skb, | |
3373 | void *accel_priv) | |
3374 | { | |
3375 | int queue_index = 0; | |
3376 | ||
3377 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3378 | u32 sender_cpu = skb->sender_cpu - 1; |
3379 | ||
3380 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3381 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3382 | #endif | |
3383 | ||
3384 | if (dev->real_num_tx_queues != 1) { | |
3385 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3386 | |
638b2a69 JP |
3387 | if (ops->ndo_select_queue) |
3388 | queue_index = ops->ndo_select_queue(dev, skb, accel_priv, | |
3389 | __netdev_pick_tx); | |
3390 | else | |
3391 | queue_index = __netdev_pick_tx(dev, skb); | |
3392 | ||
3393 | if (!accel_priv) | |
3394 | queue_index = netdev_cap_txqueue(dev, queue_index); | |
3395 | } | |
3396 | ||
3397 | skb_set_queue_mapping(skb, queue_index); | |
3398 | return netdev_get_tx_queue(dev, queue_index); | |
3399 | } | |
3400 | ||
d29f749e | 3401 | /** |
9d08dd3d | 3402 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3403 | * @skb: buffer to transmit |
9d08dd3d | 3404 | * @accel_priv: private data used for L2 forwarding offload |
d29f749e DJ |
3405 | * |
3406 | * Queue a buffer for transmission to a network device. The caller must | |
3407 | * have set the device and priority and built the buffer before calling | |
3408 | * this function. The function can be called from an interrupt. | |
3409 | * | |
3410 | * A negative errno code is returned on a failure. A success does not | |
3411 | * guarantee the frame will be transmitted as it may be dropped due | |
3412 | * to congestion or traffic shaping. | |
3413 | * | |
3414 | * ----------------------------------------------------------------------------------- | |
3415 | * I notice this method can also return errors from the queue disciplines, | |
3416 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3417 | * be positive. | |
3418 | * | |
3419 | * Regardless of the return value, the skb is consumed, so it is currently | |
3420 | * difficult to retry a send to this method. (You can bump the ref count | |
3421 | * before sending to hold a reference for retry if you are careful.) | |
3422 | * | |
3423 | * When calling this method, interrupts MUST be enabled. This is because | |
3424 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3425 | * --BLG | |
3426 | */ | |
0a59f3a9 | 3427 | static int __dev_queue_xmit(struct sk_buff *skb, void *accel_priv) |
1da177e4 LT |
3428 | { |
3429 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3430 | struct netdev_queue *txq; |
1da177e4 LT |
3431 | struct Qdisc *q; |
3432 | int rc = -ENOMEM; | |
3433 | ||
6d1ccff6 ED |
3434 | skb_reset_mac_header(skb); |
3435 | ||
e7fd2885 WB |
3436 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3437 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3438 | ||
4ec93edb YH |
3439 | /* Disable soft irqs for various locks below. Also |
3440 | * stops preemption for RCU. | |
1da177e4 | 3441 | */ |
4ec93edb | 3442 | rcu_read_lock_bh(); |
1da177e4 | 3443 | |
5bc1421e NH |
3444 | skb_update_prio(skb); |
3445 | ||
1f211a1b DB |
3446 | qdisc_pkt_len_init(skb); |
3447 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3448 | skb->tc_at_ingress = 0; |
1f211a1b DB |
3449 | # ifdef CONFIG_NET_EGRESS |
3450 | if (static_key_false(&egress_needed)) { | |
3451 | skb = sch_handle_egress(skb, &rc, dev); | |
3452 | if (!skb) | |
3453 | goto out; | |
3454 | } | |
3455 | # endif | |
3456 | #endif | |
02875878 ED |
3457 | /* If device/qdisc don't need skb->dst, release it right now while |
3458 | * its hot in this cpu cache. | |
3459 | */ | |
3460 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3461 | skb_dst_drop(skb); | |
3462 | else | |
3463 | skb_dst_force(skb); | |
3464 | ||
f663dd9a | 3465 | txq = netdev_pick_tx(dev, skb, accel_priv); |
a898def2 | 3466 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3467 | |
cf66ba58 | 3468 | trace_net_dev_queue(skb); |
1da177e4 | 3469 | if (q->enqueue) { |
bbd8a0d3 | 3470 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3471 | goto out; |
1da177e4 LT |
3472 | } |
3473 | ||
3474 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 3475 | * loopback, all the sorts of tunnels... |
1da177e4 | 3476 | |
eb13da1a | 3477 | * Really, it is unlikely that netif_tx_lock protection is necessary |
3478 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
3479 | * counters.) | |
3480 | * However, it is possible, that they rely on protection | |
3481 | * made by us here. | |
1da177e4 | 3482 | |
eb13da1a | 3483 | * Check this and shot the lock. It is not prone from deadlocks. |
3484 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
3485 | */ |
3486 | if (dev->flags & IFF_UP) { | |
3487 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3488 | ||
c773e847 | 3489 | if (txq->xmit_lock_owner != cpu) { |
a70b506e DB |
3490 | if (unlikely(__this_cpu_read(xmit_recursion) > |
3491 | XMIT_RECURSION_LIMIT)) | |
745e20f1 ED |
3492 | goto recursion_alert; |
3493 | ||
1f59533f JDB |
3494 | skb = validate_xmit_skb(skb, dev); |
3495 | if (!skb) | |
d21fd63e | 3496 | goto out; |
1f59533f | 3497 | |
c773e847 | 3498 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3499 | |
73466498 | 3500 | if (!netif_xmit_stopped(txq)) { |
745e20f1 | 3501 | __this_cpu_inc(xmit_recursion); |
ce93718f | 3502 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
745e20f1 | 3503 | __this_cpu_dec(xmit_recursion); |
572a9d7b | 3504 | if (dev_xmit_complete(rc)) { |
c773e847 | 3505 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3506 | goto out; |
3507 | } | |
3508 | } | |
c773e847 | 3509 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3510 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3511 | dev->name); | |
1da177e4 LT |
3512 | } else { |
3513 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3514 | * unfortunately |
3515 | */ | |
3516 | recursion_alert: | |
e87cc472 JP |
3517 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3518 | dev->name); | |
1da177e4 LT |
3519 | } |
3520 | } | |
3521 | ||
3522 | rc = -ENETDOWN; | |
d4828d85 | 3523 | rcu_read_unlock_bh(); |
1da177e4 | 3524 | |
015f0688 | 3525 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3526 | kfree_skb_list(skb); |
1da177e4 LT |
3527 | return rc; |
3528 | out: | |
d4828d85 | 3529 | rcu_read_unlock_bh(); |
1da177e4 LT |
3530 | return rc; |
3531 | } | |
f663dd9a | 3532 | |
2b4aa3ce | 3533 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3534 | { |
3535 | return __dev_queue_xmit(skb, NULL); | |
3536 | } | |
2b4aa3ce | 3537 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3538 | |
f663dd9a JW |
3539 | int dev_queue_xmit_accel(struct sk_buff *skb, void *accel_priv) |
3540 | { | |
3541 | return __dev_queue_xmit(skb, accel_priv); | |
3542 | } | |
3543 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3544 | ||
1da177e4 | 3545 | |
eb13da1a | 3546 | /************************************************************************* |
3547 | * Receiver routines | |
3548 | *************************************************************************/ | |
1da177e4 | 3549 | |
6b2bedc3 | 3550 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3551 | EXPORT_SYMBOL(netdev_max_backlog); |
3552 | ||
3b098e2d | 3553 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 3554 | int netdev_budget __read_mostly = 300; |
7acf8a1e | 3555 | unsigned int __read_mostly netdev_budget_usecs = 2000; |
3d48b53f MT |
3556 | int weight_p __read_mostly = 64; /* old backlog weight */ |
3557 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
3558 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
3559 | int dev_rx_weight __read_mostly = 64; | |
3560 | int dev_tx_weight __read_mostly = 64; | |
1da177e4 | 3561 | |
eecfd7c4 ED |
3562 | /* Called with irq disabled */ |
3563 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3564 | struct napi_struct *napi) | |
3565 | { | |
3566 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3567 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3568 | } | |
3569 | ||
bfb564e7 KK |
3570 | #ifdef CONFIG_RPS |
3571 | ||
3572 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3573 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3574 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3575 | u32 rps_cpu_mask __read_mostly; |
3576 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3577 | |
c5905afb | 3578 | struct static_key rps_needed __read_mostly; |
3df97ba8 | 3579 | EXPORT_SYMBOL(rps_needed); |
13bfff25 ED |
3580 | struct static_key rfs_needed __read_mostly; |
3581 | EXPORT_SYMBOL(rfs_needed); | |
adc9300e | 3582 | |
c445477d BH |
3583 | static struct rps_dev_flow * |
3584 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3585 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3586 | { | |
a31196b0 | 3587 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3588 | #ifdef CONFIG_RFS_ACCEL |
3589 | struct netdev_rx_queue *rxqueue; | |
3590 | struct rps_dev_flow_table *flow_table; | |
3591 | struct rps_dev_flow *old_rflow; | |
3592 | u32 flow_id; | |
3593 | u16 rxq_index; | |
3594 | int rc; | |
3595 | ||
3596 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
3597 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
3598 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
3599 | goto out; |
3600 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
3601 | if (rxq_index == skb_get_rx_queue(skb)) | |
3602 | goto out; | |
3603 | ||
3604 | rxqueue = dev->_rx + rxq_index; | |
3605 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3606 | if (!flow_table) | |
3607 | goto out; | |
61b905da | 3608 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
3609 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
3610 | rxq_index, flow_id); | |
3611 | if (rc < 0) | |
3612 | goto out; | |
3613 | old_rflow = rflow; | |
3614 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
3615 | rflow->filter = rc; |
3616 | if (old_rflow->filter == rflow->filter) | |
3617 | old_rflow->filter = RPS_NO_FILTER; | |
3618 | out: | |
3619 | #endif | |
3620 | rflow->last_qtail = | |
09994d1b | 3621 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
3622 | } |
3623 | ||
09994d1b | 3624 | rflow->cpu = next_cpu; |
c445477d BH |
3625 | return rflow; |
3626 | } | |
3627 | ||
bfb564e7 KK |
3628 | /* |
3629 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
3630 | * CPU from the RPS map of the receiving queue for a given skb. | |
3631 | * rcu_read_lock must be held on entry. | |
3632 | */ | |
3633 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3634 | struct rps_dev_flow **rflowp) | |
3635 | { | |
567e4b79 ED |
3636 | const struct rps_sock_flow_table *sock_flow_table; |
3637 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 3638 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 3639 | struct rps_map *map; |
bfb564e7 | 3640 | int cpu = -1; |
567e4b79 | 3641 | u32 tcpu; |
61b905da | 3642 | u32 hash; |
bfb564e7 KK |
3643 | |
3644 | if (skb_rx_queue_recorded(skb)) { | |
3645 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 3646 | |
62fe0b40 BH |
3647 | if (unlikely(index >= dev->real_num_rx_queues)) { |
3648 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
3649 | "%s received packet on queue %u, but number " | |
3650 | "of RX queues is %u\n", | |
3651 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
3652 | goto done; |
3653 | } | |
567e4b79 ED |
3654 | rxqueue += index; |
3655 | } | |
bfb564e7 | 3656 | |
567e4b79 ED |
3657 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
3658 | ||
3659 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 3660 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 3661 | if (!flow_table && !map) |
bfb564e7 KK |
3662 | goto done; |
3663 | ||
2d47b459 | 3664 | skb_reset_network_header(skb); |
61b905da TH |
3665 | hash = skb_get_hash(skb); |
3666 | if (!hash) | |
bfb564e7 KK |
3667 | goto done; |
3668 | ||
fec5e652 TH |
3669 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
3670 | if (flow_table && sock_flow_table) { | |
fec5e652 | 3671 | struct rps_dev_flow *rflow; |
567e4b79 ED |
3672 | u32 next_cpu; |
3673 | u32 ident; | |
3674 | ||
3675 | /* First check into global flow table if there is a match */ | |
3676 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
3677 | if ((ident ^ hash) & ~rps_cpu_mask) | |
3678 | goto try_rps; | |
fec5e652 | 3679 | |
567e4b79 ED |
3680 | next_cpu = ident & rps_cpu_mask; |
3681 | ||
3682 | /* OK, now we know there is a match, | |
3683 | * we can look at the local (per receive queue) flow table | |
3684 | */ | |
61b905da | 3685 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
3686 | tcpu = rflow->cpu; |
3687 | ||
fec5e652 TH |
3688 | /* |
3689 | * If the desired CPU (where last recvmsg was done) is | |
3690 | * different from current CPU (one in the rx-queue flow | |
3691 | * table entry), switch if one of the following holds: | |
a31196b0 | 3692 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
3693 | * - Current CPU is offline. |
3694 | * - The current CPU's queue tail has advanced beyond the | |
3695 | * last packet that was enqueued using this table entry. | |
3696 | * This guarantees that all previous packets for the flow | |
3697 | * have been dequeued, thus preserving in order delivery. | |
3698 | */ | |
3699 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 3700 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 3701 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
3702 | rflow->last_qtail)) >= 0)) { |
3703 | tcpu = next_cpu; | |
c445477d | 3704 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 3705 | } |
c445477d | 3706 | |
a31196b0 | 3707 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
3708 | *rflowp = rflow; |
3709 | cpu = tcpu; | |
3710 | goto done; | |
3711 | } | |
3712 | } | |
3713 | ||
567e4b79 ED |
3714 | try_rps: |
3715 | ||
0a9627f2 | 3716 | if (map) { |
8fc54f68 | 3717 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
3718 | if (cpu_online(tcpu)) { |
3719 | cpu = tcpu; | |
3720 | goto done; | |
3721 | } | |
3722 | } | |
3723 | ||
3724 | done: | |
0a9627f2 TH |
3725 | return cpu; |
3726 | } | |
3727 | ||
c445477d BH |
3728 | #ifdef CONFIG_RFS_ACCEL |
3729 | ||
3730 | /** | |
3731 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
3732 | * @dev: Device on which the filter was set | |
3733 | * @rxq_index: RX queue index | |
3734 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
3735 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
3736 | * | |
3737 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
3738 | * this function for each installed filter and remove the filters for | |
3739 | * which it returns %true. | |
3740 | */ | |
3741 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
3742 | u32 flow_id, u16 filter_id) | |
3743 | { | |
3744 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
3745 | struct rps_dev_flow_table *flow_table; | |
3746 | struct rps_dev_flow *rflow; | |
3747 | bool expire = true; | |
a31196b0 | 3748 | unsigned int cpu; |
c445477d BH |
3749 | |
3750 | rcu_read_lock(); | |
3751 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
3752 | if (flow_table && flow_id <= flow_table->mask) { | |
3753 | rflow = &flow_table->flows[flow_id]; | |
3754 | cpu = ACCESS_ONCE(rflow->cpu); | |
a31196b0 | 3755 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
3756 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
3757 | rflow->last_qtail) < | |
3758 | (int)(10 * flow_table->mask))) | |
3759 | expire = false; | |
3760 | } | |
3761 | rcu_read_unlock(); | |
3762 | return expire; | |
3763 | } | |
3764 | EXPORT_SYMBOL(rps_may_expire_flow); | |
3765 | ||
3766 | #endif /* CONFIG_RFS_ACCEL */ | |
3767 | ||
0a9627f2 | 3768 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 3769 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 3770 | { |
e36fa2f7 ED |
3771 | struct softnet_data *sd = data; |
3772 | ||
eecfd7c4 | 3773 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 3774 | sd->received_rps++; |
0a9627f2 | 3775 | } |
e36fa2f7 | 3776 | |
fec5e652 | 3777 | #endif /* CONFIG_RPS */ |
0a9627f2 | 3778 | |
e36fa2f7 ED |
3779 | /* |
3780 | * Check if this softnet_data structure is another cpu one | |
3781 | * If yes, queue it to our IPI list and return 1 | |
3782 | * If no, return 0 | |
3783 | */ | |
3784 | static int rps_ipi_queued(struct softnet_data *sd) | |
3785 | { | |
3786 | #ifdef CONFIG_RPS | |
903ceff7 | 3787 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
3788 | |
3789 | if (sd != mysd) { | |
3790 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
3791 | mysd->rps_ipi_list = sd; | |
3792 | ||
3793 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3794 | return 1; | |
3795 | } | |
3796 | #endif /* CONFIG_RPS */ | |
3797 | return 0; | |
3798 | } | |
3799 | ||
99bbc707 WB |
3800 | #ifdef CONFIG_NET_FLOW_LIMIT |
3801 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
3802 | #endif | |
3803 | ||
3804 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
3805 | { | |
3806 | #ifdef CONFIG_NET_FLOW_LIMIT | |
3807 | struct sd_flow_limit *fl; | |
3808 | struct softnet_data *sd; | |
3809 | unsigned int old_flow, new_flow; | |
3810 | ||
3811 | if (qlen < (netdev_max_backlog >> 1)) | |
3812 | return false; | |
3813 | ||
903ceff7 | 3814 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
3815 | |
3816 | rcu_read_lock(); | |
3817 | fl = rcu_dereference(sd->flow_limit); | |
3818 | if (fl) { | |
3958afa1 | 3819 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
3820 | old_flow = fl->history[fl->history_head]; |
3821 | fl->history[fl->history_head] = new_flow; | |
3822 | ||
3823 | fl->history_head++; | |
3824 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
3825 | ||
3826 | if (likely(fl->buckets[old_flow])) | |
3827 | fl->buckets[old_flow]--; | |
3828 | ||
3829 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
3830 | fl->count++; | |
3831 | rcu_read_unlock(); | |
3832 | return true; | |
3833 | } | |
3834 | } | |
3835 | rcu_read_unlock(); | |
3836 | #endif | |
3837 | return false; | |
3838 | } | |
3839 | ||
0a9627f2 TH |
3840 | /* |
3841 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
3842 | * queue (may be a remote CPU queue). | |
3843 | */ | |
fec5e652 TH |
3844 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
3845 | unsigned int *qtail) | |
0a9627f2 | 3846 | { |
e36fa2f7 | 3847 | struct softnet_data *sd; |
0a9627f2 | 3848 | unsigned long flags; |
99bbc707 | 3849 | unsigned int qlen; |
0a9627f2 | 3850 | |
e36fa2f7 | 3851 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
3852 | |
3853 | local_irq_save(flags); | |
0a9627f2 | 3854 | |
e36fa2f7 | 3855 | rps_lock(sd); |
e9e4dd32 JA |
3856 | if (!netif_running(skb->dev)) |
3857 | goto drop; | |
99bbc707 WB |
3858 | qlen = skb_queue_len(&sd->input_pkt_queue); |
3859 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 3860 | if (qlen) { |
0a9627f2 | 3861 | enqueue: |
e36fa2f7 | 3862 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 3863 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 3864 | rps_unlock(sd); |
152102c7 | 3865 | local_irq_restore(flags); |
0a9627f2 TH |
3866 | return NET_RX_SUCCESS; |
3867 | } | |
3868 | ||
ebda37c2 ED |
3869 | /* Schedule NAPI for backlog device |
3870 | * We can use non atomic operation since we own the queue lock | |
3871 | */ | |
3872 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 3873 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 3874 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
3875 | } |
3876 | goto enqueue; | |
3877 | } | |
3878 | ||
e9e4dd32 | 3879 | drop: |
dee42870 | 3880 | sd->dropped++; |
e36fa2f7 | 3881 | rps_unlock(sd); |
0a9627f2 | 3882 | |
0a9627f2 TH |
3883 | local_irq_restore(flags); |
3884 | ||
caf586e5 | 3885 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
3886 | kfree_skb(skb); |
3887 | return NET_RX_DROP; | |
3888 | } | |
1da177e4 | 3889 | |
d4455169 JF |
3890 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
3891 | struct bpf_prog *xdp_prog) | |
3892 | { | |
de8f3a83 | 3893 | u32 metalen, act = XDP_DROP; |
d4455169 | 3894 | struct xdp_buff xdp; |
d4455169 JF |
3895 | void *orig_data; |
3896 | int hlen, off; | |
3897 | u32 mac_len; | |
3898 | ||
3899 | /* Reinjected packets coming from act_mirred or similar should | |
3900 | * not get XDP generic processing. | |
3901 | */ | |
3902 | if (skb_cloned(skb)) | |
3903 | return XDP_PASS; | |
3904 | ||
de8f3a83 DB |
3905 | /* XDP packets must be linear and must have sufficient headroom |
3906 | * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also | |
3907 | * native XDP provides, thus we need to do it here as well. | |
3908 | */ | |
3909 | if (skb_is_nonlinear(skb) || | |
3910 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { | |
3911 | int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
3912 | int troom = skb->tail + skb->data_len - skb->end; | |
3913 | ||
3914 | /* In case we have to go down the path and also linearize, | |
3915 | * then lets do the pskb_expand_head() work just once here. | |
3916 | */ | |
3917 | if (pskb_expand_head(skb, | |
3918 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
3919 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) | |
3920 | goto do_drop; | |
3921 | if (troom > 0 && __skb_linearize(skb)) | |
3922 | goto do_drop; | |
3923 | } | |
d4455169 JF |
3924 | |
3925 | /* The XDP program wants to see the packet starting at the MAC | |
3926 | * header. | |
3927 | */ | |
3928 | mac_len = skb->data - skb_mac_header(skb); | |
3929 | hlen = skb_headlen(skb) + mac_len; | |
3930 | xdp.data = skb->data - mac_len; | |
de8f3a83 | 3931 | xdp.data_meta = xdp.data; |
d4455169 JF |
3932 | xdp.data_end = xdp.data + hlen; |
3933 | xdp.data_hard_start = skb->data - skb_headroom(skb); | |
3934 | orig_data = xdp.data; | |
3935 | ||
3936 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
3937 | ||
3938 | off = xdp.data - orig_data; | |
3939 | if (off > 0) | |
3940 | __skb_pull(skb, off); | |
3941 | else if (off < 0) | |
3942 | __skb_push(skb, -off); | |
92dd5452 | 3943 | skb->mac_header += off; |
d4455169 JF |
3944 | |
3945 | switch (act) { | |
6103aa96 | 3946 | case XDP_REDIRECT: |
d4455169 JF |
3947 | case XDP_TX: |
3948 | __skb_push(skb, mac_len); | |
de8f3a83 | 3949 | break; |
d4455169 | 3950 | case XDP_PASS: |
de8f3a83 DB |
3951 | metalen = xdp.data - xdp.data_meta; |
3952 | if (metalen) | |
3953 | skb_metadata_set(skb, metalen); | |
d4455169 | 3954 | break; |
d4455169 JF |
3955 | default: |
3956 | bpf_warn_invalid_xdp_action(act); | |
3957 | /* fall through */ | |
3958 | case XDP_ABORTED: | |
3959 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
3960 | /* fall through */ | |
3961 | case XDP_DROP: | |
3962 | do_drop: | |
3963 | kfree_skb(skb); | |
3964 | break; | |
3965 | } | |
3966 | ||
3967 | return act; | |
3968 | } | |
3969 | ||
3970 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
3971 | * network taps in order to match in-driver-XDP behavior. | |
3972 | */ | |
7c497478 | 3973 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
3974 | { |
3975 | struct net_device *dev = skb->dev; | |
3976 | struct netdev_queue *txq; | |
3977 | bool free_skb = true; | |
3978 | int cpu, rc; | |
3979 | ||
3980 | txq = netdev_pick_tx(dev, skb, NULL); | |
3981 | cpu = smp_processor_id(); | |
3982 | HARD_TX_LOCK(dev, txq, cpu); | |
3983 | if (!netif_xmit_stopped(txq)) { | |
3984 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
3985 | if (dev_xmit_complete(rc)) | |
3986 | free_skb = false; | |
3987 | } | |
3988 | HARD_TX_UNLOCK(dev, txq); | |
3989 | if (free_skb) { | |
3990 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
3991 | kfree_skb(skb); | |
3992 | } | |
3993 | } | |
7c497478 | 3994 | EXPORT_SYMBOL_GPL(generic_xdp_tx); |
d4455169 JF |
3995 | |
3996 | static struct static_key generic_xdp_needed __read_mostly; | |
3997 | ||
7c497478 | 3998 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 3999 | { |
d4455169 JF |
4000 | if (xdp_prog) { |
4001 | u32 act = netif_receive_generic_xdp(skb, xdp_prog); | |
6103aa96 | 4002 | int err; |
d4455169 JF |
4003 | |
4004 | if (act != XDP_PASS) { | |
6103aa96 JF |
4005 | switch (act) { |
4006 | case XDP_REDIRECT: | |
2facaad6 JDB |
4007 | err = xdp_do_generic_redirect(skb->dev, skb, |
4008 | xdp_prog); | |
6103aa96 JF |
4009 | if (err) |
4010 | goto out_redir; | |
4011 | /* fallthru to submit skb */ | |
4012 | case XDP_TX: | |
d4455169 | 4013 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
4014 | break; |
4015 | } | |
d4455169 JF |
4016 | return XDP_DROP; |
4017 | } | |
4018 | } | |
4019 | return XDP_PASS; | |
6103aa96 | 4020 | out_redir: |
6103aa96 JF |
4021 | kfree_skb(skb); |
4022 | return XDP_DROP; | |
d4455169 | 4023 | } |
7c497478 | 4024 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 4025 | |
ae78dbfa | 4026 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 4027 | { |
b0e28f1e | 4028 | int ret; |
1da177e4 | 4029 | |
588f0330 | 4030 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 4031 | |
cf66ba58 | 4032 | trace_netif_rx(skb); |
d4455169 JF |
4033 | |
4034 | if (static_key_false(&generic_xdp_needed)) { | |
bbbe211c JF |
4035 | int ret; |
4036 | ||
4037 | preempt_disable(); | |
4038 | rcu_read_lock(); | |
4039 | ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
4040 | rcu_read_unlock(); | |
4041 | preempt_enable(); | |
d4455169 | 4042 | |
6103aa96 JF |
4043 | /* Consider XDP consuming the packet a success from |
4044 | * the netdev point of view we do not want to count | |
4045 | * this as an error. | |
4046 | */ | |
d4455169 | 4047 | if (ret != XDP_PASS) |
6103aa96 | 4048 | return NET_RX_SUCCESS; |
d4455169 JF |
4049 | } |
4050 | ||
df334545 | 4051 | #ifdef CONFIG_RPS |
c5905afb | 4052 | if (static_key_false(&rps_needed)) { |
fec5e652 | 4053 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4054 | int cpu; |
4055 | ||
cece1945 | 4056 | preempt_disable(); |
b0e28f1e | 4057 | rcu_read_lock(); |
fec5e652 TH |
4058 | |
4059 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4060 | if (cpu < 0) |
4061 | cpu = smp_processor_id(); | |
fec5e652 TH |
4062 | |
4063 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4064 | ||
b0e28f1e | 4065 | rcu_read_unlock(); |
cece1945 | 4066 | preempt_enable(); |
adc9300e ED |
4067 | } else |
4068 | #endif | |
fec5e652 TH |
4069 | { |
4070 | unsigned int qtail; | |
f4563a75 | 4071 | |
fec5e652 TH |
4072 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4073 | put_cpu(); | |
4074 | } | |
b0e28f1e | 4075 | return ret; |
1da177e4 | 4076 | } |
ae78dbfa BH |
4077 | |
4078 | /** | |
4079 | * netif_rx - post buffer to the network code | |
4080 | * @skb: buffer to post | |
4081 | * | |
4082 | * This function receives a packet from a device driver and queues it for | |
4083 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4084 | * may be dropped during processing for congestion control or by the | |
4085 | * protocol layers. | |
4086 | * | |
4087 | * return values: | |
4088 | * NET_RX_SUCCESS (no congestion) | |
4089 | * NET_RX_DROP (packet was dropped) | |
4090 | * | |
4091 | */ | |
4092 | ||
4093 | int netif_rx(struct sk_buff *skb) | |
4094 | { | |
4095 | trace_netif_rx_entry(skb); | |
4096 | ||
4097 | return netif_rx_internal(skb); | |
4098 | } | |
d1b19dff | 4099 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4100 | |
4101 | int netif_rx_ni(struct sk_buff *skb) | |
4102 | { | |
4103 | int err; | |
4104 | ||
ae78dbfa BH |
4105 | trace_netif_rx_ni_entry(skb); |
4106 | ||
1da177e4 | 4107 | preempt_disable(); |
ae78dbfa | 4108 | err = netif_rx_internal(skb); |
1da177e4 LT |
4109 | if (local_softirq_pending()) |
4110 | do_softirq(); | |
4111 | preempt_enable(); | |
4112 | ||
4113 | return err; | |
4114 | } | |
1da177e4 LT |
4115 | EXPORT_SYMBOL(netif_rx_ni); |
4116 | ||
0766f788 | 4117 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4118 | { |
903ceff7 | 4119 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4120 | |
4121 | if (sd->completion_queue) { | |
4122 | struct sk_buff *clist; | |
4123 | ||
4124 | local_irq_disable(); | |
4125 | clist = sd->completion_queue; | |
4126 | sd->completion_queue = NULL; | |
4127 | local_irq_enable(); | |
4128 | ||
4129 | while (clist) { | |
4130 | struct sk_buff *skb = clist; | |
f4563a75 | 4131 | |
1da177e4 LT |
4132 | clist = clist->next; |
4133 | ||
63354797 | 4134 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4135 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4136 | trace_consume_skb(skb); | |
4137 | else | |
4138 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4139 | |
4140 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4141 | __kfree_skb(skb); | |
4142 | else | |
4143 | __kfree_skb_defer(skb); | |
1da177e4 | 4144 | } |
15fad714 JDB |
4145 | |
4146 | __kfree_skb_flush(); | |
1da177e4 LT |
4147 | } |
4148 | ||
4149 | if (sd->output_queue) { | |
37437bb2 | 4150 | struct Qdisc *head; |
1da177e4 LT |
4151 | |
4152 | local_irq_disable(); | |
4153 | head = sd->output_queue; | |
4154 | sd->output_queue = NULL; | |
a9cbd588 | 4155 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4156 | local_irq_enable(); |
4157 | ||
4158 | while (head) { | |
37437bb2 DM |
4159 | struct Qdisc *q = head; |
4160 | spinlock_t *root_lock; | |
4161 | ||
1da177e4 LT |
4162 | head = head->next_sched; |
4163 | ||
5fb66229 | 4164 | root_lock = qdisc_lock(q); |
3bcb846c ED |
4165 | spin_lock(root_lock); |
4166 | /* We need to make sure head->next_sched is read | |
4167 | * before clearing __QDISC_STATE_SCHED | |
4168 | */ | |
4169 | smp_mb__before_atomic(); | |
4170 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4171 | qdisc_run(q); | |
4172 | spin_unlock(root_lock); | |
1da177e4 LT |
4173 | } |
4174 | } | |
4175 | } | |
4176 | ||
181402a5 | 4177 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4178 | /* This hook is defined here for ATM LANE */ |
4179 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4180 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4181 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4182 | #endif |
1da177e4 | 4183 | |
1f211a1b DB |
4184 | static inline struct sk_buff * |
4185 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4186 | struct net_device *orig_dev) | |
f697c3e8 | 4187 | { |
e7582bab | 4188 | #ifdef CONFIG_NET_CLS_ACT |
d2788d34 DB |
4189 | struct tcf_proto *cl = rcu_dereference_bh(skb->dev->ingress_cl_list); |
4190 | struct tcf_result cl_res; | |
24824a09 | 4191 | |
c9e99fd0 DB |
4192 | /* If there's at least one ingress present somewhere (so |
4193 | * we get here via enabled static key), remaining devices | |
4194 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4195 | * out here. |
c9e99fd0 | 4196 | */ |
d2788d34 | 4197 | if (!cl) |
4577139b | 4198 | return skb; |
f697c3e8 HX |
4199 | if (*pt_prev) { |
4200 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4201 | *pt_prev = NULL; | |
1da177e4 LT |
4202 | } |
4203 | ||
3365495c | 4204 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 4205 | skb->tc_at_ingress = 1; |
24ea591d | 4206 | qdisc_bstats_cpu_update(cl->q, skb); |
c9e99fd0 | 4207 | |
87d83093 | 4208 | switch (tcf_classify(skb, cl, &cl_res, false)) { |
d2788d34 DB |
4209 | case TC_ACT_OK: |
4210 | case TC_ACT_RECLASSIFY: | |
4211 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
4212 | break; | |
4213 | case TC_ACT_SHOT: | |
24ea591d | 4214 | qdisc_qstats_cpu_drop(cl->q); |
8a3a4c6e ED |
4215 | kfree_skb(skb); |
4216 | return NULL; | |
d2788d34 DB |
4217 | case TC_ACT_STOLEN: |
4218 | case TC_ACT_QUEUED: | |
e25ea21f | 4219 | case TC_ACT_TRAP: |
8a3a4c6e | 4220 | consume_skb(skb); |
d2788d34 | 4221 | return NULL; |
27b29f63 AS |
4222 | case TC_ACT_REDIRECT: |
4223 | /* skb_mac_header check was done by cls/act_bpf, so | |
4224 | * we can safely push the L2 header back before | |
4225 | * redirecting to another netdev | |
4226 | */ | |
4227 | __skb_push(skb, skb->mac_len); | |
4228 | skb_do_redirect(skb); | |
4229 | return NULL; | |
d2788d34 DB |
4230 | default: |
4231 | break; | |
f697c3e8 | 4232 | } |
e7582bab | 4233 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4234 | return skb; |
4235 | } | |
1da177e4 | 4236 | |
24b27fc4 MB |
4237 | /** |
4238 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4239 | * @dev: device to check | |
4240 | * | |
4241 | * Check if a receive handler is already registered for a given device. | |
4242 | * Return true if there one. | |
4243 | * | |
4244 | * The caller must hold the rtnl_mutex. | |
4245 | */ | |
4246 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4247 | { | |
4248 | ASSERT_RTNL(); | |
4249 | return dev && rtnl_dereference(dev->rx_handler); | |
4250 | } | |
4251 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4252 | ||
ab95bfe0 JP |
4253 | /** |
4254 | * netdev_rx_handler_register - register receive handler | |
4255 | * @dev: device to register a handler for | |
4256 | * @rx_handler: receive handler to register | |
93e2c32b | 4257 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4258 | * |
e227867f | 4259 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4260 | * called from __netif_receive_skb. A negative errno code is returned |
4261 | * on a failure. | |
4262 | * | |
4263 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4264 | * |
4265 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4266 | */ |
4267 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4268 | rx_handler_func_t *rx_handler, |
4269 | void *rx_handler_data) | |
ab95bfe0 | 4270 | { |
1b7cd004 | 4271 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4272 | return -EBUSY; |
4273 | ||
00cfec37 | 4274 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4275 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4276 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4277 | ||
4278 | return 0; | |
4279 | } | |
4280 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4281 | ||
4282 | /** | |
4283 | * netdev_rx_handler_unregister - unregister receive handler | |
4284 | * @dev: device to unregister a handler from | |
4285 | * | |
166ec369 | 4286 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4287 | * |
4288 | * The caller must hold the rtnl_mutex. | |
4289 | */ | |
4290 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4291 | { | |
4292 | ||
4293 | ASSERT_RTNL(); | |
a9b3cd7f | 4294 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4295 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4296 | * section has a guarantee to see a non NULL rx_handler_data | |
4297 | * as well. | |
4298 | */ | |
4299 | synchronize_net(); | |
a9b3cd7f | 4300 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4301 | } |
4302 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4303 | ||
b4b9e355 MG |
4304 | /* |
4305 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4306 | * the special handling of PFMEMALLOC skbs. | |
4307 | */ | |
4308 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4309 | { | |
4310 | switch (skb->protocol) { | |
2b8837ae JP |
4311 | case htons(ETH_P_ARP): |
4312 | case htons(ETH_P_IP): | |
4313 | case htons(ETH_P_IPV6): | |
4314 | case htons(ETH_P_8021Q): | |
4315 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4316 | return true; |
4317 | default: | |
4318 | return false; | |
4319 | } | |
4320 | } | |
4321 | ||
e687ad60 PN |
4322 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4323 | int *ret, struct net_device *orig_dev) | |
4324 | { | |
e7582bab | 4325 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 | 4326 | if (nf_hook_ingress_active(skb)) { |
2c1e2703 AC |
4327 | int ingress_retval; |
4328 | ||
e687ad60 PN |
4329 | if (*pt_prev) { |
4330 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4331 | *pt_prev = NULL; | |
4332 | } | |
4333 | ||
2c1e2703 AC |
4334 | rcu_read_lock(); |
4335 | ingress_retval = nf_hook_ingress(skb); | |
4336 | rcu_read_unlock(); | |
4337 | return ingress_retval; | |
e687ad60 | 4338 | } |
e7582bab | 4339 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4340 | return 0; |
4341 | } | |
e687ad60 | 4342 | |
9754e293 | 4343 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc) |
1da177e4 LT |
4344 | { |
4345 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4346 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4347 | struct net_device *orig_dev; |
8a4eb573 | 4348 | bool deliver_exact = false; |
1da177e4 | 4349 | int ret = NET_RX_DROP; |
252e3346 | 4350 | __be16 type; |
1da177e4 | 4351 | |
588f0330 | 4352 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4353 | |
cf66ba58 | 4354 | trace_netif_receive_skb(skb); |
9b22ea56 | 4355 | |
cc9bd5ce | 4356 | orig_dev = skb->dev; |
8f903c70 | 4357 | |
c1d2bbe1 | 4358 | skb_reset_network_header(skb); |
fda55eca ED |
4359 | if (!skb_transport_header_was_set(skb)) |
4360 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4361 | skb_reset_mac_len(skb); |
1da177e4 LT |
4362 | |
4363 | pt_prev = NULL; | |
4364 | ||
63d8ea7f | 4365 | another_round: |
b6858177 | 4366 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4367 | |
4368 | __this_cpu_inc(softnet_data.processed); | |
4369 | ||
8ad227ff PM |
4370 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4371 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4372 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4373 | if (unlikely(!skb)) |
2c17d27c | 4374 | goto out; |
bcc6d479 JP |
4375 | } |
4376 | ||
e7246e12 WB |
4377 | if (skb_skip_tc_classify(skb)) |
4378 | goto skip_classify; | |
1da177e4 | 4379 | |
9754e293 | 4380 | if (pfmemalloc) |
b4b9e355 MG |
4381 | goto skip_taps; |
4382 | ||
1da177e4 | 4383 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4384 | if (pt_prev) |
4385 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4386 | pt_prev = ptype; | |
4387 | } | |
4388 | ||
4389 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4390 | if (pt_prev) | |
4391 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4392 | pt_prev = ptype; | |
1da177e4 LT |
4393 | } |
4394 | ||
b4b9e355 | 4395 | skip_taps: |
1cf51900 | 4396 | #ifdef CONFIG_NET_INGRESS |
4577139b | 4397 | if (static_key_false(&ingress_needed)) { |
1f211a1b | 4398 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4399 | if (!skb) |
2c17d27c | 4400 | goto out; |
e687ad60 PN |
4401 | |
4402 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4403 | goto out; |
4577139b | 4404 | } |
1cf51900 | 4405 | #endif |
a5135bcf | 4406 | skb_reset_tc(skb); |
e7246e12 | 4407 | skip_classify: |
9754e293 | 4408 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4409 | goto drop; |
4410 | ||
df8a39de | 4411 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4412 | if (pt_prev) { |
4413 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4414 | pt_prev = NULL; | |
4415 | } | |
48cc32d3 | 4416 | if (vlan_do_receive(&skb)) |
2425717b JF |
4417 | goto another_round; |
4418 | else if (unlikely(!skb)) | |
2c17d27c | 4419 | goto out; |
2425717b JF |
4420 | } |
4421 | ||
48cc32d3 | 4422 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4423 | if (rx_handler) { |
4424 | if (pt_prev) { | |
4425 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4426 | pt_prev = NULL; | |
4427 | } | |
8a4eb573 JP |
4428 | switch (rx_handler(&skb)) { |
4429 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4430 | ret = NET_RX_SUCCESS; |
2c17d27c | 4431 | goto out; |
8a4eb573 | 4432 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4433 | goto another_round; |
8a4eb573 JP |
4434 | case RX_HANDLER_EXACT: |
4435 | deliver_exact = true; | |
4436 | case RX_HANDLER_PASS: | |
4437 | break; | |
4438 | default: | |
4439 | BUG(); | |
4440 | } | |
ab95bfe0 | 4441 | } |
1da177e4 | 4442 | |
df8a39de JP |
4443 | if (unlikely(skb_vlan_tag_present(skb))) { |
4444 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4445 | skb->pkt_type = PACKET_OTHERHOST; |
4446 | /* Note: we might in the future use prio bits | |
4447 | * and set skb->priority like in vlan_do_receive() | |
4448 | * For the time being, just ignore Priority Code Point | |
4449 | */ | |
4450 | skb->vlan_tci = 0; | |
4451 | } | |
48cc32d3 | 4452 | |
7866a621 SN |
4453 | type = skb->protocol; |
4454 | ||
63d8ea7f | 4455 | /* deliver only exact match when indicated */ |
7866a621 SN |
4456 | if (likely(!deliver_exact)) { |
4457 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4458 | &ptype_base[ntohs(type) & | |
4459 | PTYPE_HASH_MASK]); | |
4460 | } | |
1f3c8804 | 4461 | |
7866a621 SN |
4462 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4463 | &orig_dev->ptype_specific); | |
4464 | ||
4465 | if (unlikely(skb->dev != orig_dev)) { | |
4466 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4467 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4468 | } |
4469 | ||
4470 | if (pt_prev) { | |
1f8b977a | 4471 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 4472 | goto drop; |
1080e512 MT |
4473 | else |
4474 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
1da177e4 | 4475 | } else { |
b4b9e355 | 4476 | drop: |
6e7333d3 JW |
4477 | if (!deliver_exact) |
4478 | atomic_long_inc(&skb->dev->rx_dropped); | |
4479 | else | |
4480 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4481 | kfree_skb(skb); |
4482 | /* Jamal, now you will not able to escape explaining | |
4483 | * me how you were going to use this. :-) | |
4484 | */ | |
4485 | ret = NET_RX_DROP; | |
4486 | } | |
4487 | ||
2c17d27c | 4488 | out: |
9754e293 DM |
4489 | return ret; |
4490 | } | |
4491 | ||
4492 | static int __netif_receive_skb(struct sk_buff *skb) | |
4493 | { | |
4494 | int ret; | |
4495 | ||
4496 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 4497 | unsigned int noreclaim_flag; |
9754e293 DM |
4498 | |
4499 | /* | |
4500 | * PFMEMALLOC skbs are special, they should | |
4501 | * - be delivered to SOCK_MEMALLOC sockets only | |
4502 | * - stay away from userspace | |
4503 | * - have bounded memory usage | |
4504 | * | |
4505 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
4506 | * context down to all allocation sites. | |
4507 | */ | |
f1083048 | 4508 | noreclaim_flag = memalloc_noreclaim_save(); |
9754e293 | 4509 | ret = __netif_receive_skb_core(skb, true); |
f1083048 | 4510 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 DM |
4511 | } else |
4512 | ret = __netif_receive_skb_core(skb, false); | |
4513 | ||
1da177e4 LT |
4514 | return ret; |
4515 | } | |
0a9627f2 | 4516 | |
b5cdae32 DM |
4517 | static int generic_xdp_install(struct net_device *dev, struct netdev_xdp *xdp) |
4518 | { | |
58038695 | 4519 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
4520 | struct bpf_prog *new = xdp->prog; |
4521 | int ret = 0; | |
4522 | ||
4523 | switch (xdp->command) { | |
58038695 | 4524 | case XDP_SETUP_PROG: |
b5cdae32 DM |
4525 | rcu_assign_pointer(dev->xdp_prog, new); |
4526 | if (old) | |
4527 | bpf_prog_put(old); | |
4528 | ||
4529 | if (old && !new) { | |
4530 | static_key_slow_dec(&generic_xdp_needed); | |
4531 | } else if (new && !old) { | |
4532 | static_key_slow_inc(&generic_xdp_needed); | |
4533 | dev_disable_lro(dev); | |
4534 | } | |
4535 | break; | |
b5cdae32 DM |
4536 | |
4537 | case XDP_QUERY_PROG: | |
58038695 MKL |
4538 | xdp->prog_attached = !!old; |
4539 | xdp->prog_id = old ? old->aux->id : 0; | |
b5cdae32 DM |
4540 | break; |
4541 | ||
4542 | default: | |
4543 | ret = -EINVAL; | |
4544 | break; | |
4545 | } | |
4546 | ||
4547 | return ret; | |
4548 | } | |
4549 | ||
ae78dbfa | 4550 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 4551 | { |
2c17d27c JA |
4552 | int ret; |
4553 | ||
588f0330 | 4554 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 4555 | |
c1f19b51 RC |
4556 | if (skb_defer_rx_timestamp(skb)) |
4557 | return NET_RX_SUCCESS; | |
4558 | ||
b5cdae32 | 4559 | if (static_key_false(&generic_xdp_needed)) { |
bbbe211c | 4560 | int ret; |
b5cdae32 | 4561 | |
bbbe211c JF |
4562 | preempt_disable(); |
4563 | rcu_read_lock(); | |
4564 | ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
4565 | rcu_read_unlock(); | |
4566 | preempt_enable(); | |
4567 | ||
4568 | if (ret != XDP_PASS) | |
d4455169 | 4569 | return NET_RX_DROP; |
b5cdae32 DM |
4570 | } |
4571 | ||
bbbe211c | 4572 | rcu_read_lock(); |
df334545 | 4573 | #ifdef CONFIG_RPS |
c5905afb | 4574 | if (static_key_false(&rps_needed)) { |
3b098e2d | 4575 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 4576 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 4577 | |
3b098e2d ED |
4578 | if (cpu >= 0) { |
4579 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4580 | rcu_read_unlock(); | |
adc9300e | 4581 | return ret; |
3b098e2d | 4582 | } |
fec5e652 | 4583 | } |
1e94d72f | 4584 | #endif |
2c17d27c JA |
4585 | ret = __netif_receive_skb(skb); |
4586 | rcu_read_unlock(); | |
4587 | return ret; | |
0a9627f2 | 4588 | } |
ae78dbfa BH |
4589 | |
4590 | /** | |
4591 | * netif_receive_skb - process receive buffer from network | |
4592 | * @skb: buffer to process | |
4593 | * | |
4594 | * netif_receive_skb() is the main receive data processing function. | |
4595 | * It always succeeds. The buffer may be dropped during processing | |
4596 | * for congestion control or by the protocol layers. | |
4597 | * | |
4598 | * This function may only be called from softirq context and interrupts | |
4599 | * should be enabled. | |
4600 | * | |
4601 | * Return values (usually ignored): | |
4602 | * NET_RX_SUCCESS: no congestion | |
4603 | * NET_RX_DROP: packet was dropped | |
4604 | */ | |
04eb4489 | 4605 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa BH |
4606 | { |
4607 | trace_netif_receive_skb_entry(skb); | |
4608 | ||
4609 | return netif_receive_skb_internal(skb); | |
4610 | } | |
04eb4489 | 4611 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 4612 | |
41852497 | 4613 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
4614 | |
4615 | /* Network device is going away, flush any packets still pending */ | |
4616 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 4617 | { |
6e583ce5 | 4618 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
4619 | struct softnet_data *sd; |
4620 | ||
4621 | local_bh_disable(); | |
4622 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 4623 | |
145dd5f9 | 4624 | local_irq_disable(); |
e36fa2f7 | 4625 | rps_lock(sd); |
6e7676c1 | 4626 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 4627 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 4628 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 4629 | kfree_skb(skb); |
76cc8b13 | 4630 | input_queue_head_incr(sd); |
6e583ce5 | 4631 | } |
6e7676c1 | 4632 | } |
e36fa2f7 | 4633 | rps_unlock(sd); |
145dd5f9 | 4634 | local_irq_enable(); |
6e7676c1 CG |
4635 | |
4636 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 4637 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
4638 | __skb_unlink(skb, &sd->process_queue); |
4639 | kfree_skb(skb); | |
76cc8b13 | 4640 | input_queue_head_incr(sd); |
6e7676c1 CG |
4641 | } |
4642 | } | |
145dd5f9 PA |
4643 | local_bh_enable(); |
4644 | } | |
4645 | ||
41852497 | 4646 | static void flush_all_backlogs(void) |
145dd5f9 PA |
4647 | { |
4648 | unsigned int cpu; | |
4649 | ||
4650 | get_online_cpus(); | |
4651 | ||
41852497 ED |
4652 | for_each_online_cpu(cpu) |
4653 | queue_work_on(cpu, system_highpri_wq, | |
4654 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
4655 | |
4656 | for_each_online_cpu(cpu) | |
41852497 | 4657 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
4658 | |
4659 | put_online_cpus(); | |
6e583ce5 SH |
4660 | } |
4661 | ||
d565b0a1 HX |
4662 | static int napi_gro_complete(struct sk_buff *skb) |
4663 | { | |
22061d80 | 4664 | struct packet_offload *ptype; |
d565b0a1 | 4665 | __be16 type = skb->protocol; |
22061d80 | 4666 | struct list_head *head = &offload_base; |
d565b0a1 HX |
4667 | int err = -ENOENT; |
4668 | ||
c3c7c254 ED |
4669 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
4670 | ||
fc59f9a3 HX |
4671 | if (NAPI_GRO_CB(skb)->count == 1) { |
4672 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 4673 | goto out; |
fc59f9a3 | 4674 | } |
d565b0a1 HX |
4675 | |
4676 | rcu_read_lock(); | |
4677 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4678 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
4679 | continue; |
4680 | ||
299603e8 | 4681 | err = ptype->callbacks.gro_complete(skb, 0); |
d565b0a1 HX |
4682 | break; |
4683 | } | |
4684 | rcu_read_unlock(); | |
4685 | ||
4686 | if (err) { | |
4687 | WARN_ON(&ptype->list == head); | |
4688 | kfree_skb(skb); | |
4689 | return NET_RX_SUCCESS; | |
4690 | } | |
4691 | ||
4692 | out: | |
ae78dbfa | 4693 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
4694 | } |
4695 | ||
2e71a6f8 ED |
4696 | /* napi->gro_list contains packets ordered by age. |
4697 | * youngest packets at the head of it. | |
4698 | * Complete skbs in reverse order to reduce latencies. | |
4699 | */ | |
4700 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
d565b0a1 | 4701 | { |
2e71a6f8 | 4702 | struct sk_buff *skb, *prev = NULL; |
d565b0a1 | 4703 | |
2e71a6f8 ED |
4704 | /* scan list and build reverse chain */ |
4705 | for (skb = napi->gro_list; skb != NULL; skb = skb->next) { | |
4706 | skb->prev = prev; | |
4707 | prev = skb; | |
4708 | } | |
4709 | ||
4710 | for (skb = prev; skb; skb = prev) { | |
d565b0a1 | 4711 | skb->next = NULL; |
2e71a6f8 ED |
4712 | |
4713 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) | |
4714 | return; | |
4715 | ||
4716 | prev = skb->prev; | |
d565b0a1 | 4717 | napi_gro_complete(skb); |
2e71a6f8 | 4718 | napi->gro_count--; |
d565b0a1 HX |
4719 | } |
4720 | ||
4721 | napi->gro_list = NULL; | |
4722 | } | |
86cac58b | 4723 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 4724 | |
89c5fa33 ED |
4725 | static void gro_list_prepare(struct napi_struct *napi, struct sk_buff *skb) |
4726 | { | |
4727 | struct sk_buff *p; | |
4728 | unsigned int maclen = skb->dev->hard_header_len; | |
0b4cec8c | 4729 | u32 hash = skb_get_hash_raw(skb); |
89c5fa33 ED |
4730 | |
4731 | for (p = napi->gro_list; p; p = p->next) { | |
4732 | unsigned long diffs; | |
4733 | ||
0b4cec8c TH |
4734 | NAPI_GRO_CB(p)->flush = 0; |
4735 | ||
4736 | if (hash != skb_get_hash_raw(p)) { | |
4737 | NAPI_GRO_CB(p)->same_flow = 0; | |
4738 | continue; | |
4739 | } | |
4740 | ||
89c5fa33 ED |
4741 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
4742 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 4743 | diffs |= skb_metadata_dst_cmp(p, skb); |
de8f3a83 | 4744 | diffs |= skb_metadata_differs(p, skb); |
89c5fa33 ED |
4745 | if (maclen == ETH_HLEN) |
4746 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 4747 | skb_mac_header(skb)); |
89c5fa33 ED |
4748 | else if (!diffs) |
4749 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 4750 | skb_mac_header(skb), |
89c5fa33 ED |
4751 | maclen); |
4752 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 ED |
4753 | } |
4754 | } | |
4755 | ||
299603e8 JC |
4756 | static void skb_gro_reset_offset(struct sk_buff *skb) |
4757 | { | |
4758 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4759 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
4760 | ||
4761 | NAPI_GRO_CB(skb)->data_offset = 0; | |
4762 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
4763 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
4764 | ||
4765 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
4766 | pinfo->nr_frags && | |
4767 | !PageHighMem(skb_frag_page(frag0))) { | |
4768 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
4769 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
4770 | skb_frag_size(frag0), | |
4771 | skb->end - skb->tail); | |
89c5fa33 ED |
4772 | } |
4773 | } | |
4774 | ||
a50e233c ED |
4775 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
4776 | { | |
4777 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
4778 | ||
4779 | BUG_ON(skb->end - skb->tail < grow); | |
4780 | ||
4781 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
4782 | ||
4783 | skb->data_len -= grow; | |
4784 | skb->tail += grow; | |
4785 | ||
4786 | pinfo->frags[0].page_offset += grow; | |
4787 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
4788 | ||
4789 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
4790 | skb_frag_unref(skb, 0); | |
4791 | memmove(pinfo->frags, pinfo->frags + 1, | |
4792 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
4793 | } | |
4794 | } | |
4795 | ||
bb728820 | 4796 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 HX |
4797 | { |
4798 | struct sk_buff **pp = NULL; | |
22061d80 | 4799 | struct packet_offload *ptype; |
d565b0a1 | 4800 | __be16 type = skb->protocol; |
22061d80 | 4801 | struct list_head *head = &offload_base; |
0da2afd5 | 4802 | int same_flow; |
5b252f0c | 4803 | enum gro_result ret; |
a50e233c | 4804 | int grow; |
d565b0a1 | 4805 | |
b5cdae32 | 4806 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
4807 | goto normal; |
4808 | ||
89c5fa33 ED |
4809 | gro_list_prepare(napi, skb); |
4810 | ||
d565b0a1 HX |
4811 | rcu_read_lock(); |
4812 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 4813 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
4814 | continue; |
4815 | ||
86911732 | 4816 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 4817 | skb_reset_mac_len(skb); |
d565b0a1 | 4818 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 4819 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 4820 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 4821 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 4822 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 4823 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 4824 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 4825 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 4826 | |
662880f4 TH |
4827 | /* Setup for GRO checksum validation */ |
4828 | switch (skb->ip_summed) { | |
4829 | case CHECKSUM_COMPLETE: | |
4830 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
4831 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
4832 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4833 | break; | |
4834 | case CHECKSUM_UNNECESSARY: | |
4835 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
4836 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4837 | break; | |
4838 | default: | |
4839 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
4840 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
4841 | } | |
d565b0a1 | 4842 | |
f191a1d1 | 4843 | pp = ptype->callbacks.gro_receive(&napi->gro_list, skb); |
d565b0a1 HX |
4844 | break; |
4845 | } | |
4846 | rcu_read_unlock(); | |
4847 | ||
4848 | if (&ptype->list == head) | |
4849 | goto normal; | |
4850 | ||
25393d3f SK |
4851 | if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { |
4852 | ret = GRO_CONSUMED; | |
4853 | goto ok; | |
4854 | } | |
4855 | ||
0da2afd5 | 4856 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 4857 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 4858 | |
d565b0a1 HX |
4859 | if (pp) { |
4860 | struct sk_buff *nskb = *pp; | |
4861 | ||
4862 | *pp = nskb->next; | |
4863 | nskb->next = NULL; | |
4864 | napi_gro_complete(nskb); | |
4ae5544f | 4865 | napi->gro_count--; |
d565b0a1 HX |
4866 | } |
4867 | ||
0da2afd5 | 4868 | if (same_flow) |
d565b0a1 HX |
4869 | goto ok; |
4870 | ||
600adc18 | 4871 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 4872 | goto normal; |
d565b0a1 | 4873 | |
600adc18 ED |
4874 | if (unlikely(napi->gro_count >= MAX_GRO_SKBS)) { |
4875 | struct sk_buff *nskb = napi->gro_list; | |
4876 | ||
4877 | /* locate the end of the list to select the 'oldest' flow */ | |
4878 | while (nskb->next) { | |
4879 | pp = &nskb->next; | |
4880 | nskb = *pp; | |
4881 | } | |
4882 | *pp = NULL; | |
4883 | nskb->next = NULL; | |
4884 | napi_gro_complete(nskb); | |
4885 | } else { | |
4886 | napi->gro_count++; | |
4887 | } | |
d565b0a1 | 4888 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 4889 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 4890 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 4891 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
d565b0a1 HX |
4892 | skb->next = napi->gro_list; |
4893 | napi->gro_list = skb; | |
5d0d9be8 | 4894 | ret = GRO_HELD; |
d565b0a1 | 4895 | |
ad0f9904 | 4896 | pull: |
a50e233c ED |
4897 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
4898 | if (grow > 0) | |
4899 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 4900 | ok: |
5d0d9be8 | 4901 | return ret; |
d565b0a1 HX |
4902 | |
4903 | normal: | |
ad0f9904 HX |
4904 | ret = GRO_NORMAL; |
4905 | goto pull; | |
5d38a079 | 4906 | } |
96e93eab | 4907 | |
bf5a755f JC |
4908 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
4909 | { | |
4910 | struct list_head *offload_head = &offload_base; | |
4911 | struct packet_offload *ptype; | |
4912 | ||
4913 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4914 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
4915 | continue; | |
4916 | return ptype; | |
4917 | } | |
4918 | return NULL; | |
4919 | } | |
e27a2f83 | 4920 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
4921 | |
4922 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
4923 | { | |
4924 | struct list_head *offload_head = &offload_base; | |
4925 | struct packet_offload *ptype; | |
4926 | ||
4927 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
4928 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
4929 | continue; | |
4930 | return ptype; | |
4931 | } | |
4932 | return NULL; | |
4933 | } | |
e27a2f83 | 4934 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 4935 | |
e44699d2 MK |
4936 | static void napi_skb_free_stolen_head(struct sk_buff *skb) |
4937 | { | |
4938 | skb_dst_drop(skb); | |
4939 | secpath_reset(skb); | |
4940 | kmem_cache_free(skbuff_head_cache, skb); | |
4941 | } | |
4942 | ||
bb728820 | 4943 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 4944 | { |
5d0d9be8 HX |
4945 | switch (ret) { |
4946 | case GRO_NORMAL: | |
ae78dbfa | 4947 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
4948 | ret = GRO_DROP; |
4949 | break; | |
5d38a079 | 4950 | |
5d0d9be8 | 4951 | case GRO_DROP: |
5d38a079 HX |
4952 | kfree_skb(skb); |
4953 | break; | |
5b252f0c | 4954 | |
daa86548 | 4955 | case GRO_MERGED_FREE: |
e44699d2 MK |
4956 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
4957 | napi_skb_free_stolen_head(skb); | |
4958 | else | |
d7e8883c | 4959 | __kfree_skb(skb); |
daa86548 ED |
4960 | break; |
4961 | ||
5b252f0c BH |
4962 | case GRO_HELD: |
4963 | case GRO_MERGED: | |
25393d3f | 4964 | case GRO_CONSUMED: |
5b252f0c | 4965 | break; |
5d38a079 HX |
4966 | } |
4967 | ||
c7c4b3b6 | 4968 | return ret; |
5d0d9be8 | 4969 | } |
5d0d9be8 | 4970 | |
c7c4b3b6 | 4971 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 4972 | { |
93f93a44 | 4973 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 4974 | trace_napi_gro_receive_entry(skb); |
86911732 | 4975 | |
a50e233c ED |
4976 | skb_gro_reset_offset(skb); |
4977 | ||
89c5fa33 | 4978 | return napi_skb_finish(dev_gro_receive(napi, skb), skb); |
d565b0a1 HX |
4979 | } |
4980 | EXPORT_SYMBOL(napi_gro_receive); | |
4981 | ||
d0c2b0d2 | 4982 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 4983 | { |
93a35f59 ED |
4984 | if (unlikely(skb->pfmemalloc)) { |
4985 | consume_skb(skb); | |
4986 | return; | |
4987 | } | |
96e93eab | 4988 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
4989 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
4990 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
3701e513 | 4991 | skb->vlan_tci = 0; |
66c46d74 | 4992 | skb->dev = napi->dev; |
6d152e23 | 4993 | skb->skb_iif = 0; |
c3caf119 JC |
4994 | skb->encapsulation = 0; |
4995 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 4996 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
f991bb9d | 4997 | secpath_reset(skb); |
96e93eab HX |
4998 | |
4999 | napi->skb = skb; | |
5000 | } | |
96e93eab | 5001 | |
76620aaf | 5002 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 5003 | { |
5d38a079 | 5004 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
5005 | |
5006 | if (!skb) { | |
fd11a83d | 5007 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
5008 | if (skb) { |
5009 | napi->skb = skb; | |
5010 | skb_mark_napi_id(skb, napi); | |
5011 | } | |
80595d59 | 5012 | } |
96e93eab HX |
5013 | return skb; |
5014 | } | |
76620aaf | 5015 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 5016 | |
a50e233c ED |
5017 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
5018 | struct sk_buff *skb, | |
5019 | gro_result_t ret) | |
96e93eab | 5020 | { |
5d0d9be8 HX |
5021 | switch (ret) { |
5022 | case GRO_NORMAL: | |
a50e233c ED |
5023 | case GRO_HELD: |
5024 | __skb_push(skb, ETH_HLEN); | |
5025 | skb->protocol = eth_type_trans(skb, skb->dev); | |
5026 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 5027 | ret = GRO_DROP; |
86911732 | 5028 | break; |
5d38a079 | 5029 | |
5d0d9be8 | 5030 | case GRO_DROP: |
5d0d9be8 HX |
5031 | napi_reuse_skb(napi, skb); |
5032 | break; | |
5b252f0c | 5033 | |
e44699d2 MK |
5034 | case GRO_MERGED_FREE: |
5035 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
5036 | napi_skb_free_stolen_head(skb); | |
5037 | else | |
5038 | napi_reuse_skb(napi, skb); | |
5039 | break; | |
5040 | ||
5b252f0c | 5041 | case GRO_MERGED: |
25393d3f | 5042 | case GRO_CONSUMED: |
5b252f0c | 5043 | break; |
5d0d9be8 | 5044 | } |
5d38a079 | 5045 | |
c7c4b3b6 | 5046 | return ret; |
5d38a079 | 5047 | } |
5d0d9be8 | 5048 | |
a50e233c ED |
5049 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
5050 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
5051 | * We copy ethernet header into skb->data to have a common layout. | |
5052 | */ | |
4adb9c4a | 5053 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
5054 | { |
5055 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
5056 | const struct ethhdr *eth; |
5057 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
5058 | |
5059 | napi->skb = NULL; | |
5060 | ||
a50e233c ED |
5061 | skb_reset_mac_header(skb); |
5062 | skb_gro_reset_offset(skb); | |
5063 | ||
5064 | eth = skb_gro_header_fast(skb, 0); | |
5065 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
5066 | eth = skb_gro_header_slow(skb, hlen, 0); | |
5067 | if (unlikely(!eth)) { | |
4da46ceb AC |
5068 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
5069 | __func__, napi->dev->name); | |
a50e233c ED |
5070 | napi_reuse_skb(napi, skb); |
5071 | return NULL; | |
5072 | } | |
5073 | } else { | |
5074 | gro_pull_from_frag0(skb, hlen); | |
5075 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
5076 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 5077 | } |
a50e233c ED |
5078 | __skb_pull(skb, hlen); |
5079 | ||
5080 | /* | |
5081 | * This works because the only protocols we care about don't require | |
5082 | * special handling. | |
5083 | * We'll fix it up properly in napi_frags_finish() | |
5084 | */ | |
5085 | skb->protocol = eth->h_proto; | |
76620aaf | 5086 | |
76620aaf HX |
5087 | return skb; |
5088 | } | |
76620aaf | 5089 | |
c7c4b3b6 | 5090 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 5091 | { |
76620aaf | 5092 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
5093 | |
5094 | if (!skb) | |
c7c4b3b6 | 5095 | return GRO_DROP; |
5d0d9be8 | 5096 | |
ae78dbfa BH |
5097 | trace_napi_gro_frags_entry(skb); |
5098 | ||
89c5fa33 | 5099 | return napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5d0d9be8 | 5100 | } |
5d38a079 HX |
5101 | EXPORT_SYMBOL(napi_gro_frags); |
5102 | ||
573e8fca TH |
5103 | /* Compute the checksum from gro_offset and return the folded value |
5104 | * after adding in any pseudo checksum. | |
5105 | */ | |
5106 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
5107 | { | |
5108 | __wsum wsum; | |
5109 | __sum16 sum; | |
5110 | ||
5111 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
5112 | ||
5113 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
5114 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
5115 | if (likely(!sum)) { | |
5116 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
5117 | !skb->csum_complete_sw) | |
5118 | netdev_rx_csum_fault(skb->dev); | |
5119 | } | |
5120 | ||
5121 | NAPI_GRO_CB(skb)->csum = wsum; | |
5122 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5123 | ||
5124 | return sum; | |
5125 | } | |
5126 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
5127 | ||
773fc8f6 | 5128 | static void net_rps_send_ipi(struct softnet_data *remsd) |
5129 | { | |
5130 | #ifdef CONFIG_RPS | |
5131 | while (remsd) { | |
5132 | struct softnet_data *next = remsd->rps_ipi_next; | |
5133 | ||
5134 | if (cpu_online(remsd->cpu)) | |
5135 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
5136 | remsd = next; | |
5137 | } | |
5138 | #endif | |
5139 | } | |
5140 | ||
e326bed2 | 5141 | /* |
855abcf0 | 5142 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
5143 | * Note: called with local irq disabled, but exits with local irq enabled. |
5144 | */ | |
5145 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
5146 | { | |
5147 | #ifdef CONFIG_RPS | |
5148 | struct softnet_data *remsd = sd->rps_ipi_list; | |
5149 | ||
5150 | if (remsd) { | |
5151 | sd->rps_ipi_list = NULL; | |
5152 | ||
5153 | local_irq_enable(); | |
5154 | ||
5155 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 5156 | net_rps_send_ipi(remsd); |
e326bed2 ED |
5157 | } else |
5158 | #endif | |
5159 | local_irq_enable(); | |
5160 | } | |
5161 | ||
d75b1ade ED |
5162 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
5163 | { | |
5164 | #ifdef CONFIG_RPS | |
5165 | return sd->rps_ipi_list != NULL; | |
5166 | #else | |
5167 | return false; | |
5168 | #endif | |
5169 | } | |
5170 | ||
bea3348e | 5171 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 5172 | { |
eecfd7c4 | 5173 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
5174 | bool again = true; |
5175 | int work = 0; | |
1da177e4 | 5176 | |
e326bed2 ED |
5177 | /* Check if we have pending ipi, its better to send them now, |
5178 | * not waiting net_rx_action() end. | |
5179 | */ | |
d75b1ade | 5180 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
5181 | local_irq_disable(); |
5182 | net_rps_action_and_irq_enable(sd); | |
5183 | } | |
d75b1ade | 5184 | |
3d48b53f | 5185 | napi->weight = dev_rx_weight; |
145dd5f9 | 5186 | while (again) { |
1da177e4 | 5187 | struct sk_buff *skb; |
6e7676c1 CG |
5188 | |
5189 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 5190 | rcu_read_lock(); |
6e7676c1 | 5191 | __netif_receive_skb(skb); |
2c17d27c | 5192 | rcu_read_unlock(); |
76cc8b13 | 5193 | input_queue_head_incr(sd); |
145dd5f9 | 5194 | if (++work >= quota) |
76cc8b13 | 5195 | return work; |
145dd5f9 | 5196 | |
6e7676c1 | 5197 | } |
1da177e4 | 5198 | |
145dd5f9 | 5199 | local_irq_disable(); |
e36fa2f7 | 5200 | rps_lock(sd); |
11ef7a89 | 5201 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
5202 | /* |
5203 | * Inline a custom version of __napi_complete(). | |
5204 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
5205 | * and NAPI_STATE_SCHED is the only possible flag set |
5206 | * on backlog. | |
5207 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
5208 | * and we dont need an smp_mb() memory barrier. |
5209 | */ | |
eecfd7c4 | 5210 | napi->state = 0; |
145dd5f9 PA |
5211 | again = false; |
5212 | } else { | |
5213 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
5214 | &sd->process_queue); | |
bea3348e | 5215 | } |
e36fa2f7 | 5216 | rps_unlock(sd); |
145dd5f9 | 5217 | local_irq_enable(); |
6e7676c1 | 5218 | } |
1da177e4 | 5219 | |
bea3348e SH |
5220 | return work; |
5221 | } | |
1da177e4 | 5222 | |
bea3348e SH |
5223 | /** |
5224 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 5225 | * @n: entry to schedule |
bea3348e | 5226 | * |
bc9ad166 ED |
5227 | * The entry's receive function will be scheduled to run. |
5228 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 5229 | */ |
b5606c2d | 5230 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
5231 | { |
5232 | unsigned long flags; | |
1da177e4 | 5233 | |
bea3348e | 5234 | local_irq_save(flags); |
903ceff7 | 5235 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 5236 | local_irq_restore(flags); |
1da177e4 | 5237 | } |
bea3348e SH |
5238 | EXPORT_SYMBOL(__napi_schedule); |
5239 | ||
39e6c820 ED |
5240 | /** |
5241 | * napi_schedule_prep - check if napi can be scheduled | |
5242 | * @n: napi context | |
5243 | * | |
5244 | * Test if NAPI routine is already running, and if not mark | |
5245 | * it as running. This is used as a condition variable | |
5246 | * insure only one NAPI poll instance runs. We also make | |
5247 | * sure there is no pending NAPI disable. | |
5248 | */ | |
5249 | bool napi_schedule_prep(struct napi_struct *n) | |
5250 | { | |
5251 | unsigned long val, new; | |
5252 | ||
5253 | do { | |
5254 | val = READ_ONCE(n->state); | |
5255 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
5256 | return false; | |
5257 | new = val | NAPIF_STATE_SCHED; | |
5258 | ||
5259 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
5260 | * This was suggested by Alexander Duyck, as compiler | |
5261 | * emits better code than : | |
5262 | * if (val & NAPIF_STATE_SCHED) | |
5263 | * new |= NAPIF_STATE_MISSED; | |
5264 | */ | |
5265 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
5266 | NAPIF_STATE_MISSED; | |
5267 | } while (cmpxchg(&n->state, val, new) != val); | |
5268 | ||
5269 | return !(val & NAPIF_STATE_SCHED); | |
5270 | } | |
5271 | EXPORT_SYMBOL(napi_schedule_prep); | |
5272 | ||
bc9ad166 ED |
5273 | /** |
5274 | * __napi_schedule_irqoff - schedule for receive | |
5275 | * @n: entry to schedule | |
5276 | * | |
5277 | * Variant of __napi_schedule() assuming hard irqs are masked | |
5278 | */ | |
5279 | void __napi_schedule_irqoff(struct napi_struct *n) | |
5280 | { | |
5281 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
5282 | } | |
5283 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
5284 | ||
364b6055 | 5285 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 5286 | { |
39e6c820 | 5287 | unsigned long flags, val, new; |
d565b0a1 HX |
5288 | |
5289 | /* | |
217f6974 ED |
5290 | * 1) Don't let napi dequeue from the cpu poll list |
5291 | * just in case its running on a different cpu. | |
5292 | * 2) If we are busy polling, do nothing here, we have | |
5293 | * the guarantee we will be called later. | |
d565b0a1 | 5294 | */ |
217f6974 ED |
5295 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
5296 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 5297 | return false; |
d565b0a1 | 5298 | |
3b47d303 ED |
5299 | if (n->gro_list) { |
5300 | unsigned long timeout = 0; | |
d75b1ade | 5301 | |
3b47d303 ED |
5302 | if (work_done) |
5303 | timeout = n->dev->gro_flush_timeout; | |
5304 | ||
5305 | if (timeout) | |
5306 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
5307 | HRTIMER_MODE_REL_PINNED); | |
5308 | else | |
5309 | napi_gro_flush(n, false); | |
5310 | } | |
02c1602e | 5311 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
5312 | /* If n->poll_list is not empty, we need to mask irqs */ |
5313 | local_irq_save(flags); | |
02c1602e | 5314 | list_del_init(&n->poll_list); |
d75b1ade ED |
5315 | local_irq_restore(flags); |
5316 | } | |
39e6c820 ED |
5317 | |
5318 | do { | |
5319 | val = READ_ONCE(n->state); | |
5320 | ||
5321 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
5322 | ||
5323 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
5324 | ||
5325 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
5326 | * because we will call napi->poll() one more time. | |
5327 | * This C code was suggested by Alexander Duyck to help gcc. | |
5328 | */ | |
5329 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
5330 | NAPIF_STATE_SCHED; | |
5331 | } while (cmpxchg(&n->state, val, new) != val); | |
5332 | ||
5333 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
5334 | __napi_schedule(n); | |
5335 | return false; | |
5336 | } | |
5337 | ||
364b6055 | 5338 | return true; |
d565b0a1 | 5339 | } |
3b47d303 | 5340 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 5341 | |
af12fa6e | 5342 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 5343 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
5344 | { |
5345 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
5346 | struct napi_struct *napi; | |
5347 | ||
5348 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
5349 | if (napi->napi_id == napi_id) | |
5350 | return napi; | |
5351 | ||
5352 | return NULL; | |
5353 | } | |
02d62e86 ED |
5354 | |
5355 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 5356 | |
ce6aea93 | 5357 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
5358 | |
5359 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
5360 | { | |
5361 | int rc; | |
5362 | ||
39e6c820 ED |
5363 | /* Busy polling means there is a high chance device driver hard irq |
5364 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
5365 | * set in napi_schedule_prep(). | |
5366 | * Since we are about to call napi->poll() once more, we can safely | |
5367 | * clear NAPI_STATE_MISSED. | |
5368 | * | |
5369 | * Note: x86 could use a single "lock and ..." instruction | |
5370 | * to perform these two clear_bit() | |
5371 | */ | |
5372 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
5373 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
5374 | ||
5375 | local_bh_disable(); | |
5376 | ||
5377 | /* All we really want here is to re-enable device interrupts. | |
5378 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
5379 | */ | |
5380 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
1e22391e | 5381 | trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); |
217f6974 ED |
5382 | netpoll_poll_unlock(have_poll_lock); |
5383 | if (rc == BUSY_POLL_BUDGET) | |
5384 | __napi_schedule(napi); | |
5385 | local_bh_enable(); | |
217f6974 ED |
5386 | } |
5387 | ||
7db6b048 SS |
5388 | void napi_busy_loop(unsigned int napi_id, |
5389 | bool (*loop_end)(void *, unsigned long), | |
5390 | void *loop_end_arg) | |
02d62e86 | 5391 | { |
7db6b048 | 5392 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 5393 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 5394 | void *have_poll_lock = NULL; |
02d62e86 | 5395 | struct napi_struct *napi; |
217f6974 ED |
5396 | |
5397 | restart: | |
217f6974 | 5398 | napi_poll = NULL; |
02d62e86 | 5399 | |
2a028ecb | 5400 | rcu_read_lock(); |
02d62e86 | 5401 | |
545cd5e5 | 5402 | napi = napi_by_id(napi_id); |
02d62e86 ED |
5403 | if (!napi) |
5404 | goto out; | |
5405 | ||
217f6974 ED |
5406 | preempt_disable(); |
5407 | for (;;) { | |
2b5cd0df AD |
5408 | int work = 0; |
5409 | ||
2a028ecb | 5410 | local_bh_disable(); |
217f6974 ED |
5411 | if (!napi_poll) { |
5412 | unsigned long val = READ_ONCE(napi->state); | |
5413 | ||
5414 | /* If multiple threads are competing for this napi, | |
5415 | * we avoid dirtying napi->state as much as we can. | |
5416 | */ | |
5417 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
5418 | NAPIF_STATE_IN_BUSY_POLL)) | |
5419 | goto count; | |
5420 | if (cmpxchg(&napi->state, val, | |
5421 | val | NAPIF_STATE_IN_BUSY_POLL | | |
5422 | NAPIF_STATE_SCHED) != val) | |
5423 | goto count; | |
5424 | have_poll_lock = netpoll_poll_lock(napi); | |
5425 | napi_poll = napi->poll; | |
5426 | } | |
2b5cd0df AD |
5427 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
5428 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
217f6974 | 5429 | count: |
2b5cd0df | 5430 | if (work > 0) |
7db6b048 | 5431 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 5432 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 5433 | local_bh_enable(); |
02d62e86 | 5434 | |
7db6b048 | 5435 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 5436 | break; |
02d62e86 | 5437 | |
217f6974 ED |
5438 | if (unlikely(need_resched())) { |
5439 | if (napi_poll) | |
5440 | busy_poll_stop(napi, have_poll_lock); | |
5441 | preempt_enable(); | |
5442 | rcu_read_unlock(); | |
5443 | cond_resched(); | |
7db6b048 | 5444 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 5445 | return; |
217f6974 ED |
5446 | goto restart; |
5447 | } | |
6cdf89b1 | 5448 | cpu_relax(); |
217f6974 ED |
5449 | } |
5450 | if (napi_poll) | |
5451 | busy_poll_stop(napi, have_poll_lock); | |
5452 | preempt_enable(); | |
02d62e86 | 5453 | out: |
2a028ecb | 5454 | rcu_read_unlock(); |
02d62e86 | 5455 | } |
7db6b048 | 5456 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
5457 | |
5458 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 5459 | |
149d6ad8 | 5460 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 5461 | { |
d64b5e85 ED |
5462 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
5463 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 5464 | return; |
af12fa6e | 5465 | |
52bd2d62 | 5466 | spin_lock(&napi_hash_lock); |
af12fa6e | 5467 | |
545cd5e5 | 5468 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 5469 | do { |
545cd5e5 AD |
5470 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
5471 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
5472 | } while (napi_by_id(napi_gen_id)); |
5473 | napi->napi_id = napi_gen_id; | |
af12fa6e | 5474 | |
52bd2d62 ED |
5475 | hlist_add_head_rcu(&napi->napi_hash_node, |
5476 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 5477 | |
52bd2d62 | 5478 | spin_unlock(&napi_hash_lock); |
af12fa6e | 5479 | } |
af12fa6e ET |
5480 | |
5481 | /* Warning : caller is responsible to make sure rcu grace period | |
5482 | * is respected before freeing memory containing @napi | |
5483 | */ | |
34cbe27e | 5484 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 5485 | { |
34cbe27e ED |
5486 | bool rcu_sync_needed = false; |
5487 | ||
af12fa6e ET |
5488 | spin_lock(&napi_hash_lock); |
5489 | ||
34cbe27e ED |
5490 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
5491 | rcu_sync_needed = true; | |
af12fa6e | 5492 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 5493 | } |
af12fa6e | 5494 | spin_unlock(&napi_hash_lock); |
34cbe27e | 5495 | return rcu_sync_needed; |
af12fa6e ET |
5496 | } |
5497 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
5498 | ||
3b47d303 ED |
5499 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
5500 | { | |
5501 | struct napi_struct *napi; | |
5502 | ||
5503 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
5504 | |
5505 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
5506 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
5507 | */ | |
5508 | if (napi->gro_list && !napi_disable_pending(napi) && | |
5509 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) | |
5510 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
5511 | |
5512 | return HRTIMER_NORESTART; | |
5513 | } | |
5514 | ||
d565b0a1 HX |
5515 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, |
5516 | int (*poll)(struct napi_struct *, int), int weight) | |
5517 | { | |
5518 | INIT_LIST_HEAD(&napi->poll_list); | |
3b47d303 ED |
5519 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); |
5520 | napi->timer.function = napi_watchdog; | |
4ae5544f | 5521 | napi->gro_count = 0; |
d565b0a1 | 5522 | napi->gro_list = NULL; |
5d38a079 | 5523 | napi->skb = NULL; |
d565b0a1 | 5524 | napi->poll = poll; |
82dc3c63 ED |
5525 | if (weight > NAPI_POLL_WEIGHT) |
5526 | pr_err_once("netif_napi_add() called with weight %d on device %s\n", | |
5527 | weight, dev->name); | |
d565b0a1 HX |
5528 | napi->weight = weight; |
5529 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 5530 | napi->dev = dev; |
5d38a079 | 5531 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
5532 | napi->poll_owner = -1; |
5533 | #endif | |
5534 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 5535 | napi_hash_add(napi); |
d565b0a1 HX |
5536 | } |
5537 | EXPORT_SYMBOL(netif_napi_add); | |
5538 | ||
3b47d303 ED |
5539 | void napi_disable(struct napi_struct *n) |
5540 | { | |
5541 | might_sleep(); | |
5542 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
5543 | ||
5544 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
5545 | msleep(1); | |
2d8bff12 NH |
5546 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
5547 | msleep(1); | |
3b47d303 ED |
5548 | |
5549 | hrtimer_cancel(&n->timer); | |
5550 | ||
5551 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
5552 | } | |
5553 | EXPORT_SYMBOL(napi_disable); | |
5554 | ||
93d05d4a | 5555 | /* Must be called in process context */ |
d565b0a1 HX |
5556 | void netif_napi_del(struct napi_struct *napi) |
5557 | { | |
93d05d4a ED |
5558 | might_sleep(); |
5559 | if (napi_hash_del(napi)) | |
5560 | synchronize_net(); | |
d7b06636 | 5561 | list_del_init(&napi->dev_list); |
76620aaf | 5562 | napi_free_frags(napi); |
d565b0a1 | 5563 | |
289dccbe | 5564 | kfree_skb_list(napi->gro_list); |
d565b0a1 | 5565 | napi->gro_list = NULL; |
4ae5544f | 5566 | napi->gro_count = 0; |
d565b0a1 HX |
5567 | } |
5568 | EXPORT_SYMBOL(netif_napi_del); | |
5569 | ||
726ce70e HX |
5570 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
5571 | { | |
5572 | void *have; | |
5573 | int work, weight; | |
5574 | ||
5575 | list_del_init(&n->poll_list); | |
5576 | ||
5577 | have = netpoll_poll_lock(n); | |
5578 | ||
5579 | weight = n->weight; | |
5580 | ||
5581 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
5582 | * with netpoll's poll_napi(). Only the entity which | |
5583 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
5584 | * actually make the ->poll() call. Therefore we avoid | |
5585 | * accidentally calling ->poll() when NAPI is not scheduled. | |
5586 | */ | |
5587 | work = 0; | |
5588 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
5589 | work = n->poll(n, weight); | |
1db19db7 | 5590 | trace_napi_poll(n, work, weight); |
726ce70e HX |
5591 | } |
5592 | ||
5593 | WARN_ON_ONCE(work > weight); | |
5594 | ||
5595 | if (likely(work < weight)) | |
5596 | goto out_unlock; | |
5597 | ||
5598 | /* Drivers must not modify the NAPI state if they | |
5599 | * consume the entire weight. In such cases this code | |
5600 | * still "owns" the NAPI instance and therefore can | |
5601 | * move the instance around on the list at-will. | |
5602 | */ | |
5603 | if (unlikely(napi_disable_pending(n))) { | |
5604 | napi_complete(n); | |
5605 | goto out_unlock; | |
5606 | } | |
5607 | ||
5608 | if (n->gro_list) { | |
5609 | /* flush too old packets | |
5610 | * If HZ < 1000, flush all packets. | |
5611 | */ | |
5612 | napi_gro_flush(n, HZ >= 1000); | |
5613 | } | |
5614 | ||
001ce546 HX |
5615 | /* Some drivers may have called napi_schedule |
5616 | * prior to exhausting their budget. | |
5617 | */ | |
5618 | if (unlikely(!list_empty(&n->poll_list))) { | |
5619 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
5620 | n->dev ? n->dev->name : "backlog"); | |
5621 | goto out_unlock; | |
5622 | } | |
5623 | ||
726ce70e HX |
5624 | list_add_tail(&n->poll_list, repoll); |
5625 | ||
5626 | out_unlock: | |
5627 | netpoll_poll_unlock(have); | |
5628 | ||
5629 | return work; | |
5630 | } | |
5631 | ||
0766f788 | 5632 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 5633 | { |
903ceff7 | 5634 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
5635 | unsigned long time_limit = jiffies + |
5636 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 5637 | int budget = netdev_budget; |
d75b1ade ED |
5638 | LIST_HEAD(list); |
5639 | LIST_HEAD(repoll); | |
53fb95d3 | 5640 | |
1da177e4 | 5641 | local_irq_disable(); |
d75b1ade ED |
5642 | list_splice_init(&sd->poll_list, &list); |
5643 | local_irq_enable(); | |
1da177e4 | 5644 | |
ceb8d5bf | 5645 | for (;;) { |
bea3348e | 5646 | struct napi_struct *n; |
1da177e4 | 5647 | |
ceb8d5bf HX |
5648 | if (list_empty(&list)) { |
5649 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 5650 | goto out; |
ceb8d5bf HX |
5651 | break; |
5652 | } | |
5653 | ||
6bd373eb HX |
5654 | n = list_first_entry(&list, struct napi_struct, poll_list); |
5655 | budget -= napi_poll(n, &repoll); | |
5656 | ||
d75b1ade | 5657 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
5658 | * Allow this to run for 2 jiffies since which will allow |
5659 | * an average latency of 1.5/HZ. | |
bea3348e | 5660 | */ |
ceb8d5bf HX |
5661 | if (unlikely(budget <= 0 || |
5662 | time_after_eq(jiffies, time_limit))) { | |
5663 | sd->time_squeeze++; | |
5664 | break; | |
5665 | } | |
1da177e4 | 5666 | } |
d75b1ade | 5667 | |
d75b1ade ED |
5668 | local_irq_disable(); |
5669 | ||
5670 | list_splice_tail_init(&sd->poll_list, &list); | |
5671 | list_splice_tail(&repoll, &list); | |
5672 | list_splice(&list, &sd->poll_list); | |
5673 | if (!list_empty(&sd->poll_list)) | |
5674 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
5675 | ||
e326bed2 | 5676 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
5677 | out: |
5678 | __kfree_skb_flush(); | |
1da177e4 LT |
5679 | } |
5680 | ||
aa9d8560 | 5681 | struct netdev_adjacent { |
9ff162a8 | 5682 | struct net_device *dev; |
5d261913 VF |
5683 | |
5684 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 5685 | bool master; |
5d261913 | 5686 | |
5d261913 VF |
5687 | /* counter for the number of times this device was added to us */ |
5688 | u16 ref_nr; | |
5689 | ||
402dae96 VF |
5690 | /* private field for the users */ |
5691 | void *private; | |
5692 | ||
9ff162a8 JP |
5693 | struct list_head list; |
5694 | struct rcu_head rcu; | |
9ff162a8 JP |
5695 | }; |
5696 | ||
6ea29da1 | 5697 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 5698 | struct list_head *adj_list) |
9ff162a8 | 5699 | { |
5d261913 | 5700 | struct netdev_adjacent *adj; |
5d261913 | 5701 | |
2f268f12 | 5702 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
5703 | if (adj->dev == adj_dev) |
5704 | return adj; | |
9ff162a8 JP |
5705 | } |
5706 | return NULL; | |
5707 | } | |
5708 | ||
f1170fd4 DA |
5709 | static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
5710 | { | |
5711 | struct net_device *dev = data; | |
5712 | ||
5713 | return upper_dev == dev; | |
5714 | } | |
5715 | ||
9ff162a8 JP |
5716 | /** |
5717 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
5718 | * @dev: device | |
5719 | * @upper_dev: upper device to check | |
5720 | * | |
5721 | * Find out if a device is linked to specified upper device and return true | |
5722 | * in case it is. Note that this checks only immediate upper device, | |
5723 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
5724 | */ | |
5725 | bool netdev_has_upper_dev(struct net_device *dev, | |
5726 | struct net_device *upper_dev) | |
5727 | { | |
5728 | ASSERT_RTNL(); | |
5729 | ||
f1170fd4 DA |
5730 | return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, |
5731 | upper_dev); | |
9ff162a8 JP |
5732 | } |
5733 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
5734 | ||
1a3f060c DA |
5735 | /** |
5736 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
5737 | * @dev: device | |
5738 | * @upper_dev: upper device to check | |
5739 | * | |
5740 | * Find out if a device is linked to specified upper device and return true | |
5741 | * in case it is. Note that this checks the entire upper device chain. | |
5742 | * The caller must hold rcu lock. | |
5743 | */ | |
5744 | ||
1a3f060c DA |
5745 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
5746 | struct net_device *upper_dev) | |
5747 | { | |
5748 | return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, | |
5749 | upper_dev); | |
5750 | } | |
5751 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
5752 | ||
9ff162a8 JP |
5753 | /** |
5754 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
5755 | * @dev: device | |
5756 | * | |
5757 | * Find out if a device is linked to an upper device and return true in case | |
5758 | * it is. The caller must hold the RTNL lock. | |
5759 | */ | |
25cc72a3 | 5760 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
5761 | { |
5762 | ASSERT_RTNL(); | |
5763 | ||
f1170fd4 | 5764 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 5765 | } |
25cc72a3 | 5766 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
5767 | |
5768 | /** | |
5769 | * netdev_master_upper_dev_get - Get master upper device | |
5770 | * @dev: device | |
5771 | * | |
5772 | * Find a master upper device and return pointer to it or NULL in case | |
5773 | * it's not there. The caller must hold the RTNL lock. | |
5774 | */ | |
5775 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
5776 | { | |
aa9d8560 | 5777 | struct netdev_adjacent *upper; |
9ff162a8 JP |
5778 | |
5779 | ASSERT_RTNL(); | |
5780 | ||
2f268f12 | 5781 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
5782 | return NULL; |
5783 | ||
2f268f12 | 5784 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 5785 | struct netdev_adjacent, list); |
9ff162a8 JP |
5786 | if (likely(upper->master)) |
5787 | return upper->dev; | |
5788 | return NULL; | |
5789 | } | |
5790 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
5791 | ||
0f524a80 DA |
5792 | /** |
5793 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
5794 | * @dev: device | |
5795 | * | |
5796 | * Find out if a device is linked to a lower device and return true in case | |
5797 | * it is. The caller must hold the RTNL lock. | |
5798 | */ | |
5799 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
5800 | { | |
5801 | ASSERT_RTNL(); | |
5802 | ||
5803 | return !list_empty(&dev->adj_list.lower); | |
5804 | } | |
5805 | ||
b6ccba4c VF |
5806 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
5807 | { | |
5808 | struct netdev_adjacent *adj; | |
5809 | ||
5810 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
5811 | ||
5812 | return adj->private; | |
5813 | } | |
5814 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
5815 | ||
44a40855 VY |
5816 | /** |
5817 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
5818 | * @dev: device | |
5819 | * @iter: list_head ** of the current position | |
5820 | * | |
5821 | * Gets the next device from the dev's upper list, starting from iter | |
5822 | * position. The caller must hold RCU read lock. | |
5823 | */ | |
5824 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
5825 | struct list_head **iter) | |
5826 | { | |
5827 | struct netdev_adjacent *upper; | |
5828 | ||
5829 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5830 | ||
5831 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5832 | ||
5833 | if (&upper->list == &dev->adj_list.upper) | |
5834 | return NULL; | |
5835 | ||
5836 | *iter = &upper->list; | |
5837 | ||
5838 | return upper->dev; | |
5839 | } | |
5840 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
5841 | ||
1a3f060c DA |
5842 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
5843 | struct list_head **iter) | |
5844 | { | |
5845 | struct netdev_adjacent *upper; | |
5846 | ||
5847 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
5848 | ||
5849 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5850 | ||
5851 | if (&upper->list == &dev->adj_list.upper) | |
5852 | return NULL; | |
5853 | ||
5854 | *iter = &upper->list; | |
5855 | ||
5856 | return upper->dev; | |
5857 | } | |
5858 | ||
5859 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, | |
5860 | int (*fn)(struct net_device *dev, | |
5861 | void *data), | |
5862 | void *data) | |
5863 | { | |
5864 | struct net_device *udev; | |
5865 | struct list_head *iter; | |
5866 | int ret; | |
5867 | ||
5868 | for (iter = &dev->adj_list.upper, | |
5869 | udev = netdev_next_upper_dev_rcu(dev, &iter); | |
5870 | udev; | |
5871 | udev = netdev_next_upper_dev_rcu(dev, &iter)) { | |
5872 | /* first is the upper device itself */ | |
5873 | ret = fn(udev, data); | |
5874 | if (ret) | |
5875 | return ret; | |
5876 | ||
5877 | /* then look at all of its upper devices */ | |
5878 | ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); | |
5879 | if (ret) | |
5880 | return ret; | |
5881 | } | |
5882 | ||
5883 | return 0; | |
5884 | } | |
5885 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
5886 | ||
31088a11 VF |
5887 | /** |
5888 | * netdev_lower_get_next_private - Get the next ->private from the | |
5889 | * lower neighbour list | |
5890 | * @dev: device | |
5891 | * @iter: list_head ** of the current position | |
5892 | * | |
5893 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5894 | * list, starting from iter position. The caller must hold either hold the | |
5895 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 5896 | * list will remain unchanged. |
31088a11 VF |
5897 | */ |
5898 | void *netdev_lower_get_next_private(struct net_device *dev, | |
5899 | struct list_head **iter) | |
5900 | { | |
5901 | struct netdev_adjacent *lower; | |
5902 | ||
5903 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
5904 | ||
5905 | if (&lower->list == &dev->adj_list.lower) | |
5906 | return NULL; | |
5907 | ||
6859e7df | 5908 | *iter = lower->list.next; |
31088a11 VF |
5909 | |
5910 | return lower->private; | |
5911 | } | |
5912 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
5913 | ||
5914 | /** | |
5915 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
5916 | * lower neighbour list, RCU | |
5917 | * variant | |
5918 | * @dev: device | |
5919 | * @iter: list_head ** of the current position | |
5920 | * | |
5921 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
5922 | * list, starting from iter position. The caller must hold RCU read lock. | |
5923 | */ | |
5924 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
5925 | struct list_head **iter) | |
5926 | { | |
5927 | struct netdev_adjacent *lower; | |
5928 | ||
5929 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
5930 | ||
5931 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
5932 | ||
5933 | if (&lower->list == &dev->adj_list.lower) | |
5934 | return NULL; | |
5935 | ||
6859e7df | 5936 | *iter = &lower->list; |
31088a11 VF |
5937 | |
5938 | return lower->private; | |
5939 | } | |
5940 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
5941 | ||
4085ebe8 VY |
5942 | /** |
5943 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
5944 | * list | |
5945 | * @dev: device | |
5946 | * @iter: list_head ** of the current position | |
5947 | * | |
5948 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
5949 | * list, starting from iter position. The caller must hold RTNL lock or | |
5950 | * its own locking that guarantees that the neighbour lower | |
b469139e | 5951 | * list will remain unchanged. |
4085ebe8 VY |
5952 | */ |
5953 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
5954 | { | |
5955 | struct netdev_adjacent *lower; | |
5956 | ||
cfdd28be | 5957 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
5958 | |
5959 | if (&lower->list == &dev->adj_list.lower) | |
5960 | return NULL; | |
5961 | ||
cfdd28be | 5962 | *iter = lower->list.next; |
4085ebe8 VY |
5963 | |
5964 | return lower->dev; | |
5965 | } | |
5966 | EXPORT_SYMBOL(netdev_lower_get_next); | |
5967 | ||
1a3f060c DA |
5968 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
5969 | struct list_head **iter) | |
5970 | { | |
5971 | struct netdev_adjacent *lower; | |
5972 | ||
46b5ab1a | 5973 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
5974 | |
5975 | if (&lower->list == &dev->adj_list.lower) | |
5976 | return NULL; | |
5977 | ||
46b5ab1a | 5978 | *iter = &lower->list; |
1a3f060c DA |
5979 | |
5980 | return lower->dev; | |
5981 | } | |
5982 | ||
5983 | int netdev_walk_all_lower_dev(struct net_device *dev, | |
5984 | int (*fn)(struct net_device *dev, | |
5985 | void *data), | |
5986 | void *data) | |
5987 | { | |
5988 | struct net_device *ldev; | |
5989 | struct list_head *iter; | |
5990 | int ret; | |
5991 | ||
5992 | for (iter = &dev->adj_list.lower, | |
5993 | ldev = netdev_next_lower_dev(dev, &iter); | |
5994 | ldev; | |
5995 | ldev = netdev_next_lower_dev(dev, &iter)) { | |
5996 | /* first is the lower device itself */ | |
5997 | ret = fn(ldev, data); | |
5998 | if (ret) | |
5999 | return ret; | |
6000 | ||
6001 | /* then look at all of its lower devices */ | |
6002 | ret = netdev_walk_all_lower_dev(ldev, fn, data); | |
6003 | if (ret) | |
6004 | return ret; | |
6005 | } | |
6006 | ||
6007 | return 0; | |
6008 | } | |
6009 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
6010 | ||
1a3f060c DA |
6011 | static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
6012 | struct list_head **iter) | |
6013 | { | |
6014 | struct netdev_adjacent *lower; | |
6015 | ||
6016 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6017 | if (&lower->list == &dev->adj_list.lower) | |
6018 | return NULL; | |
6019 | ||
6020 | *iter = &lower->list; | |
6021 | ||
6022 | return lower->dev; | |
6023 | } | |
6024 | ||
6025 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
6026 | int (*fn)(struct net_device *dev, | |
6027 | void *data), | |
6028 | void *data) | |
6029 | { | |
6030 | struct net_device *ldev; | |
6031 | struct list_head *iter; | |
6032 | int ret; | |
6033 | ||
6034 | for (iter = &dev->adj_list.lower, | |
6035 | ldev = netdev_next_lower_dev_rcu(dev, &iter); | |
6036 | ldev; | |
6037 | ldev = netdev_next_lower_dev_rcu(dev, &iter)) { | |
6038 | /* first is the lower device itself */ | |
6039 | ret = fn(ldev, data); | |
6040 | if (ret) | |
6041 | return ret; | |
6042 | ||
6043 | /* then look at all of its lower devices */ | |
6044 | ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); | |
6045 | if (ret) | |
6046 | return ret; | |
6047 | } | |
6048 | ||
6049 | return 0; | |
6050 | } | |
6051 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
6052 | ||
e001bfad | 6053 | /** |
6054 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
6055 | * lower neighbour list, RCU | |
6056 | * variant | |
6057 | * @dev: device | |
6058 | * | |
6059 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
6060 | * list. The caller must hold RCU read lock. | |
6061 | */ | |
6062 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
6063 | { | |
6064 | struct netdev_adjacent *lower; | |
6065 | ||
6066 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
6067 | struct netdev_adjacent, list); | |
6068 | if (lower) | |
6069 | return lower->private; | |
6070 | return NULL; | |
6071 | } | |
6072 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
6073 | ||
9ff162a8 JP |
6074 | /** |
6075 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
6076 | * @dev: device | |
6077 | * | |
6078 | * Find a master upper device and return pointer to it or NULL in case | |
6079 | * it's not there. The caller must hold the RCU read lock. | |
6080 | */ | |
6081 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
6082 | { | |
aa9d8560 | 6083 | struct netdev_adjacent *upper; |
9ff162a8 | 6084 | |
2f268f12 | 6085 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 6086 | struct netdev_adjacent, list); |
9ff162a8 JP |
6087 | if (upper && likely(upper->master)) |
6088 | return upper->dev; | |
6089 | return NULL; | |
6090 | } | |
6091 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
6092 | ||
0a59f3a9 | 6093 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
6094 | struct net_device *adj_dev, |
6095 | struct list_head *dev_list) | |
6096 | { | |
6097 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6098 | |
3ee32707 VF |
6099 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6100 | "upper_%s" : "lower_%s", adj_dev->name); | |
6101 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
6102 | linkname); | |
6103 | } | |
0a59f3a9 | 6104 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
6105 | char *name, |
6106 | struct list_head *dev_list) | |
6107 | { | |
6108 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6109 | |
3ee32707 VF |
6110 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6111 | "upper_%s" : "lower_%s", name); | |
6112 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
6113 | } | |
6114 | ||
7ce64c79 AF |
6115 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
6116 | struct net_device *adj_dev, | |
6117 | struct list_head *dev_list) | |
6118 | { | |
6119 | return (dev_list == &dev->adj_list.upper || | |
6120 | dev_list == &dev->adj_list.lower) && | |
6121 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
6122 | } | |
3ee32707 | 6123 | |
5d261913 VF |
6124 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
6125 | struct net_device *adj_dev, | |
7863c054 | 6126 | struct list_head *dev_list, |
402dae96 | 6127 | void *private, bool master) |
5d261913 VF |
6128 | { |
6129 | struct netdev_adjacent *adj; | |
842d67a7 | 6130 | int ret; |
5d261913 | 6131 | |
6ea29da1 | 6132 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
6133 | |
6134 | if (adj) { | |
790510d9 | 6135 | adj->ref_nr += 1; |
67b62f98 DA |
6136 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
6137 | dev->name, adj_dev->name, adj->ref_nr); | |
6138 | ||
5d261913 VF |
6139 | return 0; |
6140 | } | |
6141 | ||
6142 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
6143 | if (!adj) | |
6144 | return -ENOMEM; | |
6145 | ||
6146 | adj->dev = adj_dev; | |
6147 | adj->master = master; | |
790510d9 | 6148 | adj->ref_nr = 1; |
402dae96 | 6149 | adj->private = private; |
5d261913 | 6150 | dev_hold(adj_dev); |
2f268f12 | 6151 | |
67b62f98 DA |
6152 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
6153 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 6154 | |
7ce64c79 | 6155 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 6156 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
6157 | if (ret) |
6158 | goto free_adj; | |
6159 | } | |
6160 | ||
7863c054 | 6161 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
6162 | if (master) { |
6163 | ret = sysfs_create_link(&(dev->dev.kobj), | |
6164 | &(adj_dev->dev.kobj), "master"); | |
6165 | if (ret) | |
5831d66e | 6166 | goto remove_symlinks; |
842d67a7 | 6167 | |
7863c054 | 6168 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 6169 | } else { |
7863c054 | 6170 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 6171 | } |
5d261913 VF |
6172 | |
6173 | return 0; | |
842d67a7 | 6174 | |
5831d66e | 6175 | remove_symlinks: |
7ce64c79 | 6176 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6177 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
6178 | free_adj: |
6179 | kfree(adj); | |
974daef7 | 6180 | dev_put(adj_dev); |
842d67a7 VF |
6181 | |
6182 | return ret; | |
5d261913 VF |
6183 | } |
6184 | ||
1d143d9f | 6185 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
6186 | struct net_device *adj_dev, | |
93409033 | 6187 | u16 ref_nr, |
1d143d9f | 6188 | struct list_head *dev_list) |
5d261913 VF |
6189 | { |
6190 | struct netdev_adjacent *adj; | |
6191 | ||
67b62f98 DA |
6192 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
6193 | dev->name, adj_dev->name, ref_nr); | |
6194 | ||
6ea29da1 | 6195 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 6196 | |
2f268f12 | 6197 | if (!adj) { |
67b62f98 | 6198 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 6199 | dev->name, adj_dev->name); |
67b62f98 DA |
6200 | WARN_ON(1); |
6201 | return; | |
2f268f12 | 6202 | } |
5d261913 | 6203 | |
93409033 | 6204 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
6205 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
6206 | dev->name, adj_dev->name, ref_nr, | |
6207 | adj->ref_nr - ref_nr); | |
93409033 | 6208 | adj->ref_nr -= ref_nr; |
5d261913 VF |
6209 | return; |
6210 | } | |
6211 | ||
842d67a7 VF |
6212 | if (adj->master) |
6213 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
6214 | ||
7ce64c79 | 6215 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6216 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 6217 | |
5d261913 | 6218 | list_del_rcu(&adj->list); |
67b62f98 | 6219 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 6220 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
6221 | dev_put(adj_dev); |
6222 | kfree_rcu(adj, rcu); | |
6223 | } | |
6224 | ||
1d143d9f | 6225 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
6226 | struct net_device *upper_dev, | |
6227 | struct list_head *up_list, | |
6228 | struct list_head *down_list, | |
6229 | void *private, bool master) | |
5d261913 VF |
6230 | { |
6231 | int ret; | |
6232 | ||
790510d9 | 6233 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 6234 | private, master); |
5d261913 VF |
6235 | if (ret) |
6236 | return ret; | |
6237 | ||
790510d9 | 6238 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 6239 | private, false); |
5d261913 | 6240 | if (ret) { |
790510d9 | 6241 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
6242 | return ret; |
6243 | } | |
6244 | ||
6245 | return 0; | |
6246 | } | |
6247 | ||
1d143d9f | 6248 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
6249 | struct net_device *upper_dev, | |
93409033 | 6250 | u16 ref_nr, |
1d143d9f | 6251 | struct list_head *up_list, |
6252 | struct list_head *down_list) | |
5d261913 | 6253 | { |
93409033 AC |
6254 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
6255 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
6256 | } |
6257 | ||
1d143d9f | 6258 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
6259 | struct net_device *upper_dev, | |
6260 | void *private, bool master) | |
2f268f12 | 6261 | { |
f1170fd4 DA |
6262 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
6263 | &dev->adj_list.upper, | |
6264 | &upper_dev->adj_list.lower, | |
6265 | private, master); | |
5d261913 VF |
6266 | } |
6267 | ||
1d143d9f | 6268 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
6269 | struct net_device *upper_dev) | |
2f268f12 | 6270 | { |
93409033 | 6271 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
6272 | &dev->adj_list.upper, |
6273 | &upper_dev->adj_list.lower); | |
6274 | } | |
5d261913 | 6275 | |
9ff162a8 | 6276 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 6277 | struct net_device *upper_dev, bool master, |
29bf24af | 6278 | void *upper_priv, void *upper_info) |
9ff162a8 | 6279 | { |
0e4ead9d | 6280 | struct netdev_notifier_changeupper_info changeupper_info; |
5d261913 | 6281 | int ret = 0; |
9ff162a8 JP |
6282 | |
6283 | ASSERT_RTNL(); | |
6284 | ||
6285 | if (dev == upper_dev) | |
6286 | return -EBUSY; | |
6287 | ||
6288 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
f1170fd4 | 6289 | if (netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
6290 | return -EBUSY; |
6291 | ||
f1170fd4 | 6292 | if (netdev_has_upper_dev(dev, upper_dev)) |
9ff162a8 JP |
6293 | return -EEXIST; |
6294 | ||
6295 | if (master && netdev_master_upper_dev_get(dev)) | |
6296 | return -EBUSY; | |
6297 | ||
0e4ead9d JP |
6298 | changeupper_info.upper_dev = upper_dev; |
6299 | changeupper_info.master = master; | |
6300 | changeupper_info.linking = true; | |
29bf24af | 6301 | changeupper_info.upper_info = upper_info; |
0e4ead9d | 6302 | |
573c7ba0 JP |
6303 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6304 | &changeupper_info.info); | |
6305 | ret = notifier_to_errno(ret); | |
6306 | if (ret) | |
6307 | return ret; | |
6308 | ||
6dffb044 | 6309 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 6310 | master); |
5d261913 VF |
6311 | if (ret) |
6312 | return ret; | |
9ff162a8 | 6313 | |
b03804e7 IS |
6314 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6315 | &changeupper_info.info); | |
6316 | ret = notifier_to_errno(ret); | |
6317 | if (ret) | |
f1170fd4 | 6318 | goto rollback; |
b03804e7 | 6319 | |
9ff162a8 | 6320 | return 0; |
5d261913 | 6321 | |
f1170fd4 | 6322 | rollback: |
2f268f12 | 6323 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
6324 | |
6325 | return ret; | |
9ff162a8 JP |
6326 | } |
6327 | ||
6328 | /** | |
6329 | * netdev_upper_dev_link - Add a link to the upper device | |
6330 | * @dev: device | |
6331 | * @upper_dev: new upper device | |
6332 | * | |
6333 | * Adds a link to device which is upper to this one. The caller must hold | |
6334 | * the RTNL lock. On a failure a negative errno code is returned. | |
6335 | * On success the reference counts are adjusted and the function | |
6336 | * returns zero. | |
6337 | */ | |
6338 | int netdev_upper_dev_link(struct net_device *dev, | |
6339 | struct net_device *upper_dev) | |
6340 | { | |
29bf24af | 6341 | return __netdev_upper_dev_link(dev, upper_dev, false, NULL, NULL); |
9ff162a8 JP |
6342 | } |
6343 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
6344 | ||
6345 | /** | |
6346 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
6347 | * @dev: device | |
6348 | * @upper_dev: new upper device | |
6dffb044 | 6349 | * @upper_priv: upper device private |
29bf24af | 6350 | * @upper_info: upper info to be passed down via notifier |
9ff162a8 JP |
6351 | * |
6352 | * Adds a link to device which is upper to this one. In this case, only | |
6353 | * one master upper device can be linked, although other non-master devices | |
6354 | * might be linked as well. The caller must hold the RTNL lock. | |
6355 | * On a failure a negative errno code is returned. On success the reference | |
6356 | * counts are adjusted and the function returns zero. | |
6357 | */ | |
6358 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 6359 | struct net_device *upper_dev, |
29bf24af | 6360 | void *upper_priv, void *upper_info) |
9ff162a8 | 6361 | { |
29bf24af JP |
6362 | return __netdev_upper_dev_link(dev, upper_dev, true, |
6363 | upper_priv, upper_info); | |
9ff162a8 JP |
6364 | } |
6365 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
6366 | ||
6367 | /** | |
6368 | * netdev_upper_dev_unlink - Removes a link to upper device | |
6369 | * @dev: device | |
6370 | * @upper_dev: new upper device | |
6371 | * | |
6372 | * Removes a link to device which is upper to this one. The caller must hold | |
6373 | * the RTNL lock. | |
6374 | */ | |
6375 | void netdev_upper_dev_unlink(struct net_device *dev, | |
6376 | struct net_device *upper_dev) | |
6377 | { | |
0e4ead9d | 6378 | struct netdev_notifier_changeupper_info changeupper_info; |
f4563a75 | 6379 | |
9ff162a8 JP |
6380 | ASSERT_RTNL(); |
6381 | ||
0e4ead9d JP |
6382 | changeupper_info.upper_dev = upper_dev; |
6383 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; | |
6384 | changeupper_info.linking = false; | |
6385 | ||
573c7ba0 JP |
6386 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, dev, |
6387 | &changeupper_info.info); | |
6388 | ||
2f268f12 | 6389 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 6390 | |
0e4ead9d JP |
6391 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, dev, |
6392 | &changeupper_info.info); | |
9ff162a8 JP |
6393 | } |
6394 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
6395 | ||
61bd3857 MS |
6396 | /** |
6397 | * netdev_bonding_info_change - Dispatch event about slave change | |
6398 | * @dev: device | |
4a26e453 | 6399 | * @bonding_info: info to dispatch |
61bd3857 MS |
6400 | * |
6401 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
6402 | * The caller must hold the RTNL lock. | |
6403 | */ | |
6404 | void netdev_bonding_info_change(struct net_device *dev, | |
6405 | struct netdev_bonding_info *bonding_info) | |
6406 | { | |
6407 | struct netdev_notifier_bonding_info info; | |
6408 | ||
6409 | memcpy(&info.bonding_info, bonding_info, | |
6410 | sizeof(struct netdev_bonding_info)); | |
6411 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, dev, | |
6412 | &info.info); | |
6413 | } | |
6414 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
6415 | ||
2ce1ee17 | 6416 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
6417 | { |
6418 | struct netdev_adjacent *iter; | |
6419 | ||
6420 | struct net *net = dev_net(dev); | |
6421 | ||
6422 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6423 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6424 | continue; |
6425 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6426 | &iter->dev->adj_list.lower); | |
6427 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6428 | &dev->adj_list.upper); | |
6429 | } | |
6430 | ||
6431 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6432 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6433 | continue; |
6434 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6435 | &iter->dev->adj_list.upper); | |
6436 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
6437 | &dev->adj_list.lower); | |
6438 | } | |
6439 | } | |
6440 | ||
2ce1ee17 | 6441 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
6442 | { |
6443 | struct netdev_adjacent *iter; | |
6444 | ||
6445 | struct net *net = dev_net(dev); | |
6446 | ||
6447 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 6448 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6449 | continue; |
6450 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6451 | &iter->dev->adj_list.lower); | |
6452 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6453 | &dev->adj_list.upper); | |
6454 | } | |
6455 | ||
6456 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 6457 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
6458 | continue; |
6459 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
6460 | &iter->dev->adj_list.upper); | |
6461 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
6462 | &dev->adj_list.lower); | |
6463 | } | |
6464 | } | |
6465 | ||
5bb025fa | 6466 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 6467 | { |
5bb025fa | 6468 | struct netdev_adjacent *iter; |
402dae96 | 6469 | |
4c75431a AF |
6470 | struct net *net = dev_net(dev); |
6471 | ||
5bb025fa | 6472 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 6473 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6474 | continue; |
5bb025fa VF |
6475 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6476 | &iter->dev->adj_list.lower); | |
6477 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6478 | &iter->dev->adj_list.lower); | |
6479 | } | |
402dae96 | 6480 | |
5bb025fa | 6481 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 6482 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 6483 | continue; |
5bb025fa VF |
6484 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
6485 | &iter->dev->adj_list.upper); | |
6486 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
6487 | &iter->dev->adj_list.upper); | |
6488 | } | |
402dae96 | 6489 | } |
402dae96 VF |
6490 | |
6491 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
6492 | struct net_device *lower_dev) | |
6493 | { | |
6494 | struct netdev_adjacent *lower; | |
6495 | ||
6496 | if (!lower_dev) | |
6497 | return NULL; | |
6ea29da1 | 6498 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
6499 | if (!lower) |
6500 | return NULL; | |
6501 | ||
6502 | return lower->private; | |
6503 | } | |
6504 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
6505 | ||
4085ebe8 | 6506 | |
952fcfd0 | 6507 | int dev_get_nest_level(struct net_device *dev) |
4085ebe8 VY |
6508 | { |
6509 | struct net_device *lower = NULL; | |
6510 | struct list_head *iter; | |
6511 | int max_nest = -1; | |
6512 | int nest; | |
6513 | ||
6514 | ASSERT_RTNL(); | |
6515 | ||
6516 | netdev_for_each_lower_dev(dev, lower, iter) { | |
952fcfd0 | 6517 | nest = dev_get_nest_level(lower); |
4085ebe8 VY |
6518 | if (max_nest < nest) |
6519 | max_nest = nest; | |
6520 | } | |
6521 | ||
952fcfd0 | 6522 | return max_nest + 1; |
4085ebe8 VY |
6523 | } |
6524 | EXPORT_SYMBOL(dev_get_nest_level); | |
6525 | ||
04d48266 JP |
6526 | /** |
6527 | * netdev_lower_change - Dispatch event about lower device state change | |
6528 | * @lower_dev: device | |
6529 | * @lower_state_info: state to dispatch | |
6530 | * | |
6531 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
6532 | * The caller must hold the RTNL lock. | |
6533 | */ | |
6534 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
6535 | void *lower_state_info) | |
6536 | { | |
6537 | struct netdev_notifier_changelowerstate_info changelowerstate_info; | |
6538 | ||
6539 | ASSERT_RTNL(); | |
6540 | changelowerstate_info.lower_state_info = lower_state_info; | |
6541 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, lower_dev, | |
6542 | &changelowerstate_info.info); | |
6543 | } | |
6544 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
6545 | ||
b6c40d68 PM |
6546 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
6547 | { | |
d314774c SH |
6548 | const struct net_device_ops *ops = dev->netdev_ops; |
6549 | ||
d2615bf4 | 6550 | if (ops->ndo_change_rx_flags) |
d314774c | 6551 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
6552 | } |
6553 | ||
991fb3f7 | 6554 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6555 | { |
b536db93 | 6556 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
6557 | kuid_t uid; |
6558 | kgid_t gid; | |
1da177e4 | 6559 | |
24023451 PM |
6560 | ASSERT_RTNL(); |
6561 | ||
dad9b335 WC |
6562 | dev->flags |= IFF_PROMISC; |
6563 | dev->promiscuity += inc; | |
6564 | if (dev->promiscuity == 0) { | |
6565 | /* | |
6566 | * Avoid overflow. | |
6567 | * If inc causes overflow, untouch promisc and return error. | |
6568 | */ | |
6569 | if (inc < 0) | |
6570 | dev->flags &= ~IFF_PROMISC; | |
6571 | else { | |
6572 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
6573 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
6574 | dev->name); | |
dad9b335 WC |
6575 | return -EOVERFLOW; |
6576 | } | |
6577 | } | |
52609c0b | 6578 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
6579 | pr_info("device %s %s promiscuous mode\n", |
6580 | dev->name, | |
6581 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
6582 | if (audit_enabled) { |
6583 | current_uid_gid(&uid, &gid); | |
7759db82 KHK |
6584 | audit_log(current->audit_context, GFP_ATOMIC, |
6585 | AUDIT_ANOM_PROMISCUOUS, | |
6586 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
6587 | dev->name, (dev->flags & IFF_PROMISC), | |
6588 | (old_flags & IFF_PROMISC), | |
e1760bd5 | 6589 | from_kuid(&init_user_ns, audit_get_loginuid(current)), |
d04a48b0 EB |
6590 | from_kuid(&init_user_ns, uid), |
6591 | from_kgid(&init_user_ns, gid), | |
7759db82 | 6592 | audit_get_sessionid(current)); |
8192b0c4 | 6593 | } |
24023451 | 6594 | |
b6c40d68 | 6595 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 6596 | } |
991fb3f7 ND |
6597 | if (notify) |
6598 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 6599 | return 0; |
1da177e4 LT |
6600 | } |
6601 | ||
4417da66 PM |
6602 | /** |
6603 | * dev_set_promiscuity - update promiscuity count on a device | |
6604 | * @dev: device | |
6605 | * @inc: modifier | |
6606 | * | |
6607 | * Add or remove promiscuity from a device. While the count in the device | |
6608 | * remains above zero the interface remains promiscuous. Once it hits zero | |
6609 | * the device reverts back to normal filtering operation. A negative inc | |
6610 | * value is used to drop promiscuity on the device. | |
dad9b335 | 6611 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 6612 | */ |
dad9b335 | 6613 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 6614 | { |
b536db93 | 6615 | unsigned int old_flags = dev->flags; |
dad9b335 | 6616 | int err; |
4417da66 | 6617 | |
991fb3f7 | 6618 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 6619 | if (err < 0) |
dad9b335 | 6620 | return err; |
4417da66 PM |
6621 | if (dev->flags != old_flags) |
6622 | dev_set_rx_mode(dev); | |
dad9b335 | 6623 | return err; |
4417da66 | 6624 | } |
d1b19dff | 6625 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 6626 | |
991fb3f7 | 6627 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 6628 | { |
991fb3f7 | 6629 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 6630 | |
24023451 PM |
6631 | ASSERT_RTNL(); |
6632 | ||
1da177e4 | 6633 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
6634 | dev->allmulti += inc; |
6635 | if (dev->allmulti == 0) { | |
6636 | /* | |
6637 | * Avoid overflow. | |
6638 | * If inc causes overflow, untouch allmulti and return error. | |
6639 | */ | |
6640 | if (inc < 0) | |
6641 | dev->flags &= ~IFF_ALLMULTI; | |
6642 | else { | |
6643 | dev->allmulti -= inc; | |
7b6cd1ce JP |
6644 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
6645 | dev->name); | |
dad9b335 WC |
6646 | return -EOVERFLOW; |
6647 | } | |
6648 | } | |
24023451 | 6649 | if (dev->flags ^ old_flags) { |
b6c40d68 | 6650 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 6651 | dev_set_rx_mode(dev); |
991fb3f7 ND |
6652 | if (notify) |
6653 | __dev_notify_flags(dev, old_flags, | |
6654 | dev->gflags ^ old_gflags); | |
24023451 | 6655 | } |
dad9b335 | 6656 | return 0; |
4417da66 | 6657 | } |
991fb3f7 ND |
6658 | |
6659 | /** | |
6660 | * dev_set_allmulti - update allmulti count on a device | |
6661 | * @dev: device | |
6662 | * @inc: modifier | |
6663 | * | |
6664 | * Add or remove reception of all multicast frames to a device. While the | |
6665 | * count in the device remains above zero the interface remains listening | |
6666 | * to all interfaces. Once it hits zero the device reverts back to normal | |
6667 | * filtering operation. A negative @inc value is used to drop the counter | |
6668 | * when releasing a resource needing all multicasts. | |
6669 | * Return 0 if successful or a negative errno code on error. | |
6670 | */ | |
6671 | ||
6672 | int dev_set_allmulti(struct net_device *dev, int inc) | |
6673 | { | |
6674 | return __dev_set_allmulti(dev, inc, true); | |
6675 | } | |
d1b19dff | 6676 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
6677 | |
6678 | /* | |
6679 | * Upload unicast and multicast address lists to device and | |
6680 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 6681 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
6682 | * are present. |
6683 | */ | |
6684 | void __dev_set_rx_mode(struct net_device *dev) | |
6685 | { | |
d314774c SH |
6686 | const struct net_device_ops *ops = dev->netdev_ops; |
6687 | ||
4417da66 PM |
6688 | /* dev_open will call this function so the list will stay sane. */ |
6689 | if (!(dev->flags&IFF_UP)) | |
6690 | return; | |
6691 | ||
6692 | if (!netif_device_present(dev)) | |
40b77c94 | 6693 | return; |
4417da66 | 6694 | |
01789349 | 6695 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
6696 | /* Unicast addresses changes may only happen under the rtnl, |
6697 | * therefore calling __dev_set_promiscuity here is safe. | |
6698 | */ | |
32e7bfc4 | 6699 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 6700 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 6701 | dev->uc_promisc = true; |
32e7bfc4 | 6702 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 6703 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 6704 | dev->uc_promisc = false; |
4417da66 | 6705 | } |
4417da66 | 6706 | } |
01789349 JP |
6707 | |
6708 | if (ops->ndo_set_rx_mode) | |
6709 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
6710 | } |
6711 | ||
6712 | void dev_set_rx_mode(struct net_device *dev) | |
6713 | { | |
b9e40857 | 6714 | netif_addr_lock_bh(dev); |
4417da66 | 6715 | __dev_set_rx_mode(dev); |
b9e40857 | 6716 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
6717 | } |
6718 | ||
f0db275a SH |
6719 | /** |
6720 | * dev_get_flags - get flags reported to userspace | |
6721 | * @dev: device | |
6722 | * | |
6723 | * Get the combination of flag bits exported through APIs to userspace. | |
6724 | */ | |
95c96174 | 6725 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 6726 | { |
95c96174 | 6727 | unsigned int flags; |
1da177e4 LT |
6728 | |
6729 | flags = (dev->flags & ~(IFF_PROMISC | | |
6730 | IFF_ALLMULTI | | |
b00055aa SR |
6731 | IFF_RUNNING | |
6732 | IFF_LOWER_UP | | |
6733 | IFF_DORMANT)) | | |
1da177e4 LT |
6734 | (dev->gflags & (IFF_PROMISC | |
6735 | IFF_ALLMULTI)); | |
6736 | ||
b00055aa SR |
6737 | if (netif_running(dev)) { |
6738 | if (netif_oper_up(dev)) | |
6739 | flags |= IFF_RUNNING; | |
6740 | if (netif_carrier_ok(dev)) | |
6741 | flags |= IFF_LOWER_UP; | |
6742 | if (netif_dormant(dev)) | |
6743 | flags |= IFF_DORMANT; | |
6744 | } | |
1da177e4 LT |
6745 | |
6746 | return flags; | |
6747 | } | |
d1b19dff | 6748 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 6749 | |
bd380811 | 6750 | int __dev_change_flags(struct net_device *dev, unsigned int flags) |
1da177e4 | 6751 | { |
b536db93 | 6752 | unsigned int old_flags = dev->flags; |
bd380811 | 6753 | int ret; |
1da177e4 | 6754 | |
24023451 PM |
6755 | ASSERT_RTNL(); |
6756 | ||
1da177e4 LT |
6757 | /* |
6758 | * Set the flags on our device. | |
6759 | */ | |
6760 | ||
6761 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
6762 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
6763 | IFF_AUTOMEDIA)) | | |
6764 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
6765 | IFF_ALLMULTI)); | |
6766 | ||
6767 | /* | |
6768 | * Load in the correct multicast list now the flags have changed. | |
6769 | */ | |
6770 | ||
b6c40d68 PM |
6771 | if ((old_flags ^ flags) & IFF_MULTICAST) |
6772 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 6773 | |
4417da66 | 6774 | dev_set_rx_mode(dev); |
1da177e4 LT |
6775 | |
6776 | /* | |
6777 | * Have we downed the interface. We handle IFF_UP ourselves | |
6778 | * according to user attempts to set it, rather than blindly | |
6779 | * setting it. | |
6780 | */ | |
6781 | ||
6782 | ret = 0; | |
7051b88a | 6783 | if ((old_flags ^ flags) & IFF_UP) { |
6784 | if (old_flags & IFF_UP) | |
6785 | __dev_close(dev); | |
6786 | else | |
6787 | ret = __dev_open(dev); | |
6788 | } | |
1da177e4 | 6789 | |
1da177e4 | 6790 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 6791 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 6792 | unsigned int old_flags = dev->flags; |
d1b19dff | 6793 | |
1da177e4 | 6794 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
6795 | |
6796 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
6797 | if (dev->flags != old_flags) | |
6798 | dev_set_rx_mode(dev); | |
1da177e4 LT |
6799 | } |
6800 | ||
6801 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 6802 | * is important. Some (broken) drivers set IFF_PROMISC, when |
6803 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
6804 | */ |
6805 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
6806 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
6807 | ||
1da177e4 | 6808 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 6809 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
6810 | } |
6811 | ||
bd380811 PM |
6812 | return ret; |
6813 | } | |
6814 | ||
a528c219 ND |
6815 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
6816 | unsigned int gchanges) | |
bd380811 PM |
6817 | { |
6818 | unsigned int changes = dev->flags ^ old_flags; | |
6819 | ||
a528c219 | 6820 | if (gchanges) |
7f294054 | 6821 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 6822 | |
bd380811 PM |
6823 | if (changes & IFF_UP) { |
6824 | if (dev->flags & IFF_UP) | |
6825 | call_netdevice_notifiers(NETDEV_UP, dev); | |
6826 | else | |
6827 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
6828 | } | |
6829 | ||
6830 | if (dev->flags & IFF_UP && | |
be9efd36 JP |
6831 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
6832 | struct netdev_notifier_change_info change_info; | |
6833 | ||
6834 | change_info.flags_changed = changes; | |
6835 | call_netdevice_notifiers_info(NETDEV_CHANGE, dev, | |
6836 | &change_info.info); | |
6837 | } | |
bd380811 PM |
6838 | } |
6839 | ||
6840 | /** | |
6841 | * dev_change_flags - change device settings | |
6842 | * @dev: device | |
6843 | * @flags: device state flags | |
6844 | * | |
6845 | * Change settings on device based state flags. The flags are | |
6846 | * in the userspace exported format. | |
6847 | */ | |
b536db93 | 6848 | int dev_change_flags(struct net_device *dev, unsigned int flags) |
bd380811 | 6849 | { |
b536db93 | 6850 | int ret; |
991fb3f7 | 6851 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 PM |
6852 | |
6853 | ret = __dev_change_flags(dev, flags); | |
6854 | if (ret < 0) | |
6855 | return ret; | |
6856 | ||
991fb3f7 | 6857 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 6858 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
6859 | return ret; |
6860 | } | |
d1b19dff | 6861 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 6862 | |
f51048c3 | 6863 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
6864 | { |
6865 | const struct net_device_ops *ops = dev->netdev_ops; | |
6866 | ||
6867 | if (ops->ndo_change_mtu) | |
6868 | return ops->ndo_change_mtu(dev, new_mtu); | |
6869 | ||
6870 | dev->mtu = new_mtu; | |
6871 | return 0; | |
6872 | } | |
f51048c3 | 6873 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 6874 | |
f0db275a SH |
6875 | /** |
6876 | * dev_set_mtu - Change maximum transfer unit | |
6877 | * @dev: device | |
6878 | * @new_mtu: new transfer unit | |
6879 | * | |
6880 | * Change the maximum transfer size of the network device. | |
6881 | */ | |
1da177e4 LT |
6882 | int dev_set_mtu(struct net_device *dev, int new_mtu) |
6883 | { | |
2315dc91 | 6884 | int err, orig_mtu; |
1da177e4 LT |
6885 | |
6886 | if (new_mtu == dev->mtu) | |
6887 | return 0; | |
6888 | ||
61e84623 JW |
6889 | /* MTU must be positive, and in range */ |
6890 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
6891 | net_err_ratelimited("%s: Invalid MTU %d requested, hw min %d\n", | |
6892 | dev->name, new_mtu, dev->min_mtu); | |
1da177e4 | 6893 | return -EINVAL; |
61e84623 JW |
6894 | } |
6895 | ||
6896 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
6897 | net_err_ratelimited("%s: Invalid MTU %d requested, hw max %d\n", | |
a0e65de7 | 6898 | dev->name, new_mtu, dev->max_mtu); |
61e84623 JW |
6899 | return -EINVAL; |
6900 | } | |
1da177e4 LT |
6901 | |
6902 | if (!netif_device_present(dev)) | |
6903 | return -ENODEV; | |
6904 | ||
1d486bfb VF |
6905 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
6906 | err = notifier_to_errno(err); | |
6907 | if (err) | |
6908 | return err; | |
d314774c | 6909 | |
2315dc91 VF |
6910 | orig_mtu = dev->mtu; |
6911 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 6912 | |
2315dc91 VF |
6913 | if (!err) { |
6914 | err = call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6915 | err = notifier_to_errno(err); | |
6916 | if (err) { | |
6917 | /* setting mtu back and notifying everyone again, | |
6918 | * so that they have a chance to revert changes. | |
6919 | */ | |
6920 | __dev_set_mtu(dev, orig_mtu); | |
6921 | call_netdevice_notifiers(NETDEV_CHANGEMTU, dev); | |
6922 | } | |
6923 | } | |
1da177e4 LT |
6924 | return err; |
6925 | } | |
d1b19dff | 6926 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 6927 | |
cbda10fa VD |
6928 | /** |
6929 | * dev_set_group - Change group this device belongs to | |
6930 | * @dev: device | |
6931 | * @new_group: group this device should belong to | |
6932 | */ | |
6933 | void dev_set_group(struct net_device *dev, int new_group) | |
6934 | { | |
6935 | dev->group = new_group; | |
6936 | } | |
6937 | EXPORT_SYMBOL(dev_set_group); | |
6938 | ||
f0db275a SH |
6939 | /** |
6940 | * dev_set_mac_address - Change Media Access Control Address | |
6941 | * @dev: device | |
6942 | * @sa: new address | |
6943 | * | |
6944 | * Change the hardware (MAC) address of the device | |
6945 | */ | |
1da177e4 LT |
6946 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa) |
6947 | { | |
d314774c | 6948 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
6949 | int err; |
6950 | ||
d314774c | 6951 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
6952 | return -EOPNOTSUPP; |
6953 | if (sa->sa_family != dev->type) | |
6954 | return -EINVAL; | |
6955 | if (!netif_device_present(dev)) | |
6956 | return -ENODEV; | |
d314774c | 6957 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
6958 | if (err) |
6959 | return err; | |
fbdeca2d | 6960 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 6961 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 6962 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 6963 | return 0; |
1da177e4 | 6964 | } |
d1b19dff | 6965 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 6966 | |
4bf84c35 JP |
6967 | /** |
6968 | * dev_change_carrier - Change device carrier | |
6969 | * @dev: device | |
691b3b7e | 6970 | * @new_carrier: new value |
4bf84c35 JP |
6971 | * |
6972 | * Change device carrier | |
6973 | */ | |
6974 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
6975 | { | |
6976 | const struct net_device_ops *ops = dev->netdev_ops; | |
6977 | ||
6978 | if (!ops->ndo_change_carrier) | |
6979 | return -EOPNOTSUPP; | |
6980 | if (!netif_device_present(dev)) | |
6981 | return -ENODEV; | |
6982 | return ops->ndo_change_carrier(dev, new_carrier); | |
6983 | } | |
6984 | EXPORT_SYMBOL(dev_change_carrier); | |
6985 | ||
66b52b0d JP |
6986 | /** |
6987 | * dev_get_phys_port_id - Get device physical port ID | |
6988 | * @dev: device | |
6989 | * @ppid: port ID | |
6990 | * | |
6991 | * Get device physical port ID | |
6992 | */ | |
6993 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 6994 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
6995 | { |
6996 | const struct net_device_ops *ops = dev->netdev_ops; | |
6997 | ||
6998 | if (!ops->ndo_get_phys_port_id) | |
6999 | return -EOPNOTSUPP; | |
7000 | return ops->ndo_get_phys_port_id(dev, ppid); | |
7001 | } | |
7002 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
7003 | ||
db24a904 DA |
7004 | /** |
7005 | * dev_get_phys_port_name - Get device physical port name | |
7006 | * @dev: device | |
7007 | * @name: port name | |
ed49e650 | 7008 | * @len: limit of bytes to copy to name |
db24a904 DA |
7009 | * |
7010 | * Get device physical port name | |
7011 | */ | |
7012 | int dev_get_phys_port_name(struct net_device *dev, | |
7013 | char *name, size_t len) | |
7014 | { | |
7015 | const struct net_device_ops *ops = dev->netdev_ops; | |
7016 | ||
7017 | if (!ops->ndo_get_phys_port_name) | |
7018 | return -EOPNOTSUPP; | |
7019 | return ops->ndo_get_phys_port_name(dev, name, len); | |
7020 | } | |
7021 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
7022 | ||
d746d707 AK |
7023 | /** |
7024 | * dev_change_proto_down - update protocol port state information | |
7025 | * @dev: device | |
7026 | * @proto_down: new value | |
7027 | * | |
7028 | * This info can be used by switch drivers to set the phys state of the | |
7029 | * port. | |
7030 | */ | |
7031 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
7032 | { | |
7033 | const struct net_device_ops *ops = dev->netdev_ops; | |
7034 | ||
7035 | if (!ops->ndo_change_proto_down) | |
7036 | return -EOPNOTSUPP; | |
7037 | if (!netif_device_present(dev)) | |
7038 | return -ENODEV; | |
7039 | return ops->ndo_change_proto_down(dev, proto_down); | |
7040 | } | |
7041 | EXPORT_SYMBOL(dev_change_proto_down); | |
7042 | ||
ce158e58 | 7043 | u8 __dev_xdp_attached(struct net_device *dev, xdp_op_t xdp_op, u32 *prog_id) |
d67b9cd2 DB |
7044 | { |
7045 | struct netdev_xdp xdp; | |
7046 | ||
7047 | memset(&xdp, 0, sizeof(xdp)); | |
7048 | xdp.command = XDP_QUERY_PROG; | |
7049 | ||
7050 | /* Query must always succeed. */ | |
7051 | WARN_ON(xdp_op(dev, &xdp) < 0); | |
58038695 MKL |
7052 | if (prog_id) |
7053 | *prog_id = xdp.prog_id; | |
7054 | ||
d67b9cd2 DB |
7055 | return xdp.prog_attached; |
7056 | } | |
7057 | ||
7058 | static int dev_xdp_install(struct net_device *dev, xdp_op_t xdp_op, | |
32d60277 | 7059 | struct netlink_ext_ack *extack, u32 flags, |
d67b9cd2 DB |
7060 | struct bpf_prog *prog) |
7061 | { | |
7062 | struct netdev_xdp xdp; | |
7063 | ||
7064 | memset(&xdp, 0, sizeof(xdp)); | |
ee5d032f JK |
7065 | if (flags & XDP_FLAGS_HW_MODE) |
7066 | xdp.command = XDP_SETUP_PROG_HW; | |
7067 | else | |
7068 | xdp.command = XDP_SETUP_PROG; | |
d67b9cd2 | 7069 | xdp.extack = extack; |
32d60277 | 7070 | xdp.flags = flags; |
d67b9cd2 DB |
7071 | xdp.prog = prog; |
7072 | ||
7073 | return xdp_op(dev, &xdp); | |
7074 | } | |
7075 | ||
a7862b45 BB |
7076 | /** |
7077 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
7078 | * @dev: device | |
b5d60989 | 7079 | * @extack: netlink extended ack |
a7862b45 | 7080 | * @fd: new program fd or negative value to clear |
85de8576 | 7081 | * @flags: xdp-related flags |
a7862b45 BB |
7082 | * |
7083 | * Set or clear a bpf program for a device | |
7084 | */ | |
ddf9f970 JK |
7085 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
7086 | int fd, u32 flags) | |
a7862b45 BB |
7087 | { |
7088 | const struct net_device_ops *ops = dev->netdev_ops; | |
7089 | struct bpf_prog *prog = NULL; | |
d67b9cd2 | 7090 | xdp_op_t xdp_op, xdp_chk; |
a7862b45 BB |
7091 | int err; |
7092 | ||
85de8576 DB |
7093 | ASSERT_RTNL(); |
7094 | ||
d67b9cd2 | 7095 | xdp_op = xdp_chk = ops->ndo_xdp; |
ee5d032f | 7096 | if (!xdp_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) |
0489df9a | 7097 | return -EOPNOTSUPP; |
b5cdae32 DM |
7098 | if (!xdp_op || (flags & XDP_FLAGS_SKB_MODE)) |
7099 | xdp_op = generic_xdp_install; | |
d67b9cd2 DB |
7100 | if (xdp_op == xdp_chk) |
7101 | xdp_chk = generic_xdp_install; | |
b5cdae32 | 7102 | |
a7862b45 | 7103 | if (fd >= 0) { |
58038695 | 7104 | if (xdp_chk && __dev_xdp_attached(dev, xdp_chk, NULL)) |
d67b9cd2 DB |
7105 | return -EEXIST; |
7106 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && | |
58038695 | 7107 | __dev_xdp_attached(dev, xdp_op, NULL)) |
d67b9cd2 | 7108 | return -EBUSY; |
85de8576 | 7109 | |
a7862b45 BB |
7110 | prog = bpf_prog_get_type(fd, BPF_PROG_TYPE_XDP); |
7111 | if (IS_ERR(prog)) | |
7112 | return PTR_ERR(prog); | |
7113 | } | |
7114 | ||
32d60277 | 7115 | err = dev_xdp_install(dev, xdp_op, extack, flags, prog); |
a7862b45 BB |
7116 | if (err < 0 && prog) |
7117 | bpf_prog_put(prog); | |
7118 | ||
7119 | return err; | |
7120 | } | |
a7862b45 | 7121 | |
1da177e4 LT |
7122 | /** |
7123 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 7124 | * @net: the applicable net namespace |
1da177e4 LT |
7125 | * |
7126 | * Returns a suitable unique value for a new device interface | |
7127 | * number. The caller must hold the rtnl semaphore or the | |
7128 | * dev_base_lock to be sure it remains unique. | |
7129 | */ | |
881d966b | 7130 | static int dev_new_index(struct net *net) |
1da177e4 | 7131 | { |
aa79e66e | 7132 | int ifindex = net->ifindex; |
f4563a75 | 7133 | |
1da177e4 LT |
7134 | for (;;) { |
7135 | if (++ifindex <= 0) | |
7136 | ifindex = 1; | |
881d966b | 7137 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 7138 | return net->ifindex = ifindex; |
1da177e4 LT |
7139 | } |
7140 | } | |
7141 | ||
1da177e4 | 7142 | /* Delayed registration/unregisteration */ |
3b5b34fd | 7143 | static LIST_HEAD(net_todo_list); |
200b916f | 7144 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 7145 | |
6f05f629 | 7146 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 7147 | { |
1da177e4 | 7148 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 7149 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
7150 | } |
7151 | ||
9b5e383c | 7152 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 7153 | { |
e93737b0 | 7154 | struct net_device *dev, *tmp; |
5cde2829 | 7155 | LIST_HEAD(close_head); |
9b5e383c | 7156 | |
93ee31f1 DL |
7157 | BUG_ON(dev_boot_phase); |
7158 | ASSERT_RTNL(); | |
7159 | ||
e93737b0 | 7160 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 7161 | /* Some devices call without registering |
e93737b0 KK |
7162 | * for initialization unwind. Remove those |
7163 | * devices and proceed with the remaining. | |
9b5e383c ED |
7164 | */ |
7165 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
7166 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
7167 | dev->name, dev); | |
93ee31f1 | 7168 | |
9b5e383c | 7169 | WARN_ON(1); |
e93737b0 KK |
7170 | list_del(&dev->unreg_list); |
7171 | continue; | |
9b5e383c | 7172 | } |
449f4544 | 7173 | dev->dismantle = true; |
9b5e383c | 7174 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 7175 | } |
93ee31f1 | 7176 | |
44345724 | 7177 | /* If device is running, close it first. */ |
5cde2829 EB |
7178 | list_for_each_entry(dev, head, unreg_list) |
7179 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 7180 | dev_close_many(&close_head, true); |
93ee31f1 | 7181 | |
44345724 | 7182 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
7183 | /* And unlink it from device chain. */ |
7184 | unlist_netdevice(dev); | |
93ee31f1 | 7185 | |
9b5e383c ED |
7186 | dev->reg_state = NETREG_UNREGISTERING; |
7187 | } | |
41852497 | 7188 | flush_all_backlogs(); |
93ee31f1 DL |
7189 | |
7190 | synchronize_net(); | |
7191 | ||
9b5e383c | 7192 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
7193 | struct sk_buff *skb = NULL; |
7194 | ||
9b5e383c ED |
7195 | /* Shutdown queueing discipline. */ |
7196 | dev_shutdown(dev); | |
93ee31f1 DL |
7197 | |
7198 | ||
9b5e383c | 7199 | /* Notify protocols, that we are about to destroy |
eb13da1a | 7200 | * this device. They should clean all the things. |
7201 | */ | |
9b5e383c | 7202 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 7203 | |
395eea6c MB |
7204 | if (!dev->rtnl_link_ops || |
7205 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
3d3ea5af | 7206 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, |
395eea6c MB |
7207 | GFP_KERNEL); |
7208 | ||
9b5e383c ED |
7209 | /* |
7210 | * Flush the unicast and multicast chains | |
7211 | */ | |
a748ee24 | 7212 | dev_uc_flush(dev); |
22bedad3 | 7213 | dev_mc_flush(dev); |
93ee31f1 | 7214 | |
9b5e383c ED |
7215 | if (dev->netdev_ops->ndo_uninit) |
7216 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 7217 | |
395eea6c MB |
7218 | if (skb) |
7219 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 7220 | |
9ff162a8 JP |
7221 | /* Notifier chain MUST detach us all upper devices. */ |
7222 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 7223 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 7224 | |
9b5e383c ED |
7225 | /* Remove entries from kobject tree */ |
7226 | netdev_unregister_kobject(dev); | |
024e9679 AD |
7227 | #ifdef CONFIG_XPS |
7228 | /* Remove XPS queueing entries */ | |
7229 | netif_reset_xps_queues_gt(dev, 0); | |
7230 | #endif | |
9b5e383c | 7231 | } |
93ee31f1 | 7232 | |
850a545b | 7233 | synchronize_net(); |
395264d5 | 7234 | |
a5ee1551 | 7235 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
7236 | dev_put(dev); |
7237 | } | |
7238 | ||
7239 | static void rollback_registered(struct net_device *dev) | |
7240 | { | |
7241 | LIST_HEAD(single); | |
7242 | ||
7243 | list_add(&dev->unreg_list, &single); | |
7244 | rollback_registered_many(&single); | |
ceaaec98 | 7245 | list_del(&single); |
93ee31f1 DL |
7246 | } |
7247 | ||
fd867d51 JW |
7248 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
7249 | struct net_device *upper, netdev_features_t features) | |
7250 | { | |
7251 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7252 | netdev_features_t feature; | |
5ba3f7d6 | 7253 | int feature_bit; |
fd867d51 | 7254 | |
5ba3f7d6 JW |
7255 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7256 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7257 | if (!(upper->wanted_features & feature) |
7258 | && (features & feature)) { | |
7259 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
7260 | &feature, upper->name); | |
7261 | features &= ~feature; | |
7262 | } | |
7263 | } | |
7264 | ||
7265 | return features; | |
7266 | } | |
7267 | ||
7268 | static void netdev_sync_lower_features(struct net_device *upper, | |
7269 | struct net_device *lower, netdev_features_t features) | |
7270 | { | |
7271 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
7272 | netdev_features_t feature; | |
5ba3f7d6 | 7273 | int feature_bit; |
fd867d51 | 7274 | |
5ba3f7d6 JW |
7275 | for_each_netdev_feature(&upper_disables, feature_bit) { |
7276 | feature = __NETIF_F_BIT(feature_bit); | |
fd867d51 JW |
7277 | if (!(features & feature) && (lower->features & feature)) { |
7278 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
7279 | &feature, lower->name); | |
7280 | lower->wanted_features &= ~feature; | |
7281 | netdev_update_features(lower); | |
7282 | ||
7283 | if (unlikely(lower->features & feature)) | |
7284 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
7285 | &feature, lower->name); | |
7286 | } | |
7287 | } | |
7288 | } | |
7289 | ||
c8f44aff MM |
7290 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
7291 | netdev_features_t features) | |
b63365a2 | 7292 | { |
57422dc5 MM |
7293 | /* Fix illegal checksum combinations */ |
7294 | if ((features & NETIF_F_HW_CSUM) && | |
7295 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 7296 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
7297 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
7298 | } | |
7299 | ||
b63365a2 | 7300 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 7301 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 7302 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 7303 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
7304 | } |
7305 | ||
ec5f0615 PS |
7306 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
7307 | !(features & NETIF_F_IP_CSUM)) { | |
7308 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
7309 | features &= ~NETIF_F_TSO; | |
7310 | features &= ~NETIF_F_TSO_ECN; | |
7311 | } | |
7312 | ||
7313 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
7314 | !(features & NETIF_F_IPV6_CSUM)) { | |
7315 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
7316 | features &= ~NETIF_F_TSO6; | |
7317 | } | |
7318 | ||
b1dc497b AD |
7319 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
7320 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
7321 | features &= ~NETIF_F_TSO_MANGLEID; | |
7322 | ||
31d8b9e0 BH |
7323 | /* TSO ECN requires that TSO is present as well. */ |
7324 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
7325 | features &= ~NETIF_F_TSO_ECN; | |
7326 | ||
212b573f MM |
7327 | /* Software GSO depends on SG. */ |
7328 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 7329 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
7330 | features &= ~NETIF_F_GSO; |
7331 | } | |
7332 | ||
802ab55a AD |
7333 | /* GSO partial features require GSO partial be set */ |
7334 | if ((features & dev->gso_partial_features) && | |
7335 | !(features & NETIF_F_GSO_PARTIAL)) { | |
7336 | netdev_dbg(dev, | |
7337 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
7338 | features &= ~dev->gso_partial_features; | |
7339 | } | |
7340 | ||
b63365a2 HX |
7341 | return features; |
7342 | } | |
b63365a2 | 7343 | |
6cb6a27c | 7344 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 7345 | { |
fd867d51 | 7346 | struct net_device *upper, *lower; |
c8f44aff | 7347 | netdev_features_t features; |
fd867d51 | 7348 | struct list_head *iter; |
e7868a85 | 7349 | int err = -1; |
5455c699 | 7350 | |
87267485 MM |
7351 | ASSERT_RTNL(); |
7352 | ||
5455c699 MM |
7353 | features = netdev_get_wanted_features(dev); |
7354 | ||
7355 | if (dev->netdev_ops->ndo_fix_features) | |
7356 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
7357 | ||
7358 | /* driver might be less strict about feature dependencies */ | |
7359 | features = netdev_fix_features(dev, features); | |
7360 | ||
fd867d51 JW |
7361 | /* some features can't be enabled if they're off an an upper device */ |
7362 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
7363 | features = netdev_sync_upper_features(dev, upper, features); | |
7364 | ||
5455c699 | 7365 | if (dev->features == features) |
e7868a85 | 7366 | goto sync_lower; |
5455c699 | 7367 | |
c8f44aff MM |
7368 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
7369 | &dev->features, &features); | |
5455c699 MM |
7370 | |
7371 | if (dev->netdev_ops->ndo_set_features) | |
7372 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
7373 | else |
7374 | err = 0; | |
5455c699 | 7375 | |
6cb6a27c | 7376 | if (unlikely(err < 0)) { |
5455c699 | 7377 | netdev_err(dev, |
c8f44aff MM |
7378 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
7379 | err, &features, &dev->features); | |
17b85d29 NA |
7380 | /* return non-0 since some features might have changed and |
7381 | * it's better to fire a spurious notification than miss it | |
7382 | */ | |
7383 | return -1; | |
6cb6a27c MM |
7384 | } |
7385 | ||
e7868a85 | 7386 | sync_lower: |
fd867d51 JW |
7387 | /* some features must be disabled on lower devices when disabled |
7388 | * on an upper device (think: bonding master or bridge) | |
7389 | */ | |
7390 | netdev_for_each_lower_dev(dev, lower, iter) | |
7391 | netdev_sync_lower_features(dev, lower, features); | |
7392 | ||
ae847f40 SD |
7393 | if (!err) { |
7394 | netdev_features_t diff = features ^ dev->features; | |
7395 | ||
7396 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
7397 | /* udp_tunnel_{get,drop}_rx_info both need | |
7398 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
7399 | * device, or they won't do anything. | |
7400 | * Thus we need to update dev->features | |
7401 | * *before* calling udp_tunnel_get_rx_info, | |
7402 | * but *after* calling udp_tunnel_drop_rx_info. | |
7403 | */ | |
7404 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
7405 | dev->features = features; | |
7406 | udp_tunnel_get_rx_info(dev); | |
7407 | } else { | |
7408 | udp_tunnel_drop_rx_info(dev); | |
7409 | } | |
7410 | } | |
7411 | ||
6cb6a27c | 7412 | dev->features = features; |
ae847f40 | 7413 | } |
6cb6a27c | 7414 | |
e7868a85 | 7415 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
7416 | } |
7417 | ||
afe12cc8 MM |
7418 | /** |
7419 | * netdev_update_features - recalculate device features | |
7420 | * @dev: the device to check | |
7421 | * | |
7422 | * Recalculate dev->features set and send notifications if it | |
7423 | * has changed. Should be called after driver or hardware dependent | |
7424 | * conditions might have changed that influence the features. | |
7425 | */ | |
6cb6a27c MM |
7426 | void netdev_update_features(struct net_device *dev) |
7427 | { | |
7428 | if (__netdev_update_features(dev)) | |
7429 | netdev_features_change(dev); | |
5455c699 MM |
7430 | } |
7431 | EXPORT_SYMBOL(netdev_update_features); | |
7432 | ||
afe12cc8 MM |
7433 | /** |
7434 | * netdev_change_features - recalculate device features | |
7435 | * @dev: the device to check | |
7436 | * | |
7437 | * Recalculate dev->features set and send notifications even | |
7438 | * if they have not changed. Should be called instead of | |
7439 | * netdev_update_features() if also dev->vlan_features might | |
7440 | * have changed to allow the changes to be propagated to stacked | |
7441 | * VLAN devices. | |
7442 | */ | |
7443 | void netdev_change_features(struct net_device *dev) | |
7444 | { | |
7445 | __netdev_update_features(dev); | |
7446 | netdev_features_change(dev); | |
7447 | } | |
7448 | EXPORT_SYMBOL(netdev_change_features); | |
7449 | ||
fc4a7489 PM |
7450 | /** |
7451 | * netif_stacked_transfer_operstate - transfer operstate | |
7452 | * @rootdev: the root or lower level device to transfer state from | |
7453 | * @dev: the device to transfer operstate to | |
7454 | * | |
7455 | * Transfer operational state from root to device. This is normally | |
7456 | * called when a stacking relationship exists between the root | |
7457 | * device and the device(a leaf device). | |
7458 | */ | |
7459 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
7460 | struct net_device *dev) | |
7461 | { | |
7462 | if (rootdev->operstate == IF_OPER_DORMANT) | |
7463 | netif_dormant_on(dev); | |
7464 | else | |
7465 | netif_dormant_off(dev); | |
7466 | ||
0575c86b ZS |
7467 | if (netif_carrier_ok(rootdev)) |
7468 | netif_carrier_on(dev); | |
7469 | else | |
7470 | netif_carrier_off(dev); | |
fc4a7489 PM |
7471 | } |
7472 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
7473 | ||
a953be53 | 7474 | #ifdef CONFIG_SYSFS |
1b4bf461 ED |
7475 | static int netif_alloc_rx_queues(struct net_device *dev) |
7476 | { | |
1b4bf461 | 7477 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 7478 | struct netdev_rx_queue *rx; |
10595902 | 7479 | size_t sz = count * sizeof(*rx); |
1b4bf461 | 7480 | |
bd25fa7b | 7481 | BUG_ON(count < 1); |
1b4bf461 | 7482 | |
dcda9b04 | 7483 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
7484 | if (!rx) |
7485 | return -ENOMEM; | |
7486 | ||
bd25fa7b TH |
7487 | dev->_rx = rx; |
7488 | ||
bd25fa7b | 7489 | for (i = 0; i < count; i++) |
fe822240 | 7490 | rx[i].dev = dev; |
1b4bf461 ED |
7491 | return 0; |
7492 | } | |
bf264145 | 7493 | #endif |
1b4bf461 | 7494 | |
aa942104 CG |
7495 | static void netdev_init_one_queue(struct net_device *dev, |
7496 | struct netdev_queue *queue, void *_unused) | |
7497 | { | |
7498 | /* Initialize queue lock */ | |
7499 | spin_lock_init(&queue->_xmit_lock); | |
7500 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
7501 | queue->xmit_lock_owner = -1; | |
b236da69 | 7502 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 7503 | queue->dev = dev; |
114cf580 TH |
7504 | #ifdef CONFIG_BQL |
7505 | dql_init(&queue->dql, HZ); | |
7506 | #endif | |
aa942104 CG |
7507 | } |
7508 | ||
60877a32 ED |
7509 | static void netif_free_tx_queues(struct net_device *dev) |
7510 | { | |
4cb28970 | 7511 | kvfree(dev->_tx); |
60877a32 ED |
7512 | } |
7513 | ||
e6484930 TH |
7514 | static int netif_alloc_netdev_queues(struct net_device *dev) |
7515 | { | |
7516 | unsigned int count = dev->num_tx_queues; | |
7517 | struct netdev_queue *tx; | |
60877a32 | 7518 | size_t sz = count * sizeof(*tx); |
e6484930 | 7519 | |
d339727c ED |
7520 | if (count < 1 || count > 0xffff) |
7521 | return -EINVAL; | |
62b5942a | 7522 | |
dcda9b04 | 7523 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
7524 | if (!tx) |
7525 | return -ENOMEM; | |
7526 | ||
e6484930 | 7527 | dev->_tx = tx; |
1d24eb48 | 7528 | |
e6484930 TH |
7529 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
7530 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
7531 | |
7532 | return 0; | |
e6484930 TH |
7533 | } |
7534 | ||
a2029240 DV |
7535 | void netif_tx_stop_all_queues(struct net_device *dev) |
7536 | { | |
7537 | unsigned int i; | |
7538 | ||
7539 | for (i = 0; i < dev->num_tx_queues; i++) { | |
7540 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 7541 | |
a2029240 DV |
7542 | netif_tx_stop_queue(txq); |
7543 | } | |
7544 | } | |
7545 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
7546 | ||
1da177e4 LT |
7547 | /** |
7548 | * register_netdevice - register a network device | |
7549 | * @dev: device to register | |
7550 | * | |
7551 | * Take a completed network device structure and add it to the kernel | |
7552 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7553 | * chain. 0 is returned on success. A negative errno code is returned | |
7554 | * on a failure to set up the device, or if the name is a duplicate. | |
7555 | * | |
7556 | * Callers must hold the rtnl semaphore. You may want | |
7557 | * register_netdev() instead of this. | |
7558 | * | |
7559 | * BUGS: | |
7560 | * The locking appears insufficient to guarantee two parallel registers | |
7561 | * will not get the same name. | |
7562 | */ | |
7563 | ||
7564 | int register_netdevice(struct net_device *dev) | |
7565 | { | |
1da177e4 | 7566 | int ret; |
d314774c | 7567 | struct net *net = dev_net(dev); |
1da177e4 LT |
7568 | |
7569 | BUG_ON(dev_boot_phase); | |
7570 | ASSERT_RTNL(); | |
7571 | ||
b17a7c17 SH |
7572 | might_sleep(); |
7573 | ||
1da177e4 LT |
7574 | /* When net_device's are persistent, this will be fatal. */ |
7575 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 7576 | BUG_ON(!net); |
1da177e4 | 7577 | |
f1f28aa3 | 7578 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 7579 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 7580 | |
828de4f6 | 7581 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
7582 | if (ret < 0) |
7583 | goto out; | |
7584 | ||
1da177e4 | 7585 | /* Init, if this function is available */ |
d314774c SH |
7586 | if (dev->netdev_ops->ndo_init) { |
7587 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
7588 | if (ret) { |
7589 | if (ret > 0) | |
7590 | ret = -EIO; | |
90833aa4 | 7591 | goto out; |
1da177e4 LT |
7592 | } |
7593 | } | |
4ec93edb | 7594 | |
f646968f PM |
7595 | if (((dev->hw_features | dev->features) & |
7596 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
7597 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
7598 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
7599 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
7600 | ret = -EINVAL; | |
7601 | goto err_uninit; | |
7602 | } | |
7603 | ||
9c7dafbf PE |
7604 | ret = -EBUSY; |
7605 | if (!dev->ifindex) | |
7606 | dev->ifindex = dev_new_index(net); | |
7607 | else if (__dev_get_by_index(net, dev->ifindex)) | |
7608 | goto err_uninit; | |
7609 | ||
5455c699 MM |
7610 | /* Transfer changeable features to wanted_features and enable |
7611 | * software offloads (GSO and GRO). | |
7612 | */ | |
7613 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f | 7614 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 SD |
7615 | |
7616 | if (dev->netdev_ops->ndo_udp_tunnel_add) { | |
7617 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
7618 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
7619 | } | |
7620 | ||
14d1232f | 7621 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 7622 | |
cbc53e08 | 7623 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 7624 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 7625 | |
7f348a60 AD |
7626 | /* If IPv4 TCP segmentation offload is supported we should also |
7627 | * allow the device to enable segmenting the frame with the option | |
7628 | * of ignoring a static IP ID value. This doesn't enable the | |
7629 | * feature itself but allows the user to enable it later. | |
7630 | */ | |
cbc53e08 AD |
7631 | if (dev->hw_features & NETIF_F_TSO) |
7632 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
7633 | if (dev->vlan_features & NETIF_F_TSO) |
7634 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
7635 | if (dev->mpls_features & NETIF_F_TSO) | |
7636 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
7637 | if (dev->hw_enc_features & NETIF_F_TSO) | |
7638 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 7639 | |
1180e7d6 | 7640 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 7641 | */ |
1180e7d6 | 7642 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 7643 | |
ee579677 PS |
7644 | /* Make NETIF_F_SG inheritable to tunnel devices. |
7645 | */ | |
802ab55a | 7646 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 7647 | |
0d89d203 SH |
7648 | /* Make NETIF_F_SG inheritable to MPLS. |
7649 | */ | |
7650 | dev->mpls_features |= NETIF_F_SG; | |
7651 | ||
7ffbe3fd JB |
7652 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
7653 | ret = notifier_to_errno(ret); | |
7654 | if (ret) | |
7655 | goto err_uninit; | |
7656 | ||
8b41d188 | 7657 | ret = netdev_register_kobject(dev); |
b17a7c17 | 7658 | if (ret) |
7ce1b0ed | 7659 | goto err_uninit; |
b17a7c17 SH |
7660 | dev->reg_state = NETREG_REGISTERED; |
7661 | ||
6cb6a27c | 7662 | __netdev_update_features(dev); |
8e9b59b2 | 7663 | |
1da177e4 LT |
7664 | /* |
7665 | * Default initial state at registry is that the | |
7666 | * device is present. | |
7667 | */ | |
7668 | ||
7669 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7670 | ||
8f4cccbb BH |
7671 | linkwatch_init_dev(dev); |
7672 | ||
1da177e4 | 7673 | dev_init_scheduler(dev); |
1da177e4 | 7674 | dev_hold(dev); |
ce286d32 | 7675 | list_netdevice(dev); |
7bf23575 | 7676 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 7677 | |
948b337e JP |
7678 | /* If the device has permanent device address, driver should |
7679 | * set dev_addr and also addr_assign_type should be set to | |
7680 | * NET_ADDR_PERM (default value). | |
7681 | */ | |
7682 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
7683 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
7684 | ||
1da177e4 | 7685 | /* Notify protocols, that a new device appeared. */ |
056925ab | 7686 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 7687 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
7688 | if (ret) { |
7689 | rollback_registered(dev); | |
7690 | dev->reg_state = NETREG_UNREGISTERED; | |
7691 | } | |
d90a909e EB |
7692 | /* |
7693 | * Prevent userspace races by waiting until the network | |
7694 | * device is fully setup before sending notifications. | |
7695 | */ | |
a2835763 PM |
7696 | if (!dev->rtnl_link_ops || |
7697 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 7698 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
7699 | |
7700 | out: | |
7701 | return ret; | |
7ce1b0ed HX |
7702 | |
7703 | err_uninit: | |
d314774c SH |
7704 | if (dev->netdev_ops->ndo_uninit) |
7705 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
7706 | if (dev->priv_destructor) |
7707 | dev->priv_destructor(dev); | |
7ce1b0ed | 7708 | goto out; |
1da177e4 | 7709 | } |
d1b19dff | 7710 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 7711 | |
937f1ba5 BH |
7712 | /** |
7713 | * init_dummy_netdev - init a dummy network device for NAPI | |
7714 | * @dev: device to init | |
7715 | * | |
7716 | * This takes a network device structure and initialize the minimum | |
7717 | * amount of fields so it can be used to schedule NAPI polls without | |
7718 | * registering a full blown interface. This is to be used by drivers | |
7719 | * that need to tie several hardware interfaces to a single NAPI | |
7720 | * poll scheduler due to HW limitations. | |
7721 | */ | |
7722 | int init_dummy_netdev(struct net_device *dev) | |
7723 | { | |
7724 | /* Clear everything. Note we don't initialize spinlocks | |
7725 | * are they aren't supposed to be taken by any of the | |
7726 | * NAPI code and this dummy netdev is supposed to be | |
7727 | * only ever used for NAPI polls | |
7728 | */ | |
7729 | memset(dev, 0, sizeof(struct net_device)); | |
7730 | ||
7731 | /* make sure we BUG if trying to hit standard | |
7732 | * register/unregister code path | |
7733 | */ | |
7734 | dev->reg_state = NETREG_DUMMY; | |
7735 | ||
937f1ba5 BH |
7736 | /* NAPI wants this */ |
7737 | INIT_LIST_HEAD(&dev->napi_list); | |
7738 | ||
7739 | /* a dummy interface is started by default */ | |
7740 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
7741 | set_bit(__LINK_STATE_START, &dev->state); | |
7742 | ||
29b4433d ED |
7743 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
7744 | * because users of this 'device' dont need to change | |
7745 | * its refcount. | |
7746 | */ | |
7747 | ||
937f1ba5 BH |
7748 | return 0; |
7749 | } | |
7750 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
7751 | ||
7752 | ||
1da177e4 LT |
7753 | /** |
7754 | * register_netdev - register a network device | |
7755 | * @dev: device to register | |
7756 | * | |
7757 | * Take a completed network device structure and add it to the kernel | |
7758 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
7759 | * chain. 0 is returned on success. A negative errno code is returned | |
7760 | * on a failure to set up the device, or if the name is a duplicate. | |
7761 | * | |
38b4da38 | 7762 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
7763 | * and expands the device name if you passed a format string to |
7764 | * alloc_netdev. | |
7765 | */ | |
7766 | int register_netdev(struct net_device *dev) | |
7767 | { | |
7768 | int err; | |
7769 | ||
7770 | rtnl_lock(); | |
1da177e4 | 7771 | err = register_netdevice(dev); |
1da177e4 LT |
7772 | rtnl_unlock(); |
7773 | return err; | |
7774 | } | |
7775 | EXPORT_SYMBOL(register_netdev); | |
7776 | ||
29b4433d ED |
7777 | int netdev_refcnt_read(const struct net_device *dev) |
7778 | { | |
7779 | int i, refcnt = 0; | |
7780 | ||
7781 | for_each_possible_cpu(i) | |
7782 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
7783 | return refcnt; | |
7784 | } | |
7785 | EXPORT_SYMBOL(netdev_refcnt_read); | |
7786 | ||
2c53040f | 7787 | /** |
1da177e4 | 7788 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 7789 | * @dev: target net_device |
1da177e4 LT |
7790 | * |
7791 | * This is called when unregistering network devices. | |
7792 | * | |
7793 | * Any protocol or device that holds a reference should register | |
7794 | * for netdevice notification, and cleanup and put back the | |
7795 | * reference if they receive an UNREGISTER event. | |
7796 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 7797 | * call dev_put. |
1da177e4 LT |
7798 | */ |
7799 | static void netdev_wait_allrefs(struct net_device *dev) | |
7800 | { | |
7801 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 7802 | int refcnt; |
1da177e4 | 7803 | |
e014debe ED |
7804 | linkwatch_forget_dev(dev); |
7805 | ||
1da177e4 | 7806 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
7807 | refcnt = netdev_refcnt_read(dev); |
7808 | ||
7809 | while (refcnt != 0) { | |
1da177e4 | 7810 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 7811 | rtnl_lock(); |
1da177e4 LT |
7812 | |
7813 | /* Rebroadcast unregister notification */ | |
056925ab | 7814 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 7815 | |
748e2d93 | 7816 | __rtnl_unlock(); |
0115e8e3 | 7817 | rcu_barrier(); |
748e2d93 ED |
7818 | rtnl_lock(); |
7819 | ||
0115e8e3 | 7820 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
1da177e4 LT |
7821 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
7822 | &dev->state)) { | |
7823 | /* We must not have linkwatch events | |
7824 | * pending on unregister. If this | |
7825 | * happens, we simply run the queue | |
7826 | * unscheduled, resulting in a noop | |
7827 | * for this device. | |
7828 | */ | |
7829 | linkwatch_run_queue(); | |
7830 | } | |
7831 | ||
6756ae4b | 7832 | __rtnl_unlock(); |
1da177e4 LT |
7833 | |
7834 | rebroadcast_time = jiffies; | |
7835 | } | |
7836 | ||
7837 | msleep(250); | |
7838 | ||
29b4433d ED |
7839 | refcnt = netdev_refcnt_read(dev); |
7840 | ||
1da177e4 | 7841 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
7842 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
7843 | dev->name, refcnt); | |
1da177e4 LT |
7844 | warning_time = jiffies; |
7845 | } | |
7846 | } | |
7847 | } | |
7848 | ||
7849 | /* The sequence is: | |
7850 | * | |
7851 | * rtnl_lock(); | |
7852 | * ... | |
7853 | * register_netdevice(x1); | |
7854 | * register_netdevice(x2); | |
7855 | * ... | |
7856 | * unregister_netdevice(y1); | |
7857 | * unregister_netdevice(y2); | |
7858 | * ... | |
7859 | * rtnl_unlock(); | |
7860 | * free_netdev(y1); | |
7861 | * free_netdev(y2); | |
7862 | * | |
58ec3b4d | 7863 | * We are invoked by rtnl_unlock(). |
1da177e4 | 7864 | * This allows us to deal with problems: |
b17a7c17 | 7865 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
7866 | * without deadlocking with linkwatch via keventd. |
7867 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
7868 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
7869 | * |
7870 | * We must not return until all unregister events added during | |
7871 | * the interval the lock was held have been completed. | |
1da177e4 | 7872 | */ |
1da177e4 LT |
7873 | void netdev_run_todo(void) |
7874 | { | |
626ab0e6 | 7875 | struct list_head list; |
1da177e4 | 7876 | |
1da177e4 | 7877 | /* Snapshot list, allow later requests */ |
626ab0e6 | 7878 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
7879 | |
7880 | __rtnl_unlock(); | |
626ab0e6 | 7881 | |
0115e8e3 ED |
7882 | |
7883 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
7884 | if (!list_empty(&list)) |
7885 | rcu_barrier(); | |
7886 | ||
1da177e4 LT |
7887 | while (!list_empty(&list)) { |
7888 | struct net_device *dev | |
e5e26d75 | 7889 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
7890 | list_del(&dev->todo_list); |
7891 | ||
748e2d93 | 7892 | rtnl_lock(); |
0115e8e3 | 7893 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); |
748e2d93 | 7894 | __rtnl_unlock(); |
0115e8e3 | 7895 | |
b17a7c17 | 7896 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 7897 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
7898 | dev->name, dev->reg_state); |
7899 | dump_stack(); | |
7900 | continue; | |
7901 | } | |
1da177e4 | 7902 | |
b17a7c17 | 7903 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 7904 | |
b17a7c17 | 7905 | netdev_wait_allrefs(dev); |
1da177e4 | 7906 | |
b17a7c17 | 7907 | /* paranoia */ |
29b4433d | 7908 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
7909 | BUG_ON(!list_empty(&dev->ptype_all)); |
7910 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
7911 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
7912 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
547b792c | 7913 | WARN_ON(dev->dn_ptr); |
1da177e4 | 7914 | |
cf124db5 DM |
7915 | if (dev->priv_destructor) |
7916 | dev->priv_destructor(dev); | |
7917 | if (dev->needs_free_netdev) | |
7918 | free_netdev(dev); | |
9093bbb2 | 7919 | |
50624c93 EB |
7920 | /* Report a network device has been unregistered */ |
7921 | rtnl_lock(); | |
7922 | dev_net(dev)->dev_unreg_count--; | |
7923 | __rtnl_unlock(); | |
7924 | wake_up(&netdev_unregistering_wq); | |
7925 | ||
9093bbb2 SH |
7926 | /* Free network device */ |
7927 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 7928 | } |
1da177e4 LT |
7929 | } |
7930 | ||
9256645a JW |
7931 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
7932 | * all the same fields in the same order as net_device_stats, with only | |
7933 | * the type differing, but rtnl_link_stats64 may have additional fields | |
7934 | * at the end for newer counters. | |
3cfde79c | 7935 | */ |
77a1abf5 ED |
7936 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
7937 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
7938 | { |
7939 | #if BITS_PER_LONG == 64 | |
9256645a | 7940 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 7941 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
7942 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7943 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
7944 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 7945 | #else |
9256645a | 7946 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
7947 | const unsigned long *src = (const unsigned long *)netdev_stats; |
7948 | u64 *dst = (u64 *)stats64; | |
7949 | ||
9256645a | 7950 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
7951 | for (i = 0; i < n; i++) |
7952 | dst[i] = src[i]; | |
9256645a JW |
7953 | /* zero out counters that only exist in rtnl_link_stats64 */ |
7954 | memset((char *)stats64 + n * sizeof(u64), 0, | |
7955 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
7956 | #endif |
7957 | } | |
77a1abf5 | 7958 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 7959 | |
eeda3fd6 SH |
7960 | /** |
7961 | * dev_get_stats - get network device statistics | |
7962 | * @dev: device to get statistics from | |
28172739 | 7963 | * @storage: place to store stats |
eeda3fd6 | 7964 | * |
d7753516 BH |
7965 | * Get network statistics from device. Return @storage. |
7966 | * The device driver may provide its own method by setting | |
7967 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
7968 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 7969 | */ |
d7753516 BH |
7970 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
7971 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 7972 | { |
eeda3fd6 SH |
7973 | const struct net_device_ops *ops = dev->netdev_ops; |
7974 | ||
28172739 ED |
7975 | if (ops->ndo_get_stats64) { |
7976 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
7977 | ops->ndo_get_stats64(dev, storage); |
7978 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 7979 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
7980 | } else { |
7981 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 7982 | } |
6f64ec74 ED |
7983 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
7984 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
7985 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 7986 | return storage; |
c45d286e | 7987 | } |
eeda3fd6 | 7988 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 7989 | |
24824a09 | 7990 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 7991 | { |
24824a09 | 7992 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 7993 | |
24824a09 ED |
7994 | #ifdef CONFIG_NET_CLS_ACT |
7995 | if (queue) | |
7996 | return queue; | |
7997 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
7998 | if (!queue) | |
7999 | return NULL; | |
8000 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 8001 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
8002 | queue->qdisc_sleeping = &noop_qdisc; |
8003 | rcu_assign_pointer(dev->ingress_queue, queue); | |
8004 | #endif | |
8005 | return queue; | |
bb949fbd DM |
8006 | } |
8007 | ||
2c60db03 ED |
8008 | static const struct ethtool_ops default_ethtool_ops; |
8009 | ||
d07d7507 SG |
8010 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
8011 | const struct ethtool_ops *ops) | |
8012 | { | |
8013 | if (dev->ethtool_ops == &default_ethtool_ops) | |
8014 | dev->ethtool_ops = ops; | |
8015 | } | |
8016 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
8017 | ||
74d332c1 ED |
8018 | void netdev_freemem(struct net_device *dev) |
8019 | { | |
8020 | char *addr = (char *)dev - dev->padded; | |
8021 | ||
4cb28970 | 8022 | kvfree(addr); |
74d332c1 ED |
8023 | } |
8024 | ||
1da177e4 | 8025 | /** |
722c9a0c | 8026 | * alloc_netdev_mqs - allocate network device |
8027 | * @sizeof_priv: size of private data to allocate space for | |
8028 | * @name: device name format string | |
8029 | * @name_assign_type: origin of device name | |
8030 | * @setup: callback to initialize device | |
8031 | * @txqs: the number of TX subqueues to allocate | |
8032 | * @rxqs: the number of RX subqueues to allocate | |
8033 | * | |
8034 | * Allocates a struct net_device with private data area for driver use | |
8035 | * and performs basic initialization. Also allocates subqueue structs | |
8036 | * for each queue on the device. | |
1da177e4 | 8037 | */ |
36909ea4 | 8038 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 8039 | unsigned char name_assign_type, |
36909ea4 TH |
8040 | void (*setup)(struct net_device *), |
8041 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 8042 | { |
1da177e4 | 8043 | struct net_device *dev; |
52a59bd5 | 8044 | unsigned int alloc_size; |
1ce8e7b5 | 8045 | struct net_device *p; |
1da177e4 | 8046 | |
b6fe17d6 SH |
8047 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
8048 | ||
36909ea4 | 8049 | if (txqs < 1) { |
7b6cd1ce | 8050 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
8051 | return NULL; |
8052 | } | |
8053 | ||
a953be53 | 8054 | #ifdef CONFIG_SYSFS |
36909ea4 | 8055 | if (rxqs < 1) { |
7b6cd1ce | 8056 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
8057 | return NULL; |
8058 | } | |
8059 | #endif | |
8060 | ||
fd2ea0a7 | 8061 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
8062 | if (sizeof_priv) { |
8063 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 8064 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
8065 | alloc_size += sizeof_priv; |
8066 | } | |
8067 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 8068 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 8069 | |
dcda9b04 | 8070 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 8071 | if (!p) |
1da177e4 | 8072 | return NULL; |
1da177e4 | 8073 | |
1ce8e7b5 | 8074 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 8075 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 8076 | |
29b4433d ED |
8077 | dev->pcpu_refcnt = alloc_percpu(int); |
8078 | if (!dev->pcpu_refcnt) | |
74d332c1 | 8079 | goto free_dev; |
ab9c73cc | 8080 | |
ab9c73cc | 8081 | if (dev_addr_init(dev)) |
29b4433d | 8082 | goto free_pcpu; |
ab9c73cc | 8083 | |
22bedad3 | 8084 | dev_mc_init(dev); |
a748ee24 | 8085 | dev_uc_init(dev); |
ccffad25 | 8086 | |
c346dca1 | 8087 | dev_net_set(dev, &init_net); |
1da177e4 | 8088 | |
8d3bdbd5 | 8089 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 8090 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 8091 | |
8d3bdbd5 DM |
8092 | INIT_LIST_HEAD(&dev->napi_list); |
8093 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 8094 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 8095 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
8096 | INIT_LIST_HEAD(&dev->adj_list.upper); |
8097 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
8098 | INIT_LIST_HEAD(&dev->ptype_all); |
8099 | INIT_LIST_HEAD(&dev->ptype_specific); | |
59cc1f61 JK |
8100 | #ifdef CONFIG_NET_SCHED |
8101 | hash_init(dev->qdisc_hash); | |
8102 | #endif | |
02875878 | 8103 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
8104 | setup(dev); |
8105 | ||
a813104d | 8106 | if (!dev->tx_queue_len) { |
f84bb1ea | 8107 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 8108 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 8109 | } |
906470c1 | 8110 | |
36909ea4 TH |
8111 | dev->num_tx_queues = txqs; |
8112 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 8113 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 8114 | goto free_all; |
e8a0464c | 8115 | |
a953be53 | 8116 | #ifdef CONFIG_SYSFS |
36909ea4 TH |
8117 | dev->num_rx_queues = rxqs; |
8118 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 8119 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 8120 | goto free_all; |
df334545 | 8121 | #endif |
0a9627f2 | 8122 | |
1da177e4 | 8123 | strcpy(dev->name, name); |
c835a677 | 8124 | dev->name_assign_type = name_assign_type; |
cbda10fa | 8125 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
8126 | if (!dev->ethtool_ops) |
8127 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
8128 | |
8129 | nf_hook_ingress_init(dev); | |
8130 | ||
1da177e4 | 8131 | return dev; |
ab9c73cc | 8132 | |
8d3bdbd5 DM |
8133 | free_all: |
8134 | free_netdev(dev); | |
8135 | return NULL; | |
8136 | ||
29b4433d ED |
8137 | free_pcpu: |
8138 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
8139 | free_dev: |
8140 | netdev_freemem(dev); | |
ab9c73cc | 8141 | return NULL; |
1da177e4 | 8142 | } |
36909ea4 | 8143 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
8144 | |
8145 | /** | |
722c9a0c | 8146 | * free_netdev - free network device |
8147 | * @dev: device | |
1da177e4 | 8148 | * |
722c9a0c | 8149 | * This function does the last stage of destroying an allocated device |
8150 | * interface. The reference to the device object is released. If this | |
8151 | * is the last reference then it will be freed.Must be called in process | |
8152 | * context. | |
1da177e4 LT |
8153 | */ |
8154 | void free_netdev(struct net_device *dev) | |
8155 | { | |
d565b0a1 | 8156 | struct napi_struct *p, *n; |
b5cdae32 | 8157 | struct bpf_prog *prog; |
d565b0a1 | 8158 | |
93d05d4a | 8159 | might_sleep(); |
60877a32 | 8160 | netif_free_tx_queues(dev); |
a953be53 | 8161 | #ifdef CONFIG_SYSFS |
10595902 | 8162 | kvfree(dev->_rx); |
fe822240 | 8163 | #endif |
e8a0464c | 8164 | |
33d480ce | 8165 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 8166 | |
f001fde5 JP |
8167 | /* Flush device addresses */ |
8168 | dev_addr_flush(dev); | |
8169 | ||
d565b0a1 HX |
8170 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
8171 | netif_napi_del(p); | |
8172 | ||
29b4433d ED |
8173 | free_percpu(dev->pcpu_refcnt); |
8174 | dev->pcpu_refcnt = NULL; | |
8175 | ||
b5cdae32 DM |
8176 | prog = rcu_dereference_protected(dev->xdp_prog, 1); |
8177 | if (prog) { | |
8178 | bpf_prog_put(prog); | |
8179 | static_key_slow_dec(&generic_xdp_needed); | |
8180 | } | |
8181 | ||
3041a069 | 8182 | /* Compatibility with error handling in drivers */ |
1da177e4 | 8183 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 8184 | netdev_freemem(dev); |
1da177e4 LT |
8185 | return; |
8186 | } | |
8187 | ||
8188 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
8189 | dev->reg_state = NETREG_RELEASED; | |
8190 | ||
43cb76d9 GKH |
8191 | /* will free via device release */ |
8192 | put_device(&dev->dev); | |
1da177e4 | 8193 | } |
d1b19dff | 8194 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 8195 | |
f0db275a SH |
8196 | /** |
8197 | * synchronize_net - Synchronize with packet receive processing | |
8198 | * | |
8199 | * Wait for packets currently being received to be done. | |
8200 | * Does not block later packets from starting. | |
8201 | */ | |
4ec93edb | 8202 | void synchronize_net(void) |
1da177e4 LT |
8203 | { |
8204 | might_sleep(); | |
be3fc413 ED |
8205 | if (rtnl_is_locked()) |
8206 | synchronize_rcu_expedited(); | |
8207 | else | |
8208 | synchronize_rcu(); | |
1da177e4 | 8209 | } |
d1b19dff | 8210 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
8211 | |
8212 | /** | |
44a0873d | 8213 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 8214 | * @dev: device |
44a0873d | 8215 | * @head: list |
6ebfbc06 | 8216 | * |
1da177e4 | 8217 | * This function shuts down a device interface and removes it |
d59b54b1 | 8218 | * from the kernel tables. |
44a0873d | 8219 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
8220 | * |
8221 | * Callers must hold the rtnl semaphore. You may want | |
8222 | * unregister_netdev() instead of this. | |
8223 | */ | |
8224 | ||
44a0873d | 8225 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 8226 | { |
a6620712 HX |
8227 | ASSERT_RTNL(); |
8228 | ||
44a0873d | 8229 | if (head) { |
9fdce099 | 8230 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
8231 | } else { |
8232 | rollback_registered(dev); | |
8233 | /* Finish processing unregister after unlock */ | |
8234 | net_set_todo(dev); | |
8235 | } | |
1da177e4 | 8236 | } |
44a0873d | 8237 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 8238 | |
9b5e383c ED |
8239 | /** |
8240 | * unregister_netdevice_many - unregister many devices | |
8241 | * @head: list of devices | |
87757a91 ED |
8242 | * |
8243 | * Note: As most callers use a stack allocated list_head, | |
8244 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
8245 | */ |
8246 | void unregister_netdevice_many(struct list_head *head) | |
8247 | { | |
8248 | struct net_device *dev; | |
8249 | ||
8250 | if (!list_empty(head)) { | |
8251 | rollback_registered_many(head); | |
8252 | list_for_each_entry(dev, head, unreg_list) | |
8253 | net_set_todo(dev); | |
87757a91 | 8254 | list_del(head); |
9b5e383c ED |
8255 | } |
8256 | } | |
63c8099d | 8257 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 8258 | |
1da177e4 LT |
8259 | /** |
8260 | * unregister_netdev - remove device from the kernel | |
8261 | * @dev: device | |
8262 | * | |
8263 | * This function shuts down a device interface and removes it | |
d59b54b1 | 8264 | * from the kernel tables. |
1da177e4 LT |
8265 | * |
8266 | * This is just a wrapper for unregister_netdevice that takes | |
8267 | * the rtnl semaphore. In general you want to use this and not | |
8268 | * unregister_netdevice. | |
8269 | */ | |
8270 | void unregister_netdev(struct net_device *dev) | |
8271 | { | |
8272 | rtnl_lock(); | |
8273 | unregister_netdevice(dev); | |
8274 | rtnl_unlock(); | |
8275 | } | |
1da177e4 LT |
8276 | EXPORT_SYMBOL(unregister_netdev); |
8277 | ||
ce286d32 EB |
8278 | /** |
8279 | * dev_change_net_namespace - move device to different nethost namespace | |
8280 | * @dev: device | |
8281 | * @net: network namespace | |
8282 | * @pat: If not NULL name pattern to try if the current device name | |
8283 | * is already taken in the destination network namespace. | |
8284 | * | |
8285 | * This function shuts down a device interface and moves it | |
8286 | * to a new network namespace. On success 0 is returned, on | |
8287 | * a failure a netagive errno code is returned. | |
8288 | * | |
8289 | * Callers must hold the rtnl semaphore. | |
8290 | */ | |
8291 | ||
8292 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
8293 | { | |
ce286d32 EB |
8294 | int err; |
8295 | ||
8296 | ASSERT_RTNL(); | |
8297 | ||
8298 | /* Don't allow namespace local devices to be moved. */ | |
8299 | err = -EINVAL; | |
8300 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8301 | goto out; | |
8302 | ||
8303 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
8304 | if (dev->reg_state != NETREG_REGISTERED) |
8305 | goto out; | |
8306 | ||
8307 | /* Get out if there is nothing todo */ | |
8308 | err = 0; | |
878628fb | 8309 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
8310 | goto out; |
8311 | ||
8312 | /* Pick the destination device name, and ensure | |
8313 | * we can use it in the destination network namespace. | |
8314 | */ | |
8315 | err = -EEXIST; | |
d9031024 | 8316 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
8317 | /* We get here if we can't use the current device name */ |
8318 | if (!pat) | |
8319 | goto out; | |
828de4f6 | 8320 | if (dev_get_valid_name(net, dev, pat) < 0) |
ce286d32 EB |
8321 | goto out; |
8322 | } | |
8323 | ||
8324 | /* | |
8325 | * And now a mini version of register_netdevice unregister_netdevice. | |
8326 | */ | |
8327 | ||
8328 | /* If device is running close it first. */ | |
9b772652 | 8329 | dev_close(dev); |
ce286d32 EB |
8330 | |
8331 | /* And unlink it from device chain */ | |
8332 | err = -ENODEV; | |
8333 | unlist_netdevice(dev); | |
8334 | ||
8335 | synchronize_net(); | |
8336 | ||
8337 | /* Shutdown queueing discipline. */ | |
8338 | dev_shutdown(dev); | |
8339 | ||
8340 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 8341 | * this device. They should clean all the things. |
8342 | * | |
8343 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
8344 | * This is wanted because this way 8021q and macvlan know | |
8345 | * the device is just moving and can keep their slaves up. | |
8346 | */ | |
ce286d32 | 8347 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 G |
8348 | rcu_barrier(); |
8349 | call_netdevice_notifiers(NETDEV_UNREGISTER_FINAL, dev); | |
7f294054 | 8350 | rtmsg_ifinfo(RTM_DELLINK, dev, ~0U, GFP_KERNEL); |
ce286d32 EB |
8351 | |
8352 | /* | |
8353 | * Flush the unicast and multicast chains | |
8354 | */ | |
a748ee24 | 8355 | dev_uc_flush(dev); |
22bedad3 | 8356 | dev_mc_flush(dev); |
ce286d32 | 8357 | |
4e66ae2e SH |
8358 | /* Send a netdev-removed uevent to the old namespace */ |
8359 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 8360 | netdev_adjacent_del_links(dev); |
4e66ae2e | 8361 | |
ce286d32 | 8362 | /* Actually switch the network namespace */ |
c346dca1 | 8363 | dev_net_set(dev, net); |
ce286d32 | 8364 | |
ce286d32 | 8365 | /* If there is an ifindex conflict assign a new one */ |
7a66bbc9 | 8366 | if (__dev_get_by_index(net, dev->ifindex)) |
ce286d32 | 8367 | dev->ifindex = dev_new_index(net); |
ce286d32 | 8368 | |
4e66ae2e SH |
8369 | /* Send a netdev-add uevent to the new namespace */ |
8370 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 8371 | netdev_adjacent_add_links(dev); |
4e66ae2e | 8372 | |
8b41d188 | 8373 | /* Fixup kobjects */ |
a1b3f594 | 8374 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 8375 | WARN_ON(err); |
ce286d32 EB |
8376 | |
8377 | /* Add the device back in the hashes */ | |
8378 | list_netdevice(dev); | |
8379 | ||
8380 | /* Notify protocols, that a new device appeared. */ | |
8381 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
8382 | ||
d90a909e EB |
8383 | /* |
8384 | * Prevent userspace races by waiting until the network | |
8385 | * device is fully setup before sending notifications. | |
8386 | */ | |
7f294054 | 8387 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 8388 | |
ce286d32 EB |
8389 | synchronize_net(); |
8390 | err = 0; | |
8391 | out: | |
8392 | return err; | |
8393 | } | |
463d0183 | 8394 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 8395 | |
f0bf90de | 8396 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
8397 | { |
8398 | struct sk_buff **list_skb; | |
1da177e4 | 8399 | struct sk_buff *skb; |
f0bf90de | 8400 | unsigned int cpu; |
97d8b6e3 | 8401 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 8402 | |
1da177e4 LT |
8403 | local_irq_disable(); |
8404 | cpu = smp_processor_id(); | |
8405 | sd = &per_cpu(softnet_data, cpu); | |
8406 | oldsd = &per_cpu(softnet_data, oldcpu); | |
8407 | ||
8408 | /* Find end of our completion_queue. */ | |
8409 | list_skb = &sd->completion_queue; | |
8410 | while (*list_skb) | |
8411 | list_skb = &(*list_skb)->next; | |
8412 | /* Append completion queue from offline CPU. */ | |
8413 | *list_skb = oldsd->completion_queue; | |
8414 | oldsd->completion_queue = NULL; | |
8415 | ||
1da177e4 | 8416 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
8417 | if (oldsd->output_queue) { |
8418 | *sd->output_queue_tailp = oldsd->output_queue; | |
8419 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
8420 | oldsd->output_queue = NULL; | |
8421 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
8422 | } | |
ac64da0b ED |
8423 | /* Append NAPI poll list from offline CPU, with one exception : |
8424 | * process_backlog() must be called by cpu owning percpu backlog. | |
8425 | * We properly handle process_queue & input_pkt_queue later. | |
8426 | */ | |
8427 | while (!list_empty(&oldsd->poll_list)) { | |
8428 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
8429 | struct napi_struct, | |
8430 | poll_list); | |
8431 | ||
8432 | list_del_init(&napi->poll_list); | |
8433 | if (napi->poll == process_backlog) | |
8434 | napi->state = 0; | |
8435 | else | |
8436 | ____napi_schedule(sd, napi); | |
264524d5 | 8437 | } |
1da177e4 LT |
8438 | |
8439 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
8440 | local_irq_enable(); | |
8441 | ||
773fc8f6 | 8442 | #ifdef CONFIG_RPS |
8443 | remsd = oldsd->rps_ipi_list; | |
8444 | oldsd->rps_ipi_list = NULL; | |
8445 | #endif | |
8446 | /* send out pending IPI's on offline CPU */ | |
8447 | net_rps_send_ipi(remsd); | |
8448 | ||
1da177e4 | 8449 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 8450 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 8451 | netif_rx_ni(skb); |
76cc8b13 | 8452 | input_queue_head_incr(oldsd); |
fec5e652 | 8453 | } |
ac64da0b | 8454 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 8455 | netif_rx_ni(skb); |
76cc8b13 TH |
8456 | input_queue_head_incr(oldsd); |
8457 | } | |
1da177e4 | 8458 | |
f0bf90de | 8459 | return 0; |
1da177e4 | 8460 | } |
1da177e4 | 8461 | |
7f353bf2 | 8462 | /** |
b63365a2 HX |
8463 | * netdev_increment_features - increment feature set by one |
8464 | * @all: current feature set | |
8465 | * @one: new feature set | |
8466 | * @mask: mask feature set | |
7f353bf2 HX |
8467 | * |
8468 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
8469 | * @one to the master device with current feature set @all. Will not |
8470 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 8471 | */ |
c8f44aff MM |
8472 | netdev_features_t netdev_increment_features(netdev_features_t all, |
8473 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 8474 | { |
c8cd0989 | 8475 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 8476 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 8477 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 8478 | |
a188222b | 8479 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 8480 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 8481 | |
1742f183 | 8482 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
8483 | if (all & NETIF_F_HW_CSUM) |
8484 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
8485 | |
8486 | return all; | |
8487 | } | |
b63365a2 | 8488 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 8489 | |
430f03cd | 8490 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
8491 | { |
8492 | int i; | |
8493 | struct hlist_head *hash; | |
8494 | ||
8495 | hash = kmalloc(sizeof(*hash) * NETDEV_HASHENTRIES, GFP_KERNEL); | |
8496 | if (hash != NULL) | |
8497 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
8498 | INIT_HLIST_HEAD(&hash[i]); | |
8499 | ||
8500 | return hash; | |
8501 | } | |
8502 | ||
881d966b | 8503 | /* Initialize per network namespace state */ |
4665079c | 8504 | static int __net_init netdev_init(struct net *net) |
881d966b | 8505 | { |
734b6541 RM |
8506 | if (net != &init_net) |
8507 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 8508 | |
30d97d35 PE |
8509 | net->dev_name_head = netdev_create_hash(); |
8510 | if (net->dev_name_head == NULL) | |
8511 | goto err_name; | |
881d966b | 8512 | |
30d97d35 PE |
8513 | net->dev_index_head = netdev_create_hash(); |
8514 | if (net->dev_index_head == NULL) | |
8515 | goto err_idx; | |
881d966b EB |
8516 | |
8517 | return 0; | |
30d97d35 PE |
8518 | |
8519 | err_idx: | |
8520 | kfree(net->dev_name_head); | |
8521 | err_name: | |
8522 | return -ENOMEM; | |
881d966b EB |
8523 | } |
8524 | ||
f0db275a SH |
8525 | /** |
8526 | * netdev_drivername - network driver for the device | |
8527 | * @dev: network device | |
f0db275a SH |
8528 | * |
8529 | * Determine network driver for device. | |
8530 | */ | |
3019de12 | 8531 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 8532 | { |
cf04a4c7 SH |
8533 | const struct device_driver *driver; |
8534 | const struct device *parent; | |
3019de12 | 8535 | const char *empty = ""; |
6579e57b AV |
8536 | |
8537 | parent = dev->dev.parent; | |
6579e57b | 8538 | if (!parent) |
3019de12 | 8539 | return empty; |
6579e57b AV |
8540 | |
8541 | driver = parent->driver; | |
8542 | if (driver && driver->name) | |
3019de12 DM |
8543 | return driver->name; |
8544 | return empty; | |
6579e57b AV |
8545 | } |
8546 | ||
6ea754eb JP |
8547 | static void __netdev_printk(const char *level, const struct net_device *dev, |
8548 | struct va_format *vaf) | |
256df2f3 | 8549 | { |
b004ff49 | 8550 | if (dev && dev->dev.parent) { |
6ea754eb JP |
8551 | dev_printk_emit(level[1] - '0', |
8552 | dev->dev.parent, | |
8553 | "%s %s %s%s: %pV", | |
8554 | dev_driver_string(dev->dev.parent), | |
8555 | dev_name(dev->dev.parent), | |
8556 | netdev_name(dev), netdev_reg_state(dev), | |
8557 | vaf); | |
b004ff49 | 8558 | } else if (dev) { |
6ea754eb JP |
8559 | printk("%s%s%s: %pV", |
8560 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 8561 | } else { |
6ea754eb | 8562 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 8563 | } |
256df2f3 JP |
8564 | } |
8565 | ||
6ea754eb JP |
8566 | void netdev_printk(const char *level, const struct net_device *dev, |
8567 | const char *format, ...) | |
256df2f3 JP |
8568 | { |
8569 | struct va_format vaf; | |
8570 | va_list args; | |
256df2f3 JP |
8571 | |
8572 | va_start(args, format); | |
8573 | ||
8574 | vaf.fmt = format; | |
8575 | vaf.va = &args; | |
8576 | ||
6ea754eb | 8577 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 8578 | |
256df2f3 | 8579 | va_end(args); |
256df2f3 JP |
8580 | } |
8581 | EXPORT_SYMBOL(netdev_printk); | |
8582 | ||
8583 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 8584 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 8585 | { \ |
256df2f3 JP |
8586 | struct va_format vaf; \ |
8587 | va_list args; \ | |
8588 | \ | |
8589 | va_start(args, fmt); \ | |
8590 | \ | |
8591 | vaf.fmt = fmt; \ | |
8592 | vaf.va = &args; \ | |
8593 | \ | |
6ea754eb | 8594 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 8595 | \ |
256df2f3 | 8596 | va_end(args); \ |
256df2f3 JP |
8597 | } \ |
8598 | EXPORT_SYMBOL(func); | |
8599 | ||
8600 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
8601 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
8602 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
8603 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
8604 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
8605 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
8606 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
8607 | ||
4665079c | 8608 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
8609 | { |
8610 | kfree(net->dev_name_head); | |
8611 | kfree(net->dev_index_head); | |
8612 | } | |
8613 | ||
022cbae6 | 8614 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
8615 | .init = netdev_init, |
8616 | .exit = netdev_exit, | |
8617 | }; | |
8618 | ||
4665079c | 8619 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 8620 | { |
e008b5fc | 8621 | struct net_device *dev, *aux; |
ce286d32 | 8622 | /* |
e008b5fc | 8623 | * Push all migratable network devices back to the |
ce286d32 EB |
8624 | * initial network namespace |
8625 | */ | |
8626 | rtnl_lock(); | |
e008b5fc | 8627 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 8628 | int err; |
aca51397 | 8629 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
8630 | |
8631 | /* Ignore unmoveable devices (i.e. loopback) */ | |
8632 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
8633 | continue; | |
8634 | ||
e008b5fc EB |
8635 | /* Leave virtual devices for the generic cleanup */ |
8636 | if (dev->rtnl_link_ops) | |
8637 | continue; | |
d0c082ce | 8638 | |
25985edc | 8639 | /* Push remaining network devices to init_net */ |
aca51397 PE |
8640 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
8641 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 8642 | if (err) { |
7b6cd1ce JP |
8643 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
8644 | __func__, dev->name, err); | |
aca51397 | 8645 | BUG(); |
ce286d32 EB |
8646 | } |
8647 | } | |
8648 | rtnl_unlock(); | |
8649 | } | |
8650 | ||
50624c93 EB |
8651 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
8652 | { | |
8653 | /* Return with the rtnl_lock held when there are no network | |
8654 | * devices unregistering in any network namespace in net_list. | |
8655 | */ | |
8656 | struct net *net; | |
8657 | bool unregistering; | |
ff960a73 | 8658 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 8659 | |
ff960a73 | 8660 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 8661 | for (;;) { |
50624c93 EB |
8662 | unregistering = false; |
8663 | rtnl_lock(); | |
8664 | list_for_each_entry(net, net_list, exit_list) { | |
8665 | if (net->dev_unreg_count > 0) { | |
8666 | unregistering = true; | |
8667 | break; | |
8668 | } | |
8669 | } | |
8670 | if (!unregistering) | |
8671 | break; | |
8672 | __rtnl_unlock(); | |
ff960a73 PZ |
8673 | |
8674 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 8675 | } |
ff960a73 | 8676 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
8677 | } |
8678 | ||
04dc7f6b EB |
8679 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
8680 | { | |
8681 | /* At exit all network devices most be removed from a network | |
b595076a | 8682 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
8683 | * Do this across as many network namespaces as possible to |
8684 | * improve batching efficiency. | |
8685 | */ | |
8686 | struct net_device *dev; | |
8687 | struct net *net; | |
8688 | LIST_HEAD(dev_kill_list); | |
8689 | ||
50624c93 EB |
8690 | /* To prevent network device cleanup code from dereferencing |
8691 | * loopback devices or network devices that have been freed | |
8692 | * wait here for all pending unregistrations to complete, | |
8693 | * before unregistring the loopback device and allowing the | |
8694 | * network namespace be freed. | |
8695 | * | |
8696 | * The netdev todo list containing all network devices | |
8697 | * unregistrations that happen in default_device_exit_batch | |
8698 | * will run in the rtnl_unlock() at the end of | |
8699 | * default_device_exit_batch. | |
8700 | */ | |
8701 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
8702 | list_for_each_entry(net, net_list, exit_list) { |
8703 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 8704 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
8705 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
8706 | else | |
8707 | unregister_netdevice_queue(dev, &dev_kill_list); | |
8708 | } | |
8709 | } | |
8710 | unregister_netdevice_many(&dev_kill_list); | |
8711 | rtnl_unlock(); | |
8712 | } | |
8713 | ||
022cbae6 | 8714 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 8715 | .exit = default_device_exit, |
04dc7f6b | 8716 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
8717 | }; |
8718 | ||
1da177e4 LT |
8719 | /* |
8720 | * Initialize the DEV module. At boot time this walks the device list and | |
8721 | * unhooks any devices that fail to initialise (normally hardware not | |
8722 | * present) and leaves us with a valid list of present and active devices. | |
8723 | * | |
8724 | */ | |
8725 | ||
8726 | /* | |
8727 | * This is called single threaded during boot, so no need | |
8728 | * to take the rtnl semaphore. | |
8729 | */ | |
8730 | static int __init net_dev_init(void) | |
8731 | { | |
8732 | int i, rc = -ENOMEM; | |
8733 | ||
8734 | BUG_ON(!dev_boot_phase); | |
8735 | ||
1da177e4 LT |
8736 | if (dev_proc_init()) |
8737 | goto out; | |
8738 | ||
8b41d188 | 8739 | if (netdev_kobject_init()) |
1da177e4 LT |
8740 | goto out; |
8741 | ||
8742 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 8743 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
8744 | INIT_LIST_HEAD(&ptype_base[i]); |
8745 | ||
62532da9 VY |
8746 | INIT_LIST_HEAD(&offload_base); |
8747 | ||
881d966b EB |
8748 | if (register_pernet_subsys(&netdev_net_ops)) |
8749 | goto out; | |
1da177e4 LT |
8750 | |
8751 | /* | |
8752 | * Initialise the packet receive queues. | |
8753 | */ | |
8754 | ||
6f912042 | 8755 | for_each_possible_cpu(i) { |
41852497 | 8756 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 8757 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 8758 | |
41852497 ED |
8759 | INIT_WORK(flush, flush_backlog); |
8760 | ||
e36fa2f7 | 8761 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 8762 | skb_queue_head_init(&sd->process_queue); |
e36fa2f7 | 8763 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 8764 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 8765 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
8766 | sd->csd.func = rps_trigger_softirq; |
8767 | sd->csd.info = sd; | |
e36fa2f7 | 8768 | sd->cpu = i; |
1e94d72f | 8769 | #endif |
0a9627f2 | 8770 | |
e36fa2f7 ED |
8771 | sd->backlog.poll = process_backlog; |
8772 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
8773 | } |
8774 | ||
1da177e4 LT |
8775 | dev_boot_phase = 0; |
8776 | ||
505d4f73 EB |
8777 | /* The loopback device is special if any other network devices |
8778 | * is present in a network namespace the loopback device must | |
8779 | * be present. Since we now dynamically allocate and free the | |
8780 | * loopback device ensure this invariant is maintained by | |
8781 | * keeping the loopback device as the first device on the | |
8782 | * list of network devices. Ensuring the loopback devices | |
8783 | * is the first device that appears and the last network device | |
8784 | * that disappears. | |
8785 | */ | |
8786 | if (register_pernet_device(&loopback_net_ops)) | |
8787 | goto out; | |
8788 | ||
8789 | if (register_pernet_device(&default_device_ops)) | |
8790 | goto out; | |
8791 | ||
962cf36c CM |
8792 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
8793 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 8794 | |
f0bf90de SAS |
8795 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
8796 | NULL, dev_cpu_dead); | |
8797 | WARN_ON(rc < 0); | |
1da177e4 LT |
8798 | rc = 0; |
8799 | out: | |
8800 | return rc; | |
8801 | } | |
8802 | ||
8803 | subsys_initcall(net_dev_init); |