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 | 96 | #include <linux/skbuff.h> |
a7862b45 | 97 | #include <linux/bpf.h> |
b5cdae32 | 98 | #include <linux/bpf_trace.h> |
457c4cbc | 99 | #include <net/net_namespace.h> |
1da177e4 | 100 | #include <net/sock.h> |
02d62e86 | 101 | #include <net/busy_poll.h> |
1da177e4 | 102 | #include <linux/rtnetlink.h> |
1da177e4 | 103 | #include <linux/stat.h> |
1da177e4 | 104 | #include <net/dst.h> |
fc4099f1 | 105 | #include <net/dst_metadata.h> |
1da177e4 | 106 | #include <net/pkt_sched.h> |
87d83093 | 107 | #include <net/pkt_cls.h> |
1da177e4 | 108 | #include <net/checksum.h> |
44540960 | 109 | #include <net/xfrm.h> |
1da177e4 LT |
110 | #include <linux/highmem.h> |
111 | #include <linux/init.h> | |
1da177e4 | 112 | #include <linux/module.h> |
1da177e4 LT |
113 | #include <linux/netpoll.h> |
114 | #include <linux/rcupdate.h> | |
115 | #include <linux/delay.h> | |
1da177e4 | 116 | #include <net/iw_handler.h> |
1da177e4 | 117 | #include <asm/current.h> |
5bdb9886 | 118 | #include <linux/audit.h> |
db217334 | 119 | #include <linux/dmaengine.h> |
f6a78bfc | 120 | #include <linux/err.h> |
c7fa9d18 | 121 | #include <linux/ctype.h> |
723e98b7 | 122 | #include <linux/if_arp.h> |
6de329e2 | 123 | #include <linux/if_vlan.h> |
8f0f2223 | 124 | #include <linux/ip.h> |
ad55dcaf | 125 | #include <net/ip.h> |
25cd9ba0 | 126 | #include <net/mpls.h> |
8f0f2223 DM |
127 | #include <linux/ipv6.h> |
128 | #include <linux/in.h> | |
b6b2fed1 DM |
129 | #include <linux/jhash.h> |
130 | #include <linux/random.h> | |
9cbc1cb8 | 131 | #include <trace/events/napi.h> |
cf66ba58 | 132 | #include <trace/events/net.h> |
07dc22e7 | 133 | #include <trace/events/skb.h> |
5acbbd42 | 134 | #include <linux/pci.h> |
caeda9b9 | 135 | #include <linux/inetdevice.h> |
c445477d | 136 | #include <linux/cpu_rmap.h> |
c5905afb | 137 | #include <linux/static_key.h> |
af12fa6e | 138 | #include <linux/hashtable.h> |
60877a32 | 139 | #include <linux/vmalloc.h> |
529d0489 | 140 | #include <linux/if_macvlan.h> |
e7fd2885 | 141 | #include <linux/errqueue.h> |
3b47d303 | 142 | #include <linux/hrtimer.h> |
e687ad60 | 143 | #include <linux/netfilter_ingress.h> |
40e4e713 | 144 | #include <linux/crash_dump.h> |
b72b5bf6 | 145 | #include <linux/sctp.h> |
ae847f40 | 146 | #include <net/udp_tunnel.h> |
6621dd29 | 147 | #include <linux/net_namespace.h> |
aaa5d90b | 148 | #include <linux/indirect_call_wrapper.h> |
af3836df | 149 | #include <net/devlink.h> |
1da177e4 | 150 | |
342709ef PE |
151 | #include "net-sysfs.h" |
152 | ||
d565b0a1 HX |
153 | #define MAX_GRO_SKBS 8 |
154 | ||
5d38a079 HX |
155 | /* This should be increased if a protocol with a bigger head is added. */ |
156 | #define GRO_MAX_HEAD (MAX_HEADER + 128) | |
157 | ||
1da177e4 | 158 | static DEFINE_SPINLOCK(ptype_lock); |
62532da9 | 159 | static DEFINE_SPINLOCK(offload_lock); |
900ff8c6 CW |
160 | struct list_head ptype_base[PTYPE_HASH_SIZE] __read_mostly; |
161 | struct list_head ptype_all __read_mostly; /* Taps */ | |
62532da9 | 162 | static struct list_head offload_base __read_mostly; |
1da177e4 | 163 | |
ae78dbfa | 164 | static int netif_rx_internal(struct sk_buff *skb); |
54951194 | 165 | static int call_netdevice_notifiers_info(unsigned long val, |
54951194 | 166 | struct netdev_notifier_info *info); |
26372605 PM |
167 | static int call_netdevice_notifiers_extack(unsigned long val, |
168 | struct net_device *dev, | |
169 | struct netlink_ext_ack *extack); | |
90b602f8 | 170 | static struct napi_struct *napi_by_id(unsigned int napi_id); |
ae78dbfa | 171 | |
1da177e4 | 172 | /* |
7562f876 | 173 | * The @dev_base_head list is protected by @dev_base_lock and the rtnl |
1da177e4 LT |
174 | * semaphore. |
175 | * | |
c6d14c84 | 176 | * Pure readers hold dev_base_lock for reading, or rcu_read_lock() |
1da177e4 LT |
177 | * |
178 | * Writers must hold the rtnl semaphore while they loop through the | |
7562f876 | 179 | * dev_base_head list, and hold dev_base_lock for writing when they do the |
1da177e4 LT |
180 | * actual updates. This allows pure readers to access the list even |
181 | * while a writer is preparing to update it. | |
182 | * | |
183 | * To put it another way, dev_base_lock is held for writing only to | |
184 | * protect against pure readers; the rtnl semaphore provides the | |
185 | * protection against other writers. | |
186 | * | |
187 | * See, for example usages, register_netdevice() and | |
188 | * unregister_netdevice(), which must be called with the rtnl | |
189 | * semaphore held. | |
190 | */ | |
1da177e4 | 191 | DEFINE_RWLOCK(dev_base_lock); |
1da177e4 LT |
192 | EXPORT_SYMBOL(dev_base_lock); |
193 | ||
6c557001 FW |
194 | static DEFINE_MUTEX(ifalias_mutex); |
195 | ||
af12fa6e ET |
196 | /* protects napi_hash addition/deletion and napi_gen_id */ |
197 | static DEFINE_SPINLOCK(napi_hash_lock); | |
198 | ||
52bd2d62 | 199 | static unsigned int napi_gen_id = NR_CPUS; |
6180d9de | 200 | static DEFINE_READ_MOSTLY_HASHTABLE(napi_hash, 8); |
af12fa6e | 201 | |
18afa4b0 | 202 | static seqcount_t devnet_rename_seq; |
c91f6df2 | 203 | |
4e985ada TG |
204 | static inline void dev_base_seq_inc(struct net *net) |
205 | { | |
643aa9cb | 206 | while (++net->dev_base_seq == 0) |
207 | ; | |
4e985ada TG |
208 | } |
209 | ||
881d966b | 210 | static inline struct hlist_head *dev_name_hash(struct net *net, const char *name) |
1da177e4 | 211 | { |
8387ff25 | 212 | unsigned int hash = full_name_hash(net, name, strnlen(name, IFNAMSIZ)); |
95c96174 | 213 | |
08e9897d | 214 | return &net->dev_name_head[hash_32(hash, NETDEV_HASHBITS)]; |
1da177e4 LT |
215 | } |
216 | ||
881d966b | 217 | static inline struct hlist_head *dev_index_hash(struct net *net, int ifindex) |
1da177e4 | 218 | { |
7c28bd0b | 219 | return &net->dev_index_head[ifindex & (NETDEV_HASHENTRIES - 1)]; |
1da177e4 LT |
220 | } |
221 | ||
e36fa2f7 | 222 | static inline void rps_lock(struct softnet_data *sd) |
152102c7 CG |
223 | { |
224 | #ifdef CONFIG_RPS | |
e36fa2f7 | 225 | spin_lock(&sd->input_pkt_queue.lock); |
152102c7 CG |
226 | #endif |
227 | } | |
228 | ||
e36fa2f7 | 229 | static inline void rps_unlock(struct softnet_data *sd) |
152102c7 CG |
230 | { |
231 | #ifdef CONFIG_RPS | |
e36fa2f7 | 232 | spin_unlock(&sd->input_pkt_queue.lock); |
152102c7 CG |
233 | #endif |
234 | } | |
235 | ||
ce286d32 | 236 | /* Device list insertion */ |
53759be9 | 237 | static void list_netdevice(struct net_device *dev) |
ce286d32 | 238 | { |
c346dca1 | 239 | struct net *net = dev_net(dev); |
ce286d32 EB |
240 | |
241 | ASSERT_RTNL(); | |
242 | ||
243 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 244 | list_add_tail_rcu(&dev->dev_list, &net->dev_base_head); |
72c9528b | 245 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); |
fb699dfd ED |
246 | hlist_add_head_rcu(&dev->index_hlist, |
247 | dev_index_hash(net, dev->ifindex)); | |
ce286d32 | 248 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
249 | |
250 | dev_base_seq_inc(net); | |
ce286d32 EB |
251 | } |
252 | ||
fb699dfd ED |
253 | /* Device list removal |
254 | * caller must respect a RCU grace period before freeing/reusing dev | |
255 | */ | |
ce286d32 EB |
256 | static void unlist_netdevice(struct net_device *dev) |
257 | { | |
258 | ASSERT_RTNL(); | |
259 | ||
260 | /* Unlink dev from the device chain */ | |
261 | write_lock_bh(&dev_base_lock); | |
c6d14c84 | 262 | list_del_rcu(&dev->dev_list); |
72c9528b | 263 | hlist_del_rcu(&dev->name_hlist); |
fb699dfd | 264 | hlist_del_rcu(&dev->index_hlist); |
ce286d32 | 265 | write_unlock_bh(&dev_base_lock); |
4e985ada TG |
266 | |
267 | dev_base_seq_inc(dev_net(dev)); | |
ce286d32 EB |
268 | } |
269 | ||
1da177e4 LT |
270 | /* |
271 | * Our notifier list | |
272 | */ | |
273 | ||
f07d5b94 | 274 | static RAW_NOTIFIER_HEAD(netdev_chain); |
1da177e4 LT |
275 | |
276 | /* | |
277 | * Device drivers call our routines to queue packets here. We empty the | |
278 | * queue in the local softnet handler. | |
279 | */ | |
bea3348e | 280 | |
9958da05 | 281 | DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data); |
d1b19dff | 282 | EXPORT_PER_CPU_SYMBOL(softnet_data); |
1da177e4 | 283 | |
cf508b12 | 284 | #ifdef CONFIG_LOCKDEP |
723e98b7 | 285 | /* |
c773e847 | 286 | * register_netdevice() inits txq->_xmit_lock and sets lockdep class |
723e98b7 JP |
287 | * according to dev->type |
288 | */ | |
643aa9cb | 289 | static const unsigned short netdev_lock_type[] = { |
290 | ARPHRD_NETROM, ARPHRD_ETHER, ARPHRD_EETHER, ARPHRD_AX25, | |
723e98b7 JP |
291 | ARPHRD_PRONET, ARPHRD_CHAOS, ARPHRD_IEEE802, ARPHRD_ARCNET, |
292 | ARPHRD_APPLETLK, ARPHRD_DLCI, ARPHRD_ATM, ARPHRD_METRICOM, | |
293 | ARPHRD_IEEE1394, ARPHRD_EUI64, ARPHRD_INFINIBAND, ARPHRD_SLIP, | |
294 | ARPHRD_CSLIP, ARPHRD_SLIP6, ARPHRD_CSLIP6, ARPHRD_RSRVD, | |
295 | ARPHRD_ADAPT, ARPHRD_ROSE, ARPHRD_X25, ARPHRD_HWX25, | |
296 | ARPHRD_PPP, ARPHRD_CISCO, ARPHRD_LAPB, ARPHRD_DDCMP, | |
297 | ARPHRD_RAWHDLC, ARPHRD_TUNNEL, ARPHRD_TUNNEL6, ARPHRD_FRAD, | |
298 | ARPHRD_SKIP, ARPHRD_LOOPBACK, ARPHRD_LOCALTLK, ARPHRD_FDDI, | |
299 | ARPHRD_BIF, ARPHRD_SIT, ARPHRD_IPDDP, ARPHRD_IPGRE, | |
300 | ARPHRD_PIMREG, ARPHRD_HIPPI, ARPHRD_ASH, ARPHRD_ECONET, | |
301 | ARPHRD_IRDA, ARPHRD_FCPP, ARPHRD_FCAL, ARPHRD_FCPL, | |
211ed865 PG |
302 | ARPHRD_FCFABRIC, ARPHRD_IEEE80211, ARPHRD_IEEE80211_PRISM, |
303 | ARPHRD_IEEE80211_RADIOTAP, ARPHRD_PHONET, ARPHRD_PHONET_PIPE, | |
304 | ARPHRD_IEEE802154, ARPHRD_VOID, ARPHRD_NONE}; | |
723e98b7 | 305 | |
643aa9cb | 306 | static const char *const netdev_lock_name[] = { |
307 | "_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25", | |
308 | "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET", | |
309 | "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM", | |
310 | "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP", | |
311 | "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD", | |
312 | "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25", | |
313 | "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP", | |
314 | "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD", | |
315 | "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI", | |
316 | "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE", | |
317 | "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET", | |
318 | "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL", | |
319 | "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM", | |
320 | "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE", | |
321 | "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"}; | |
723e98b7 JP |
322 | |
323 | static struct lock_class_key netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)]; | |
cf508b12 | 324 | static struct lock_class_key netdev_addr_lock_key[ARRAY_SIZE(netdev_lock_type)]; |
723e98b7 JP |
325 | |
326 | static inline unsigned short netdev_lock_pos(unsigned short dev_type) | |
327 | { | |
328 | int i; | |
329 | ||
330 | for (i = 0; i < ARRAY_SIZE(netdev_lock_type); i++) | |
331 | if (netdev_lock_type[i] == dev_type) | |
332 | return i; | |
333 | /* the last key is used by default */ | |
334 | return ARRAY_SIZE(netdev_lock_type) - 1; | |
335 | } | |
336 | ||
cf508b12 DM |
337 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
338 | unsigned short dev_type) | |
723e98b7 JP |
339 | { |
340 | int i; | |
341 | ||
342 | i = netdev_lock_pos(dev_type); | |
343 | lockdep_set_class_and_name(lock, &netdev_xmit_lock_key[i], | |
344 | netdev_lock_name[i]); | |
345 | } | |
cf508b12 DM |
346 | |
347 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
348 | { | |
349 | int i; | |
350 | ||
351 | i = netdev_lock_pos(dev->type); | |
352 | lockdep_set_class_and_name(&dev->addr_list_lock, | |
353 | &netdev_addr_lock_key[i], | |
354 | netdev_lock_name[i]); | |
355 | } | |
723e98b7 | 356 | #else |
cf508b12 DM |
357 | static inline void netdev_set_xmit_lockdep_class(spinlock_t *lock, |
358 | unsigned short dev_type) | |
359 | { | |
360 | } | |
361 | static inline void netdev_set_addr_lockdep_class(struct net_device *dev) | |
723e98b7 JP |
362 | { |
363 | } | |
364 | #endif | |
1da177e4 LT |
365 | |
366 | /******************************************************************************* | |
eb13da1a | 367 | * |
368 | * Protocol management and registration routines | |
369 | * | |
370 | *******************************************************************************/ | |
1da177e4 | 371 | |
1da177e4 | 372 | |
1da177e4 LT |
373 | /* |
374 | * Add a protocol ID to the list. Now that the input handler is | |
375 | * smarter we can dispense with all the messy stuff that used to be | |
376 | * here. | |
377 | * | |
378 | * BEWARE!!! Protocol handlers, mangling input packets, | |
379 | * MUST BE last in hash buckets and checking protocol handlers | |
380 | * MUST start from promiscuous ptype_all chain in net_bh. | |
381 | * It is true now, do not change it. | |
382 | * Explanation follows: if protocol handler, mangling packet, will | |
383 | * be the first on list, it is not able to sense, that packet | |
384 | * is cloned and should be copied-on-write, so that it will | |
385 | * change it and subsequent readers will get broken packet. | |
386 | * --ANK (980803) | |
387 | */ | |
388 | ||
c07b68e8 ED |
389 | static inline struct list_head *ptype_head(const struct packet_type *pt) |
390 | { | |
391 | if (pt->type == htons(ETH_P_ALL)) | |
7866a621 | 392 | return pt->dev ? &pt->dev->ptype_all : &ptype_all; |
c07b68e8 | 393 | else |
7866a621 SN |
394 | return pt->dev ? &pt->dev->ptype_specific : |
395 | &ptype_base[ntohs(pt->type) & PTYPE_HASH_MASK]; | |
c07b68e8 ED |
396 | } |
397 | ||
1da177e4 LT |
398 | /** |
399 | * dev_add_pack - add packet handler | |
400 | * @pt: packet type declaration | |
401 | * | |
402 | * Add a protocol handler to the networking stack. The passed &packet_type | |
403 | * is linked into kernel lists and may not be freed until it has been | |
404 | * removed from the kernel lists. | |
405 | * | |
4ec93edb | 406 | * This call does not sleep therefore it can not |
1da177e4 LT |
407 | * guarantee all CPU's that are in middle of receiving packets |
408 | * will see the new packet type (until the next received packet). | |
409 | */ | |
410 | ||
411 | void dev_add_pack(struct packet_type *pt) | |
412 | { | |
c07b68e8 | 413 | struct list_head *head = ptype_head(pt); |
1da177e4 | 414 | |
c07b68e8 ED |
415 | spin_lock(&ptype_lock); |
416 | list_add_rcu(&pt->list, head); | |
417 | spin_unlock(&ptype_lock); | |
1da177e4 | 418 | } |
d1b19dff | 419 | EXPORT_SYMBOL(dev_add_pack); |
1da177e4 | 420 | |
1da177e4 LT |
421 | /** |
422 | * __dev_remove_pack - remove packet handler | |
423 | * @pt: packet type declaration | |
424 | * | |
425 | * Remove a protocol handler that was previously added to the kernel | |
426 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
427 | * from the kernel lists and can be freed or reused once this function | |
4ec93edb | 428 | * returns. |
1da177e4 LT |
429 | * |
430 | * The packet type might still be in use by receivers | |
431 | * and must not be freed until after all the CPU's have gone | |
432 | * through a quiescent state. | |
433 | */ | |
434 | void __dev_remove_pack(struct packet_type *pt) | |
435 | { | |
c07b68e8 | 436 | struct list_head *head = ptype_head(pt); |
1da177e4 LT |
437 | struct packet_type *pt1; |
438 | ||
c07b68e8 | 439 | spin_lock(&ptype_lock); |
1da177e4 LT |
440 | |
441 | list_for_each_entry(pt1, head, list) { | |
442 | if (pt == pt1) { | |
443 | list_del_rcu(&pt->list); | |
444 | goto out; | |
445 | } | |
446 | } | |
447 | ||
7b6cd1ce | 448 | pr_warn("dev_remove_pack: %p not found\n", pt); |
1da177e4 | 449 | out: |
c07b68e8 | 450 | spin_unlock(&ptype_lock); |
1da177e4 | 451 | } |
d1b19dff ED |
452 | EXPORT_SYMBOL(__dev_remove_pack); |
453 | ||
1da177e4 LT |
454 | /** |
455 | * dev_remove_pack - remove packet handler | |
456 | * @pt: packet type declaration | |
457 | * | |
458 | * Remove a protocol handler that was previously added to the kernel | |
459 | * protocol handlers by dev_add_pack(). The passed &packet_type is removed | |
460 | * from the kernel lists and can be freed or reused once this function | |
461 | * returns. | |
462 | * | |
463 | * This call sleeps to guarantee that no CPU is looking at the packet | |
464 | * type after return. | |
465 | */ | |
466 | void dev_remove_pack(struct packet_type *pt) | |
467 | { | |
468 | __dev_remove_pack(pt); | |
4ec93edb | 469 | |
1da177e4 LT |
470 | synchronize_net(); |
471 | } | |
d1b19dff | 472 | EXPORT_SYMBOL(dev_remove_pack); |
1da177e4 | 473 | |
62532da9 VY |
474 | |
475 | /** | |
476 | * dev_add_offload - register offload handlers | |
477 | * @po: protocol offload declaration | |
478 | * | |
479 | * Add protocol offload handlers to the networking stack. The passed | |
480 | * &proto_offload is linked into kernel lists and may not be freed until | |
481 | * it has been removed from the kernel lists. | |
482 | * | |
483 | * This call does not sleep therefore it can not | |
484 | * guarantee all CPU's that are in middle of receiving packets | |
485 | * will see the new offload handlers (until the next received packet). | |
486 | */ | |
487 | void dev_add_offload(struct packet_offload *po) | |
488 | { | |
bdef7de4 | 489 | struct packet_offload *elem; |
62532da9 VY |
490 | |
491 | spin_lock(&offload_lock); | |
bdef7de4 DM |
492 | list_for_each_entry(elem, &offload_base, list) { |
493 | if (po->priority < elem->priority) | |
494 | break; | |
495 | } | |
496 | list_add_rcu(&po->list, elem->list.prev); | |
62532da9 VY |
497 | spin_unlock(&offload_lock); |
498 | } | |
499 | EXPORT_SYMBOL(dev_add_offload); | |
500 | ||
501 | /** | |
502 | * __dev_remove_offload - remove offload handler | |
503 | * @po: packet offload declaration | |
504 | * | |
505 | * Remove a protocol offload handler that was previously added to the | |
506 | * kernel offload handlers by dev_add_offload(). The passed &offload_type | |
507 | * is removed from the kernel lists and can be freed or reused once this | |
508 | * function returns. | |
509 | * | |
510 | * The packet type might still be in use by receivers | |
511 | * and must not be freed until after all the CPU's have gone | |
512 | * through a quiescent state. | |
513 | */ | |
1d143d9f | 514 | static void __dev_remove_offload(struct packet_offload *po) |
62532da9 VY |
515 | { |
516 | struct list_head *head = &offload_base; | |
517 | struct packet_offload *po1; | |
518 | ||
c53aa505 | 519 | spin_lock(&offload_lock); |
62532da9 VY |
520 | |
521 | list_for_each_entry(po1, head, list) { | |
522 | if (po == po1) { | |
523 | list_del_rcu(&po->list); | |
524 | goto out; | |
525 | } | |
526 | } | |
527 | ||
528 | pr_warn("dev_remove_offload: %p not found\n", po); | |
529 | out: | |
c53aa505 | 530 | spin_unlock(&offload_lock); |
62532da9 | 531 | } |
62532da9 VY |
532 | |
533 | /** | |
534 | * dev_remove_offload - remove packet offload handler | |
535 | * @po: packet offload declaration | |
536 | * | |
537 | * Remove a packet offload handler that was previously added to the kernel | |
538 | * offload handlers by dev_add_offload(). The passed &offload_type is | |
539 | * removed from the kernel lists and can be freed or reused once this | |
540 | * function returns. | |
541 | * | |
542 | * This call sleeps to guarantee that no CPU is looking at the packet | |
543 | * type after return. | |
544 | */ | |
545 | void dev_remove_offload(struct packet_offload *po) | |
546 | { | |
547 | __dev_remove_offload(po); | |
548 | ||
549 | synchronize_net(); | |
550 | } | |
551 | EXPORT_SYMBOL(dev_remove_offload); | |
552 | ||
1da177e4 | 553 | /****************************************************************************** |
eb13da1a | 554 | * |
555 | * Device Boot-time Settings Routines | |
556 | * | |
557 | ******************************************************************************/ | |
1da177e4 LT |
558 | |
559 | /* Boot time configuration table */ | |
560 | static struct netdev_boot_setup dev_boot_setup[NETDEV_BOOT_SETUP_MAX]; | |
561 | ||
562 | /** | |
563 | * netdev_boot_setup_add - add new setup entry | |
564 | * @name: name of the device | |
565 | * @map: configured settings for the device | |
566 | * | |
567 | * Adds new setup entry to the dev_boot_setup list. The function | |
568 | * returns 0 on error and 1 on success. This is a generic routine to | |
569 | * all netdevices. | |
570 | */ | |
571 | static int netdev_boot_setup_add(char *name, struct ifmap *map) | |
572 | { | |
573 | struct netdev_boot_setup *s; | |
574 | int i; | |
575 | ||
576 | s = dev_boot_setup; | |
577 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
578 | if (s[i].name[0] == '\0' || s[i].name[0] == ' ') { | |
579 | memset(s[i].name, 0, sizeof(s[i].name)); | |
93b3cff9 | 580 | strlcpy(s[i].name, name, IFNAMSIZ); |
1da177e4 LT |
581 | memcpy(&s[i].map, map, sizeof(s[i].map)); |
582 | break; | |
583 | } | |
584 | } | |
585 | ||
586 | return i >= NETDEV_BOOT_SETUP_MAX ? 0 : 1; | |
587 | } | |
588 | ||
589 | /** | |
722c9a0c | 590 | * netdev_boot_setup_check - check boot time settings |
591 | * @dev: the netdevice | |
1da177e4 | 592 | * |
722c9a0c | 593 | * Check boot time settings for the device. |
594 | * The found settings are set for the device to be used | |
595 | * later in the device probing. | |
596 | * Returns 0 if no settings found, 1 if they are. | |
1da177e4 LT |
597 | */ |
598 | int netdev_boot_setup_check(struct net_device *dev) | |
599 | { | |
600 | struct netdev_boot_setup *s = dev_boot_setup; | |
601 | int i; | |
602 | ||
603 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) { | |
604 | if (s[i].name[0] != '\0' && s[i].name[0] != ' ' && | |
93b3cff9 | 605 | !strcmp(dev->name, s[i].name)) { |
722c9a0c | 606 | dev->irq = s[i].map.irq; |
607 | dev->base_addr = s[i].map.base_addr; | |
608 | dev->mem_start = s[i].map.mem_start; | |
609 | dev->mem_end = s[i].map.mem_end; | |
1da177e4 LT |
610 | return 1; |
611 | } | |
612 | } | |
613 | return 0; | |
614 | } | |
d1b19dff | 615 | EXPORT_SYMBOL(netdev_boot_setup_check); |
1da177e4 LT |
616 | |
617 | ||
618 | /** | |
722c9a0c | 619 | * netdev_boot_base - get address from boot time settings |
620 | * @prefix: prefix for network device | |
621 | * @unit: id for network device | |
622 | * | |
623 | * Check boot time settings for the base address of device. | |
624 | * The found settings are set for the device to be used | |
625 | * later in the device probing. | |
626 | * Returns 0 if no settings found. | |
1da177e4 LT |
627 | */ |
628 | unsigned long netdev_boot_base(const char *prefix, int unit) | |
629 | { | |
630 | const struct netdev_boot_setup *s = dev_boot_setup; | |
631 | char name[IFNAMSIZ]; | |
632 | int i; | |
633 | ||
634 | sprintf(name, "%s%d", prefix, unit); | |
635 | ||
636 | /* | |
637 | * If device already registered then return base of 1 | |
638 | * to indicate not to probe for this interface | |
639 | */ | |
881d966b | 640 | if (__dev_get_by_name(&init_net, name)) |
1da177e4 LT |
641 | return 1; |
642 | ||
643 | for (i = 0; i < NETDEV_BOOT_SETUP_MAX; i++) | |
644 | if (!strcmp(name, s[i].name)) | |
645 | return s[i].map.base_addr; | |
646 | return 0; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Saves at boot time configured settings for any netdevice. | |
651 | */ | |
652 | int __init netdev_boot_setup(char *str) | |
653 | { | |
654 | int ints[5]; | |
655 | struct ifmap map; | |
656 | ||
657 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
658 | if (!str || !*str) | |
659 | return 0; | |
660 | ||
661 | /* Save settings */ | |
662 | memset(&map, 0, sizeof(map)); | |
663 | if (ints[0] > 0) | |
664 | map.irq = ints[1]; | |
665 | if (ints[0] > 1) | |
666 | map.base_addr = ints[2]; | |
667 | if (ints[0] > 2) | |
668 | map.mem_start = ints[3]; | |
669 | if (ints[0] > 3) | |
670 | map.mem_end = ints[4]; | |
671 | ||
672 | /* Add new entry to the list */ | |
673 | return netdev_boot_setup_add(str, &map); | |
674 | } | |
675 | ||
676 | __setup("netdev=", netdev_boot_setup); | |
677 | ||
678 | /******************************************************************************* | |
eb13da1a | 679 | * |
680 | * Device Interface Subroutines | |
681 | * | |
682 | *******************************************************************************/ | |
1da177e4 | 683 | |
a54acb3a ND |
684 | /** |
685 | * dev_get_iflink - get 'iflink' value of a interface | |
686 | * @dev: targeted interface | |
687 | * | |
688 | * Indicates the ifindex the interface is linked to. | |
689 | * Physical interfaces have the same 'ifindex' and 'iflink' values. | |
690 | */ | |
691 | ||
692 | int dev_get_iflink(const struct net_device *dev) | |
693 | { | |
694 | if (dev->netdev_ops && dev->netdev_ops->ndo_get_iflink) | |
695 | return dev->netdev_ops->ndo_get_iflink(dev); | |
696 | ||
7a66bbc9 | 697 | return dev->ifindex; |
a54acb3a ND |
698 | } |
699 | EXPORT_SYMBOL(dev_get_iflink); | |
700 | ||
fc4099f1 PS |
701 | /** |
702 | * dev_fill_metadata_dst - Retrieve tunnel egress information. | |
703 | * @dev: targeted interface | |
704 | * @skb: The packet. | |
705 | * | |
706 | * For better visibility of tunnel traffic OVS needs to retrieve | |
707 | * egress tunnel information for a packet. Following API allows | |
708 | * user to get this info. | |
709 | */ | |
710 | int dev_fill_metadata_dst(struct net_device *dev, struct sk_buff *skb) | |
711 | { | |
712 | struct ip_tunnel_info *info; | |
713 | ||
714 | if (!dev->netdev_ops || !dev->netdev_ops->ndo_fill_metadata_dst) | |
715 | return -EINVAL; | |
716 | ||
717 | info = skb_tunnel_info_unclone(skb); | |
718 | if (!info) | |
719 | return -ENOMEM; | |
720 | if (unlikely(!(info->mode & IP_TUNNEL_INFO_TX))) | |
721 | return -EINVAL; | |
722 | ||
723 | return dev->netdev_ops->ndo_fill_metadata_dst(dev, skb); | |
724 | } | |
725 | EXPORT_SYMBOL_GPL(dev_fill_metadata_dst); | |
726 | ||
1da177e4 LT |
727 | /** |
728 | * __dev_get_by_name - find a device by its name | |
c4ea43c5 | 729 | * @net: the applicable net namespace |
1da177e4 LT |
730 | * @name: name to find |
731 | * | |
732 | * Find an interface by name. Must be called under RTNL semaphore | |
733 | * or @dev_base_lock. If the name is found a pointer to the device | |
734 | * is returned. If the name is not found then %NULL is returned. The | |
735 | * reference counters are not incremented so the caller must be | |
736 | * careful with locks. | |
737 | */ | |
738 | ||
881d966b | 739 | struct net_device *__dev_get_by_name(struct net *net, const char *name) |
1da177e4 | 740 | { |
0bd8d536 ED |
741 | struct net_device *dev; |
742 | struct hlist_head *head = dev_name_hash(net, name); | |
1da177e4 | 743 | |
b67bfe0d | 744 | hlist_for_each_entry(dev, head, name_hlist) |
1da177e4 LT |
745 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
746 | return dev; | |
0bd8d536 | 747 | |
1da177e4 LT |
748 | return NULL; |
749 | } | |
d1b19dff | 750 | EXPORT_SYMBOL(__dev_get_by_name); |
1da177e4 | 751 | |
72c9528b | 752 | /** |
722c9a0c | 753 | * dev_get_by_name_rcu - find a device by its name |
754 | * @net: the applicable net namespace | |
755 | * @name: name to find | |
756 | * | |
757 | * Find an interface by name. | |
758 | * If the name is found a pointer to the device is returned. | |
759 | * If the name is not found then %NULL is returned. | |
760 | * The reference counters are not incremented so the caller must be | |
761 | * careful with locks. The caller must hold RCU lock. | |
72c9528b ED |
762 | */ |
763 | ||
764 | struct net_device *dev_get_by_name_rcu(struct net *net, const char *name) | |
765 | { | |
72c9528b ED |
766 | struct net_device *dev; |
767 | struct hlist_head *head = dev_name_hash(net, name); | |
768 | ||
b67bfe0d | 769 | hlist_for_each_entry_rcu(dev, head, name_hlist) |
72c9528b ED |
770 | if (!strncmp(dev->name, name, IFNAMSIZ)) |
771 | return dev; | |
772 | ||
773 | return NULL; | |
774 | } | |
775 | EXPORT_SYMBOL(dev_get_by_name_rcu); | |
776 | ||
1da177e4 LT |
777 | /** |
778 | * dev_get_by_name - find a device by its name | |
c4ea43c5 | 779 | * @net: the applicable net namespace |
1da177e4 LT |
780 | * @name: name to find |
781 | * | |
782 | * Find an interface by name. This can be called from any | |
783 | * context and does its own locking. The returned handle has | |
784 | * the usage count incremented and the caller must use dev_put() to | |
785 | * release it when it is no longer needed. %NULL is returned if no | |
786 | * matching device is found. | |
787 | */ | |
788 | ||
881d966b | 789 | struct net_device *dev_get_by_name(struct net *net, const char *name) |
1da177e4 LT |
790 | { |
791 | struct net_device *dev; | |
792 | ||
72c9528b ED |
793 | rcu_read_lock(); |
794 | dev = dev_get_by_name_rcu(net, name); | |
1da177e4 LT |
795 | if (dev) |
796 | dev_hold(dev); | |
72c9528b | 797 | rcu_read_unlock(); |
1da177e4 LT |
798 | return dev; |
799 | } | |
d1b19dff | 800 | EXPORT_SYMBOL(dev_get_by_name); |
1da177e4 LT |
801 | |
802 | /** | |
803 | * __dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 804 | * @net: the applicable net namespace |
1da177e4 LT |
805 | * @ifindex: index of device |
806 | * | |
807 | * Search for an interface by index. Returns %NULL if the device | |
808 | * is not found or a pointer to the device. The device has not | |
809 | * had its reference counter increased so the caller must be careful | |
810 | * about locking. The caller must hold either the RTNL semaphore | |
811 | * or @dev_base_lock. | |
812 | */ | |
813 | ||
881d966b | 814 | struct net_device *__dev_get_by_index(struct net *net, int ifindex) |
1da177e4 | 815 | { |
0bd8d536 ED |
816 | struct net_device *dev; |
817 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
1da177e4 | 818 | |
b67bfe0d | 819 | hlist_for_each_entry(dev, head, index_hlist) |
1da177e4 LT |
820 | if (dev->ifindex == ifindex) |
821 | return dev; | |
0bd8d536 | 822 | |
1da177e4 LT |
823 | return NULL; |
824 | } | |
d1b19dff | 825 | EXPORT_SYMBOL(__dev_get_by_index); |
1da177e4 | 826 | |
fb699dfd ED |
827 | /** |
828 | * dev_get_by_index_rcu - find a device by its ifindex | |
829 | * @net: the applicable net namespace | |
830 | * @ifindex: index of device | |
831 | * | |
832 | * Search for an interface by index. Returns %NULL if the device | |
833 | * is not found or a pointer to the device. The device has not | |
834 | * had its reference counter increased so the caller must be careful | |
835 | * about locking. The caller must hold RCU lock. | |
836 | */ | |
837 | ||
838 | struct net_device *dev_get_by_index_rcu(struct net *net, int ifindex) | |
839 | { | |
fb699dfd ED |
840 | struct net_device *dev; |
841 | struct hlist_head *head = dev_index_hash(net, ifindex); | |
842 | ||
b67bfe0d | 843 | hlist_for_each_entry_rcu(dev, head, index_hlist) |
fb699dfd ED |
844 | if (dev->ifindex == ifindex) |
845 | return dev; | |
846 | ||
847 | return NULL; | |
848 | } | |
849 | EXPORT_SYMBOL(dev_get_by_index_rcu); | |
850 | ||
1da177e4 LT |
851 | |
852 | /** | |
853 | * dev_get_by_index - find a device by its ifindex | |
c4ea43c5 | 854 | * @net: the applicable net namespace |
1da177e4 LT |
855 | * @ifindex: index of device |
856 | * | |
857 | * Search for an interface by index. Returns NULL if the device | |
858 | * is not found or a pointer to the device. The device returned has | |
859 | * had a reference added and the pointer is safe until the user calls | |
860 | * dev_put to indicate they have finished with it. | |
861 | */ | |
862 | ||
881d966b | 863 | struct net_device *dev_get_by_index(struct net *net, int ifindex) |
1da177e4 LT |
864 | { |
865 | struct net_device *dev; | |
866 | ||
fb699dfd ED |
867 | rcu_read_lock(); |
868 | dev = dev_get_by_index_rcu(net, ifindex); | |
1da177e4 LT |
869 | if (dev) |
870 | dev_hold(dev); | |
fb699dfd | 871 | rcu_read_unlock(); |
1da177e4 LT |
872 | return dev; |
873 | } | |
d1b19dff | 874 | EXPORT_SYMBOL(dev_get_by_index); |
1da177e4 | 875 | |
90b602f8 ML |
876 | /** |
877 | * dev_get_by_napi_id - find a device by napi_id | |
878 | * @napi_id: ID of the NAPI struct | |
879 | * | |
880 | * Search for an interface by NAPI ID. Returns %NULL if the device | |
881 | * is not found or a pointer to the device. The device has not had | |
882 | * its reference counter increased so the caller must be careful | |
883 | * about locking. The caller must hold RCU lock. | |
884 | */ | |
885 | ||
886 | struct net_device *dev_get_by_napi_id(unsigned int napi_id) | |
887 | { | |
888 | struct napi_struct *napi; | |
889 | ||
890 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
891 | ||
892 | if (napi_id < MIN_NAPI_ID) | |
893 | return NULL; | |
894 | ||
895 | napi = napi_by_id(napi_id); | |
896 | ||
897 | return napi ? napi->dev : NULL; | |
898 | } | |
899 | EXPORT_SYMBOL(dev_get_by_napi_id); | |
900 | ||
5dbe7c17 NS |
901 | /** |
902 | * netdev_get_name - get a netdevice name, knowing its ifindex. | |
903 | * @net: network namespace | |
904 | * @name: a pointer to the buffer where the name will be stored. | |
905 | * @ifindex: the ifindex of the interface to get the name from. | |
906 | * | |
907 | * The use of raw_seqcount_begin() and cond_resched() before | |
908 | * retrying is required as we want to give the writers a chance | |
909 | * to complete when CONFIG_PREEMPT is not set. | |
910 | */ | |
911 | int netdev_get_name(struct net *net, char *name, int ifindex) | |
912 | { | |
913 | struct net_device *dev; | |
914 | unsigned int seq; | |
915 | ||
916 | retry: | |
917 | seq = raw_seqcount_begin(&devnet_rename_seq); | |
918 | rcu_read_lock(); | |
919 | dev = dev_get_by_index_rcu(net, ifindex); | |
920 | if (!dev) { | |
921 | rcu_read_unlock(); | |
922 | return -ENODEV; | |
923 | } | |
924 | ||
925 | strcpy(name, dev->name); | |
926 | rcu_read_unlock(); | |
927 | if (read_seqcount_retry(&devnet_rename_seq, seq)) { | |
928 | cond_resched(); | |
929 | goto retry; | |
930 | } | |
931 | ||
932 | return 0; | |
933 | } | |
934 | ||
1da177e4 | 935 | /** |
941666c2 | 936 | * dev_getbyhwaddr_rcu - find a device by its hardware address |
c4ea43c5 | 937 | * @net: the applicable net namespace |
1da177e4 LT |
938 | * @type: media type of device |
939 | * @ha: hardware address | |
940 | * | |
941 | * Search for an interface by MAC address. Returns NULL if the device | |
c506653d ED |
942 | * is not found or a pointer to the device. |
943 | * The caller must hold RCU or RTNL. | |
941666c2 | 944 | * The returned device has not had its ref count increased |
1da177e4 LT |
945 | * and the caller must therefore be careful about locking |
946 | * | |
1da177e4 LT |
947 | */ |
948 | ||
941666c2 ED |
949 | struct net_device *dev_getbyhwaddr_rcu(struct net *net, unsigned short type, |
950 | const char *ha) | |
1da177e4 LT |
951 | { |
952 | struct net_device *dev; | |
953 | ||
941666c2 | 954 | for_each_netdev_rcu(net, dev) |
1da177e4 LT |
955 | if (dev->type == type && |
956 | !memcmp(dev->dev_addr, ha, dev->addr_len)) | |
7562f876 PE |
957 | return dev; |
958 | ||
959 | return NULL; | |
1da177e4 | 960 | } |
941666c2 | 961 | EXPORT_SYMBOL(dev_getbyhwaddr_rcu); |
cf309e3f | 962 | |
881d966b | 963 | struct net_device *__dev_getfirstbyhwtype(struct net *net, unsigned short type) |
1da177e4 LT |
964 | { |
965 | struct net_device *dev; | |
966 | ||
4e9cac2b | 967 | ASSERT_RTNL(); |
881d966b | 968 | for_each_netdev(net, dev) |
4e9cac2b | 969 | if (dev->type == type) |
7562f876 PE |
970 | return dev; |
971 | ||
972 | return NULL; | |
4e9cac2b | 973 | } |
4e9cac2b PM |
974 | EXPORT_SYMBOL(__dev_getfirstbyhwtype); |
975 | ||
881d966b | 976 | struct net_device *dev_getfirstbyhwtype(struct net *net, unsigned short type) |
4e9cac2b | 977 | { |
99fe3c39 | 978 | struct net_device *dev, *ret = NULL; |
4e9cac2b | 979 | |
99fe3c39 ED |
980 | rcu_read_lock(); |
981 | for_each_netdev_rcu(net, dev) | |
982 | if (dev->type == type) { | |
983 | dev_hold(dev); | |
984 | ret = dev; | |
985 | break; | |
986 | } | |
987 | rcu_read_unlock(); | |
988 | return ret; | |
1da177e4 | 989 | } |
1da177e4 LT |
990 | EXPORT_SYMBOL(dev_getfirstbyhwtype); |
991 | ||
992 | /** | |
6c555490 | 993 | * __dev_get_by_flags - find any device with given flags |
c4ea43c5 | 994 | * @net: the applicable net namespace |
1da177e4 LT |
995 | * @if_flags: IFF_* values |
996 | * @mask: bitmask of bits in if_flags to check | |
997 | * | |
998 | * Search for any interface with the given flags. Returns NULL if a device | |
bb69ae04 | 999 | * is not found or a pointer to the device. Must be called inside |
6c555490 | 1000 | * rtnl_lock(), and result refcount is unchanged. |
1da177e4 LT |
1001 | */ |
1002 | ||
6c555490 WC |
1003 | struct net_device *__dev_get_by_flags(struct net *net, unsigned short if_flags, |
1004 | unsigned short mask) | |
1da177e4 | 1005 | { |
7562f876 | 1006 | struct net_device *dev, *ret; |
1da177e4 | 1007 | |
6c555490 WC |
1008 | ASSERT_RTNL(); |
1009 | ||
7562f876 | 1010 | ret = NULL; |
6c555490 | 1011 | for_each_netdev(net, dev) { |
1da177e4 | 1012 | if (((dev->flags ^ if_flags) & mask) == 0) { |
7562f876 | 1013 | ret = dev; |
1da177e4 LT |
1014 | break; |
1015 | } | |
1016 | } | |
7562f876 | 1017 | return ret; |
1da177e4 | 1018 | } |
6c555490 | 1019 | EXPORT_SYMBOL(__dev_get_by_flags); |
1da177e4 LT |
1020 | |
1021 | /** | |
1022 | * dev_valid_name - check if name is okay for network device | |
1023 | * @name: name string | |
1024 | * | |
1025 | * Network device names need to be valid file names to | |
c7fa9d18 DM |
1026 | * to allow sysfs to work. We also disallow any kind of |
1027 | * whitespace. | |
1da177e4 | 1028 | */ |
95f050bf | 1029 | bool dev_valid_name(const char *name) |
1da177e4 | 1030 | { |
c7fa9d18 | 1031 | if (*name == '\0') |
95f050bf | 1032 | return false; |
a9d48205 | 1033 | if (strnlen(name, IFNAMSIZ) == IFNAMSIZ) |
95f050bf | 1034 | return false; |
c7fa9d18 | 1035 | if (!strcmp(name, ".") || !strcmp(name, "..")) |
95f050bf | 1036 | return false; |
c7fa9d18 DM |
1037 | |
1038 | while (*name) { | |
a4176a93 | 1039 | if (*name == '/' || *name == ':' || isspace(*name)) |
95f050bf | 1040 | return false; |
c7fa9d18 DM |
1041 | name++; |
1042 | } | |
95f050bf | 1043 | return true; |
1da177e4 | 1044 | } |
d1b19dff | 1045 | EXPORT_SYMBOL(dev_valid_name); |
1da177e4 LT |
1046 | |
1047 | /** | |
b267b179 EB |
1048 | * __dev_alloc_name - allocate a name for a device |
1049 | * @net: network namespace to allocate the device name in | |
1da177e4 | 1050 | * @name: name format string |
b267b179 | 1051 | * @buf: scratch buffer and result name string |
1da177e4 LT |
1052 | * |
1053 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
3041a069 SH |
1054 | * id. It scans list of devices to build up a free map, then chooses |
1055 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1056 | * while allocating the name and adding the device in order to avoid | |
1057 | * duplicates. | |
1058 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1059 | * Returns the number of the unit assigned or a negative errno code. | |
1da177e4 LT |
1060 | */ |
1061 | ||
b267b179 | 1062 | static int __dev_alloc_name(struct net *net, const char *name, char *buf) |
1da177e4 LT |
1063 | { |
1064 | int i = 0; | |
1da177e4 LT |
1065 | const char *p; |
1066 | const int max_netdevices = 8*PAGE_SIZE; | |
cfcabdcc | 1067 | unsigned long *inuse; |
1da177e4 LT |
1068 | struct net_device *d; |
1069 | ||
93809105 RV |
1070 | if (!dev_valid_name(name)) |
1071 | return -EINVAL; | |
1072 | ||
51f299dd | 1073 | p = strchr(name, '%'); |
1da177e4 LT |
1074 | if (p) { |
1075 | /* | |
1076 | * Verify the string as this thing may have come from | |
1077 | * the user. There must be either one "%d" and no other "%" | |
1078 | * characters. | |
1079 | */ | |
1080 | if (p[1] != 'd' || strchr(p + 2, '%')) | |
1081 | return -EINVAL; | |
1082 | ||
1083 | /* Use one page as a bit array of possible slots */ | |
cfcabdcc | 1084 | inuse = (unsigned long *) get_zeroed_page(GFP_ATOMIC); |
1da177e4 LT |
1085 | if (!inuse) |
1086 | return -ENOMEM; | |
1087 | ||
881d966b | 1088 | for_each_netdev(net, d) { |
1da177e4 LT |
1089 | if (!sscanf(d->name, name, &i)) |
1090 | continue; | |
1091 | if (i < 0 || i >= max_netdevices) | |
1092 | continue; | |
1093 | ||
1094 | /* avoid cases where sscanf is not exact inverse of printf */ | |
b267b179 | 1095 | snprintf(buf, IFNAMSIZ, name, i); |
1da177e4 LT |
1096 | if (!strncmp(buf, d->name, IFNAMSIZ)) |
1097 | set_bit(i, inuse); | |
1098 | } | |
1099 | ||
1100 | i = find_first_zero_bit(inuse, max_netdevices); | |
1101 | free_page((unsigned long) inuse); | |
1102 | } | |
1103 | ||
6224abda | 1104 | snprintf(buf, IFNAMSIZ, name, i); |
b267b179 | 1105 | if (!__dev_get_by_name(net, buf)) |
1da177e4 | 1106 | return i; |
1da177e4 LT |
1107 | |
1108 | /* It is possible to run out of possible slots | |
1109 | * when the name is long and there isn't enough space left | |
1110 | * for the digits, or if all bits are used. | |
1111 | */ | |
029b6d14 | 1112 | return -ENFILE; |
1da177e4 LT |
1113 | } |
1114 | ||
2c88b855 RV |
1115 | static int dev_alloc_name_ns(struct net *net, |
1116 | struct net_device *dev, | |
1117 | const char *name) | |
1118 | { | |
1119 | char buf[IFNAMSIZ]; | |
1120 | int ret; | |
1121 | ||
c46d7642 | 1122 | BUG_ON(!net); |
2c88b855 RV |
1123 | ret = __dev_alloc_name(net, name, buf); |
1124 | if (ret >= 0) | |
1125 | strlcpy(dev->name, buf, IFNAMSIZ); | |
1126 | return ret; | |
1da177e4 LT |
1127 | } |
1128 | ||
b267b179 EB |
1129 | /** |
1130 | * dev_alloc_name - allocate a name for a device | |
1131 | * @dev: device | |
1132 | * @name: name format string | |
1133 | * | |
1134 | * Passed a format string - eg "lt%d" it will try and find a suitable | |
1135 | * id. It scans list of devices to build up a free map, then chooses | |
1136 | * the first empty slot. The caller must hold the dev_base or rtnl lock | |
1137 | * while allocating the name and adding the device in order to avoid | |
1138 | * duplicates. | |
1139 | * Limited to bits_per_byte * page size devices (ie 32K on most platforms). | |
1140 | * Returns the number of the unit assigned or a negative errno code. | |
1141 | */ | |
1142 | ||
1143 | int dev_alloc_name(struct net_device *dev, const char *name) | |
1144 | { | |
c46d7642 | 1145 | return dev_alloc_name_ns(dev_net(dev), dev, name); |
b267b179 | 1146 | } |
d1b19dff | 1147 | EXPORT_SYMBOL(dev_alloc_name); |
b267b179 | 1148 | |
0ad646c8 CW |
1149 | int dev_get_valid_name(struct net *net, struct net_device *dev, |
1150 | const char *name) | |
828de4f6 | 1151 | { |
55a5ec9b DM |
1152 | BUG_ON(!net); |
1153 | ||
1154 | if (!dev_valid_name(name)) | |
1155 | return -EINVAL; | |
1156 | ||
1157 | if (strchr(name, '%')) | |
1158 | return dev_alloc_name_ns(net, dev, name); | |
1159 | else if (__dev_get_by_name(net, name)) | |
1160 | return -EEXIST; | |
1161 | else if (dev->name != name) | |
1162 | strlcpy(dev->name, name, IFNAMSIZ); | |
1163 | ||
1164 | return 0; | |
d9031024 | 1165 | } |
0ad646c8 | 1166 | EXPORT_SYMBOL(dev_get_valid_name); |
1da177e4 LT |
1167 | |
1168 | /** | |
1169 | * dev_change_name - change name of a device | |
1170 | * @dev: device | |
1171 | * @newname: name (or format string) must be at least IFNAMSIZ | |
1172 | * | |
1173 | * Change name of a device, can pass format strings "eth%d". | |
1174 | * for wildcarding. | |
1175 | */ | |
cf04a4c7 | 1176 | int dev_change_name(struct net_device *dev, const char *newname) |
1da177e4 | 1177 | { |
238fa362 | 1178 | unsigned char old_assign_type; |
fcc5a03a | 1179 | char oldname[IFNAMSIZ]; |
1da177e4 | 1180 | int err = 0; |
fcc5a03a | 1181 | int ret; |
881d966b | 1182 | struct net *net; |
1da177e4 LT |
1183 | |
1184 | ASSERT_RTNL(); | |
c346dca1 | 1185 | BUG_ON(!dev_net(dev)); |
1da177e4 | 1186 | |
c346dca1 | 1187 | net = dev_net(dev); |
1da177e4 LT |
1188 | if (dev->flags & IFF_UP) |
1189 | return -EBUSY; | |
1190 | ||
30e6c9fa | 1191 | write_seqcount_begin(&devnet_rename_seq); |
c91f6df2 BH |
1192 | |
1193 | if (strncmp(newname, dev->name, IFNAMSIZ) == 0) { | |
30e6c9fa | 1194 | write_seqcount_end(&devnet_rename_seq); |
c8d90dca | 1195 | return 0; |
c91f6df2 | 1196 | } |
c8d90dca | 1197 | |
fcc5a03a HX |
1198 | memcpy(oldname, dev->name, IFNAMSIZ); |
1199 | ||
828de4f6 | 1200 | err = dev_get_valid_name(net, dev, newname); |
c91f6df2 | 1201 | if (err < 0) { |
30e6c9fa | 1202 | write_seqcount_end(&devnet_rename_seq); |
d9031024 | 1203 | return err; |
c91f6df2 | 1204 | } |
1da177e4 | 1205 | |
6fe82a39 VF |
1206 | if (oldname[0] && !strchr(oldname, '%')) |
1207 | netdev_info(dev, "renamed from %s\n", oldname); | |
1208 | ||
238fa362 TG |
1209 | old_assign_type = dev->name_assign_type; |
1210 | dev->name_assign_type = NET_NAME_RENAMED; | |
1211 | ||
fcc5a03a | 1212 | rollback: |
a1b3f594 EB |
1213 | ret = device_rename(&dev->dev, dev->name); |
1214 | if (ret) { | |
1215 | memcpy(dev->name, oldname, IFNAMSIZ); | |
238fa362 | 1216 | dev->name_assign_type = old_assign_type; |
30e6c9fa | 1217 | write_seqcount_end(&devnet_rename_seq); |
a1b3f594 | 1218 | return ret; |
dcc99773 | 1219 | } |
7f988eab | 1220 | |
30e6c9fa | 1221 | write_seqcount_end(&devnet_rename_seq); |
c91f6df2 | 1222 | |
5bb025fa VF |
1223 | netdev_adjacent_rename_links(dev, oldname); |
1224 | ||
7f988eab | 1225 | write_lock_bh(&dev_base_lock); |
372b2312 | 1226 | hlist_del_rcu(&dev->name_hlist); |
72c9528b ED |
1227 | write_unlock_bh(&dev_base_lock); |
1228 | ||
1229 | synchronize_rcu(); | |
1230 | ||
1231 | write_lock_bh(&dev_base_lock); | |
1232 | hlist_add_head_rcu(&dev->name_hlist, dev_name_hash(net, dev->name)); | |
7f988eab HX |
1233 | write_unlock_bh(&dev_base_lock); |
1234 | ||
056925ab | 1235 | ret = call_netdevice_notifiers(NETDEV_CHANGENAME, dev); |
fcc5a03a HX |
1236 | ret = notifier_to_errno(ret); |
1237 | ||
1238 | if (ret) { | |
91e9c07b ED |
1239 | /* err >= 0 after dev_alloc_name() or stores the first errno */ |
1240 | if (err >= 0) { | |
fcc5a03a | 1241 | err = ret; |
30e6c9fa | 1242 | write_seqcount_begin(&devnet_rename_seq); |
fcc5a03a | 1243 | memcpy(dev->name, oldname, IFNAMSIZ); |
5bb025fa | 1244 | memcpy(oldname, newname, IFNAMSIZ); |
238fa362 TG |
1245 | dev->name_assign_type = old_assign_type; |
1246 | old_assign_type = NET_NAME_RENAMED; | |
fcc5a03a | 1247 | goto rollback; |
91e9c07b | 1248 | } else { |
7b6cd1ce | 1249 | pr_err("%s: name change rollback failed: %d\n", |
91e9c07b | 1250 | dev->name, ret); |
fcc5a03a HX |
1251 | } |
1252 | } | |
1da177e4 LT |
1253 | |
1254 | return err; | |
1255 | } | |
1256 | ||
0b815a1a SH |
1257 | /** |
1258 | * dev_set_alias - change ifalias of a device | |
1259 | * @dev: device | |
1260 | * @alias: name up to IFALIASZ | |
f0db275a | 1261 | * @len: limit of bytes to copy from info |
0b815a1a SH |
1262 | * |
1263 | * Set ifalias for a device, | |
1264 | */ | |
1265 | int dev_set_alias(struct net_device *dev, const char *alias, size_t len) | |
1266 | { | |
6c557001 | 1267 | struct dev_ifalias *new_alias = NULL; |
0b815a1a SH |
1268 | |
1269 | if (len >= IFALIASZ) | |
1270 | return -EINVAL; | |
1271 | ||
6c557001 FW |
1272 | if (len) { |
1273 | new_alias = kmalloc(sizeof(*new_alias) + len + 1, GFP_KERNEL); | |
1274 | if (!new_alias) | |
1275 | return -ENOMEM; | |
1276 | ||
1277 | memcpy(new_alias->ifalias, alias, len); | |
1278 | new_alias->ifalias[len] = 0; | |
96ca4a2c OH |
1279 | } |
1280 | ||
6c557001 FW |
1281 | mutex_lock(&ifalias_mutex); |
1282 | rcu_swap_protected(dev->ifalias, new_alias, | |
1283 | mutex_is_locked(&ifalias_mutex)); | |
1284 | mutex_unlock(&ifalias_mutex); | |
1285 | ||
1286 | if (new_alias) | |
1287 | kfree_rcu(new_alias, rcuhead); | |
0b815a1a | 1288 | |
0b815a1a SH |
1289 | return len; |
1290 | } | |
0fe554a4 | 1291 | EXPORT_SYMBOL(dev_set_alias); |
0b815a1a | 1292 | |
6c557001 FW |
1293 | /** |
1294 | * dev_get_alias - get ifalias of a device | |
1295 | * @dev: device | |
20e88320 | 1296 | * @name: buffer to store name of ifalias |
6c557001 FW |
1297 | * @len: size of buffer |
1298 | * | |
1299 | * get ifalias for a device. Caller must make sure dev cannot go | |
1300 | * away, e.g. rcu read lock or own a reference count to device. | |
1301 | */ | |
1302 | int dev_get_alias(const struct net_device *dev, char *name, size_t len) | |
1303 | { | |
1304 | const struct dev_ifalias *alias; | |
1305 | int ret = 0; | |
1306 | ||
1307 | rcu_read_lock(); | |
1308 | alias = rcu_dereference(dev->ifalias); | |
1309 | if (alias) | |
1310 | ret = snprintf(name, len, "%s", alias->ifalias); | |
1311 | rcu_read_unlock(); | |
1312 | ||
1313 | return ret; | |
1314 | } | |
0b815a1a | 1315 | |
d8a33ac4 | 1316 | /** |
3041a069 | 1317 | * netdev_features_change - device changes features |
d8a33ac4 SH |
1318 | * @dev: device to cause notification |
1319 | * | |
1320 | * Called to indicate a device has changed features. | |
1321 | */ | |
1322 | void netdev_features_change(struct net_device *dev) | |
1323 | { | |
056925ab | 1324 | call_netdevice_notifiers(NETDEV_FEAT_CHANGE, dev); |
d8a33ac4 SH |
1325 | } |
1326 | EXPORT_SYMBOL(netdev_features_change); | |
1327 | ||
1da177e4 LT |
1328 | /** |
1329 | * netdev_state_change - device changes state | |
1330 | * @dev: device to cause notification | |
1331 | * | |
1332 | * Called to indicate a device has changed state. This function calls | |
1333 | * the notifier chains for netdev_chain and sends a NEWLINK message | |
1334 | * to the routing socket. | |
1335 | */ | |
1336 | void netdev_state_change(struct net_device *dev) | |
1337 | { | |
1338 | if (dev->flags & IFF_UP) { | |
51d0c047 DA |
1339 | struct netdev_notifier_change_info change_info = { |
1340 | .info.dev = dev, | |
1341 | }; | |
54951194 | 1342 | |
51d0c047 | 1343 | call_netdevice_notifiers_info(NETDEV_CHANGE, |
54951194 | 1344 | &change_info.info); |
7f294054 | 1345 | rtmsg_ifinfo(RTM_NEWLINK, dev, 0, GFP_KERNEL); |
1da177e4 LT |
1346 | } |
1347 | } | |
d1b19dff | 1348 | EXPORT_SYMBOL(netdev_state_change); |
1da177e4 | 1349 | |
ee89bab1 | 1350 | /** |
722c9a0c | 1351 | * netdev_notify_peers - notify network peers about existence of @dev |
1352 | * @dev: network device | |
ee89bab1 AW |
1353 | * |
1354 | * Generate traffic such that interested network peers are aware of | |
1355 | * @dev, such as by generating a gratuitous ARP. This may be used when | |
1356 | * a device wants to inform the rest of the network about some sort of | |
1357 | * reconfiguration such as a failover event or virtual machine | |
1358 | * migration. | |
1359 | */ | |
1360 | void netdev_notify_peers(struct net_device *dev) | |
c1da4ac7 | 1361 | { |
ee89bab1 AW |
1362 | rtnl_lock(); |
1363 | call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); | |
37c343b4 | 1364 | call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); |
ee89bab1 | 1365 | rtnl_unlock(); |
c1da4ac7 | 1366 | } |
ee89bab1 | 1367 | EXPORT_SYMBOL(netdev_notify_peers); |
c1da4ac7 | 1368 | |
40c900aa | 1369 | static int __dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
1da177e4 | 1370 | { |
d314774c | 1371 | const struct net_device_ops *ops = dev->netdev_ops; |
3b8bcfd5 | 1372 | int ret; |
1da177e4 | 1373 | |
e46b66bc BH |
1374 | ASSERT_RTNL(); |
1375 | ||
1da177e4 LT |
1376 | if (!netif_device_present(dev)) |
1377 | return -ENODEV; | |
1378 | ||
ca99ca14 NH |
1379 | /* Block netpoll from trying to do any rx path servicing. |
1380 | * If we don't do this there is a chance ndo_poll_controller | |
1381 | * or ndo_poll may be running while we open the device | |
1382 | */ | |
66b5552f | 1383 | netpoll_poll_disable(dev); |
ca99ca14 | 1384 | |
40c900aa | 1385 | ret = call_netdevice_notifiers_extack(NETDEV_PRE_UP, dev, extack); |
3b8bcfd5 JB |
1386 | ret = notifier_to_errno(ret); |
1387 | if (ret) | |
1388 | return ret; | |
1389 | ||
1da177e4 | 1390 | set_bit(__LINK_STATE_START, &dev->state); |
bada339b | 1391 | |
d314774c SH |
1392 | if (ops->ndo_validate_addr) |
1393 | ret = ops->ndo_validate_addr(dev); | |
bada339b | 1394 | |
d314774c SH |
1395 | if (!ret && ops->ndo_open) |
1396 | ret = ops->ndo_open(dev); | |
1da177e4 | 1397 | |
66b5552f | 1398 | netpoll_poll_enable(dev); |
ca99ca14 | 1399 | |
bada339b JG |
1400 | if (ret) |
1401 | clear_bit(__LINK_STATE_START, &dev->state); | |
1402 | else { | |
1da177e4 | 1403 | dev->flags |= IFF_UP; |
4417da66 | 1404 | dev_set_rx_mode(dev); |
1da177e4 | 1405 | dev_activate(dev); |
7bf23575 | 1406 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 1407 | } |
bada339b | 1408 | |
1da177e4 LT |
1409 | return ret; |
1410 | } | |
1411 | ||
1412 | /** | |
bd380811 | 1413 | * dev_open - prepare an interface for use. |
00f54e68 PM |
1414 | * @dev: device to open |
1415 | * @extack: netlink extended ack | |
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 | */ |
00f54e68 | 1425 | int dev_open(struct net_device *dev, struct netlink_ext_ack *extack) |
bd380811 PM |
1426 | { |
1427 | int ret; | |
1428 | ||
bd380811 PM |
1429 | if (dev->flags & IFF_UP) |
1430 | return 0; | |
1431 | ||
40c900aa | 1432 | ret = __dev_open(dev, extack); |
bd380811 PM |
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 | ||
56f5aa77 MC |
1562 | /** |
1563 | * dev_disable_gro_hw - disable HW Generic Receive Offload on a device | |
1564 | * @dev: device | |
1565 | * | |
1566 | * Disable HW Generic Receive Offload (GRO_HW) on a net device. Must be | |
1567 | * called under RTNL. This is needed if Generic XDP is installed on | |
1568 | * the device. | |
1569 | */ | |
1570 | static void dev_disable_gro_hw(struct net_device *dev) | |
1571 | { | |
1572 | dev->wanted_features &= ~NETIF_F_GRO_HW; | |
1573 | netdev_update_features(dev); | |
1574 | ||
1575 | if (unlikely(dev->features & NETIF_F_GRO_HW)) | |
1576 | netdev_WARN(dev, "failed to disable GRO_HW!\n"); | |
1577 | } | |
1578 | ||
ede2762d KT |
1579 | const char *netdev_cmd_to_name(enum netdev_cmd cmd) |
1580 | { | |
1581 | #define N(val) \ | |
1582 | case NETDEV_##val: \ | |
1583 | return "NETDEV_" __stringify(val); | |
1584 | switch (cmd) { | |
1585 | N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) N(UNREGISTER) | |
1586 | N(CHANGEMTU) N(CHANGEADDR) N(GOING_DOWN) N(CHANGENAME) N(FEAT_CHANGE) | |
1587 | N(BONDING_FAILOVER) N(PRE_UP) N(PRE_TYPE_CHANGE) N(POST_TYPE_CHANGE) | |
1588 | N(POST_INIT) N(RELEASE) N(NOTIFY_PEERS) N(JOIN) N(CHANGEUPPER) | |
1589 | N(RESEND_IGMP) N(PRECHANGEMTU) N(CHANGEINFODATA) N(BONDING_INFO) | |
1590 | N(PRECHANGEUPPER) N(CHANGELOWERSTATE) N(UDP_TUNNEL_PUSH_INFO) | |
1591 | N(UDP_TUNNEL_DROP_INFO) N(CHANGE_TX_QUEUE_LEN) | |
9daae9bd GP |
1592 | N(CVLAN_FILTER_PUSH_INFO) N(CVLAN_FILTER_DROP_INFO) |
1593 | N(SVLAN_FILTER_PUSH_INFO) N(SVLAN_FILTER_DROP_INFO) | |
1570415f | 1594 | N(PRE_CHANGEADDR) |
3f5ecd8a | 1595 | } |
ede2762d KT |
1596 | #undef N |
1597 | return "UNKNOWN_NETDEV_EVENT"; | |
1598 | } | |
1599 | EXPORT_SYMBOL_GPL(netdev_cmd_to_name); | |
1600 | ||
351638e7 JP |
1601 | static int call_netdevice_notifier(struct notifier_block *nb, unsigned long val, |
1602 | struct net_device *dev) | |
1603 | { | |
51d0c047 DA |
1604 | struct netdev_notifier_info info = { |
1605 | .dev = dev, | |
1606 | }; | |
351638e7 | 1607 | |
351638e7 JP |
1608 | return nb->notifier_call(nb, val, &info); |
1609 | } | |
0187bdfb | 1610 | |
881d966b EB |
1611 | static int dev_boot_phase = 1; |
1612 | ||
1da177e4 | 1613 | /** |
722c9a0c | 1614 | * register_netdevice_notifier - register a network notifier block |
1615 | * @nb: notifier | |
1da177e4 | 1616 | * |
722c9a0c | 1617 | * Register a notifier to be called when network device events occur. |
1618 | * The notifier passed is linked into the kernel structures and must | |
1619 | * not be reused until it has been unregistered. A negative errno code | |
1620 | * is returned on a failure. | |
1da177e4 | 1621 | * |
722c9a0c | 1622 | * When registered all registration and up events are replayed |
1623 | * to the new notifier to allow device to have a race free | |
1624 | * view of the network device list. | |
1da177e4 LT |
1625 | */ |
1626 | ||
1627 | int register_netdevice_notifier(struct notifier_block *nb) | |
1628 | { | |
1629 | struct net_device *dev; | |
fcc5a03a | 1630 | struct net_device *last; |
881d966b | 1631 | struct net *net; |
1da177e4 LT |
1632 | int err; |
1633 | ||
328fbe74 KT |
1634 | /* Close race with setup_net() and cleanup_net() */ |
1635 | down_write(&pernet_ops_rwsem); | |
1da177e4 | 1636 | rtnl_lock(); |
f07d5b94 | 1637 | err = raw_notifier_chain_register(&netdev_chain, nb); |
fcc5a03a HX |
1638 | if (err) |
1639 | goto unlock; | |
881d966b EB |
1640 | if (dev_boot_phase) |
1641 | goto unlock; | |
1642 | for_each_net(net) { | |
1643 | for_each_netdev(net, dev) { | |
351638e7 | 1644 | err = call_netdevice_notifier(nb, NETDEV_REGISTER, dev); |
881d966b EB |
1645 | err = notifier_to_errno(err); |
1646 | if (err) | |
1647 | goto rollback; | |
1648 | ||
1649 | if (!(dev->flags & IFF_UP)) | |
1650 | continue; | |
1da177e4 | 1651 | |
351638e7 | 1652 | call_netdevice_notifier(nb, NETDEV_UP, dev); |
881d966b | 1653 | } |
1da177e4 | 1654 | } |
fcc5a03a HX |
1655 | |
1656 | unlock: | |
1da177e4 | 1657 | rtnl_unlock(); |
328fbe74 | 1658 | up_write(&pernet_ops_rwsem); |
1da177e4 | 1659 | return err; |
fcc5a03a HX |
1660 | |
1661 | rollback: | |
1662 | last = dev; | |
881d966b EB |
1663 | for_each_net(net) { |
1664 | for_each_netdev(net, dev) { | |
1665 | if (dev == last) | |
8f891489 | 1666 | goto outroll; |
fcc5a03a | 1667 | |
881d966b | 1668 | if (dev->flags & IFF_UP) { |
351638e7 JP |
1669 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1670 | dev); | |
1671 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
881d966b | 1672 | } |
351638e7 | 1673 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
fcc5a03a | 1674 | } |
fcc5a03a | 1675 | } |
c67625a1 | 1676 | |
8f891489 | 1677 | outroll: |
c67625a1 | 1678 | raw_notifier_chain_unregister(&netdev_chain, nb); |
fcc5a03a | 1679 | goto unlock; |
1da177e4 | 1680 | } |
d1b19dff | 1681 | EXPORT_SYMBOL(register_netdevice_notifier); |
1da177e4 LT |
1682 | |
1683 | /** | |
722c9a0c | 1684 | * unregister_netdevice_notifier - unregister a network notifier block |
1685 | * @nb: notifier | |
1da177e4 | 1686 | * |
722c9a0c | 1687 | * Unregister a notifier previously registered by |
1688 | * register_netdevice_notifier(). The notifier is unlinked into the | |
1689 | * kernel structures and may then be reused. A negative errno code | |
1690 | * is returned on a failure. | |
7d3d43da | 1691 | * |
722c9a0c | 1692 | * After unregistering unregister and down device events are synthesized |
1693 | * for all devices on the device list to the removed notifier to remove | |
1694 | * the need for special case cleanup code. | |
1da177e4 LT |
1695 | */ |
1696 | ||
1697 | int unregister_netdevice_notifier(struct notifier_block *nb) | |
1698 | { | |
7d3d43da EB |
1699 | struct net_device *dev; |
1700 | struct net *net; | |
9f514950 HX |
1701 | int err; |
1702 | ||
328fbe74 KT |
1703 | /* Close race with setup_net() and cleanup_net() */ |
1704 | down_write(&pernet_ops_rwsem); | |
9f514950 | 1705 | rtnl_lock(); |
f07d5b94 | 1706 | err = raw_notifier_chain_unregister(&netdev_chain, nb); |
7d3d43da EB |
1707 | if (err) |
1708 | goto unlock; | |
1709 | ||
1710 | for_each_net(net) { | |
1711 | for_each_netdev(net, dev) { | |
1712 | if (dev->flags & IFF_UP) { | |
351638e7 JP |
1713 | call_netdevice_notifier(nb, NETDEV_GOING_DOWN, |
1714 | dev); | |
1715 | call_netdevice_notifier(nb, NETDEV_DOWN, dev); | |
7d3d43da | 1716 | } |
351638e7 | 1717 | call_netdevice_notifier(nb, NETDEV_UNREGISTER, dev); |
7d3d43da EB |
1718 | } |
1719 | } | |
1720 | unlock: | |
9f514950 | 1721 | rtnl_unlock(); |
328fbe74 | 1722 | up_write(&pernet_ops_rwsem); |
9f514950 | 1723 | return err; |
1da177e4 | 1724 | } |
d1b19dff | 1725 | EXPORT_SYMBOL(unregister_netdevice_notifier); |
1da177e4 | 1726 | |
351638e7 JP |
1727 | /** |
1728 | * call_netdevice_notifiers_info - call all network notifier blocks | |
1729 | * @val: value passed unmodified to notifier function | |
351638e7 JP |
1730 | * @info: notifier information data |
1731 | * | |
1732 | * Call all network notifier blocks. Parameters and return value | |
1733 | * are as for raw_notifier_call_chain(). | |
1734 | */ | |
1735 | ||
1d143d9f | 1736 | static int call_netdevice_notifiers_info(unsigned long val, |
1d143d9f | 1737 | struct netdev_notifier_info *info) |
351638e7 JP |
1738 | { |
1739 | ASSERT_RTNL(); | |
351638e7 JP |
1740 | return raw_notifier_call_chain(&netdev_chain, val, info); |
1741 | } | |
351638e7 | 1742 | |
26372605 PM |
1743 | static int call_netdevice_notifiers_extack(unsigned long val, |
1744 | struct net_device *dev, | |
1745 | struct netlink_ext_ack *extack) | |
1746 | { | |
1747 | struct netdev_notifier_info info = { | |
1748 | .dev = dev, | |
1749 | .extack = extack, | |
1750 | }; | |
1751 | ||
1752 | return call_netdevice_notifiers_info(val, &info); | |
1753 | } | |
1754 | ||
1da177e4 LT |
1755 | /** |
1756 | * call_netdevice_notifiers - call all network notifier blocks | |
1757 | * @val: value passed unmodified to notifier function | |
c4ea43c5 | 1758 | * @dev: net_device pointer passed unmodified to notifier function |
1da177e4 LT |
1759 | * |
1760 | * Call all network notifier blocks. Parameters and return value | |
f07d5b94 | 1761 | * are as for raw_notifier_call_chain(). |
1da177e4 LT |
1762 | */ |
1763 | ||
ad7379d4 | 1764 | int call_netdevice_notifiers(unsigned long val, struct net_device *dev) |
1da177e4 | 1765 | { |
26372605 | 1766 | return call_netdevice_notifiers_extack(val, dev, NULL); |
1da177e4 | 1767 | } |
edf947f1 | 1768 | EXPORT_SYMBOL(call_netdevice_notifiers); |
1da177e4 | 1769 | |
af7d6cce SD |
1770 | /** |
1771 | * call_netdevice_notifiers_mtu - call all network notifier blocks | |
1772 | * @val: value passed unmodified to notifier function | |
1773 | * @dev: net_device pointer passed unmodified to notifier function | |
1774 | * @arg: additional u32 argument passed to the notifier function | |
1775 | * | |
1776 | * Call all network notifier blocks. Parameters and return value | |
1777 | * are as for raw_notifier_call_chain(). | |
1778 | */ | |
1779 | static int call_netdevice_notifiers_mtu(unsigned long val, | |
1780 | struct net_device *dev, u32 arg) | |
1781 | { | |
1782 | struct netdev_notifier_info_ext info = { | |
1783 | .info.dev = dev, | |
1784 | .ext.mtu = arg, | |
1785 | }; | |
1786 | ||
1787 | BUILD_BUG_ON(offsetof(struct netdev_notifier_info_ext, info) != 0); | |
1788 | ||
1789 | return call_netdevice_notifiers_info(val, &info.info); | |
1790 | } | |
1791 | ||
1cf51900 | 1792 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 1793 | static DEFINE_STATIC_KEY_FALSE(ingress_needed_key); |
4577139b DB |
1794 | |
1795 | void net_inc_ingress_queue(void) | |
1796 | { | |
aabf6772 | 1797 | static_branch_inc(&ingress_needed_key); |
4577139b DB |
1798 | } |
1799 | EXPORT_SYMBOL_GPL(net_inc_ingress_queue); | |
1800 | ||
1801 | void net_dec_ingress_queue(void) | |
1802 | { | |
aabf6772 | 1803 | static_branch_dec(&ingress_needed_key); |
4577139b DB |
1804 | } |
1805 | EXPORT_SYMBOL_GPL(net_dec_ingress_queue); | |
1806 | #endif | |
1807 | ||
1f211a1b | 1808 | #ifdef CONFIG_NET_EGRESS |
aabf6772 | 1809 | static DEFINE_STATIC_KEY_FALSE(egress_needed_key); |
1f211a1b DB |
1810 | |
1811 | void net_inc_egress_queue(void) | |
1812 | { | |
aabf6772 | 1813 | static_branch_inc(&egress_needed_key); |
1f211a1b DB |
1814 | } |
1815 | EXPORT_SYMBOL_GPL(net_inc_egress_queue); | |
1816 | ||
1817 | void net_dec_egress_queue(void) | |
1818 | { | |
aabf6772 | 1819 | static_branch_dec(&egress_needed_key); |
1f211a1b DB |
1820 | } |
1821 | EXPORT_SYMBOL_GPL(net_dec_egress_queue); | |
1822 | #endif | |
1823 | ||
39e83922 | 1824 | static DEFINE_STATIC_KEY_FALSE(netstamp_needed_key); |
e9666d10 | 1825 | #ifdef CONFIG_JUMP_LABEL |
b90e5794 | 1826 | static atomic_t netstamp_needed_deferred; |
13baa00a | 1827 | static atomic_t netstamp_wanted; |
5fa8bbda | 1828 | static void netstamp_clear(struct work_struct *work) |
1da177e4 | 1829 | { |
b90e5794 | 1830 | int deferred = atomic_xchg(&netstamp_needed_deferred, 0); |
13baa00a | 1831 | int wanted; |
b90e5794 | 1832 | |
13baa00a ED |
1833 | wanted = atomic_add_return(deferred, &netstamp_wanted); |
1834 | if (wanted > 0) | |
39e83922 | 1835 | static_branch_enable(&netstamp_needed_key); |
13baa00a | 1836 | else |
39e83922 | 1837 | static_branch_disable(&netstamp_needed_key); |
5fa8bbda ED |
1838 | } |
1839 | static DECLARE_WORK(netstamp_work, netstamp_clear); | |
b90e5794 | 1840 | #endif |
5fa8bbda ED |
1841 | |
1842 | void net_enable_timestamp(void) | |
1843 | { | |
e9666d10 | 1844 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
1845 | int wanted; |
1846 | ||
1847 | while (1) { | |
1848 | wanted = atomic_read(&netstamp_wanted); | |
1849 | if (wanted <= 0) | |
1850 | break; | |
1851 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted + 1) == wanted) | |
1852 | return; | |
1853 | } | |
1854 | atomic_inc(&netstamp_needed_deferred); | |
1855 | schedule_work(&netstamp_work); | |
1856 | #else | |
39e83922 | 1857 | static_branch_inc(&netstamp_needed_key); |
13baa00a | 1858 | #endif |
1da177e4 | 1859 | } |
d1b19dff | 1860 | EXPORT_SYMBOL(net_enable_timestamp); |
1da177e4 LT |
1861 | |
1862 | void net_disable_timestamp(void) | |
1863 | { | |
e9666d10 | 1864 | #ifdef CONFIG_JUMP_LABEL |
13baa00a ED |
1865 | int wanted; |
1866 | ||
1867 | while (1) { | |
1868 | wanted = atomic_read(&netstamp_wanted); | |
1869 | if (wanted <= 1) | |
1870 | break; | |
1871 | if (atomic_cmpxchg(&netstamp_wanted, wanted, wanted - 1) == wanted) | |
1872 | return; | |
1873 | } | |
1874 | atomic_dec(&netstamp_needed_deferred); | |
5fa8bbda ED |
1875 | schedule_work(&netstamp_work); |
1876 | #else | |
39e83922 | 1877 | static_branch_dec(&netstamp_needed_key); |
5fa8bbda | 1878 | #endif |
1da177e4 | 1879 | } |
d1b19dff | 1880 | EXPORT_SYMBOL(net_disable_timestamp); |
1da177e4 | 1881 | |
3b098e2d | 1882 | static inline void net_timestamp_set(struct sk_buff *skb) |
1da177e4 | 1883 | { |
2456e855 | 1884 | skb->tstamp = 0; |
39e83922 | 1885 | if (static_branch_unlikely(&netstamp_needed_key)) |
a61bbcf2 | 1886 | __net_timestamp(skb); |
1da177e4 LT |
1887 | } |
1888 | ||
39e83922 DB |
1889 | #define net_timestamp_check(COND, SKB) \ |
1890 | if (static_branch_unlikely(&netstamp_needed_key)) { \ | |
1891 | if ((COND) && !(SKB)->tstamp) \ | |
1892 | __net_timestamp(SKB); \ | |
1893 | } \ | |
3b098e2d | 1894 | |
f4b05d27 | 1895 | bool is_skb_forwardable(const struct net_device *dev, const struct sk_buff *skb) |
79b569f0 DL |
1896 | { |
1897 | unsigned int len; | |
1898 | ||
1899 | if (!(dev->flags & IFF_UP)) | |
1900 | return false; | |
1901 | ||
1902 | len = dev->mtu + dev->hard_header_len + VLAN_HLEN; | |
1903 | if (skb->len <= len) | |
1904 | return true; | |
1905 | ||
1906 | /* if TSO is enabled, we don't care about the length as the packet | |
1907 | * could be forwarded without being segmented before | |
1908 | */ | |
1909 | if (skb_is_gso(skb)) | |
1910 | return true; | |
1911 | ||
1912 | return false; | |
1913 | } | |
1ee481fb | 1914 | EXPORT_SYMBOL_GPL(is_skb_forwardable); |
79b569f0 | 1915 | |
a0265d28 HX |
1916 | int __dev_forward_skb(struct net_device *dev, struct sk_buff *skb) |
1917 | { | |
4e3264d2 | 1918 | int ret = ____dev_forward_skb(dev, skb); |
a0265d28 | 1919 | |
4e3264d2 MKL |
1920 | if (likely(!ret)) { |
1921 | skb->protocol = eth_type_trans(skb, dev); | |
1922 | skb_postpull_rcsum(skb, eth_hdr(skb), ETH_HLEN); | |
1923 | } | |
a0265d28 | 1924 | |
4e3264d2 | 1925 | return ret; |
a0265d28 HX |
1926 | } |
1927 | EXPORT_SYMBOL_GPL(__dev_forward_skb); | |
1928 | ||
44540960 AB |
1929 | /** |
1930 | * dev_forward_skb - loopback an skb to another netif | |
1931 | * | |
1932 | * @dev: destination network device | |
1933 | * @skb: buffer to forward | |
1934 | * | |
1935 | * return values: | |
1936 | * NET_RX_SUCCESS (no congestion) | |
6ec82562 | 1937 | * NET_RX_DROP (packet was dropped, but freed) |
44540960 AB |
1938 | * |
1939 | * dev_forward_skb can be used for injecting an skb from the | |
1940 | * start_xmit function of one device into the receive queue | |
1941 | * of another device. | |
1942 | * | |
1943 | * The receiving device may be in another namespace, so | |
1944 | * we have to clear all information in the skb that could | |
1945 | * impact namespace isolation. | |
1946 | */ | |
1947 | int dev_forward_skb(struct net_device *dev, struct sk_buff *skb) | |
1948 | { | |
a0265d28 | 1949 | return __dev_forward_skb(dev, skb) ?: netif_rx_internal(skb); |
44540960 AB |
1950 | } |
1951 | EXPORT_SYMBOL_GPL(dev_forward_skb); | |
1952 | ||
71d9dec2 CG |
1953 | static inline int deliver_skb(struct sk_buff *skb, |
1954 | struct packet_type *pt_prev, | |
1955 | struct net_device *orig_dev) | |
1956 | { | |
1f8b977a | 1957 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
1080e512 | 1958 | return -ENOMEM; |
63354797 | 1959 | refcount_inc(&skb->users); |
71d9dec2 CG |
1960 | return pt_prev->func(skb, skb->dev, pt_prev, orig_dev); |
1961 | } | |
1962 | ||
7866a621 SN |
1963 | static inline void deliver_ptype_list_skb(struct sk_buff *skb, |
1964 | struct packet_type **pt, | |
fbcb2170 JP |
1965 | struct net_device *orig_dev, |
1966 | __be16 type, | |
7866a621 SN |
1967 | struct list_head *ptype_list) |
1968 | { | |
1969 | struct packet_type *ptype, *pt_prev = *pt; | |
1970 | ||
1971 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
1972 | if (ptype->type != type) | |
1973 | continue; | |
1974 | if (pt_prev) | |
fbcb2170 | 1975 | deliver_skb(skb, pt_prev, orig_dev); |
7866a621 SN |
1976 | pt_prev = ptype; |
1977 | } | |
1978 | *pt = pt_prev; | |
1979 | } | |
1980 | ||
c0de08d0 EL |
1981 | static inline bool skb_loop_sk(struct packet_type *ptype, struct sk_buff *skb) |
1982 | { | |
a3d744e9 | 1983 | if (!ptype->af_packet_priv || !skb->sk) |
c0de08d0 EL |
1984 | return false; |
1985 | ||
1986 | if (ptype->id_match) | |
1987 | return ptype->id_match(ptype, skb->sk); | |
1988 | else if ((struct sock *)ptype->af_packet_priv == skb->sk) | |
1989 | return true; | |
1990 | ||
1991 | return false; | |
1992 | } | |
1993 | ||
9f9a742d MR |
1994 | /** |
1995 | * dev_nit_active - return true if any network interface taps are in use | |
1996 | * | |
1997 | * @dev: network device to check for the presence of taps | |
1998 | */ | |
1999 | bool dev_nit_active(struct net_device *dev) | |
2000 | { | |
2001 | return !list_empty(&ptype_all) || !list_empty(&dev->ptype_all); | |
2002 | } | |
2003 | EXPORT_SYMBOL_GPL(dev_nit_active); | |
2004 | ||
1da177e4 LT |
2005 | /* |
2006 | * Support routine. Sends outgoing frames to any network | |
2007 | * taps currently in use. | |
2008 | */ | |
2009 | ||
74b20582 | 2010 | void dev_queue_xmit_nit(struct sk_buff *skb, struct net_device *dev) |
1da177e4 LT |
2011 | { |
2012 | struct packet_type *ptype; | |
71d9dec2 CG |
2013 | struct sk_buff *skb2 = NULL; |
2014 | struct packet_type *pt_prev = NULL; | |
7866a621 | 2015 | struct list_head *ptype_list = &ptype_all; |
a61bbcf2 | 2016 | |
1da177e4 | 2017 | rcu_read_lock(); |
7866a621 SN |
2018 | again: |
2019 | list_for_each_entry_rcu(ptype, ptype_list, list) { | |
fa788d98 VW |
2020 | if (ptype->ignore_outgoing) |
2021 | continue; | |
2022 | ||
1da177e4 LT |
2023 | /* Never send packets back to the socket |
2024 | * they originated from - MvS (miquels@drinkel.ow.org) | |
2025 | */ | |
7866a621 SN |
2026 | if (skb_loop_sk(ptype, skb)) |
2027 | continue; | |
71d9dec2 | 2028 | |
7866a621 SN |
2029 | if (pt_prev) { |
2030 | deliver_skb(skb2, pt_prev, skb->dev); | |
2031 | pt_prev = ptype; | |
2032 | continue; | |
2033 | } | |
1da177e4 | 2034 | |
7866a621 SN |
2035 | /* need to clone skb, done only once */ |
2036 | skb2 = skb_clone(skb, GFP_ATOMIC); | |
2037 | if (!skb2) | |
2038 | goto out_unlock; | |
70978182 | 2039 | |
7866a621 | 2040 | net_timestamp_set(skb2); |
1da177e4 | 2041 | |
7866a621 SN |
2042 | /* skb->nh should be correctly |
2043 | * set by sender, so that the second statement is | |
2044 | * just protection against buggy protocols. | |
2045 | */ | |
2046 | skb_reset_mac_header(skb2); | |
2047 | ||
2048 | if (skb_network_header(skb2) < skb2->data || | |
2049 | skb_network_header(skb2) > skb_tail_pointer(skb2)) { | |
2050 | net_crit_ratelimited("protocol %04x is buggy, dev %s\n", | |
2051 | ntohs(skb2->protocol), | |
2052 | dev->name); | |
2053 | skb_reset_network_header(skb2); | |
1da177e4 | 2054 | } |
7866a621 SN |
2055 | |
2056 | skb2->transport_header = skb2->network_header; | |
2057 | skb2->pkt_type = PACKET_OUTGOING; | |
2058 | pt_prev = ptype; | |
2059 | } | |
2060 | ||
2061 | if (ptype_list == &ptype_all) { | |
2062 | ptype_list = &dev->ptype_all; | |
2063 | goto again; | |
1da177e4 | 2064 | } |
7866a621 | 2065 | out_unlock: |
581fe0ea WB |
2066 | if (pt_prev) { |
2067 | if (!skb_orphan_frags_rx(skb2, GFP_ATOMIC)) | |
2068 | pt_prev->func(skb2, skb->dev, pt_prev, skb->dev); | |
2069 | else | |
2070 | kfree_skb(skb2); | |
2071 | } | |
1da177e4 LT |
2072 | rcu_read_unlock(); |
2073 | } | |
74b20582 | 2074 | EXPORT_SYMBOL_GPL(dev_queue_xmit_nit); |
1da177e4 | 2075 | |
2c53040f BH |
2076 | /** |
2077 | * netif_setup_tc - Handle tc mappings on real_num_tx_queues change | |
4f57c087 JF |
2078 | * @dev: Network device |
2079 | * @txq: number of queues available | |
2080 | * | |
2081 | * If real_num_tx_queues is changed the tc mappings may no longer be | |
2082 | * valid. To resolve this verify the tc mapping remains valid and if | |
2083 | * not NULL the mapping. With no priorities mapping to this | |
2084 | * offset/count pair it will no longer be used. In the worst case TC0 | |
2085 | * is invalid nothing can be done so disable priority mappings. If is | |
2086 | * expected that drivers will fix this mapping if they can before | |
2087 | * calling netif_set_real_num_tx_queues. | |
2088 | */ | |
bb134d22 | 2089 | static void netif_setup_tc(struct net_device *dev, unsigned int txq) |
4f57c087 JF |
2090 | { |
2091 | int i; | |
2092 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2093 | ||
2094 | /* If TC0 is invalidated disable TC mapping */ | |
2095 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce | 2096 | pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n"); |
4f57c087 JF |
2097 | dev->num_tc = 0; |
2098 | return; | |
2099 | } | |
2100 | ||
2101 | /* Invalidated prio to tc mappings set to TC0 */ | |
2102 | for (i = 1; i < TC_BITMASK + 1; i++) { | |
2103 | int q = netdev_get_prio_tc_map(dev, i); | |
2104 | ||
2105 | tc = &dev->tc_to_txq[q]; | |
2106 | if (tc->offset + tc->count > txq) { | |
7b6cd1ce JP |
2107 | pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n", |
2108 | i, q); | |
4f57c087 JF |
2109 | netdev_set_prio_tc_map(dev, i, 0); |
2110 | } | |
2111 | } | |
2112 | } | |
2113 | ||
8d059b0f AD |
2114 | int netdev_txq_to_tc(struct net_device *dev, unsigned int txq) |
2115 | { | |
2116 | if (dev->num_tc) { | |
2117 | struct netdev_tc_txq *tc = &dev->tc_to_txq[0]; | |
2118 | int i; | |
2119 | ||
ffcfe25b | 2120 | /* walk through the TCs and see if it falls into any of them */ |
8d059b0f AD |
2121 | for (i = 0; i < TC_MAX_QUEUE; i++, tc++) { |
2122 | if ((txq - tc->offset) < tc->count) | |
2123 | return i; | |
2124 | } | |
2125 | ||
ffcfe25b | 2126 | /* didn't find it, just return -1 to indicate no match */ |
8d059b0f AD |
2127 | return -1; |
2128 | } | |
2129 | ||
2130 | return 0; | |
2131 | } | |
8a5f2166 | 2132 | EXPORT_SYMBOL(netdev_txq_to_tc); |
8d059b0f | 2133 | |
537c00de | 2134 | #ifdef CONFIG_XPS |
04157469 AN |
2135 | struct static_key xps_needed __read_mostly; |
2136 | EXPORT_SYMBOL(xps_needed); | |
2137 | struct static_key xps_rxqs_needed __read_mostly; | |
2138 | EXPORT_SYMBOL(xps_rxqs_needed); | |
537c00de AD |
2139 | static DEFINE_MUTEX(xps_map_mutex); |
2140 | #define xmap_dereference(P) \ | |
2141 | rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex)) | |
2142 | ||
6234f874 AD |
2143 | static bool remove_xps_queue(struct xps_dev_maps *dev_maps, |
2144 | int tci, u16 index) | |
537c00de | 2145 | { |
10cdc3f3 AD |
2146 | struct xps_map *map = NULL; |
2147 | int pos; | |
537c00de | 2148 | |
10cdc3f3 | 2149 | if (dev_maps) |
80d19669 | 2150 | map = xmap_dereference(dev_maps->attr_map[tci]); |
6234f874 AD |
2151 | if (!map) |
2152 | return false; | |
537c00de | 2153 | |
6234f874 AD |
2154 | for (pos = map->len; pos--;) { |
2155 | if (map->queues[pos] != index) | |
2156 | continue; | |
2157 | ||
2158 | if (map->len > 1) { | |
2159 | map->queues[pos] = map->queues[--map->len]; | |
10cdc3f3 | 2160 | break; |
537c00de | 2161 | } |
6234f874 | 2162 | |
80d19669 | 2163 | RCU_INIT_POINTER(dev_maps->attr_map[tci], NULL); |
6234f874 AD |
2164 | kfree_rcu(map, rcu); |
2165 | return false; | |
537c00de AD |
2166 | } |
2167 | ||
6234f874 | 2168 | return true; |
10cdc3f3 AD |
2169 | } |
2170 | ||
6234f874 AD |
2171 | static bool remove_xps_queue_cpu(struct net_device *dev, |
2172 | struct xps_dev_maps *dev_maps, | |
2173 | int cpu, u16 offset, u16 count) | |
2174 | { | |
184c449f AD |
2175 | int num_tc = dev->num_tc ? : 1; |
2176 | bool active = false; | |
2177 | int tci; | |
6234f874 | 2178 | |
184c449f AD |
2179 | for (tci = cpu * num_tc; num_tc--; tci++) { |
2180 | int i, j; | |
2181 | ||
2182 | for (i = count, j = offset; i--; j++) { | |
6358d49a | 2183 | if (!remove_xps_queue(dev_maps, tci, j)) |
184c449f AD |
2184 | break; |
2185 | } | |
2186 | ||
2187 | active |= i < 0; | |
6234f874 AD |
2188 | } |
2189 | ||
184c449f | 2190 | return active; |
6234f874 AD |
2191 | } |
2192 | ||
867d0ad4 SD |
2193 | static void reset_xps_maps(struct net_device *dev, |
2194 | struct xps_dev_maps *dev_maps, | |
2195 | bool is_rxqs_map) | |
2196 | { | |
2197 | if (is_rxqs_map) { | |
2198 | static_key_slow_dec_cpuslocked(&xps_rxqs_needed); | |
2199 | RCU_INIT_POINTER(dev->xps_rxqs_map, NULL); | |
2200 | } else { | |
2201 | RCU_INIT_POINTER(dev->xps_cpus_map, NULL); | |
2202 | } | |
2203 | static_key_slow_dec_cpuslocked(&xps_needed); | |
2204 | kfree_rcu(dev_maps, rcu); | |
2205 | } | |
2206 | ||
80d19669 AN |
2207 | static void clean_xps_maps(struct net_device *dev, const unsigned long *mask, |
2208 | struct xps_dev_maps *dev_maps, unsigned int nr_ids, | |
2209 | u16 offset, u16 count, bool is_rxqs_map) | |
2210 | { | |
2211 | bool active = false; | |
2212 | int i, j; | |
2213 | ||
2214 | for (j = -1; j = netif_attrmask_next(j, mask, nr_ids), | |
2215 | j < nr_ids;) | |
2216 | active |= remove_xps_queue_cpu(dev, dev_maps, j, offset, | |
2217 | count); | |
867d0ad4 SD |
2218 | if (!active) |
2219 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
80d19669 | 2220 | |
f28c020f SD |
2221 | if (!is_rxqs_map) { |
2222 | for (i = offset + (count - 1); count--; i--) { | |
2223 | netdev_queue_numa_node_write( | |
2224 | netdev_get_tx_queue(dev, i), | |
2225 | NUMA_NO_NODE); | |
80d19669 | 2226 | } |
80d19669 AN |
2227 | } |
2228 | } | |
2229 | ||
6234f874 AD |
2230 | static void netif_reset_xps_queues(struct net_device *dev, u16 offset, |
2231 | u16 count) | |
10cdc3f3 | 2232 | { |
80d19669 | 2233 | const unsigned long *possible_mask = NULL; |
10cdc3f3 | 2234 | struct xps_dev_maps *dev_maps; |
80d19669 | 2235 | unsigned int nr_ids; |
10cdc3f3 | 2236 | |
04157469 AN |
2237 | if (!static_key_false(&xps_needed)) |
2238 | return; | |
10cdc3f3 | 2239 | |
4d99f660 | 2240 | cpus_read_lock(); |
04157469 | 2241 | mutex_lock(&xps_map_mutex); |
10cdc3f3 | 2242 | |
04157469 AN |
2243 | if (static_key_false(&xps_rxqs_needed)) { |
2244 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2245 | if (dev_maps) { | |
2246 | nr_ids = dev->num_rx_queues; | |
2247 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, | |
2248 | offset, count, true); | |
2249 | } | |
537c00de AD |
2250 | } |
2251 | ||
80d19669 AN |
2252 | dev_maps = xmap_dereference(dev->xps_cpus_map); |
2253 | if (!dev_maps) | |
2254 | goto out_no_maps; | |
2255 | ||
2256 | if (num_possible_cpus() > 1) | |
2257 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2258 | nr_ids = nr_cpu_ids; | |
2259 | clean_xps_maps(dev, possible_mask, dev_maps, nr_ids, offset, count, | |
2260 | false); | |
024e9679 | 2261 | |
537c00de AD |
2262 | out_no_maps: |
2263 | mutex_unlock(&xps_map_mutex); | |
4d99f660 | 2264 | cpus_read_unlock(); |
537c00de AD |
2265 | } |
2266 | ||
6234f874 AD |
2267 | static void netif_reset_xps_queues_gt(struct net_device *dev, u16 index) |
2268 | { | |
2269 | netif_reset_xps_queues(dev, index, dev->num_tx_queues - index); | |
2270 | } | |
2271 | ||
80d19669 AN |
2272 | static struct xps_map *expand_xps_map(struct xps_map *map, int attr_index, |
2273 | u16 index, bool is_rxqs_map) | |
01c5f864 AD |
2274 | { |
2275 | struct xps_map *new_map; | |
2276 | int alloc_len = XPS_MIN_MAP_ALLOC; | |
2277 | int i, pos; | |
2278 | ||
2279 | for (pos = 0; map && pos < map->len; pos++) { | |
2280 | if (map->queues[pos] != index) | |
2281 | continue; | |
2282 | return map; | |
2283 | } | |
2284 | ||
80d19669 | 2285 | /* Need to add tx-queue to this CPU's/rx-queue's existing map */ |
01c5f864 AD |
2286 | if (map) { |
2287 | if (pos < map->alloc_len) | |
2288 | return map; | |
2289 | ||
2290 | alloc_len = map->alloc_len * 2; | |
2291 | } | |
2292 | ||
80d19669 AN |
2293 | /* Need to allocate new map to store tx-queue on this CPU's/rx-queue's |
2294 | * map | |
2295 | */ | |
2296 | if (is_rxqs_map) | |
2297 | new_map = kzalloc(XPS_MAP_SIZE(alloc_len), GFP_KERNEL); | |
2298 | else | |
2299 | new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len), GFP_KERNEL, | |
2300 | cpu_to_node(attr_index)); | |
01c5f864 AD |
2301 | if (!new_map) |
2302 | return NULL; | |
2303 | ||
2304 | for (i = 0; i < pos; i++) | |
2305 | new_map->queues[i] = map->queues[i]; | |
2306 | new_map->alloc_len = alloc_len; | |
2307 | new_map->len = pos; | |
2308 | ||
2309 | return new_map; | |
2310 | } | |
2311 | ||
4d99f660 | 2312 | /* Must be called under cpus_read_lock */ |
80d19669 AN |
2313 | int __netif_set_xps_queue(struct net_device *dev, const unsigned long *mask, |
2314 | u16 index, bool is_rxqs_map) | |
537c00de | 2315 | { |
80d19669 | 2316 | const unsigned long *online_mask = NULL, *possible_mask = NULL; |
01c5f864 | 2317 | struct xps_dev_maps *dev_maps, *new_dev_maps = NULL; |
80d19669 | 2318 | int i, j, tci, numa_node_id = -2; |
184c449f | 2319 | int maps_sz, num_tc = 1, tc = 0; |
537c00de | 2320 | struct xps_map *map, *new_map; |
01c5f864 | 2321 | bool active = false; |
80d19669 | 2322 | unsigned int nr_ids; |
537c00de | 2323 | |
184c449f | 2324 | if (dev->num_tc) { |
ffcfe25b | 2325 | /* Do not allow XPS on subordinate device directly */ |
184c449f | 2326 | num_tc = dev->num_tc; |
ffcfe25b AD |
2327 | if (num_tc < 0) |
2328 | return -EINVAL; | |
2329 | ||
2330 | /* If queue belongs to subordinate dev use its map */ | |
2331 | dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev; | |
2332 | ||
184c449f AD |
2333 | tc = netdev_txq_to_tc(dev, index); |
2334 | if (tc < 0) | |
2335 | return -EINVAL; | |
2336 | } | |
2337 | ||
537c00de | 2338 | mutex_lock(&xps_map_mutex); |
80d19669 AN |
2339 | if (is_rxqs_map) { |
2340 | maps_sz = XPS_RXQ_DEV_MAPS_SIZE(num_tc, dev->num_rx_queues); | |
2341 | dev_maps = xmap_dereference(dev->xps_rxqs_map); | |
2342 | nr_ids = dev->num_rx_queues; | |
2343 | } else { | |
2344 | maps_sz = XPS_CPU_DEV_MAPS_SIZE(num_tc); | |
2345 | if (num_possible_cpus() > 1) { | |
2346 | online_mask = cpumask_bits(cpu_online_mask); | |
2347 | possible_mask = cpumask_bits(cpu_possible_mask); | |
2348 | } | |
2349 | dev_maps = xmap_dereference(dev->xps_cpus_map); | |
2350 | nr_ids = nr_cpu_ids; | |
2351 | } | |
537c00de | 2352 | |
80d19669 AN |
2353 | if (maps_sz < L1_CACHE_BYTES) |
2354 | maps_sz = L1_CACHE_BYTES; | |
537c00de | 2355 | |
01c5f864 | 2356 | /* allocate memory for queue storage */ |
80d19669 AN |
2357 | for (j = -1; j = netif_attrmask_next_and(j, online_mask, mask, nr_ids), |
2358 | j < nr_ids;) { | |
01c5f864 AD |
2359 | if (!new_dev_maps) |
2360 | new_dev_maps = kzalloc(maps_sz, GFP_KERNEL); | |
2bb60cb9 AD |
2361 | if (!new_dev_maps) { |
2362 | mutex_unlock(&xps_map_mutex); | |
01c5f864 | 2363 | return -ENOMEM; |
2bb60cb9 | 2364 | } |
01c5f864 | 2365 | |
80d19669 AN |
2366 | tci = j * num_tc + tc; |
2367 | map = dev_maps ? xmap_dereference(dev_maps->attr_map[tci]) : | |
01c5f864 AD |
2368 | NULL; |
2369 | ||
80d19669 | 2370 | map = expand_xps_map(map, j, index, is_rxqs_map); |
01c5f864 AD |
2371 | if (!map) |
2372 | goto error; | |
2373 | ||
80d19669 | 2374 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); |
01c5f864 AD |
2375 | } |
2376 | ||
2377 | if (!new_dev_maps) | |
2378 | goto out_no_new_maps; | |
2379 | ||
867d0ad4 SD |
2380 | if (!dev_maps) { |
2381 | /* Increment static keys at most once per type */ | |
2382 | static_key_slow_inc_cpuslocked(&xps_needed); | |
2383 | if (is_rxqs_map) | |
2384 | static_key_slow_inc_cpuslocked(&xps_rxqs_needed); | |
2385 | } | |
04157469 | 2386 | |
80d19669 AN |
2387 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2388 | j < nr_ids;) { | |
184c449f | 2389 | /* copy maps belonging to foreign traffic classes */ |
80d19669 | 2390 | for (i = tc, tci = j * num_tc; dev_maps && i--; tci++) { |
184c449f | 2391 | /* fill in the new device map from the old device map */ |
80d19669 AN |
2392 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2393 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f AD |
2394 | } |
2395 | ||
2396 | /* We need to explicitly update tci as prevous loop | |
2397 | * could break out early if dev_maps is NULL. | |
2398 | */ | |
80d19669 | 2399 | tci = j * num_tc + tc; |
184c449f | 2400 | |
80d19669 AN |
2401 | if (netif_attr_test_mask(j, mask, nr_ids) && |
2402 | netif_attr_test_online(j, online_mask, nr_ids)) { | |
2403 | /* add tx-queue to CPU/rx-queue maps */ | |
01c5f864 AD |
2404 | int pos = 0; |
2405 | ||
80d19669 | 2406 | map = xmap_dereference(new_dev_maps->attr_map[tci]); |
01c5f864 AD |
2407 | while ((pos < map->len) && (map->queues[pos] != index)) |
2408 | pos++; | |
2409 | ||
2410 | if (pos == map->len) | |
2411 | map->queues[map->len++] = index; | |
537c00de | 2412 | #ifdef CONFIG_NUMA |
80d19669 AN |
2413 | if (!is_rxqs_map) { |
2414 | if (numa_node_id == -2) | |
2415 | numa_node_id = cpu_to_node(j); | |
2416 | else if (numa_node_id != cpu_to_node(j)) | |
2417 | numa_node_id = -1; | |
2418 | } | |
537c00de | 2419 | #endif |
01c5f864 AD |
2420 | } else if (dev_maps) { |
2421 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2422 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2423 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
537c00de | 2424 | } |
01c5f864 | 2425 | |
184c449f AD |
2426 | /* copy maps belonging to foreign traffic classes */ |
2427 | for (i = num_tc - tc, tci++; dev_maps && --i; tci++) { | |
2428 | /* fill in the new device map from the old device map */ | |
80d19669 AN |
2429 | map = xmap_dereference(dev_maps->attr_map[tci]); |
2430 | RCU_INIT_POINTER(new_dev_maps->attr_map[tci], map); | |
184c449f | 2431 | } |
537c00de AD |
2432 | } |
2433 | ||
80d19669 AN |
2434 | if (is_rxqs_map) |
2435 | rcu_assign_pointer(dev->xps_rxqs_map, new_dev_maps); | |
2436 | else | |
2437 | rcu_assign_pointer(dev->xps_cpus_map, new_dev_maps); | |
01c5f864 | 2438 | |
537c00de | 2439 | /* Cleanup old maps */ |
184c449f AD |
2440 | if (!dev_maps) |
2441 | goto out_no_old_maps; | |
2442 | ||
80d19669 AN |
2443 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2444 | j < nr_ids;) { | |
2445 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2446 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
2447 | map = xmap_dereference(dev_maps->attr_map[tci]); | |
01c5f864 AD |
2448 | if (map && map != new_map) |
2449 | kfree_rcu(map, rcu); | |
2450 | } | |
537c00de AD |
2451 | } |
2452 | ||
184c449f AD |
2453 | kfree_rcu(dev_maps, rcu); |
2454 | ||
2455 | out_no_old_maps: | |
01c5f864 AD |
2456 | dev_maps = new_dev_maps; |
2457 | active = true; | |
537c00de | 2458 | |
01c5f864 | 2459 | out_no_new_maps: |
80d19669 AN |
2460 | if (!is_rxqs_map) { |
2461 | /* update Tx queue numa node */ | |
2462 | netdev_queue_numa_node_write(netdev_get_tx_queue(dev, index), | |
2463 | (numa_node_id >= 0) ? | |
2464 | numa_node_id : NUMA_NO_NODE); | |
2465 | } | |
537c00de | 2466 | |
01c5f864 AD |
2467 | if (!dev_maps) |
2468 | goto out_no_maps; | |
2469 | ||
80d19669 AN |
2470 | /* removes tx-queue from unused CPUs/rx-queues */ |
2471 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), | |
2472 | j < nr_ids;) { | |
2473 | for (i = tc, tci = j * num_tc; i--; tci++) | |
184c449f | 2474 | active |= remove_xps_queue(dev_maps, tci, index); |
80d19669 AN |
2475 | if (!netif_attr_test_mask(j, mask, nr_ids) || |
2476 | !netif_attr_test_online(j, online_mask, nr_ids)) | |
184c449f AD |
2477 | active |= remove_xps_queue(dev_maps, tci, index); |
2478 | for (i = num_tc - tc, tci++; --i; tci++) | |
2479 | active |= remove_xps_queue(dev_maps, tci, index); | |
01c5f864 AD |
2480 | } |
2481 | ||
2482 | /* free map if not active */ | |
867d0ad4 SD |
2483 | if (!active) |
2484 | reset_xps_maps(dev, dev_maps, is_rxqs_map); | |
01c5f864 AD |
2485 | |
2486 | out_no_maps: | |
537c00de AD |
2487 | mutex_unlock(&xps_map_mutex); |
2488 | ||
2489 | return 0; | |
2490 | error: | |
01c5f864 | 2491 | /* remove any maps that we added */ |
80d19669 AN |
2492 | for (j = -1; j = netif_attrmask_next(j, possible_mask, nr_ids), |
2493 | j < nr_ids;) { | |
2494 | for (i = num_tc, tci = j * num_tc; i--; tci++) { | |
2495 | new_map = xmap_dereference(new_dev_maps->attr_map[tci]); | |
184c449f | 2496 | map = dev_maps ? |
80d19669 | 2497 | xmap_dereference(dev_maps->attr_map[tci]) : |
184c449f AD |
2498 | NULL; |
2499 | if (new_map && new_map != map) | |
2500 | kfree(new_map); | |
2501 | } | |
01c5f864 AD |
2502 | } |
2503 | ||
537c00de AD |
2504 | mutex_unlock(&xps_map_mutex); |
2505 | ||
537c00de AD |
2506 | kfree(new_dev_maps); |
2507 | return -ENOMEM; | |
2508 | } | |
4d99f660 | 2509 | EXPORT_SYMBOL_GPL(__netif_set_xps_queue); |
80d19669 AN |
2510 | |
2511 | int netif_set_xps_queue(struct net_device *dev, const struct cpumask *mask, | |
2512 | u16 index) | |
2513 | { | |
4d99f660 AV |
2514 | int ret; |
2515 | ||
2516 | cpus_read_lock(); | |
2517 | ret = __netif_set_xps_queue(dev, cpumask_bits(mask), index, false); | |
2518 | cpus_read_unlock(); | |
2519 | ||
2520 | return ret; | |
80d19669 | 2521 | } |
537c00de AD |
2522 | EXPORT_SYMBOL(netif_set_xps_queue); |
2523 | ||
2524 | #endif | |
ffcfe25b AD |
2525 | static void netdev_unbind_all_sb_channels(struct net_device *dev) |
2526 | { | |
2527 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2528 | ||
2529 | /* Unbind any subordinate channels */ | |
2530 | while (txq-- != &dev->_tx[0]) { | |
2531 | if (txq->sb_dev) | |
2532 | netdev_unbind_sb_channel(dev, txq->sb_dev); | |
2533 | } | |
2534 | } | |
2535 | ||
9cf1f6a8 AD |
2536 | void netdev_reset_tc(struct net_device *dev) |
2537 | { | |
6234f874 AD |
2538 | #ifdef CONFIG_XPS |
2539 | netif_reset_xps_queues_gt(dev, 0); | |
2540 | #endif | |
ffcfe25b AD |
2541 | netdev_unbind_all_sb_channels(dev); |
2542 | ||
2543 | /* Reset TC configuration of device */ | |
9cf1f6a8 AD |
2544 | dev->num_tc = 0; |
2545 | memset(dev->tc_to_txq, 0, sizeof(dev->tc_to_txq)); | |
2546 | memset(dev->prio_tc_map, 0, sizeof(dev->prio_tc_map)); | |
2547 | } | |
2548 | EXPORT_SYMBOL(netdev_reset_tc); | |
2549 | ||
2550 | int netdev_set_tc_queue(struct net_device *dev, u8 tc, u16 count, u16 offset) | |
2551 | { | |
2552 | if (tc >= dev->num_tc) | |
2553 | return -EINVAL; | |
2554 | ||
6234f874 AD |
2555 | #ifdef CONFIG_XPS |
2556 | netif_reset_xps_queues(dev, offset, count); | |
2557 | #endif | |
9cf1f6a8 AD |
2558 | dev->tc_to_txq[tc].count = count; |
2559 | dev->tc_to_txq[tc].offset = offset; | |
2560 | return 0; | |
2561 | } | |
2562 | EXPORT_SYMBOL(netdev_set_tc_queue); | |
2563 | ||
2564 | int netdev_set_num_tc(struct net_device *dev, u8 num_tc) | |
2565 | { | |
2566 | if (num_tc > TC_MAX_QUEUE) | |
2567 | return -EINVAL; | |
2568 | ||
6234f874 AD |
2569 | #ifdef CONFIG_XPS |
2570 | netif_reset_xps_queues_gt(dev, 0); | |
2571 | #endif | |
ffcfe25b AD |
2572 | netdev_unbind_all_sb_channels(dev); |
2573 | ||
9cf1f6a8 AD |
2574 | dev->num_tc = num_tc; |
2575 | return 0; | |
2576 | } | |
2577 | EXPORT_SYMBOL(netdev_set_num_tc); | |
2578 | ||
ffcfe25b AD |
2579 | void netdev_unbind_sb_channel(struct net_device *dev, |
2580 | struct net_device *sb_dev) | |
2581 | { | |
2582 | struct netdev_queue *txq = &dev->_tx[dev->num_tx_queues]; | |
2583 | ||
2584 | #ifdef CONFIG_XPS | |
2585 | netif_reset_xps_queues_gt(sb_dev, 0); | |
2586 | #endif | |
2587 | memset(sb_dev->tc_to_txq, 0, sizeof(sb_dev->tc_to_txq)); | |
2588 | memset(sb_dev->prio_tc_map, 0, sizeof(sb_dev->prio_tc_map)); | |
2589 | ||
2590 | while (txq-- != &dev->_tx[0]) { | |
2591 | if (txq->sb_dev == sb_dev) | |
2592 | txq->sb_dev = NULL; | |
2593 | } | |
2594 | } | |
2595 | EXPORT_SYMBOL(netdev_unbind_sb_channel); | |
2596 | ||
2597 | int netdev_bind_sb_channel_queue(struct net_device *dev, | |
2598 | struct net_device *sb_dev, | |
2599 | u8 tc, u16 count, u16 offset) | |
2600 | { | |
2601 | /* Make certain the sb_dev and dev are already configured */ | |
2602 | if (sb_dev->num_tc >= 0 || tc >= dev->num_tc) | |
2603 | return -EINVAL; | |
2604 | ||
2605 | /* We cannot hand out queues we don't have */ | |
2606 | if ((offset + count) > dev->real_num_tx_queues) | |
2607 | return -EINVAL; | |
2608 | ||
2609 | /* Record the mapping */ | |
2610 | sb_dev->tc_to_txq[tc].count = count; | |
2611 | sb_dev->tc_to_txq[tc].offset = offset; | |
2612 | ||
2613 | /* Provide a way for Tx queue to find the tc_to_txq map or | |
2614 | * XPS map for itself. | |
2615 | */ | |
2616 | while (count--) | |
2617 | netdev_get_tx_queue(dev, count + offset)->sb_dev = sb_dev; | |
2618 | ||
2619 | return 0; | |
2620 | } | |
2621 | EXPORT_SYMBOL(netdev_bind_sb_channel_queue); | |
2622 | ||
2623 | int netdev_set_sb_channel(struct net_device *dev, u16 channel) | |
2624 | { | |
2625 | /* Do not use a multiqueue device to represent a subordinate channel */ | |
2626 | if (netif_is_multiqueue(dev)) | |
2627 | return -ENODEV; | |
2628 | ||
2629 | /* We allow channels 1 - 32767 to be used for subordinate channels. | |
2630 | * Channel 0 is meant to be "native" mode and used only to represent | |
2631 | * the main root device. We allow writing 0 to reset the device back | |
2632 | * to normal mode after being used as a subordinate channel. | |
2633 | */ | |
2634 | if (channel > S16_MAX) | |
2635 | return -EINVAL; | |
2636 | ||
2637 | dev->num_tc = -channel; | |
2638 | ||
2639 | return 0; | |
2640 | } | |
2641 | EXPORT_SYMBOL(netdev_set_sb_channel); | |
2642 | ||
f0796d5c JF |
2643 | /* |
2644 | * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues | |
3a053b1a | 2645 | * greater than real_num_tx_queues stale skbs on the qdisc must be flushed. |
f0796d5c | 2646 | */ |
e6484930 | 2647 | int netif_set_real_num_tx_queues(struct net_device *dev, unsigned int txq) |
f0796d5c | 2648 | { |
ac5b7019 | 2649 | bool disabling; |
1d24eb48 TH |
2650 | int rc; |
2651 | ||
ac5b7019 JK |
2652 | disabling = txq < dev->real_num_tx_queues; |
2653 | ||
e6484930 TH |
2654 | if (txq < 1 || txq > dev->num_tx_queues) |
2655 | return -EINVAL; | |
f0796d5c | 2656 | |
5c56580b BH |
2657 | if (dev->reg_state == NETREG_REGISTERED || |
2658 | dev->reg_state == NETREG_UNREGISTERING) { | |
e6484930 TH |
2659 | ASSERT_RTNL(); |
2660 | ||
1d24eb48 TH |
2661 | rc = netdev_queue_update_kobjects(dev, dev->real_num_tx_queues, |
2662 | txq); | |
bf264145 TH |
2663 | if (rc) |
2664 | return rc; | |
2665 | ||
4f57c087 JF |
2666 | if (dev->num_tc) |
2667 | netif_setup_tc(dev, txq); | |
2668 | ||
ac5b7019 JK |
2669 | dev->real_num_tx_queues = txq; |
2670 | ||
2671 | if (disabling) { | |
2672 | synchronize_net(); | |
e6484930 | 2673 | qdisc_reset_all_tx_gt(dev, txq); |
024e9679 AD |
2674 | #ifdef CONFIG_XPS |
2675 | netif_reset_xps_queues_gt(dev, txq); | |
2676 | #endif | |
2677 | } | |
ac5b7019 JK |
2678 | } else { |
2679 | dev->real_num_tx_queues = txq; | |
f0796d5c | 2680 | } |
e6484930 | 2681 | |
e6484930 | 2682 | return 0; |
f0796d5c JF |
2683 | } |
2684 | EXPORT_SYMBOL(netif_set_real_num_tx_queues); | |
56079431 | 2685 | |
a953be53 | 2686 | #ifdef CONFIG_SYSFS |
62fe0b40 BH |
2687 | /** |
2688 | * netif_set_real_num_rx_queues - set actual number of RX queues used | |
2689 | * @dev: Network device | |
2690 | * @rxq: Actual number of RX queues | |
2691 | * | |
2692 | * This must be called either with the rtnl_lock held or before | |
2693 | * registration of the net device. Returns 0 on success, or a | |
4e7f7951 BH |
2694 | * negative error code. If called before registration, it always |
2695 | * succeeds. | |
62fe0b40 BH |
2696 | */ |
2697 | int netif_set_real_num_rx_queues(struct net_device *dev, unsigned int rxq) | |
2698 | { | |
2699 | int rc; | |
2700 | ||
bd25fa7b TH |
2701 | if (rxq < 1 || rxq > dev->num_rx_queues) |
2702 | return -EINVAL; | |
2703 | ||
62fe0b40 BH |
2704 | if (dev->reg_state == NETREG_REGISTERED) { |
2705 | ASSERT_RTNL(); | |
2706 | ||
62fe0b40 BH |
2707 | rc = net_rx_queue_update_kobjects(dev, dev->real_num_rx_queues, |
2708 | rxq); | |
2709 | if (rc) | |
2710 | return rc; | |
62fe0b40 BH |
2711 | } |
2712 | ||
2713 | dev->real_num_rx_queues = rxq; | |
2714 | return 0; | |
2715 | } | |
2716 | EXPORT_SYMBOL(netif_set_real_num_rx_queues); | |
2717 | #endif | |
2718 | ||
2c53040f BH |
2719 | /** |
2720 | * netif_get_num_default_rss_queues - default number of RSS queues | |
16917b87 YM |
2721 | * |
2722 | * This routine should set an upper limit on the number of RSS queues | |
2723 | * used by default by multiqueue devices. | |
2724 | */ | |
a55b138b | 2725 | int netif_get_num_default_rss_queues(void) |
16917b87 | 2726 | { |
40e4e713 HS |
2727 | return is_kdump_kernel() ? |
2728 | 1 : min_t(int, DEFAULT_MAX_NUM_RSS_QUEUES, num_online_cpus()); | |
16917b87 YM |
2729 | } |
2730 | EXPORT_SYMBOL(netif_get_num_default_rss_queues); | |
2731 | ||
3bcb846c | 2732 | static void __netif_reschedule(struct Qdisc *q) |
56079431 | 2733 | { |
def82a1d JP |
2734 | struct softnet_data *sd; |
2735 | unsigned long flags; | |
56079431 | 2736 | |
def82a1d | 2737 | local_irq_save(flags); |
903ceff7 | 2738 | sd = this_cpu_ptr(&softnet_data); |
a9cbd588 CG |
2739 | q->next_sched = NULL; |
2740 | *sd->output_queue_tailp = q; | |
2741 | sd->output_queue_tailp = &q->next_sched; | |
def82a1d JP |
2742 | raise_softirq_irqoff(NET_TX_SOFTIRQ); |
2743 | local_irq_restore(flags); | |
2744 | } | |
2745 | ||
2746 | void __netif_schedule(struct Qdisc *q) | |
2747 | { | |
2748 | if (!test_and_set_bit(__QDISC_STATE_SCHED, &q->state)) | |
2749 | __netif_reschedule(q); | |
56079431 DV |
2750 | } |
2751 | EXPORT_SYMBOL(__netif_schedule); | |
2752 | ||
e6247027 ED |
2753 | struct dev_kfree_skb_cb { |
2754 | enum skb_free_reason reason; | |
2755 | }; | |
2756 | ||
2757 | static struct dev_kfree_skb_cb *get_kfree_skb_cb(const struct sk_buff *skb) | |
56079431 | 2758 | { |
e6247027 ED |
2759 | return (struct dev_kfree_skb_cb *)skb->cb; |
2760 | } | |
2761 | ||
46e5da40 JF |
2762 | void netif_schedule_queue(struct netdev_queue *txq) |
2763 | { | |
2764 | rcu_read_lock(); | |
2765 | if (!(txq->state & QUEUE_STATE_ANY_XOFF)) { | |
2766 | struct Qdisc *q = rcu_dereference(txq->qdisc); | |
2767 | ||
2768 | __netif_schedule(q); | |
2769 | } | |
2770 | rcu_read_unlock(); | |
2771 | } | |
2772 | EXPORT_SYMBOL(netif_schedule_queue); | |
2773 | ||
46e5da40 JF |
2774 | void netif_tx_wake_queue(struct netdev_queue *dev_queue) |
2775 | { | |
2776 | if (test_and_clear_bit(__QUEUE_STATE_DRV_XOFF, &dev_queue->state)) { | |
2777 | struct Qdisc *q; | |
2778 | ||
2779 | rcu_read_lock(); | |
2780 | q = rcu_dereference(dev_queue->qdisc); | |
2781 | __netif_schedule(q); | |
2782 | rcu_read_unlock(); | |
2783 | } | |
2784 | } | |
2785 | EXPORT_SYMBOL(netif_tx_wake_queue); | |
2786 | ||
e6247027 | 2787 | void __dev_kfree_skb_irq(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 | 2788 | { |
e6247027 | 2789 | unsigned long flags; |
56079431 | 2790 | |
9899886d MJ |
2791 | if (unlikely(!skb)) |
2792 | return; | |
2793 | ||
63354797 | 2794 | if (likely(refcount_read(&skb->users) == 1)) { |
e6247027 | 2795 | smp_rmb(); |
63354797 RE |
2796 | refcount_set(&skb->users, 0); |
2797 | } else if (likely(!refcount_dec_and_test(&skb->users))) { | |
e6247027 | 2798 | return; |
bea3348e | 2799 | } |
e6247027 ED |
2800 | get_kfree_skb_cb(skb)->reason = reason; |
2801 | local_irq_save(flags); | |
2802 | skb->next = __this_cpu_read(softnet_data.completion_queue); | |
2803 | __this_cpu_write(softnet_data.completion_queue, skb); | |
2804 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
2805 | local_irq_restore(flags); | |
56079431 | 2806 | } |
e6247027 | 2807 | EXPORT_SYMBOL(__dev_kfree_skb_irq); |
56079431 | 2808 | |
e6247027 | 2809 | void __dev_kfree_skb_any(struct sk_buff *skb, enum skb_free_reason reason) |
56079431 DV |
2810 | { |
2811 | if (in_irq() || irqs_disabled()) | |
e6247027 | 2812 | __dev_kfree_skb_irq(skb, reason); |
56079431 DV |
2813 | else |
2814 | dev_kfree_skb(skb); | |
2815 | } | |
e6247027 | 2816 | EXPORT_SYMBOL(__dev_kfree_skb_any); |
56079431 DV |
2817 | |
2818 | ||
bea3348e SH |
2819 | /** |
2820 | * netif_device_detach - mark device as removed | |
2821 | * @dev: network device | |
2822 | * | |
2823 | * Mark device as removed from system and therefore no longer available. | |
2824 | */ | |
56079431 DV |
2825 | void netif_device_detach(struct net_device *dev) |
2826 | { | |
2827 | if (test_and_clear_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2828 | netif_running(dev)) { | |
d543103a | 2829 | netif_tx_stop_all_queues(dev); |
56079431 DV |
2830 | } |
2831 | } | |
2832 | EXPORT_SYMBOL(netif_device_detach); | |
2833 | ||
bea3348e SH |
2834 | /** |
2835 | * netif_device_attach - mark device as attached | |
2836 | * @dev: network device | |
2837 | * | |
2838 | * Mark device as attached from system and restart if needed. | |
2839 | */ | |
56079431 DV |
2840 | void netif_device_attach(struct net_device *dev) |
2841 | { | |
2842 | if (!test_and_set_bit(__LINK_STATE_PRESENT, &dev->state) && | |
2843 | netif_running(dev)) { | |
d543103a | 2844 | netif_tx_wake_all_queues(dev); |
4ec93edb | 2845 | __netdev_watchdog_up(dev); |
56079431 DV |
2846 | } |
2847 | } | |
2848 | EXPORT_SYMBOL(netif_device_attach); | |
2849 | ||
5605c762 JP |
2850 | /* |
2851 | * Returns a Tx hash based on the given packet descriptor a Tx queues' number | |
2852 | * to be used as a distribution range. | |
2853 | */ | |
eadec877 AD |
2854 | static u16 skb_tx_hash(const struct net_device *dev, |
2855 | const struct net_device *sb_dev, | |
2856 | struct sk_buff *skb) | |
5605c762 JP |
2857 | { |
2858 | u32 hash; | |
2859 | u16 qoffset = 0; | |
1b837d48 | 2860 | u16 qcount = dev->real_num_tx_queues; |
5605c762 | 2861 | |
eadec877 AD |
2862 | if (dev->num_tc) { |
2863 | u8 tc = netdev_get_prio_tc_map(dev, skb->priority); | |
2864 | ||
2865 | qoffset = sb_dev->tc_to_txq[tc].offset; | |
2866 | qcount = sb_dev->tc_to_txq[tc].count; | |
2867 | } | |
2868 | ||
5605c762 JP |
2869 | if (skb_rx_queue_recorded(skb)) { |
2870 | hash = skb_get_rx_queue(skb); | |
1b837d48 AD |
2871 | while (unlikely(hash >= qcount)) |
2872 | hash -= qcount; | |
eadec877 | 2873 | return hash + qoffset; |
5605c762 JP |
2874 | } |
2875 | ||
2876 | return (u16) reciprocal_scale(skb_get_hash(skb), qcount) + qoffset; | |
2877 | } | |
5605c762 | 2878 | |
36c92474 BH |
2879 | static void skb_warn_bad_offload(const struct sk_buff *skb) |
2880 | { | |
84d15ae5 | 2881 | static const netdev_features_t null_features; |
36c92474 | 2882 | struct net_device *dev = skb->dev; |
88ad4175 | 2883 | const char *name = ""; |
36c92474 | 2884 | |
c846ad9b BG |
2885 | if (!net_ratelimit()) |
2886 | return; | |
2887 | ||
88ad4175 BM |
2888 | if (dev) { |
2889 | if (dev->dev.parent) | |
2890 | name = dev_driver_string(dev->dev.parent); | |
2891 | else | |
2892 | name = netdev_name(dev); | |
2893 | } | |
36c92474 BH |
2894 | WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d " |
2895 | "gso_type=%d ip_summed=%d\n", | |
88ad4175 | 2896 | name, dev ? &dev->features : &null_features, |
65e9d2fa | 2897 | skb->sk ? &skb->sk->sk_route_caps : &null_features, |
36c92474 BH |
2898 | skb->len, skb->data_len, skb_shinfo(skb)->gso_size, |
2899 | skb_shinfo(skb)->gso_type, skb->ip_summed); | |
2900 | } | |
2901 | ||
1da177e4 LT |
2902 | /* |
2903 | * Invalidate hardware checksum when packet is to be mangled, and | |
2904 | * complete checksum manually on outgoing path. | |
2905 | */ | |
84fa7933 | 2906 | int skb_checksum_help(struct sk_buff *skb) |
1da177e4 | 2907 | { |
d3bc23e7 | 2908 | __wsum csum; |
663ead3b | 2909 | int ret = 0, offset; |
1da177e4 | 2910 | |
84fa7933 | 2911 | if (skb->ip_summed == CHECKSUM_COMPLETE) |
a430a43d HX |
2912 | goto out_set_summed; |
2913 | ||
2914 | if (unlikely(skb_shinfo(skb)->gso_size)) { | |
36c92474 BH |
2915 | skb_warn_bad_offload(skb); |
2916 | return -EINVAL; | |
1da177e4 LT |
2917 | } |
2918 | ||
cef401de ED |
2919 | /* Before computing a checksum, we should make sure no frag could |
2920 | * be modified by an external entity : checksum could be wrong. | |
2921 | */ | |
2922 | if (skb_has_shared_frag(skb)) { | |
2923 | ret = __skb_linearize(skb); | |
2924 | if (ret) | |
2925 | goto out; | |
2926 | } | |
2927 | ||
55508d60 | 2928 | offset = skb_checksum_start_offset(skb); |
a030847e HX |
2929 | BUG_ON(offset >= skb_headlen(skb)); |
2930 | csum = skb_checksum(skb, offset, skb->len - offset, 0); | |
2931 | ||
2932 | offset += skb->csum_offset; | |
2933 | BUG_ON(offset + sizeof(__sum16) > skb_headlen(skb)); | |
2934 | ||
2935 | if (skb_cloned(skb) && | |
2936 | !skb_clone_writable(skb, offset + sizeof(__sum16))) { | |
1da177e4 LT |
2937 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); |
2938 | if (ret) | |
2939 | goto out; | |
2940 | } | |
2941 | ||
4f2e4ad5 | 2942 | *(__sum16 *)(skb->data + offset) = csum_fold(csum) ?: CSUM_MANGLED_0; |
a430a43d | 2943 | out_set_summed: |
1da177e4 | 2944 | skb->ip_summed = CHECKSUM_NONE; |
4ec93edb | 2945 | out: |
1da177e4 LT |
2946 | return ret; |
2947 | } | |
d1b19dff | 2948 | EXPORT_SYMBOL(skb_checksum_help); |
1da177e4 | 2949 | |
b72b5bf6 DC |
2950 | int skb_crc32c_csum_help(struct sk_buff *skb) |
2951 | { | |
2952 | __le32 crc32c_csum; | |
2953 | int ret = 0, offset, start; | |
2954 | ||
2955 | if (skb->ip_summed != CHECKSUM_PARTIAL) | |
2956 | goto out; | |
2957 | ||
2958 | if (unlikely(skb_is_gso(skb))) | |
2959 | goto out; | |
2960 | ||
2961 | /* Before computing a checksum, we should make sure no frag could | |
2962 | * be modified by an external entity : checksum could be wrong. | |
2963 | */ | |
2964 | if (unlikely(skb_has_shared_frag(skb))) { | |
2965 | ret = __skb_linearize(skb); | |
2966 | if (ret) | |
2967 | goto out; | |
2968 | } | |
2969 | start = skb_checksum_start_offset(skb); | |
2970 | offset = start + offsetof(struct sctphdr, checksum); | |
2971 | if (WARN_ON_ONCE(offset >= skb_headlen(skb))) { | |
2972 | ret = -EINVAL; | |
2973 | goto out; | |
2974 | } | |
2975 | if (skb_cloned(skb) && | |
2976 | !skb_clone_writable(skb, offset + sizeof(__le32))) { | |
2977 | ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC); | |
2978 | if (ret) | |
2979 | goto out; | |
2980 | } | |
2981 | crc32c_csum = cpu_to_le32(~__skb_checksum(skb, start, | |
2982 | skb->len - start, ~(__u32)0, | |
2983 | crc32c_csum_stub)); | |
2984 | *(__le32 *)(skb->data + offset) = crc32c_csum; | |
2985 | skb->ip_summed = CHECKSUM_NONE; | |
dba00306 | 2986 | skb->csum_not_inet = 0; |
b72b5bf6 DC |
2987 | out: |
2988 | return ret; | |
2989 | } | |
2990 | ||
53d6471c | 2991 | __be16 skb_network_protocol(struct sk_buff *skb, int *depth) |
f6a78bfc | 2992 | { |
252e3346 | 2993 | __be16 type = skb->protocol; |
f6a78bfc | 2994 | |
19acc327 PS |
2995 | /* Tunnel gso handlers can set protocol to ethernet. */ |
2996 | if (type == htons(ETH_P_TEB)) { | |
2997 | struct ethhdr *eth; | |
2998 | ||
2999 | if (unlikely(!pskb_may_pull(skb, sizeof(struct ethhdr)))) | |
3000 | return 0; | |
3001 | ||
1dfe82eb | 3002 | eth = (struct ethhdr *)skb->data; |
19acc327 PS |
3003 | type = eth->h_proto; |
3004 | } | |
3005 | ||
d4bcef3f | 3006 | return __vlan_get_protocol(skb, type, depth); |
ec5f0615 PS |
3007 | } |
3008 | ||
3009 | /** | |
3010 | * skb_mac_gso_segment - mac layer segmentation handler. | |
3011 | * @skb: buffer to segment | |
3012 | * @features: features for the output path (see dev->features) | |
3013 | */ | |
3014 | struct sk_buff *skb_mac_gso_segment(struct sk_buff *skb, | |
3015 | netdev_features_t features) | |
3016 | { | |
3017 | struct sk_buff *segs = ERR_PTR(-EPROTONOSUPPORT); | |
3018 | struct packet_offload *ptype; | |
53d6471c VY |
3019 | int vlan_depth = skb->mac_len; |
3020 | __be16 type = skb_network_protocol(skb, &vlan_depth); | |
ec5f0615 PS |
3021 | |
3022 | if (unlikely(!type)) | |
3023 | return ERR_PTR(-EINVAL); | |
3024 | ||
53d6471c | 3025 | __skb_pull(skb, vlan_depth); |
f6a78bfc HX |
3026 | |
3027 | rcu_read_lock(); | |
22061d80 | 3028 | list_for_each_entry_rcu(ptype, &offload_base, list) { |
f191a1d1 | 3029 | if (ptype->type == type && ptype->callbacks.gso_segment) { |
f191a1d1 | 3030 | segs = ptype->callbacks.gso_segment(skb, features); |
f6a78bfc HX |
3031 | break; |
3032 | } | |
3033 | } | |
3034 | rcu_read_unlock(); | |
3035 | ||
98e399f8 | 3036 | __skb_push(skb, skb->data - skb_mac_header(skb)); |
576a30eb | 3037 | |
f6a78bfc HX |
3038 | return segs; |
3039 | } | |
05e8ef4a PS |
3040 | EXPORT_SYMBOL(skb_mac_gso_segment); |
3041 | ||
3042 | ||
3043 | /* openvswitch calls this on rx path, so we need a different check. | |
3044 | */ | |
3045 | static inline bool skb_needs_check(struct sk_buff *skb, bool tx_path) | |
3046 | { | |
3047 | if (tx_path) | |
0c19f846 WB |
3048 | return skb->ip_summed != CHECKSUM_PARTIAL && |
3049 | skb->ip_summed != CHECKSUM_UNNECESSARY; | |
6e7bc478 ED |
3050 | |
3051 | return skb->ip_summed == CHECKSUM_NONE; | |
05e8ef4a PS |
3052 | } |
3053 | ||
3054 | /** | |
3055 | * __skb_gso_segment - Perform segmentation on skb. | |
3056 | * @skb: buffer to segment | |
3057 | * @features: features for the output path (see dev->features) | |
3058 | * @tx_path: whether it is called in TX path | |
3059 | * | |
3060 | * This function segments the given skb and returns a list of segments. | |
3061 | * | |
3062 | * It may return NULL if the skb requires no segmentation. This is | |
3063 | * only possible when GSO is used for verifying header integrity. | |
9207f9d4 KK |
3064 | * |
3065 | * Segmentation preserves SKB_SGO_CB_OFFSET bytes of previous skb cb. | |
05e8ef4a PS |
3066 | */ |
3067 | struct sk_buff *__skb_gso_segment(struct sk_buff *skb, | |
3068 | netdev_features_t features, bool tx_path) | |
3069 | { | |
b2504a5d ED |
3070 | struct sk_buff *segs; |
3071 | ||
05e8ef4a PS |
3072 | if (unlikely(skb_needs_check(skb, tx_path))) { |
3073 | int err; | |
3074 | ||
b2504a5d | 3075 | /* We're going to init ->check field in TCP or UDP header */ |
a40e0a66 | 3076 | err = skb_cow_head(skb, 0); |
3077 | if (err < 0) | |
05e8ef4a PS |
3078 | return ERR_PTR(err); |
3079 | } | |
3080 | ||
802ab55a AD |
3081 | /* Only report GSO partial support if it will enable us to |
3082 | * support segmentation on this frame without needing additional | |
3083 | * work. | |
3084 | */ | |
3085 | if (features & NETIF_F_GSO_PARTIAL) { | |
3086 | netdev_features_t partial_features = NETIF_F_GSO_ROBUST; | |
3087 | struct net_device *dev = skb->dev; | |
3088 | ||
3089 | partial_features |= dev->features & dev->gso_partial_features; | |
3090 | if (!skb_gso_ok(skb, features | partial_features)) | |
3091 | features &= ~NETIF_F_GSO_PARTIAL; | |
3092 | } | |
3093 | ||
9207f9d4 KK |
3094 | BUILD_BUG_ON(SKB_SGO_CB_OFFSET + |
3095 | sizeof(*SKB_GSO_CB(skb)) > sizeof(skb->cb)); | |
3096 | ||
68c33163 | 3097 | SKB_GSO_CB(skb)->mac_offset = skb_headroom(skb); |
3347c960 ED |
3098 | SKB_GSO_CB(skb)->encap_level = 0; |
3099 | ||
05e8ef4a PS |
3100 | skb_reset_mac_header(skb); |
3101 | skb_reset_mac_len(skb); | |
3102 | ||
b2504a5d ED |
3103 | segs = skb_mac_gso_segment(skb, features); |
3104 | ||
8d74e9f8 | 3105 | if (unlikely(skb_needs_check(skb, tx_path) && !IS_ERR(segs))) |
b2504a5d ED |
3106 | skb_warn_bad_offload(skb); |
3107 | ||
3108 | return segs; | |
05e8ef4a | 3109 | } |
12b0004d | 3110 | EXPORT_SYMBOL(__skb_gso_segment); |
f6a78bfc | 3111 | |
fb286bb2 HX |
3112 | /* Take action when hardware reception checksum errors are detected. */ |
3113 | #ifdef CONFIG_BUG | |
7fe50ac8 | 3114 | void netdev_rx_csum_fault(struct net_device *dev, struct sk_buff *skb) |
fb286bb2 HX |
3115 | { |
3116 | if (net_ratelimit()) { | |
7b6cd1ce | 3117 | pr_err("%s: hw csum failure\n", dev ? dev->name : "<unknown>"); |
7fe50ac8 CW |
3118 | if (dev) |
3119 | pr_err("dev features: %pNF\n", &dev->features); | |
3120 | pr_err("skb len=%u data_len=%u pkt_type=%u gso_size=%u gso_type=%u nr_frags=%u ip_summed=%u csum=%x csum_complete_sw=%d csum_valid=%d csum_level=%u\n", | |
3121 | skb->len, skb->data_len, skb->pkt_type, | |
3122 | skb_shinfo(skb)->gso_size, skb_shinfo(skb)->gso_type, | |
3123 | skb_shinfo(skb)->nr_frags, skb->ip_summed, skb->csum, | |
3124 | skb->csum_complete_sw, skb->csum_valid, skb->csum_level); | |
fb286bb2 HX |
3125 | dump_stack(); |
3126 | } | |
3127 | } | |
3128 | EXPORT_SYMBOL(netdev_rx_csum_fault); | |
3129 | #endif | |
3130 | ||
ab74cfeb | 3131 | /* XXX: check that highmem exists at all on the given machine. */ |
c1e756bf | 3132 | static int illegal_highdma(struct net_device *dev, struct sk_buff *skb) |
1da177e4 | 3133 | { |
3d3a8533 | 3134 | #ifdef CONFIG_HIGHMEM |
1da177e4 | 3135 | int i; |
f4563a75 | 3136 | |
5acbbd42 | 3137 | if (!(dev->features & NETIF_F_HIGHDMA)) { |
ea2ab693 IC |
3138 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { |
3139 | skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
f4563a75 | 3140 | |
ea2ab693 | 3141 | if (PageHighMem(skb_frag_page(frag))) |
5acbbd42 | 3142 | return 1; |
ea2ab693 | 3143 | } |
5acbbd42 | 3144 | } |
3d3a8533 | 3145 | #endif |
1da177e4 LT |
3146 | return 0; |
3147 | } | |
1da177e4 | 3148 | |
3b392ddb SH |
3149 | /* If MPLS offload request, verify we are testing hardware MPLS features |
3150 | * instead of standard features for the netdev. | |
3151 | */ | |
d0edc7bf | 3152 | #if IS_ENABLED(CONFIG_NET_MPLS_GSO) |
3b392ddb SH |
3153 | static netdev_features_t net_mpls_features(struct sk_buff *skb, |
3154 | netdev_features_t features, | |
3155 | __be16 type) | |
3156 | { | |
25cd9ba0 | 3157 | if (eth_p_mpls(type)) |
3b392ddb SH |
3158 | features &= skb->dev->mpls_features; |
3159 | ||
3160 | return features; | |
3161 | } | |
3162 | #else | |
3163 | static netdev_features_t net_mpls_features(struct sk_buff *skb, | |
3164 | netdev_features_t features, | |
3165 | __be16 type) | |
3166 | { | |
3167 | return features; | |
3168 | } | |
3169 | #endif | |
3170 | ||
c8f44aff | 3171 | static netdev_features_t harmonize_features(struct sk_buff *skb, |
c1e756bf | 3172 | netdev_features_t features) |
f01a5236 | 3173 | { |
53d6471c | 3174 | int tmp; |
3b392ddb SH |
3175 | __be16 type; |
3176 | ||
3177 | type = skb_network_protocol(skb, &tmp); | |
3178 | features = net_mpls_features(skb, features, type); | |
53d6471c | 3179 | |
c0d680e5 | 3180 | if (skb->ip_summed != CHECKSUM_NONE && |
3b392ddb | 3181 | !can_checksum_protocol(features, type)) { |
996e8021 | 3182 | features &= ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK); |
f01a5236 | 3183 | } |
7be2c82c ED |
3184 | if (illegal_highdma(skb->dev, skb)) |
3185 | features &= ~NETIF_F_SG; | |
f01a5236 JG |
3186 | |
3187 | return features; | |
3188 | } | |
3189 | ||
e38f3025 TM |
3190 | netdev_features_t passthru_features_check(struct sk_buff *skb, |
3191 | struct net_device *dev, | |
3192 | netdev_features_t features) | |
3193 | { | |
3194 | return features; | |
3195 | } | |
3196 | EXPORT_SYMBOL(passthru_features_check); | |
3197 | ||
7ce23672 | 3198 | static netdev_features_t dflt_features_check(struct sk_buff *skb, |
8cb65d00 TM |
3199 | struct net_device *dev, |
3200 | netdev_features_t features) | |
3201 | { | |
3202 | return vlan_features_check(skb, features); | |
3203 | } | |
3204 | ||
cbc53e08 AD |
3205 | static netdev_features_t gso_features_check(const struct sk_buff *skb, |
3206 | struct net_device *dev, | |
3207 | netdev_features_t features) | |
3208 | { | |
3209 | u16 gso_segs = skb_shinfo(skb)->gso_segs; | |
3210 | ||
3211 | if (gso_segs > dev->gso_max_segs) | |
3212 | return features & ~NETIF_F_GSO_MASK; | |
3213 | ||
802ab55a AD |
3214 | /* Support for GSO partial features requires software |
3215 | * intervention before we can actually process the packets | |
3216 | * so we need to strip support for any partial features now | |
3217 | * and we can pull them back in after we have partially | |
3218 | * segmented the frame. | |
3219 | */ | |
3220 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)) | |
3221 | features &= ~dev->gso_partial_features; | |
3222 | ||
3223 | /* Make sure to clear the IPv4 ID mangling feature if the | |
3224 | * IPv4 header has the potential to be fragmented. | |
cbc53e08 AD |
3225 | */ |
3226 | if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) { | |
3227 | struct iphdr *iph = skb->encapsulation ? | |
3228 | inner_ip_hdr(skb) : ip_hdr(skb); | |
3229 | ||
3230 | if (!(iph->frag_off & htons(IP_DF))) | |
3231 | features &= ~NETIF_F_TSO_MANGLEID; | |
3232 | } | |
3233 | ||
3234 | return features; | |
3235 | } | |
3236 | ||
c1e756bf | 3237 | netdev_features_t netif_skb_features(struct sk_buff *skb) |
58e998c6 | 3238 | { |
5f35227e | 3239 | struct net_device *dev = skb->dev; |
fcbeb976 | 3240 | netdev_features_t features = dev->features; |
58e998c6 | 3241 | |
cbc53e08 AD |
3242 | if (skb_is_gso(skb)) |
3243 | features = gso_features_check(skb, dev, features); | |
30b678d8 | 3244 | |
5f35227e JG |
3245 | /* If encapsulation offload request, verify we are testing |
3246 | * hardware encapsulation features instead of standard | |
3247 | * features for the netdev | |
3248 | */ | |
3249 | if (skb->encapsulation) | |
3250 | features &= dev->hw_enc_features; | |
3251 | ||
f5a7fb88 TM |
3252 | if (skb_vlan_tagged(skb)) |
3253 | features = netdev_intersect_features(features, | |
3254 | dev->vlan_features | | |
3255 | NETIF_F_HW_VLAN_CTAG_TX | | |
3256 | NETIF_F_HW_VLAN_STAG_TX); | |
f01a5236 | 3257 | |
5f35227e JG |
3258 | if (dev->netdev_ops->ndo_features_check) |
3259 | features &= dev->netdev_ops->ndo_features_check(skb, dev, | |
3260 | features); | |
8cb65d00 TM |
3261 | else |
3262 | features &= dflt_features_check(skb, dev, features); | |
5f35227e | 3263 | |
c1e756bf | 3264 | return harmonize_features(skb, features); |
58e998c6 | 3265 | } |
c1e756bf | 3266 | EXPORT_SYMBOL(netif_skb_features); |
58e998c6 | 3267 | |
2ea25513 | 3268 | static int xmit_one(struct sk_buff *skb, struct net_device *dev, |
95f6b3dd | 3269 | struct netdev_queue *txq, bool more) |
f6a78bfc | 3270 | { |
2ea25513 DM |
3271 | unsigned int len; |
3272 | int rc; | |
00829823 | 3273 | |
9f9a742d | 3274 | if (dev_nit_active(dev)) |
2ea25513 | 3275 | dev_queue_xmit_nit(skb, dev); |
fc741216 | 3276 | |
2ea25513 DM |
3277 | len = skb->len; |
3278 | trace_net_dev_start_xmit(skb, dev); | |
95f6b3dd | 3279 | rc = netdev_start_xmit(skb, dev, txq, more); |
2ea25513 | 3280 | trace_net_dev_xmit(skb, rc, dev, len); |
adf30907 | 3281 | |
2ea25513 DM |
3282 | return rc; |
3283 | } | |
7b9c6090 | 3284 | |
8dcda22a DM |
3285 | struct sk_buff *dev_hard_start_xmit(struct sk_buff *first, struct net_device *dev, |
3286 | struct netdev_queue *txq, int *ret) | |
7f2e870f DM |
3287 | { |
3288 | struct sk_buff *skb = first; | |
3289 | int rc = NETDEV_TX_OK; | |
7b9c6090 | 3290 | |
7f2e870f DM |
3291 | while (skb) { |
3292 | struct sk_buff *next = skb->next; | |
fc70fb64 | 3293 | |
a8305bff | 3294 | skb_mark_not_on_list(skb); |
95f6b3dd | 3295 | rc = xmit_one(skb, dev, txq, next != NULL); |
7f2e870f DM |
3296 | if (unlikely(!dev_xmit_complete(rc))) { |
3297 | skb->next = next; | |
3298 | goto out; | |
3299 | } | |
6afff0ca | 3300 | |
7f2e870f | 3301 | skb = next; |
fe60faa5 | 3302 | if (netif_tx_queue_stopped(txq) && skb) { |
7f2e870f DM |
3303 | rc = NETDEV_TX_BUSY; |
3304 | break; | |
9ccb8975 | 3305 | } |
7f2e870f | 3306 | } |
9ccb8975 | 3307 | |
7f2e870f DM |
3308 | out: |
3309 | *ret = rc; | |
3310 | return skb; | |
3311 | } | |
b40863c6 | 3312 | |
1ff0dc94 ED |
3313 | static struct sk_buff *validate_xmit_vlan(struct sk_buff *skb, |
3314 | netdev_features_t features) | |
f6a78bfc | 3315 | { |
df8a39de | 3316 | if (skb_vlan_tag_present(skb) && |
5968250c JP |
3317 | !vlan_hw_offload_capable(features, skb->vlan_proto)) |
3318 | skb = __vlan_hwaccel_push_inside(skb); | |
eae3f88e DM |
3319 | return skb; |
3320 | } | |
f6a78bfc | 3321 | |
43c26a1a DC |
3322 | int skb_csum_hwoffload_help(struct sk_buff *skb, |
3323 | const netdev_features_t features) | |
3324 | { | |
3325 | if (unlikely(skb->csum_not_inet)) | |
3326 | return !!(features & NETIF_F_SCTP_CRC) ? 0 : | |
3327 | skb_crc32c_csum_help(skb); | |
3328 | ||
3329 | return !!(features & NETIF_F_CSUM_MASK) ? 0 : skb_checksum_help(skb); | |
3330 | } | |
3331 | EXPORT_SYMBOL(skb_csum_hwoffload_help); | |
3332 | ||
f53c7239 | 3333 | static struct sk_buff *validate_xmit_skb(struct sk_buff *skb, struct net_device *dev, bool *again) |
eae3f88e DM |
3334 | { |
3335 | netdev_features_t features; | |
f6a78bfc | 3336 | |
eae3f88e DM |
3337 | features = netif_skb_features(skb); |
3338 | skb = validate_xmit_vlan(skb, features); | |
3339 | if (unlikely(!skb)) | |
3340 | goto out_null; | |
7b9c6090 | 3341 | |
ebf4e808 IL |
3342 | skb = sk_validate_xmit_skb(skb, dev); |
3343 | if (unlikely(!skb)) | |
3344 | goto out_null; | |
3345 | ||
8b86a61d | 3346 | if (netif_needs_gso(skb, features)) { |
ce93718f DM |
3347 | struct sk_buff *segs; |
3348 | ||
3349 | segs = skb_gso_segment(skb, features); | |
cecda693 | 3350 | if (IS_ERR(segs)) { |
af6dabc9 | 3351 | goto out_kfree_skb; |
cecda693 JW |
3352 | } else if (segs) { |
3353 | consume_skb(skb); | |
3354 | skb = segs; | |
f6a78bfc | 3355 | } |
eae3f88e DM |
3356 | } else { |
3357 | if (skb_needs_linearize(skb, features) && | |
3358 | __skb_linearize(skb)) | |
3359 | goto out_kfree_skb; | |
4ec93edb | 3360 | |
eae3f88e DM |
3361 | /* If packet is not checksummed and device does not |
3362 | * support checksumming for this protocol, complete | |
3363 | * checksumming here. | |
3364 | */ | |
3365 | if (skb->ip_summed == CHECKSUM_PARTIAL) { | |
3366 | if (skb->encapsulation) | |
3367 | skb_set_inner_transport_header(skb, | |
3368 | skb_checksum_start_offset(skb)); | |
3369 | else | |
3370 | skb_set_transport_header(skb, | |
3371 | skb_checksum_start_offset(skb)); | |
43c26a1a | 3372 | if (skb_csum_hwoffload_help(skb, features)) |
eae3f88e | 3373 | goto out_kfree_skb; |
7b9c6090 | 3374 | } |
0c772159 | 3375 | } |
7b9c6090 | 3376 | |
f53c7239 | 3377 | skb = validate_xmit_xfrm(skb, features, again); |
3dca3f38 | 3378 | |
eae3f88e | 3379 | return skb; |
fc70fb64 | 3380 | |
f6a78bfc HX |
3381 | out_kfree_skb: |
3382 | kfree_skb(skb); | |
eae3f88e | 3383 | out_null: |
d21fd63e | 3384 | atomic_long_inc(&dev->tx_dropped); |
eae3f88e DM |
3385 | return NULL; |
3386 | } | |
6afff0ca | 3387 | |
f53c7239 | 3388 | struct sk_buff *validate_xmit_skb_list(struct sk_buff *skb, struct net_device *dev, bool *again) |
55a93b3e ED |
3389 | { |
3390 | struct sk_buff *next, *head = NULL, *tail; | |
3391 | ||
bec3cfdc | 3392 | for (; skb != NULL; skb = next) { |
55a93b3e | 3393 | next = skb->next; |
a8305bff | 3394 | skb_mark_not_on_list(skb); |
bec3cfdc ED |
3395 | |
3396 | /* in case skb wont be segmented, point to itself */ | |
3397 | skb->prev = skb; | |
3398 | ||
f53c7239 | 3399 | skb = validate_xmit_skb(skb, dev, again); |
bec3cfdc ED |
3400 | if (!skb) |
3401 | continue; | |
55a93b3e | 3402 | |
bec3cfdc ED |
3403 | if (!head) |
3404 | head = skb; | |
3405 | else | |
3406 | tail->next = skb; | |
3407 | /* If skb was segmented, skb->prev points to | |
3408 | * the last segment. If not, it still contains skb. | |
3409 | */ | |
3410 | tail = skb->prev; | |
55a93b3e ED |
3411 | } |
3412 | return head; | |
f6a78bfc | 3413 | } |
104ba78c | 3414 | EXPORT_SYMBOL_GPL(validate_xmit_skb_list); |
f6a78bfc | 3415 | |
1def9238 ED |
3416 | static void qdisc_pkt_len_init(struct sk_buff *skb) |
3417 | { | |
3418 | const struct skb_shared_info *shinfo = skb_shinfo(skb); | |
3419 | ||
3420 | qdisc_skb_cb(skb)->pkt_len = skb->len; | |
3421 | ||
3422 | /* To get more precise estimation of bytes sent on wire, | |
3423 | * we add to pkt_len the headers size of all segments | |
3424 | */ | |
a0dce875 | 3425 | if (shinfo->gso_size && skb_transport_header_was_set(skb)) { |
757b8b1d | 3426 | unsigned int hdr_len; |
15e5a030 | 3427 | u16 gso_segs = shinfo->gso_segs; |
1def9238 | 3428 | |
757b8b1d ED |
3429 | /* mac layer + network layer */ |
3430 | hdr_len = skb_transport_header(skb) - skb_mac_header(skb); | |
3431 | ||
3432 | /* + transport layer */ | |
7c68d1a6 ED |
3433 | if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) { |
3434 | const struct tcphdr *th; | |
3435 | struct tcphdr _tcphdr; | |
3436 | ||
3437 | th = skb_header_pointer(skb, skb_transport_offset(skb), | |
3438 | sizeof(_tcphdr), &_tcphdr); | |
3439 | if (likely(th)) | |
3440 | hdr_len += __tcp_hdrlen(th); | |
3441 | } else { | |
3442 | struct udphdr _udphdr; | |
3443 | ||
3444 | if (skb_header_pointer(skb, skb_transport_offset(skb), | |
3445 | sizeof(_udphdr), &_udphdr)) | |
3446 | hdr_len += sizeof(struct udphdr); | |
3447 | } | |
15e5a030 JW |
3448 | |
3449 | if (shinfo->gso_type & SKB_GSO_DODGY) | |
3450 | gso_segs = DIV_ROUND_UP(skb->len - hdr_len, | |
3451 | shinfo->gso_size); | |
3452 | ||
3453 | qdisc_skb_cb(skb)->pkt_len += (gso_segs - 1) * hdr_len; | |
1def9238 ED |
3454 | } |
3455 | } | |
3456 | ||
bbd8a0d3 KK |
3457 | static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q, |
3458 | struct net_device *dev, | |
3459 | struct netdev_queue *txq) | |
3460 | { | |
3461 | spinlock_t *root_lock = qdisc_lock(q); | |
520ac30f | 3462 | struct sk_buff *to_free = NULL; |
a2da570d | 3463 | bool contended; |
bbd8a0d3 KK |
3464 | int rc; |
3465 | ||
a2da570d | 3466 | qdisc_calculate_pkt_len(skb, q); |
6b3ba914 JF |
3467 | |
3468 | if (q->flags & TCQ_F_NOLOCK) { | |
3469 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
3470 | __qdisc_drop(skb, &to_free); | |
3471 | rc = NET_XMIT_DROP; | |
ba27b4cd PA |
3472 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && q->empty && |
3473 | qdisc_run_begin(q)) { | |
3474 | qdisc_bstats_cpu_update(q, skb); | |
3475 | ||
3476 | if (sch_direct_xmit(skb, q, dev, txq, NULL, true)) | |
3477 | __qdisc_run(q); | |
3478 | ||
3479 | qdisc_run_end(q); | |
3480 | rc = NET_XMIT_SUCCESS; | |
6b3ba914 JF |
3481 | } else { |
3482 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; | |
32f7b44d | 3483 | qdisc_run(q); |
6b3ba914 JF |
3484 | } |
3485 | ||
3486 | if (unlikely(to_free)) | |
3487 | kfree_skb_list(to_free); | |
3488 | return rc; | |
3489 | } | |
3490 | ||
79640a4c ED |
3491 | /* |
3492 | * Heuristic to force contended enqueues to serialize on a | |
3493 | * separate lock before trying to get qdisc main lock. | |
f9eb8aea | 3494 | * This permits qdisc->running owner to get the lock more |
9bf2b8c2 | 3495 | * often and dequeue packets faster. |
79640a4c | 3496 | */ |
a2da570d | 3497 | contended = qdisc_is_running(q); |
79640a4c ED |
3498 | if (unlikely(contended)) |
3499 | spin_lock(&q->busylock); | |
3500 | ||
bbd8a0d3 KK |
3501 | spin_lock(root_lock); |
3502 | if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED, &q->state))) { | |
520ac30f | 3503 | __qdisc_drop(skb, &to_free); |
bbd8a0d3 KK |
3504 | rc = NET_XMIT_DROP; |
3505 | } else if ((q->flags & TCQ_F_CAN_BYPASS) && !qdisc_qlen(q) && | |
bc135b23 | 3506 | qdisc_run_begin(q)) { |
bbd8a0d3 KK |
3507 | /* |
3508 | * This is a work-conserving queue; there are no old skbs | |
3509 | * waiting to be sent out; and the qdisc is not running - | |
3510 | * xmit the skb directly. | |
3511 | */ | |
bfe0d029 | 3512 | |
bfe0d029 ED |
3513 | qdisc_bstats_update(q, skb); |
3514 | ||
55a93b3e | 3515 | if (sch_direct_xmit(skb, q, dev, txq, root_lock, true)) { |
79640a4c ED |
3516 | if (unlikely(contended)) { |
3517 | spin_unlock(&q->busylock); | |
3518 | contended = false; | |
3519 | } | |
bbd8a0d3 | 3520 | __qdisc_run(q); |
6c148184 | 3521 | } |
bbd8a0d3 | 3522 | |
6c148184 | 3523 | qdisc_run_end(q); |
bbd8a0d3 KK |
3524 | rc = NET_XMIT_SUCCESS; |
3525 | } else { | |
520ac30f | 3526 | rc = q->enqueue(skb, q, &to_free) & NET_XMIT_MASK; |
79640a4c ED |
3527 | if (qdisc_run_begin(q)) { |
3528 | if (unlikely(contended)) { | |
3529 | spin_unlock(&q->busylock); | |
3530 | contended = false; | |
3531 | } | |
3532 | __qdisc_run(q); | |
6c148184 | 3533 | qdisc_run_end(q); |
79640a4c | 3534 | } |
bbd8a0d3 KK |
3535 | } |
3536 | spin_unlock(root_lock); | |
520ac30f ED |
3537 | if (unlikely(to_free)) |
3538 | kfree_skb_list(to_free); | |
79640a4c ED |
3539 | if (unlikely(contended)) |
3540 | spin_unlock(&q->busylock); | |
bbd8a0d3 KK |
3541 | return rc; |
3542 | } | |
3543 | ||
86f8515f | 3544 | #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO) |
5bc1421e NH |
3545 | static void skb_update_prio(struct sk_buff *skb) |
3546 | { | |
4dcb31d4 ED |
3547 | const struct netprio_map *map; |
3548 | const struct sock *sk; | |
3549 | unsigned int prioidx; | |
5bc1421e | 3550 | |
4dcb31d4 ED |
3551 | if (skb->priority) |
3552 | return; | |
3553 | map = rcu_dereference_bh(skb->dev->priomap); | |
3554 | if (!map) | |
3555 | return; | |
3556 | sk = skb_to_full_sk(skb); | |
3557 | if (!sk) | |
3558 | return; | |
91c68ce2 | 3559 | |
4dcb31d4 ED |
3560 | prioidx = sock_cgroup_prioidx(&sk->sk_cgrp_data); |
3561 | ||
3562 | if (prioidx < map->priomap_len) | |
3563 | skb->priority = map->priomap[prioidx]; | |
5bc1421e NH |
3564 | } |
3565 | #else | |
3566 | #define skb_update_prio(skb) | |
3567 | #endif | |
3568 | ||
95603e22 MM |
3569 | /** |
3570 | * dev_loopback_xmit - loop back @skb | |
0c4b51f0 EB |
3571 | * @net: network namespace this loopback is happening in |
3572 | * @sk: sk needed to be a netfilter okfn | |
95603e22 MM |
3573 | * @skb: buffer to transmit |
3574 | */ | |
0c4b51f0 | 3575 | int dev_loopback_xmit(struct net *net, struct sock *sk, struct sk_buff *skb) |
95603e22 MM |
3576 | { |
3577 | skb_reset_mac_header(skb); | |
3578 | __skb_pull(skb, skb_network_offset(skb)); | |
3579 | skb->pkt_type = PACKET_LOOPBACK; | |
3580 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
3581 | WARN_ON(!skb_dst(skb)); | |
3582 | skb_dst_force(skb); | |
3583 | netif_rx_ni(skb); | |
3584 | return 0; | |
3585 | } | |
3586 | EXPORT_SYMBOL(dev_loopback_xmit); | |
3587 | ||
1f211a1b DB |
3588 | #ifdef CONFIG_NET_EGRESS |
3589 | static struct sk_buff * | |
3590 | sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev) | |
3591 | { | |
46209401 | 3592 | struct mini_Qdisc *miniq = rcu_dereference_bh(dev->miniq_egress); |
1f211a1b DB |
3593 | struct tcf_result cl_res; |
3594 | ||
46209401 | 3595 | if (!miniq) |
1f211a1b DB |
3596 | return skb; |
3597 | ||
8dc07fdb | 3598 | /* qdisc_skb_cb(skb)->pkt_len was already set by the caller. */ |
46209401 | 3599 | mini_qdisc_bstats_cpu_update(miniq, skb); |
1f211a1b | 3600 | |
46209401 | 3601 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
1f211a1b DB |
3602 | case TC_ACT_OK: |
3603 | case TC_ACT_RECLASSIFY: | |
3604 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
3605 | break; | |
3606 | case TC_ACT_SHOT: | |
46209401 | 3607 | mini_qdisc_qstats_cpu_drop(miniq); |
1f211a1b | 3608 | *ret = NET_XMIT_DROP; |
7e2c3aea DB |
3609 | kfree_skb(skb); |
3610 | return NULL; | |
1f211a1b DB |
3611 | case TC_ACT_STOLEN: |
3612 | case TC_ACT_QUEUED: | |
e25ea21f | 3613 | case TC_ACT_TRAP: |
1f211a1b | 3614 | *ret = NET_XMIT_SUCCESS; |
7e2c3aea | 3615 | consume_skb(skb); |
1f211a1b DB |
3616 | return NULL; |
3617 | case TC_ACT_REDIRECT: | |
3618 | /* No need to push/pop skb's mac_header here on egress! */ | |
3619 | skb_do_redirect(skb); | |
3620 | *ret = NET_XMIT_SUCCESS; | |
3621 | return NULL; | |
3622 | default: | |
3623 | break; | |
3624 | } | |
3625 | ||
3626 | return skb; | |
3627 | } | |
3628 | #endif /* CONFIG_NET_EGRESS */ | |
3629 | ||
fc9bab24 AN |
3630 | #ifdef CONFIG_XPS |
3631 | static int __get_xps_queue_idx(struct net_device *dev, struct sk_buff *skb, | |
3632 | struct xps_dev_maps *dev_maps, unsigned int tci) | |
3633 | { | |
3634 | struct xps_map *map; | |
3635 | int queue_index = -1; | |
3636 | ||
3637 | if (dev->num_tc) { | |
3638 | tci *= dev->num_tc; | |
3639 | tci += netdev_get_prio_tc_map(dev, skb->priority); | |
3640 | } | |
3641 | ||
3642 | map = rcu_dereference(dev_maps->attr_map[tci]); | |
3643 | if (map) { | |
3644 | if (map->len == 1) | |
3645 | queue_index = map->queues[0]; | |
3646 | else | |
3647 | queue_index = map->queues[reciprocal_scale( | |
3648 | skb_get_hash(skb), map->len)]; | |
3649 | if (unlikely(queue_index >= dev->real_num_tx_queues)) | |
3650 | queue_index = -1; | |
3651 | } | |
3652 | return queue_index; | |
3653 | } | |
3654 | #endif | |
3655 | ||
eadec877 AD |
3656 | static int get_xps_queue(struct net_device *dev, struct net_device *sb_dev, |
3657 | struct sk_buff *skb) | |
638b2a69 JP |
3658 | { |
3659 | #ifdef CONFIG_XPS | |
3660 | struct xps_dev_maps *dev_maps; | |
fc9bab24 | 3661 | struct sock *sk = skb->sk; |
638b2a69 JP |
3662 | int queue_index = -1; |
3663 | ||
04157469 AN |
3664 | if (!static_key_false(&xps_needed)) |
3665 | return -1; | |
3666 | ||
638b2a69 | 3667 | rcu_read_lock(); |
fc9bab24 AN |
3668 | if (!static_key_false(&xps_rxqs_needed)) |
3669 | goto get_cpus_map; | |
3670 | ||
eadec877 | 3671 | dev_maps = rcu_dereference(sb_dev->xps_rxqs_map); |
638b2a69 | 3672 | if (dev_maps) { |
fc9bab24 | 3673 | int tci = sk_rx_queue_get(sk); |
184c449f | 3674 | |
fc9bab24 AN |
3675 | if (tci >= 0 && tci < dev->num_rx_queues) |
3676 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3677 | tci); | |
3678 | } | |
184c449f | 3679 | |
fc9bab24 AN |
3680 | get_cpus_map: |
3681 | if (queue_index < 0) { | |
eadec877 | 3682 | dev_maps = rcu_dereference(sb_dev->xps_cpus_map); |
fc9bab24 AN |
3683 | if (dev_maps) { |
3684 | unsigned int tci = skb->sender_cpu - 1; | |
3685 | ||
3686 | queue_index = __get_xps_queue_idx(dev, skb, dev_maps, | |
3687 | tci); | |
638b2a69 JP |
3688 | } |
3689 | } | |
3690 | rcu_read_unlock(); | |
3691 | ||
3692 | return queue_index; | |
3693 | #else | |
3694 | return -1; | |
3695 | #endif | |
3696 | } | |
3697 | ||
a4ea8a3d | 3698 | u16 dev_pick_tx_zero(struct net_device *dev, struct sk_buff *skb, |
a350ecce | 3699 | struct net_device *sb_dev) |
a4ea8a3d AD |
3700 | { |
3701 | return 0; | |
3702 | } | |
3703 | EXPORT_SYMBOL(dev_pick_tx_zero); | |
3704 | ||
3705 | u16 dev_pick_tx_cpu_id(struct net_device *dev, struct sk_buff *skb, | |
a350ecce | 3706 | struct net_device *sb_dev) |
a4ea8a3d AD |
3707 | { |
3708 | return (u16)raw_smp_processor_id() % dev->real_num_tx_queues; | |
3709 | } | |
3710 | EXPORT_SYMBOL(dev_pick_tx_cpu_id); | |
3711 | ||
b71b5837 PA |
3712 | u16 netdev_pick_tx(struct net_device *dev, struct sk_buff *skb, |
3713 | struct net_device *sb_dev) | |
638b2a69 JP |
3714 | { |
3715 | struct sock *sk = skb->sk; | |
3716 | int queue_index = sk_tx_queue_get(sk); | |
3717 | ||
eadec877 AD |
3718 | sb_dev = sb_dev ? : dev; |
3719 | ||
638b2a69 JP |
3720 | if (queue_index < 0 || skb->ooo_okay || |
3721 | queue_index >= dev->real_num_tx_queues) { | |
eadec877 | 3722 | int new_index = get_xps_queue(dev, sb_dev, skb); |
f4563a75 | 3723 | |
638b2a69 | 3724 | if (new_index < 0) |
eadec877 | 3725 | new_index = skb_tx_hash(dev, sb_dev, skb); |
638b2a69 JP |
3726 | |
3727 | if (queue_index != new_index && sk && | |
004a5d01 | 3728 | sk_fullsock(sk) && |
638b2a69 JP |
3729 | rcu_access_pointer(sk->sk_dst_cache)) |
3730 | sk_tx_queue_set(sk, new_index); | |
3731 | ||
3732 | queue_index = new_index; | |
3733 | } | |
3734 | ||
3735 | return queue_index; | |
3736 | } | |
b71b5837 | 3737 | EXPORT_SYMBOL(netdev_pick_tx); |
638b2a69 | 3738 | |
4bd97d51 PA |
3739 | struct netdev_queue *netdev_core_pick_tx(struct net_device *dev, |
3740 | struct sk_buff *skb, | |
3741 | struct net_device *sb_dev) | |
638b2a69 JP |
3742 | { |
3743 | int queue_index = 0; | |
3744 | ||
3745 | #ifdef CONFIG_XPS | |
52bd2d62 ED |
3746 | u32 sender_cpu = skb->sender_cpu - 1; |
3747 | ||
3748 | if (sender_cpu >= (u32)NR_CPUS) | |
638b2a69 JP |
3749 | skb->sender_cpu = raw_smp_processor_id() + 1; |
3750 | #endif | |
3751 | ||
3752 | if (dev->real_num_tx_queues != 1) { | |
3753 | const struct net_device_ops *ops = dev->netdev_ops; | |
f4563a75 | 3754 | |
638b2a69 | 3755 | if (ops->ndo_select_queue) |
a350ecce | 3756 | queue_index = ops->ndo_select_queue(dev, skb, sb_dev); |
638b2a69 | 3757 | else |
4bd97d51 | 3758 | queue_index = netdev_pick_tx(dev, skb, sb_dev); |
638b2a69 | 3759 | |
d584527c | 3760 | queue_index = netdev_cap_txqueue(dev, queue_index); |
638b2a69 JP |
3761 | } |
3762 | ||
3763 | skb_set_queue_mapping(skb, queue_index); | |
3764 | return netdev_get_tx_queue(dev, queue_index); | |
3765 | } | |
3766 | ||
d29f749e | 3767 | /** |
9d08dd3d | 3768 | * __dev_queue_xmit - transmit a buffer |
d29f749e | 3769 | * @skb: buffer to transmit |
eadec877 | 3770 | * @sb_dev: suboordinate device used for L2 forwarding offload |
d29f749e DJ |
3771 | * |
3772 | * Queue a buffer for transmission to a network device. The caller must | |
3773 | * have set the device and priority and built the buffer before calling | |
3774 | * this function. The function can be called from an interrupt. | |
3775 | * | |
3776 | * A negative errno code is returned on a failure. A success does not | |
3777 | * guarantee the frame will be transmitted as it may be dropped due | |
3778 | * to congestion or traffic shaping. | |
3779 | * | |
3780 | * ----------------------------------------------------------------------------------- | |
3781 | * I notice this method can also return errors from the queue disciplines, | |
3782 | * including NET_XMIT_DROP, which is a positive value. So, errors can also | |
3783 | * be positive. | |
3784 | * | |
3785 | * Regardless of the return value, the skb is consumed, so it is currently | |
3786 | * difficult to retry a send to this method. (You can bump the ref count | |
3787 | * before sending to hold a reference for retry if you are careful.) | |
3788 | * | |
3789 | * When calling this method, interrupts MUST be enabled. This is because | |
3790 | * the BH enable code must have IRQs enabled so that it will not deadlock. | |
3791 | * --BLG | |
3792 | */ | |
eadec877 | 3793 | static int __dev_queue_xmit(struct sk_buff *skb, struct net_device *sb_dev) |
1da177e4 LT |
3794 | { |
3795 | struct net_device *dev = skb->dev; | |
dc2b4847 | 3796 | struct netdev_queue *txq; |
1da177e4 LT |
3797 | struct Qdisc *q; |
3798 | int rc = -ENOMEM; | |
f53c7239 | 3799 | bool again = false; |
1da177e4 | 3800 | |
6d1ccff6 ED |
3801 | skb_reset_mac_header(skb); |
3802 | ||
e7fd2885 WB |
3803 | if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_SCHED_TSTAMP)) |
3804 | __skb_tstamp_tx(skb, NULL, skb->sk, SCM_TSTAMP_SCHED); | |
3805 | ||
4ec93edb YH |
3806 | /* Disable soft irqs for various locks below. Also |
3807 | * stops preemption for RCU. | |
1da177e4 | 3808 | */ |
4ec93edb | 3809 | rcu_read_lock_bh(); |
1da177e4 | 3810 | |
5bc1421e NH |
3811 | skb_update_prio(skb); |
3812 | ||
1f211a1b DB |
3813 | qdisc_pkt_len_init(skb); |
3814 | #ifdef CONFIG_NET_CLS_ACT | |
8dc07fdb | 3815 | skb->tc_at_ingress = 0; |
1f211a1b | 3816 | # ifdef CONFIG_NET_EGRESS |
aabf6772 | 3817 | if (static_branch_unlikely(&egress_needed_key)) { |
1f211a1b DB |
3818 | skb = sch_handle_egress(skb, &rc, dev); |
3819 | if (!skb) | |
3820 | goto out; | |
3821 | } | |
3822 | # endif | |
3823 | #endif | |
02875878 ED |
3824 | /* If device/qdisc don't need skb->dst, release it right now while |
3825 | * its hot in this cpu cache. | |
3826 | */ | |
3827 | if (dev->priv_flags & IFF_XMIT_DST_RELEASE) | |
3828 | skb_dst_drop(skb); | |
3829 | else | |
3830 | skb_dst_force(skb); | |
3831 | ||
4bd97d51 | 3832 | txq = netdev_core_pick_tx(dev, skb, sb_dev); |
a898def2 | 3833 | q = rcu_dereference_bh(txq->qdisc); |
37437bb2 | 3834 | |
cf66ba58 | 3835 | trace_net_dev_queue(skb); |
1da177e4 | 3836 | if (q->enqueue) { |
bbd8a0d3 | 3837 | rc = __dev_xmit_skb(skb, q, dev, txq); |
37437bb2 | 3838 | goto out; |
1da177e4 LT |
3839 | } |
3840 | ||
3841 | /* The device has no queue. Common case for software devices: | |
eb13da1a | 3842 | * loopback, all the sorts of tunnels... |
1da177e4 | 3843 | |
eb13da1a | 3844 | * Really, it is unlikely that netif_tx_lock protection is necessary |
3845 | * here. (f.e. loopback and IP tunnels are clean ignoring statistics | |
3846 | * counters.) | |
3847 | * However, it is possible, that they rely on protection | |
3848 | * made by us here. | |
1da177e4 | 3849 | |
eb13da1a | 3850 | * Check this and shot the lock. It is not prone from deadlocks. |
3851 | *Either shot noqueue qdisc, it is even simpler 8) | |
1da177e4 LT |
3852 | */ |
3853 | if (dev->flags & IFF_UP) { | |
3854 | int cpu = smp_processor_id(); /* ok because BHs are off */ | |
3855 | ||
c773e847 | 3856 | if (txq->xmit_lock_owner != cpu) { |
97cdcf37 | 3857 | if (dev_xmit_recursion()) |
745e20f1 ED |
3858 | goto recursion_alert; |
3859 | ||
f53c7239 | 3860 | skb = validate_xmit_skb(skb, dev, &again); |
1f59533f | 3861 | if (!skb) |
d21fd63e | 3862 | goto out; |
1f59533f | 3863 | |
c773e847 | 3864 | HARD_TX_LOCK(dev, txq, cpu); |
1da177e4 | 3865 | |
73466498 | 3866 | if (!netif_xmit_stopped(txq)) { |
97cdcf37 | 3867 | dev_xmit_recursion_inc(); |
ce93718f | 3868 | skb = dev_hard_start_xmit(skb, dev, txq, &rc); |
97cdcf37 | 3869 | dev_xmit_recursion_dec(); |
572a9d7b | 3870 | if (dev_xmit_complete(rc)) { |
c773e847 | 3871 | HARD_TX_UNLOCK(dev, txq); |
1da177e4 LT |
3872 | goto out; |
3873 | } | |
3874 | } | |
c773e847 | 3875 | HARD_TX_UNLOCK(dev, txq); |
e87cc472 JP |
3876 | net_crit_ratelimited("Virtual device %s asks to queue packet!\n", |
3877 | dev->name); | |
1da177e4 LT |
3878 | } else { |
3879 | /* Recursion is detected! It is possible, | |
745e20f1 ED |
3880 | * unfortunately |
3881 | */ | |
3882 | recursion_alert: | |
e87cc472 JP |
3883 | net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n", |
3884 | dev->name); | |
1da177e4 LT |
3885 | } |
3886 | } | |
3887 | ||
3888 | rc = -ENETDOWN; | |
d4828d85 | 3889 | rcu_read_unlock_bh(); |
1da177e4 | 3890 | |
015f0688 | 3891 | atomic_long_inc(&dev->tx_dropped); |
1f59533f | 3892 | kfree_skb_list(skb); |
1da177e4 LT |
3893 | return rc; |
3894 | out: | |
d4828d85 | 3895 | rcu_read_unlock_bh(); |
1da177e4 LT |
3896 | return rc; |
3897 | } | |
f663dd9a | 3898 | |
2b4aa3ce | 3899 | int dev_queue_xmit(struct sk_buff *skb) |
f663dd9a JW |
3900 | { |
3901 | return __dev_queue_xmit(skb, NULL); | |
3902 | } | |
2b4aa3ce | 3903 | EXPORT_SYMBOL(dev_queue_xmit); |
1da177e4 | 3904 | |
eadec877 | 3905 | int dev_queue_xmit_accel(struct sk_buff *skb, struct net_device *sb_dev) |
f663dd9a | 3906 | { |
eadec877 | 3907 | return __dev_queue_xmit(skb, sb_dev); |
f663dd9a JW |
3908 | } |
3909 | EXPORT_SYMBOL(dev_queue_xmit_accel); | |
3910 | ||
865b03f2 MK |
3911 | int dev_direct_xmit(struct sk_buff *skb, u16 queue_id) |
3912 | { | |
3913 | struct net_device *dev = skb->dev; | |
3914 | struct sk_buff *orig_skb = skb; | |
3915 | struct netdev_queue *txq; | |
3916 | int ret = NETDEV_TX_BUSY; | |
3917 | bool again = false; | |
3918 | ||
3919 | if (unlikely(!netif_running(dev) || | |
3920 | !netif_carrier_ok(dev))) | |
3921 | goto drop; | |
3922 | ||
3923 | skb = validate_xmit_skb_list(skb, dev, &again); | |
3924 | if (skb != orig_skb) | |
3925 | goto drop; | |
3926 | ||
3927 | skb_set_queue_mapping(skb, queue_id); | |
3928 | txq = skb_get_tx_queue(dev, skb); | |
3929 | ||
3930 | local_bh_disable(); | |
3931 | ||
3932 | HARD_TX_LOCK(dev, txq, smp_processor_id()); | |
3933 | if (!netif_xmit_frozen_or_drv_stopped(txq)) | |
3934 | ret = netdev_start_xmit(skb, dev, txq, false); | |
3935 | HARD_TX_UNLOCK(dev, txq); | |
3936 | ||
3937 | local_bh_enable(); | |
3938 | ||
3939 | if (!dev_xmit_complete(ret)) | |
3940 | kfree_skb(skb); | |
3941 | ||
3942 | return ret; | |
3943 | drop: | |
3944 | atomic_long_inc(&dev->tx_dropped); | |
3945 | kfree_skb_list(skb); | |
3946 | return NET_XMIT_DROP; | |
3947 | } | |
3948 | EXPORT_SYMBOL(dev_direct_xmit); | |
1da177e4 | 3949 | |
eb13da1a | 3950 | /************************************************************************* |
3951 | * Receiver routines | |
3952 | *************************************************************************/ | |
1da177e4 | 3953 | |
6b2bedc3 | 3954 | int netdev_max_backlog __read_mostly = 1000; |
c9e6bc64 ED |
3955 | EXPORT_SYMBOL(netdev_max_backlog); |
3956 | ||
3b098e2d | 3957 | int netdev_tstamp_prequeue __read_mostly = 1; |
6b2bedc3 | 3958 | int netdev_budget __read_mostly = 300; |
7acf8a1e | 3959 | unsigned int __read_mostly netdev_budget_usecs = 2000; |
3d48b53f MT |
3960 | int weight_p __read_mostly = 64; /* old backlog weight */ |
3961 | int dev_weight_rx_bias __read_mostly = 1; /* bias for backlog weight */ | |
3962 | int dev_weight_tx_bias __read_mostly = 1; /* bias for output_queue quota */ | |
3963 | int dev_rx_weight __read_mostly = 64; | |
3964 | int dev_tx_weight __read_mostly = 64; | |
1da177e4 | 3965 | |
eecfd7c4 ED |
3966 | /* Called with irq disabled */ |
3967 | static inline void ____napi_schedule(struct softnet_data *sd, | |
3968 | struct napi_struct *napi) | |
3969 | { | |
3970 | list_add_tail(&napi->poll_list, &sd->poll_list); | |
3971 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
3972 | } | |
3973 | ||
bfb564e7 KK |
3974 | #ifdef CONFIG_RPS |
3975 | ||
3976 | /* One global table that all flow-based protocols share. */ | |
6e3f7faf | 3977 | struct rps_sock_flow_table __rcu *rps_sock_flow_table __read_mostly; |
bfb564e7 | 3978 | EXPORT_SYMBOL(rps_sock_flow_table); |
567e4b79 ED |
3979 | u32 rps_cpu_mask __read_mostly; |
3980 | EXPORT_SYMBOL(rps_cpu_mask); | |
bfb564e7 | 3981 | |
dc05360f | 3982 | struct static_key_false rps_needed __read_mostly; |
3df97ba8 | 3983 | EXPORT_SYMBOL(rps_needed); |
dc05360f | 3984 | struct static_key_false rfs_needed __read_mostly; |
13bfff25 | 3985 | EXPORT_SYMBOL(rfs_needed); |
adc9300e | 3986 | |
c445477d BH |
3987 | static struct rps_dev_flow * |
3988 | set_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
3989 | struct rps_dev_flow *rflow, u16 next_cpu) | |
3990 | { | |
a31196b0 | 3991 | if (next_cpu < nr_cpu_ids) { |
c445477d BH |
3992 | #ifdef CONFIG_RFS_ACCEL |
3993 | struct netdev_rx_queue *rxqueue; | |
3994 | struct rps_dev_flow_table *flow_table; | |
3995 | struct rps_dev_flow *old_rflow; | |
3996 | u32 flow_id; | |
3997 | u16 rxq_index; | |
3998 | int rc; | |
3999 | ||
4000 | /* Should we steer this flow to a different hardware queue? */ | |
69a19ee6 BH |
4001 | if (!skb_rx_queue_recorded(skb) || !dev->rx_cpu_rmap || |
4002 | !(dev->features & NETIF_F_NTUPLE)) | |
c445477d BH |
4003 | goto out; |
4004 | rxq_index = cpu_rmap_lookup_index(dev->rx_cpu_rmap, next_cpu); | |
4005 | if (rxq_index == skb_get_rx_queue(skb)) | |
4006 | goto out; | |
4007 | ||
4008 | rxqueue = dev->_rx + rxq_index; | |
4009 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4010 | if (!flow_table) | |
4011 | goto out; | |
61b905da | 4012 | flow_id = skb_get_hash(skb) & flow_table->mask; |
c445477d BH |
4013 | rc = dev->netdev_ops->ndo_rx_flow_steer(dev, skb, |
4014 | rxq_index, flow_id); | |
4015 | if (rc < 0) | |
4016 | goto out; | |
4017 | old_rflow = rflow; | |
4018 | rflow = &flow_table->flows[flow_id]; | |
c445477d BH |
4019 | rflow->filter = rc; |
4020 | if (old_rflow->filter == rflow->filter) | |
4021 | old_rflow->filter = RPS_NO_FILTER; | |
4022 | out: | |
4023 | #endif | |
4024 | rflow->last_qtail = | |
09994d1b | 4025 | per_cpu(softnet_data, next_cpu).input_queue_head; |
c445477d BH |
4026 | } |
4027 | ||
09994d1b | 4028 | rflow->cpu = next_cpu; |
c445477d BH |
4029 | return rflow; |
4030 | } | |
4031 | ||
bfb564e7 KK |
4032 | /* |
4033 | * get_rps_cpu is called from netif_receive_skb and returns the target | |
4034 | * CPU from the RPS map of the receiving queue for a given skb. | |
4035 | * rcu_read_lock must be held on entry. | |
4036 | */ | |
4037 | static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb, | |
4038 | struct rps_dev_flow **rflowp) | |
4039 | { | |
567e4b79 ED |
4040 | const struct rps_sock_flow_table *sock_flow_table; |
4041 | struct netdev_rx_queue *rxqueue = dev->_rx; | |
bfb564e7 | 4042 | struct rps_dev_flow_table *flow_table; |
567e4b79 | 4043 | struct rps_map *map; |
bfb564e7 | 4044 | int cpu = -1; |
567e4b79 | 4045 | u32 tcpu; |
61b905da | 4046 | u32 hash; |
bfb564e7 KK |
4047 | |
4048 | if (skb_rx_queue_recorded(skb)) { | |
4049 | u16 index = skb_get_rx_queue(skb); | |
567e4b79 | 4050 | |
62fe0b40 BH |
4051 | if (unlikely(index >= dev->real_num_rx_queues)) { |
4052 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4053 | "%s received packet on queue %u, but number " | |
4054 | "of RX queues is %u\n", | |
4055 | dev->name, index, dev->real_num_rx_queues); | |
bfb564e7 KK |
4056 | goto done; |
4057 | } | |
567e4b79 ED |
4058 | rxqueue += index; |
4059 | } | |
bfb564e7 | 4060 | |
567e4b79 ED |
4061 | /* Avoid computing hash if RFS/RPS is not active for this rxqueue */ |
4062 | ||
4063 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
6e3f7faf | 4064 | map = rcu_dereference(rxqueue->rps_map); |
567e4b79 | 4065 | if (!flow_table && !map) |
bfb564e7 KK |
4066 | goto done; |
4067 | ||
2d47b459 | 4068 | skb_reset_network_header(skb); |
61b905da TH |
4069 | hash = skb_get_hash(skb); |
4070 | if (!hash) | |
bfb564e7 KK |
4071 | goto done; |
4072 | ||
fec5e652 TH |
4073 | sock_flow_table = rcu_dereference(rps_sock_flow_table); |
4074 | if (flow_table && sock_flow_table) { | |
fec5e652 | 4075 | struct rps_dev_flow *rflow; |
567e4b79 ED |
4076 | u32 next_cpu; |
4077 | u32 ident; | |
4078 | ||
4079 | /* First check into global flow table if there is a match */ | |
4080 | ident = sock_flow_table->ents[hash & sock_flow_table->mask]; | |
4081 | if ((ident ^ hash) & ~rps_cpu_mask) | |
4082 | goto try_rps; | |
fec5e652 | 4083 | |
567e4b79 ED |
4084 | next_cpu = ident & rps_cpu_mask; |
4085 | ||
4086 | /* OK, now we know there is a match, | |
4087 | * we can look at the local (per receive queue) flow table | |
4088 | */ | |
61b905da | 4089 | rflow = &flow_table->flows[hash & flow_table->mask]; |
fec5e652 TH |
4090 | tcpu = rflow->cpu; |
4091 | ||
fec5e652 TH |
4092 | /* |
4093 | * If the desired CPU (where last recvmsg was done) is | |
4094 | * different from current CPU (one in the rx-queue flow | |
4095 | * table entry), switch if one of the following holds: | |
a31196b0 | 4096 | * - Current CPU is unset (>= nr_cpu_ids). |
fec5e652 TH |
4097 | * - Current CPU is offline. |
4098 | * - The current CPU's queue tail has advanced beyond the | |
4099 | * last packet that was enqueued using this table entry. | |
4100 | * This guarantees that all previous packets for the flow | |
4101 | * have been dequeued, thus preserving in order delivery. | |
4102 | */ | |
4103 | if (unlikely(tcpu != next_cpu) && | |
a31196b0 | 4104 | (tcpu >= nr_cpu_ids || !cpu_online(tcpu) || |
fec5e652 | 4105 | ((int)(per_cpu(softnet_data, tcpu).input_queue_head - |
baefa31d TH |
4106 | rflow->last_qtail)) >= 0)) { |
4107 | tcpu = next_cpu; | |
c445477d | 4108 | rflow = set_rps_cpu(dev, skb, rflow, next_cpu); |
baefa31d | 4109 | } |
c445477d | 4110 | |
a31196b0 | 4111 | if (tcpu < nr_cpu_ids && cpu_online(tcpu)) { |
fec5e652 TH |
4112 | *rflowp = rflow; |
4113 | cpu = tcpu; | |
4114 | goto done; | |
4115 | } | |
4116 | } | |
4117 | ||
567e4b79 ED |
4118 | try_rps: |
4119 | ||
0a9627f2 | 4120 | if (map) { |
8fc54f68 | 4121 | tcpu = map->cpus[reciprocal_scale(hash, map->len)]; |
0a9627f2 TH |
4122 | if (cpu_online(tcpu)) { |
4123 | cpu = tcpu; | |
4124 | goto done; | |
4125 | } | |
4126 | } | |
4127 | ||
4128 | done: | |
0a9627f2 TH |
4129 | return cpu; |
4130 | } | |
4131 | ||
c445477d BH |
4132 | #ifdef CONFIG_RFS_ACCEL |
4133 | ||
4134 | /** | |
4135 | * rps_may_expire_flow - check whether an RFS hardware filter may be removed | |
4136 | * @dev: Device on which the filter was set | |
4137 | * @rxq_index: RX queue index | |
4138 | * @flow_id: Flow ID passed to ndo_rx_flow_steer() | |
4139 | * @filter_id: Filter ID returned by ndo_rx_flow_steer() | |
4140 | * | |
4141 | * Drivers that implement ndo_rx_flow_steer() should periodically call | |
4142 | * this function for each installed filter and remove the filters for | |
4143 | * which it returns %true. | |
4144 | */ | |
4145 | bool rps_may_expire_flow(struct net_device *dev, u16 rxq_index, | |
4146 | u32 flow_id, u16 filter_id) | |
4147 | { | |
4148 | struct netdev_rx_queue *rxqueue = dev->_rx + rxq_index; | |
4149 | struct rps_dev_flow_table *flow_table; | |
4150 | struct rps_dev_flow *rflow; | |
4151 | bool expire = true; | |
a31196b0 | 4152 | unsigned int cpu; |
c445477d BH |
4153 | |
4154 | rcu_read_lock(); | |
4155 | flow_table = rcu_dereference(rxqueue->rps_flow_table); | |
4156 | if (flow_table && flow_id <= flow_table->mask) { | |
4157 | rflow = &flow_table->flows[flow_id]; | |
6aa7de05 | 4158 | cpu = READ_ONCE(rflow->cpu); |
a31196b0 | 4159 | if (rflow->filter == filter_id && cpu < nr_cpu_ids && |
c445477d BH |
4160 | ((int)(per_cpu(softnet_data, cpu).input_queue_head - |
4161 | rflow->last_qtail) < | |
4162 | (int)(10 * flow_table->mask))) | |
4163 | expire = false; | |
4164 | } | |
4165 | rcu_read_unlock(); | |
4166 | return expire; | |
4167 | } | |
4168 | EXPORT_SYMBOL(rps_may_expire_flow); | |
4169 | ||
4170 | #endif /* CONFIG_RFS_ACCEL */ | |
4171 | ||
0a9627f2 | 4172 | /* Called from hardirq (IPI) context */ |
e36fa2f7 | 4173 | static void rps_trigger_softirq(void *data) |
0a9627f2 | 4174 | { |
e36fa2f7 ED |
4175 | struct softnet_data *sd = data; |
4176 | ||
eecfd7c4 | 4177 | ____napi_schedule(sd, &sd->backlog); |
dee42870 | 4178 | sd->received_rps++; |
0a9627f2 | 4179 | } |
e36fa2f7 | 4180 | |
fec5e652 | 4181 | #endif /* CONFIG_RPS */ |
0a9627f2 | 4182 | |
e36fa2f7 ED |
4183 | /* |
4184 | * Check if this softnet_data structure is another cpu one | |
4185 | * If yes, queue it to our IPI list and return 1 | |
4186 | * If no, return 0 | |
4187 | */ | |
4188 | static int rps_ipi_queued(struct softnet_data *sd) | |
4189 | { | |
4190 | #ifdef CONFIG_RPS | |
903ceff7 | 4191 | struct softnet_data *mysd = this_cpu_ptr(&softnet_data); |
e36fa2f7 ED |
4192 | |
4193 | if (sd != mysd) { | |
4194 | sd->rps_ipi_next = mysd->rps_ipi_list; | |
4195 | mysd->rps_ipi_list = sd; | |
4196 | ||
4197 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
4198 | return 1; | |
4199 | } | |
4200 | #endif /* CONFIG_RPS */ | |
4201 | return 0; | |
4202 | } | |
4203 | ||
99bbc707 WB |
4204 | #ifdef CONFIG_NET_FLOW_LIMIT |
4205 | int netdev_flow_limit_table_len __read_mostly = (1 << 12); | |
4206 | #endif | |
4207 | ||
4208 | static bool skb_flow_limit(struct sk_buff *skb, unsigned int qlen) | |
4209 | { | |
4210 | #ifdef CONFIG_NET_FLOW_LIMIT | |
4211 | struct sd_flow_limit *fl; | |
4212 | struct softnet_data *sd; | |
4213 | unsigned int old_flow, new_flow; | |
4214 | ||
4215 | if (qlen < (netdev_max_backlog >> 1)) | |
4216 | return false; | |
4217 | ||
903ceff7 | 4218 | sd = this_cpu_ptr(&softnet_data); |
99bbc707 WB |
4219 | |
4220 | rcu_read_lock(); | |
4221 | fl = rcu_dereference(sd->flow_limit); | |
4222 | if (fl) { | |
3958afa1 | 4223 | new_flow = skb_get_hash(skb) & (fl->num_buckets - 1); |
99bbc707 WB |
4224 | old_flow = fl->history[fl->history_head]; |
4225 | fl->history[fl->history_head] = new_flow; | |
4226 | ||
4227 | fl->history_head++; | |
4228 | fl->history_head &= FLOW_LIMIT_HISTORY - 1; | |
4229 | ||
4230 | if (likely(fl->buckets[old_flow])) | |
4231 | fl->buckets[old_flow]--; | |
4232 | ||
4233 | if (++fl->buckets[new_flow] > (FLOW_LIMIT_HISTORY >> 1)) { | |
4234 | fl->count++; | |
4235 | rcu_read_unlock(); | |
4236 | return true; | |
4237 | } | |
4238 | } | |
4239 | rcu_read_unlock(); | |
4240 | #endif | |
4241 | return false; | |
4242 | } | |
4243 | ||
0a9627f2 TH |
4244 | /* |
4245 | * enqueue_to_backlog is called to queue an skb to a per CPU backlog | |
4246 | * queue (may be a remote CPU queue). | |
4247 | */ | |
fec5e652 TH |
4248 | static int enqueue_to_backlog(struct sk_buff *skb, int cpu, |
4249 | unsigned int *qtail) | |
0a9627f2 | 4250 | { |
e36fa2f7 | 4251 | struct softnet_data *sd; |
0a9627f2 | 4252 | unsigned long flags; |
99bbc707 | 4253 | unsigned int qlen; |
0a9627f2 | 4254 | |
e36fa2f7 | 4255 | sd = &per_cpu(softnet_data, cpu); |
0a9627f2 TH |
4256 | |
4257 | local_irq_save(flags); | |
0a9627f2 | 4258 | |
e36fa2f7 | 4259 | rps_lock(sd); |
e9e4dd32 JA |
4260 | if (!netif_running(skb->dev)) |
4261 | goto drop; | |
99bbc707 WB |
4262 | qlen = skb_queue_len(&sd->input_pkt_queue); |
4263 | if (qlen <= netdev_max_backlog && !skb_flow_limit(skb, qlen)) { | |
e008f3f0 | 4264 | if (qlen) { |
0a9627f2 | 4265 | enqueue: |
e36fa2f7 | 4266 | __skb_queue_tail(&sd->input_pkt_queue, skb); |
76cc8b13 | 4267 | input_queue_tail_incr_save(sd, qtail); |
e36fa2f7 | 4268 | rps_unlock(sd); |
152102c7 | 4269 | local_irq_restore(flags); |
0a9627f2 TH |
4270 | return NET_RX_SUCCESS; |
4271 | } | |
4272 | ||
ebda37c2 ED |
4273 | /* Schedule NAPI for backlog device |
4274 | * We can use non atomic operation since we own the queue lock | |
4275 | */ | |
4276 | if (!__test_and_set_bit(NAPI_STATE_SCHED, &sd->backlog.state)) { | |
e36fa2f7 | 4277 | if (!rps_ipi_queued(sd)) |
eecfd7c4 | 4278 | ____napi_schedule(sd, &sd->backlog); |
0a9627f2 TH |
4279 | } |
4280 | goto enqueue; | |
4281 | } | |
4282 | ||
e9e4dd32 | 4283 | drop: |
dee42870 | 4284 | sd->dropped++; |
e36fa2f7 | 4285 | rps_unlock(sd); |
0a9627f2 | 4286 | |
0a9627f2 TH |
4287 | local_irq_restore(flags); |
4288 | ||
caf586e5 | 4289 | atomic_long_inc(&skb->dev->rx_dropped); |
0a9627f2 TH |
4290 | kfree_skb(skb); |
4291 | return NET_RX_DROP; | |
4292 | } | |
1da177e4 | 4293 | |
e817f856 JDB |
4294 | static struct netdev_rx_queue *netif_get_rxqueue(struct sk_buff *skb) |
4295 | { | |
4296 | struct net_device *dev = skb->dev; | |
4297 | struct netdev_rx_queue *rxqueue; | |
4298 | ||
4299 | rxqueue = dev->_rx; | |
4300 | ||
4301 | if (skb_rx_queue_recorded(skb)) { | |
4302 | u16 index = skb_get_rx_queue(skb); | |
4303 | ||
4304 | if (unlikely(index >= dev->real_num_rx_queues)) { | |
4305 | WARN_ONCE(dev->real_num_rx_queues > 1, | |
4306 | "%s received packet on queue %u, but number " | |
4307 | "of RX queues is %u\n", | |
4308 | dev->name, index, dev->real_num_rx_queues); | |
4309 | ||
4310 | return rxqueue; /* Return first rxqueue */ | |
4311 | } | |
4312 | rxqueue += index; | |
4313 | } | |
4314 | return rxqueue; | |
4315 | } | |
4316 | ||
d4455169 | 4317 | static u32 netif_receive_generic_xdp(struct sk_buff *skb, |
02671e23 | 4318 | struct xdp_buff *xdp, |
d4455169 JF |
4319 | struct bpf_prog *xdp_prog) |
4320 | { | |
e817f856 | 4321 | struct netdev_rx_queue *rxqueue; |
198d83bb | 4322 | void *orig_data, *orig_data_end; |
de8f3a83 | 4323 | u32 metalen, act = XDP_DROP; |
29724956 JDB |
4324 | __be16 orig_eth_type; |
4325 | struct ethhdr *eth; | |
4326 | bool orig_bcast; | |
d4455169 JF |
4327 | int hlen, off; |
4328 | u32 mac_len; | |
4329 | ||
4330 | /* Reinjected packets coming from act_mirred or similar should | |
4331 | * not get XDP generic processing. | |
4332 | */ | |
cd11b164 | 4333 | if (skb_cloned(skb) || skb_is_tc_redirected(skb)) |
d4455169 JF |
4334 | return XDP_PASS; |
4335 | ||
de8f3a83 DB |
4336 | /* XDP packets must be linear and must have sufficient headroom |
4337 | * of XDP_PACKET_HEADROOM bytes. This is the guarantee that also | |
4338 | * native XDP provides, thus we need to do it here as well. | |
4339 | */ | |
4340 | if (skb_is_nonlinear(skb) || | |
4341 | skb_headroom(skb) < XDP_PACKET_HEADROOM) { | |
4342 | int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb); | |
4343 | int troom = skb->tail + skb->data_len - skb->end; | |
4344 | ||
4345 | /* In case we have to go down the path and also linearize, | |
4346 | * then lets do the pskb_expand_head() work just once here. | |
4347 | */ | |
4348 | if (pskb_expand_head(skb, | |
4349 | hroom > 0 ? ALIGN(hroom, NET_SKB_PAD) : 0, | |
4350 | troom > 0 ? troom + 128 : 0, GFP_ATOMIC)) | |
4351 | goto do_drop; | |
2d17d8d7 | 4352 | if (skb_linearize(skb)) |
de8f3a83 DB |
4353 | goto do_drop; |
4354 | } | |
d4455169 JF |
4355 | |
4356 | /* The XDP program wants to see the packet starting at the MAC | |
4357 | * header. | |
4358 | */ | |
4359 | mac_len = skb->data - skb_mac_header(skb); | |
4360 | hlen = skb_headlen(skb) + mac_len; | |
02671e23 BT |
4361 | xdp->data = skb->data - mac_len; |
4362 | xdp->data_meta = xdp->data; | |
4363 | xdp->data_end = xdp->data + hlen; | |
4364 | xdp->data_hard_start = skb->data - skb_headroom(skb); | |
4365 | orig_data_end = xdp->data_end; | |
4366 | orig_data = xdp->data; | |
29724956 JDB |
4367 | eth = (struct ethhdr *)xdp->data; |
4368 | orig_bcast = is_multicast_ether_addr_64bits(eth->h_dest); | |
4369 | orig_eth_type = eth->h_proto; | |
d4455169 | 4370 | |
e817f856 | 4371 | rxqueue = netif_get_rxqueue(skb); |
02671e23 | 4372 | xdp->rxq = &rxqueue->xdp_rxq; |
e817f856 | 4373 | |
02671e23 | 4374 | act = bpf_prog_run_xdp(xdp_prog, xdp); |
d4455169 | 4375 | |
02671e23 | 4376 | off = xdp->data - orig_data; |
d4455169 JF |
4377 | if (off > 0) |
4378 | __skb_pull(skb, off); | |
4379 | else if (off < 0) | |
4380 | __skb_push(skb, -off); | |
92dd5452 | 4381 | skb->mac_header += off; |
d4455169 | 4382 | |
198d83bb NS |
4383 | /* check if bpf_xdp_adjust_tail was used. it can only "shrink" |
4384 | * pckt. | |
4385 | */ | |
02671e23 | 4386 | off = orig_data_end - xdp->data_end; |
f7613120 | 4387 | if (off != 0) { |
02671e23 | 4388 | skb_set_tail_pointer(skb, xdp->data_end - xdp->data); |
f7613120 | 4389 | skb->len -= off; |
02671e23 | 4390 | |
f7613120 | 4391 | } |
198d83bb | 4392 | |
29724956 JDB |
4393 | /* check if XDP changed eth hdr such SKB needs update */ |
4394 | eth = (struct ethhdr *)xdp->data; | |
4395 | if ((orig_eth_type != eth->h_proto) || | |
4396 | (orig_bcast != is_multicast_ether_addr_64bits(eth->h_dest))) { | |
4397 | __skb_push(skb, ETH_HLEN); | |
4398 | skb->protocol = eth_type_trans(skb, skb->dev); | |
4399 | } | |
4400 | ||
d4455169 | 4401 | switch (act) { |
6103aa96 | 4402 | case XDP_REDIRECT: |
d4455169 JF |
4403 | case XDP_TX: |
4404 | __skb_push(skb, mac_len); | |
de8f3a83 | 4405 | break; |
d4455169 | 4406 | case XDP_PASS: |
02671e23 | 4407 | metalen = xdp->data - xdp->data_meta; |
de8f3a83 DB |
4408 | if (metalen) |
4409 | skb_metadata_set(skb, metalen); | |
d4455169 | 4410 | break; |
d4455169 JF |
4411 | default: |
4412 | bpf_warn_invalid_xdp_action(act); | |
4413 | /* fall through */ | |
4414 | case XDP_ABORTED: | |
4415 | trace_xdp_exception(skb->dev, xdp_prog, act); | |
4416 | /* fall through */ | |
4417 | case XDP_DROP: | |
4418 | do_drop: | |
4419 | kfree_skb(skb); | |
4420 | break; | |
4421 | } | |
4422 | ||
4423 | return act; | |
4424 | } | |
4425 | ||
4426 | /* When doing generic XDP we have to bypass the qdisc layer and the | |
4427 | * network taps in order to match in-driver-XDP behavior. | |
4428 | */ | |
7c497478 | 4429 | void generic_xdp_tx(struct sk_buff *skb, struct bpf_prog *xdp_prog) |
d4455169 JF |
4430 | { |
4431 | struct net_device *dev = skb->dev; | |
4432 | struct netdev_queue *txq; | |
4433 | bool free_skb = true; | |
4434 | int cpu, rc; | |
4435 | ||
4bd97d51 | 4436 | txq = netdev_core_pick_tx(dev, skb, NULL); |
d4455169 JF |
4437 | cpu = smp_processor_id(); |
4438 | HARD_TX_LOCK(dev, txq, cpu); | |
4439 | if (!netif_xmit_stopped(txq)) { | |
4440 | rc = netdev_start_xmit(skb, dev, txq, 0); | |
4441 | if (dev_xmit_complete(rc)) | |
4442 | free_skb = false; | |
4443 | } | |
4444 | HARD_TX_UNLOCK(dev, txq); | |
4445 | if (free_skb) { | |
4446 | trace_xdp_exception(dev, xdp_prog, XDP_TX); | |
4447 | kfree_skb(skb); | |
4448 | } | |
4449 | } | |
7c497478 | 4450 | EXPORT_SYMBOL_GPL(generic_xdp_tx); |
d4455169 | 4451 | |
02786475 | 4452 | static DEFINE_STATIC_KEY_FALSE(generic_xdp_needed_key); |
d4455169 | 4453 | |
7c497478 | 4454 | int do_xdp_generic(struct bpf_prog *xdp_prog, struct sk_buff *skb) |
d4455169 | 4455 | { |
d4455169 | 4456 | if (xdp_prog) { |
02671e23 BT |
4457 | struct xdp_buff xdp; |
4458 | u32 act; | |
6103aa96 | 4459 | int err; |
d4455169 | 4460 | |
02671e23 | 4461 | act = netif_receive_generic_xdp(skb, &xdp, xdp_prog); |
d4455169 | 4462 | if (act != XDP_PASS) { |
6103aa96 JF |
4463 | switch (act) { |
4464 | case XDP_REDIRECT: | |
2facaad6 | 4465 | err = xdp_do_generic_redirect(skb->dev, skb, |
02671e23 | 4466 | &xdp, xdp_prog); |
6103aa96 JF |
4467 | if (err) |
4468 | goto out_redir; | |
02671e23 | 4469 | break; |
6103aa96 | 4470 | case XDP_TX: |
d4455169 | 4471 | generic_xdp_tx(skb, xdp_prog); |
6103aa96 JF |
4472 | break; |
4473 | } | |
d4455169 JF |
4474 | return XDP_DROP; |
4475 | } | |
4476 | } | |
4477 | return XDP_PASS; | |
6103aa96 | 4478 | out_redir: |
6103aa96 JF |
4479 | kfree_skb(skb); |
4480 | return XDP_DROP; | |
d4455169 | 4481 | } |
7c497478 | 4482 | EXPORT_SYMBOL_GPL(do_xdp_generic); |
d4455169 | 4483 | |
ae78dbfa | 4484 | static int netif_rx_internal(struct sk_buff *skb) |
1da177e4 | 4485 | { |
b0e28f1e | 4486 | int ret; |
1da177e4 | 4487 | |
588f0330 | 4488 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
1da177e4 | 4489 | |
cf66ba58 | 4490 | trace_netif_rx(skb); |
d4455169 | 4491 | |
02786475 | 4492 | if (static_branch_unlikely(&generic_xdp_needed_key)) { |
bbbe211c JF |
4493 | int ret; |
4494 | ||
4495 | preempt_disable(); | |
4496 | rcu_read_lock(); | |
4497 | ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
4498 | rcu_read_unlock(); | |
4499 | preempt_enable(); | |
d4455169 | 4500 | |
6103aa96 JF |
4501 | /* Consider XDP consuming the packet a success from |
4502 | * the netdev point of view we do not want to count | |
4503 | * this as an error. | |
4504 | */ | |
d4455169 | 4505 | if (ret != XDP_PASS) |
6103aa96 | 4506 | return NET_RX_SUCCESS; |
d4455169 JF |
4507 | } |
4508 | ||
df334545 | 4509 | #ifdef CONFIG_RPS |
dc05360f | 4510 | if (static_branch_unlikely(&rps_needed)) { |
fec5e652 | 4511 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
b0e28f1e ED |
4512 | int cpu; |
4513 | ||
cece1945 | 4514 | preempt_disable(); |
b0e28f1e | 4515 | rcu_read_lock(); |
fec5e652 TH |
4516 | |
4517 | cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
b0e28f1e ED |
4518 | if (cpu < 0) |
4519 | cpu = smp_processor_id(); | |
fec5e652 TH |
4520 | |
4521 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
4522 | ||
b0e28f1e | 4523 | rcu_read_unlock(); |
cece1945 | 4524 | preempt_enable(); |
adc9300e ED |
4525 | } else |
4526 | #endif | |
fec5e652 TH |
4527 | { |
4528 | unsigned int qtail; | |
f4563a75 | 4529 | |
fec5e652 TH |
4530 | ret = enqueue_to_backlog(skb, get_cpu(), &qtail); |
4531 | put_cpu(); | |
4532 | } | |
b0e28f1e | 4533 | return ret; |
1da177e4 | 4534 | } |
ae78dbfa BH |
4535 | |
4536 | /** | |
4537 | * netif_rx - post buffer to the network code | |
4538 | * @skb: buffer to post | |
4539 | * | |
4540 | * This function receives a packet from a device driver and queues it for | |
4541 | * the upper (protocol) levels to process. It always succeeds. The buffer | |
4542 | * may be dropped during processing for congestion control or by the | |
4543 | * protocol layers. | |
4544 | * | |
4545 | * return values: | |
4546 | * NET_RX_SUCCESS (no congestion) | |
4547 | * NET_RX_DROP (packet was dropped) | |
4548 | * | |
4549 | */ | |
4550 | ||
4551 | int netif_rx(struct sk_buff *skb) | |
4552 | { | |
b0e3f1bd GB |
4553 | int ret; |
4554 | ||
ae78dbfa BH |
4555 | trace_netif_rx_entry(skb); |
4556 | ||
b0e3f1bd GB |
4557 | ret = netif_rx_internal(skb); |
4558 | trace_netif_rx_exit(ret); | |
4559 | ||
4560 | return ret; | |
ae78dbfa | 4561 | } |
d1b19dff | 4562 | EXPORT_SYMBOL(netif_rx); |
1da177e4 LT |
4563 | |
4564 | int netif_rx_ni(struct sk_buff *skb) | |
4565 | { | |
4566 | int err; | |
4567 | ||
ae78dbfa BH |
4568 | trace_netif_rx_ni_entry(skb); |
4569 | ||
1da177e4 | 4570 | preempt_disable(); |
ae78dbfa | 4571 | err = netif_rx_internal(skb); |
1da177e4 LT |
4572 | if (local_softirq_pending()) |
4573 | do_softirq(); | |
4574 | preempt_enable(); | |
b0e3f1bd | 4575 | trace_netif_rx_ni_exit(err); |
1da177e4 LT |
4576 | |
4577 | return err; | |
4578 | } | |
1da177e4 LT |
4579 | EXPORT_SYMBOL(netif_rx_ni); |
4580 | ||
0766f788 | 4581 | static __latent_entropy void net_tx_action(struct softirq_action *h) |
1da177e4 | 4582 | { |
903ceff7 | 4583 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
1da177e4 LT |
4584 | |
4585 | if (sd->completion_queue) { | |
4586 | struct sk_buff *clist; | |
4587 | ||
4588 | local_irq_disable(); | |
4589 | clist = sd->completion_queue; | |
4590 | sd->completion_queue = NULL; | |
4591 | local_irq_enable(); | |
4592 | ||
4593 | while (clist) { | |
4594 | struct sk_buff *skb = clist; | |
f4563a75 | 4595 | |
1da177e4 LT |
4596 | clist = clist->next; |
4597 | ||
63354797 | 4598 | WARN_ON(refcount_read(&skb->users)); |
e6247027 ED |
4599 | if (likely(get_kfree_skb_cb(skb)->reason == SKB_REASON_CONSUMED)) |
4600 | trace_consume_skb(skb); | |
4601 | else | |
4602 | trace_kfree_skb(skb, net_tx_action); | |
15fad714 JDB |
4603 | |
4604 | if (skb->fclone != SKB_FCLONE_UNAVAILABLE) | |
4605 | __kfree_skb(skb); | |
4606 | else | |
4607 | __kfree_skb_defer(skb); | |
1da177e4 | 4608 | } |
15fad714 JDB |
4609 | |
4610 | __kfree_skb_flush(); | |
1da177e4 LT |
4611 | } |
4612 | ||
4613 | if (sd->output_queue) { | |
37437bb2 | 4614 | struct Qdisc *head; |
1da177e4 LT |
4615 | |
4616 | local_irq_disable(); | |
4617 | head = sd->output_queue; | |
4618 | sd->output_queue = NULL; | |
a9cbd588 | 4619 | sd->output_queue_tailp = &sd->output_queue; |
1da177e4 LT |
4620 | local_irq_enable(); |
4621 | ||
4622 | while (head) { | |
37437bb2 | 4623 | struct Qdisc *q = head; |
6b3ba914 | 4624 | spinlock_t *root_lock = NULL; |
37437bb2 | 4625 | |
1da177e4 LT |
4626 | head = head->next_sched; |
4627 | ||
6b3ba914 JF |
4628 | if (!(q->flags & TCQ_F_NOLOCK)) { |
4629 | root_lock = qdisc_lock(q); | |
4630 | spin_lock(root_lock); | |
4631 | } | |
3bcb846c ED |
4632 | /* We need to make sure head->next_sched is read |
4633 | * before clearing __QDISC_STATE_SCHED | |
4634 | */ | |
4635 | smp_mb__before_atomic(); | |
4636 | clear_bit(__QDISC_STATE_SCHED, &q->state); | |
4637 | qdisc_run(q); | |
6b3ba914 JF |
4638 | if (root_lock) |
4639 | spin_unlock(root_lock); | |
1da177e4 LT |
4640 | } |
4641 | } | |
f53c7239 SK |
4642 | |
4643 | xfrm_dev_backlog(sd); | |
1da177e4 LT |
4644 | } |
4645 | ||
181402a5 | 4646 | #if IS_ENABLED(CONFIG_BRIDGE) && IS_ENABLED(CONFIG_ATM_LANE) |
da678292 MM |
4647 | /* This hook is defined here for ATM LANE */ |
4648 | int (*br_fdb_test_addr_hook)(struct net_device *dev, | |
4649 | unsigned char *addr) __read_mostly; | |
4fb019a0 | 4650 | EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook); |
da678292 | 4651 | #endif |
1da177e4 | 4652 | |
1f211a1b DB |
4653 | static inline struct sk_buff * |
4654 | sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret, | |
4655 | struct net_device *orig_dev) | |
f697c3e8 | 4656 | { |
e7582bab | 4657 | #ifdef CONFIG_NET_CLS_ACT |
46209401 | 4658 | struct mini_Qdisc *miniq = rcu_dereference_bh(skb->dev->miniq_ingress); |
d2788d34 | 4659 | struct tcf_result cl_res; |
24824a09 | 4660 | |
c9e99fd0 DB |
4661 | /* If there's at least one ingress present somewhere (so |
4662 | * we get here via enabled static key), remaining devices | |
4663 | * that are not configured with an ingress qdisc will bail | |
d2788d34 | 4664 | * out here. |
c9e99fd0 | 4665 | */ |
46209401 | 4666 | if (!miniq) |
4577139b | 4667 | return skb; |
46209401 | 4668 | |
f697c3e8 HX |
4669 | if (*pt_prev) { |
4670 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4671 | *pt_prev = NULL; | |
1da177e4 LT |
4672 | } |
4673 | ||
3365495c | 4674 | qdisc_skb_cb(skb)->pkt_len = skb->len; |
8dc07fdb | 4675 | skb->tc_at_ingress = 1; |
46209401 | 4676 | mini_qdisc_bstats_cpu_update(miniq, skb); |
c9e99fd0 | 4677 | |
46209401 | 4678 | switch (tcf_classify(skb, miniq->filter_list, &cl_res, false)) { |
d2788d34 DB |
4679 | case TC_ACT_OK: |
4680 | case TC_ACT_RECLASSIFY: | |
4681 | skb->tc_index = TC_H_MIN(cl_res.classid); | |
4682 | break; | |
4683 | case TC_ACT_SHOT: | |
46209401 | 4684 | mini_qdisc_qstats_cpu_drop(miniq); |
8a3a4c6e ED |
4685 | kfree_skb(skb); |
4686 | return NULL; | |
d2788d34 DB |
4687 | case TC_ACT_STOLEN: |
4688 | case TC_ACT_QUEUED: | |
e25ea21f | 4689 | case TC_ACT_TRAP: |
8a3a4c6e | 4690 | consume_skb(skb); |
d2788d34 | 4691 | return NULL; |
27b29f63 AS |
4692 | case TC_ACT_REDIRECT: |
4693 | /* skb_mac_header check was done by cls/act_bpf, so | |
4694 | * we can safely push the L2 header back before | |
4695 | * redirecting to another netdev | |
4696 | */ | |
4697 | __skb_push(skb, skb->mac_len); | |
4698 | skb_do_redirect(skb); | |
4699 | return NULL; | |
cd11b164 PA |
4700 | case TC_ACT_REINSERT: |
4701 | /* this does not scrub the packet, and updates stats on error */ | |
4702 | skb_tc_reinsert(skb, &cl_res); | |
4703 | return NULL; | |
d2788d34 DB |
4704 | default: |
4705 | break; | |
f697c3e8 | 4706 | } |
e7582bab | 4707 | #endif /* CONFIG_NET_CLS_ACT */ |
e687ad60 PN |
4708 | return skb; |
4709 | } | |
1da177e4 | 4710 | |
24b27fc4 MB |
4711 | /** |
4712 | * netdev_is_rx_handler_busy - check if receive handler is registered | |
4713 | * @dev: device to check | |
4714 | * | |
4715 | * Check if a receive handler is already registered for a given device. | |
4716 | * Return true if there one. | |
4717 | * | |
4718 | * The caller must hold the rtnl_mutex. | |
4719 | */ | |
4720 | bool netdev_is_rx_handler_busy(struct net_device *dev) | |
4721 | { | |
4722 | ASSERT_RTNL(); | |
4723 | return dev && rtnl_dereference(dev->rx_handler); | |
4724 | } | |
4725 | EXPORT_SYMBOL_GPL(netdev_is_rx_handler_busy); | |
4726 | ||
ab95bfe0 JP |
4727 | /** |
4728 | * netdev_rx_handler_register - register receive handler | |
4729 | * @dev: device to register a handler for | |
4730 | * @rx_handler: receive handler to register | |
93e2c32b | 4731 | * @rx_handler_data: data pointer that is used by rx handler |
ab95bfe0 | 4732 | * |
e227867f | 4733 | * Register a receive handler for a device. This handler will then be |
ab95bfe0 JP |
4734 | * called from __netif_receive_skb. A negative errno code is returned |
4735 | * on a failure. | |
4736 | * | |
4737 | * The caller must hold the rtnl_mutex. | |
8a4eb573 JP |
4738 | * |
4739 | * For a general description of rx_handler, see enum rx_handler_result. | |
ab95bfe0 JP |
4740 | */ |
4741 | int netdev_rx_handler_register(struct net_device *dev, | |
93e2c32b JP |
4742 | rx_handler_func_t *rx_handler, |
4743 | void *rx_handler_data) | |
ab95bfe0 | 4744 | { |
1b7cd004 | 4745 | if (netdev_is_rx_handler_busy(dev)) |
ab95bfe0 JP |
4746 | return -EBUSY; |
4747 | ||
f5426250 PA |
4748 | if (dev->priv_flags & IFF_NO_RX_HANDLER) |
4749 | return -EINVAL; | |
4750 | ||
00cfec37 | 4751 | /* Note: rx_handler_data must be set before rx_handler */ |
93e2c32b | 4752 | rcu_assign_pointer(dev->rx_handler_data, rx_handler_data); |
ab95bfe0 JP |
4753 | rcu_assign_pointer(dev->rx_handler, rx_handler); |
4754 | ||
4755 | return 0; | |
4756 | } | |
4757 | EXPORT_SYMBOL_GPL(netdev_rx_handler_register); | |
4758 | ||
4759 | /** | |
4760 | * netdev_rx_handler_unregister - unregister receive handler | |
4761 | * @dev: device to unregister a handler from | |
4762 | * | |
166ec369 | 4763 | * Unregister a receive handler from a device. |
ab95bfe0 JP |
4764 | * |
4765 | * The caller must hold the rtnl_mutex. | |
4766 | */ | |
4767 | void netdev_rx_handler_unregister(struct net_device *dev) | |
4768 | { | |
4769 | ||
4770 | ASSERT_RTNL(); | |
a9b3cd7f | 4771 | RCU_INIT_POINTER(dev->rx_handler, NULL); |
00cfec37 ED |
4772 | /* a reader seeing a non NULL rx_handler in a rcu_read_lock() |
4773 | * section has a guarantee to see a non NULL rx_handler_data | |
4774 | * as well. | |
4775 | */ | |
4776 | synchronize_net(); | |
a9b3cd7f | 4777 | RCU_INIT_POINTER(dev->rx_handler_data, NULL); |
ab95bfe0 JP |
4778 | } |
4779 | EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister); | |
4780 | ||
b4b9e355 MG |
4781 | /* |
4782 | * Limit the use of PFMEMALLOC reserves to those protocols that implement | |
4783 | * the special handling of PFMEMALLOC skbs. | |
4784 | */ | |
4785 | static bool skb_pfmemalloc_protocol(struct sk_buff *skb) | |
4786 | { | |
4787 | switch (skb->protocol) { | |
2b8837ae JP |
4788 | case htons(ETH_P_ARP): |
4789 | case htons(ETH_P_IP): | |
4790 | case htons(ETH_P_IPV6): | |
4791 | case htons(ETH_P_8021Q): | |
4792 | case htons(ETH_P_8021AD): | |
b4b9e355 MG |
4793 | return true; |
4794 | default: | |
4795 | return false; | |
4796 | } | |
4797 | } | |
4798 | ||
e687ad60 PN |
4799 | static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev, |
4800 | int *ret, struct net_device *orig_dev) | |
4801 | { | |
e7582bab | 4802 | #ifdef CONFIG_NETFILTER_INGRESS |
e687ad60 | 4803 | if (nf_hook_ingress_active(skb)) { |
2c1e2703 AC |
4804 | int ingress_retval; |
4805 | ||
e687ad60 PN |
4806 | if (*pt_prev) { |
4807 | *ret = deliver_skb(skb, *pt_prev, orig_dev); | |
4808 | *pt_prev = NULL; | |
4809 | } | |
4810 | ||
2c1e2703 AC |
4811 | rcu_read_lock(); |
4812 | ingress_retval = nf_hook_ingress(skb); | |
4813 | rcu_read_unlock(); | |
4814 | return ingress_retval; | |
e687ad60 | 4815 | } |
e7582bab | 4816 | #endif /* CONFIG_NETFILTER_INGRESS */ |
e687ad60 PN |
4817 | return 0; |
4818 | } | |
e687ad60 | 4819 | |
88eb1944 EC |
4820 | static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, |
4821 | struct packet_type **ppt_prev) | |
1da177e4 LT |
4822 | { |
4823 | struct packet_type *ptype, *pt_prev; | |
ab95bfe0 | 4824 | rx_handler_func_t *rx_handler; |
f2ccd8fa | 4825 | struct net_device *orig_dev; |
8a4eb573 | 4826 | bool deliver_exact = false; |
1da177e4 | 4827 | int ret = NET_RX_DROP; |
252e3346 | 4828 | __be16 type; |
1da177e4 | 4829 | |
588f0330 | 4830 | net_timestamp_check(!netdev_tstamp_prequeue, skb); |
81bbb3d4 | 4831 | |
cf66ba58 | 4832 | trace_netif_receive_skb(skb); |
9b22ea56 | 4833 | |
cc9bd5ce | 4834 | orig_dev = skb->dev; |
8f903c70 | 4835 | |
c1d2bbe1 | 4836 | skb_reset_network_header(skb); |
fda55eca ED |
4837 | if (!skb_transport_header_was_set(skb)) |
4838 | skb_reset_transport_header(skb); | |
0b5c9db1 | 4839 | skb_reset_mac_len(skb); |
1da177e4 LT |
4840 | |
4841 | pt_prev = NULL; | |
4842 | ||
63d8ea7f | 4843 | another_round: |
b6858177 | 4844 | skb->skb_iif = skb->dev->ifindex; |
63d8ea7f DM |
4845 | |
4846 | __this_cpu_inc(softnet_data.processed); | |
4847 | ||
8ad227ff PM |
4848 | if (skb->protocol == cpu_to_be16(ETH_P_8021Q) || |
4849 | skb->protocol == cpu_to_be16(ETH_P_8021AD)) { | |
0d5501c1 | 4850 | skb = skb_vlan_untag(skb); |
bcc6d479 | 4851 | if (unlikely(!skb)) |
2c17d27c | 4852 | goto out; |
bcc6d479 JP |
4853 | } |
4854 | ||
e7246e12 WB |
4855 | if (skb_skip_tc_classify(skb)) |
4856 | goto skip_classify; | |
1da177e4 | 4857 | |
9754e293 | 4858 | if (pfmemalloc) |
b4b9e355 MG |
4859 | goto skip_taps; |
4860 | ||
1da177e4 | 4861 | list_for_each_entry_rcu(ptype, &ptype_all, list) { |
7866a621 SN |
4862 | if (pt_prev) |
4863 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4864 | pt_prev = ptype; | |
4865 | } | |
4866 | ||
4867 | list_for_each_entry_rcu(ptype, &skb->dev->ptype_all, list) { | |
4868 | if (pt_prev) | |
4869 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4870 | pt_prev = ptype; | |
1da177e4 LT |
4871 | } |
4872 | ||
b4b9e355 | 4873 | skip_taps: |
1cf51900 | 4874 | #ifdef CONFIG_NET_INGRESS |
aabf6772 | 4875 | if (static_branch_unlikely(&ingress_needed_key)) { |
1f211a1b | 4876 | skb = sch_handle_ingress(skb, &pt_prev, &ret, orig_dev); |
4577139b | 4877 | if (!skb) |
2c17d27c | 4878 | goto out; |
e687ad60 PN |
4879 | |
4880 | if (nf_ingress(skb, &pt_prev, &ret, orig_dev) < 0) | |
2c17d27c | 4881 | goto out; |
4577139b | 4882 | } |
1cf51900 | 4883 | #endif |
a5135bcf | 4884 | skb_reset_tc(skb); |
e7246e12 | 4885 | skip_classify: |
9754e293 | 4886 | if (pfmemalloc && !skb_pfmemalloc_protocol(skb)) |
b4b9e355 MG |
4887 | goto drop; |
4888 | ||
df8a39de | 4889 | if (skb_vlan_tag_present(skb)) { |
2425717b JF |
4890 | if (pt_prev) { |
4891 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4892 | pt_prev = NULL; | |
4893 | } | |
48cc32d3 | 4894 | if (vlan_do_receive(&skb)) |
2425717b JF |
4895 | goto another_round; |
4896 | else if (unlikely(!skb)) | |
2c17d27c | 4897 | goto out; |
2425717b JF |
4898 | } |
4899 | ||
48cc32d3 | 4900 | rx_handler = rcu_dereference(skb->dev->rx_handler); |
ab95bfe0 JP |
4901 | if (rx_handler) { |
4902 | if (pt_prev) { | |
4903 | ret = deliver_skb(skb, pt_prev, orig_dev); | |
4904 | pt_prev = NULL; | |
4905 | } | |
8a4eb573 JP |
4906 | switch (rx_handler(&skb)) { |
4907 | case RX_HANDLER_CONSUMED: | |
3bc1b1ad | 4908 | ret = NET_RX_SUCCESS; |
2c17d27c | 4909 | goto out; |
8a4eb573 | 4910 | case RX_HANDLER_ANOTHER: |
63d8ea7f | 4911 | goto another_round; |
8a4eb573 JP |
4912 | case RX_HANDLER_EXACT: |
4913 | deliver_exact = true; | |
4914 | case RX_HANDLER_PASS: | |
4915 | break; | |
4916 | default: | |
4917 | BUG(); | |
4918 | } | |
ab95bfe0 | 4919 | } |
1da177e4 | 4920 | |
df8a39de JP |
4921 | if (unlikely(skb_vlan_tag_present(skb))) { |
4922 | if (skb_vlan_tag_get_id(skb)) | |
d4b812de ED |
4923 | skb->pkt_type = PACKET_OTHERHOST; |
4924 | /* Note: we might in the future use prio bits | |
4925 | * and set skb->priority like in vlan_do_receive() | |
4926 | * For the time being, just ignore Priority Code Point | |
4927 | */ | |
b1817524 | 4928 | __vlan_hwaccel_clear_tag(skb); |
d4b812de | 4929 | } |
48cc32d3 | 4930 | |
7866a621 SN |
4931 | type = skb->protocol; |
4932 | ||
63d8ea7f | 4933 | /* deliver only exact match when indicated */ |
7866a621 SN |
4934 | if (likely(!deliver_exact)) { |
4935 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4936 | &ptype_base[ntohs(type) & | |
4937 | PTYPE_HASH_MASK]); | |
4938 | } | |
1f3c8804 | 4939 | |
7866a621 SN |
4940 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, |
4941 | &orig_dev->ptype_specific); | |
4942 | ||
4943 | if (unlikely(skb->dev != orig_dev)) { | |
4944 | deliver_ptype_list_skb(skb, &pt_prev, orig_dev, type, | |
4945 | &skb->dev->ptype_specific); | |
1da177e4 LT |
4946 | } |
4947 | ||
4948 | if (pt_prev) { | |
1f8b977a | 4949 | if (unlikely(skb_orphan_frags_rx(skb, GFP_ATOMIC))) |
0e698bf6 | 4950 | goto drop; |
88eb1944 | 4951 | *ppt_prev = pt_prev; |
1da177e4 | 4952 | } else { |
b4b9e355 | 4953 | drop: |
6e7333d3 JW |
4954 | if (!deliver_exact) |
4955 | atomic_long_inc(&skb->dev->rx_dropped); | |
4956 | else | |
4957 | atomic_long_inc(&skb->dev->rx_nohandler); | |
1da177e4 LT |
4958 | kfree_skb(skb); |
4959 | /* Jamal, now you will not able to escape explaining | |
4960 | * me how you were going to use this. :-) | |
4961 | */ | |
4962 | ret = NET_RX_DROP; | |
4963 | } | |
4964 | ||
2c17d27c | 4965 | out: |
9754e293 DM |
4966 | return ret; |
4967 | } | |
4968 | ||
88eb1944 EC |
4969 | static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc) |
4970 | { | |
4971 | struct net_device *orig_dev = skb->dev; | |
4972 | struct packet_type *pt_prev = NULL; | |
4973 | int ret; | |
4974 | ||
4975 | ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); | |
4976 | if (pt_prev) | |
4977 | ret = pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
4978 | return ret; | |
4979 | } | |
4980 | ||
1c601d82 JDB |
4981 | /** |
4982 | * netif_receive_skb_core - special purpose version of netif_receive_skb | |
4983 | * @skb: buffer to process | |
4984 | * | |
4985 | * More direct receive version of netif_receive_skb(). It should | |
4986 | * only be used by callers that have a need to skip RPS and Generic XDP. | |
4987 | * Caller must also take care of handling if (page_is_)pfmemalloc. | |
4988 | * | |
4989 | * This function may only be called from softirq context and interrupts | |
4990 | * should be enabled. | |
4991 | * | |
4992 | * Return values (usually ignored): | |
4993 | * NET_RX_SUCCESS: no congestion | |
4994 | * NET_RX_DROP: packet was dropped | |
4995 | */ | |
4996 | int netif_receive_skb_core(struct sk_buff *skb) | |
4997 | { | |
4998 | int ret; | |
4999 | ||
5000 | rcu_read_lock(); | |
88eb1944 | 5001 | ret = __netif_receive_skb_one_core(skb, false); |
1c601d82 JDB |
5002 | rcu_read_unlock(); |
5003 | ||
5004 | return ret; | |
5005 | } | |
5006 | EXPORT_SYMBOL(netif_receive_skb_core); | |
5007 | ||
88eb1944 EC |
5008 | static inline void __netif_receive_skb_list_ptype(struct list_head *head, |
5009 | struct packet_type *pt_prev, | |
5010 | struct net_device *orig_dev) | |
4ce0017a EC |
5011 | { |
5012 | struct sk_buff *skb, *next; | |
5013 | ||
88eb1944 EC |
5014 | if (!pt_prev) |
5015 | return; | |
5016 | if (list_empty(head)) | |
5017 | return; | |
17266ee9 EC |
5018 | if (pt_prev->list_func != NULL) |
5019 | pt_prev->list_func(head, pt_prev, orig_dev); | |
5020 | else | |
5021 | list_for_each_entry_safe(skb, next, head, list) | |
5022 | pt_prev->func(skb, skb->dev, pt_prev, orig_dev); | |
88eb1944 EC |
5023 | } |
5024 | ||
5025 | static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemalloc) | |
5026 | { | |
5027 | /* Fast-path assumptions: | |
5028 | * - There is no RX handler. | |
5029 | * - Only one packet_type matches. | |
5030 | * If either of these fails, we will end up doing some per-packet | |
5031 | * processing in-line, then handling the 'last ptype' for the whole | |
5032 | * sublist. This can't cause out-of-order delivery to any single ptype, | |
5033 | * because the 'last ptype' must be constant across the sublist, and all | |
5034 | * other ptypes are handled per-packet. | |
5035 | */ | |
5036 | /* Current (common) ptype of sublist */ | |
5037 | struct packet_type *pt_curr = NULL; | |
5038 | /* Current (common) orig_dev of sublist */ | |
5039 | struct net_device *od_curr = NULL; | |
5040 | struct list_head sublist; | |
5041 | struct sk_buff *skb, *next; | |
5042 | ||
9af86f93 | 5043 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5044 | list_for_each_entry_safe(skb, next, head, list) { |
5045 | struct net_device *orig_dev = skb->dev; | |
5046 | struct packet_type *pt_prev = NULL; | |
5047 | ||
22f6bbb7 | 5048 | skb_list_del_init(skb); |
88eb1944 | 5049 | __netif_receive_skb_core(skb, pfmemalloc, &pt_prev); |
9af86f93 EC |
5050 | if (!pt_prev) |
5051 | continue; | |
88eb1944 EC |
5052 | if (pt_curr != pt_prev || od_curr != orig_dev) { |
5053 | /* dispatch old sublist */ | |
88eb1944 EC |
5054 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
5055 | /* start new sublist */ | |
9af86f93 | 5056 | INIT_LIST_HEAD(&sublist); |
88eb1944 EC |
5057 | pt_curr = pt_prev; |
5058 | od_curr = orig_dev; | |
5059 | } | |
9af86f93 | 5060 | list_add_tail(&skb->list, &sublist); |
88eb1944 EC |
5061 | } |
5062 | ||
5063 | /* dispatch final sublist */ | |
9af86f93 | 5064 | __netif_receive_skb_list_ptype(&sublist, pt_curr, od_curr); |
4ce0017a EC |
5065 | } |
5066 | ||
9754e293 DM |
5067 | static int __netif_receive_skb(struct sk_buff *skb) |
5068 | { | |
5069 | int ret; | |
5070 | ||
5071 | if (sk_memalloc_socks() && skb_pfmemalloc(skb)) { | |
f1083048 | 5072 | unsigned int noreclaim_flag; |
9754e293 DM |
5073 | |
5074 | /* | |
5075 | * PFMEMALLOC skbs are special, they should | |
5076 | * - be delivered to SOCK_MEMALLOC sockets only | |
5077 | * - stay away from userspace | |
5078 | * - have bounded memory usage | |
5079 | * | |
5080 | * Use PF_MEMALLOC as this saves us from propagating the allocation | |
5081 | * context down to all allocation sites. | |
5082 | */ | |
f1083048 | 5083 | noreclaim_flag = memalloc_noreclaim_save(); |
88eb1944 | 5084 | ret = __netif_receive_skb_one_core(skb, true); |
f1083048 | 5085 | memalloc_noreclaim_restore(noreclaim_flag); |
9754e293 | 5086 | } else |
88eb1944 | 5087 | ret = __netif_receive_skb_one_core(skb, false); |
9754e293 | 5088 | |
1da177e4 LT |
5089 | return ret; |
5090 | } | |
0a9627f2 | 5091 | |
4ce0017a EC |
5092 | static void __netif_receive_skb_list(struct list_head *head) |
5093 | { | |
5094 | unsigned long noreclaim_flag = 0; | |
5095 | struct sk_buff *skb, *next; | |
5096 | bool pfmemalloc = false; /* Is current sublist PF_MEMALLOC? */ | |
5097 | ||
5098 | list_for_each_entry_safe(skb, next, head, list) { | |
5099 | if ((sk_memalloc_socks() && skb_pfmemalloc(skb)) != pfmemalloc) { | |
5100 | struct list_head sublist; | |
5101 | ||
5102 | /* Handle the previous sublist */ | |
5103 | list_cut_before(&sublist, head, &skb->list); | |
b9f463d6 EC |
5104 | if (!list_empty(&sublist)) |
5105 | __netif_receive_skb_list_core(&sublist, pfmemalloc); | |
4ce0017a EC |
5106 | pfmemalloc = !pfmemalloc; |
5107 | /* See comments in __netif_receive_skb */ | |
5108 | if (pfmemalloc) | |
5109 | noreclaim_flag = memalloc_noreclaim_save(); | |
5110 | else | |
5111 | memalloc_noreclaim_restore(noreclaim_flag); | |
5112 | } | |
5113 | } | |
5114 | /* Handle the remaining sublist */ | |
b9f463d6 EC |
5115 | if (!list_empty(head)) |
5116 | __netif_receive_skb_list_core(head, pfmemalloc); | |
4ce0017a EC |
5117 | /* Restore pflags */ |
5118 | if (pfmemalloc) | |
5119 | memalloc_noreclaim_restore(noreclaim_flag); | |
5120 | } | |
5121 | ||
f4e63525 | 5122 | static int generic_xdp_install(struct net_device *dev, struct netdev_bpf *xdp) |
b5cdae32 | 5123 | { |
58038695 | 5124 | struct bpf_prog *old = rtnl_dereference(dev->xdp_prog); |
b5cdae32 DM |
5125 | struct bpf_prog *new = xdp->prog; |
5126 | int ret = 0; | |
5127 | ||
5128 | switch (xdp->command) { | |
58038695 | 5129 | case XDP_SETUP_PROG: |
b5cdae32 DM |
5130 | rcu_assign_pointer(dev->xdp_prog, new); |
5131 | if (old) | |
5132 | bpf_prog_put(old); | |
5133 | ||
5134 | if (old && !new) { | |
02786475 | 5135 | static_branch_dec(&generic_xdp_needed_key); |
b5cdae32 | 5136 | } else if (new && !old) { |
02786475 | 5137 | static_branch_inc(&generic_xdp_needed_key); |
b5cdae32 | 5138 | dev_disable_lro(dev); |
56f5aa77 | 5139 | dev_disable_gro_hw(dev); |
b5cdae32 DM |
5140 | } |
5141 | break; | |
b5cdae32 DM |
5142 | |
5143 | case XDP_QUERY_PROG: | |
58038695 | 5144 | xdp->prog_id = old ? old->aux->id : 0; |
b5cdae32 DM |
5145 | break; |
5146 | ||
5147 | default: | |
5148 | ret = -EINVAL; | |
5149 | break; | |
5150 | } | |
5151 | ||
5152 | return ret; | |
5153 | } | |
5154 | ||
ae78dbfa | 5155 | static int netif_receive_skb_internal(struct sk_buff *skb) |
0a9627f2 | 5156 | { |
2c17d27c JA |
5157 | int ret; |
5158 | ||
588f0330 | 5159 | net_timestamp_check(netdev_tstamp_prequeue, skb); |
3b098e2d | 5160 | |
c1f19b51 RC |
5161 | if (skb_defer_rx_timestamp(skb)) |
5162 | return NET_RX_SUCCESS; | |
5163 | ||
02786475 | 5164 | if (static_branch_unlikely(&generic_xdp_needed_key)) { |
bbbe211c | 5165 | int ret; |
b5cdae32 | 5166 | |
bbbe211c JF |
5167 | preempt_disable(); |
5168 | rcu_read_lock(); | |
5169 | ret = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb); | |
5170 | rcu_read_unlock(); | |
5171 | preempt_enable(); | |
5172 | ||
5173 | if (ret != XDP_PASS) | |
d4455169 | 5174 | return NET_RX_DROP; |
b5cdae32 DM |
5175 | } |
5176 | ||
bbbe211c | 5177 | rcu_read_lock(); |
df334545 | 5178 | #ifdef CONFIG_RPS |
dc05360f | 5179 | if (static_branch_unlikely(&rps_needed)) { |
3b098e2d | 5180 | struct rps_dev_flow voidflow, *rflow = &voidflow; |
2c17d27c | 5181 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); |
0a9627f2 | 5182 | |
3b098e2d ED |
5183 | if (cpu >= 0) { |
5184 | ret = enqueue_to_backlog(skb, cpu, &rflow->last_qtail); | |
5185 | rcu_read_unlock(); | |
adc9300e | 5186 | return ret; |
3b098e2d | 5187 | } |
fec5e652 | 5188 | } |
1e94d72f | 5189 | #endif |
2c17d27c JA |
5190 | ret = __netif_receive_skb(skb); |
5191 | rcu_read_unlock(); | |
5192 | return ret; | |
0a9627f2 | 5193 | } |
ae78dbfa | 5194 | |
7da517a3 EC |
5195 | static void netif_receive_skb_list_internal(struct list_head *head) |
5196 | { | |
5197 | struct bpf_prog *xdp_prog = NULL; | |
5198 | struct sk_buff *skb, *next; | |
8c057efa | 5199 | struct list_head sublist; |
7da517a3 | 5200 | |
8c057efa | 5201 | INIT_LIST_HEAD(&sublist); |
7da517a3 EC |
5202 | list_for_each_entry_safe(skb, next, head, list) { |
5203 | net_timestamp_check(netdev_tstamp_prequeue, skb); | |
22f6bbb7 | 5204 | skb_list_del_init(skb); |
8c057efa EC |
5205 | if (!skb_defer_rx_timestamp(skb)) |
5206 | list_add_tail(&skb->list, &sublist); | |
7da517a3 | 5207 | } |
8c057efa | 5208 | list_splice_init(&sublist, head); |
7da517a3 EC |
5209 | |
5210 | if (static_branch_unlikely(&generic_xdp_needed_key)) { | |
5211 | preempt_disable(); | |
5212 | rcu_read_lock(); | |
5213 | list_for_each_entry_safe(skb, next, head, list) { | |
5214 | xdp_prog = rcu_dereference(skb->dev->xdp_prog); | |
22f6bbb7 | 5215 | skb_list_del_init(skb); |
8c057efa EC |
5216 | if (do_xdp_generic(xdp_prog, skb) == XDP_PASS) |
5217 | list_add_tail(&skb->list, &sublist); | |
7da517a3 EC |
5218 | } |
5219 | rcu_read_unlock(); | |
5220 | preempt_enable(); | |
8c057efa EC |
5221 | /* Put passed packets back on main list */ |
5222 | list_splice_init(&sublist, head); | |
7da517a3 EC |
5223 | } |
5224 | ||
5225 | rcu_read_lock(); | |
5226 | #ifdef CONFIG_RPS | |
dc05360f | 5227 | if (static_branch_unlikely(&rps_needed)) { |
7da517a3 EC |
5228 | list_for_each_entry_safe(skb, next, head, list) { |
5229 | struct rps_dev_flow voidflow, *rflow = &voidflow; | |
5230 | int cpu = get_rps_cpu(skb->dev, skb, &rflow); | |
5231 | ||
5232 | if (cpu >= 0) { | |
8c057efa | 5233 | /* Will be handled, remove from list */ |
22f6bbb7 | 5234 | skb_list_del_init(skb); |
8c057efa | 5235 | enqueue_to_backlog(skb, cpu, &rflow->last_qtail); |
7da517a3 EC |
5236 | } |
5237 | } | |
5238 | } | |
5239 | #endif | |
5240 | __netif_receive_skb_list(head); | |
5241 | rcu_read_unlock(); | |
5242 | } | |
5243 | ||
ae78dbfa BH |
5244 | /** |
5245 | * netif_receive_skb - process receive buffer from network | |
5246 | * @skb: buffer to process | |
5247 | * | |
5248 | * netif_receive_skb() is the main receive data processing function. | |
5249 | * It always succeeds. The buffer may be dropped during processing | |
5250 | * for congestion control or by the protocol layers. | |
5251 | * | |
5252 | * This function may only be called from softirq context and interrupts | |
5253 | * should be enabled. | |
5254 | * | |
5255 | * Return values (usually ignored): | |
5256 | * NET_RX_SUCCESS: no congestion | |
5257 | * NET_RX_DROP: packet was dropped | |
5258 | */ | |
04eb4489 | 5259 | int netif_receive_skb(struct sk_buff *skb) |
ae78dbfa | 5260 | { |
b0e3f1bd GB |
5261 | int ret; |
5262 | ||
ae78dbfa BH |
5263 | trace_netif_receive_skb_entry(skb); |
5264 | ||
b0e3f1bd GB |
5265 | ret = netif_receive_skb_internal(skb); |
5266 | trace_netif_receive_skb_exit(ret); | |
5267 | ||
5268 | return ret; | |
ae78dbfa | 5269 | } |
04eb4489 | 5270 | EXPORT_SYMBOL(netif_receive_skb); |
1da177e4 | 5271 | |
f6ad8c1b EC |
5272 | /** |
5273 | * netif_receive_skb_list - process many receive buffers from network | |
5274 | * @head: list of skbs to process. | |
5275 | * | |
7da517a3 EC |
5276 | * Since return value of netif_receive_skb() is normally ignored, and |
5277 | * wouldn't be meaningful for a list, this function returns void. | |
f6ad8c1b EC |
5278 | * |
5279 | * This function may only be called from softirq context and interrupts | |
5280 | * should be enabled. | |
5281 | */ | |
5282 | void netif_receive_skb_list(struct list_head *head) | |
5283 | { | |
7da517a3 | 5284 | struct sk_buff *skb; |
f6ad8c1b | 5285 | |
b9f463d6 EC |
5286 | if (list_empty(head)) |
5287 | return; | |
b0e3f1bd GB |
5288 | if (trace_netif_receive_skb_list_entry_enabled()) { |
5289 | list_for_each_entry(skb, head, list) | |
5290 | trace_netif_receive_skb_list_entry(skb); | |
5291 | } | |
7da517a3 | 5292 | netif_receive_skb_list_internal(head); |
b0e3f1bd | 5293 | trace_netif_receive_skb_list_exit(0); |
f6ad8c1b EC |
5294 | } |
5295 | EXPORT_SYMBOL(netif_receive_skb_list); | |
5296 | ||
41852497 | 5297 | DEFINE_PER_CPU(struct work_struct, flush_works); |
145dd5f9 PA |
5298 | |
5299 | /* Network device is going away, flush any packets still pending */ | |
5300 | static void flush_backlog(struct work_struct *work) | |
6e583ce5 | 5301 | { |
6e583ce5 | 5302 | struct sk_buff *skb, *tmp; |
145dd5f9 PA |
5303 | struct softnet_data *sd; |
5304 | ||
5305 | local_bh_disable(); | |
5306 | sd = this_cpu_ptr(&softnet_data); | |
6e583ce5 | 5307 | |
145dd5f9 | 5308 | local_irq_disable(); |
e36fa2f7 | 5309 | rps_lock(sd); |
6e7676c1 | 5310 | skb_queue_walk_safe(&sd->input_pkt_queue, skb, tmp) { |
41852497 | 5311 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
e36fa2f7 | 5312 | __skb_unlink(skb, &sd->input_pkt_queue); |
6e583ce5 | 5313 | kfree_skb(skb); |
76cc8b13 | 5314 | input_queue_head_incr(sd); |
6e583ce5 | 5315 | } |
6e7676c1 | 5316 | } |
e36fa2f7 | 5317 | rps_unlock(sd); |
145dd5f9 | 5318 | local_irq_enable(); |
6e7676c1 CG |
5319 | |
5320 | skb_queue_walk_safe(&sd->process_queue, skb, tmp) { | |
41852497 | 5321 | if (skb->dev->reg_state == NETREG_UNREGISTERING) { |
6e7676c1 CG |
5322 | __skb_unlink(skb, &sd->process_queue); |
5323 | kfree_skb(skb); | |
76cc8b13 | 5324 | input_queue_head_incr(sd); |
6e7676c1 CG |
5325 | } |
5326 | } | |
145dd5f9 PA |
5327 | local_bh_enable(); |
5328 | } | |
5329 | ||
41852497 | 5330 | static void flush_all_backlogs(void) |
145dd5f9 PA |
5331 | { |
5332 | unsigned int cpu; | |
5333 | ||
5334 | get_online_cpus(); | |
5335 | ||
41852497 ED |
5336 | for_each_online_cpu(cpu) |
5337 | queue_work_on(cpu, system_highpri_wq, | |
5338 | per_cpu_ptr(&flush_works, cpu)); | |
145dd5f9 PA |
5339 | |
5340 | for_each_online_cpu(cpu) | |
41852497 | 5341 | flush_work(per_cpu_ptr(&flush_works, cpu)); |
145dd5f9 PA |
5342 | |
5343 | put_online_cpus(); | |
6e583ce5 SH |
5344 | } |
5345 | ||
aaa5d90b PA |
5346 | INDIRECT_CALLABLE_DECLARE(int inet_gro_complete(struct sk_buff *, int)); |
5347 | INDIRECT_CALLABLE_DECLARE(int ipv6_gro_complete(struct sk_buff *, int)); | |
d565b0a1 HX |
5348 | static int napi_gro_complete(struct sk_buff *skb) |
5349 | { | |
22061d80 | 5350 | struct packet_offload *ptype; |
d565b0a1 | 5351 | __be16 type = skb->protocol; |
22061d80 | 5352 | struct list_head *head = &offload_base; |
d565b0a1 HX |
5353 | int err = -ENOENT; |
5354 | ||
c3c7c254 ED |
5355 | BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb)); |
5356 | ||
fc59f9a3 HX |
5357 | if (NAPI_GRO_CB(skb)->count == 1) { |
5358 | skb_shinfo(skb)->gso_size = 0; | |
d565b0a1 | 5359 | goto out; |
fc59f9a3 | 5360 | } |
d565b0a1 HX |
5361 | |
5362 | rcu_read_lock(); | |
5363 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5364 | if (ptype->type != type || !ptype->callbacks.gro_complete) |
d565b0a1 HX |
5365 | continue; |
5366 | ||
aaa5d90b PA |
5367 | err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete, |
5368 | ipv6_gro_complete, inet_gro_complete, | |
5369 | skb, 0); | |
d565b0a1 HX |
5370 | break; |
5371 | } | |
5372 | rcu_read_unlock(); | |
5373 | ||
5374 | if (err) { | |
5375 | WARN_ON(&ptype->list == head); | |
5376 | kfree_skb(skb); | |
5377 | return NET_RX_SUCCESS; | |
5378 | } | |
5379 | ||
5380 | out: | |
ae78dbfa | 5381 | return netif_receive_skb_internal(skb); |
d565b0a1 HX |
5382 | } |
5383 | ||
6312fe77 | 5384 | static void __napi_gro_flush_chain(struct napi_struct *napi, u32 index, |
07d78363 | 5385 | bool flush_old) |
d565b0a1 | 5386 | { |
6312fe77 | 5387 | struct list_head *head = &napi->gro_hash[index].list; |
d4546c25 | 5388 | struct sk_buff *skb, *p; |
2e71a6f8 | 5389 | |
07d78363 | 5390 | list_for_each_entry_safe_reverse(skb, p, head, list) { |
2e71a6f8 ED |
5391 | if (flush_old && NAPI_GRO_CB(skb)->age == jiffies) |
5392 | return; | |
992cba7e | 5393 | skb_list_del_init(skb); |
d565b0a1 | 5394 | napi_gro_complete(skb); |
6312fe77 | 5395 | napi->gro_hash[index].count--; |
d565b0a1 | 5396 | } |
d9f37d01 LR |
5397 | |
5398 | if (!napi->gro_hash[index].count) | |
5399 | __clear_bit(index, &napi->gro_bitmask); | |
d565b0a1 | 5400 | } |
07d78363 | 5401 | |
6312fe77 | 5402 | /* napi->gro_hash[].list contains packets ordered by age. |
07d78363 DM |
5403 | * youngest packets at the head of it. |
5404 | * Complete skbs in reverse order to reduce latencies. | |
5405 | */ | |
5406 | void napi_gro_flush(struct napi_struct *napi, bool flush_old) | |
5407 | { | |
42519ede ED |
5408 | unsigned long bitmask = napi->gro_bitmask; |
5409 | unsigned int i, base = ~0U; | |
07d78363 | 5410 | |
42519ede ED |
5411 | while ((i = ffs(bitmask)) != 0) { |
5412 | bitmask >>= i; | |
5413 | base += i; | |
5414 | __napi_gro_flush_chain(napi, base, flush_old); | |
d9f37d01 | 5415 | } |
07d78363 | 5416 | } |
86cac58b | 5417 | EXPORT_SYMBOL(napi_gro_flush); |
d565b0a1 | 5418 | |
07d78363 DM |
5419 | static struct list_head *gro_list_prepare(struct napi_struct *napi, |
5420 | struct sk_buff *skb) | |
89c5fa33 | 5421 | { |
89c5fa33 | 5422 | unsigned int maclen = skb->dev->hard_header_len; |
0b4cec8c | 5423 | u32 hash = skb_get_hash_raw(skb); |
07d78363 | 5424 | struct list_head *head; |
d4546c25 | 5425 | struct sk_buff *p; |
89c5fa33 | 5426 | |
6312fe77 | 5427 | head = &napi->gro_hash[hash & (GRO_HASH_BUCKETS - 1)].list; |
07d78363 | 5428 | list_for_each_entry(p, head, list) { |
89c5fa33 ED |
5429 | unsigned long diffs; |
5430 | ||
0b4cec8c TH |
5431 | NAPI_GRO_CB(p)->flush = 0; |
5432 | ||
5433 | if (hash != skb_get_hash_raw(p)) { | |
5434 | NAPI_GRO_CB(p)->same_flow = 0; | |
5435 | continue; | |
5436 | } | |
5437 | ||
89c5fa33 | 5438 | diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev; |
b1817524 MM |
5439 | diffs |= skb_vlan_tag_present(p) ^ skb_vlan_tag_present(skb); |
5440 | if (skb_vlan_tag_present(p)) | |
5441 | diffs |= p->vlan_tci ^ skb->vlan_tci; | |
ce87fc6c | 5442 | diffs |= skb_metadata_dst_cmp(p, skb); |
de8f3a83 | 5443 | diffs |= skb_metadata_differs(p, skb); |
89c5fa33 ED |
5444 | if (maclen == ETH_HLEN) |
5445 | diffs |= compare_ether_header(skb_mac_header(p), | |
a50e233c | 5446 | skb_mac_header(skb)); |
89c5fa33 ED |
5447 | else if (!diffs) |
5448 | diffs = memcmp(skb_mac_header(p), | |
a50e233c | 5449 | skb_mac_header(skb), |
89c5fa33 ED |
5450 | maclen); |
5451 | NAPI_GRO_CB(p)->same_flow = !diffs; | |
89c5fa33 | 5452 | } |
07d78363 DM |
5453 | |
5454 | return head; | |
89c5fa33 ED |
5455 | } |
5456 | ||
299603e8 JC |
5457 | static void skb_gro_reset_offset(struct sk_buff *skb) |
5458 | { | |
5459 | const struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5460 | const skb_frag_t *frag0 = &pinfo->frags[0]; | |
5461 | ||
5462 | NAPI_GRO_CB(skb)->data_offset = 0; | |
5463 | NAPI_GRO_CB(skb)->frag0 = NULL; | |
5464 | NAPI_GRO_CB(skb)->frag0_len = 0; | |
5465 | ||
5466 | if (skb_mac_header(skb) == skb_tail_pointer(skb) && | |
5467 | pinfo->nr_frags && | |
5468 | !PageHighMem(skb_frag_page(frag0))) { | |
5469 | NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0); | |
7cfd5fd5 ED |
5470 | NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int, |
5471 | skb_frag_size(frag0), | |
5472 | skb->end - skb->tail); | |
89c5fa33 ED |
5473 | } |
5474 | } | |
5475 | ||
a50e233c ED |
5476 | static void gro_pull_from_frag0(struct sk_buff *skb, int grow) |
5477 | { | |
5478 | struct skb_shared_info *pinfo = skb_shinfo(skb); | |
5479 | ||
5480 | BUG_ON(skb->end - skb->tail < grow); | |
5481 | ||
5482 | memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow); | |
5483 | ||
5484 | skb->data_len -= grow; | |
5485 | skb->tail += grow; | |
5486 | ||
5487 | pinfo->frags[0].page_offset += grow; | |
5488 | skb_frag_size_sub(&pinfo->frags[0], grow); | |
5489 | ||
5490 | if (unlikely(!skb_frag_size(&pinfo->frags[0]))) { | |
5491 | skb_frag_unref(skb, 0); | |
5492 | memmove(pinfo->frags, pinfo->frags + 1, | |
5493 | --pinfo->nr_frags * sizeof(pinfo->frags[0])); | |
5494 | } | |
5495 | } | |
5496 | ||
6312fe77 | 5497 | static void gro_flush_oldest(struct list_head *head) |
07d78363 | 5498 | { |
6312fe77 | 5499 | struct sk_buff *oldest; |
07d78363 | 5500 | |
6312fe77 | 5501 | oldest = list_last_entry(head, struct sk_buff, list); |
07d78363 | 5502 | |
6312fe77 | 5503 | /* We are called with head length >= MAX_GRO_SKBS, so this is |
07d78363 DM |
5504 | * impossible. |
5505 | */ | |
5506 | if (WARN_ON_ONCE(!oldest)) | |
5507 | return; | |
5508 | ||
d9f37d01 LR |
5509 | /* Do not adjust napi->gro_hash[].count, caller is adding a new |
5510 | * SKB to the chain. | |
07d78363 | 5511 | */ |
ece23711 | 5512 | skb_list_del_init(oldest); |
07d78363 DM |
5513 | napi_gro_complete(oldest); |
5514 | } | |
5515 | ||
aaa5d90b PA |
5516 | INDIRECT_CALLABLE_DECLARE(struct sk_buff *inet_gro_receive(struct list_head *, |
5517 | struct sk_buff *)); | |
5518 | INDIRECT_CALLABLE_DECLARE(struct sk_buff *ipv6_gro_receive(struct list_head *, | |
5519 | struct sk_buff *)); | |
bb728820 | 5520 | static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
d565b0a1 | 5521 | { |
6312fe77 | 5522 | u32 hash = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1); |
d4546c25 | 5523 | struct list_head *head = &offload_base; |
22061d80 | 5524 | struct packet_offload *ptype; |
d565b0a1 | 5525 | __be16 type = skb->protocol; |
07d78363 | 5526 | struct list_head *gro_head; |
d4546c25 | 5527 | struct sk_buff *pp = NULL; |
5b252f0c | 5528 | enum gro_result ret; |
d4546c25 | 5529 | int same_flow; |
a50e233c | 5530 | int grow; |
d565b0a1 | 5531 | |
b5cdae32 | 5532 | if (netif_elide_gro(skb->dev)) |
d565b0a1 HX |
5533 | goto normal; |
5534 | ||
07d78363 | 5535 | gro_head = gro_list_prepare(napi, skb); |
89c5fa33 | 5536 | |
d565b0a1 HX |
5537 | rcu_read_lock(); |
5538 | list_for_each_entry_rcu(ptype, head, list) { | |
f191a1d1 | 5539 | if (ptype->type != type || !ptype->callbacks.gro_receive) |
d565b0a1 HX |
5540 | continue; |
5541 | ||
86911732 | 5542 | skb_set_network_header(skb, skb_gro_offset(skb)); |
efd9450e | 5543 | skb_reset_mac_len(skb); |
d565b0a1 | 5544 | NAPI_GRO_CB(skb)->same_flow = 0; |
d61d072e | 5545 | NAPI_GRO_CB(skb)->flush = skb_is_gso(skb) || skb_has_frag_list(skb); |
5d38a079 | 5546 | NAPI_GRO_CB(skb)->free = 0; |
fac8e0f5 | 5547 | NAPI_GRO_CB(skb)->encap_mark = 0; |
fcd91dd4 | 5548 | NAPI_GRO_CB(skb)->recursion_counter = 0; |
a0ca153f | 5549 | NAPI_GRO_CB(skb)->is_fou = 0; |
1530545e | 5550 | NAPI_GRO_CB(skb)->is_atomic = 1; |
15e2396d | 5551 | NAPI_GRO_CB(skb)->gro_remcsum_start = 0; |
d565b0a1 | 5552 | |
662880f4 TH |
5553 | /* Setup for GRO checksum validation */ |
5554 | switch (skb->ip_summed) { | |
5555 | case CHECKSUM_COMPLETE: | |
5556 | NAPI_GRO_CB(skb)->csum = skb->csum; | |
5557 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5558 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5559 | break; | |
5560 | case CHECKSUM_UNNECESSARY: | |
5561 | NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1; | |
5562 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
5563 | break; | |
5564 | default: | |
5565 | NAPI_GRO_CB(skb)->csum_cnt = 0; | |
5566 | NAPI_GRO_CB(skb)->csum_valid = 0; | |
5567 | } | |
d565b0a1 | 5568 | |
aaa5d90b PA |
5569 | pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive, |
5570 | ipv6_gro_receive, inet_gro_receive, | |
5571 | gro_head, skb); | |
d565b0a1 HX |
5572 | break; |
5573 | } | |
5574 | rcu_read_unlock(); | |
5575 | ||
5576 | if (&ptype->list == head) | |
5577 | goto normal; | |
5578 | ||
25393d3f SK |
5579 | if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) { |
5580 | ret = GRO_CONSUMED; | |
5581 | goto ok; | |
5582 | } | |
5583 | ||
0da2afd5 | 5584 | same_flow = NAPI_GRO_CB(skb)->same_flow; |
5d0d9be8 | 5585 | ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED; |
0da2afd5 | 5586 | |
d565b0a1 | 5587 | if (pp) { |
992cba7e | 5588 | skb_list_del_init(pp); |
d4546c25 | 5589 | napi_gro_complete(pp); |
6312fe77 | 5590 | napi->gro_hash[hash].count--; |
d565b0a1 HX |
5591 | } |
5592 | ||
0da2afd5 | 5593 | if (same_flow) |
d565b0a1 HX |
5594 | goto ok; |
5595 | ||
600adc18 | 5596 | if (NAPI_GRO_CB(skb)->flush) |
d565b0a1 | 5597 | goto normal; |
d565b0a1 | 5598 | |
6312fe77 LR |
5599 | if (unlikely(napi->gro_hash[hash].count >= MAX_GRO_SKBS)) { |
5600 | gro_flush_oldest(gro_head); | |
600adc18 | 5601 | } else { |
6312fe77 | 5602 | napi->gro_hash[hash].count++; |
600adc18 | 5603 | } |
d565b0a1 | 5604 | NAPI_GRO_CB(skb)->count = 1; |
2e71a6f8 | 5605 | NAPI_GRO_CB(skb)->age = jiffies; |
29e98242 | 5606 | NAPI_GRO_CB(skb)->last = skb; |
86911732 | 5607 | skb_shinfo(skb)->gso_size = skb_gro_len(skb); |
07d78363 | 5608 | list_add(&skb->list, gro_head); |
5d0d9be8 | 5609 | ret = GRO_HELD; |
d565b0a1 | 5610 | |
ad0f9904 | 5611 | pull: |
a50e233c ED |
5612 | grow = skb_gro_offset(skb) - skb_headlen(skb); |
5613 | if (grow > 0) | |
5614 | gro_pull_from_frag0(skb, grow); | |
d565b0a1 | 5615 | ok: |
d9f37d01 LR |
5616 | if (napi->gro_hash[hash].count) { |
5617 | if (!test_bit(hash, &napi->gro_bitmask)) | |
5618 | __set_bit(hash, &napi->gro_bitmask); | |
5619 | } else if (test_bit(hash, &napi->gro_bitmask)) { | |
5620 | __clear_bit(hash, &napi->gro_bitmask); | |
5621 | } | |
5622 | ||
5d0d9be8 | 5623 | return ret; |
d565b0a1 HX |
5624 | |
5625 | normal: | |
ad0f9904 HX |
5626 | ret = GRO_NORMAL; |
5627 | goto pull; | |
5d38a079 | 5628 | } |
96e93eab | 5629 | |
bf5a755f JC |
5630 | struct packet_offload *gro_find_receive_by_type(__be16 type) |
5631 | { | |
5632 | struct list_head *offload_head = &offload_base; | |
5633 | struct packet_offload *ptype; | |
5634 | ||
5635 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
5636 | if (ptype->type != type || !ptype->callbacks.gro_receive) | |
5637 | continue; | |
5638 | return ptype; | |
5639 | } | |
5640 | return NULL; | |
5641 | } | |
e27a2f83 | 5642 | EXPORT_SYMBOL(gro_find_receive_by_type); |
bf5a755f JC |
5643 | |
5644 | struct packet_offload *gro_find_complete_by_type(__be16 type) | |
5645 | { | |
5646 | struct list_head *offload_head = &offload_base; | |
5647 | struct packet_offload *ptype; | |
5648 | ||
5649 | list_for_each_entry_rcu(ptype, offload_head, list) { | |
5650 | if (ptype->type != type || !ptype->callbacks.gro_complete) | |
5651 | continue; | |
5652 | return ptype; | |
5653 | } | |
5654 | return NULL; | |
5655 | } | |
e27a2f83 | 5656 | EXPORT_SYMBOL(gro_find_complete_by_type); |
5d38a079 | 5657 | |
e44699d2 MK |
5658 | static void napi_skb_free_stolen_head(struct sk_buff *skb) |
5659 | { | |
5660 | skb_dst_drop(skb); | |
5661 | secpath_reset(skb); | |
5662 | kmem_cache_free(skbuff_head_cache, skb); | |
5663 | } | |
5664 | ||
bb728820 | 5665 | static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb) |
5d38a079 | 5666 | { |
5d0d9be8 HX |
5667 | switch (ret) { |
5668 | case GRO_NORMAL: | |
ae78dbfa | 5669 | if (netif_receive_skb_internal(skb)) |
c7c4b3b6 BH |
5670 | ret = GRO_DROP; |
5671 | break; | |
5d38a079 | 5672 | |
5d0d9be8 | 5673 | case GRO_DROP: |
5d38a079 HX |
5674 | kfree_skb(skb); |
5675 | break; | |
5b252f0c | 5676 | |
daa86548 | 5677 | case GRO_MERGED_FREE: |
e44699d2 MK |
5678 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) |
5679 | napi_skb_free_stolen_head(skb); | |
5680 | else | |
d7e8883c | 5681 | __kfree_skb(skb); |
daa86548 ED |
5682 | break; |
5683 | ||
5b252f0c BH |
5684 | case GRO_HELD: |
5685 | case GRO_MERGED: | |
25393d3f | 5686 | case GRO_CONSUMED: |
5b252f0c | 5687 | break; |
5d38a079 HX |
5688 | } |
5689 | ||
c7c4b3b6 | 5690 | return ret; |
5d0d9be8 | 5691 | } |
5d0d9be8 | 5692 | |
c7c4b3b6 | 5693 | gro_result_t napi_gro_receive(struct napi_struct *napi, struct sk_buff *skb) |
5d0d9be8 | 5694 | { |
b0e3f1bd GB |
5695 | gro_result_t ret; |
5696 | ||
93f93a44 | 5697 | skb_mark_napi_id(skb, napi); |
ae78dbfa | 5698 | trace_napi_gro_receive_entry(skb); |
86911732 | 5699 | |
a50e233c ED |
5700 | skb_gro_reset_offset(skb); |
5701 | ||
b0e3f1bd GB |
5702 | ret = napi_skb_finish(dev_gro_receive(napi, skb), skb); |
5703 | trace_napi_gro_receive_exit(ret); | |
5704 | ||
5705 | return ret; | |
d565b0a1 HX |
5706 | } |
5707 | EXPORT_SYMBOL(napi_gro_receive); | |
5708 | ||
d0c2b0d2 | 5709 | static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) |
96e93eab | 5710 | { |
93a35f59 ED |
5711 | if (unlikely(skb->pfmemalloc)) { |
5712 | consume_skb(skb); | |
5713 | return; | |
5714 | } | |
96e93eab | 5715 | __skb_pull(skb, skb_headlen(skb)); |
2a2a459e ED |
5716 | /* restore the reserve we had after netdev_alloc_skb_ip_align() */ |
5717 | skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb)); | |
b1817524 | 5718 | __vlan_hwaccel_clear_tag(skb); |
66c46d74 | 5719 | skb->dev = napi->dev; |
6d152e23 | 5720 | skb->skb_iif = 0; |
33d9a2c7 ED |
5721 | |
5722 | /* eth_type_trans() assumes pkt_type is PACKET_HOST */ | |
5723 | skb->pkt_type = PACKET_HOST; | |
5724 | ||
c3caf119 JC |
5725 | skb->encapsulation = 0; |
5726 | skb_shinfo(skb)->gso_type = 0; | |
e33d0ba8 | 5727 | skb->truesize = SKB_TRUESIZE(skb_end_offset(skb)); |
f991bb9d | 5728 | secpath_reset(skb); |
96e93eab HX |
5729 | |
5730 | napi->skb = skb; | |
5731 | } | |
96e93eab | 5732 | |
76620aaf | 5733 | struct sk_buff *napi_get_frags(struct napi_struct *napi) |
5d38a079 | 5734 | { |
5d38a079 | 5735 | struct sk_buff *skb = napi->skb; |
5d38a079 HX |
5736 | |
5737 | if (!skb) { | |
fd11a83d | 5738 | skb = napi_alloc_skb(napi, GRO_MAX_HEAD); |
e2f9dc3b ED |
5739 | if (skb) { |
5740 | napi->skb = skb; | |
5741 | skb_mark_napi_id(skb, napi); | |
5742 | } | |
80595d59 | 5743 | } |
96e93eab HX |
5744 | return skb; |
5745 | } | |
76620aaf | 5746 | EXPORT_SYMBOL(napi_get_frags); |
96e93eab | 5747 | |
a50e233c ED |
5748 | static gro_result_t napi_frags_finish(struct napi_struct *napi, |
5749 | struct sk_buff *skb, | |
5750 | gro_result_t ret) | |
96e93eab | 5751 | { |
5d0d9be8 HX |
5752 | switch (ret) { |
5753 | case GRO_NORMAL: | |
a50e233c ED |
5754 | case GRO_HELD: |
5755 | __skb_push(skb, ETH_HLEN); | |
5756 | skb->protocol = eth_type_trans(skb, skb->dev); | |
5757 | if (ret == GRO_NORMAL && netif_receive_skb_internal(skb)) | |
c7c4b3b6 | 5758 | ret = GRO_DROP; |
86911732 | 5759 | break; |
5d38a079 | 5760 | |
5d0d9be8 | 5761 | case GRO_DROP: |
5d0d9be8 HX |
5762 | napi_reuse_skb(napi, skb); |
5763 | break; | |
5b252f0c | 5764 | |
e44699d2 MK |
5765 | case GRO_MERGED_FREE: |
5766 | if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD) | |
5767 | napi_skb_free_stolen_head(skb); | |
5768 | else | |
5769 | napi_reuse_skb(napi, skb); | |
5770 | break; | |
5771 | ||
5b252f0c | 5772 | case GRO_MERGED: |
25393d3f | 5773 | case GRO_CONSUMED: |
5b252f0c | 5774 | break; |
5d0d9be8 | 5775 | } |
5d38a079 | 5776 | |
c7c4b3b6 | 5777 | return ret; |
5d38a079 | 5778 | } |
5d0d9be8 | 5779 | |
a50e233c ED |
5780 | /* Upper GRO stack assumes network header starts at gro_offset=0 |
5781 | * Drivers could call both napi_gro_frags() and napi_gro_receive() | |
5782 | * We copy ethernet header into skb->data to have a common layout. | |
5783 | */ | |
4adb9c4a | 5784 | static struct sk_buff *napi_frags_skb(struct napi_struct *napi) |
76620aaf HX |
5785 | { |
5786 | struct sk_buff *skb = napi->skb; | |
a50e233c ED |
5787 | const struct ethhdr *eth; |
5788 | unsigned int hlen = sizeof(*eth); | |
76620aaf HX |
5789 | |
5790 | napi->skb = NULL; | |
5791 | ||
a50e233c ED |
5792 | skb_reset_mac_header(skb); |
5793 | skb_gro_reset_offset(skb); | |
5794 | ||
5795 | eth = skb_gro_header_fast(skb, 0); | |
5796 | if (unlikely(skb_gro_header_hard(skb, hlen))) { | |
5797 | eth = skb_gro_header_slow(skb, hlen, 0); | |
5798 | if (unlikely(!eth)) { | |
4da46ceb AC |
5799 | net_warn_ratelimited("%s: dropping impossible skb from %s\n", |
5800 | __func__, napi->dev->name); | |
a50e233c ED |
5801 | napi_reuse_skb(napi, skb); |
5802 | return NULL; | |
5803 | } | |
5804 | } else { | |
5805 | gro_pull_from_frag0(skb, hlen); | |
5806 | NAPI_GRO_CB(skb)->frag0 += hlen; | |
5807 | NAPI_GRO_CB(skb)->frag0_len -= hlen; | |
76620aaf | 5808 | } |
a50e233c ED |
5809 | __skb_pull(skb, hlen); |
5810 | ||
5811 | /* | |
5812 | * This works because the only protocols we care about don't require | |
5813 | * special handling. | |
5814 | * We'll fix it up properly in napi_frags_finish() | |
5815 | */ | |
5816 | skb->protocol = eth->h_proto; | |
76620aaf | 5817 | |
76620aaf HX |
5818 | return skb; |
5819 | } | |
76620aaf | 5820 | |
c7c4b3b6 | 5821 | gro_result_t napi_gro_frags(struct napi_struct *napi) |
5d0d9be8 | 5822 | { |
b0e3f1bd | 5823 | gro_result_t ret; |
76620aaf | 5824 | struct sk_buff *skb = napi_frags_skb(napi); |
5d0d9be8 HX |
5825 | |
5826 | if (!skb) | |
c7c4b3b6 | 5827 | return GRO_DROP; |
5d0d9be8 | 5828 | |
ae78dbfa BH |
5829 | trace_napi_gro_frags_entry(skb); |
5830 | ||
b0e3f1bd GB |
5831 | ret = napi_frags_finish(napi, skb, dev_gro_receive(napi, skb)); |
5832 | trace_napi_gro_frags_exit(ret); | |
5833 | ||
5834 | return ret; | |
5d0d9be8 | 5835 | } |
5d38a079 HX |
5836 | EXPORT_SYMBOL(napi_gro_frags); |
5837 | ||
573e8fca TH |
5838 | /* Compute the checksum from gro_offset and return the folded value |
5839 | * after adding in any pseudo checksum. | |
5840 | */ | |
5841 | __sum16 __skb_gro_checksum_complete(struct sk_buff *skb) | |
5842 | { | |
5843 | __wsum wsum; | |
5844 | __sum16 sum; | |
5845 | ||
5846 | wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0); | |
5847 | ||
5848 | /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */ | |
5849 | sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum)); | |
14641931 | 5850 | /* See comments in __skb_checksum_complete(). */ |
573e8fca TH |
5851 | if (likely(!sum)) { |
5852 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && | |
5853 | !skb->csum_complete_sw) | |
7fe50ac8 | 5854 | netdev_rx_csum_fault(skb->dev, skb); |
573e8fca TH |
5855 | } |
5856 | ||
5857 | NAPI_GRO_CB(skb)->csum = wsum; | |
5858 | NAPI_GRO_CB(skb)->csum_valid = 1; | |
5859 | ||
5860 | return sum; | |
5861 | } | |
5862 | EXPORT_SYMBOL(__skb_gro_checksum_complete); | |
5863 | ||
773fc8f6 | 5864 | static void net_rps_send_ipi(struct softnet_data *remsd) |
5865 | { | |
5866 | #ifdef CONFIG_RPS | |
5867 | while (remsd) { | |
5868 | struct softnet_data *next = remsd->rps_ipi_next; | |
5869 | ||
5870 | if (cpu_online(remsd->cpu)) | |
5871 | smp_call_function_single_async(remsd->cpu, &remsd->csd); | |
5872 | remsd = next; | |
5873 | } | |
5874 | #endif | |
5875 | } | |
5876 | ||
e326bed2 | 5877 | /* |
855abcf0 | 5878 | * net_rps_action_and_irq_enable sends any pending IPI's for rps. |
e326bed2 ED |
5879 | * Note: called with local irq disabled, but exits with local irq enabled. |
5880 | */ | |
5881 | static void net_rps_action_and_irq_enable(struct softnet_data *sd) | |
5882 | { | |
5883 | #ifdef CONFIG_RPS | |
5884 | struct softnet_data *remsd = sd->rps_ipi_list; | |
5885 | ||
5886 | if (remsd) { | |
5887 | sd->rps_ipi_list = NULL; | |
5888 | ||
5889 | local_irq_enable(); | |
5890 | ||
5891 | /* Send pending IPI's to kick RPS processing on remote cpus. */ | |
773fc8f6 | 5892 | net_rps_send_ipi(remsd); |
e326bed2 ED |
5893 | } else |
5894 | #endif | |
5895 | local_irq_enable(); | |
5896 | } | |
5897 | ||
d75b1ade ED |
5898 | static bool sd_has_rps_ipi_waiting(struct softnet_data *sd) |
5899 | { | |
5900 | #ifdef CONFIG_RPS | |
5901 | return sd->rps_ipi_list != NULL; | |
5902 | #else | |
5903 | return false; | |
5904 | #endif | |
5905 | } | |
5906 | ||
bea3348e | 5907 | static int process_backlog(struct napi_struct *napi, int quota) |
1da177e4 | 5908 | { |
eecfd7c4 | 5909 | struct softnet_data *sd = container_of(napi, struct softnet_data, backlog); |
145dd5f9 PA |
5910 | bool again = true; |
5911 | int work = 0; | |
1da177e4 | 5912 | |
e326bed2 ED |
5913 | /* Check if we have pending ipi, its better to send them now, |
5914 | * not waiting net_rx_action() end. | |
5915 | */ | |
d75b1ade | 5916 | if (sd_has_rps_ipi_waiting(sd)) { |
e326bed2 ED |
5917 | local_irq_disable(); |
5918 | net_rps_action_and_irq_enable(sd); | |
5919 | } | |
d75b1ade | 5920 | |
3d48b53f | 5921 | napi->weight = dev_rx_weight; |
145dd5f9 | 5922 | while (again) { |
1da177e4 | 5923 | struct sk_buff *skb; |
6e7676c1 CG |
5924 | |
5925 | while ((skb = __skb_dequeue(&sd->process_queue))) { | |
2c17d27c | 5926 | rcu_read_lock(); |
6e7676c1 | 5927 | __netif_receive_skb(skb); |
2c17d27c | 5928 | rcu_read_unlock(); |
76cc8b13 | 5929 | input_queue_head_incr(sd); |
145dd5f9 | 5930 | if (++work >= quota) |
76cc8b13 | 5931 | return work; |
145dd5f9 | 5932 | |
6e7676c1 | 5933 | } |
1da177e4 | 5934 | |
145dd5f9 | 5935 | local_irq_disable(); |
e36fa2f7 | 5936 | rps_lock(sd); |
11ef7a89 | 5937 | if (skb_queue_empty(&sd->input_pkt_queue)) { |
eecfd7c4 ED |
5938 | /* |
5939 | * Inline a custom version of __napi_complete(). | |
5940 | * only current cpu owns and manipulates this napi, | |
11ef7a89 TH |
5941 | * and NAPI_STATE_SCHED is the only possible flag set |
5942 | * on backlog. | |
5943 | * We can use a plain write instead of clear_bit(), | |
eecfd7c4 ED |
5944 | * and we dont need an smp_mb() memory barrier. |
5945 | */ | |
eecfd7c4 | 5946 | napi->state = 0; |
145dd5f9 PA |
5947 | again = false; |
5948 | } else { | |
5949 | skb_queue_splice_tail_init(&sd->input_pkt_queue, | |
5950 | &sd->process_queue); | |
bea3348e | 5951 | } |
e36fa2f7 | 5952 | rps_unlock(sd); |
145dd5f9 | 5953 | local_irq_enable(); |
6e7676c1 | 5954 | } |
1da177e4 | 5955 | |
bea3348e SH |
5956 | return work; |
5957 | } | |
1da177e4 | 5958 | |
bea3348e SH |
5959 | /** |
5960 | * __napi_schedule - schedule for receive | |
c4ea43c5 | 5961 | * @n: entry to schedule |
bea3348e | 5962 | * |
bc9ad166 ED |
5963 | * The entry's receive function will be scheduled to run. |
5964 | * Consider using __napi_schedule_irqoff() if hard irqs are masked. | |
bea3348e | 5965 | */ |
b5606c2d | 5966 | void __napi_schedule(struct napi_struct *n) |
bea3348e SH |
5967 | { |
5968 | unsigned long flags; | |
1da177e4 | 5969 | |
bea3348e | 5970 | local_irq_save(flags); |
903ceff7 | 5971 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); |
bea3348e | 5972 | local_irq_restore(flags); |
1da177e4 | 5973 | } |
bea3348e SH |
5974 | EXPORT_SYMBOL(__napi_schedule); |
5975 | ||
39e6c820 ED |
5976 | /** |
5977 | * napi_schedule_prep - check if napi can be scheduled | |
5978 | * @n: napi context | |
5979 | * | |
5980 | * Test if NAPI routine is already running, and if not mark | |
5981 | * it as running. This is used as a condition variable | |
5982 | * insure only one NAPI poll instance runs. We also make | |
5983 | * sure there is no pending NAPI disable. | |
5984 | */ | |
5985 | bool napi_schedule_prep(struct napi_struct *n) | |
5986 | { | |
5987 | unsigned long val, new; | |
5988 | ||
5989 | do { | |
5990 | val = READ_ONCE(n->state); | |
5991 | if (unlikely(val & NAPIF_STATE_DISABLE)) | |
5992 | return false; | |
5993 | new = val | NAPIF_STATE_SCHED; | |
5994 | ||
5995 | /* Sets STATE_MISSED bit if STATE_SCHED was already set | |
5996 | * This was suggested by Alexander Duyck, as compiler | |
5997 | * emits better code than : | |
5998 | * if (val & NAPIF_STATE_SCHED) | |
5999 | * new |= NAPIF_STATE_MISSED; | |
6000 | */ | |
6001 | new |= (val & NAPIF_STATE_SCHED) / NAPIF_STATE_SCHED * | |
6002 | NAPIF_STATE_MISSED; | |
6003 | } while (cmpxchg(&n->state, val, new) != val); | |
6004 | ||
6005 | return !(val & NAPIF_STATE_SCHED); | |
6006 | } | |
6007 | EXPORT_SYMBOL(napi_schedule_prep); | |
6008 | ||
bc9ad166 ED |
6009 | /** |
6010 | * __napi_schedule_irqoff - schedule for receive | |
6011 | * @n: entry to schedule | |
6012 | * | |
6013 | * Variant of __napi_schedule() assuming hard irqs are masked | |
6014 | */ | |
6015 | void __napi_schedule_irqoff(struct napi_struct *n) | |
6016 | { | |
6017 | ____napi_schedule(this_cpu_ptr(&softnet_data), n); | |
6018 | } | |
6019 | EXPORT_SYMBOL(__napi_schedule_irqoff); | |
6020 | ||
364b6055 | 6021 | bool napi_complete_done(struct napi_struct *n, int work_done) |
d565b0a1 | 6022 | { |
39e6c820 | 6023 | unsigned long flags, val, new; |
d565b0a1 HX |
6024 | |
6025 | /* | |
217f6974 ED |
6026 | * 1) Don't let napi dequeue from the cpu poll list |
6027 | * just in case its running on a different cpu. | |
6028 | * 2) If we are busy polling, do nothing here, we have | |
6029 | * the guarantee we will be called later. | |
d565b0a1 | 6030 | */ |
217f6974 ED |
6031 | if (unlikely(n->state & (NAPIF_STATE_NPSVC | |
6032 | NAPIF_STATE_IN_BUSY_POLL))) | |
364b6055 | 6033 | return false; |
d565b0a1 | 6034 | |
d9f37d01 | 6035 | if (n->gro_bitmask) { |
3b47d303 | 6036 | unsigned long timeout = 0; |
d75b1ade | 6037 | |
3b47d303 ED |
6038 | if (work_done) |
6039 | timeout = n->dev->gro_flush_timeout; | |
6040 | ||
605108ac PA |
6041 | /* When the NAPI instance uses a timeout and keeps postponing |
6042 | * it, we need to bound somehow the time packets are kept in | |
6043 | * the GRO layer | |
6044 | */ | |
6045 | napi_gro_flush(n, !!timeout); | |
3b47d303 ED |
6046 | if (timeout) |
6047 | hrtimer_start(&n->timer, ns_to_ktime(timeout), | |
6048 | HRTIMER_MODE_REL_PINNED); | |
3b47d303 | 6049 | } |
02c1602e | 6050 | if (unlikely(!list_empty(&n->poll_list))) { |
d75b1ade ED |
6051 | /* If n->poll_list is not empty, we need to mask irqs */ |
6052 | local_irq_save(flags); | |
02c1602e | 6053 | list_del_init(&n->poll_list); |
d75b1ade ED |
6054 | local_irq_restore(flags); |
6055 | } | |
39e6c820 ED |
6056 | |
6057 | do { | |
6058 | val = READ_ONCE(n->state); | |
6059 | ||
6060 | WARN_ON_ONCE(!(val & NAPIF_STATE_SCHED)); | |
6061 | ||
6062 | new = val & ~(NAPIF_STATE_MISSED | NAPIF_STATE_SCHED); | |
6063 | ||
6064 | /* If STATE_MISSED was set, leave STATE_SCHED set, | |
6065 | * because we will call napi->poll() one more time. | |
6066 | * This C code was suggested by Alexander Duyck to help gcc. | |
6067 | */ | |
6068 | new |= (val & NAPIF_STATE_MISSED) / NAPIF_STATE_MISSED * | |
6069 | NAPIF_STATE_SCHED; | |
6070 | } while (cmpxchg(&n->state, val, new) != val); | |
6071 | ||
6072 | if (unlikely(val & NAPIF_STATE_MISSED)) { | |
6073 | __napi_schedule(n); | |
6074 | return false; | |
6075 | } | |
6076 | ||
364b6055 | 6077 | return true; |
d565b0a1 | 6078 | } |
3b47d303 | 6079 | EXPORT_SYMBOL(napi_complete_done); |
d565b0a1 | 6080 | |
af12fa6e | 6081 | /* must be called under rcu_read_lock(), as we dont take a reference */ |
02d62e86 | 6082 | static struct napi_struct *napi_by_id(unsigned int napi_id) |
af12fa6e ET |
6083 | { |
6084 | unsigned int hash = napi_id % HASH_SIZE(napi_hash); | |
6085 | struct napi_struct *napi; | |
6086 | ||
6087 | hlist_for_each_entry_rcu(napi, &napi_hash[hash], napi_hash_node) | |
6088 | if (napi->napi_id == napi_id) | |
6089 | return napi; | |
6090 | ||
6091 | return NULL; | |
6092 | } | |
02d62e86 ED |
6093 | |
6094 | #if defined(CONFIG_NET_RX_BUSY_POLL) | |
217f6974 | 6095 | |
ce6aea93 | 6096 | #define BUSY_POLL_BUDGET 8 |
217f6974 ED |
6097 | |
6098 | static void busy_poll_stop(struct napi_struct *napi, void *have_poll_lock) | |
6099 | { | |
6100 | int rc; | |
6101 | ||
39e6c820 ED |
6102 | /* Busy polling means there is a high chance device driver hard irq |
6103 | * could not grab NAPI_STATE_SCHED, and that NAPI_STATE_MISSED was | |
6104 | * set in napi_schedule_prep(). | |
6105 | * Since we are about to call napi->poll() once more, we can safely | |
6106 | * clear NAPI_STATE_MISSED. | |
6107 | * | |
6108 | * Note: x86 could use a single "lock and ..." instruction | |
6109 | * to perform these two clear_bit() | |
6110 | */ | |
6111 | clear_bit(NAPI_STATE_MISSED, &napi->state); | |
217f6974 ED |
6112 | clear_bit(NAPI_STATE_IN_BUSY_POLL, &napi->state); |
6113 | ||
6114 | local_bh_disable(); | |
6115 | ||
6116 | /* All we really want here is to re-enable device interrupts. | |
6117 | * Ideally, a new ndo_busy_poll_stop() could avoid another round. | |
6118 | */ | |
6119 | rc = napi->poll(napi, BUSY_POLL_BUDGET); | |
1e22391e | 6120 | trace_napi_poll(napi, rc, BUSY_POLL_BUDGET); |
217f6974 ED |
6121 | netpoll_poll_unlock(have_poll_lock); |
6122 | if (rc == BUSY_POLL_BUDGET) | |
6123 | __napi_schedule(napi); | |
6124 | local_bh_enable(); | |
217f6974 ED |
6125 | } |
6126 | ||
7db6b048 SS |
6127 | void napi_busy_loop(unsigned int napi_id, |
6128 | bool (*loop_end)(void *, unsigned long), | |
6129 | void *loop_end_arg) | |
02d62e86 | 6130 | { |
7db6b048 | 6131 | unsigned long start_time = loop_end ? busy_loop_current_time() : 0; |
217f6974 | 6132 | int (*napi_poll)(struct napi_struct *napi, int budget); |
217f6974 | 6133 | void *have_poll_lock = NULL; |
02d62e86 | 6134 | struct napi_struct *napi; |
217f6974 ED |
6135 | |
6136 | restart: | |
217f6974 | 6137 | napi_poll = NULL; |
02d62e86 | 6138 | |
2a028ecb | 6139 | rcu_read_lock(); |
02d62e86 | 6140 | |
545cd5e5 | 6141 | napi = napi_by_id(napi_id); |
02d62e86 ED |
6142 | if (!napi) |
6143 | goto out; | |
6144 | ||
217f6974 ED |
6145 | preempt_disable(); |
6146 | for (;;) { | |
2b5cd0df AD |
6147 | int work = 0; |
6148 | ||
2a028ecb | 6149 | local_bh_disable(); |
217f6974 ED |
6150 | if (!napi_poll) { |
6151 | unsigned long val = READ_ONCE(napi->state); | |
6152 | ||
6153 | /* If multiple threads are competing for this napi, | |
6154 | * we avoid dirtying napi->state as much as we can. | |
6155 | */ | |
6156 | if (val & (NAPIF_STATE_DISABLE | NAPIF_STATE_SCHED | | |
6157 | NAPIF_STATE_IN_BUSY_POLL)) | |
6158 | goto count; | |
6159 | if (cmpxchg(&napi->state, val, | |
6160 | val | NAPIF_STATE_IN_BUSY_POLL | | |
6161 | NAPIF_STATE_SCHED) != val) | |
6162 | goto count; | |
6163 | have_poll_lock = netpoll_poll_lock(napi); | |
6164 | napi_poll = napi->poll; | |
6165 | } | |
2b5cd0df AD |
6166 | work = napi_poll(napi, BUSY_POLL_BUDGET); |
6167 | trace_napi_poll(napi, work, BUSY_POLL_BUDGET); | |
217f6974 | 6168 | count: |
2b5cd0df | 6169 | if (work > 0) |
7db6b048 | 6170 | __NET_ADD_STATS(dev_net(napi->dev), |
2b5cd0df | 6171 | LINUX_MIB_BUSYPOLLRXPACKETS, work); |
2a028ecb | 6172 | local_bh_enable(); |
02d62e86 | 6173 | |
7db6b048 | 6174 | if (!loop_end || loop_end(loop_end_arg, start_time)) |
217f6974 | 6175 | break; |
02d62e86 | 6176 | |
217f6974 ED |
6177 | if (unlikely(need_resched())) { |
6178 | if (napi_poll) | |
6179 | busy_poll_stop(napi, have_poll_lock); | |
6180 | preempt_enable(); | |
6181 | rcu_read_unlock(); | |
6182 | cond_resched(); | |
7db6b048 | 6183 | if (loop_end(loop_end_arg, start_time)) |
2b5cd0df | 6184 | return; |
217f6974 ED |
6185 | goto restart; |
6186 | } | |
6cdf89b1 | 6187 | cpu_relax(); |
217f6974 ED |
6188 | } |
6189 | if (napi_poll) | |
6190 | busy_poll_stop(napi, have_poll_lock); | |
6191 | preempt_enable(); | |
02d62e86 | 6192 | out: |
2a028ecb | 6193 | rcu_read_unlock(); |
02d62e86 | 6194 | } |
7db6b048 | 6195 | EXPORT_SYMBOL(napi_busy_loop); |
02d62e86 ED |
6196 | |
6197 | #endif /* CONFIG_NET_RX_BUSY_POLL */ | |
af12fa6e | 6198 | |
149d6ad8 | 6199 | static void napi_hash_add(struct napi_struct *napi) |
af12fa6e | 6200 | { |
d64b5e85 ED |
6201 | if (test_bit(NAPI_STATE_NO_BUSY_POLL, &napi->state) || |
6202 | test_and_set_bit(NAPI_STATE_HASHED, &napi->state)) | |
52bd2d62 | 6203 | return; |
af12fa6e | 6204 | |
52bd2d62 | 6205 | spin_lock(&napi_hash_lock); |
af12fa6e | 6206 | |
545cd5e5 | 6207 | /* 0..NR_CPUS range is reserved for sender_cpu use */ |
52bd2d62 | 6208 | do { |
545cd5e5 AD |
6209 | if (unlikely(++napi_gen_id < MIN_NAPI_ID)) |
6210 | napi_gen_id = MIN_NAPI_ID; | |
52bd2d62 ED |
6211 | } while (napi_by_id(napi_gen_id)); |
6212 | napi->napi_id = napi_gen_id; | |
af12fa6e | 6213 | |
52bd2d62 ED |
6214 | hlist_add_head_rcu(&napi->napi_hash_node, |
6215 | &napi_hash[napi->napi_id % HASH_SIZE(napi_hash)]); | |
af12fa6e | 6216 | |
52bd2d62 | 6217 | spin_unlock(&napi_hash_lock); |
af12fa6e | 6218 | } |
af12fa6e ET |
6219 | |
6220 | /* Warning : caller is responsible to make sure rcu grace period | |
6221 | * is respected before freeing memory containing @napi | |
6222 | */ | |
34cbe27e | 6223 | bool napi_hash_del(struct napi_struct *napi) |
af12fa6e | 6224 | { |
34cbe27e ED |
6225 | bool rcu_sync_needed = false; |
6226 | ||
af12fa6e ET |
6227 | spin_lock(&napi_hash_lock); |
6228 | ||
34cbe27e ED |
6229 | if (test_and_clear_bit(NAPI_STATE_HASHED, &napi->state)) { |
6230 | rcu_sync_needed = true; | |
af12fa6e | 6231 | hlist_del_rcu(&napi->napi_hash_node); |
34cbe27e | 6232 | } |
af12fa6e | 6233 | spin_unlock(&napi_hash_lock); |
34cbe27e | 6234 | return rcu_sync_needed; |
af12fa6e ET |
6235 | } |
6236 | EXPORT_SYMBOL_GPL(napi_hash_del); | |
6237 | ||
3b47d303 ED |
6238 | static enum hrtimer_restart napi_watchdog(struct hrtimer *timer) |
6239 | { | |
6240 | struct napi_struct *napi; | |
6241 | ||
6242 | napi = container_of(timer, struct napi_struct, timer); | |
39e6c820 ED |
6243 | |
6244 | /* Note : we use a relaxed variant of napi_schedule_prep() not setting | |
6245 | * NAPI_STATE_MISSED, since we do not react to a device IRQ. | |
6246 | */ | |
d9f37d01 | 6247 | if (napi->gro_bitmask && !napi_disable_pending(napi) && |
39e6c820 ED |
6248 | !test_and_set_bit(NAPI_STATE_SCHED, &napi->state)) |
6249 | __napi_schedule_irqoff(napi); | |
3b47d303 ED |
6250 | |
6251 | return HRTIMER_NORESTART; | |
6252 | } | |
6253 | ||
7c4ec749 | 6254 | static void init_gro_hash(struct napi_struct *napi) |
d565b0a1 | 6255 | { |
07d78363 DM |
6256 | int i; |
6257 | ||
6312fe77 LR |
6258 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6259 | INIT_LIST_HEAD(&napi->gro_hash[i].list); | |
6260 | napi->gro_hash[i].count = 0; | |
6261 | } | |
7c4ec749 DM |
6262 | napi->gro_bitmask = 0; |
6263 | } | |
6264 | ||
6265 | void netif_napi_add(struct net_device *dev, struct napi_struct *napi, | |
6266 | int (*poll)(struct napi_struct *, int), int weight) | |
6267 | { | |
6268 | INIT_LIST_HEAD(&napi->poll_list); | |
6269 | hrtimer_init(&napi->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL_PINNED); | |
6270 | napi->timer.function = napi_watchdog; | |
6271 | init_gro_hash(napi); | |
5d38a079 | 6272 | napi->skb = NULL; |
d565b0a1 | 6273 | napi->poll = poll; |
82dc3c63 | 6274 | if (weight > NAPI_POLL_WEIGHT) |
bf29e9e9 QC |
6275 | netdev_err_once(dev, "%s() called with weight %d\n", __func__, |
6276 | weight); | |
d565b0a1 HX |
6277 | napi->weight = weight; |
6278 | list_add(&napi->dev_list, &dev->napi_list); | |
d565b0a1 | 6279 | napi->dev = dev; |
5d38a079 | 6280 | #ifdef CONFIG_NETPOLL |
d565b0a1 HX |
6281 | napi->poll_owner = -1; |
6282 | #endif | |
6283 | set_bit(NAPI_STATE_SCHED, &napi->state); | |
93d05d4a | 6284 | napi_hash_add(napi); |
d565b0a1 HX |
6285 | } |
6286 | EXPORT_SYMBOL(netif_napi_add); | |
6287 | ||
3b47d303 ED |
6288 | void napi_disable(struct napi_struct *n) |
6289 | { | |
6290 | might_sleep(); | |
6291 | set_bit(NAPI_STATE_DISABLE, &n->state); | |
6292 | ||
6293 | while (test_and_set_bit(NAPI_STATE_SCHED, &n->state)) | |
6294 | msleep(1); | |
2d8bff12 NH |
6295 | while (test_and_set_bit(NAPI_STATE_NPSVC, &n->state)) |
6296 | msleep(1); | |
3b47d303 ED |
6297 | |
6298 | hrtimer_cancel(&n->timer); | |
6299 | ||
6300 | clear_bit(NAPI_STATE_DISABLE, &n->state); | |
6301 | } | |
6302 | EXPORT_SYMBOL(napi_disable); | |
6303 | ||
07d78363 | 6304 | static void flush_gro_hash(struct napi_struct *napi) |
d4546c25 | 6305 | { |
07d78363 | 6306 | int i; |
d4546c25 | 6307 | |
07d78363 DM |
6308 | for (i = 0; i < GRO_HASH_BUCKETS; i++) { |
6309 | struct sk_buff *skb, *n; | |
6310 | ||
6312fe77 | 6311 | list_for_each_entry_safe(skb, n, &napi->gro_hash[i].list, list) |
07d78363 | 6312 | kfree_skb(skb); |
6312fe77 | 6313 | napi->gro_hash[i].count = 0; |
07d78363 | 6314 | } |
d4546c25 DM |
6315 | } |
6316 | ||
93d05d4a | 6317 | /* Must be called in process context */ |
d565b0a1 HX |
6318 | void netif_napi_del(struct napi_struct *napi) |
6319 | { | |
93d05d4a ED |
6320 | might_sleep(); |
6321 | if (napi_hash_del(napi)) | |
6322 | synchronize_net(); | |
d7b06636 | 6323 | list_del_init(&napi->dev_list); |
76620aaf | 6324 | napi_free_frags(napi); |
d565b0a1 | 6325 | |
07d78363 | 6326 | flush_gro_hash(napi); |
d9f37d01 | 6327 | napi->gro_bitmask = 0; |
d565b0a1 HX |
6328 | } |
6329 | EXPORT_SYMBOL(netif_napi_del); | |
6330 | ||
726ce70e HX |
6331 | static int napi_poll(struct napi_struct *n, struct list_head *repoll) |
6332 | { | |
6333 | void *have; | |
6334 | int work, weight; | |
6335 | ||
6336 | list_del_init(&n->poll_list); | |
6337 | ||
6338 | have = netpoll_poll_lock(n); | |
6339 | ||
6340 | weight = n->weight; | |
6341 | ||
6342 | /* This NAPI_STATE_SCHED test is for avoiding a race | |
6343 | * with netpoll's poll_napi(). Only the entity which | |
6344 | * obtains the lock and sees NAPI_STATE_SCHED set will | |
6345 | * actually make the ->poll() call. Therefore we avoid | |
6346 | * accidentally calling ->poll() when NAPI is not scheduled. | |
6347 | */ | |
6348 | work = 0; | |
6349 | if (test_bit(NAPI_STATE_SCHED, &n->state)) { | |
6350 | work = n->poll(n, weight); | |
1db19db7 | 6351 | trace_napi_poll(n, work, weight); |
726ce70e HX |
6352 | } |
6353 | ||
6354 | WARN_ON_ONCE(work > weight); | |
6355 | ||
6356 | if (likely(work < weight)) | |
6357 | goto out_unlock; | |
6358 | ||
6359 | /* Drivers must not modify the NAPI state if they | |
6360 | * consume the entire weight. In such cases this code | |
6361 | * still "owns" the NAPI instance and therefore can | |
6362 | * move the instance around on the list at-will. | |
6363 | */ | |
6364 | if (unlikely(napi_disable_pending(n))) { | |
6365 | napi_complete(n); | |
6366 | goto out_unlock; | |
6367 | } | |
6368 | ||
d9f37d01 | 6369 | if (n->gro_bitmask) { |
726ce70e HX |
6370 | /* flush too old packets |
6371 | * If HZ < 1000, flush all packets. | |
6372 | */ | |
6373 | napi_gro_flush(n, HZ >= 1000); | |
6374 | } | |
6375 | ||
001ce546 HX |
6376 | /* Some drivers may have called napi_schedule |
6377 | * prior to exhausting their budget. | |
6378 | */ | |
6379 | if (unlikely(!list_empty(&n->poll_list))) { | |
6380 | pr_warn_once("%s: Budget exhausted after napi rescheduled\n", | |
6381 | n->dev ? n->dev->name : "backlog"); | |
6382 | goto out_unlock; | |
6383 | } | |
6384 | ||
726ce70e HX |
6385 | list_add_tail(&n->poll_list, repoll); |
6386 | ||
6387 | out_unlock: | |
6388 | netpoll_poll_unlock(have); | |
6389 | ||
6390 | return work; | |
6391 | } | |
6392 | ||
0766f788 | 6393 | static __latent_entropy void net_rx_action(struct softirq_action *h) |
1da177e4 | 6394 | { |
903ceff7 | 6395 | struct softnet_data *sd = this_cpu_ptr(&softnet_data); |
7acf8a1e MW |
6396 | unsigned long time_limit = jiffies + |
6397 | usecs_to_jiffies(netdev_budget_usecs); | |
51b0bded | 6398 | int budget = netdev_budget; |
d75b1ade ED |
6399 | LIST_HEAD(list); |
6400 | LIST_HEAD(repoll); | |
53fb95d3 | 6401 | |
1da177e4 | 6402 | local_irq_disable(); |
d75b1ade ED |
6403 | list_splice_init(&sd->poll_list, &list); |
6404 | local_irq_enable(); | |
1da177e4 | 6405 | |
ceb8d5bf | 6406 | for (;;) { |
bea3348e | 6407 | struct napi_struct *n; |
1da177e4 | 6408 | |
ceb8d5bf HX |
6409 | if (list_empty(&list)) { |
6410 | if (!sd_has_rps_ipi_waiting(sd) && list_empty(&repoll)) | |
f52dffe0 | 6411 | goto out; |
ceb8d5bf HX |
6412 | break; |
6413 | } | |
6414 | ||
6bd373eb HX |
6415 | n = list_first_entry(&list, struct napi_struct, poll_list); |
6416 | budget -= napi_poll(n, &repoll); | |
6417 | ||
d75b1ade | 6418 | /* If softirq window is exhausted then punt. |
24f8b238 SH |
6419 | * Allow this to run for 2 jiffies since which will allow |
6420 | * an average latency of 1.5/HZ. | |
bea3348e | 6421 | */ |
ceb8d5bf HX |
6422 | if (unlikely(budget <= 0 || |
6423 | time_after_eq(jiffies, time_limit))) { | |
6424 | sd->time_squeeze++; | |
6425 | break; | |
6426 | } | |
1da177e4 | 6427 | } |
d75b1ade | 6428 | |
d75b1ade ED |
6429 | local_irq_disable(); |
6430 | ||
6431 | list_splice_tail_init(&sd->poll_list, &list); | |
6432 | list_splice_tail(&repoll, &list); | |
6433 | list_splice(&list, &sd->poll_list); | |
6434 | if (!list_empty(&sd->poll_list)) | |
6435 | __raise_softirq_irqoff(NET_RX_SOFTIRQ); | |
6436 | ||
e326bed2 | 6437 | net_rps_action_and_irq_enable(sd); |
f52dffe0 ED |
6438 | out: |
6439 | __kfree_skb_flush(); | |
1da177e4 LT |
6440 | } |
6441 | ||
aa9d8560 | 6442 | struct netdev_adjacent { |
9ff162a8 | 6443 | struct net_device *dev; |
5d261913 VF |
6444 | |
6445 | /* upper master flag, there can only be one master device per list */ | |
9ff162a8 | 6446 | bool master; |
5d261913 | 6447 | |
5d261913 VF |
6448 | /* counter for the number of times this device was added to us */ |
6449 | u16 ref_nr; | |
6450 | ||
402dae96 VF |
6451 | /* private field for the users */ |
6452 | void *private; | |
6453 | ||
9ff162a8 JP |
6454 | struct list_head list; |
6455 | struct rcu_head rcu; | |
9ff162a8 JP |
6456 | }; |
6457 | ||
6ea29da1 | 6458 | static struct netdev_adjacent *__netdev_find_adj(struct net_device *adj_dev, |
2f268f12 | 6459 | struct list_head *adj_list) |
9ff162a8 | 6460 | { |
5d261913 | 6461 | struct netdev_adjacent *adj; |
5d261913 | 6462 | |
2f268f12 | 6463 | list_for_each_entry(adj, adj_list, list) { |
5d261913 VF |
6464 | if (adj->dev == adj_dev) |
6465 | return adj; | |
9ff162a8 JP |
6466 | } |
6467 | return NULL; | |
6468 | } | |
6469 | ||
f1170fd4 DA |
6470 | static int __netdev_has_upper_dev(struct net_device *upper_dev, void *data) |
6471 | { | |
6472 | struct net_device *dev = data; | |
6473 | ||
6474 | return upper_dev == dev; | |
6475 | } | |
6476 | ||
9ff162a8 JP |
6477 | /** |
6478 | * netdev_has_upper_dev - Check if device is linked to an upper device | |
6479 | * @dev: device | |
6480 | * @upper_dev: upper device to check | |
6481 | * | |
6482 | * Find out if a device is linked to specified upper device and return true | |
6483 | * in case it is. Note that this checks only immediate upper device, | |
6484 | * not through a complete stack of devices. The caller must hold the RTNL lock. | |
6485 | */ | |
6486 | bool netdev_has_upper_dev(struct net_device *dev, | |
6487 | struct net_device *upper_dev) | |
6488 | { | |
6489 | ASSERT_RTNL(); | |
6490 | ||
f1170fd4 DA |
6491 | return netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, |
6492 | upper_dev); | |
9ff162a8 JP |
6493 | } |
6494 | EXPORT_SYMBOL(netdev_has_upper_dev); | |
6495 | ||
1a3f060c DA |
6496 | /** |
6497 | * netdev_has_upper_dev_all - Check if device is linked to an upper device | |
6498 | * @dev: device | |
6499 | * @upper_dev: upper device to check | |
6500 | * | |
6501 | * Find out if a device is linked to specified upper device and return true | |
6502 | * in case it is. Note that this checks the entire upper device chain. | |
6503 | * The caller must hold rcu lock. | |
6504 | */ | |
6505 | ||
1a3f060c DA |
6506 | bool netdev_has_upper_dev_all_rcu(struct net_device *dev, |
6507 | struct net_device *upper_dev) | |
6508 | { | |
6509 | return !!netdev_walk_all_upper_dev_rcu(dev, __netdev_has_upper_dev, | |
6510 | upper_dev); | |
6511 | } | |
6512 | EXPORT_SYMBOL(netdev_has_upper_dev_all_rcu); | |
6513 | ||
9ff162a8 JP |
6514 | /** |
6515 | * netdev_has_any_upper_dev - Check if device is linked to some device | |
6516 | * @dev: device | |
6517 | * | |
6518 | * Find out if a device is linked to an upper device and return true in case | |
6519 | * it is. The caller must hold the RTNL lock. | |
6520 | */ | |
25cc72a3 | 6521 | bool netdev_has_any_upper_dev(struct net_device *dev) |
9ff162a8 JP |
6522 | { |
6523 | ASSERT_RTNL(); | |
6524 | ||
f1170fd4 | 6525 | return !list_empty(&dev->adj_list.upper); |
9ff162a8 | 6526 | } |
25cc72a3 | 6527 | EXPORT_SYMBOL(netdev_has_any_upper_dev); |
9ff162a8 JP |
6528 | |
6529 | /** | |
6530 | * netdev_master_upper_dev_get - Get master upper device | |
6531 | * @dev: device | |
6532 | * | |
6533 | * Find a master upper device and return pointer to it or NULL in case | |
6534 | * it's not there. The caller must hold the RTNL lock. | |
6535 | */ | |
6536 | struct net_device *netdev_master_upper_dev_get(struct net_device *dev) | |
6537 | { | |
aa9d8560 | 6538 | struct netdev_adjacent *upper; |
9ff162a8 JP |
6539 | |
6540 | ASSERT_RTNL(); | |
6541 | ||
2f268f12 | 6542 | if (list_empty(&dev->adj_list.upper)) |
9ff162a8 JP |
6543 | return NULL; |
6544 | ||
2f268f12 | 6545 | upper = list_first_entry(&dev->adj_list.upper, |
aa9d8560 | 6546 | struct netdev_adjacent, list); |
9ff162a8 JP |
6547 | if (likely(upper->master)) |
6548 | return upper->dev; | |
6549 | return NULL; | |
6550 | } | |
6551 | EXPORT_SYMBOL(netdev_master_upper_dev_get); | |
6552 | ||
0f524a80 DA |
6553 | /** |
6554 | * netdev_has_any_lower_dev - Check if device is linked to some device | |
6555 | * @dev: device | |
6556 | * | |
6557 | * Find out if a device is linked to a lower device and return true in case | |
6558 | * it is. The caller must hold the RTNL lock. | |
6559 | */ | |
6560 | static bool netdev_has_any_lower_dev(struct net_device *dev) | |
6561 | { | |
6562 | ASSERT_RTNL(); | |
6563 | ||
6564 | return !list_empty(&dev->adj_list.lower); | |
6565 | } | |
6566 | ||
b6ccba4c VF |
6567 | void *netdev_adjacent_get_private(struct list_head *adj_list) |
6568 | { | |
6569 | struct netdev_adjacent *adj; | |
6570 | ||
6571 | adj = list_entry(adj_list, struct netdev_adjacent, list); | |
6572 | ||
6573 | return adj->private; | |
6574 | } | |
6575 | EXPORT_SYMBOL(netdev_adjacent_get_private); | |
6576 | ||
44a40855 VY |
6577 | /** |
6578 | * netdev_upper_get_next_dev_rcu - Get the next dev from upper list | |
6579 | * @dev: device | |
6580 | * @iter: list_head ** of the current position | |
6581 | * | |
6582 | * Gets the next device from the dev's upper list, starting from iter | |
6583 | * position. The caller must hold RCU read lock. | |
6584 | */ | |
6585 | struct net_device *netdev_upper_get_next_dev_rcu(struct net_device *dev, | |
6586 | struct list_head **iter) | |
6587 | { | |
6588 | struct netdev_adjacent *upper; | |
6589 | ||
6590 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
6591 | ||
6592 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6593 | ||
6594 | if (&upper->list == &dev->adj_list.upper) | |
6595 | return NULL; | |
6596 | ||
6597 | *iter = &upper->list; | |
6598 | ||
6599 | return upper->dev; | |
6600 | } | |
6601 | EXPORT_SYMBOL(netdev_upper_get_next_dev_rcu); | |
6602 | ||
1a3f060c DA |
6603 | static struct net_device *netdev_next_upper_dev_rcu(struct net_device *dev, |
6604 | struct list_head **iter) | |
6605 | { | |
6606 | struct netdev_adjacent *upper; | |
6607 | ||
6608 | WARN_ON_ONCE(!rcu_read_lock_held() && !lockdep_rtnl_is_held()); | |
6609 | ||
6610 | upper = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6611 | ||
6612 | if (&upper->list == &dev->adj_list.upper) | |
6613 | return NULL; | |
6614 | ||
6615 | *iter = &upper->list; | |
6616 | ||
6617 | return upper->dev; | |
6618 | } | |
6619 | ||
6620 | int netdev_walk_all_upper_dev_rcu(struct net_device *dev, | |
6621 | int (*fn)(struct net_device *dev, | |
6622 | void *data), | |
6623 | void *data) | |
6624 | { | |
6625 | struct net_device *udev; | |
6626 | struct list_head *iter; | |
6627 | int ret; | |
6628 | ||
6629 | for (iter = &dev->adj_list.upper, | |
6630 | udev = netdev_next_upper_dev_rcu(dev, &iter); | |
6631 | udev; | |
6632 | udev = netdev_next_upper_dev_rcu(dev, &iter)) { | |
6633 | /* first is the upper device itself */ | |
6634 | ret = fn(udev, data); | |
6635 | if (ret) | |
6636 | return ret; | |
6637 | ||
6638 | /* then look at all of its upper devices */ | |
6639 | ret = netdev_walk_all_upper_dev_rcu(udev, fn, data); | |
6640 | if (ret) | |
6641 | return ret; | |
6642 | } | |
6643 | ||
6644 | return 0; | |
6645 | } | |
6646 | EXPORT_SYMBOL_GPL(netdev_walk_all_upper_dev_rcu); | |
6647 | ||
31088a11 VF |
6648 | /** |
6649 | * netdev_lower_get_next_private - Get the next ->private from the | |
6650 | * lower neighbour list | |
6651 | * @dev: device | |
6652 | * @iter: list_head ** of the current position | |
6653 | * | |
6654 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
6655 | * list, starting from iter position. The caller must hold either hold the | |
6656 | * RTNL lock or its own locking that guarantees that the neighbour lower | |
b469139e | 6657 | * list will remain unchanged. |
31088a11 VF |
6658 | */ |
6659 | void *netdev_lower_get_next_private(struct net_device *dev, | |
6660 | struct list_head **iter) | |
6661 | { | |
6662 | struct netdev_adjacent *lower; | |
6663 | ||
6664 | lower = list_entry(*iter, struct netdev_adjacent, list); | |
6665 | ||
6666 | if (&lower->list == &dev->adj_list.lower) | |
6667 | return NULL; | |
6668 | ||
6859e7df | 6669 | *iter = lower->list.next; |
31088a11 VF |
6670 | |
6671 | return lower->private; | |
6672 | } | |
6673 | EXPORT_SYMBOL(netdev_lower_get_next_private); | |
6674 | ||
6675 | /** | |
6676 | * netdev_lower_get_next_private_rcu - Get the next ->private from the | |
6677 | * lower neighbour list, RCU | |
6678 | * variant | |
6679 | * @dev: device | |
6680 | * @iter: list_head ** of the current position | |
6681 | * | |
6682 | * Gets the next netdev_adjacent->private from the dev's lower neighbour | |
6683 | * list, starting from iter position. The caller must hold RCU read lock. | |
6684 | */ | |
6685 | void *netdev_lower_get_next_private_rcu(struct net_device *dev, | |
6686 | struct list_head **iter) | |
6687 | { | |
6688 | struct netdev_adjacent *lower; | |
6689 | ||
6690 | WARN_ON_ONCE(!rcu_read_lock_held()); | |
6691 | ||
6692 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6693 | ||
6694 | if (&lower->list == &dev->adj_list.lower) | |
6695 | return NULL; | |
6696 | ||
6859e7df | 6697 | *iter = &lower->list; |
31088a11 VF |
6698 | |
6699 | return lower->private; | |
6700 | } | |
6701 | EXPORT_SYMBOL(netdev_lower_get_next_private_rcu); | |
6702 | ||
4085ebe8 VY |
6703 | /** |
6704 | * netdev_lower_get_next - Get the next device from the lower neighbour | |
6705 | * list | |
6706 | * @dev: device | |
6707 | * @iter: list_head ** of the current position | |
6708 | * | |
6709 | * Gets the next netdev_adjacent from the dev's lower neighbour | |
6710 | * list, starting from iter position. The caller must hold RTNL lock or | |
6711 | * its own locking that guarantees that the neighbour lower | |
b469139e | 6712 | * list will remain unchanged. |
4085ebe8 VY |
6713 | */ |
6714 | void *netdev_lower_get_next(struct net_device *dev, struct list_head **iter) | |
6715 | { | |
6716 | struct netdev_adjacent *lower; | |
6717 | ||
cfdd28be | 6718 | lower = list_entry(*iter, struct netdev_adjacent, list); |
4085ebe8 VY |
6719 | |
6720 | if (&lower->list == &dev->adj_list.lower) | |
6721 | return NULL; | |
6722 | ||
cfdd28be | 6723 | *iter = lower->list.next; |
4085ebe8 VY |
6724 | |
6725 | return lower->dev; | |
6726 | } | |
6727 | EXPORT_SYMBOL(netdev_lower_get_next); | |
6728 | ||
1a3f060c DA |
6729 | static struct net_device *netdev_next_lower_dev(struct net_device *dev, |
6730 | struct list_head **iter) | |
6731 | { | |
6732 | struct netdev_adjacent *lower; | |
6733 | ||
46b5ab1a | 6734 | lower = list_entry((*iter)->next, struct netdev_adjacent, list); |
1a3f060c DA |
6735 | |
6736 | if (&lower->list == &dev->adj_list.lower) | |
6737 | return NULL; | |
6738 | ||
46b5ab1a | 6739 | *iter = &lower->list; |
1a3f060c DA |
6740 | |
6741 | return lower->dev; | |
6742 | } | |
6743 | ||
6744 | int netdev_walk_all_lower_dev(struct net_device *dev, | |
6745 | int (*fn)(struct net_device *dev, | |
6746 | void *data), | |
6747 | void *data) | |
6748 | { | |
6749 | struct net_device *ldev; | |
6750 | struct list_head *iter; | |
6751 | int ret; | |
6752 | ||
6753 | for (iter = &dev->adj_list.lower, | |
6754 | ldev = netdev_next_lower_dev(dev, &iter); | |
6755 | ldev; | |
6756 | ldev = netdev_next_lower_dev(dev, &iter)) { | |
6757 | /* first is the lower device itself */ | |
6758 | ret = fn(ldev, data); | |
6759 | if (ret) | |
6760 | return ret; | |
6761 | ||
6762 | /* then look at all of its lower devices */ | |
6763 | ret = netdev_walk_all_lower_dev(ldev, fn, data); | |
6764 | if (ret) | |
6765 | return ret; | |
6766 | } | |
6767 | ||
6768 | return 0; | |
6769 | } | |
6770 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev); | |
6771 | ||
1a3f060c DA |
6772 | static struct net_device *netdev_next_lower_dev_rcu(struct net_device *dev, |
6773 | struct list_head **iter) | |
6774 | { | |
6775 | struct netdev_adjacent *lower; | |
6776 | ||
6777 | lower = list_entry_rcu((*iter)->next, struct netdev_adjacent, list); | |
6778 | if (&lower->list == &dev->adj_list.lower) | |
6779 | return NULL; | |
6780 | ||
6781 | *iter = &lower->list; | |
6782 | ||
6783 | return lower->dev; | |
6784 | } | |
6785 | ||
6786 | int netdev_walk_all_lower_dev_rcu(struct net_device *dev, | |
6787 | int (*fn)(struct net_device *dev, | |
6788 | void *data), | |
6789 | void *data) | |
6790 | { | |
6791 | struct net_device *ldev; | |
6792 | struct list_head *iter; | |
6793 | int ret; | |
6794 | ||
6795 | for (iter = &dev->adj_list.lower, | |
6796 | ldev = netdev_next_lower_dev_rcu(dev, &iter); | |
6797 | ldev; | |
6798 | ldev = netdev_next_lower_dev_rcu(dev, &iter)) { | |
6799 | /* first is the lower device itself */ | |
6800 | ret = fn(ldev, data); | |
6801 | if (ret) | |
6802 | return ret; | |
6803 | ||
6804 | /* then look at all of its lower devices */ | |
6805 | ret = netdev_walk_all_lower_dev_rcu(ldev, fn, data); | |
6806 | if (ret) | |
6807 | return ret; | |
6808 | } | |
6809 | ||
6810 | return 0; | |
6811 | } | |
6812 | EXPORT_SYMBOL_GPL(netdev_walk_all_lower_dev_rcu); | |
6813 | ||
e001bfad | 6814 | /** |
6815 | * netdev_lower_get_first_private_rcu - Get the first ->private from the | |
6816 | * lower neighbour list, RCU | |
6817 | * variant | |
6818 | * @dev: device | |
6819 | * | |
6820 | * Gets the first netdev_adjacent->private from the dev's lower neighbour | |
6821 | * list. The caller must hold RCU read lock. | |
6822 | */ | |
6823 | void *netdev_lower_get_first_private_rcu(struct net_device *dev) | |
6824 | { | |
6825 | struct netdev_adjacent *lower; | |
6826 | ||
6827 | lower = list_first_or_null_rcu(&dev->adj_list.lower, | |
6828 | struct netdev_adjacent, list); | |
6829 | if (lower) | |
6830 | return lower->private; | |
6831 | return NULL; | |
6832 | } | |
6833 | EXPORT_SYMBOL(netdev_lower_get_first_private_rcu); | |
6834 | ||
9ff162a8 JP |
6835 | /** |
6836 | * netdev_master_upper_dev_get_rcu - Get master upper device | |
6837 | * @dev: device | |
6838 | * | |
6839 | * Find a master upper device and return pointer to it or NULL in case | |
6840 | * it's not there. The caller must hold the RCU read lock. | |
6841 | */ | |
6842 | struct net_device *netdev_master_upper_dev_get_rcu(struct net_device *dev) | |
6843 | { | |
aa9d8560 | 6844 | struct netdev_adjacent *upper; |
9ff162a8 | 6845 | |
2f268f12 | 6846 | upper = list_first_or_null_rcu(&dev->adj_list.upper, |
aa9d8560 | 6847 | struct netdev_adjacent, list); |
9ff162a8 JP |
6848 | if (upper && likely(upper->master)) |
6849 | return upper->dev; | |
6850 | return NULL; | |
6851 | } | |
6852 | EXPORT_SYMBOL(netdev_master_upper_dev_get_rcu); | |
6853 | ||
0a59f3a9 | 6854 | static int netdev_adjacent_sysfs_add(struct net_device *dev, |
3ee32707 VF |
6855 | struct net_device *adj_dev, |
6856 | struct list_head *dev_list) | |
6857 | { | |
6858 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6859 | |
3ee32707 VF |
6860 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6861 | "upper_%s" : "lower_%s", adj_dev->name); | |
6862 | return sysfs_create_link(&(dev->dev.kobj), &(adj_dev->dev.kobj), | |
6863 | linkname); | |
6864 | } | |
0a59f3a9 | 6865 | static void netdev_adjacent_sysfs_del(struct net_device *dev, |
3ee32707 VF |
6866 | char *name, |
6867 | struct list_head *dev_list) | |
6868 | { | |
6869 | char linkname[IFNAMSIZ+7]; | |
f4563a75 | 6870 | |
3ee32707 VF |
6871 | sprintf(linkname, dev_list == &dev->adj_list.upper ? |
6872 | "upper_%s" : "lower_%s", name); | |
6873 | sysfs_remove_link(&(dev->dev.kobj), linkname); | |
6874 | } | |
6875 | ||
7ce64c79 AF |
6876 | static inline bool netdev_adjacent_is_neigh_list(struct net_device *dev, |
6877 | struct net_device *adj_dev, | |
6878 | struct list_head *dev_list) | |
6879 | { | |
6880 | return (dev_list == &dev->adj_list.upper || | |
6881 | dev_list == &dev->adj_list.lower) && | |
6882 | net_eq(dev_net(dev), dev_net(adj_dev)); | |
6883 | } | |
3ee32707 | 6884 | |
5d261913 VF |
6885 | static int __netdev_adjacent_dev_insert(struct net_device *dev, |
6886 | struct net_device *adj_dev, | |
7863c054 | 6887 | struct list_head *dev_list, |
402dae96 | 6888 | void *private, bool master) |
5d261913 VF |
6889 | { |
6890 | struct netdev_adjacent *adj; | |
842d67a7 | 6891 | int ret; |
5d261913 | 6892 | |
6ea29da1 | 6893 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 VF |
6894 | |
6895 | if (adj) { | |
790510d9 | 6896 | adj->ref_nr += 1; |
67b62f98 DA |
6897 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d\n", |
6898 | dev->name, adj_dev->name, adj->ref_nr); | |
6899 | ||
5d261913 VF |
6900 | return 0; |
6901 | } | |
6902 | ||
6903 | adj = kmalloc(sizeof(*adj), GFP_KERNEL); | |
6904 | if (!adj) | |
6905 | return -ENOMEM; | |
6906 | ||
6907 | adj->dev = adj_dev; | |
6908 | adj->master = master; | |
790510d9 | 6909 | adj->ref_nr = 1; |
402dae96 | 6910 | adj->private = private; |
5d261913 | 6911 | dev_hold(adj_dev); |
2f268f12 | 6912 | |
67b62f98 DA |
6913 | pr_debug("Insert adjacency: dev %s adj_dev %s adj->ref_nr %d; dev_hold on %s\n", |
6914 | dev->name, adj_dev->name, adj->ref_nr, adj_dev->name); | |
5d261913 | 6915 | |
7ce64c79 | 6916 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) { |
3ee32707 | 6917 | ret = netdev_adjacent_sysfs_add(dev, adj_dev, dev_list); |
5831d66e VF |
6918 | if (ret) |
6919 | goto free_adj; | |
6920 | } | |
6921 | ||
7863c054 | 6922 | /* Ensure that master link is always the first item in list. */ |
842d67a7 VF |
6923 | if (master) { |
6924 | ret = sysfs_create_link(&(dev->dev.kobj), | |
6925 | &(adj_dev->dev.kobj), "master"); | |
6926 | if (ret) | |
5831d66e | 6927 | goto remove_symlinks; |
842d67a7 | 6928 | |
7863c054 | 6929 | list_add_rcu(&adj->list, dev_list); |
842d67a7 | 6930 | } else { |
7863c054 | 6931 | list_add_tail_rcu(&adj->list, dev_list); |
842d67a7 | 6932 | } |
5d261913 VF |
6933 | |
6934 | return 0; | |
842d67a7 | 6935 | |
5831d66e | 6936 | remove_symlinks: |
7ce64c79 | 6937 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6938 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
842d67a7 VF |
6939 | free_adj: |
6940 | kfree(adj); | |
974daef7 | 6941 | dev_put(adj_dev); |
842d67a7 VF |
6942 | |
6943 | return ret; | |
5d261913 VF |
6944 | } |
6945 | ||
1d143d9f | 6946 | static void __netdev_adjacent_dev_remove(struct net_device *dev, |
6947 | struct net_device *adj_dev, | |
93409033 | 6948 | u16 ref_nr, |
1d143d9f | 6949 | struct list_head *dev_list) |
5d261913 VF |
6950 | { |
6951 | struct netdev_adjacent *adj; | |
6952 | ||
67b62f98 DA |
6953 | pr_debug("Remove adjacency: dev %s adj_dev %s ref_nr %d\n", |
6954 | dev->name, adj_dev->name, ref_nr); | |
6955 | ||
6ea29da1 | 6956 | adj = __netdev_find_adj(adj_dev, dev_list); |
5d261913 | 6957 | |
2f268f12 | 6958 | if (!adj) { |
67b62f98 | 6959 | pr_err("Adjacency does not exist for device %s from %s\n", |
2f268f12 | 6960 | dev->name, adj_dev->name); |
67b62f98 DA |
6961 | WARN_ON(1); |
6962 | return; | |
2f268f12 | 6963 | } |
5d261913 | 6964 | |
93409033 | 6965 | if (adj->ref_nr > ref_nr) { |
67b62f98 DA |
6966 | pr_debug("adjacency: %s to %s ref_nr - %d = %d\n", |
6967 | dev->name, adj_dev->name, ref_nr, | |
6968 | adj->ref_nr - ref_nr); | |
93409033 | 6969 | adj->ref_nr -= ref_nr; |
5d261913 VF |
6970 | return; |
6971 | } | |
6972 | ||
842d67a7 VF |
6973 | if (adj->master) |
6974 | sysfs_remove_link(&(dev->dev.kobj), "master"); | |
6975 | ||
7ce64c79 | 6976 | if (netdev_adjacent_is_neigh_list(dev, adj_dev, dev_list)) |
3ee32707 | 6977 | netdev_adjacent_sysfs_del(dev, adj_dev->name, dev_list); |
5831d66e | 6978 | |
5d261913 | 6979 | list_del_rcu(&adj->list); |
67b62f98 | 6980 | pr_debug("adjacency: dev_put for %s, because link removed from %s to %s\n", |
2f268f12 | 6981 | adj_dev->name, dev->name, adj_dev->name); |
5d261913 VF |
6982 | dev_put(adj_dev); |
6983 | kfree_rcu(adj, rcu); | |
6984 | } | |
6985 | ||
1d143d9f | 6986 | static int __netdev_adjacent_dev_link_lists(struct net_device *dev, |
6987 | struct net_device *upper_dev, | |
6988 | struct list_head *up_list, | |
6989 | struct list_head *down_list, | |
6990 | void *private, bool master) | |
5d261913 VF |
6991 | { |
6992 | int ret; | |
6993 | ||
790510d9 | 6994 | ret = __netdev_adjacent_dev_insert(dev, upper_dev, up_list, |
93409033 | 6995 | private, master); |
5d261913 VF |
6996 | if (ret) |
6997 | return ret; | |
6998 | ||
790510d9 | 6999 | ret = __netdev_adjacent_dev_insert(upper_dev, dev, down_list, |
93409033 | 7000 | private, false); |
5d261913 | 7001 | if (ret) { |
790510d9 | 7002 | __netdev_adjacent_dev_remove(dev, upper_dev, 1, up_list); |
5d261913 VF |
7003 | return ret; |
7004 | } | |
7005 | ||
7006 | return 0; | |
7007 | } | |
7008 | ||
1d143d9f | 7009 | static void __netdev_adjacent_dev_unlink_lists(struct net_device *dev, |
7010 | struct net_device *upper_dev, | |
93409033 | 7011 | u16 ref_nr, |
1d143d9f | 7012 | struct list_head *up_list, |
7013 | struct list_head *down_list) | |
5d261913 | 7014 | { |
93409033 AC |
7015 | __netdev_adjacent_dev_remove(dev, upper_dev, ref_nr, up_list); |
7016 | __netdev_adjacent_dev_remove(upper_dev, dev, ref_nr, down_list); | |
5d261913 VF |
7017 | } |
7018 | ||
1d143d9f | 7019 | static int __netdev_adjacent_dev_link_neighbour(struct net_device *dev, |
7020 | struct net_device *upper_dev, | |
7021 | void *private, bool master) | |
2f268f12 | 7022 | { |
f1170fd4 DA |
7023 | return __netdev_adjacent_dev_link_lists(dev, upper_dev, |
7024 | &dev->adj_list.upper, | |
7025 | &upper_dev->adj_list.lower, | |
7026 | private, master); | |
5d261913 VF |
7027 | } |
7028 | ||
1d143d9f | 7029 | static void __netdev_adjacent_dev_unlink_neighbour(struct net_device *dev, |
7030 | struct net_device *upper_dev) | |
2f268f12 | 7031 | { |
93409033 | 7032 | __netdev_adjacent_dev_unlink_lists(dev, upper_dev, 1, |
2f268f12 VF |
7033 | &dev->adj_list.upper, |
7034 | &upper_dev->adj_list.lower); | |
7035 | } | |
5d261913 | 7036 | |
9ff162a8 | 7037 | static int __netdev_upper_dev_link(struct net_device *dev, |
402dae96 | 7038 | struct net_device *upper_dev, bool master, |
42ab19ee DA |
7039 | void *upper_priv, void *upper_info, |
7040 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7041 | { |
51d0c047 DA |
7042 | struct netdev_notifier_changeupper_info changeupper_info = { |
7043 | .info = { | |
7044 | .dev = dev, | |
42ab19ee | 7045 | .extack = extack, |
51d0c047 DA |
7046 | }, |
7047 | .upper_dev = upper_dev, | |
7048 | .master = master, | |
7049 | .linking = true, | |
7050 | .upper_info = upper_info, | |
7051 | }; | |
50d629e7 | 7052 | struct net_device *master_dev; |
5d261913 | 7053 | int ret = 0; |
9ff162a8 JP |
7054 | |
7055 | ASSERT_RTNL(); | |
7056 | ||
7057 | if (dev == upper_dev) | |
7058 | return -EBUSY; | |
7059 | ||
7060 | /* To prevent loops, check if dev is not upper device to upper_dev. */ | |
f1170fd4 | 7061 | if (netdev_has_upper_dev(upper_dev, dev)) |
9ff162a8 JP |
7062 | return -EBUSY; |
7063 | ||
50d629e7 MM |
7064 | if (!master) { |
7065 | if (netdev_has_upper_dev(dev, upper_dev)) | |
7066 | return -EEXIST; | |
7067 | } else { | |
7068 | master_dev = netdev_master_upper_dev_get(dev); | |
7069 | if (master_dev) | |
7070 | return master_dev == upper_dev ? -EEXIST : -EBUSY; | |
7071 | } | |
9ff162a8 | 7072 | |
51d0c047 | 7073 | ret = call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7074 | &changeupper_info.info); |
7075 | ret = notifier_to_errno(ret); | |
7076 | if (ret) | |
7077 | return ret; | |
7078 | ||
6dffb044 | 7079 | ret = __netdev_adjacent_dev_link_neighbour(dev, upper_dev, upper_priv, |
402dae96 | 7080 | master); |
5d261913 VF |
7081 | if (ret) |
7082 | return ret; | |
9ff162a8 | 7083 | |
51d0c047 | 7084 | ret = call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
b03804e7 IS |
7085 | &changeupper_info.info); |
7086 | ret = notifier_to_errno(ret); | |
7087 | if (ret) | |
f1170fd4 | 7088 | goto rollback; |
b03804e7 | 7089 | |
9ff162a8 | 7090 | return 0; |
5d261913 | 7091 | |
f1170fd4 | 7092 | rollback: |
2f268f12 | 7093 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 VF |
7094 | |
7095 | return ret; | |
9ff162a8 JP |
7096 | } |
7097 | ||
7098 | /** | |
7099 | * netdev_upper_dev_link - Add a link to the upper device | |
7100 | * @dev: device | |
7101 | * @upper_dev: new upper device | |
7a006d59 | 7102 | * @extack: netlink extended ack |
9ff162a8 JP |
7103 | * |
7104 | * Adds a link to device which is upper to this one. The caller must hold | |
7105 | * the RTNL lock. On a failure a negative errno code is returned. | |
7106 | * On success the reference counts are adjusted and the function | |
7107 | * returns zero. | |
7108 | */ | |
7109 | int netdev_upper_dev_link(struct net_device *dev, | |
42ab19ee DA |
7110 | struct net_device *upper_dev, |
7111 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7112 | { |
42ab19ee DA |
7113 | return __netdev_upper_dev_link(dev, upper_dev, false, |
7114 | NULL, NULL, extack); | |
9ff162a8 JP |
7115 | } |
7116 | EXPORT_SYMBOL(netdev_upper_dev_link); | |
7117 | ||
7118 | /** | |
7119 | * netdev_master_upper_dev_link - Add a master link to the upper device | |
7120 | * @dev: device | |
7121 | * @upper_dev: new upper device | |
6dffb044 | 7122 | * @upper_priv: upper device private |
29bf24af | 7123 | * @upper_info: upper info to be passed down via notifier |
7a006d59 | 7124 | * @extack: netlink extended ack |
9ff162a8 JP |
7125 | * |
7126 | * Adds a link to device which is upper to this one. In this case, only | |
7127 | * one master upper device can be linked, although other non-master devices | |
7128 | * might be linked as well. The caller must hold the RTNL lock. | |
7129 | * On a failure a negative errno code is returned. On success the reference | |
7130 | * counts are adjusted and the function returns zero. | |
7131 | */ | |
7132 | int netdev_master_upper_dev_link(struct net_device *dev, | |
6dffb044 | 7133 | struct net_device *upper_dev, |
42ab19ee DA |
7134 | void *upper_priv, void *upper_info, |
7135 | struct netlink_ext_ack *extack) | |
9ff162a8 | 7136 | { |
29bf24af | 7137 | return __netdev_upper_dev_link(dev, upper_dev, true, |
42ab19ee | 7138 | upper_priv, upper_info, extack); |
9ff162a8 JP |
7139 | } |
7140 | EXPORT_SYMBOL(netdev_master_upper_dev_link); | |
7141 | ||
7142 | /** | |
7143 | * netdev_upper_dev_unlink - Removes a link to upper device | |
7144 | * @dev: device | |
7145 | * @upper_dev: new upper device | |
7146 | * | |
7147 | * Removes a link to device which is upper to this one. The caller must hold | |
7148 | * the RTNL lock. | |
7149 | */ | |
7150 | void netdev_upper_dev_unlink(struct net_device *dev, | |
7151 | struct net_device *upper_dev) | |
7152 | { | |
51d0c047 DA |
7153 | struct netdev_notifier_changeupper_info changeupper_info = { |
7154 | .info = { | |
7155 | .dev = dev, | |
7156 | }, | |
7157 | .upper_dev = upper_dev, | |
7158 | .linking = false, | |
7159 | }; | |
f4563a75 | 7160 | |
9ff162a8 JP |
7161 | ASSERT_RTNL(); |
7162 | ||
0e4ead9d | 7163 | changeupper_info.master = netdev_master_upper_dev_get(dev) == upper_dev; |
0e4ead9d | 7164 | |
51d0c047 | 7165 | call_netdevice_notifiers_info(NETDEV_PRECHANGEUPPER, |
573c7ba0 JP |
7166 | &changeupper_info.info); |
7167 | ||
2f268f12 | 7168 | __netdev_adjacent_dev_unlink_neighbour(dev, upper_dev); |
5d261913 | 7169 | |
51d0c047 | 7170 | call_netdevice_notifiers_info(NETDEV_CHANGEUPPER, |
0e4ead9d | 7171 | &changeupper_info.info); |
9ff162a8 JP |
7172 | } |
7173 | EXPORT_SYMBOL(netdev_upper_dev_unlink); | |
7174 | ||
61bd3857 MS |
7175 | /** |
7176 | * netdev_bonding_info_change - Dispatch event about slave change | |
7177 | * @dev: device | |
4a26e453 | 7178 | * @bonding_info: info to dispatch |
61bd3857 MS |
7179 | * |
7180 | * Send NETDEV_BONDING_INFO to netdev notifiers with info. | |
7181 | * The caller must hold the RTNL lock. | |
7182 | */ | |
7183 | void netdev_bonding_info_change(struct net_device *dev, | |
7184 | struct netdev_bonding_info *bonding_info) | |
7185 | { | |
51d0c047 DA |
7186 | struct netdev_notifier_bonding_info info = { |
7187 | .info.dev = dev, | |
7188 | }; | |
61bd3857 MS |
7189 | |
7190 | memcpy(&info.bonding_info, bonding_info, | |
7191 | sizeof(struct netdev_bonding_info)); | |
51d0c047 | 7192 | call_netdevice_notifiers_info(NETDEV_BONDING_INFO, |
61bd3857 MS |
7193 | &info.info); |
7194 | } | |
7195 | EXPORT_SYMBOL(netdev_bonding_info_change); | |
7196 | ||
2ce1ee17 | 7197 | static void netdev_adjacent_add_links(struct net_device *dev) |
4c75431a AF |
7198 | { |
7199 | struct netdev_adjacent *iter; | |
7200 | ||
7201 | struct net *net = dev_net(dev); | |
7202 | ||
7203 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 7204 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7205 | continue; |
7206 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7207 | &iter->dev->adj_list.lower); | |
7208 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
7209 | &dev->adj_list.upper); | |
7210 | } | |
7211 | ||
7212 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 7213 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7214 | continue; |
7215 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7216 | &iter->dev->adj_list.upper); | |
7217 | netdev_adjacent_sysfs_add(dev, iter->dev, | |
7218 | &dev->adj_list.lower); | |
7219 | } | |
7220 | } | |
7221 | ||
2ce1ee17 | 7222 | static void netdev_adjacent_del_links(struct net_device *dev) |
4c75431a AF |
7223 | { |
7224 | struct netdev_adjacent *iter; | |
7225 | ||
7226 | struct net *net = dev_net(dev); | |
7227 | ||
7228 | list_for_each_entry(iter, &dev->adj_list.upper, list) { | |
be4da0e3 | 7229 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7230 | continue; |
7231 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
7232 | &iter->dev->adj_list.lower); | |
7233 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
7234 | &dev->adj_list.upper); | |
7235 | } | |
7236 | ||
7237 | list_for_each_entry(iter, &dev->adj_list.lower, list) { | |
be4da0e3 | 7238 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a AF |
7239 | continue; |
7240 | netdev_adjacent_sysfs_del(iter->dev, dev->name, | |
7241 | &iter->dev->adj_list.upper); | |
7242 | netdev_adjacent_sysfs_del(dev, iter->dev->name, | |
7243 | &dev->adj_list.lower); | |
7244 | } | |
7245 | } | |
7246 | ||
5bb025fa | 7247 | void netdev_adjacent_rename_links(struct net_device *dev, char *oldname) |
402dae96 | 7248 | { |
5bb025fa | 7249 | struct netdev_adjacent *iter; |
402dae96 | 7250 | |
4c75431a AF |
7251 | struct net *net = dev_net(dev); |
7252 | ||
5bb025fa | 7253 | list_for_each_entry(iter, &dev->adj_list.upper, list) { |
be4da0e3 | 7254 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 7255 | continue; |
5bb025fa VF |
7256 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
7257 | &iter->dev->adj_list.lower); | |
7258 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7259 | &iter->dev->adj_list.lower); | |
7260 | } | |
402dae96 | 7261 | |
5bb025fa | 7262 | list_for_each_entry(iter, &dev->adj_list.lower, list) { |
be4da0e3 | 7263 | if (!net_eq(net, dev_net(iter->dev))) |
4c75431a | 7264 | continue; |
5bb025fa VF |
7265 | netdev_adjacent_sysfs_del(iter->dev, oldname, |
7266 | &iter->dev->adj_list.upper); | |
7267 | netdev_adjacent_sysfs_add(iter->dev, dev, | |
7268 | &iter->dev->adj_list.upper); | |
7269 | } | |
402dae96 | 7270 | } |
402dae96 VF |
7271 | |
7272 | void *netdev_lower_dev_get_private(struct net_device *dev, | |
7273 | struct net_device *lower_dev) | |
7274 | { | |
7275 | struct netdev_adjacent *lower; | |
7276 | ||
7277 | if (!lower_dev) | |
7278 | return NULL; | |
6ea29da1 | 7279 | lower = __netdev_find_adj(lower_dev, &dev->adj_list.lower); |
402dae96 VF |
7280 | if (!lower) |
7281 | return NULL; | |
7282 | ||
7283 | return lower->private; | |
7284 | } | |
7285 | EXPORT_SYMBOL(netdev_lower_dev_get_private); | |
7286 | ||
4085ebe8 | 7287 | |
952fcfd0 | 7288 | int dev_get_nest_level(struct net_device *dev) |
4085ebe8 VY |
7289 | { |
7290 | struct net_device *lower = NULL; | |
7291 | struct list_head *iter; | |
7292 | int max_nest = -1; | |
7293 | int nest; | |
7294 | ||
7295 | ASSERT_RTNL(); | |
7296 | ||
7297 | netdev_for_each_lower_dev(dev, lower, iter) { | |
952fcfd0 | 7298 | nest = dev_get_nest_level(lower); |
4085ebe8 VY |
7299 | if (max_nest < nest) |
7300 | max_nest = nest; | |
7301 | } | |
7302 | ||
952fcfd0 | 7303 | return max_nest + 1; |
4085ebe8 VY |
7304 | } |
7305 | EXPORT_SYMBOL(dev_get_nest_level); | |
7306 | ||
04d48266 JP |
7307 | /** |
7308 | * netdev_lower_change - Dispatch event about lower device state change | |
7309 | * @lower_dev: device | |
7310 | * @lower_state_info: state to dispatch | |
7311 | * | |
7312 | * Send NETDEV_CHANGELOWERSTATE to netdev notifiers with info. | |
7313 | * The caller must hold the RTNL lock. | |
7314 | */ | |
7315 | void netdev_lower_state_changed(struct net_device *lower_dev, | |
7316 | void *lower_state_info) | |
7317 | { | |
51d0c047 DA |
7318 | struct netdev_notifier_changelowerstate_info changelowerstate_info = { |
7319 | .info.dev = lower_dev, | |
7320 | }; | |
04d48266 JP |
7321 | |
7322 | ASSERT_RTNL(); | |
7323 | changelowerstate_info.lower_state_info = lower_state_info; | |
51d0c047 | 7324 | call_netdevice_notifiers_info(NETDEV_CHANGELOWERSTATE, |
04d48266 JP |
7325 | &changelowerstate_info.info); |
7326 | } | |
7327 | EXPORT_SYMBOL(netdev_lower_state_changed); | |
7328 | ||
b6c40d68 PM |
7329 | static void dev_change_rx_flags(struct net_device *dev, int flags) |
7330 | { | |
d314774c SH |
7331 | const struct net_device_ops *ops = dev->netdev_ops; |
7332 | ||
d2615bf4 | 7333 | if (ops->ndo_change_rx_flags) |
d314774c | 7334 | ops->ndo_change_rx_flags(dev, flags); |
b6c40d68 PM |
7335 | } |
7336 | ||
991fb3f7 | 7337 | static int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify) |
1da177e4 | 7338 | { |
b536db93 | 7339 | unsigned int old_flags = dev->flags; |
d04a48b0 EB |
7340 | kuid_t uid; |
7341 | kgid_t gid; | |
1da177e4 | 7342 | |
24023451 PM |
7343 | ASSERT_RTNL(); |
7344 | ||
dad9b335 WC |
7345 | dev->flags |= IFF_PROMISC; |
7346 | dev->promiscuity += inc; | |
7347 | if (dev->promiscuity == 0) { | |
7348 | /* | |
7349 | * Avoid overflow. | |
7350 | * If inc causes overflow, untouch promisc and return error. | |
7351 | */ | |
7352 | if (inc < 0) | |
7353 | dev->flags &= ~IFF_PROMISC; | |
7354 | else { | |
7355 | dev->promiscuity -= inc; | |
7b6cd1ce JP |
7356 | pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n", |
7357 | dev->name); | |
dad9b335 WC |
7358 | return -EOVERFLOW; |
7359 | } | |
7360 | } | |
52609c0b | 7361 | if (dev->flags != old_flags) { |
7b6cd1ce JP |
7362 | pr_info("device %s %s promiscuous mode\n", |
7363 | dev->name, | |
7364 | dev->flags & IFF_PROMISC ? "entered" : "left"); | |
8192b0c4 DH |
7365 | if (audit_enabled) { |
7366 | current_uid_gid(&uid, &gid); | |
cdfb6b34 RGB |
7367 | audit_log(audit_context(), GFP_ATOMIC, |
7368 | AUDIT_ANOM_PROMISCUOUS, | |
7369 | "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u", | |
7370 | dev->name, (dev->flags & IFF_PROMISC), | |
7371 | (old_flags & IFF_PROMISC), | |
7372 | from_kuid(&init_user_ns, audit_get_loginuid(current)), | |
7373 | from_kuid(&init_user_ns, uid), | |
7374 | from_kgid(&init_user_ns, gid), | |
7375 | audit_get_sessionid(current)); | |
8192b0c4 | 7376 | } |
24023451 | 7377 | |
b6c40d68 | 7378 | dev_change_rx_flags(dev, IFF_PROMISC); |
1da177e4 | 7379 | } |
991fb3f7 ND |
7380 | if (notify) |
7381 | __dev_notify_flags(dev, old_flags, IFF_PROMISC); | |
dad9b335 | 7382 | return 0; |
1da177e4 LT |
7383 | } |
7384 | ||
4417da66 PM |
7385 | /** |
7386 | * dev_set_promiscuity - update promiscuity count on a device | |
7387 | * @dev: device | |
7388 | * @inc: modifier | |
7389 | * | |
7390 | * Add or remove promiscuity from a device. While the count in the device | |
7391 | * remains above zero the interface remains promiscuous. Once it hits zero | |
7392 | * the device reverts back to normal filtering operation. A negative inc | |
7393 | * value is used to drop promiscuity on the device. | |
dad9b335 | 7394 | * Return 0 if successful or a negative errno code on error. |
4417da66 | 7395 | */ |
dad9b335 | 7396 | int dev_set_promiscuity(struct net_device *dev, int inc) |
4417da66 | 7397 | { |
b536db93 | 7398 | unsigned int old_flags = dev->flags; |
dad9b335 | 7399 | int err; |
4417da66 | 7400 | |
991fb3f7 | 7401 | err = __dev_set_promiscuity(dev, inc, true); |
4b5a698e | 7402 | if (err < 0) |
dad9b335 | 7403 | return err; |
4417da66 PM |
7404 | if (dev->flags != old_flags) |
7405 | dev_set_rx_mode(dev); | |
dad9b335 | 7406 | return err; |
4417da66 | 7407 | } |
d1b19dff | 7408 | EXPORT_SYMBOL(dev_set_promiscuity); |
4417da66 | 7409 | |
991fb3f7 | 7410 | static int __dev_set_allmulti(struct net_device *dev, int inc, bool notify) |
1da177e4 | 7411 | { |
991fb3f7 | 7412 | unsigned int old_flags = dev->flags, old_gflags = dev->gflags; |
1da177e4 | 7413 | |
24023451 PM |
7414 | ASSERT_RTNL(); |
7415 | ||
1da177e4 | 7416 | dev->flags |= IFF_ALLMULTI; |
dad9b335 WC |
7417 | dev->allmulti += inc; |
7418 | if (dev->allmulti == 0) { | |
7419 | /* | |
7420 | * Avoid overflow. | |
7421 | * If inc causes overflow, untouch allmulti and return error. | |
7422 | */ | |
7423 | if (inc < 0) | |
7424 | dev->flags &= ~IFF_ALLMULTI; | |
7425 | else { | |
7426 | dev->allmulti -= inc; | |
7b6cd1ce JP |
7427 | pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n", |
7428 | dev->name); | |
dad9b335 WC |
7429 | return -EOVERFLOW; |
7430 | } | |
7431 | } | |
24023451 | 7432 | if (dev->flags ^ old_flags) { |
b6c40d68 | 7433 | dev_change_rx_flags(dev, IFF_ALLMULTI); |
4417da66 | 7434 | dev_set_rx_mode(dev); |
991fb3f7 ND |
7435 | if (notify) |
7436 | __dev_notify_flags(dev, old_flags, | |
7437 | dev->gflags ^ old_gflags); | |
24023451 | 7438 | } |
dad9b335 | 7439 | return 0; |
4417da66 | 7440 | } |
991fb3f7 ND |
7441 | |
7442 | /** | |
7443 | * dev_set_allmulti - update allmulti count on a device | |
7444 | * @dev: device | |
7445 | * @inc: modifier | |
7446 | * | |
7447 | * Add or remove reception of all multicast frames to a device. While the | |
7448 | * count in the device remains above zero the interface remains listening | |
7449 | * to all interfaces. Once it hits zero the device reverts back to normal | |
7450 | * filtering operation. A negative @inc value is used to drop the counter | |
7451 | * when releasing a resource needing all multicasts. | |
7452 | * Return 0 if successful or a negative errno code on error. | |
7453 | */ | |
7454 | ||
7455 | int dev_set_allmulti(struct net_device *dev, int inc) | |
7456 | { | |
7457 | return __dev_set_allmulti(dev, inc, true); | |
7458 | } | |
d1b19dff | 7459 | EXPORT_SYMBOL(dev_set_allmulti); |
4417da66 PM |
7460 | |
7461 | /* | |
7462 | * Upload unicast and multicast address lists to device and | |
7463 | * configure RX filtering. When the device doesn't support unicast | |
53ccaae1 | 7464 | * filtering it is put in promiscuous mode while unicast addresses |
4417da66 PM |
7465 | * are present. |
7466 | */ | |
7467 | void __dev_set_rx_mode(struct net_device *dev) | |
7468 | { | |
d314774c SH |
7469 | const struct net_device_ops *ops = dev->netdev_ops; |
7470 | ||
4417da66 PM |
7471 | /* dev_open will call this function so the list will stay sane. */ |
7472 | if (!(dev->flags&IFF_UP)) | |
7473 | return; | |
7474 | ||
7475 | if (!netif_device_present(dev)) | |
40b77c94 | 7476 | return; |
4417da66 | 7477 | |
01789349 | 7478 | if (!(dev->priv_flags & IFF_UNICAST_FLT)) { |
4417da66 PM |
7479 | /* Unicast addresses changes may only happen under the rtnl, |
7480 | * therefore calling __dev_set_promiscuity here is safe. | |
7481 | */ | |
32e7bfc4 | 7482 | if (!netdev_uc_empty(dev) && !dev->uc_promisc) { |
991fb3f7 | 7483 | __dev_set_promiscuity(dev, 1, false); |
2d348d1f | 7484 | dev->uc_promisc = true; |
32e7bfc4 | 7485 | } else if (netdev_uc_empty(dev) && dev->uc_promisc) { |
991fb3f7 | 7486 | __dev_set_promiscuity(dev, -1, false); |
2d348d1f | 7487 | dev->uc_promisc = false; |
4417da66 | 7488 | } |
4417da66 | 7489 | } |
01789349 JP |
7490 | |
7491 | if (ops->ndo_set_rx_mode) | |
7492 | ops->ndo_set_rx_mode(dev); | |
4417da66 PM |
7493 | } |
7494 | ||
7495 | void dev_set_rx_mode(struct net_device *dev) | |
7496 | { | |
b9e40857 | 7497 | netif_addr_lock_bh(dev); |
4417da66 | 7498 | __dev_set_rx_mode(dev); |
b9e40857 | 7499 | netif_addr_unlock_bh(dev); |
1da177e4 LT |
7500 | } |
7501 | ||
f0db275a SH |
7502 | /** |
7503 | * dev_get_flags - get flags reported to userspace | |
7504 | * @dev: device | |
7505 | * | |
7506 | * Get the combination of flag bits exported through APIs to userspace. | |
7507 | */ | |
95c96174 | 7508 | unsigned int dev_get_flags(const struct net_device *dev) |
1da177e4 | 7509 | { |
95c96174 | 7510 | unsigned int flags; |
1da177e4 LT |
7511 | |
7512 | flags = (dev->flags & ~(IFF_PROMISC | | |
7513 | IFF_ALLMULTI | | |
b00055aa SR |
7514 | IFF_RUNNING | |
7515 | IFF_LOWER_UP | | |
7516 | IFF_DORMANT)) | | |
1da177e4 LT |
7517 | (dev->gflags & (IFF_PROMISC | |
7518 | IFF_ALLMULTI)); | |
7519 | ||
b00055aa SR |
7520 | if (netif_running(dev)) { |
7521 | if (netif_oper_up(dev)) | |
7522 | flags |= IFF_RUNNING; | |
7523 | if (netif_carrier_ok(dev)) | |
7524 | flags |= IFF_LOWER_UP; | |
7525 | if (netif_dormant(dev)) | |
7526 | flags |= IFF_DORMANT; | |
7527 | } | |
1da177e4 LT |
7528 | |
7529 | return flags; | |
7530 | } | |
d1b19dff | 7531 | EXPORT_SYMBOL(dev_get_flags); |
1da177e4 | 7532 | |
6d040321 PM |
7533 | int __dev_change_flags(struct net_device *dev, unsigned int flags, |
7534 | struct netlink_ext_ack *extack) | |
1da177e4 | 7535 | { |
b536db93 | 7536 | unsigned int old_flags = dev->flags; |
bd380811 | 7537 | int ret; |
1da177e4 | 7538 | |
24023451 PM |
7539 | ASSERT_RTNL(); |
7540 | ||
1da177e4 LT |
7541 | /* |
7542 | * Set the flags on our device. | |
7543 | */ | |
7544 | ||
7545 | dev->flags = (flags & (IFF_DEBUG | IFF_NOTRAILERS | IFF_NOARP | | |
7546 | IFF_DYNAMIC | IFF_MULTICAST | IFF_PORTSEL | | |
7547 | IFF_AUTOMEDIA)) | | |
7548 | (dev->flags & (IFF_UP | IFF_VOLATILE | IFF_PROMISC | | |
7549 | IFF_ALLMULTI)); | |
7550 | ||
7551 | /* | |
7552 | * Load in the correct multicast list now the flags have changed. | |
7553 | */ | |
7554 | ||
b6c40d68 PM |
7555 | if ((old_flags ^ flags) & IFF_MULTICAST) |
7556 | dev_change_rx_flags(dev, IFF_MULTICAST); | |
24023451 | 7557 | |
4417da66 | 7558 | dev_set_rx_mode(dev); |
1da177e4 LT |
7559 | |
7560 | /* | |
7561 | * Have we downed the interface. We handle IFF_UP ourselves | |
7562 | * according to user attempts to set it, rather than blindly | |
7563 | * setting it. | |
7564 | */ | |
7565 | ||
7566 | ret = 0; | |
7051b88a | 7567 | if ((old_flags ^ flags) & IFF_UP) { |
7568 | if (old_flags & IFF_UP) | |
7569 | __dev_close(dev); | |
7570 | else | |
40c900aa | 7571 | ret = __dev_open(dev, extack); |
7051b88a | 7572 | } |
1da177e4 | 7573 | |
1da177e4 | 7574 | if ((flags ^ dev->gflags) & IFF_PROMISC) { |
d1b19dff | 7575 | int inc = (flags & IFF_PROMISC) ? 1 : -1; |
991fb3f7 | 7576 | unsigned int old_flags = dev->flags; |
d1b19dff | 7577 | |
1da177e4 | 7578 | dev->gflags ^= IFF_PROMISC; |
991fb3f7 ND |
7579 | |
7580 | if (__dev_set_promiscuity(dev, inc, false) >= 0) | |
7581 | if (dev->flags != old_flags) | |
7582 | dev_set_rx_mode(dev); | |
1da177e4 LT |
7583 | } |
7584 | ||
7585 | /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI | |
eb13da1a | 7586 | * is important. Some (broken) drivers set IFF_PROMISC, when |
7587 | * IFF_ALLMULTI is requested not asking us and not reporting. | |
1da177e4 LT |
7588 | */ |
7589 | if ((flags ^ dev->gflags) & IFF_ALLMULTI) { | |
d1b19dff ED |
7590 | int inc = (flags & IFF_ALLMULTI) ? 1 : -1; |
7591 | ||
1da177e4 | 7592 | dev->gflags ^= IFF_ALLMULTI; |
991fb3f7 | 7593 | __dev_set_allmulti(dev, inc, false); |
1da177e4 LT |
7594 | } |
7595 | ||
bd380811 PM |
7596 | return ret; |
7597 | } | |
7598 | ||
a528c219 ND |
7599 | void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, |
7600 | unsigned int gchanges) | |
bd380811 PM |
7601 | { |
7602 | unsigned int changes = dev->flags ^ old_flags; | |
7603 | ||
a528c219 | 7604 | if (gchanges) |
7f294054 | 7605 | rtmsg_ifinfo(RTM_NEWLINK, dev, gchanges, GFP_ATOMIC); |
a528c219 | 7606 | |
bd380811 PM |
7607 | if (changes & IFF_UP) { |
7608 | if (dev->flags & IFF_UP) | |
7609 | call_netdevice_notifiers(NETDEV_UP, dev); | |
7610 | else | |
7611 | call_netdevice_notifiers(NETDEV_DOWN, dev); | |
7612 | } | |
7613 | ||
7614 | if (dev->flags & IFF_UP && | |
be9efd36 | 7615 | (changes & ~(IFF_UP | IFF_PROMISC | IFF_ALLMULTI | IFF_VOLATILE))) { |
51d0c047 DA |
7616 | struct netdev_notifier_change_info change_info = { |
7617 | .info = { | |
7618 | .dev = dev, | |
7619 | }, | |
7620 | .flags_changed = changes, | |
7621 | }; | |
be9efd36 | 7622 | |
51d0c047 | 7623 | call_netdevice_notifiers_info(NETDEV_CHANGE, &change_info.info); |
be9efd36 | 7624 | } |
bd380811 PM |
7625 | } |
7626 | ||
7627 | /** | |
7628 | * dev_change_flags - change device settings | |
7629 | * @dev: device | |
7630 | * @flags: device state flags | |
567c5e13 | 7631 | * @extack: netlink extended ack |
bd380811 PM |
7632 | * |
7633 | * Change settings on device based state flags. The flags are | |
7634 | * in the userspace exported format. | |
7635 | */ | |
567c5e13 PM |
7636 | int dev_change_flags(struct net_device *dev, unsigned int flags, |
7637 | struct netlink_ext_ack *extack) | |
bd380811 | 7638 | { |
b536db93 | 7639 | int ret; |
991fb3f7 | 7640 | unsigned int changes, old_flags = dev->flags, old_gflags = dev->gflags; |
bd380811 | 7641 | |
6d040321 | 7642 | ret = __dev_change_flags(dev, flags, extack); |
bd380811 PM |
7643 | if (ret < 0) |
7644 | return ret; | |
7645 | ||
991fb3f7 | 7646 | changes = (old_flags ^ dev->flags) | (old_gflags ^ dev->gflags); |
a528c219 | 7647 | __dev_notify_flags(dev, old_flags, changes); |
1da177e4 LT |
7648 | return ret; |
7649 | } | |
d1b19dff | 7650 | EXPORT_SYMBOL(dev_change_flags); |
1da177e4 | 7651 | |
f51048c3 | 7652 | int __dev_set_mtu(struct net_device *dev, int new_mtu) |
2315dc91 VF |
7653 | { |
7654 | const struct net_device_ops *ops = dev->netdev_ops; | |
7655 | ||
7656 | if (ops->ndo_change_mtu) | |
7657 | return ops->ndo_change_mtu(dev, new_mtu); | |
7658 | ||
7659 | dev->mtu = new_mtu; | |
7660 | return 0; | |
7661 | } | |
f51048c3 | 7662 | EXPORT_SYMBOL(__dev_set_mtu); |
2315dc91 | 7663 | |
f0db275a | 7664 | /** |
7a4c53be | 7665 | * dev_set_mtu_ext - Change maximum transfer unit |
f0db275a SH |
7666 | * @dev: device |
7667 | * @new_mtu: new transfer unit | |
7a4c53be | 7668 | * @extack: netlink extended ack |
f0db275a SH |
7669 | * |
7670 | * Change the maximum transfer size of the network device. | |
7671 | */ | |
7a4c53be SH |
7672 | int dev_set_mtu_ext(struct net_device *dev, int new_mtu, |
7673 | struct netlink_ext_ack *extack) | |
1da177e4 | 7674 | { |
2315dc91 | 7675 | int err, orig_mtu; |
1da177e4 LT |
7676 | |
7677 | if (new_mtu == dev->mtu) | |
7678 | return 0; | |
7679 | ||
61e84623 JW |
7680 | /* MTU must be positive, and in range */ |
7681 | if (new_mtu < 0 || new_mtu < dev->min_mtu) { | |
7a4c53be | 7682 | NL_SET_ERR_MSG(extack, "mtu less than device minimum"); |
1da177e4 | 7683 | return -EINVAL; |
61e84623 JW |
7684 | } |
7685 | ||
7686 | if (dev->max_mtu > 0 && new_mtu > dev->max_mtu) { | |
7a4c53be | 7687 | NL_SET_ERR_MSG(extack, "mtu greater than device maximum"); |
61e84623 JW |
7688 | return -EINVAL; |
7689 | } | |
1da177e4 LT |
7690 | |
7691 | if (!netif_device_present(dev)) | |
7692 | return -ENODEV; | |
7693 | ||
1d486bfb VF |
7694 | err = call_netdevice_notifiers(NETDEV_PRECHANGEMTU, dev); |
7695 | err = notifier_to_errno(err); | |
7696 | if (err) | |
7697 | return err; | |
d314774c | 7698 | |
2315dc91 VF |
7699 | orig_mtu = dev->mtu; |
7700 | err = __dev_set_mtu(dev, new_mtu); | |
d314774c | 7701 | |
2315dc91 | 7702 | if (!err) { |
af7d6cce SD |
7703 | err = call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
7704 | orig_mtu); | |
2315dc91 VF |
7705 | err = notifier_to_errno(err); |
7706 | if (err) { | |
7707 | /* setting mtu back and notifying everyone again, | |
7708 | * so that they have a chance to revert changes. | |
7709 | */ | |
7710 | __dev_set_mtu(dev, orig_mtu); | |
af7d6cce SD |
7711 | call_netdevice_notifiers_mtu(NETDEV_CHANGEMTU, dev, |
7712 | new_mtu); | |
2315dc91 VF |
7713 | } |
7714 | } | |
1da177e4 LT |
7715 | return err; |
7716 | } | |
7a4c53be SH |
7717 | |
7718 | int dev_set_mtu(struct net_device *dev, int new_mtu) | |
7719 | { | |
7720 | struct netlink_ext_ack extack; | |
7721 | int err; | |
7722 | ||
a6bcfc89 | 7723 | memset(&extack, 0, sizeof(extack)); |
7a4c53be | 7724 | err = dev_set_mtu_ext(dev, new_mtu, &extack); |
a6bcfc89 | 7725 | if (err && extack._msg) |
7a4c53be SH |
7726 | net_err_ratelimited("%s: %s\n", dev->name, extack._msg); |
7727 | return err; | |
7728 | } | |
d1b19dff | 7729 | EXPORT_SYMBOL(dev_set_mtu); |
1da177e4 | 7730 | |
6a643ddb CW |
7731 | /** |
7732 | * dev_change_tx_queue_len - Change TX queue length of a netdevice | |
7733 | * @dev: device | |
7734 | * @new_len: new tx queue length | |
7735 | */ | |
7736 | int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len) | |
7737 | { | |
7738 | unsigned int orig_len = dev->tx_queue_len; | |
7739 | int res; | |
7740 | ||
7741 | if (new_len != (unsigned int)new_len) | |
7742 | return -ERANGE; | |
7743 | ||
7744 | if (new_len != orig_len) { | |
7745 | dev->tx_queue_len = new_len; | |
7746 | res = call_netdevice_notifiers(NETDEV_CHANGE_TX_QUEUE_LEN, dev); | |
7747 | res = notifier_to_errno(res); | |
7effaf06 TT |
7748 | if (res) |
7749 | goto err_rollback; | |
7750 | res = dev_qdisc_change_tx_queue_len(dev); | |
7751 | if (res) | |
7752 | goto err_rollback; | |
6a643ddb CW |
7753 | } |
7754 | ||
7755 | return 0; | |
7effaf06 TT |
7756 | |
7757 | err_rollback: | |
7758 | netdev_err(dev, "refused to change device tx_queue_len\n"); | |
7759 | dev->tx_queue_len = orig_len; | |
7760 | return res; | |
6a643ddb CW |
7761 | } |
7762 | ||
cbda10fa VD |
7763 | /** |
7764 | * dev_set_group - Change group this device belongs to | |
7765 | * @dev: device | |
7766 | * @new_group: group this device should belong to | |
7767 | */ | |
7768 | void dev_set_group(struct net_device *dev, int new_group) | |
7769 | { | |
7770 | dev->group = new_group; | |
7771 | } | |
7772 | EXPORT_SYMBOL(dev_set_group); | |
7773 | ||
d59cdf94 PM |
7774 | /** |
7775 | * dev_pre_changeaddr_notify - Call NETDEV_PRE_CHANGEADDR. | |
7776 | * @dev: device | |
7777 | * @addr: new address | |
7778 | * @extack: netlink extended ack | |
7779 | */ | |
7780 | int dev_pre_changeaddr_notify(struct net_device *dev, const char *addr, | |
7781 | struct netlink_ext_ack *extack) | |
7782 | { | |
7783 | struct netdev_notifier_pre_changeaddr_info info = { | |
7784 | .info.dev = dev, | |
7785 | .info.extack = extack, | |
7786 | .dev_addr = addr, | |
7787 | }; | |
7788 | int rc; | |
7789 | ||
7790 | rc = call_netdevice_notifiers_info(NETDEV_PRE_CHANGEADDR, &info.info); | |
7791 | return notifier_to_errno(rc); | |
7792 | } | |
7793 | EXPORT_SYMBOL(dev_pre_changeaddr_notify); | |
7794 | ||
f0db275a SH |
7795 | /** |
7796 | * dev_set_mac_address - Change Media Access Control Address | |
7797 | * @dev: device | |
7798 | * @sa: new address | |
3a37a963 | 7799 | * @extack: netlink extended ack |
f0db275a SH |
7800 | * |
7801 | * Change the hardware (MAC) address of the device | |
7802 | */ | |
3a37a963 PM |
7803 | int dev_set_mac_address(struct net_device *dev, struct sockaddr *sa, |
7804 | struct netlink_ext_ack *extack) | |
1da177e4 | 7805 | { |
d314774c | 7806 | const struct net_device_ops *ops = dev->netdev_ops; |
1da177e4 LT |
7807 | int err; |
7808 | ||
d314774c | 7809 | if (!ops->ndo_set_mac_address) |
1da177e4 LT |
7810 | return -EOPNOTSUPP; |
7811 | if (sa->sa_family != dev->type) | |
7812 | return -EINVAL; | |
7813 | if (!netif_device_present(dev)) | |
7814 | return -ENODEV; | |
d59cdf94 PM |
7815 | err = dev_pre_changeaddr_notify(dev, sa->sa_data, extack); |
7816 | if (err) | |
7817 | return err; | |
d314774c | 7818 | err = ops->ndo_set_mac_address(dev, sa); |
f6521516 JP |
7819 | if (err) |
7820 | return err; | |
fbdeca2d | 7821 | dev->addr_assign_type = NET_ADDR_SET; |
f6521516 | 7822 | call_netdevice_notifiers(NETDEV_CHANGEADDR, dev); |
7bf23575 | 7823 | add_device_randomness(dev->dev_addr, dev->addr_len); |
f6521516 | 7824 | return 0; |
1da177e4 | 7825 | } |
d1b19dff | 7826 | EXPORT_SYMBOL(dev_set_mac_address); |
1da177e4 | 7827 | |
4bf84c35 JP |
7828 | /** |
7829 | * dev_change_carrier - Change device carrier | |
7830 | * @dev: device | |
691b3b7e | 7831 | * @new_carrier: new value |
4bf84c35 JP |
7832 | * |
7833 | * Change device carrier | |
7834 | */ | |
7835 | int dev_change_carrier(struct net_device *dev, bool new_carrier) | |
7836 | { | |
7837 | const struct net_device_ops *ops = dev->netdev_ops; | |
7838 | ||
7839 | if (!ops->ndo_change_carrier) | |
7840 | return -EOPNOTSUPP; | |
7841 | if (!netif_device_present(dev)) | |
7842 | return -ENODEV; | |
7843 | return ops->ndo_change_carrier(dev, new_carrier); | |
7844 | } | |
7845 | EXPORT_SYMBOL(dev_change_carrier); | |
7846 | ||
66b52b0d JP |
7847 | /** |
7848 | * dev_get_phys_port_id - Get device physical port ID | |
7849 | * @dev: device | |
7850 | * @ppid: port ID | |
7851 | * | |
7852 | * Get device physical port ID | |
7853 | */ | |
7854 | int dev_get_phys_port_id(struct net_device *dev, | |
02637fce | 7855 | struct netdev_phys_item_id *ppid) |
66b52b0d JP |
7856 | { |
7857 | const struct net_device_ops *ops = dev->netdev_ops; | |
7858 | ||
7859 | if (!ops->ndo_get_phys_port_id) | |
7860 | return -EOPNOTSUPP; | |
7861 | return ops->ndo_get_phys_port_id(dev, ppid); | |
7862 | } | |
7863 | EXPORT_SYMBOL(dev_get_phys_port_id); | |
7864 | ||
db24a904 DA |
7865 | /** |
7866 | * dev_get_phys_port_name - Get device physical port name | |
7867 | * @dev: device | |
7868 | * @name: port name | |
ed49e650 | 7869 | * @len: limit of bytes to copy to name |
db24a904 DA |
7870 | * |
7871 | * Get device physical port name | |
7872 | */ | |
7873 | int dev_get_phys_port_name(struct net_device *dev, | |
7874 | char *name, size_t len) | |
7875 | { | |
7876 | const struct net_device_ops *ops = dev->netdev_ops; | |
af3836df | 7877 | int err; |
db24a904 | 7878 | |
af3836df JP |
7879 | if (ops->ndo_get_phys_port_name) { |
7880 | err = ops->ndo_get_phys_port_name(dev, name, len); | |
7881 | if (err != -EOPNOTSUPP) | |
7882 | return err; | |
7883 | } | |
7884 | return devlink_compat_phys_port_name_get(dev, name, len); | |
db24a904 DA |
7885 | } |
7886 | EXPORT_SYMBOL(dev_get_phys_port_name); | |
7887 | ||
d6abc596 FF |
7888 | /** |
7889 | * dev_get_port_parent_id - Get the device's port parent identifier | |
7890 | * @dev: network device | |
7891 | * @ppid: pointer to a storage for the port's parent identifier | |
7892 | * @recurse: allow/disallow recursion to lower devices | |
7893 | * | |
7894 | * Get the devices's port parent identifier | |
7895 | */ | |
7896 | int dev_get_port_parent_id(struct net_device *dev, | |
7897 | struct netdev_phys_item_id *ppid, | |
7898 | bool recurse) | |
7899 | { | |
7900 | const struct net_device_ops *ops = dev->netdev_ops; | |
7901 | struct netdev_phys_item_id first = { }; | |
7902 | struct net_device *lower_dev; | |
7903 | struct list_head *iter; | |
7904 | int err = -EOPNOTSUPP; | |
7905 | ||
7906 | if (ops->ndo_get_port_parent_id) | |
7907 | return ops->ndo_get_port_parent_id(dev, ppid); | |
7908 | ||
7909 | if (!recurse) | |
7910 | return err; | |
7911 | ||
7912 | netdev_for_each_lower_dev(dev, lower_dev, iter) { | |
7913 | err = dev_get_port_parent_id(lower_dev, ppid, recurse); | |
7914 | if (err) | |
7915 | break; | |
7916 | if (!first.id_len) | |
7917 | first = *ppid; | |
7918 | else if (memcmp(&first, ppid, sizeof(*ppid))) | |
7919 | return -ENODATA; | |
7920 | } | |
7921 | ||
7922 | return err; | |
7923 | } | |
7924 | EXPORT_SYMBOL(dev_get_port_parent_id); | |
7925 | ||
7926 | /** | |
7927 | * netdev_port_same_parent_id - Indicate if two network devices have | |
7928 | * the same port parent identifier | |
7929 | * @a: first network device | |
7930 | * @b: second network device | |
7931 | */ | |
7932 | bool netdev_port_same_parent_id(struct net_device *a, struct net_device *b) | |
7933 | { | |
7934 | struct netdev_phys_item_id a_id = { }; | |
7935 | struct netdev_phys_item_id b_id = { }; | |
7936 | ||
7937 | if (dev_get_port_parent_id(a, &a_id, true) || | |
7938 | dev_get_port_parent_id(b, &b_id, true)) | |
7939 | return false; | |
7940 | ||
7941 | return netdev_phys_item_id_same(&a_id, &b_id); | |
7942 | } | |
7943 | EXPORT_SYMBOL(netdev_port_same_parent_id); | |
7944 | ||
d746d707 AK |
7945 | /** |
7946 | * dev_change_proto_down - update protocol port state information | |
7947 | * @dev: device | |
7948 | * @proto_down: new value | |
7949 | * | |
7950 | * This info can be used by switch drivers to set the phys state of the | |
7951 | * port. | |
7952 | */ | |
7953 | int dev_change_proto_down(struct net_device *dev, bool proto_down) | |
7954 | { | |
7955 | const struct net_device_ops *ops = dev->netdev_ops; | |
7956 | ||
7957 | if (!ops->ndo_change_proto_down) | |
7958 | return -EOPNOTSUPP; | |
7959 | if (!netif_device_present(dev)) | |
7960 | return -ENODEV; | |
7961 | return ops->ndo_change_proto_down(dev, proto_down); | |
7962 | } | |
7963 | EXPORT_SYMBOL(dev_change_proto_down); | |
7964 | ||
b5899679 AR |
7965 | /** |
7966 | * dev_change_proto_down_generic - generic implementation for | |
7967 | * ndo_change_proto_down that sets carrier according to | |
7968 | * proto_down. | |
7969 | * | |
7970 | * @dev: device | |
7971 | * @proto_down: new value | |
7972 | */ | |
7973 | int dev_change_proto_down_generic(struct net_device *dev, bool proto_down) | |
7974 | { | |
7975 | if (proto_down) | |
7976 | netif_carrier_off(dev); | |
7977 | else | |
7978 | netif_carrier_on(dev); | |
7979 | dev->proto_down = proto_down; | |
7980 | return 0; | |
7981 | } | |
7982 | EXPORT_SYMBOL(dev_change_proto_down_generic); | |
7983 | ||
a25717d2 JK |
7984 | u32 __dev_xdp_query(struct net_device *dev, bpf_op_t bpf_op, |
7985 | enum bpf_netdev_command cmd) | |
d67b9cd2 | 7986 | { |
a25717d2 | 7987 | struct netdev_bpf xdp; |
d67b9cd2 | 7988 | |
a25717d2 JK |
7989 | if (!bpf_op) |
7990 | return 0; | |
118b4aa2 | 7991 | |
a25717d2 JK |
7992 | memset(&xdp, 0, sizeof(xdp)); |
7993 | xdp.command = cmd; | |
118b4aa2 | 7994 | |
a25717d2 JK |
7995 | /* Query must always succeed. */ |
7996 | WARN_ON(bpf_op(dev, &xdp) < 0 && cmd == XDP_QUERY_PROG); | |
58038695 | 7997 | |
6b867589 | 7998 | return xdp.prog_id; |
d67b9cd2 DB |
7999 | } |
8000 | ||
f4e63525 | 8001 | static int dev_xdp_install(struct net_device *dev, bpf_op_t bpf_op, |
32d60277 | 8002 | struct netlink_ext_ack *extack, u32 flags, |
d67b9cd2 DB |
8003 | struct bpf_prog *prog) |
8004 | { | |
f4e63525 | 8005 | struct netdev_bpf xdp; |
d67b9cd2 DB |
8006 | |
8007 | memset(&xdp, 0, sizeof(xdp)); | |
ee5d032f JK |
8008 | if (flags & XDP_FLAGS_HW_MODE) |
8009 | xdp.command = XDP_SETUP_PROG_HW; | |
8010 | else | |
8011 | xdp.command = XDP_SETUP_PROG; | |
d67b9cd2 | 8012 | xdp.extack = extack; |
32d60277 | 8013 | xdp.flags = flags; |
d67b9cd2 DB |
8014 | xdp.prog = prog; |
8015 | ||
f4e63525 | 8016 | return bpf_op(dev, &xdp); |
d67b9cd2 DB |
8017 | } |
8018 | ||
bd0b2e7f JK |
8019 | static void dev_xdp_uninstall(struct net_device *dev) |
8020 | { | |
8021 | struct netdev_bpf xdp; | |
8022 | bpf_op_t ndo_bpf; | |
8023 | ||
8024 | /* Remove generic XDP */ | |
8025 | WARN_ON(dev_xdp_install(dev, generic_xdp_install, NULL, 0, NULL)); | |
8026 | ||
8027 | /* Remove from the driver */ | |
8028 | ndo_bpf = dev->netdev_ops->ndo_bpf; | |
8029 | if (!ndo_bpf) | |
8030 | return; | |
8031 | ||
a25717d2 JK |
8032 | memset(&xdp, 0, sizeof(xdp)); |
8033 | xdp.command = XDP_QUERY_PROG; | |
8034 | WARN_ON(ndo_bpf(dev, &xdp)); | |
8035 | if (xdp.prog_id) | |
8036 | WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, | |
8037 | NULL)); | |
bd0b2e7f | 8038 | |
a25717d2 JK |
8039 | /* Remove HW offload */ |
8040 | memset(&xdp, 0, sizeof(xdp)); | |
8041 | xdp.command = XDP_QUERY_PROG_HW; | |
8042 | if (!ndo_bpf(dev, &xdp) && xdp.prog_id) | |
8043 | WARN_ON(dev_xdp_install(dev, ndo_bpf, NULL, xdp.prog_flags, | |
8044 | NULL)); | |
bd0b2e7f JK |
8045 | } |
8046 | ||
a7862b45 BB |
8047 | /** |
8048 | * dev_change_xdp_fd - set or clear a bpf program for a device rx path | |
8049 | * @dev: device | |
b5d60989 | 8050 | * @extack: netlink extended ack |
a7862b45 | 8051 | * @fd: new program fd or negative value to clear |
85de8576 | 8052 | * @flags: xdp-related flags |
a7862b45 BB |
8053 | * |
8054 | * Set or clear a bpf program for a device | |
8055 | */ | |
ddf9f970 JK |
8056 | int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, |
8057 | int fd, u32 flags) | |
a7862b45 BB |
8058 | { |
8059 | const struct net_device_ops *ops = dev->netdev_ops; | |
a25717d2 | 8060 | enum bpf_netdev_command query; |
a7862b45 | 8061 | struct bpf_prog *prog = NULL; |
f4e63525 | 8062 | bpf_op_t bpf_op, bpf_chk; |
9ee963d6 | 8063 | bool offload; |
a7862b45 BB |
8064 | int err; |
8065 | ||
85de8576 DB |
8066 | ASSERT_RTNL(); |
8067 | ||
9ee963d6 JK |
8068 | offload = flags & XDP_FLAGS_HW_MODE; |
8069 | query = offload ? XDP_QUERY_PROG_HW : XDP_QUERY_PROG; | |
a25717d2 | 8070 | |
f4e63525 | 8071 | bpf_op = bpf_chk = ops->ndo_bpf; |
01dde20c MF |
8072 | if (!bpf_op && (flags & (XDP_FLAGS_DRV_MODE | XDP_FLAGS_HW_MODE))) { |
8073 | NL_SET_ERR_MSG(extack, "underlying driver does not support XDP in native mode"); | |
0489df9a | 8074 | return -EOPNOTSUPP; |
01dde20c | 8075 | } |
f4e63525 JK |
8076 | if (!bpf_op || (flags & XDP_FLAGS_SKB_MODE)) |
8077 | bpf_op = generic_xdp_install; | |
8078 | if (bpf_op == bpf_chk) | |
8079 | bpf_chk = generic_xdp_install; | |
b5cdae32 | 8080 | |
a7862b45 | 8081 | if (fd >= 0) { |
9ee963d6 | 8082 | if (!offload && __dev_xdp_query(dev, bpf_chk, XDP_QUERY_PROG)) { |
01dde20c | 8083 | NL_SET_ERR_MSG(extack, "native and generic XDP can't be active at the same time"); |
d67b9cd2 | 8084 | return -EEXIST; |
01dde20c | 8085 | } |
d67b9cd2 | 8086 | if ((flags & XDP_FLAGS_UPDATE_IF_NOEXIST) && |
01dde20c MF |
8087 | __dev_xdp_query(dev, bpf_op, query)) { |
8088 | NL_SET_ERR_MSG(extack, "XDP program already attached"); | |
d67b9cd2 | 8089 | return -EBUSY; |
01dde20c | 8090 | } |
85de8576 | 8091 | |
288b3de5 JK |
8092 | prog = bpf_prog_get_type_dev(fd, BPF_PROG_TYPE_XDP, |
8093 | bpf_op == ops->ndo_bpf); | |
a7862b45 BB |
8094 | if (IS_ERR(prog)) |
8095 | return PTR_ERR(prog); | |
441a3303 | 8096 | |
9ee963d6 | 8097 | if (!offload && bpf_prog_is_dev_bound(prog->aux)) { |
441a3303 JK |
8098 | NL_SET_ERR_MSG(extack, "using device-bound program without HW_MODE flag is not supported"); |
8099 | bpf_prog_put(prog); | |
8100 | return -EINVAL; | |
8101 | } | |
a7862b45 BB |
8102 | } |
8103 | ||
f4e63525 | 8104 | err = dev_xdp_install(dev, bpf_op, extack, flags, prog); |
a7862b45 BB |
8105 | if (err < 0 && prog) |
8106 | bpf_prog_put(prog); | |
8107 | ||
8108 | return err; | |
8109 | } | |
a7862b45 | 8110 | |
1da177e4 LT |
8111 | /** |
8112 | * dev_new_index - allocate an ifindex | |
c4ea43c5 | 8113 | * @net: the applicable net namespace |
1da177e4 LT |
8114 | * |
8115 | * Returns a suitable unique value for a new device interface | |
8116 | * number. The caller must hold the rtnl semaphore or the | |
8117 | * dev_base_lock to be sure it remains unique. | |
8118 | */ | |
881d966b | 8119 | static int dev_new_index(struct net *net) |
1da177e4 | 8120 | { |
aa79e66e | 8121 | int ifindex = net->ifindex; |
f4563a75 | 8122 | |
1da177e4 LT |
8123 | for (;;) { |
8124 | if (++ifindex <= 0) | |
8125 | ifindex = 1; | |
881d966b | 8126 | if (!__dev_get_by_index(net, ifindex)) |
aa79e66e | 8127 | return net->ifindex = ifindex; |
1da177e4 LT |
8128 | } |
8129 | } | |
8130 | ||
1da177e4 | 8131 | /* Delayed registration/unregisteration */ |
3b5b34fd | 8132 | static LIST_HEAD(net_todo_list); |
200b916f | 8133 | DECLARE_WAIT_QUEUE_HEAD(netdev_unregistering_wq); |
1da177e4 | 8134 | |
6f05f629 | 8135 | static void net_set_todo(struct net_device *dev) |
1da177e4 | 8136 | { |
1da177e4 | 8137 | list_add_tail(&dev->todo_list, &net_todo_list); |
50624c93 | 8138 | dev_net(dev)->dev_unreg_count++; |
1da177e4 LT |
8139 | } |
8140 | ||
9b5e383c | 8141 | static void rollback_registered_many(struct list_head *head) |
93ee31f1 | 8142 | { |
e93737b0 | 8143 | struct net_device *dev, *tmp; |
5cde2829 | 8144 | LIST_HEAD(close_head); |
9b5e383c | 8145 | |
93ee31f1 DL |
8146 | BUG_ON(dev_boot_phase); |
8147 | ASSERT_RTNL(); | |
8148 | ||
e93737b0 | 8149 | list_for_each_entry_safe(dev, tmp, head, unreg_list) { |
9b5e383c | 8150 | /* Some devices call without registering |
e93737b0 KK |
8151 | * for initialization unwind. Remove those |
8152 | * devices and proceed with the remaining. | |
9b5e383c ED |
8153 | */ |
8154 | if (dev->reg_state == NETREG_UNINITIALIZED) { | |
7b6cd1ce JP |
8155 | pr_debug("unregister_netdevice: device %s/%p never was registered\n", |
8156 | dev->name, dev); | |
93ee31f1 | 8157 | |
9b5e383c | 8158 | WARN_ON(1); |
e93737b0 KK |
8159 | list_del(&dev->unreg_list); |
8160 | continue; | |
9b5e383c | 8161 | } |
449f4544 | 8162 | dev->dismantle = true; |
9b5e383c | 8163 | BUG_ON(dev->reg_state != NETREG_REGISTERED); |
44345724 | 8164 | } |
93ee31f1 | 8165 | |
44345724 | 8166 | /* If device is running, close it first. */ |
5cde2829 EB |
8167 | list_for_each_entry(dev, head, unreg_list) |
8168 | list_add_tail(&dev->close_list, &close_head); | |
99c4a26a | 8169 | dev_close_many(&close_head, true); |
93ee31f1 | 8170 | |
44345724 | 8171 | list_for_each_entry(dev, head, unreg_list) { |
9b5e383c ED |
8172 | /* And unlink it from device chain. */ |
8173 | unlist_netdevice(dev); | |
93ee31f1 | 8174 | |
9b5e383c ED |
8175 | dev->reg_state = NETREG_UNREGISTERING; |
8176 | } | |
41852497 | 8177 | flush_all_backlogs(); |
93ee31f1 DL |
8178 | |
8179 | synchronize_net(); | |
8180 | ||
9b5e383c | 8181 | list_for_each_entry(dev, head, unreg_list) { |
395eea6c MB |
8182 | struct sk_buff *skb = NULL; |
8183 | ||
9b5e383c ED |
8184 | /* Shutdown queueing discipline. */ |
8185 | dev_shutdown(dev); | |
93ee31f1 | 8186 | |
bd0b2e7f | 8187 | dev_xdp_uninstall(dev); |
93ee31f1 | 8188 | |
9b5e383c | 8189 | /* Notify protocols, that we are about to destroy |
eb13da1a | 8190 | * this device. They should clean all the things. |
8191 | */ | |
9b5e383c | 8192 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
93ee31f1 | 8193 | |
395eea6c MB |
8194 | if (!dev->rtnl_link_ops || |
8195 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
3d3ea5af | 8196 | skb = rtmsg_ifinfo_build_skb(RTM_DELLINK, dev, ~0U, 0, |
38e01b30 | 8197 | GFP_KERNEL, NULL, 0); |
395eea6c | 8198 | |
9b5e383c ED |
8199 | /* |
8200 | * Flush the unicast and multicast chains | |
8201 | */ | |
a748ee24 | 8202 | dev_uc_flush(dev); |
22bedad3 | 8203 | dev_mc_flush(dev); |
93ee31f1 | 8204 | |
9b5e383c ED |
8205 | if (dev->netdev_ops->ndo_uninit) |
8206 | dev->netdev_ops->ndo_uninit(dev); | |
93ee31f1 | 8207 | |
395eea6c MB |
8208 | if (skb) |
8209 | rtmsg_ifinfo_send(skb, dev, GFP_KERNEL); | |
56bfa7ee | 8210 | |
9ff162a8 JP |
8211 | /* Notifier chain MUST detach us all upper devices. */ |
8212 | WARN_ON(netdev_has_any_upper_dev(dev)); | |
0f524a80 | 8213 | WARN_ON(netdev_has_any_lower_dev(dev)); |
93ee31f1 | 8214 | |
9b5e383c ED |
8215 | /* Remove entries from kobject tree */ |
8216 | netdev_unregister_kobject(dev); | |
024e9679 AD |
8217 | #ifdef CONFIG_XPS |
8218 | /* Remove XPS queueing entries */ | |
8219 | netif_reset_xps_queues_gt(dev, 0); | |
8220 | #endif | |
9b5e383c | 8221 | } |
93ee31f1 | 8222 | |
850a545b | 8223 | synchronize_net(); |
395264d5 | 8224 | |
a5ee1551 | 8225 | list_for_each_entry(dev, head, unreg_list) |
9b5e383c ED |
8226 | dev_put(dev); |
8227 | } | |
8228 | ||
8229 | static void rollback_registered(struct net_device *dev) | |
8230 | { | |
8231 | LIST_HEAD(single); | |
8232 | ||
8233 | list_add(&dev->unreg_list, &single); | |
8234 | rollback_registered_many(&single); | |
ceaaec98 | 8235 | list_del(&single); |
93ee31f1 DL |
8236 | } |
8237 | ||
fd867d51 JW |
8238 | static netdev_features_t netdev_sync_upper_features(struct net_device *lower, |
8239 | struct net_device *upper, netdev_features_t features) | |
8240 | { | |
8241 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
8242 | netdev_features_t feature; | |
5ba3f7d6 | 8243 | int feature_bit; |
fd867d51 | 8244 | |
3b89ea9c | 8245 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 8246 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
8247 | if (!(upper->wanted_features & feature) |
8248 | && (features & feature)) { | |
8249 | netdev_dbg(lower, "Dropping feature %pNF, upper dev %s has it off.\n", | |
8250 | &feature, upper->name); | |
8251 | features &= ~feature; | |
8252 | } | |
8253 | } | |
8254 | ||
8255 | return features; | |
8256 | } | |
8257 | ||
8258 | static void netdev_sync_lower_features(struct net_device *upper, | |
8259 | struct net_device *lower, netdev_features_t features) | |
8260 | { | |
8261 | netdev_features_t upper_disables = NETIF_F_UPPER_DISABLES; | |
8262 | netdev_features_t feature; | |
5ba3f7d6 | 8263 | int feature_bit; |
fd867d51 | 8264 | |
3b89ea9c | 8265 | for_each_netdev_feature(upper_disables, feature_bit) { |
5ba3f7d6 | 8266 | feature = __NETIF_F_BIT(feature_bit); |
fd867d51 JW |
8267 | if (!(features & feature) && (lower->features & feature)) { |
8268 | netdev_dbg(upper, "Disabling feature %pNF on lower dev %s.\n", | |
8269 | &feature, lower->name); | |
8270 | lower->wanted_features &= ~feature; | |
8271 | netdev_update_features(lower); | |
8272 | ||
8273 | if (unlikely(lower->features & feature)) | |
8274 | netdev_WARN(upper, "failed to disable %pNF on %s!\n", | |
8275 | &feature, lower->name); | |
8276 | } | |
8277 | } | |
8278 | } | |
8279 | ||
c8f44aff MM |
8280 | static netdev_features_t netdev_fix_features(struct net_device *dev, |
8281 | netdev_features_t features) | |
b63365a2 | 8282 | { |
57422dc5 MM |
8283 | /* Fix illegal checksum combinations */ |
8284 | if ((features & NETIF_F_HW_CSUM) && | |
8285 | (features & (NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM))) { | |
6f404e44 | 8286 | netdev_warn(dev, "mixed HW and IP checksum settings.\n"); |
57422dc5 MM |
8287 | features &= ~(NETIF_F_IP_CSUM|NETIF_F_IPV6_CSUM); |
8288 | } | |
8289 | ||
b63365a2 | 8290 | /* TSO requires that SG is present as well. */ |
ea2d3688 | 8291 | if ((features & NETIF_F_ALL_TSO) && !(features & NETIF_F_SG)) { |
6f404e44 | 8292 | netdev_dbg(dev, "Dropping TSO features since no SG feature.\n"); |
ea2d3688 | 8293 | features &= ~NETIF_F_ALL_TSO; |
b63365a2 HX |
8294 | } |
8295 | ||
ec5f0615 PS |
8296 | if ((features & NETIF_F_TSO) && !(features & NETIF_F_HW_CSUM) && |
8297 | !(features & NETIF_F_IP_CSUM)) { | |
8298 | netdev_dbg(dev, "Dropping TSO features since no CSUM feature.\n"); | |
8299 | features &= ~NETIF_F_TSO; | |
8300 | features &= ~NETIF_F_TSO_ECN; | |
8301 | } | |
8302 | ||
8303 | if ((features & NETIF_F_TSO6) && !(features & NETIF_F_HW_CSUM) && | |
8304 | !(features & NETIF_F_IPV6_CSUM)) { | |
8305 | netdev_dbg(dev, "Dropping TSO6 features since no CSUM feature.\n"); | |
8306 | features &= ~NETIF_F_TSO6; | |
8307 | } | |
8308 | ||
b1dc497b AD |
8309 | /* TSO with IPv4 ID mangling requires IPv4 TSO be enabled */ |
8310 | if ((features & NETIF_F_TSO_MANGLEID) && !(features & NETIF_F_TSO)) | |
8311 | features &= ~NETIF_F_TSO_MANGLEID; | |
8312 | ||
31d8b9e0 BH |
8313 | /* TSO ECN requires that TSO is present as well. */ |
8314 | if ((features & NETIF_F_ALL_TSO) == NETIF_F_TSO_ECN) | |
8315 | features &= ~NETIF_F_TSO_ECN; | |
8316 | ||
212b573f MM |
8317 | /* Software GSO depends on SG. */ |
8318 | if ((features & NETIF_F_GSO) && !(features & NETIF_F_SG)) { | |
6f404e44 | 8319 | netdev_dbg(dev, "Dropping NETIF_F_GSO since no SG feature.\n"); |
212b573f MM |
8320 | features &= ~NETIF_F_GSO; |
8321 | } | |
8322 | ||
802ab55a AD |
8323 | /* GSO partial features require GSO partial be set */ |
8324 | if ((features & dev->gso_partial_features) && | |
8325 | !(features & NETIF_F_GSO_PARTIAL)) { | |
8326 | netdev_dbg(dev, | |
8327 | "Dropping partially supported GSO features since no GSO partial.\n"); | |
8328 | features &= ~dev->gso_partial_features; | |
8329 | } | |
8330 | ||
fb1f5f79 MC |
8331 | if (!(features & NETIF_F_RXCSUM)) { |
8332 | /* NETIF_F_GRO_HW implies doing RXCSUM since every packet | |
8333 | * successfully merged by hardware must also have the | |
8334 | * checksum verified by hardware. If the user does not | |
8335 | * want to enable RXCSUM, logically, we should disable GRO_HW. | |
8336 | */ | |
8337 | if (features & NETIF_F_GRO_HW) { | |
8338 | netdev_dbg(dev, "Dropping NETIF_F_GRO_HW since no RXCSUM feature.\n"); | |
8339 | features &= ~NETIF_F_GRO_HW; | |
8340 | } | |
8341 | } | |
8342 | ||
de8d5ab2 GP |
8343 | /* LRO/HW-GRO features cannot be combined with RX-FCS */ |
8344 | if (features & NETIF_F_RXFCS) { | |
8345 | if (features & NETIF_F_LRO) { | |
8346 | netdev_dbg(dev, "Dropping LRO feature since RX-FCS is requested.\n"); | |
8347 | features &= ~NETIF_F_LRO; | |
8348 | } | |
8349 | ||
8350 | if (features & NETIF_F_GRO_HW) { | |
8351 | netdev_dbg(dev, "Dropping HW-GRO feature since RX-FCS is requested.\n"); | |
8352 | features &= ~NETIF_F_GRO_HW; | |
8353 | } | |
e6c6a929 GP |
8354 | } |
8355 | ||
b63365a2 HX |
8356 | return features; |
8357 | } | |
b63365a2 | 8358 | |
6cb6a27c | 8359 | int __netdev_update_features(struct net_device *dev) |
5455c699 | 8360 | { |
fd867d51 | 8361 | struct net_device *upper, *lower; |
c8f44aff | 8362 | netdev_features_t features; |
fd867d51 | 8363 | struct list_head *iter; |
e7868a85 | 8364 | int err = -1; |
5455c699 | 8365 | |
87267485 MM |
8366 | ASSERT_RTNL(); |
8367 | ||
5455c699 MM |
8368 | features = netdev_get_wanted_features(dev); |
8369 | ||
8370 | if (dev->netdev_ops->ndo_fix_features) | |
8371 | features = dev->netdev_ops->ndo_fix_features(dev, features); | |
8372 | ||
8373 | /* driver might be less strict about feature dependencies */ | |
8374 | features = netdev_fix_features(dev, features); | |
8375 | ||
fd867d51 JW |
8376 | /* some features can't be enabled if they're off an an upper device */ |
8377 | netdev_for_each_upper_dev_rcu(dev, upper, iter) | |
8378 | features = netdev_sync_upper_features(dev, upper, features); | |
8379 | ||
5455c699 | 8380 | if (dev->features == features) |
e7868a85 | 8381 | goto sync_lower; |
5455c699 | 8382 | |
c8f44aff MM |
8383 | netdev_dbg(dev, "Features changed: %pNF -> %pNF\n", |
8384 | &dev->features, &features); | |
5455c699 MM |
8385 | |
8386 | if (dev->netdev_ops->ndo_set_features) | |
8387 | err = dev->netdev_ops->ndo_set_features(dev, features); | |
5f8dc33e NA |
8388 | else |
8389 | err = 0; | |
5455c699 | 8390 | |
6cb6a27c | 8391 | if (unlikely(err < 0)) { |
5455c699 | 8392 | netdev_err(dev, |
c8f44aff MM |
8393 | "set_features() failed (%d); wanted %pNF, left %pNF\n", |
8394 | err, &features, &dev->features); | |
17b85d29 NA |
8395 | /* return non-0 since some features might have changed and |
8396 | * it's better to fire a spurious notification than miss it | |
8397 | */ | |
8398 | return -1; | |
6cb6a27c MM |
8399 | } |
8400 | ||
e7868a85 | 8401 | sync_lower: |
fd867d51 JW |
8402 | /* some features must be disabled on lower devices when disabled |
8403 | * on an upper device (think: bonding master or bridge) | |
8404 | */ | |
8405 | netdev_for_each_lower_dev(dev, lower, iter) | |
8406 | netdev_sync_lower_features(dev, lower, features); | |
8407 | ||
ae847f40 SD |
8408 | if (!err) { |
8409 | netdev_features_t diff = features ^ dev->features; | |
8410 | ||
8411 | if (diff & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
8412 | /* udp_tunnel_{get,drop}_rx_info both need | |
8413 | * NETIF_F_RX_UDP_TUNNEL_PORT enabled on the | |
8414 | * device, or they won't do anything. | |
8415 | * Thus we need to update dev->features | |
8416 | * *before* calling udp_tunnel_get_rx_info, | |
8417 | * but *after* calling udp_tunnel_drop_rx_info. | |
8418 | */ | |
8419 | if (features & NETIF_F_RX_UDP_TUNNEL_PORT) { | |
8420 | dev->features = features; | |
8421 | udp_tunnel_get_rx_info(dev); | |
8422 | } else { | |
8423 | udp_tunnel_drop_rx_info(dev); | |
8424 | } | |
8425 | } | |
8426 | ||
9daae9bd GP |
8427 | if (diff & NETIF_F_HW_VLAN_CTAG_FILTER) { |
8428 | if (features & NETIF_F_HW_VLAN_CTAG_FILTER) { | |
8429 | dev->features = features; | |
8430 | err |= vlan_get_rx_ctag_filter_info(dev); | |
8431 | } else { | |
8432 | vlan_drop_rx_ctag_filter_info(dev); | |
8433 | } | |
8434 | } | |
8435 | ||
8436 | if (diff & NETIF_F_HW_VLAN_STAG_FILTER) { | |
8437 | if (features & NETIF_F_HW_VLAN_STAG_FILTER) { | |
8438 | dev->features = features; | |
8439 | err |= vlan_get_rx_stag_filter_info(dev); | |
8440 | } else { | |
8441 | vlan_drop_rx_stag_filter_info(dev); | |
8442 | } | |
8443 | } | |
8444 | ||
6cb6a27c | 8445 | dev->features = features; |
ae847f40 | 8446 | } |
6cb6a27c | 8447 | |
e7868a85 | 8448 | return err < 0 ? 0 : 1; |
6cb6a27c MM |
8449 | } |
8450 | ||
afe12cc8 MM |
8451 | /** |
8452 | * netdev_update_features - recalculate device features | |
8453 | * @dev: the device to check | |
8454 | * | |
8455 | * Recalculate dev->features set and send notifications if it | |
8456 | * has changed. Should be called after driver or hardware dependent | |
8457 | * conditions might have changed that influence the features. | |
8458 | */ | |
6cb6a27c MM |
8459 | void netdev_update_features(struct net_device *dev) |
8460 | { | |
8461 | if (__netdev_update_features(dev)) | |
8462 | netdev_features_change(dev); | |
5455c699 MM |
8463 | } |
8464 | EXPORT_SYMBOL(netdev_update_features); | |
8465 | ||
afe12cc8 MM |
8466 | /** |
8467 | * netdev_change_features - recalculate device features | |
8468 | * @dev: the device to check | |
8469 | * | |
8470 | * Recalculate dev->features set and send notifications even | |
8471 | * if they have not changed. Should be called instead of | |
8472 | * netdev_update_features() if also dev->vlan_features might | |
8473 | * have changed to allow the changes to be propagated to stacked | |
8474 | * VLAN devices. | |
8475 | */ | |
8476 | void netdev_change_features(struct net_device *dev) | |
8477 | { | |
8478 | __netdev_update_features(dev); | |
8479 | netdev_features_change(dev); | |
8480 | } | |
8481 | EXPORT_SYMBOL(netdev_change_features); | |
8482 | ||
fc4a7489 PM |
8483 | /** |
8484 | * netif_stacked_transfer_operstate - transfer operstate | |
8485 | * @rootdev: the root or lower level device to transfer state from | |
8486 | * @dev: the device to transfer operstate to | |
8487 | * | |
8488 | * Transfer operational state from root to device. This is normally | |
8489 | * called when a stacking relationship exists between the root | |
8490 | * device and the device(a leaf device). | |
8491 | */ | |
8492 | void netif_stacked_transfer_operstate(const struct net_device *rootdev, | |
8493 | struct net_device *dev) | |
8494 | { | |
8495 | if (rootdev->operstate == IF_OPER_DORMANT) | |
8496 | netif_dormant_on(dev); | |
8497 | else | |
8498 | netif_dormant_off(dev); | |
8499 | ||
0575c86b ZS |
8500 | if (netif_carrier_ok(rootdev)) |
8501 | netif_carrier_on(dev); | |
8502 | else | |
8503 | netif_carrier_off(dev); | |
fc4a7489 PM |
8504 | } |
8505 | EXPORT_SYMBOL(netif_stacked_transfer_operstate); | |
8506 | ||
1b4bf461 ED |
8507 | static int netif_alloc_rx_queues(struct net_device *dev) |
8508 | { | |
1b4bf461 | 8509 | unsigned int i, count = dev->num_rx_queues; |
bd25fa7b | 8510 | struct netdev_rx_queue *rx; |
10595902 | 8511 | size_t sz = count * sizeof(*rx); |
e817f856 | 8512 | int err = 0; |
1b4bf461 | 8513 | |
bd25fa7b | 8514 | BUG_ON(count < 1); |
1b4bf461 | 8515 | |
dcda9b04 | 8516 | rx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
8517 | if (!rx) |
8518 | return -ENOMEM; | |
8519 | ||
bd25fa7b TH |
8520 | dev->_rx = rx; |
8521 | ||
e817f856 | 8522 | for (i = 0; i < count; i++) { |
fe822240 | 8523 | rx[i].dev = dev; |
e817f856 JDB |
8524 | |
8525 | /* XDP RX-queue setup */ | |
8526 | err = xdp_rxq_info_reg(&rx[i].xdp_rxq, dev, i); | |
8527 | if (err < 0) | |
8528 | goto err_rxq_info; | |
8529 | } | |
1b4bf461 | 8530 | return 0; |
e817f856 JDB |
8531 | |
8532 | err_rxq_info: | |
8533 | /* Rollback successful reg's and free other resources */ | |
8534 | while (i--) | |
8535 | xdp_rxq_info_unreg(&rx[i].xdp_rxq); | |
141b52a9 | 8536 | kvfree(dev->_rx); |
e817f856 JDB |
8537 | dev->_rx = NULL; |
8538 | return err; | |
8539 | } | |
8540 | ||
8541 | static void netif_free_rx_queues(struct net_device *dev) | |
8542 | { | |
8543 | unsigned int i, count = dev->num_rx_queues; | |
e817f856 JDB |
8544 | |
8545 | /* netif_alloc_rx_queues alloc failed, resources have been unreg'ed */ | |
8546 | if (!dev->_rx) | |
8547 | return; | |
8548 | ||
e817f856 | 8549 | for (i = 0; i < count; i++) |
82aaff2f JK |
8550 | xdp_rxq_info_unreg(&dev->_rx[i].xdp_rxq); |
8551 | ||
8552 | kvfree(dev->_rx); | |
1b4bf461 ED |
8553 | } |
8554 | ||
aa942104 CG |
8555 | static void netdev_init_one_queue(struct net_device *dev, |
8556 | struct netdev_queue *queue, void *_unused) | |
8557 | { | |
8558 | /* Initialize queue lock */ | |
8559 | spin_lock_init(&queue->_xmit_lock); | |
8560 | netdev_set_xmit_lockdep_class(&queue->_xmit_lock, dev->type); | |
8561 | queue->xmit_lock_owner = -1; | |
b236da69 | 8562 | netdev_queue_numa_node_write(queue, NUMA_NO_NODE); |
aa942104 | 8563 | queue->dev = dev; |
114cf580 TH |
8564 | #ifdef CONFIG_BQL |
8565 | dql_init(&queue->dql, HZ); | |
8566 | #endif | |
aa942104 CG |
8567 | } |
8568 | ||
60877a32 ED |
8569 | static void netif_free_tx_queues(struct net_device *dev) |
8570 | { | |
4cb28970 | 8571 | kvfree(dev->_tx); |
60877a32 ED |
8572 | } |
8573 | ||
e6484930 TH |
8574 | static int netif_alloc_netdev_queues(struct net_device *dev) |
8575 | { | |
8576 | unsigned int count = dev->num_tx_queues; | |
8577 | struct netdev_queue *tx; | |
60877a32 | 8578 | size_t sz = count * sizeof(*tx); |
e6484930 | 8579 | |
d339727c ED |
8580 | if (count < 1 || count > 0xffff) |
8581 | return -EINVAL; | |
62b5942a | 8582 | |
dcda9b04 | 8583 | tx = kvzalloc(sz, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
da6bc57a MH |
8584 | if (!tx) |
8585 | return -ENOMEM; | |
8586 | ||
e6484930 | 8587 | dev->_tx = tx; |
1d24eb48 | 8588 | |
e6484930 TH |
8589 | netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL); |
8590 | spin_lock_init(&dev->tx_global_lock); | |
aa942104 CG |
8591 | |
8592 | return 0; | |
e6484930 TH |
8593 | } |
8594 | ||
a2029240 DV |
8595 | void netif_tx_stop_all_queues(struct net_device *dev) |
8596 | { | |
8597 | unsigned int i; | |
8598 | ||
8599 | for (i = 0; i < dev->num_tx_queues; i++) { | |
8600 | struct netdev_queue *txq = netdev_get_tx_queue(dev, i); | |
f4563a75 | 8601 | |
a2029240 DV |
8602 | netif_tx_stop_queue(txq); |
8603 | } | |
8604 | } | |
8605 | EXPORT_SYMBOL(netif_tx_stop_all_queues); | |
8606 | ||
1da177e4 LT |
8607 | /** |
8608 | * register_netdevice - register a network device | |
8609 | * @dev: device to register | |
8610 | * | |
8611 | * Take a completed network device structure and add it to the kernel | |
8612 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
8613 | * chain. 0 is returned on success. A negative errno code is returned | |
8614 | * on a failure to set up the device, or if the name is a duplicate. | |
8615 | * | |
8616 | * Callers must hold the rtnl semaphore. You may want | |
8617 | * register_netdev() instead of this. | |
8618 | * | |
8619 | * BUGS: | |
8620 | * The locking appears insufficient to guarantee two parallel registers | |
8621 | * will not get the same name. | |
8622 | */ | |
8623 | ||
8624 | int register_netdevice(struct net_device *dev) | |
8625 | { | |
1da177e4 | 8626 | int ret; |
d314774c | 8627 | struct net *net = dev_net(dev); |
1da177e4 | 8628 | |
e283de3a FF |
8629 | BUILD_BUG_ON(sizeof(netdev_features_t) * BITS_PER_BYTE < |
8630 | NETDEV_FEATURE_COUNT); | |
1da177e4 LT |
8631 | BUG_ON(dev_boot_phase); |
8632 | ASSERT_RTNL(); | |
8633 | ||
b17a7c17 SH |
8634 | might_sleep(); |
8635 | ||
1da177e4 LT |
8636 | /* When net_device's are persistent, this will be fatal. */ |
8637 | BUG_ON(dev->reg_state != NETREG_UNINITIALIZED); | |
d314774c | 8638 | BUG_ON(!net); |
1da177e4 | 8639 | |
f1f28aa3 | 8640 | spin_lock_init(&dev->addr_list_lock); |
cf508b12 | 8641 | netdev_set_addr_lockdep_class(dev); |
1da177e4 | 8642 | |
828de4f6 | 8643 | ret = dev_get_valid_name(net, dev, dev->name); |
0696c3a8 PP |
8644 | if (ret < 0) |
8645 | goto out; | |
8646 | ||
1da177e4 | 8647 | /* Init, if this function is available */ |
d314774c SH |
8648 | if (dev->netdev_ops->ndo_init) { |
8649 | ret = dev->netdev_ops->ndo_init(dev); | |
1da177e4 LT |
8650 | if (ret) { |
8651 | if (ret > 0) | |
8652 | ret = -EIO; | |
90833aa4 | 8653 | goto out; |
1da177e4 LT |
8654 | } |
8655 | } | |
4ec93edb | 8656 | |
f646968f PM |
8657 | if (((dev->hw_features | dev->features) & |
8658 | NETIF_F_HW_VLAN_CTAG_FILTER) && | |
d2ed273d MM |
8659 | (!dev->netdev_ops->ndo_vlan_rx_add_vid || |
8660 | !dev->netdev_ops->ndo_vlan_rx_kill_vid)) { | |
8661 | netdev_WARN(dev, "Buggy VLAN acceleration in driver!\n"); | |
8662 | ret = -EINVAL; | |
8663 | goto err_uninit; | |
8664 | } | |
8665 | ||
9c7dafbf PE |
8666 | ret = -EBUSY; |
8667 | if (!dev->ifindex) | |
8668 | dev->ifindex = dev_new_index(net); | |
8669 | else if (__dev_get_by_index(net, dev->ifindex)) | |
8670 | goto err_uninit; | |
8671 | ||
5455c699 MM |
8672 | /* Transfer changeable features to wanted_features and enable |
8673 | * software offloads (GSO and GRO). | |
8674 | */ | |
8675 | dev->hw_features |= NETIF_F_SOFT_FEATURES; | |
14d1232f | 8676 | dev->features |= NETIF_F_SOFT_FEATURES; |
d764a122 SD |
8677 | |
8678 | if (dev->netdev_ops->ndo_udp_tunnel_add) { | |
8679 | dev->features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
8680 | dev->hw_features |= NETIF_F_RX_UDP_TUNNEL_PORT; | |
8681 | } | |
8682 | ||
14d1232f | 8683 | dev->wanted_features = dev->features & dev->hw_features; |
1da177e4 | 8684 | |
cbc53e08 | 8685 | if (!(dev->flags & IFF_LOOPBACK)) |
34324dc2 | 8686 | dev->hw_features |= NETIF_F_NOCACHE_COPY; |
cbc53e08 | 8687 | |
7f348a60 AD |
8688 | /* If IPv4 TCP segmentation offload is supported we should also |
8689 | * allow the device to enable segmenting the frame with the option | |
8690 | * of ignoring a static IP ID value. This doesn't enable the | |
8691 | * feature itself but allows the user to enable it later. | |
8692 | */ | |
cbc53e08 AD |
8693 | if (dev->hw_features & NETIF_F_TSO) |
8694 | dev->hw_features |= NETIF_F_TSO_MANGLEID; | |
7f348a60 AD |
8695 | if (dev->vlan_features & NETIF_F_TSO) |
8696 | dev->vlan_features |= NETIF_F_TSO_MANGLEID; | |
8697 | if (dev->mpls_features & NETIF_F_TSO) | |
8698 | dev->mpls_features |= NETIF_F_TSO_MANGLEID; | |
8699 | if (dev->hw_enc_features & NETIF_F_TSO) | |
8700 | dev->hw_enc_features |= NETIF_F_TSO_MANGLEID; | |
c6e1a0d1 | 8701 | |
1180e7d6 | 8702 | /* Make NETIF_F_HIGHDMA inheritable to VLAN devices. |
16c3ea78 | 8703 | */ |
1180e7d6 | 8704 | dev->vlan_features |= NETIF_F_HIGHDMA; |
16c3ea78 | 8705 | |
ee579677 PS |
8706 | /* Make NETIF_F_SG inheritable to tunnel devices. |
8707 | */ | |
802ab55a | 8708 | dev->hw_enc_features |= NETIF_F_SG | NETIF_F_GSO_PARTIAL; |
ee579677 | 8709 | |
0d89d203 SH |
8710 | /* Make NETIF_F_SG inheritable to MPLS. |
8711 | */ | |
8712 | dev->mpls_features |= NETIF_F_SG; | |
8713 | ||
7ffbe3fd JB |
8714 | ret = call_netdevice_notifiers(NETDEV_POST_INIT, dev); |
8715 | ret = notifier_to_errno(ret); | |
8716 | if (ret) | |
8717 | goto err_uninit; | |
8718 | ||
8b41d188 | 8719 | ret = netdev_register_kobject(dev); |
b17a7c17 | 8720 | if (ret) |
7ce1b0ed | 8721 | goto err_uninit; |
b17a7c17 SH |
8722 | dev->reg_state = NETREG_REGISTERED; |
8723 | ||
6cb6a27c | 8724 | __netdev_update_features(dev); |
8e9b59b2 | 8725 | |
1da177e4 LT |
8726 | /* |
8727 | * Default initial state at registry is that the | |
8728 | * device is present. | |
8729 | */ | |
8730 | ||
8731 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
8732 | ||
8f4cccbb BH |
8733 | linkwatch_init_dev(dev); |
8734 | ||
1da177e4 | 8735 | dev_init_scheduler(dev); |
1da177e4 | 8736 | dev_hold(dev); |
ce286d32 | 8737 | list_netdevice(dev); |
7bf23575 | 8738 | add_device_randomness(dev->dev_addr, dev->addr_len); |
1da177e4 | 8739 | |
948b337e JP |
8740 | /* If the device has permanent device address, driver should |
8741 | * set dev_addr and also addr_assign_type should be set to | |
8742 | * NET_ADDR_PERM (default value). | |
8743 | */ | |
8744 | if (dev->addr_assign_type == NET_ADDR_PERM) | |
8745 | memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len); | |
8746 | ||
1da177e4 | 8747 | /* Notify protocols, that a new device appeared. */ |
056925ab | 8748 | ret = call_netdevice_notifiers(NETDEV_REGISTER, dev); |
fcc5a03a | 8749 | ret = notifier_to_errno(ret); |
93ee31f1 DL |
8750 | if (ret) { |
8751 | rollback_registered(dev); | |
8752 | dev->reg_state = NETREG_UNREGISTERED; | |
8753 | } | |
d90a909e EB |
8754 | /* |
8755 | * Prevent userspace races by waiting until the network | |
8756 | * device is fully setup before sending notifications. | |
8757 | */ | |
a2835763 PM |
8758 | if (!dev->rtnl_link_ops || |
8759 | dev->rtnl_link_state == RTNL_LINK_INITIALIZED) | |
7f294054 | 8760 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
1da177e4 LT |
8761 | |
8762 | out: | |
8763 | return ret; | |
7ce1b0ed HX |
8764 | |
8765 | err_uninit: | |
d314774c SH |
8766 | if (dev->netdev_ops->ndo_uninit) |
8767 | dev->netdev_ops->ndo_uninit(dev); | |
cf124db5 DM |
8768 | if (dev->priv_destructor) |
8769 | dev->priv_destructor(dev); | |
7ce1b0ed | 8770 | goto out; |
1da177e4 | 8771 | } |
d1b19dff | 8772 | EXPORT_SYMBOL(register_netdevice); |
1da177e4 | 8773 | |
937f1ba5 BH |
8774 | /** |
8775 | * init_dummy_netdev - init a dummy network device for NAPI | |
8776 | * @dev: device to init | |
8777 | * | |
8778 | * This takes a network device structure and initialize the minimum | |
8779 | * amount of fields so it can be used to schedule NAPI polls without | |
8780 | * registering a full blown interface. This is to be used by drivers | |
8781 | * that need to tie several hardware interfaces to a single NAPI | |
8782 | * poll scheduler due to HW limitations. | |
8783 | */ | |
8784 | int init_dummy_netdev(struct net_device *dev) | |
8785 | { | |
8786 | /* Clear everything. Note we don't initialize spinlocks | |
8787 | * are they aren't supposed to be taken by any of the | |
8788 | * NAPI code and this dummy netdev is supposed to be | |
8789 | * only ever used for NAPI polls | |
8790 | */ | |
8791 | memset(dev, 0, sizeof(struct net_device)); | |
8792 | ||
8793 | /* make sure we BUG if trying to hit standard | |
8794 | * register/unregister code path | |
8795 | */ | |
8796 | dev->reg_state = NETREG_DUMMY; | |
8797 | ||
937f1ba5 BH |
8798 | /* NAPI wants this */ |
8799 | INIT_LIST_HEAD(&dev->napi_list); | |
8800 | ||
8801 | /* a dummy interface is started by default */ | |
8802 | set_bit(__LINK_STATE_PRESENT, &dev->state); | |
8803 | set_bit(__LINK_STATE_START, &dev->state); | |
8804 | ||
35edfdc7 JE |
8805 | /* napi_busy_loop stats accounting wants this */ |
8806 | dev_net_set(dev, &init_net); | |
8807 | ||
29b4433d ED |
8808 | /* Note : We dont allocate pcpu_refcnt for dummy devices, |
8809 | * because users of this 'device' dont need to change | |
8810 | * its refcount. | |
8811 | */ | |
8812 | ||
937f1ba5 BH |
8813 | return 0; |
8814 | } | |
8815 | EXPORT_SYMBOL_GPL(init_dummy_netdev); | |
8816 | ||
8817 | ||
1da177e4 LT |
8818 | /** |
8819 | * register_netdev - register a network device | |
8820 | * @dev: device to register | |
8821 | * | |
8822 | * Take a completed network device structure and add it to the kernel | |
8823 | * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier | |
8824 | * chain. 0 is returned on success. A negative errno code is returned | |
8825 | * on a failure to set up the device, or if the name is a duplicate. | |
8826 | * | |
38b4da38 | 8827 | * This is a wrapper around register_netdevice that takes the rtnl semaphore |
1da177e4 LT |
8828 | * and expands the device name if you passed a format string to |
8829 | * alloc_netdev. | |
8830 | */ | |
8831 | int register_netdev(struct net_device *dev) | |
8832 | { | |
8833 | int err; | |
8834 | ||
b0f3debc KT |
8835 | if (rtnl_lock_killable()) |
8836 | return -EINTR; | |
1da177e4 | 8837 | err = register_netdevice(dev); |
1da177e4 LT |
8838 | rtnl_unlock(); |
8839 | return err; | |
8840 | } | |
8841 | EXPORT_SYMBOL(register_netdev); | |
8842 | ||
29b4433d ED |
8843 | int netdev_refcnt_read(const struct net_device *dev) |
8844 | { | |
8845 | int i, refcnt = 0; | |
8846 | ||
8847 | for_each_possible_cpu(i) | |
8848 | refcnt += *per_cpu_ptr(dev->pcpu_refcnt, i); | |
8849 | return refcnt; | |
8850 | } | |
8851 | EXPORT_SYMBOL(netdev_refcnt_read); | |
8852 | ||
2c53040f | 8853 | /** |
1da177e4 | 8854 | * netdev_wait_allrefs - wait until all references are gone. |
3de7a37b | 8855 | * @dev: target net_device |
1da177e4 LT |
8856 | * |
8857 | * This is called when unregistering network devices. | |
8858 | * | |
8859 | * Any protocol or device that holds a reference should register | |
8860 | * for netdevice notification, and cleanup and put back the | |
8861 | * reference if they receive an UNREGISTER event. | |
8862 | * We can get stuck here if buggy protocols don't correctly | |
4ec93edb | 8863 | * call dev_put. |
1da177e4 LT |
8864 | */ |
8865 | static void netdev_wait_allrefs(struct net_device *dev) | |
8866 | { | |
8867 | unsigned long rebroadcast_time, warning_time; | |
29b4433d | 8868 | int refcnt; |
1da177e4 | 8869 | |
e014debe ED |
8870 | linkwatch_forget_dev(dev); |
8871 | ||
1da177e4 | 8872 | rebroadcast_time = warning_time = jiffies; |
29b4433d ED |
8873 | refcnt = netdev_refcnt_read(dev); |
8874 | ||
8875 | while (refcnt != 0) { | |
1da177e4 | 8876 | if (time_after(jiffies, rebroadcast_time + 1 * HZ)) { |
6756ae4b | 8877 | rtnl_lock(); |
1da177e4 LT |
8878 | |
8879 | /* Rebroadcast unregister notification */ | |
056925ab | 8880 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
1da177e4 | 8881 | |
748e2d93 | 8882 | __rtnl_unlock(); |
0115e8e3 | 8883 | rcu_barrier(); |
748e2d93 ED |
8884 | rtnl_lock(); |
8885 | ||
1da177e4 LT |
8886 | if (test_bit(__LINK_STATE_LINKWATCH_PENDING, |
8887 | &dev->state)) { | |
8888 | /* We must not have linkwatch events | |
8889 | * pending on unregister. If this | |
8890 | * happens, we simply run the queue | |
8891 | * unscheduled, resulting in a noop | |
8892 | * for this device. | |
8893 | */ | |
8894 | linkwatch_run_queue(); | |
8895 | } | |
8896 | ||
6756ae4b | 8897 | __rtnl_unlock(); |
1da177e4 LT |
8898 | |
8899 | rebroadcast_time = jiffies; | |
8900 | } | |
8901 | ||
8902 | msleep(250); | |
8903 | ||
29b4433d ED |
8904 | refcnt = netdev_refcnt_read(dev); |
8905 | ||
1da177e4 | 8906 | if (time_after(jiffies, warning_time + 10 * HZ)) { |
7b6cd1ce JP |
8907 | pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n", |
8908 | dev->name, refcnt); | |
1da177e4 LT |
8909 | warning_time = jiffies; |
8910 | } | |
8911 | } | |
8912 | } | |
8913 | ||
8914 | /* The sequence is: | |
8915 | * | |
8916 | * rtnl_lock(); | |
8917 | * ... | |
8918 | * register_netdevice(x1); | |
8919 | * register_netdevice(x2); | |
8920 | * ... | |
8921 | * unregister_netdevice(y1); | |
8922 | * unregister_netdevice(y2); | |
8923 | * ... | |
8924 | * rtnl_unlock(); | |
8925 | * free_netdev(y1); | |
8926 | * free_netdev(y2); | |
8927 | * | |
58ec3b4d | 8928 | * We are invoked by rtnl_unlock(). |
1da177e4 | 8929 | * This allows us to deal with problems: |
b17a7c17 | 8930 | * 1) We can delete sysfs objects which invoke hotplug |
1da177e4 LT |
8931 | * without deadlocking with linkwatch via keventd. |
8932 | * 2) Since we run with the RTNL semaphore not held, we can sleep | |
8933 | * safely in order to wait for the netdev refcnt to drop to zero. | |
58ec3b4d HX |
8934 | * |
8935 | * We must not return until all unregister events added during | |
8936 | * the interval the lock was held have been completed. | |
1da177e4 | 8937 | */ |
1da177e4 LT |
8938 | void netdev_run_todo(void) |
8939 | { | |
626ab0e6 | 8940 | struct list_head list; |
1da177e4 | 8941 | |
1da177e4 | 8942 | /* Snapshot list, allow later requests */ |
626ab0e6 | 8943 | list_replace_init(&net_todo_list, &list); |
58ec3b4d HX |
8944 | |
8945 | __rtnl_unlock(); | |
626ab0e6 | 8946 | |
0115e8e3 ED |
8947 | |
8948 | /* Wait for rcu callbacks to finish before next phase */ | |
850a545b EB |
8949 | if (!list_empty(&list)) |
8950 | rcu_barrier(); | |
8951 | ||
1da177e4 LT |
8952 | while (!list_empty(&list)) { |
8953 | struct net_device *dev | |
e5e26d75 | 8954 | = list_first_entry(&list, struct net_device, todo_list); |
1da177e4 LT |
8955 | list_del(&dev->todo_list); |
8956 | ||
b17a7c17 | 8957 | if (unlikely(dev->reg_state != NETREG_UNREGISTERING)) { |
7b6cd1ce | 8958 | pr_err("network todo '%s' but state %d\n", |
b17a7c17 SH |
8959 | dev->name, dev->reg_state); |
8960 | dump_stack(); | |
8961 | continue; | |
8962 | } | |
1da177e4 | 8963 | |
b17a7c17 | 8964 | dev->reg_state = NETREG_UNREGISTERED; |
1da177e4 | 8965 | |
b17a7c17 | 8966 | netdev_wait_allrefs(dev); |
1da177e4 | 8967 | |
b17a7c17 | 8968 | /* paranoia */ |
29b4433d | 8969 | BUG_ON(netdev_refcnt_read(dev)); |
7866a621 SN |
8970 | BUG_ON(!list_empty(&dev->ptype_all)); |
8971 | BUG_ON(!list_empty(&dev->ptype_specific)); | |
33d480ce ED |
8972 | WARN_ON(rcu_access_pointer(dev->ip_ptr)); |
8973 | WARN_ON(rcu_access_pointer(dev->ip6_ptr)); | |
330c7272 | 8974 | #if IS_ENABLED(CONFIG_DECNET) |
547b792c | 8975 | WARN_ON(dev->dn_ptr); |
330c7272 | 8976 | #endif |
cf124db5 DM |
8977 | if (dev->priv_destructor) |
8978 | dev->priv_destructor(dev); | |
8979 | if (dev->needs_free_netdev) | |
8980 | free_netdev(dev); | |
9093bbb2 | 8981 | |
50624c93 EB |
8982 | /* Report a network device has been unregistered */ |
8983 | rtnl_lock(); | |
8984 | dev_net(dev)->dev_unreg_count--; | |
8985 | __rtnl_unlock(); | |
8986 | wake_up(&netdev_unregistering_wq); | |
8987 | ||
9093bbb2 SH |
8988 | /* Free network device */ |
8989 | kobject_put(&dev->dev.kobj); | |
1da177e4 | 8990 | } |
1da177e4 LT |
8991 | } |
8992 | ||
9256645a JW |
8993 | /* Convert net_device_stats to rtnl_link_stats64. rtnl_link_stats64 has |
8994 | * all the same fields in the same order as net_device_stats, with only | |
8995 | * the type differing, but rtnl_link_stats64 may have additional fields | |
8996 | * at the end for newer counters. | |
3cfde79c | 8997 | */ |
77a1abf5 ED |
8998 | void netdev_stats_to_stats64(struct rtnl_link_stats64 *stats64, |
8999 | const struct net_device_stats *netdev_stats) | |
3cfde79c BH |
9000 | { |
9001 | #if BITS_PER_LONG == 64 | |
9256645a | 9002 | BUILD_BUG_ON(sizeof(*stats64) < sizeof(*netdev_stats)); |
9af9959e | 9003 | memcpy(stats64, netdev_stats, sizeof(*netdev_stats)); |
9256645a JW |
9004 | /* zero out counters that only exist in rtnl_link_stats64 */ |
9005 | memset((char *)stats64 + sizeof(*netdev_stats), 0, | |
9006 | sizeof(*stats64) - sizeof(*netdev_stats)); | |
3cfde79c | 9007 | #else |
9256645a | 9008 | size_t i, n = sizeof(*netdev_stats) / sizeof(unsigned long); |
3cfde79c BH |
9009 | const unsigned long *src = (const unsigned long *)netdev_stats; |
9010 | u64 *dst = (u64 *)stats64; | |
9011 | ||
9256645a | 9012 | BUILD_BUG_ON(n > sizeof(*stats64) / sizeof(u64)); |
3cfde79c BH |
9013 | for (i = 0; i < n; i++) |
9014 | dst[i] = src[i]; | |
9256645a JW |
9015 | /* zero out counters that only exist in rtnl_link_stats64 */ |
9016 | memset((char *)stats64 + n * sizeof(u64), 0, | |
9017 | sizeof(*stats64) - n * sizeof(u64)); | |
3cfde79c BH |
9018 | #endif |
9019 | } | |
77a1abf5 | 9020 | EXPORT_SYMBOL(netdev_stats_to_stats64); |
3cfde79c | 9021 | |
eeda3fd6 SH |
9022 | /** |
9023 | * dev_get_stats - get network device statistics | |
9024 | * @dev: device to get statistics from | |
28172739 | 9025 | * @storage: place to store stats |
eeda3fd6 | 9026 | * |
d7753516 BH |
9027 | * Get network statistics from device. Return @storage. |
9028 | * The device driver may provide its own method by setting | |
9029 | * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats; | |
9030 | * otherwise the internal statistics structure is used. | |
eeda3fd6 | 9031 | */ |
d7753516 BH |
9032 | struct rtnl_link_stats64 *dev_get_stats(struct net_device *dev, |
9033 | struct rtnl_link_stats64 *storage) | |
7004bf25 | 9034 | { |
eeda3fd6 SH |
9035 | const struct net_device_ops *ops = dev->netdev_ops; |
9036 | ||
28172739 ED |
9037 | if (ops->ndo_get_stats64) { |
9038 | memset(storage, 0, sizeof(*storage)); | |
caf586e5 ED |
9039 | ops->ndo_get_stats64(dev, storage); |
9040 | } else if (ops->ndo_get_stats) { | |
3cfde79c | 9041 | netdev_stats_to_stats64(storage, ops->ndo_get_stats(dev)); |
caf586e5 ED |
9042 | } else { |
9043 | netdev_stats_to_stats64(storage, &dev->stats); | |
28172739 | 9044 | } |
6f64ec74 ED |
9045 | storage->rx_dropped += (unsigned long)atomic_long_read(&dev->rx_dropped); |
9046 | storage->tx_dropped += (unsigned long)atomic_long_read(&dev->tx_dropped); | |
9047 | storage->rx_nohandler += (unsigned long)atomic_long_read(&dev->rx_nohandler); | |
28172739 | 9048 | return storage; |
c45d286e | 9049 | } |
eeda3fd6 | 9050 | EXPORT_SYMBOL(dev_get_stats); |
c45d286e | 9051 | |
24824a09 | 9052 | struct netdev_queue *dev_ingress_queue_create(struct net_device *dev) |
dc2b4847 | 9053 | { |
24824a09 | 9054 | struct netdev_queue *queue = dev_ingress_queue(dev); |
dc2b4847 | 9055 | |
24824a09 ED |
9056 | #ifdef CONFIG_NET_CLS_ACT |
9057 | if (queue) | |
9058 | return queue; | |
9059 | queue = kzalloc(sizeof(*queue), GFP_KERNEL); | |
9060 | if (!queue) | |
9061 | return NULL; | |
9062 | netdev_init_one_queue(dev, queue, NULL); | |
2ce1ee17 | 9063 | RCU_INIT_POINTER(queue->qdisc, &noop_qdisc); |
24824a09 ED |
9064 | queue->qdisc_sleeping = &noop_qdisc; |
9065 | rcu_assign_pointer(dev->ingress_queue, queue); | |
9066 | #endif | |
9067 | return queue; | |
bb949fbd DM |
9068 | } |
9069 | ||
2c60db03 ED |
9070 | static const struct ethtool_ops default_ethtool_ops; |
9071 | ||
d07d7507 SG |
9072 | void netdev_set_default_ethtool_ops(struct net_device *dev, |
9073 | const struct ethtool_ops *ops) | |
9074 | { | |
9075 | if (dev->ethtool_ops == &default_ethtool_ops) | |
9076 | dev->ethtool_ops = ops; | |
9077 | } | |
9078 | EXPORT_SYMBOL_GPL(netdev_set_default_ethtool_ops); | |
9079 | ||
74d332c1 ED |
9080 | void netdev_freemem(struct net_device *dev) |
9081 | { | |
9082 | char *addr = (char *)dev - dev->padded; | |
9083 | ||
4cb28970 | 9084 | kvfree(addr); |
74d332c1 ED |
9085 | } |
9086 | ||
1da177e4 | 9087 | /** |
722c9a0c | 9088 | * alloc_netdev_mqs - allocate network device |
9089 | * @sizeof_priv: size of private data to allocate space for | |
9090 | * @name: device name format string | |
9091 | * @name_assign_type: origin of device name | |
9092 | * @setup: callback to initialize device | |
9093 | * @txqs: the number of TX subqueues to allocate | |
9094 | * @rxqs: the number of RX subqueues to allocate | |
9095 | * | |
9096 | * Allocates a struct net_device with private data area for driver use | |
9097 | * and performs basic initialization. Also allocates subqueue structs | |
9098 | * for each queue on the device. | |
1da177e4 | 9099 | */ |
36909ea4 | 9100 | struct net_device *alloc_netdev_mqs(int sizeof_priv, const char *name, |
c835a677 | 9101 | unsigned char name_assign_type, |
36909ea4 TH |
9102 | void (*setup)(struct net_device *), |
9103 | unsigned int txqs, unsigned int rxqs) | |
1da177e4 | 9104 | { |
1da177e4 | 9105 | struct net_device *dev; |
52a59bd5 | 9106 | unsigned int alloc_size; |
1ce8e7b5 | 9107 | struct net_device *p; |
1da177e4 | 9108 | |
b6fe17d6 SH |
9109 | BUG_ON(strlen(name) >= sizeof(dev->name)); |
9110 | ||
36909ea4 | 9111 | if (txqs < 1) { |
7b6cd1ce | 9112 | pr_err("alloc_netdev: Unable to allocate device with zero queues\n"); |
55513fb4 TH |
9113 | return NULL; |
9114 | } | |
9115 | ||
36909ea4 | 9116 | if (rxqs < 1) { |
7b6cd1ce | 9117 | pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n"); |
36909ea4 TH |
9118 | return NULL; |
9119 | } | |
36909ea4 | 9120 | |
fd2ea0a7 | 9121 | alloc_size = sizeof(struct net_device); |
d1643d24 AD |
9122 | if (sizeof_priv) { |
9123 | /* ensure 32-byte alignment of private area */ | |
1ce8e7b5 | 9124 | alloc_size = ALIGN(alloc_size, NETDEV_ALIGN); |
d1643d24 AD |
9125 | alloc_size += sizeof_priv; |
9126 | } | |
9127 | /* ensure 32-byte alignment of whole construct */ | |
1ce8e7b5 | 9128 | alloc_size += NETDEV_ALIGN - 1; |
1da177e4 | 9129 | |
dcda9b04 | 9130 | p = kvzalloc(alloc_size, GFP_KERNEL | __GFP_RETRY_MAYFAIL); |
62b5942a | 9131 | if (!p) |
1da177e4 | 9132 | return NULL; |
1da177e4 | 9133 | |
1ce8e7b5 | 9134 | dev = PTR_ALIGN(p, NETDEV_ALIGN); |
1da177e4 | 9135 | dev->padded = (char *)dev - (char *)p; |
ab9c73cc | 9136 | |
29b4433d ED |
9137 | dev->pcpu_refcnt = alloc_percpu(int); |
9138 | if (!dev->pcpu_refcnt) | |
74d332c1 | 9139 | goto free_dev; |
ab9c73cc | 9140 | |
ab9c73cc | 9141 | if (dev_addr_init(dev)) |
29b4433d | 9142 | goto free_pcpu; |
ab9c73cc | 9143 | |
22bedad3 | 9144 | dev_mc_init(dev); |
a748ee24 | 9145 | dev_uc_init(dev); |
ccffad25 | 9146 | |
c346dca1 | 9147 | dev_net_set(dev, &init_net); |
1da177e4 | 9148 | |
8d3bdbd5 | 9149 | dev->gso_max_size = GSO_MAX_SIZE; |
30b678d8 | 9150 | dev->gso_max_segs = GSO_MAX_SEGS; |
8d3bdbd5 | 9151 | |
8d3bdbd5 DM |
9152 | INIT_LIST_HEAD(&dev->napi_list); |
9153 | INIT_LIST_HEAD(&dev->unreg_list); | |
5cde2829 | 9154 | INIT_LIST_HEAD(&dev->close_list); |
8d3bdbd5 | 9155 | INIT_LIST_HEAD(&dev->link_watch_list); |
2f268f12 VF |
9156 | INIT_LIST_HEAD(&dev->adj_list.upper); |
9157 | INIT_LIST_HEAD(&dev->adj_list.lower); | |
7866a621 SN |
9158 | INIT_LIST_HEAD(&dev->ptype_all); |
9159 | INIT_LIST_HEAD(&dev->ptype_specific); | |
59cc1f61 JK |
9160 | #ifdef CONFIG_NET_SCHED |
9161 | hash_init(dev->qdisc_hash); | |
9162 | #endif | |
02875878 | 9163 | dev->priv_flags = IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM; |
8d3bdbd5 DM |
9164 | setup(dev); |
9165 | ||
a813104d | 9166 | if (!dev->tx_queue_len) { |
f84bb1ea | 9167 | dev->priv_flags |= IFF_NO_QUEUE; |
11597084 | 9168 | dev->tx_queue_len = DEFAULT_TX_QUEUE_LEN; |
a813104d | 9169 | } |
906470c1 | 9170 | |
36909ea4 TH |
9171 | dev->num_tx_queues = txqs; |
9172 | dev->real_num_tx_queues = txqs; | |
ed9af2e8 | 9173 | if (netif_alloc_netdev_queues(dev)) |
8d3bdbd5 | 9174 | goto free_all; |
e8a0464c | 9175 | |
36909ea4 TH |
9176 | dev->num_rx_queues = rxqs; |
9177 | dev->real_num_rx_queues = rxqs; | |
fe822240 | 9178 | if (netif_alloc_rx_queues(dev)) |
8d3bdbd5 | 9179 | goto free_all; |
0a9627f2 | 9180 | |
1da177e4 | 9181 | strcpy(dev->name, name); |
c835a677 | 9182 | dev->name_assign_type = name_assign_type; |
cbda10fa | 9183 | dev->group = INIT_NETDEV_GROUP; |
2c60db03 ED |
9184 | if (!dev->ethtool_ops) |
9185 | dev->ethtool_ops = &default_ethtool_ops; | |
e687ad60 PN |
9186 | |
9187 | nf_hook_ingress_init(dev); | |
9188 | ||
1da177e4 | 9189 | return dev; |
ab9c73cc | 9190 | |
8d3bdbd5 DM |
9191 | free_all: |
9192 | free_netdev(dev); | |
9193 | return NULL; | |
9194 | ||
29b4433d ED |
9195 | free_pcpu: |
9196 | free_percpu(dev->pcpu_refcnt); | |
74d332c1 ED |
9197 | free_dev: |
9198 | netdev_freemem(dev); | |
ab9c73cc | 9199 | return NULL; |
1da177e4 | 9200 | } |
36909ea4 | 9201 | EXPORT_SYMBOL(alloc_netdev_mqs); |
1da177e4 LT |
9202 | |
9203 | /** | |
722c9a0c | 9204 | * free_netdev - free network device |
9205 | * @dev: device | |
1da177e4 | 9206 | * |
722c9a0c | 9207 | * This function does the last stage of destroying an allocated device |
9208 | * interface. The reference to the device object is released. If this | |
9209 | * is the last reference then it will be freed.Must be called in process | |
9210 | * context. | |
1da177e4 LT |
9211 | */ |
9212 | void free_netdev(struct net_device *dev) | |
9213 | { | |
d565b0a1 HX |
9214 | struct napi_struct *p, *n; |
9215 | ||
93d05d4a | 9216 | might_sleep(); |
60877a32 | 9217 | netif_free_tx_queues(dev); |
e817f856 | 9218 | netif_free_rx_queues(dev); |
e8a0464c | 9219 | |
33d480ce | 9220 | kfree(rcu_dereference_protected(dev->ingress_queue, 1)); |
24824a09 | 9221 | |
f001fde5 JP |
9222 | /* Flush device addresses */ |
9223 | dev_addr_flush(dev); | |
9224 | ||
d565b0a1 HX |
9225 | list_for_each_entry_safe(p, n, &dev->napi_list, dev_list) |
9226 | netif_napi_del(p); | |
9227 | ||
29b4433d ED |
9228 | free_percpu(dev->pcpu_refcnt); |
9229 | dev->pcpu_refcnt = NULL; | |
9230 | ||
3041a069 | 9231 | /* Compatibility with error handling in drivers */ |
1da177e4 | 9232 | if (dev->reg_state == NETREG_UNINITIALIZED) { |
74d332c1 | 9233 | netdev_freemem(dev); |
1da177e4 LT |
9234 | return; |
9235 | } | |
9236 | ||
9237 | BUG_ON(dev->reg_state != NETREG_UNREGISTERED); | |
9238 | dev->reg_state = NETREG_RELEASED; | |
9239 | ||
43cb76d9 GKH |
9240 | /* will free via device release */ |
9241 | put_device(&dev->dev); | |
1da177e4 | 9242 | } |
d1b19dff | 9243 | EXPORT_SYMBOL(free_netdev); |
4ec93edb | 9244 | |
f0db275a SH |
9245 | /** |
9246 | * synchronize_net - Synchronize with packet receive processing | |
9247 | * | |
9248 | * Wait for packets currently being received to be done. | |
9249 | * Does not block later packets from starting. | |
9250 | */ | |
4ec93edb | 9251 | void synchronize_net(void) |
1da177e4 LT |
9252 | { |
9253 | might_sleep(); | |
be3fc413 ED |
9254 | if (rtnl_is_locked()) |
9255 | synchronize_rcu_expedited(); | |
9256 | else | |
9257 | synchronize_rcu(); | |
1da177e4 | 9258 | } |
d1b19dff | 9259 | EXPORT_SYMBOL(synchronize_net); |
1da177e4 LT |
9260 | |
9261 | /** | |
44a0873d | 9262 | * unregister_netdevice_queue - remove device from the kernel |
1da177e4 | 9263 | * @dev: device |
44a0873d | 9264 | * @head: list |
6ebfbc06 | 9265 | * |
1da177e4 | 9266 | * This function shuts down a device interface and removes it |
d59b54b1 | 9267 | * from the kernel tables. |
44a0873d | 9268 | * If head not NULL, device is queued to be unregistered later. |
1da177e4 LT |
9269 | * |
9270 | * Callers must hold the rtnl semaphore. You may want | |
9271 | * unregister_netdev() instead of this. | |
9272 | */ | |
9273 | ||
44a0873d | 9274 | void unregister_netdevice_queue(struct net_device *dev, struct list_head *head) |
1da177e4 | 9275 | { |
a6620712 HX |
9276 | ASSERT_RTNL(); |
9277 | ||
44a0873d | 9278 | if (head) { |
9fdce099 | 9279 | list_move_tail(&dev->unreg_list, head); |
44a0873d ED |
9280 | } else { |
9281 | rollback_registered(dev); | |
9282 | /* Finish processing unregister after unlock */ | |
9283 | net_set_todo(dev); | |
9284 | } | |
1da177e4 | 9285 | } |
44a0873d | 9286 | EXPORT_SYMBOL(unregister_netdevice_queue); |
1da177e4 | 9287 | |
9b5e383c ED |
9288 | /** |
9289 | * unregister_netdevice_many - unregister many devices | |
9290 | * @head: list of devices | |
87757a91 ED |
9291 | * |
9292 | * Note: As most callers use a stack allocated list_head, | |
9293 | * we force a list_del() to make sure stack wont be corrupted later. | |
9b5e383c ED |
9294 | */ |
9295 | void unregister_netdevice_many(struct list_head *head) | |
9296 | { | |
9297 | struct net_device *dev; | |
9298 | ||
9299 | if (!list_empty(head)) { | |
9300 | rollback_registered_many(head); | |
9301 | list_for_each_entry(dev, head, unreg_list) | |
9302 | net_set_todo(dev); | |
87757a91 | 9303 | list_del(head); |
9b5e383c ED |
9304 | } |
9305 | } | |
63c8099d | 9306 | EXPORT_SYMBOL(unregister_netdevice_many); |
9b5e383c | 9307 | |
1da177e4 LT |
9308 | /** |
9309 | * unregister_netdev - remove device from the kernel | |
9310 | * @dev: device | |
9311 | * | |
9312 | * This function shuts down a device interface and removes it | |
d59b54b1 | 9313 | * from the kernel tables. |
1da177e4 LT |
9314 | * |
9315 | * This is just a wrapper for unregister_netdevice that takes | |
9316 | * the rtnl semaphore. In general you want to use this and not | |
9317 | * unregister_netdevice. | |
9318 | */ | |
9319 | void unregister_netdev(struct net_device *dev) | |
9320 | { | |
9321 | rtnl_lock(); | |
9322 | unregister_netdevice(dev); | |
9323 | rtnl_unlock(); | |
9324 | } | |
1da177e4 LT |
9325 | EXPORT_SYMBOL(unregister_netdev); |
9326 | ||
ce286d32 EB |
9327 | /** |
9328 | * dev_change_net_namespace - move device to different nethost namespace | |
9329 | * @dev: device | |
9330 | * @net: network namespace | |
9331 | * @pat: If not NULL name pattern to try if the current device name | |
9332 | * is already taken in the destination network namespace. | |
9333 | * | |
9334 | * This function shuts down a device interface and moves it | |
9335 | * to a new network namespace. On success 0 is returned, on | |
9336 | * a failure a netagive errno code is returned. | |
9337 | * | |
9338 | * Callers must hold the rtnl semaphore. | |
9339 | */ | |
9340 | ||
9341 | int dev_change_net_namespace(struct net_device *dev, struct net *net, const char *pat) | |
9342 | { | |
38e01b30 | 9343 | int err, new_nsid, new_ifindex; |
ce286d32 EB |
9344 | |
9345 | ASSERT_RTNL(); | |
9346 | ||
9347 | /* Don't allow namespace local devices to be moved. */ | |
9348 | err = -EINVAL; | |
9349 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
9350 | goto out; | |
9351 | ||
9352 | /* Ensure the device has been registrered */ | |
ce286d32 EB |
9353 | if (dev->reg_state != NETREG_REGISTERED) |
9354 | goto out; | |
9355 | ||
9356 | /* Get out if there is nothing todo */ | |
9357 | err = 0; | |
878628fb | 9358 | if (net_eq(dev_net(dev), net)) |
ce286d32 EB |
9359 | goto out; |
9360 | ||
9361 | /* Pick the destination device name, and ensure | |
9362 | * we can use it in the destination network namespace. | |
9363 | */ | |
9364 | err = -EEXIST; | |
d9031024 | 9365 | if (__dev_get_by_name(net, dev->name)) { |
ce286d32 EB |
9366 | /* We get here if we can't use the current device name */ |
9367 | if (!pat) | |
9368 | goto out; | |
7892bd08 LR |
9369 | err = dev_get_valid_name(net, dev, pat); |
9370 | if (err < 0) | |
ce286d32 EB |
9371 | goto out; |
9372 | } | |
9373 | ||
9374 | /* | |
9375 | * And now a mini version of register_netdevice unregister_netdevice. | |
9376 | */ | |
9377 | ||
9378 | /* If device is running close it first. */ | |
9b772652 | 9379 | dev_close(dev); |
ce286d32 EB |
9380 | |
9381 | /* And unlink it from device chain */ | |
ce286d32 EB |
9382 | unlist_netdevice(dev); |
9383 | ||
9384 | synchronize_net(); | |
9385 | ||
9386 | /* Shutdown queueing discipline. */ | |
9387 | dev_shutdown(dev); | |
9388 | ||
9389 | /* Notify protocols, that we are about to destroy | |
eb13da1a | 9390 | * this device. They should clean all the things. |
9391 | * | |
9392 | * Note that dev->reg_state stays at NETREG_REGISTERED. | |
9393 | * This is wanted because this way 8021q and macvlan know | |
9394 | * the device is just moving and can keep their slaves up. | |
9395 | */ | |
ce286d32 | 9396 | call_netdevice_notifiers(NETDEV_UNREGISTER, dev); |
6549dd43 | 9397 | rcu_barrier(); |
38e01b30 | 9398 | |
c36ac8e2 | 9399 | new_nsid = peernet2id_alloc(dev_net(dev), net); |
38e01b30 ND |
9400 | /* If there is an ifindex conflict assign a new one */ |
9401 | if (__dev_get_by_index(net, dev->ifindex)) | |
9402 | new_ifindex = dev_new_index(net); | |
9403 | else | |
9404 | new_ifindex = dev->ifindex; | |
9405 | ||
9406 | rtmsg_ifinfo_newnet(RTM_DELLINK, dev, ~0U, GFP_KERNEL, &new_nsid, | |
9407 | new_ifindex); | |
ce286d32 EB |
9408 | |
9409 | /* | |
9410 | * Flush the unicast and multicast chains | |
9411 | */ | |
a748ee24 | 9412 | dev_uc_flush(dev); |
22bedad3 | 9413 | dev_mc_flush(dev); |
ce286d32 | 9414 | |
4e66ae2e SH |
9415 | /* Send a netdev-removed uevent to the old namespace */ |
9416 | kobject_uevent(&dev->dev.kobj, KOBJ_REMOVE); | |
4c75431a | 9417 | netdev_adjacent_del_links(dev); |
4e66ae2e | 9418 | |
ce286d32 | 9419 | /* Actually switch the network namespace */ |
c346dca1 | 9420 | dev_net_set(dev, net); |
38e01b30 | 9421 | dev->ifindex = new_ifindex; |
ce286d32 | 9422 | |
4e66ae2e SH |
9423 | /* Send a netdev-add uevent to the new namespace */ |
9424 | kobject_uevent(&dev->dev.kobj, KOBJ_ADD); | |
4c75431a | 9425 | netdev_adjacent_add_links(dev); |
4e66ae2e | 9426 | |
8b41d188 | 9427 | /* Fixup kobjects */ |
a1b3f594 | 9428 | err = device_rename(&dev->dev, dev->name); |
8b41d188 | 9429 | WARN_ON(err); |
ce286d32 EB |
9430 | |
9431 | /* Add the device back in the hashes */ | |
9432 | list_netdevice(dev); | |
9433 | ||
9434 | /* Notify protocols, that a new device appeared. */ | |
9435 | call_netdevice_notifiers(NETDEV_REGISTER, dev); | |
9436 | ||
d90a909e EB |
9437 | /* |
9438 | * Prevent userspace races by waiting until the network | |
9439 | * device is fully setup before sending notifications. | |
9440 | */ | |
7f294054 | 9441 | rtmsg_ifinfo(RTM_NEWLINK, dev, ~0U, GFP_KERNEL); |
d90a909e | 9442 | |
ce286d32 EB |
9443 | synchronize_net(); |
9444 | err = 0; | |
9445 | out: | |
9446 | return err; | |
9447 | } | |
463d0183 | 9448 | EXPORT_SYMBOL_GPL(dev_change_net_namespace); |
ce286d32 | 9449 | |
f0bf90de | 9450 | static int dev_cpu_dead(unsigned int oldcpu) |
1da177e4 LT |
9451 | { |
9452 | struct sk_buff **list_skb; | |
1da177e4 | 9453 | struct sk_buff *skb; |
f0bf90de | 9454 | unsigned int cpu; |
97d8b6e3 | 9455 | struct softnet_data *sd, *oldsd, *remsd = NULL; |
1da177e4 | 9456 | |
1da177e4 LT |
9457 | local_irq_disable(); |
9458 | cpu = smp_processor_id(); | |
9459 | sd = &per_cpu(softnet_data, cpu); | |
9460 | oldsd = &per_cpu(softnet_data, oldcpu); | |
9461 | ||
9462 | /* Find end of our completion_queue. */ | |
9463 | list_skb = &sd->completion_queue; | |
9464 | while (*list_skb) | |
9465 | list_skb = &(*list_skb)->next; | |
9466 | /* Append completion queue from offline CPU. */ | |
9467 | *list_skb = oldsd->completion_queue; | |
9468 | oldsd->completion_queue = NULL; | |
9469 | ||
1da177e4 | 9470 | /* Append output queue from offline CPU. */ |
a9cbd588 CG |
9471 | if (oldsd->output_queue) { |
9472 | *sd->output_queue_tailp = oldsd->output_queue; | |
9473 | sd->output_queue_tailp = oldsd->output_queue_tailp; | |
9474 | oldsd->output_queue = NULL; | |
9475 | oldsd->output_queue_tailp = &oldsd->output_queue; | |
9476 | } | |
ac64da0b ED |
9477 | /* Append NAPI poll list from offline CPU, with one exception : |
9478 | * process_backlog() must be called by cpu owning percpu backlog. | |
9479 | * We properly handle process_queue & input_pkt_queue later. | |
9480 | */ | |
9481 | while (!list_empty(&oldsd->poll_list)) { | |
9482 | struct napi_struct *napi = list_first_entry(&oldsd->poll_list, | |
9483 | struct napi_struct, | |
9484 | poll_list); | |
9485 | ||
9486 | list_del_init(&napi->poll_list); | |
9487 | if (napi->poll == process_backlog) | |
9488 | napi->state = 0; | |
9489 | else | |
9490 | ____napi_schedule(sd, napi); | |
264524d5 | 9491 | } |
1da177e4 LT |
9492 | |
9493 | raise_softirq_irqoff(NET_TX_SOFTIRQ); | |
9494 | local_irq_enable(); | |
9495 | ||
773fc8f6 | 9496 | #ifdef CONFIG_RPS |
9497 | remsd = oldsd->rps_ipi_list; | |
9498 | oldsd->rps_ipi_list = NULL; | |
9499 | #endif | |
9500 | /* send out pending IPI's on offline CPU */ | |
9501 | net_rps_send_ipi(remsd); | |
9502 | ||
1da177e4 | 9503 | /* Process offline CPU's input_pkt_queue */ |
76cc8b13 | 9504 | while ((skb = __skb_dequeue(&oldsd->process_queue))) { |
91e83133 | 9505 | netif_rx_ni(skb); |
76cc8b13 | 9506 | input_queue_head_incr(oldsd); |
fec5e652 | 9507 | } |
ac64da0b | 9508 | while ((skb = skb_dequeue(&oldsd->input_pkt_queue))) { |
91e83133 | 9509 | netif_rx_ni(skb); |
76cc8b13 TH |
9510 | input_queue_head_incr(oldsd); |
9511 | } | |
1da177e4 | 9512 | |
f0bf90de | 9513 | return 0; |
1da177e4 | 9514 | } |
1da177e4 | 9515 | |
7f353bf2 | 9516 | /** |
b63365a2 HX |
9517 | * netdev_increment_features - increment feature set by one |
9518 | * @all: current feature set | |
9519 | * @one: new feature set | |
9520 | * @mask: mask feature set | |
7f353bf2 HX |
9521 | * |
9522 | * Computes a new feature set after adding a device with feature set | |
b63365a2 HX |
9523 | * @one to the master device with current feature set @all. Will not |
9524 | * enable anything that is off in @mask. Returns the new feature set. | |
7f353bf2 | 9525 | */ |
c8f44aff MM |
9526 | netdev_features_t netdev_increment_features(netdev_features_t all, |
9527 | netdev_features_t one, netdev_features_t mask) | |
b63365a2 | 9528 | { |
c8cd0989 | 9529 | if (mask & NETIF_F_HW_CSUM) |
a188222b | 9530 | mask |= NETIF_F_CSUM_MASK; |
1742f183 | 9531 | mask |= NETIF_F_VLAN_CHALLENGED; |
7f353bf2 | 9532 | |
a188222b | 9533 | all |= one & (NETIF_F_ONE_FOR_ALL | NETIF_F_CSUM_MASK) & mask; |
1742f183 | 9534 | all &= one | ~NETIF_F_ALL_FOR_ALL; |
c6e1a0d1 | 9535 | |
1742f183 | 9536 | /* If one device supports hw checksumming, set for all. */ |
c8cd0989 TH |
9537 | if (all & NETIF_F_HW_CSUM) |
9538 | all &= ~(NETIF_F_CSUM_MASK & ~NETIF_F_HW_CSUM); | |
7f353bf2 HX |
9539 | |
9540 | return all; | |
9541 | } | |
b63365a2 | 9542 | EXPORT_SYMBOL(netdev_increment_features); |
7f353bf2 | 9543 | |
430f03cd | 9544 | static struct hlist_head * __net_init netdev_create_hash(void) |
30d97d35 PE |
9545 | { |
9546 | int i; | |
9547 | struct hlist_head *hash; | |
9548 | ||
6da2ec56 | 9549 | hash = kmalloc_array(NETDEV_HASHENTRIES, sizeof(*hash), GFP_KERNEL); |
30d97d35 PE |
9550 | if (hash != NULL) |
9551 | for (i = 0; i < NETDEV_HASHENTRIES; i++) | |
9552 | INIT_HLIST_HEAD(&hash[i]); | |
9553 | ||
9554 | return hash; | |
9555 | } | |
9556 | ||
881d966b | 9557 | /* Initialize per network namespace state */ |
4665079c | 9558 | static int __net_init netdev_init(struct net *net) |
881d966b | 9559 | { |
d9f37d01 | 9560 | BUILD_BUG_ON(GRO_HASH_BUCKETS > |
ccdb5171 | 9561 | 8 * FIELD_SIZEOF(struct napi_struct, gro_bitmask)); |
d9f37d01 | 9562 | |
734b6541 RM |
9563 | if (net != &init_net) |
9564 | INIT_LIST_HEAD(&net->dev_base_head); | |
881d966b | 9565 | |
30d97d35 PE |
9566 | net->dev_name_head = netdev_create_hash(); |
9567 | if (net->dev_name_head == NULL) | |
9568 | goto err_name; | |
881d966b | 9569 | |
30d97d35 PE |
9570 | net->dev_index_head = netdev_create_hash(); |
9571 | if (net->dev_index_head == NULL) | |
9572 | goto err_idx; | |
881d966b EB |
9573 | |
9574 | return 0; | |
30d97d35 PE |
9575 | |
9576 | err_idx: | |
9577 | kfree(net->dev_name_head); | |
9578 | err_name: | |
9579 | return -ENOMEM; | |
881d966b EB |
9580 | } |
9581 | ||
f0db275a SH |
9582 | /** |
9583 | * netdev_drivername - network driver for the device | |
9584 | * @dev: network device | |
f0db275a SH |
9585 | * |
9586 | * Determine network driver for device. | |
9587 | */ | |
3019de12 | 9588 | const char *netdev_drivername(const struct net_device *dev) |
6579e57b | 9589 | { |
cf04a4c7 SH |
9590 | const struct device_driver *driver; |
9591 | const struct device *parent; | |
3019de12 | 9592 | const char *empty = ""; |
6579e57b AV |
9593 | |
9594 | parent = dev->dev.parent; | |
6579e57b | 9595 | if (!parent) |
3019de12 | 9596 | return empty; |
6579e57b AV |
9597 | |
9598 | driver = parent->driver; | |
9599 | if (driver && driver->name) | |
3019de12 DM |
9600 | return driver->name; |
9601 | return empty; | |
6579e57b AV |
9602 | } |
9603 | ||
6ea754eb JP |
9604 | static void __netdev_printk(const char *level, const struct net_device *dev, |
9605 | struct va_format *vaf) | |
256df2f3 | 9606 | { |
b004ff49 | 9607 | if (dev && dev->dev.parent) { |
6ea754eb JP |
9608 | dev_printk_emit(level[1] - '0', |
9609 | dev->dev.parent, | |
9610 | "%s %s %s%s: %pV", | |
9611 | dev_driver_string(dev->dev.parent), | |
9612 | dev_name(dev->dev.parent), | |
9613 | netdev_name(dev), netdev_reg_state(dev), | |
9614 | vaf); | |
b004ff49 | 9615 | } else if (dev) { |
6ea754eb JP |
9616 | printk("%s%s%s: %pV", |
9617 | level, netdev_name(dev), netdev_reg_state(dev), vaf); | |
b004ff49 | 9618 | } else { |
6ea754eb | 9619 | printk("%s(NULL net_device): %pV", level, vaf); |
b004ff49 | 9620 | } |
256df2f3 JP |
9621 | } |
9622 | ||
6ea754eb JP |
9623 | void netdev_printk(const char *level, const struct net_device *dev, |
9624 | const char *format, ...) | |
256df2f3 JP |
9625 | { |
9626 | struct va_format vaf; | |
9627 | va_list args; | |
256df2f3 JP |
9628 | |
9629 | va_start(args, format); | |
9630 | ||
9631 | vaf.fmt = format; | |
9632 | vaf.va = &args; | |
9633 | ||
6ea754eb | 9634 | __netdev_printk(level, dev, &vaf); |
b004ff49 | 9635 | |
256df2f3 | 9636 | va_end(args); |
256df2f3 JP |
9637 | } |
9638 | EXPORT_SYMBOL(netdev_printk); | |
9639 | ||
9640 | #define define_netdev_printk_level(func, level) \ | |
6ea754eb | 9641 | void func(const struct net_device *dev, const char *fmt, ...) \ |
256df2f3 | 9642 | { \ |
256df2f3 JP |
9643 | struct va_format vaf; \ |
9644 | va_list args; \ | |
9645 | \ | |
9646 | va_start(args, fmt); \ | |
9647 | \ | |
9648 | vaf.fmt = fmt; \ | |
9649 | vaf.va = &args; \ | |
9650 | \ | |
6ea754eb | 9651 | __netdev_printk(level, dev, &vaf); \ |
b004ff49 | 9652 | \ |
256df2f3 | 9653 | va_end(args); \ |
256df2f3 JP |
9654 | } \ |
9655 | EXPORT_SYMBOL(func); | |
9656 | ||
9657 | define_netdev_printk_level(netdev_emerg, KERN_EMERG); | |
9658 | define_netdev_printk_level(netdev_alert, KERN_ALERT); | |
9659 | define_netdev_printk_level(netdev_crit, KERN_CRIT); | |
9660 | define_netdev_printk_level(netdev_err, KERN_ERR); | |
9661 | define_netdev_printk_level(netdev_warn, KERN_WARNING); | |
9662 | define_netdev_printk_level(netdev_notice, KERN_NOTICE); | |
9663 | define_netdev_printk_level(netdev_info, KERN_INFO); | |
9664 | ||
4665079c | 9665 | static void __net_exit netdev_exit(struct net *net) |
881d966b EB |
9666 | { |
9667 | kfree(net->dev_name_head); | |
9668 | kfree(net->dev_index_head); | |
ee21b18b VA |
9669 | if (net != &init_net) |
9670 | WARN_ON_ONCE(!list_empty(&net->dev_base_head)); | |
881d966b EB |
9671 | } |
9672 | ||
022cbae6 | 9673 | static struct pernet_operations __net_initdata netdev_net_ops = { |
881d966b EB |
9674 | .init = netdev_init, |
9675 | .exit = netdev_exit, | |
9676 | }; | |
9677 | ||
4665079c | 9678 | static void __net_exit default_device_exit(struct net *net) |
ce286d32 | 9679 | { |
e008b5fc | 9680 | struct net_device *dev, *aux; |
ce286d32 | 9681 | /* |
e008b5fc | 9682 | * Push all migratable network devices back to the |
ce286d32 EB |
9683 | * initial network namespace |
9684 | */ | |
9685 | rtnl_lock(); | |
e008b5fc | 9686 | for_each_netdev_safe(net, dev, aux) { |
ce286d32 | 9687 | int err; |
aca51397 | 9688 | char fb_name[IFNAMSIZ]; |
ce286d32 EB |
9689 | |
9690 | /* Ignore unmoveable devices (i.e. loopback) */ | |
9691 | if (dev->features & NETIF_F_NETNS_LOCAL) | |
9692 | continue; | |
9693 | ||
e008b5fc EB |
9694 | /* Leave virtual devices for the generic cleanup */ |
9695 | if (dev->rtnl_link_ops) | |
9696 | continue; | |
d0c082ce | 9697 | |
25985edc | 9698 | /* Push remaining network devices to init_net */ |
aca51397 PE |
9699 | snprintf(fb_name, IFNAMSIZ, "dev%d", dev->ifindex); |
9700 | err = dev_change_net_namespace(dev, &init_net, fb_name); | |
ce286d32 | 9701 | if (err) { |
7b6cd1ce JP |
9702 | pr_emerg("%s: failed to move %s to init_net: %d\n", |
9703 | __func__, dev->name, err); | |
aca51397 | 9704 | BUG(); |
ce286d32 EB |
9705 | } |
9706 | } | |
9707 | rtnl_unlock(); | |
9708 | } | |
9709 | ||
50624c93 EB |
9710 | static void __net_exit rtnl_lock_unregistering(struct list_head *net_list) |
9711 | { | |
9712 | /* Return with the rtnl_lock held when there are no network | |
9713 | * devices unregistering in any network namespace in net_list. | |
9714 | */ | |
9715 | struct net *net; | |
9716 | bool unregistering; | |
ff960a73 | 9717 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
50624c93 | 9718 | |
ff960a73 | 9719 | add_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 | 9720 | for (;;) { |
50624c93 EB |
9721 | unregistering = false; |
9722 | rtnl_lock(); | |
9723 | list_for_each_entry(net, net_list, exit_list) { | |
9724 | if (net->dev_unreg_count > 0) { | |
9725 | unregistering = true; | |
9726 | break; | |
9727 | } | |
9728 | } | |
9729 | if (!unregistering) | |
9730 | break; | |
9731 | __rtnl_unlock(); | |
ff960a73 PZ |
9732 | |
9733 | wait_woken(&wait, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
50624c93 | 9734 | } |
ff960a73 | 9735 | remove_wait_queue(&netdev_unregistering_wq, &wait); |
50624c93 EB |
9736 | } |
9737 | ||
04dc7f6b EB |
9738 | static void __net_exit default_device_exit_batch(struct list_head *net_list) |
9739 | { | |
9740 | /* At exit all network devices most be removed from a network | |
b595076a | 9741 | * namespace. Do this in the reverse order of registration. |
04dc7f6b EB |
9742 | * Do this across as many network namespaces as possible to |
9743 | * improve batching efficiency. | |
9744 | */ | |
9745 | struct net_device *dev; | |
9746 | struct net *net; | |
9747 | LIST_HEAD(dev_kill_list); | |
9748 | ||
50624c93 EB |
9749 | /* To prevent network device cleanup code from dereferencing |
9750 | * loopback devices or network devices that have been freed | |
9751 | * wait here for all pending unregistrations to complete, | |
9752 | * before unregistring the loopback device and allowing the | |
9753 | * network namespace be freed. | |
9754 | * | |
9755 | * The netdev todo list containing all network devices | |
9756 | * unregistrations that happen in default_device_exit_batch | |
9757 | * will run in the rtnl_unlock() at the end of | |
9758 | * default_device_exit_batch. | |
9759 | */ | |
9760 | rtnl_lock_unregistering(net_list); | |
04dc7f6b EB |
9761 | list_for_each_entry(net, net_list, exit_list) { |
9762 | for_each_netdev_reverse(net, dev) { | |
b0ab2fab | 9763 | if (dev->rtnl_link_ops && dev->rtnl_link_ops->dellink) |
04dc7f6b EB |
9764 | dev->rtnl_link_ops->dellink(dev, &dev_kill_list); |
9765 | else | |
9766 | unregister_netdevice_queue(dev, &dev_kill_list); | |
9767 | } | |
9768 | } | |
9769 | unregister_netdevice_many(&dev_kill_list); | |
9770 | rtnl_unlock(); | |
9771 | } | |
9772 | ||
022cbae6 | 9773 | static struct pernet_operations __net_initdata default_device_ops = { |
ce286d32 | 9774 | .exit = default_device_exit, |
04dc7f6b | 9775 | .exit_batch = default_device_exit_batch, |
ce286d32 EB |
9776 | }; |
9777 | ||
1da177e4 LT |
9778 | /* |
9779 | * Initialize the DEV module. At boot time this walks the device list and | |
9780 | * unhooks any devices that fail to initialise (normally hardware not | |
9781 | * present) and leaves us with a valid list of present and active devices. | |
9782 | * | |
9783 | */ | |
9784 | ||
9785 | /* | |
9786 | * This is called single threaded during boot, so no need | |
9787 | * to take the rtnl semaphore. | |
9788 | */ | |
9789 | static int __init net_dev_init(void) | |
9790 | { | |
9791 | int i, rc = -ENOMEM; | |
9792 | ||
9793 | BUG_ON(!dev_boot_phase); | |
9794 | ||
1da177e4 LT |
9795 | if (dev_proc_init()) |
9796 | goto out; | |
9797 | ||
8b41d188 | 9798 | if (netdev_kobject_init()) |
1da177e4 LT |
9799 | goto out; |
9800 | ||
9801 | INIT_LIST_HEAD(&ptype_all); | |
82d8a867 | 9802 | for (i = 0; i < PTYPE_HASH_SIZE; i++) |
1da177e4 LT |
9803 | INIT_LIST_HEAD(&ptype_base[i]); |
9804 | ||
62532da9 VY |
9805 | INIT_LIST_HEAD(&offload_base); |
9806 | ||
881d966b EB |
9807 | if (register_pernet_subsys(&netdev_net_ops)) |
9808 | goto out; | |
1da177e4 LT |
9809 | |
9810 | /* | |
9811 | * Initialise the packet receive queues. | |
9812 | */ | |
9813 | ||
6f912042 | 9814 | for_each_possible_cpu(i) { |
41852497 | 9815 | struct work_struct *flush = per_cpu_ptr(&flush_works, i); |
e36fa2f7 | 9816 | struct softnet_data *sd = &per_cpu(softnet_data, i); |
1da177e4 | 9817 | |
41852497 ED |
9818 | INIT_WORK(flush, flush_backlog); |
9819 | ||
e36fa2f7 | 9820 | skb_queue_head_init(&sd->input_pkt_queue); |
6e7676c1 | 9821 | skb_queue_head_init(&sd->process_queue); |
f53c7239 SK |
9822 | #ifdef CONFIG_XFRM_OFFLOAD |
9823 | skb_queue_head_init(&sd->xfrm_backlog); | |
9824 | #endif | |
e36fa2f7 | 9825 | INIT_LIST_HEAD(&sd->poll_list); |
a9cbd588 | 9826 | sd->output_queue_tailp = &sd->output_queue; |
df334545 | 9827 | #ifdef CONFIG_RPS |
e36fa2f7 ED |
9828 | sd->csd.func = rps_trigger_softirq; |
9829 | sd->csd.info = sd; | |
e36fa2f7 | 9830 | sd->cpu = i; |
1e94d72f | 9831 | #endif |
0a9627f2 | 9832 | |
7c4ec749 | 9833 | init_gro_hash(&sd->backlog); |
e36fa2f7 ED |
9834 | sd->backlog.poll = process_backlog; |
9835 | sd->backlog.weight = weight_p; | |
1da177e4 LT |
9836 | } |
9837 | ||
1da177e4 LT |
9838 | dev_boot_phase = 0; |
9839 | ||
505d4f73 EB |
9840 | /* The loopback device is special if any other network devices |
9841 | * is present in a network namespace the loopback device must | |
9842 | * be present. Since we now dynamically allocate and free the | |
9843 | * loopback device ensure this invariant is maintained by | |
9844 | * keeping the loopback device as the first device on the | |
9845 | * list of network devices. Ensuring the loopback devices | |
9846 | * is the first device that appears and the last network device | |
9847 | * that disappears. | |
9848 | */ | |
9849 | if (register_pernet_device(&loopback_net_ops)) | |
9850 | goto out; | |
9851 | ||
9852 | if (register_pernet_device(&default_device_ops)) | |
9853 | goto out; | |
9854 | ||
962cf36c CM |
9855 | open_softirq(NET_TX_SOFTIRQ, net_tx_action); |
9856 | open_softirq(NET_RX_SOFTIRQ, net_rx_action); | |
1da177e4 | 9857 | |
f0bf90de SAS |
9858 | rc = cpuhp_setup_state_nocalls(CPUHP_NET_DEV_DEAD, "net/dev:dead", |
9859 | NULL, dev_cpu_dead); | |
9860 | WARN_ON(rc < 0); | |
1da177e4 LT |
9861 | rc = 0; |
9862 | out: | |
9863 | return rc; | |
9864 | } | |
9865 | ||
9866 | subsys_initcall(net_dev_init); |