blob: af399cac5205402480300ae7b7d6fbb2094095fb [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PACKET - implements raw packet sockets.
7 *
Jesper Juhl02c30a82005-05-05 16:16:16 -07008 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 *
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +090012 * Fixes:
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +090035 * Ulises Alonso : Frame number limit removal and
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 * packet_set_ring memory leak.
Eric W. Biederman0fb375f2005-09-21 00:11:37 -070037 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +090040 * byte arrays at the end of sockaddr_ll
Eric W. Biederman0fb375f2005-09-21 00:11:37 -070041 * and packet_mreq.
Johann Baudy69e3c752009-05-18 22:11:22 -070042 * Johann Baudy : Added TX RING.
chetan lokef6fb8f12011-08-19 10:18:16 +000043 * Chetan Loke : Implemented TPACKET_V3 block abstraction
44 * layer.
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
46 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070047 *
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
52 *
53 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +090054
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/types.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#include <linux/mm.h>
Randy Dunlap4fc268d2006-01-11 12:17:47 -080057#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <linux/fcntl.h>
59#include <linux/socket.h>
60#include <linux/in.h>
61#include <linux/inet.h>
62#include <linux/netdevice.h>
63#include <linux/if_packet.h>
64#include <linux/wireless.h>
Herbert Xuffbc6112007-02-04 23:33:10 -080065#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#include <linux/kmod.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090067#include <linux/slab.h>
Neil Horman0e3125c2010-11-16 10:26:47 -080068#include <linux/vmalloc.h>
Eric W. Biederman457c4cb2007-09-12 12:01:34 +020069#include <net/net_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <net/ip.h>
71#include <net/protocol.h>
72#include <linux/skbuff.h>
73#include <net/sock.h>
74#include <linux/errno.h>
75#include <linux/timer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#include <asm/uaccess.h>
77#include <asm/ioctls.h>
78#include <asm/page.h>
Al Viroa1f8e7f72006-10-19 16:08:53 -040079#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070080#include <asm/io.h>
81#include <linux/proc_fs.h>
82#include <linux/seq_file.h>
83#include <linux/poll.h>
84#include <linux/module.h>
85#include <linux/init.h>
Herbert Xu905db442009-01-30 14:12:06 -080086#include <linux/mutex.h>
Eric Dumazet05423b22009-10-26 18:40:35 -070087#include <linux/if_vlan.h>
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -080088#include <linux/virtio_net.h>
Richard Cochraned85b562010-04-07 22:41:28 +000089#include <linux/errqueue.h>
Scott McMillan614f60f2010-06-02 05:53:56 -070090#include <linux/net_tstamp.h>
Daniel Borkmannb0138402014-01-15 16:25:36 +010091#include <linux/percpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070092#ifdef CONFIG_INET
93#include <net/inet_common.h>
94#endif
Willem de Bruijn47dceb82015-08-14 22:31:34 -040095#include <linux/bpf.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
Pavel Emelyanov2787b042012-08-13 05:49:39 +000097#include "internal.h"
98
Linus Torvalds1da177e2005-04-16 15:20:36 -070099/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 Assumptions:
101 - if device has no dev->hard_header routine, it adds and removes ll header
102 inside itself. In this case ll header is invisible outside of device,
103 but higher levels still should reserve dev->hard_header_len.
104 Some devices are enough clever to reallocate skb, when header
105 will not fit to reserved space (tunnel), another ones are silly
106 (PPP).
107 - packet socket receives packets with pulled ll header,
108 so that SOCK_RAW should push it back.
109
110On receive:
111-----------
112
113Incoming, dev->hard_header!=NULL
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -0700114 mac_header -> ll header
115 data -> data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116
117Outgoing, dev->hard_header!=NULL
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -0700118 mac_header -> ll header
119 data -> ll header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121Incoming, dev->hard_header==NULL
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -0700122 mac_header -> UNKNOWN position. It is very likely, that it points to ll
123 header. PPP makes it, that is wrong, because introduce
YOSHIFUJI Hideakidb0c58f2007-07-19 10:44:35 +0900124 assymetry between rx and tx paths.
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -0700125 data -> data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127Outgoing, dev->hard_header==NULL
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -0700128 mac_header -> data. ll header is still not built!
129 data -> data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
131Resume
132 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
133
134
135On transmit:
136------------
137
138dev->hard_header != NULL
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -0700139 mac_header -> ll header
140 data -> ll header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141
142dev->hard_header == NULL (ll header is added by device, we cannot control it)
Arnaldo Carvalho de Melob0e380b12007-04-10 21:21:55 -0700143 mac_header -> data
144 data -> data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
146 We should set nh.raw on output to correct posistion,
147 packet classifier depends on it.
148 */
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150/* Private packet socket structures. */
151
Eric W. Biederman0fb375f2005-09-21 00:11:37 -0700152/* identical to struct packet_mreq except it has
153 * a longer address field.
154 */
Eric Dumazet40d4e3d2009-07-21 21:57:59 +0000155struct packet_mreq_max {
Eric W. Biederman0fb375f2005-09-21 00:11:37 -0700156 int mr_ifindex;
157 unsigned short mr_type;
158 unsigned short mr_alen;
159 unsigned char mr_address[MAX_ADDR_LEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160};
David S. Millera2efcfa2007-05-29 13:12:50 -0700161
Daniel Borkmann184f4892013-04-16 01:57:46 +0000162union tpacket_uhdr {
163 struct tpacket_hdr *h1;
164 struct tpacket2_hdr *h2;
165 struct tpacket3_hdr *h3;
166 void *raw;
167};
168
chetan lokef6fb8f12011-08-19 10:18:16 +0000169static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
Johann Baudy69e3c752009-05-18 22:11:22 -0700170 int closing, int tx_ring);
171
chetan lokef6fb8f12011-08-19 10:18:16 +0000172#define V3_ALIGNMENT (8)
173
chetan lokebc59ba32011-08-25 10:43:30 +0000174#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
chetan lokef6fb8f12011-08-19 10:18:16 +0000175
176#define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
chetan lokef6fb8f12011-08-19 10:18:16 +0000179#define PGV_FROM_VMALLOC 1
Johann Baudy69e3c752009-05-18 22:11:22 -0700180
chetan lokef6fb8f12011-08-19 10:18:16 +0000181#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
183#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
184#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
185#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
186#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187#define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
188
Johann Baudy69e3c752009-05-18 22:11:22 -0700189struct packet_sock;
190static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +0000191static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
chetan lokef6fb8f12011-08-19 10:18:16 +0000194static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
196 int status);
197static void packet_increment_head(struct packet_ring_buffer *buff);
chetan lokebc59ba32011-08-25 10:43:30 +0000198static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
199 struct tpacket_block_desc *);
200static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
chetan lokef6fb8f12011-08-19 10:18:16 +0000201 struct packet_sock *);
chetan lokebc59ba32011-08-25 10:43:30 +0000202static void prb_retire_current_block(struct tpacket_kbdq_core *,
chetan lokef6fb8f12011-08-19 10:18:16 +0000203 struct packet_sock *, unsigned int status);
chetan lokebc59ba32011-08-25 10:43:30 +0000204static int prb_queue_frozen(struct tpacket_kbdq_core *);
205static void prb_open_block(struct tpacket_kbdq_core *,
206 struct tpacket_block_desc *);
chetan lokef6fb8f12011-08-19 10:18:16 +0000207static void prb_retire_rx_blk_timer_expired(unsigned long);
chetan lokebc59ba32011-08-25 10:43:30 +0000208static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
209static void prb_init_blk_timer(struct packet_sock *,
210 struct tpacket_kbdq_core *,
211 void (*func) (unsigned long));
212static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
213static void prb_clear_rxhash(struct tpacket_kbdq_core *,
214 struct tpacket3_hdr *);
215static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
216 struct tpacket3_hdr *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217static void packet_flush_mclist(struct sock *sk);
218
Herbert Xuffbc6112007-02-04 23:33:10 -0800219struct packet_skb_cb {
Herbert Xuffbc6112007-02-04 23:33:10 -0800220 union {
221 struct sockaddr_pkt pkt;
Eyal Birger2472d762015-03-01 14:58:28 +0200222 union {
223 /* Trick: alias skb original length with
224 * ll.sll_family and ll.protocol in order
225 * to save room.
226 */
227 unsigned int origlen;
228 struct sockaddr_ll ll;
229 };
Herbert Xuffbc6112007-02-04 23:33:10 -0800230 } sa;
231};
232
David Woodhoused3869ef2015-09-23 19:45:08 +0100233#define vio_le() virtio_legacy_is_little_endian()
234
Herbert Xuffbc6112007-02-04 23:33:10 -0800235#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
Herbert Xu8dc41942007-02-04 23:31:32 -0800236
chetan lokebc59ba32011-08-25 10:43:30 +0000237#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
chetan lokef6fb8f12011-08-19 10:18:16 +0000238#define GET_PBLOCK_DESC(x, bid) \
chetan lokebc59ba32011-08-25 10:43:30 +0000239 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
chetan lokef6fb8f12011-08-19 10:18:16 +0000240#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
chetan lokebc59ba32011-08-25 10:43:30 +0000241 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
chetan lokef6fb8f12011-08-19 10:18:16 +0000242#define GET_NEXT_PRB_BLK_NUM(x) \
243 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244 ((x)->kactive_blk_num+1) : 0)
245
David S. Millerdc99f602011-07-05 01:45:05 -0700246static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
247static void __fanout_link(struct sock *sk, struct packet_sock *po);
248
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100249static int packet_direct_xmit(struct sk_buff *skb)
250{
251 struct net_device *dev = skb->dev;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100252 netdev_features_t features;
253 struct netdev_queue *txq;
Daniel Borkmann43279502014-03-27 16:38:30 +0100254 int ret = NETDEV_TX_BUSY;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100255
256 if (unlikely(!netif_running(dev) ||
Daniel Borkmann43279502014-03-27 16:38:30 +0100257 !netif_carrier_ok(dev)))
258 goto drop;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100259
260 features = netif_skb_features(skb);
261 if (skb_needs_linearize(skb, features) &&
Daniel Borkmann43279502014-03-27 16:38:30 +0100262 __skb_linearize(skb))
263 goto drop;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100264
Daniel Borkmann10c51b56232014-08-27 11:11:27 +0200265 txq = skb_get_tx_queue(dev, skb);
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100266
Daniel Borkmann43279502014-03-27 16:38:30 +0100267 local_bh_disable();
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100268
Daniel Borkmann43279502014-03-27 16:38:30 +0100269 HARD_TX_LOCK(dev, txq, smp_processor_id());
David S. Miller10b3ad82014-08-29 21:07:24 -0700270 if (!netif_xmit_frozen_or_drv_stopped(txq))
David S. Millerfa2dbdc22014-08-29 21:55:22 -0700271 ret = netdev_start_xmit(skb, dev, txq, false);
Daniel Borkmann43279502014-03-27 16:38:30 +0100272 HARD_TX_UNLOCK(dev, txq);
273
274 local_bh_enable();
275
276 if (!dev_xmit_complete(ret))
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100277 kfree_skb(skb);
Daniel Borkmann43279502014-03-27 16:38:30 +0100278
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100279 return ret;
Daniel Borkmann43279502014-03-27 16:38:30 +0100280drop:
Daniel Borkmann0f97ede2014-04-02 20:52:56 +0200281 atomic_long_inc(&dev->tx_dropped);
Daniel Borkmann43279502014-03-27 16:38:30 +0100282 kfree_skb(skb);
283 return NET_XMIT_DROP;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100284}
285
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100286static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287{
288 struct net_device *dev;
289
290 rcu_read_lock();
291 dev = rcu_dereference(po->cached_dev);
292 if (likely(dev))
293 dev_hold(dev);
294 rcu_read_unlock();
295
296 return dev;
297}
298
299static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
301{
302 rcu_assign_pointer(po->cached_dev, dev);
303}
304
305static void packet_cached_dev_reset(struct packet_sock *po)
306{
307 RCU_INIT_POINTER(po->cached_dev, NULL);
308}
309
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100310static bool packet_use_direct_xmit(const struct packet_sock *po)
311{
312 return po->xmit == packet_direct_xmit;
313}
314
Daniel Borkmann0fd5d572014-02-16 15:55:22 +0100315static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100316{
Li Zhong1cbac012013-12-12 22:39:55 +0100317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100318}
319
Daniel Borkmann0fd5d572014-02-16 15:55:22 +0100320static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
321{
322 const struct net_device_ops *ops = dev->netdev_ops;
323 u16 queue_index;
324
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
329 } else {
330 queue_index = __packet_pick_tx_queue(dev, skb);
331 }
332
333 skb_set_queue_mapping(skb, queue_index);
334}
335
David S. Millerce06b032011-07-04 01:44:29 -0700336/* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
339 */
340static void register_prot_hook(struct sock *sk)
341{
342 struct packet_sock *po = pkt_sk(sk);
Daniel Borkmanne40526c2013-11-21 16:50:58 +0100343
David S. Millerce06b032011-07-04 01:44:29 -0700344 if (!po->running) {
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100345 if (po->fanout)
David S. Millerdc99f602011-07-05 01:45:05 -0700346 __fanout_link(sk, po);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100347 else
David S. Millerdc99f602011-07-05 01:45:05 -0700348 dev_add_pack(&po->prot_hook);
Daniel Borkmanne40526c2013-11-21 16:50:58 +0100349
David S. Millerce06b032011-07-04 01:44:29 -0700350 sock_hold(sk);
351 po->running = 1;
352 }
353}
354
355/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
361 */
362static void __unregister_prot_hook(struct sock *sk, bool sync)
363{
364 struct packet_sock *po = pkt_sk(sk);
365
366 po->running = 0;
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100367
368 if (po->fanout)
David S. Millerdc99f602011-07-05 01:45:05 -0700369 __fanout_unlink(sk, po);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100370 else
David S. Millerdc99f602011-07-05 01:45:05 -0700371 __dev_remove_pack(&po->prot_hook);
Daniel Borkmanne40526c2013-11-21 16:50:58 +0100372
David S. Millerce06b032011-07-04 01:44:29 -0700373 __sock_put(sk);
374
375 if (sync) {
376 spin_unlock(&po->bind_lock);
377 synchronize_net();
378 spin_lock(&po->bind_lock);
379 }
380}
381
382static void unregister_prot_hook(struct sock *sk, bool sync)
383{
384 struct packet_sock *po = pkt_sk(sk);
385
386 if (po->running)
387 __unregister_prot_hook(sk, sync);
388}
389
Michael S. Tsirkin6e580402014-11-24 13:32:16 +0200390static inline struct page * __pure pgv_to_page(void *addr)
Changli Gao0af55bb2010-12-01 02:52:20 +0000391{
392 if (is_vmalloc_addr(addr))
393 return vmalloc_to_page(addr);
394 return virt_to_page(addr);
395}
396
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700397static void __packet_set_status(struct packet_sock *po, void *frame, int status)
398{
Daniel Borkmann184f4892013-04-16 01:57:46 +0000399 union tpacket_uhdr h;
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700400
401 h.raw = frame;
402 switch (po->tp_version) {
403 case TPACKET_V1:
404 h.h1->tp_status = status;
Changli Gao0af55bb2010-12-01 02:52:20 +0000405 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700406 break;
407 case TPACKET_V2:
408 h.h2->tp_status = status;
Changli Gao0af55bb2010-12-01 02:52:20 +0000409 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700410 break;
chetan lokef6fb8f12011-08-19 10:18:16 +0000411 case TPACKET_V3:
Johann Baudy69e3c752009-05-18 22:11:22 -0700412 default:
chetan lokef6fb8f12011-08-19 10:18:16 +0000413 WARN(1, "TPACKET version not supported.\n");
Johann Baudy69e3c752009-05-18 22:11:22 -0700414 BUG();
415 }
416
417 smp_wmb();
418}
419
420static int __packet_get_status(struct packet_sock *po, void *frame)
421{
Daniel Borkmann184f4892013-04-16 01:57:46 +0000422 union tpacket_uhdr h;
Johann Baudy69e3c752009-05-18 22:11:22 -0700423
424 smp_rmb();
425
426 h.raw = frame;
427 switch (po->tp_version) {
428 case TPACKET_V1:
Changli Gao0af55bb2010-12-01 02:52:20 +0000429 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
Johann Baudy69e3c752009-05-18 22:11:22 -0700430 return h.h1->tp_status;
431 case TPACKET_V2:
Changli Gao0af55bb2010-12-01 02:52:20 +0000432 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
Johann Baudy69e3c752009-05-18 22:11:22 -0700433 return h.h2->tp_status;
chetan lokef6fb8f12011-08-19 10:18:16 +0000434 case TPACKET_V3:
Johann Baudy69e3c752009-05-18 22:11:22 -0700435 default:
chetan lokef6fb8f12011-08-19 10:18:16 +0000436 WARN(1, "TPACKET version not supported.\n");
Johann Baudy69e3c752009-05-18 22:11:22 -0700437 BUG();
438 return 0;
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440}
Johann Baudy69e3c752009-05-18 22:11:22 -0700441
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000442static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
443 unsigned int flags)
Daniel Borkmann7a513842013-04-23 00:39:29 +0000444{
445 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
446
Willem de Bruijn68a360e2014-07-25 18:01:31 -0400447 if (shhwtstamps &&
448 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450 return TP_STATUS_TS_RAW_HARDWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000451
452 if (ktime_to_timespec_cond(skb->tstamp, ts))
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000453 return TP_STATUS_TS_SOFTWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000454
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000455 return 0;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000456}
457
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000458static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
459 struct sk_buff *skb)
Willem de Bruijn2e313962013-04-23 00:39:28 +0000460{
461 union tpacket_uhdr h;
462 struct timespec ts;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000463 __u32 ts_status;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000464
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000465 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
466 return 0;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000467
468 h.raw = frame;
469 switch (po->tp_version) {
470 case TPACKET_V1:
471 h.h1->tp_sec = ts.tv_sec;
472 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
473 break;
474 case TPACKET_V2:
475 h.h2->tp_sec = ts.tv_sec;
476 h.h2->tp_nsec = ts.tv_nsec;
477 break;
478 case TPACKET_V3:
479 default:
480 WARN(1, "TPACKET version not supported.\n");
481 BUG();
482 }
483
484 /* one flush is safe, as both fields always lie on the same cacheline */
485 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
486 smp_wmb();
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000487
488 return ts_status;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000489}
490
Johann Baudy69e3c752009-05-18 22:11:22 -0700491static void *packet_lookup_frame(struct packet_sock *po,
492 struct packet_ring_buffer *rb,
493 unsigned int position,
494 int status)
495{
496 unsigned int pg_vec_pos, frame_offset;
Daniel Borkmann184f4892013-04-16 01:57:46 +0000497 union tpacket_uhdr h;
Johann Baudy69e3c752009-05-18 22:11:22 -0700498
499 pg_vec_pos = position / rb->frames_per_block;
500 frame_offset = position % rb->frames_per_block;
501
Neil Horman0e3125c2010-11-16 10:26:47 -0800502 h.raw = rb->pg_vec[pg_vec_pos].buffer +
503 (frame_offset * rb->frame_size);
Johann Baudy69e3c752009-05-18 22:11:22 -0700504
505 if (status != __packet_get_status(po, h.raw))
506 return NULL;
507
508 return h.raw;
509}
510
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000511static void *packet_current_frame(struct packet_sock *po,
Johann Baudy69e3c752009-05-18 22:11:22 -0700512 struct packet_ring_buffer *rb,
513 int status)
514{
515 return packet_lookup_frame(po, rb, rb->head, status);
516}
517
chetan lokebc59ba32011-08-25 10:43:30 +0000518static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f12011-08-19 10:18:16 +0000519{
520 del_timer_sync(&pkc->retire_blk_timer);
521}
522
523static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
chetan lokef6fb8f12011-08-19 10:18:16 +0000524 struct sk_buff_head *rb_queue)
525{
chetan lokebc59ba32011-08-25 10:43:30 +0000526 struct tpacket_kbdq_core *pkc;
chetan lokef6fb8f12011-08-19 10:18:16 +0000527
Tobias Klauser73d0fcf2015-07-28 14:21:26 +0200528 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f12011-08-19 10:18:16 +0000529
Veaceslav Falicoec6f8092013-11-29 09:53:23 +0100530 spin_lock_bh(&rb_queue->lock);
chetan lokef6fb8f12011-08-19 10:18:16 +0000531 pkc->delete_blk_timer = 1;
Veaceslav Falicoec6f8092013-11-29 09:53:23 +0100532 spin_unlock_bh(&rb_queue->lock);
chetan lokef6fb8f12011-08-19 10:18:16 +0000533
534 prb_del_retire_blk_timer(pkc);
535}
536
537static void prb_init_blk_timer(struct packet_sock *po,
chetan lokebc59ba32011-08-25 10:43:30 +0000538 struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f12011-08-19 10:18:16 +0000539 void (*func) (unsigned long))
540{
541 init_timer(&pkc->retire_blk_timer);
542 pkc->retire_blk_timer.data = (long)po;
543 pkc->retire_blk_timer.function = func;
544 pkc->retire_blk_timer.expires = jiffies;
545}
546
Maninder Singhe8e85cc2015-06-22 12:39:16 +0530547static void prb_setup_retire_blk_timer(struct packet_sock *po)
chetan lokef6fb8f12011-08-19 10:18:16 +0000548{
chetan lokebc59ba32011-08-25 10:43:30 +0000549 struct tpacket_kbdq_core *pkc;
chetan lokef6fb8f12011-08-19 10:18:16 +0000550
Maninder Singhe8e85cc2015-06-22 12:39:16 +0530551 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f12011-08-19 10:18:16 +0000552 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
553}
554
555static int prb_calc_retire_blk_tmo(struct packet_sock *po,
556 int blk_size_in_bytes)
557{
558 struct net_device *dev;
559 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000560 struct ethtool_cmd ecmd;
561 int err;
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000562 u32 speed;
chetan lokef6fb8f12011-08-19 10:18:16 +0000563
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000564 rtnl_lock();
565 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
566 if (unlikely(!dev)) {
567 rtnl_unlock();
chetan lokef6fb8f12011-08-19 10:18:16 +0000568 return DEFAULT_PRB_RETIRE_TOV;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000569 }
570 err = __ethtool_get_settings(dev, &ecmd);
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000571 speed = ethtool_cmd_speed(&ecmd);
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000572 rtnl_unlock();
573 if (!err) {
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000574 /*
575 * If the link speed is so slow you don't really
576 * need to worry about perf anyways
577 */
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000578 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000579 return DEFAULT_PRB_RETIRE_TOV;
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000580 } else {
581 msec = 1;
582 div = speed / 1000;
chetan lokef6fb8f12011-08-19 10:18:16 +0000583 }
584 }
585
586 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
587
588 if (div)
589 mbits /= div;
590
591 tmo = mbits * msec;
592
593 if (div)
594 return tmo+1;
595 return tmo;
596}
597
chetan lokebc59ba32011-08-25 10:43:30 +0000598static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
chetan lokef6fb8f12011-08-19 10:18:16 +0000599 union tpacket_req_u *req_u)
600{
601 p1->feature_req_word = req_u->req3.tp_feature_req_word;
602}
603
604static void init_prb_bdqc(struct packet_sock *po,
605 struct packet_ring_buffer *rb,
606 struct pgv *pg_vec,
Maninder Singhe8e85cc2015-06-22 12:39:16 +0530607 union tpacket_req_u *req_u)
chetan lokef6fb8f12011-08-19 10:18:16 +0000608{
Duan Jiong22781a52013-12-06 13:29:36 +0800609 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
chetan lokebc59ba32011-08-25 10:43:30 +0000610 struct tpacket_block_desc *pbd;
chetan lokef6fb8f12011-08-19 10:18:16 +0000611
612 memset(p1, 0x0, sizeof(*p1));
613
614 p1->knxt_seq_num = 1;
615 p1->pkbdq = pg_vec;
chetan lokebc59ba32011-08-25 10:43:30 +0000616 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
Joe Perchese3192692012-06-03 17:41:40 +0000617 p1->pkblk_start = pg_vec[0].buffer;
chetan lokef6fb8f12011-08-19 10:18:16 +0000618 p1->kblk_size = req_u->req3.tp_block_size;
619 p1->knum_blocks = req_u->req3.tp_block_nr;
620 p1->hdrlen = po->tp_hdrlen;
621 p1->version = po->tp_version;
622 p1->last_kactive_blk_num = 0;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +0000623 po->stats.stats3.tp_freeze_q_cnt = 0;
chetan lokef6fb8f12011-08-19 10:18:16 +0000624 if (req_u->req3.tp_retire_blk_tov)
625 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
626 else
627 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
628 req_u->req3.tp_block_size);
629 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
630 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
631
Eric Dumazetdc808112014-08-15 09:16:04 -0700632 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
chetan lokef6fb8f12011-08-19 10:18:16 +0000633 prb_init_ft_ops(p1, req_u);
Maninder Singhe8e85cc2015-06-22 12:39:16 +0530634 prb_setup_retire_blk_timer(po);
chetan lokef6fb8f12011-08-19 10:18:16 +0000635 prb_open_block(p1, pbd);
636}
637
638/* Do NOT update the last_blk_num first.
639 * Assumes sk_buff_head lock is held.
640 */
chetan lokebc59ba32011-08-25 10:43:30 +0000641static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f12011-08-19 10:18:16 +0000642{
643 mod_timer(&pkc->retire_blk_timer,
644 jiffies + pkc->tov_in_jiffies);
645 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
646}
647
648/*
649 * Timer logic:
650 * 1) We refresh the timer only when we open a block.
651 * By doing this we don't waste cycles refreshing the timer
652 * on packet-by-packet basis.
653 *
654 * With a 1MB block-size, on a 1Gbps line, it will take
655 * i) ~8 ms to fill a block + ii) memcpy etc.
656 * In this cut we are not accounting for the memcpy time.
657 *
658 * So, if the user sets the 'tmo' to 10ms then the timer
659 * will never fire while the block is still getting filled
660 * (which is what we want). However, the user could choose
661 * to close a block early and that's fine.
662 *
663 * But when the timer does fire, we check whether or not to refresh it.
664 * Since the tmo granularity is in msecs, it is not too expensive
665 * to refresh the timer, lets say every '8' msecs.
666 * Either the user can set the 'tmo' or we can derive it based on
667 * a) line-speed and b) block-size.
668 * prb_calc_retire_blk_tmo() calculates the tmo.
669 *
670 */
671static void prb_retire_rx_blk_timer_expired(unsigned long data)
672{
673 struct packet_sock *po = (struct packet_sock *)data;
Duan Jiong22781a52013-12-06 13:29:36 +0800674 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f12011-08-19 10:18:16 +0000675 unsigned int frozen;
chetan lokebc59ba32011-08-25 10:43:30 +0000676 struct tpacket_block_desc *pbd;
chetan lokef6fb8f12011-08-19 10:18:16 +0000677
678 spin_lock(&po->sk.sk_receive_queue.lock);
679
680 frozen = prb_queue_frozen(pkc);
681 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
682
683 if (unlikely(pkc->delete_blk_timer))
684 goto out;
685
686 /* We only need to plug the race when the block is partially filled.
687 * tpacket_rcv:
688 * lock(); increment BLOCK_NUM_PKTS; unlock()
689 * copy_bits() is in progress ...
690 * timer fires on other cpu:
691 * we can't retire the current block because copy_bits
692 * is in progress.
693 *
694 */
695 if (BLOCK_NUM_PKTS(pbd)) {
696 while (atomic_read(&pkc->blk_fill_in_prog)) {
697 /* Waiting for skb_copy_bits to finish... */
698 cpu_relax();
699 }
700 }
701
702 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
703 if (!frozen) {
Alexander Drozdov41a50d62015-02-24 08:18:28 +0300704 if (!BLOCK_NUM_PKTS(pbd)) {
705 /* An empty block. Just refresh the timer. */
706 goto refresh_timer;
707 }
chetan lokef6fb8f12011-08-19 10:18:16 +0000708 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
709 if (!prb_dispatch_next_block(pkc, po))
710 goto refresh_timer;
711 else
712 goto out;
713 } else {
714 /* Case 1. Queue was frozen because user-space was
715 * lagging behind.
716 */
717 if (prb_curr_blk_in_use(pkc, pbd)) {
718 /*
719 * Ok, user-space is still behind.
720 * So just refresh the timer.
721 */
722 goto refresh_timer;
723 } else {
724 /* Case 2. queue was frozen,user-space caught up,
725 * now the link went idle && the timer fired.
726 * We don't have a block to close.So we open this
727 * block and restart the timer.
728 * opening a block thaws the queue,restarts timer
729 * Thawing/timer-refresh is a side effect.
730 */
731 prb_open_block(pkc, pbd);
732 goto out;
733 }
734 }
735 }
736
737refresh_timer:
738 _prb_refresh_rx_retire_blk_timer(pkc);
739
740out:
741 spin_unlock(&po->sk.sk_receive_queue.lock);
742}
743
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000744static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
chetan lokebc59ba32011-08-25 10:43:30 +0000745 struct tpacket_block_desc *pbd1, __u32 status)
chetan lokef6fb8f12011-08-19 10:18:16 +0000746{
747 /* Flush everything minus the block header */
748
749#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
750 u8 *start, *end;
751
752 start = (u8 *)pbd1;
753
754 /* Skip the block header(we know header WILL fit in 4K) */
755 start += PAGE_SIZE;
756
757 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
758 for (; start < end; start += PAGE_SIZE)
759 flush_dcache_page(pgv_to_page(start));
760
761 smp_wmb();
762#endif
763
764 /* Now update the block status. */
765
766 BLOCK_STATUS(pbd1) = status;
767
768 /* Flush the block header */
769
770#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
771 start = (u8 *)pbd1;
772 flush_dcache_page(pgv_to_page(start));
773
774 smp_wmb();
775#endif
776}
777
778/*
779 * Side effect:
780 *
781 * 1) flush the block
782 * 2) Increment active_blk_num
783 *
784 * Note:We DONT refresh the timer on purpose.
785 * Because almost always the next block will be opened.
786 */
chetan lokebc59ba32011-08-25 10:43:30 +0000787static void prb_close_block(struct tpacket_kbdq_core *pkc1,
788 struct tpacket_block_desc *pbd1,
chetan lokef6fb8f12011-08-19 10:18:16 +0000789 struct packet_sock *po, unsigned int stat)
790{
791 __u32 status = TP_STATUS_USER | stat;
792
793 struct tpacket3_hdr *last_pkt;
chetan lokebc59ba32011-08-25 10:43:30 +0000794 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
Dan Collinsda413ee2014-12-19 16:49:25 +1300795 struct sock *sk = &po->sk;
chetan lokef6fb8f12011-08-19 10:18:16 +0000796
Daniel Borkmannee80fbf2013-04-19 06:12:29 +0000797 if (po->stats.stats3.tp_drops)
chetan lokef6fb8f12011-08-19 10:18:16 +0000798 status |= TP_STATUS_LOSING;
799
800 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
801 last_pkt->tp_next_offset = 0;
802
803 /* Get the ts of the last pkt */
804 if (BLOCK_NUM_PKTS(pbd1)) {
805 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
806 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
807 } else {
Alexander Drozdov41a50d62015-02-24 08:18:28 +0300808 /* Ok, we tmo'd - so get the current time.
809 *
810 * It shouldn't really happen as we don't close empty
811 * blocks. See prb_retire_rx_blk_timer_expired().
812 */
chetan lokef6fb8f12011-08-19 10:18:16 +0000813 struct timespec ts;
814 getnstimeofday(&ts);
815 h1->ts_last_pkt.ts_sec = ts.tv_sec;
816 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
817 }
818
819 smp_wmb();
820
821 /* Flush the block */
822 prb_flush_block(pkc1, pbd1, status);
823
Dan Collinsda413ee2014-12-19 16:49:25 +1300824 sk->sk_data_ready(sk);
825
chetan lokef6fb8f12011-08-19 10:18:16 +0000826 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
827}
828
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000829static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f12011-08-19 10:18:16 +0000830{
831 pkc->reset_pending_on_curr_blk = 0;
832}
833
834/*
835 * Side effect of opening a block:
836 *
837 * 1) prb_queue is thawed.
838 * 2) retire_blk_timer is refreshed.
839 *
840 */
chetan lokebc59ba32011-08-25 10:43:30 +0000841static void prb_open_block(struct tpacket_kbdq_core *pkc1,
842 struct tpacket_block_desc *pbd1)
chetan lokef6fb8f12011-08-19 10:18:16 +0000843{
844 struct timespec ts;
chetan lokebc59ba32011-08-25 10:43:30 +0000845 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
chetan lokef6fb8f12011-08-19 10:18:16 +0000846
847 smp_rmb();
848
Daniel Borkmann8da30562013-05-03 02:57:00 +0000849 /* We could have just memset this but we will lose the
850 * flexibility of making the priv area sticky
851 */
chetan lokef6fb8f12011-08-19 10:18:16 +0000852
Daniel Borkmann8da30562013-05-03 02:57:00 +0000853 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
854 BLOCK_NUM_PKTS(pbd1) = 0;
855 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
chetan lokef6fb8f12011-08-19 10:18:16 +0000856
Daniel Borkmann8da30562013-05-03 02:57:00 +0000857 getnstimeofday(&ts);
chetan lokef6fb8f12011-08-19 10:18:16 +0000858
Daniel Borkmann8da30562013-05-03 02:57:00 +0000859 h1->ts_first_pkt.ts_sec = ts.tv_sec;
860 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
chetan lokef6fb8f12011-08-19 10:18:16 +0000861
Daniel Borkmann8da30562013-05-03 02:57:00 +0000862 pkc1->pkblk_start = (char *)pbd1;
863 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864
865 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
867
868 pbd1->version = pkc1->version;
869 pkc1->prev = pkc1->nxt_offset;
870 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
871
872 prb_thaw_queue(pkc1);
873 _prb_refresh_rx_retire_blk_timer(pkc1);
874
875 smp_wmb();
chetan lokef6fb8f12011-08-19 10:18:16 +0000876}
877
878/*
879 * Queue freeze logic:
880 * 1) Assume tp_block_nr = 8 blocks.
881 * 2) At time 't0', user opens Rx ring.
882 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
883 * 4) user-space is either sleeping or processing block '0'.
884 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
885 * it will close block-7,loop around and try to fill block '0'.
886 * call-flow:
887 * __packet_lookup_frame_in_block
888 * prb_retire_current_block()
889 * prb_dispatch_next_block()
890 * |->(BLOCK_STATUS == USER) evaluates to true
891 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
892 * 6) Now there are two cases:
893 * 6.1) Link goes idle right after the queue is frozen.
894 * But remember, the last open_block() refreshed the timer.
895 * When this timer expires,it will refresh itself so that we can
896 * re-open block-0 in near future.
897 * 6.2) Link is busy and keeps on receiving packets. This is a simple
898 * case and __packet_lookup_frame_in_block will check if block-0
899 * is free and can now be re-used.
900 */
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000901static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f12011-08-19 10:18:16 +0000902 struct packet_sock *po)
903{
904 pkc->reset_pending_on_curr_blk = 1;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +0000905 po->stats.stats3.tp_freeze_q_cnt++;
chetan lokef6fb8f12011-08-19 10:18:16 +0000906}
907
908#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
909
910/*
911 * If the next block is free then we will dispatch it
912 * and return a good offset.
913 * Else, we will freeze the queue.
914 * So, caller must check the return value.
915 */
chetan lokebc59ba32011-08-25 10:43:30 +0000916static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f12011-08-19 10:18:16 +0000917 struct packet_sock *po)
918{
chetan lokebc59ba32011-08-25 10:43:30 +0000919 struct tpacket_block_desc *pbd;
chetan lokef6fb8f12011-08-19 10:18:16 +0000920
921 smp_rmb();
922
923 /* 1. Get current block num */
924 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
925
926 /* 2. If this block is currently in_use then freeze the queue */
927 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
928 prb_freeze_queue(pkc, po);
929 return NULL;
930 }
931
932 /*
933 * 3.
934 * open this block and return the offset where the first packet
935 * needs to get stored.
936 */
937 prb_open_block(pkc, pbd);
938 return (void *)pkc->nxt_offset;
939}
940
chetan lokebc59ba32011-08-25 10:43:30 +0000941static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f12011-08-19 10:18:16 +0000942 struct packet_sock *po, unsigned int status)
943{
chetan lokebc59ba32011-08-25 10:43:30 +0000944 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
chetan lokef6fb8f12011-08-19 10:18:16 +0000945
946 /* retire/close the current block */
947 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
948 /*
949 * Plug the case where copy_bits() is in progress on
950 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
951 * have space to copy the pkt in the current block and
952 * called prb_retire_current_block()
953 *
954 * We don't need to worry about the TMO case because
955 * the timer-handler already handled this case.
956 */
957 if (!(status & TP_STATUS_BLK_TMO)) {
958 while (atomic_read(&pkc->blk_fill_in_prog)) {
959 /* Waiting for skb_copy_bits to finish... */
960 cpu_relax();
961 }
962 }
963 prb_close_block(pkc, pbd, po, status);
964 return;
965 }
chetan lokef6fb8f12011-08-19 10:18:16 +0000966}
967
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000968static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
chetan lokebc59ba32011-08-25 10:43:30 +0000969 struct tpacket_block_desc *pbd)
chetan lokef6fb8f12011-08-19 10:18:16 +0000970{
971 return TP_STATUS_USER & BLOCK_STATUS(pbd);
972}
973
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000974static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f12011-08-19 10:18:16 +0000975{
976 return pkc->reset_pending_on_curr_blk;
977}
978
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000979static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
chetan lokef6fb8f12011-08-19 10:18:16 +0000980{
chetan lokebc59ba32011-08-25 10:43:30 +0000981 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
chetan lokef6fb8f12011-08-19 10:18:16 +0000982 atomic_dec(&pkc->blk_fill_in_prog);
983}
984
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000985static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f12011-08-19 10:18:16 +0000986 struct tpacket3_hdr *ppd)
987{
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800988 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
chetan lokef6fb8f12011-08-19 10:18:16 +0000989}
990
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000991static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f12011-08-19 10:18:16 +0000992 struct tpacket3_hdr *ppd)
993{
994 ppd->hv1.tp_rxhash = 0;
995}
996
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000997static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f12011-08-19 10:18:16 +0000998 struct tpacket3_hdr *ppd)
999{
Jiri Pirkodf8a39de2015-01-13 17:13:44 +01001000 if (skb_vlan_tag_present(pkc->skb)) {
1001 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09001002 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1003 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
chetan lokef6fb8f12011-08-19 10:18:16 +00001004 } else {
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +00001005 ppd->hv1.tp_vlan_tci = 0;
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09001006 ppd->hv1.tp_vlan_tpid = 0;
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +00001007 ppd->tp_status = TP_STATUS_AVAILABLE;
chetan lokef6fb8f12011-08-19 10:18:16 +00001008 }
1009}
1010
chetan lokebc59ba32011-08-25 10:43:30 +00001011static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f12011-08-19 10:18:16 +00001012 struct tpacket3_hdr *ppd)
1013{
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09001014 ppd->hv1.tp_padding = 0;
chetan lokef6fb8f12011-08-19 10:18:16 +00001015 prb_fill_vlan_info(pkc, ppd);
1016
1017 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1018 prb_fill_rxhash(pkc, ppd);
1019 else
1020 prb_clear_rxhash(pkc, ppd);
1021}
1022
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001023static void prb_fill_curr_block(char *curr,
chetan lokebc59ba32011-08-25 10:43:30 +00001024 struct tpacket_kbdq_core *pkc,
1025 struct tpacket_block_desc *pbd,
chetan lokef6fb8f12011-08-19 10:18:16 +00001026 unsigned int len)
1027{
1028 struct tpacket3_hdr *ppd;
1029
1030 ppd = (struct tpacket3_hdr *)curr;
1031 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1032 pkc->prev = curr;
1033 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1035 BLOCK_NUM_PKTS(pbd) += 1;
1036 atomic_inc(&pkc->blk_fill_in_prog);
1037 prb_run_all_ft_ops(pkc, ppd);
1038}
1039
1040/* Assumes caller has the sk->rx_queue.lock */
1041static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1042 struct sk_buff *skb,
1043 int status,
1044 unsigned int len
1045 )
1046{
chetan lokebc59ba32011-08-25 10:43:30 +00001047 struct tpacket_kbdq_core *pkc;
1048 struct tpacket_block_desc *pbd;
chetan lokef6fb8f12011-08-19 10:18:16 +00001049 char *curr, *end;
1050
Joe Perchese3192692012-06-03 17:41:40 +00001051 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f12011-08-19 10:18:16 +00001052 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1053
1054 /* Queue is frozen when user space is lagging behind */
1055 if (prb_queue_frozen(pkc)) {
1056 /*
1057 * Check if that last block which caused the queue to freeze,
1058 * is still in_use by user-space.
1059 */
1060 if (prb_curr_blk_in_use(pkc, pbd)) {
1061 /* Can't record this packet */
1062 return NULL;
1063 } else {
1064 /*
1065 * Ok, the block was released by user-space.
1066 * Now let's open that block.
1067 * opening a block also thaws the queue.
1068 * Thawing is a side effect.
1069 */
1070 prb_open_block(pkc, pbd);
1071 }
1072 }
1073
1074 smp_mb();
1075 curr = pkc->nxt_offset;
1076 pkc->skb = skb;
Joe Perchese3192692012-06-03 17:41:40 +00001077 end = (char *)pbd + pkc->kblk_size;
chetan lokef6fb8f12011-08-19 10:18:16 +00001078
1079 /* first try the current block */
1080 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1081 prb_fill_curr_block(curr, pkc, pbd, len);
1082 return (void *)curr;
1083 }
1084
1085 /* Ok, close the current block */
1086 prb_retire_current_block(pkc, po, 0);
1087
1088 /* Now, try to dispatch the next block */
1089 curr = (char *)prb_dispatch_next_block(pkc, po);
1090 if (curr) {
1091 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1092 prb_fill_curr_block(curr, pkc, pbd, len);
1093 return (void *)curr;
1094 }
1095
1096 /*
1097 * No free blocks are available.user_space hasn't caught up yet.
1098 * Queue was just frozen and now this packet will get dropped.
1099 */
1100 return NULL;
1101}
1102
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001103static void *packet_current_rx_frame(struct packet_sock *po,
chetan lokef6fb8f12011-08-19 10:18:16 +00001104 struct sk_buff *skb,
1105 int status, unsigned int len)
1106{
1107 char *curr = NULL;
1108 switch (po->tp_version) {
1109 case TPACKET_V1:
1110 case TPACKET_V2:
1111 curr = packet_lookup_frame(po, &po->rx_ring,
1112 po->rx_ring.head, status);
1113 return curr;
1114 case TPACKET_V3:
1115 return __packet_lookup_frame_in_block(po, skb, status, len);
1116 default:
1117 WARN(1, "TPACKET version not supported\n");
1118 BUG();
Ying Xue99aa3472012-08-06 16:27:10 +00001119 return NULL;
chetan lokef6fb8f12011-08-19 10:18:16 +00001120 }
1121}
1122
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001123static void *prb_lookup_block(struct packet_sock *po,
chetan lokef6fb8f12011-08-19 10:18:16 +00001124 struct packet_ring_buffer *rb,
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001125 unsigned int idx,
chetan lokef6fb8f12011-08-19 10:18:16 +00001126 int status)
1127{
chetan lokebc59ba32011-08-25 10:43:30 +00001128 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001129 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
chetan lokef6fb8f12011-08-19 10:18:16 +00001130
1131 if (status != BLOCK_STATUS(pbd))
1132 return NULL;
1133 return pbd;
1134}
1135
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001136static int prb_previous_blk_num(struct packet_ring_buffer *rb)
chetan lokef6fb8f12011-08-19 10:18:16 +00001137{
1138 unsigned int prev;
1139 if (rb->prb_bdqc.kactive_blk_num)
1140 prev = rb->prb_bdqc.kactive_blk_num-1;
1141 else
1142 prev = rb->prb_bdqc.knum_blocks-1;
1143 return prev;
1144}
1145
1146/* Assumes caller has held the rx_queue.lock */
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001147static void *__prb_previous_block(struct packet_sock *po,
chetan lokef6fb8f12011-08-19 10:18:16 +00001148 struct packet_ring_buffer *rb,
1149 int status)
1150{
1151 unsigned int previous = prb_previous_blk_num(rb);
1152 return prb_lookup_block(po, rb, previous, status);
1153}
1154
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001155static void *packet_previous_rx_frame(struct packet_sock *po,
chetan lokef6fb8f12011-08-19 10:18:16 +00001156 struct packet_ring_buffer *rb,
1157 int status)
1158{
1159 if (po->tp_version <= TPACKET_V2)
1160 return packet_previous_frame(po, rb, status);
1161
1162 return __prb_previous_block(po, rb, status);
1163}
1164
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001165static void packet_increment_rx_head(struct packet_sock *po,
chetan lokef6fb8f12011-08-19 10:18:16 +00001166 struct packet_ring_buffer *rb)
1167{
1168 switch (po->tp_version) {
1169 case TPACKET_V1:
1170 case TPACKET_V2:
1171 return packet_increment_head(rb);
1172 case TPACKET_V3:
1173 default:
1174 WARN(1, "TPACKET version not supported.\n");
1175 BUG();
1176 return;
1177 }
1178}
1179
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001180static void *packet_previous_frame(struct packet_sock *po,
Johann Baudy69e3c752009-05-18 22:11:22 -07001181 struct packet_ring_buffer *rb,
1182 int status)
1183{
1184 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1185 return packet_lookup_frame(po, rb, previous, status);
1186}
1187
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001188static void packet_increment_head(struct packet_ring_buffer *buff)
Johann Baudy69e3c752009-05-18 22:11:22 -07001189{
1190 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1191}
1192
Daniel Borkmannb0138402014-01-15 16:25:36 +01001193static void packet_inc_pending(struct packet_ring_buffer *rb)
1194{
1195 this_cpu_inc(*rb->pending_refcnt);
1196}
1197
1198static void packet_dec_pending(struct packet_ring_buffer *rb)
1199{
1200 this_cpu_dec(*rb->pending_refcnt);
1201}
1202
1203static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1204{
1205 unsigned int refcnt = 0;
1206 int cpu;
1207
1208 /* We don't use pending refcount in rx_ring. */
1209 if (rb->pending_refcnt == NULL)
1210 return 0;
1211
1212 for_each_possible_cpu(cpu)
1213 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1214
1215 return refcnt;
1216}
1217
1218static int packet_alloc_pending(struct packet_sock *po)
1219{
1220 po->rx_ring.pending_refcnt = NULL;
1221
1222 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1223 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1224 return -ENOBUFS;
1225
1226 return 0;
1227}
1228
1229static void packet_free_pending(struct packet_sock *po)
1230{
1231 free_percpu(po->tx_ring.pending_refcnt);
1232}
1233
Willem de Bruijn99547292015-05-12 11:56:47 -04001234#define ROOM_POW_OFF 2
1235#define ROOM_NONE 0x0
1236#define ROOM_LOW 0x1
1237#define ROOM_NORMAL 0x2
1238
1239static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1240{
1241 int idx, len;
1242
1243 len = po->rx_ring.frame_max + 1;
1244 idx = po->rx_ring.head;
1245 if (pow_off)
1246 idx += len >> pow_off;
1247 if (idx >= len)
1248 idx -= len;
1249 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1250}
1251
1252static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1253{
1254 int idx, len;
1255
1256 len = po->rx_ring.prb_bdqc.knum_blocks;
1257 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1258 if (pow_off)
1259 idx += len >> pow_off;
1260 if (idx >= len)
1261 idx -= len;
1262 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1263}
1264
Willem de Bruijn2ccdbaa2015-05-12 11:56:48 -04001265static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001266{
1267 struct sock *sk = &po->sk;
Willem de Bruijn99547292015-05-12 11:56:47 -04001268 int ret = ROOM_NONE;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001269
Willem de Bruijn99547292015-05-12 11:56:47 -04001270 if (po->prot_hook.func != tpacket_rcv) {
1271 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
Willem de Bruijn2ccdbaa2015-05-12 11:56:48 -04001272 - (skb ? skb->truesize : 0);
Willem de Bruijn99547292015-05-12 11:56:47 -04001273 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1274 return ROOM_NORMAL;
1275 else if (avail > 0)
1276 return ROOM_LOW;
1277 else
1278 return ROOM_NONE;
1279 }
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001280
Willem de Bruijn99547292015-05-12 11:56:47 -04001281 if (po->tp_version == TPACKET_V3) {
1282 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1283 ret = ROOM_NORMAL;
1284 else if (__tpacket_v3_has_room(po, 0))
1285 ret = ROOM_LOW;
1286 } else {
1287 if (__tpacket_has_room(po, ROOM_POW_OFF))
1288 ret = ROOM_NORMAL;
1289 else if (__tpacket_has_room(po, 0))
1290 ret = ROOM_LOW;
1291 }
Willem de Bruijn2ccdbaa2015-05-12 11:56:48 -04001292
1293 return ret;
1294}
1295
1296static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1297{
1298 int ret;
1299 bool has_room;
1300
Willem de Bruijn54d7c012015-05-14 15:25:02 -04001301 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1302 ret = __packet_rcv_has_room(po, skb);
Willem de Bruijn2ccdbaa2015-05-12 11:56:48 -04001303 has_room = ret == ROOM_NORMAL;
1304 if (po->pressure == has_room)
Willem de Bruijn54d7c012015-05-14 15:25:02 -04001305 po->pressure = !has_room;
1306 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001307
Willem de Bruijn99547292015-05-12 11:56:47 -04001308 return ret;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001309}
1310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311static void packet_sock_destruct(struct sock *sk)
1312{
Richard Cochraned85b562010-04-07 22:41:28 +00001313 skb_queue_purge(&sk->sk_error_queue);
1314
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001315 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1316 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317
1318 if (!sock_flag(sk, SOCK_DEAD)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001319 pr_err("Attempt to release alive packet socket: %p\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 return;
1321 }
1322
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08001323 sk_refcnt_debug_dec(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324}
1325
Willem de Bruijn3b3a5b02015-05-12 11:56:49 -04001326static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1327{
1328 u32 rxhash;
1329 int i, count = 0;
1330
1331 rxhash = skb_get_hash(skb);
1332 for (i = 0; i < ROLLOVER_HLEN; i++)
1333 if (po->rollover->history[i] == rxhash)
1334 count++;
1335
1336 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1337 return count > (ROLLOVER_HLEN >> 1);
1338}
1339
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001340static unsigned int fanout_demux_hash(struct packet_fanout *f,
1341 struct sk_buff *skb,
1342 unsigned int num)
David S. Millerdc99f602011-07-05 01:45:05 -07001343{
Tom Herbert61b905d2014-03-24 15:34:47 -07001344 return reciprocal_scale(skb_get_hash(skb), num);
David S. Millerdc99f602011-07-05 01:45:05 -07001345}
1346
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001347static unsigned int fanout_demux_lb(struct packet_fanout *f,
1348 struct sk_buff *skb,
1349 unsigned int num)
David S. Millerdc99f602011-07-05 01:45:05 -07001350{
Willem de Bruijn468479e2015-06-17 15:59:34 -04001351 unsigned int val = atomic_inc_return(&f->rr_cur);
David S. Millerdc99f602011-07-05 01:45:05 -07001352
Willem de Bruijn468479e2015-06-17 15:59:34 -04001353 return val % num;
David S. Millerdc99f602011-07-05 01:45:05 -07001354}
1355
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001356static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1357 struct sk_buff *skb,
1358 unsigned int num)
David S. Miller95ec3eb2011-07-06 01:56:38 -07001359{
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001360 return smp_processor_id() % num;
1361}
David S. Miller95ec3eb2011-07-06 01:56:38 -07001362
Daniel Borkmann5df0ddf2013-08-28 22:13:09 +02001363static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1364 struct sk_buff *skb,
1365 unsigned int num)
1366{
Daniel Borkmannf337db62014-01-22 02:29:39 +01001367 return prandom_u32_max(num);
Daniel Borkmann5df0ddf2013-08-28 22:13:09 +02001368}
1369
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001370static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1371 struct sk_buff *skb,
Willem de Bruijnad377ca2015-05-12 11:56:45 -04001372 unsigned int idx, bool try_self,
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001373 unsigned int num)
1374{
Willem de Bruijn4633c9e2015-05-17 19:44:02 -04001375 struct packet_sock *po, *po_next, *po_skip = NULL;
Willem de Bruijna9b63912015-05-12 11:56:50 -04001376 unsigned int i, j, room = ROOM_NONE;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001377
Willem de Bruijn0648ab72015-05-12 11:56:46 -04001378 po = pkt_sk(f->arr[idx]);
Willem de Bruijn3b3a5b02015-05-12 11:56:49 -04001379
1380 if (try_self) {
1381 room = packet_rcv_has_room(po, skb);
1382 if (room == ROOM_NORMAL ||
1383 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1384 return idx;
Willem de Bruijn4633c9e2015-05-17 19:44:02 -04001385 po_skip = po;
Willem de Bruijn3b3a5b02015-05-12 11:56:49 -04001386 }
Willem de Bruijnad377ca2015-05-12 11:56:45 -04001387
Willem de Bruijn0648ab72015-05-12 11:56:46 -04001388 i = j = min_t(int, po->rollover->sock, num - 1);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001389 do {
Willem de Bruijn2ccdbaa2015-05-12 11:56:48 -04001390 po_next = pkt_sk(f->arr[i]);
Willem de Bruijn4633c9e2015-05-17 19:44:02 -04001391 if (po_next != po_skip && !po_next->pressure &&
Willem de Bruijn2ccdbaa2015-05-12 11:56:48 -04001392 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001393 if (i != j)
Willem de Bruijn0648ab72015-05-12 11:56:46 -04001394 po->rollover->sock = i;
Willem de Bruijna9b63912015-05-12 11:56:50 -04001395 atomic_long_inc(&po->rollover->num);
1396 if (room == ROOM_LOW)
1397 atomic_long_inc(&po->rollover->num_huge);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001398 return i;
1399 }
Willem de Bruijnad377ca2015-05-12 11:56:45 -04001400
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001401 if (++i == num)
1402 i = 0;
1403 } while (i != j);
1404
Willem de Bruijna9b63912015-05-12 11:56:50 -04001405 atomic_long_inc(&po->rollover->num_failed);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001406 return idx;
1407}
1408
Neil Horman2d360972014-01-22 16:01:44 -05001409static unsigned int fanout_demux_qm(struct packet_fanout *f,
1410 struct sk_buff *skb,
1411 unsigned int num)
1412{
1413 return skb_get_queue_mapping(skb) % num;
1414}
1415
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001416static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1417 struct sk_buff *skb,
1418 unsigned int num)
1419{
1420 struct bpf_prog *prog;
1421 unsigned int ret = 0;
1422
1423 rcu_read_lock();
1424 prog = rcu_dereference(f->bpf_prog);
1425 if (prog)
Alexei Starovoitovff936a02015-10-07 10:55:41 -07001426 ret = bpf_prog_run_clear_cb(prog, skb) % num;
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001427 rcu_read_unlock();
1428
1429 return ret;
1430}
1431
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001432static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1433{
1434 return f->flags & (flag >> 8);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001435}
1436
David S. Miller95ec3eb2011-07-06 01:56:38 -07001437static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1438 struct packet_type *pt, struct net_device *orig_dev)
David S. Millerdc99f602011-07-05 01:45:05 -07001439{
1440 struct packet_fanout *f = pt->af_packet_priv;
Eric Dumazetf98f4512015-06-16 07:59:11 -07001441 unsigned int num = READ_ONCE(f->num_members);
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -05001442 struct net *net = read_pnet(&f->net);
David S. Millerdc99f602011-07-05 01:45:05 -07001443 struct packet_sock *po;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001444 unsigned int idx;
David S. Millerdc99f602011-07-05 01:45:05 -07001445
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -05001446 if (!net_eq(dev_net(dev), net) || !num) {
David S. Millerdc99f602011-07-05 01:45:05 -07001447 kfree_skb(skb);
1448 return 0;
1449 }
1450
Alexander Drozdov3f34b242015-02-20 08:24:27 +03001451 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
Eric W. Biederman19bcf9f2015-10-09 13:44:54 -05001452 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
Alexander Drozdov3f34b242015-02-20 08:24:27 +03001453 if (!skb)
1454 return 0;
1455 }
David S. Miller95ec3eb2011-07-06 01:56:38 -07001456 switch (f->type) {
1457 case PACKET_FANOUT_HASH:
1458 default:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001459 idx = fanout_demux_hash(f, skb, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001460 break;
1461 case PACKET_FANOUT_LB:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001462 idx = fanout_demux_lb(f, skb, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001463 break;
1464 case PACKET_FANOUT_CPU:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001465 idx = fanout_demux_cpu(f, skb, num);
1466 break;
Daniel Borkmann5df0ddf2013-08-28 22:13:09 +02001467 case PACKET_FANOUT_RND:
1468 idx = fanout_demux_rnd(f, skb, num);
1469 break;
Neil Horman2d360972014-01-22 16:01:44 -05001470 case PACKET_FANOUT_QM:
1471 idx = fanout_demux_qm(f, skb, num);
1472 break;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001473 case PACKET_FANOUT_ROLLOVER:
Willem de Bruijnad377ca2015-05-12 11:56:45 -04001474 idx = fanout_demux_rollover(f, skb, 0, false, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001475 break;
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001476 case PACKET_FANOUT_CBPF:
Willem de Bruijnf2e52092015-08-14 22:31:35 -04001477 case PACKET_FANOUT_EBPF:
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001478 idx = fanout_demux_bpf(f, skb, num);
1479 break;
David S. Miller7736d332011-07-05 01:43:20 -07001480 }
1481
Willem de Bruijnad377ca2015-05-12 11:56:45 -04001482 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1483 idx = fanout_demux_rollover(f, skb, idx, true, num);
David S. Millerdc99f602011-07-05 01:45:05 -07001484
Willem de Bruijnad377ca2015-05-12 11:56:45 -04001485 po = pkt_sk(f->arr[idx]);
David S. Millerdc99f602011-07-05 01:45:05 -07001486 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1487}
1488
Pavel Emelyanovfff33212012-08-16 05:36:48 +00001489DEFINE_MUTEX(fanout_mutex);
1490EXPORT_SYMBOL_GPL(fanout_mutex);
David S. Millerdc99f602011-07-05 01:45:05 -07001491static LIST_HEAD(fanout_list);
1492
1493static void __fanout_link(struct sock *sk, struct packet_sock *po)
1494{
1495 struct packet_fanout *f = po->fanout;
1496
1497 spin_lock(&f->lock);
1498 f->arr[f->num_members] = sk;
1499 smp_wmb();
1500 f->num_members++;
1501 spin_unlock(&f->lock);
1502}
1503
1504static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1505{
1506 struct packet_fanout *f = po->fanout;
1507 int i;
1508
1509 spin_lock(&f->lock);
1510 for (i = 0; i < f->num_members; i++) {
1511 if (f->arr[i] == sk)
1512 break;
1513 }
1514 BUG_ON(i >= f->num_members);
1515 f->arr[i] = f->arr[f->num_members - 1];
1516 f->num_members--;
1517 spin_unlock(&f->lock);
1518}
1519
Weilong Chend4dd8ae2013-12-23 11:31:38 +08001520static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
Eric Leblondc0de08d2012-08-16 22:02:58 +00001521{
Eric Dumazet161642e2015-10-09 11:29:32 -07001522 if (sk->sk_family != PF_PACKET)
1523 return false;
Eric Leblondc0de08d2012-08-16 22:02:58 +00001524
Eric Dumazet161642e2015-10-09 11:29:32 -07001525 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
Eric Leblondc0de08d2012-08-16 22:02:58 +00001526}
1527
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001528static void fanout_init_data(struct packet_fanout *f)
1529{
1530 switch (f->type) {
1531 case PACKET_FANOUT_LB:
1532 atomic_set(&f->rr_cur, 0);
1533 break;
1534 case PACKET_FANOUT_CBPF:
Willem de Bruijnf2e52092015-08-14 22:31:35 -04001535 case PACKET_FANOUT_EBPF:
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001536 RCU_INIT_POINTER(f->bpf_prog, NULL);
1537 break;
1538 }
1539}
1540
1541static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1542{
1543 struct bpf_prog *old;
1544
1545 spin_lock(&f->lock);
1546 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1547 rcu_assign_pointer(f->bpf_prog, new);
1548 spin_unlock(&f->lock);
1549
1550 if (old) {
1551 synchronize_net();
1552 bpf_prog_destroy(old);
1553 }
1554}
1555
1556static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1557 unsigned int len)
1558{
1559 struct bpf_prog *new;
1560 struct sock_fprog fprog;
1561 int ret;
1562
1563 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1564 return -EPERM;
1565 if (len != sizeof(fprog))
1566 return -EINVAL;
1567 if (copy_from_user(&fprog, data, len))
1568 return -EFAULT;
1569
Daniel Borkmannbab18992015-10-02 15:17:33 +02001570 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001571 if (ret)
1572 return ret;
1573
1574 __fanout_set_data_bpf(po->fanout, new);
1575 return 0;
1576}
1577
Willem de Bruijnf2e52092015-08-14 22:31:35 -04001578static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1579 unsigned int len)
1580{
1581 struct bpf_prog *new;
1582 u32 fd;
1583
1584 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1585 return -EPERM;
1586 if (len != sizeof(fd))
1587 return -EINVAL;
1588 if (copy_from_user(&fd, data, len))
1589 return -EFAULT;
1590
1591 new = bpf_prog_get(fd);
1592 if (IS_ERR(new))
1593 return PTR_ERR(new);
1594 if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1595 bpf_prog_put(new);
1596 return -EINVAL;
1597 }
1598
1599 __fanout_set_data_bpf(po->fanout, new);
1600 return 0;
1601}
1602
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001603static int fanout_set_data(struct packet_sock *po, char __user *data,
1604 unsigned int len)
1605{
1606 switch (po->fanout->type) {
1607 case PACKET_FANOUT_CBPF:
1608 return fanout_set_data_cbpf(po, data, len);
Willem de Bruijnf2e52092015-08-14 22:31:35 -04001609 case PACKET_FANOUT_EBPF:
1610 return fanout_set_data_ebpf(po, data, len);
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001611 default:
1612 return -EINVAL;
1613 };
1614}
1615
1616static void fanout_release_data(struct packet_fanout *f)
1617{
1618 switch (f->type) {
1619 case PACKET_FANOUT_CBPF:
Willem de Bruijnf2e52092015-08-14 22:31:35 -04001620 case PACKET_FANOUT_EBPF:
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001621 __fanout_set_data_bpf(f, NULL);
1622 };
1623}
1624
David S. Miller7736d332011-07-05 01:43:20 -07001625static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
David S. Millerdc99f602011-07-05 01:45:05 -07001626{
1627 struct packet_sock *po = pkt_sk(sk);
1628 struct packet_fanout *f, *match;
David S. Miller7736d332011-07-05 01:43:20 -07001629 u8 type = type_flags & 0xff;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001630 u8 flags = type_flags >> 8;
David S. Millerdc99f602011-07-05 01:45:05 -07001631 int err;
1632
1633 switch (type) {
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001634 case PACKET_FANOUT_ROLLOVER:
1635 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1636 return -EINVAL;
David S. Millerdc99f602011-07-05 01:45:05 -07001637 case PACKET_FANOUT_HASH:
1638 case PACKET_FANOUT_LB:
David S. Miller95ec3eb2011-07-06 01:56:38 -07001639 case PACKET_FANOUT_CPU:
Daniel Borkmann5df0ddf2013-08-28 22:13:09 +02001640 case PACKET_FANOUT_RND:
Neil Horman2d360972014-01-22 16:01:44 -05001641 case PACKET_FANOUT_QM:
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001642 case PACKET_FANOUT_CBPF:
Willem de Bruijnf2e52092015-08-14 22:31:35 -04001643 case PACKET_FANOUT_EBPF:
David S. Millerdc99f602011-07-05 01:45:05 -07001644 break;
1645 default:
1646 return -EINVAL;
1647 }
1648
1649 if (!po->running)
1650 return -EINVAL;
1651
1652 if (po->fanout)
1653 return -EALREADY;
1654
Willem de Bruijn4633c9e2015-05-17 19:44:02 -04001655 if (type == PACKET_FANOUT_ROLLOVER ||
1656 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
Willem de Bruijn0648ab72015-05-12 11:56:46 -04001657 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1658 if (!po->rollover)
1659 return -ENOMEM;
Willem de Bruijna9b63912015-05-12 11:56:50 -04001660 atomic_long_set(&po->rollover->num, 0);
1661 atomic_long_set(&po->rollover->num_huge, 0);
1662 atomic_long_set(&po->rollover->num_failed, 0);
Willem de Bruijn0648ab72015-05-12 11:56:46 -04001663 }
1664
David S. Millerdc99f602011-07-05 01:45:05 -07001665 mutex_lock(&fanout_mutex);
1666 match = NULL;
1667 list_for_each_entry(f, &fanout_list, list) {
1668 if (f->id == id &&
1669 read_pnet(&f->net) == sock_net(sk)) {
1670 match = f;
1671 break;
1672 }
1673 }
Eric Dumazetafe62c62011-07-07 06:41:29 -07001674 err = -EINVAL;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001675 if (match && match->flags != flags)
Eric Dumazetafe62c62011-07-07 06:41:29 -07001676 goto out;
David S. Millerdc99f602011-07-05 01:45:05 -07001677 if (!match) {
Eric Dumazetafe62c62011-07-07 06:41:29 -07001678 err = -ENOMEM;
David S. Millerdc99f602011-07-05 01:45:05 -07001679 match = kzalloc(sizeof(*match), GFP_KERNEL);
Eric Dumazetafe62c62011-07-07 06:41:29 -07001680 if (!match)
1681 goto out;
1682 write_pnet(&match->net, sock_net(sk));
1683 match->id = id;
1684 match->type = type;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001685 match->flags = flags;
Eric Dumazetafe62c62011-07-07 06:41:29 -07001686 INIT_LIST_HEAD(&match->list);
1687 spin_lock_init(&match->lock);
1688 atomic_set(&match->sk_ref, 0);
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001689 fanout_init_data(match);
Eric Dumazetafe62c62011-07-07 06:41:29 -07001690 match->prot_hook.type = po->prot_hook.type;
1691 match->prot_hook.dev = po->prot_hook.dev;
1692 match->prot_hook.func = packet_rcv_fanout;
1693 match->prot_hook.af_packet_priv = match;
Eric Leblondc0de08d2012-08-16 22:02:58 +00001694 match->prot_hook.id_match = match_fanout_group;
Eric Dumazetafe62c62011-07-07 06:41:29 -07001695 dev_add_pack(&match->prot_hook);
1696 list_add(&match->list, &fanout_list);
1697 }
1698 err = -EINVAL;
1699 if (match->type == type &&
1700 match->prot_hook.type == po->prot_hook.type &&
1701 match->prot_hook.dev == po->prot_hook.dev) {
1702 err = -ENOSPC;
1703 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1704 __dev_remove_pack(&po->prot_hook);
1705 po->fanout = match;
1706 atomic_inc(&match->sk_ref);
1707 __fanout_link(sk, po);
1708 err = 0;
David S. Millerdc99f602011-07-05 01:45:05 -07001709 }
1710 }
Eric Dumazetafe62c62011-07-07 06:41:29 -07001711out:
David S. Millerdc99f602011-07-05 01:45:05 -07001712 mutex_unlock(&fanout_mutex);
Willem de Bruijn0648ab72015-05-12 11:56:46 -04001713 if (err) {
1714 kfree(po->rollover);
1715 po->rollover = NULL;
1716 }
David S. Millerdc99f602011-07-05 01:45:05 -07001717 return err;
1718}
1719
1720static void fanout_release(struct sock *sk)
1721{
1722 struct packet_sock *po = pkt_sk(sk);
1723 struct packet_fanout *f;
1724
1725 f = po->fanout;
1726 if (!f)
1727 return;
1728
Pavel Emelyanovfff33212012-08-16 05:36:48 +00001729 mutex_lock(&fanout_mutex);
David S. Millerdc99f602011-07-05 01:45:05 -07001730 po->fanout = NULL;
1731
David S. Millerdc99f602011-07-05 01:45:05 -07001732 if (atomic_dec_and_test(&f->sk_ref)) {
1733 list_del(&f->list);
1734 dev_remove_pack(&f->prot_hook);
Willem de Bruijn47dceb82015-08-14 22:31:34 -04001735 fanout_release_data(f);
David S. Millerdc99f602011-07-05 01:45:05 -07001736 kfree(f);
1737 }
1738 mutex_unlock(&fanout_mutex);
Willem de Bruijn0648ab72015-05-12 11:56:46 -04001739
Willem de Bruijn59f21112015-06-16 12:51:37 -04001740 if (po->rollover)
1741 kfree_rcu(po->rollover, rcu);
David S. Millerdc99f602011-07-05 01:45:05 -07001742}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08001744static const struct proto_ops packet_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08001746static const struct proto_ops packet_ops_spkt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001748static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1749 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750{
1751 struct sock *sk;
1752 struct sockaddr_pkt *spkt;
1753
1754 /*
1755 * When we registered the protocol we saved the socket in the data
1756 * field for just this event.
1757 */
1758
1759 sk = pt->af_packet_priv;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001760
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 /*
1762 * Yank back the headers [hope the device set this
1763 * right or kerboom...]
1764 *
1765 * Incoming packets have ll header pulled,
1766 * push it back.
1767 *
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001768 * For outgoing ones skb->data == skb_mac_header(skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 * so that this procedure is noop.
1770 */
1771
1772 if (skb->pkt_type == PACKET_LOOPBACK)
1773 goto out;
1774
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001775 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08001776 goto out;
1777
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001778 skb = skb_share_check(skb, GFP_ATOMIC);
1779 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 goto oom;
1781
1782 /* drop any routing info */
Eric Dumazetadf30902009-06-02 05:19:30 +00001783 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
Phil Oester84531c22005-07-12 11:57:52 -07001785 /* drop conntrack reference */
1786 nf_reset(skb);
1787
Herbert Xuffbc6112007-02-04 23:33:10 -08001788 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001790 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
1792 /*
1793 * The SOCK_PACKET socket receives _all_ frames.
1794 */
1795
1796 spkt->spkt_family = dev->type;
1797 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1798 spkt->spkt_protocol = skb->protocol;
1799
1800 /*
1801 * Charge the memory to the socket. This is done specifically
1802 * to prevent sockets using all the memory up.
1803 */
1804
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001805 if (sock_queue_rcv_skb(sk, skb) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 return 0;
1807
1808out:
1809 kfree_skb(skb);
1810oom:
1811 return 0;
1812}
1813
1814
1815/*
1816 * Output a raw packet to a device layer. This bypasses all the other
1817 * protocol layers and you must therefore supply it with a complete frame
1818 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001819
Ying Xue1b784142015-03-02 15:37:48 +08001820static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1821 size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822{
1823 struct sock *sk = sock->sk;
Steffen Hurrle342dfc32014-01-17 22:53:15 +01001824 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001825 struct sk_buff *skb = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 struct net_device *dev;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001827 __be16 proto = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 int err;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001829 int extra_len = 0;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001830
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001832 * Get and verify the address.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 */
1834
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001835 if (saddr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 if (msg->msg_namelen < sizeof(struct sockaddr))
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001837 return -EINVAL;
1838 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1839 proto = saddr->spkt_protocol;
1840 } else
1841 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
1843 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001844 * Find the device first to size check it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 */
1846
danborkmann@iogearbox.netde74e922012-06-10 08:59:28 +00001847 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001848retry:
Eric Dumazet654d1f82009-11-02 10:43:32 +01001849 rcu_read_lock();
1850 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 err = -ENODEV;
1852 if (dev == NULL)
1853 goto out_unlock;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001854
David S. Millerd5e76b02007-01-25 19:30:36 -08001855 err = -ENETDOWN;
1856 if (!(dev->flags & IFF_UP))
1857 goto out_unlock;
1858
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 /*
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001860 * You may not queue a frame bigger than the mtu. This is the lowest level
1861 * raw protocol and you must do your own fragmentation at this level.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001863
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001864 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1865 if (!netif_supports_nofcs(dev)) {
1866 err = -EPROTONOSUPPORT;
1867 goto out_unlock;
1868 }
1869 extra_len = 4; /* We're doing our own CRC */
1870 }
1871
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 err = -EMSGSIZE;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001873 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 goto out_unlock;
1875
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001876 if (!skb) {
1877 size_t reserved = LL_RESERVED_SPACE(dev);
Herbert Xu4ce40912011-11-18 02:20:05 +00001878 int tlen = dev->needed_tailroom;
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001879 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001881 rcu_read_unlock();
Herbert Xu4ce40912011-11-18 02:20:05 +00001882 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001883 if (skb == NULL)
1884 return -ENOBUFS;
1885 /* FIXME: Save some space for broken drivers that write a hard
1886 * header at transmission time by themselves. PPP is the notable
1887 * one here. This should really be fixed at the driver level.
1888 */
1889 skb_reserve(skb, reserved);
1890 skb_reset_network_header(skb);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001891
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001892 /* Try to align data part correctly */
1893 if (hhlen) {
1894 skb->data -= hhlen;
1895 skb->tail -= hhlen;
1896 if (len < hhlen)
1897 skb_reset_network_header(skb);
1898 }
Al Viro6ce8e9c2014-04-06 21:25:44 -04001899 err = memcpy_from_msg(skb_put(skb, len), msg, len);
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001900 if (err)
1901 goto out_free;
1902 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 }
1904
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001905 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
Ben Greear57f89bf2011-02-11 09:35:18 +00001906 /* Earlier code assumed this would be a VLAN pkt,
1907 * double-check this now that we have the actual
1908 * packet in hand.
1909 */
1910 struct ethhdr *ehdr;
1911 skb_reset_mac_header(skb);
1912 ehdr = eth_hdr(skb);
1913 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1914 err = -EMSGSIZE;
1915 goto out_unlock;
1916 }
1917 }
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001918
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 skb->protocol = proto;
1920 skb->dev = dev;
1921 skb->priority = sk->sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00001922 skb->mark = sk->sk_mark;
Daniel Borkmannbf84a0102013-04-14 08:08:13 +00001923
1924 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001926 if (unlikely(extra_len == 4))
1927 skb->no_fcs = 1;
1928
Jason Wang40893fd2013-03-26 23:11:22 +00001929 skb_probe_transport_header(skb, 0);
Jason Wangc1aad272013-03-25 20:19:57 +00001930
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 dev_queue_xmit(skb);
Eric Dumazet654d1f82009-11-02 10:43:32 +01001932 rcu_read_unlock();
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001933 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935out_unlock:
Eric Dumazet654d1f82009-11-02 10:43:32 +01001936 rcu_read_unlock();
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001937out_free:
1938 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 return err;
1940}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
Alexei Starovoitovff936a02015-10-07 10:55:41 -07001942static unsigned int run_filter(struct sk_buff *skb,
1943 const struct sock *sk,
1944 unsigned int res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945{
1946 struct sk_filter *filter;
1947
Eric Dumazet80f8f102011-01-18 07:46:52 +00001948 rcu_read_lock();
1949 filter = rcu_dereference(sk->sk_filter);
David S. Millerdbcb5852007-01-24 15:21:02 -08001950 if (filter != NULL)
Alexei Starovoitovff936a02015-10-07 10:55:41 -07001951 res = bpf_prog_run_clear_cb(filter->prog, skb);
Eric Dumazet80f8f102011-01-18 07:46:52 +00001952 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
David S. Millerdbcb5852007-01-24 15:21:02 -08001954 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955}
1956
1957/*
Eric Dumazet62ab0812010-12-06 20:50:09 +00001958 * This function makes lazy skb cloning in hope that most of packets
1959 * are discarded by BPF.
1960 *
1961 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1962 * and skb->cb are mangled. It works because (and until) packets
1963 * falling here are owned by current CPU. Output packets are cloned
1964 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1965 * sequencially, so that if we return skb to original state on exit,
1966 * we will not harm anyone.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 */
1968
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001969static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1970 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971{
1972 struct sock *sk;
1973 struct sockaddr_ll *sll;
1974 struct packet_sock *po;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001975 u8 *skb_head = skb->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 int skb_len = skb->len;
David S. Millerdbcb5852007-01-24 15:21:02 -08001977 unsigned int snaplen, res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
1979 if (skb->pkt_type == PACKET_LOOPBACK)
1980 goto drop;
1981
1982 sk = pt->af_packet_priv;
1983 po = pkt_sk(sk);
1984
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001985 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08001986 goto drop;
1987
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 skb->dev = dev;
1989
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -07001990 if (dev->header_ops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 /* The device has an explicit notion of ll header,
Eric Dumazet62ab0812010-12-06 20:50:09 +00001992 * exported to higher levels.
1993 *
1994 * Otherwise, the device hides details of its frame
1995 * structure, so that corresponding packet head is
1996 * never delivered to user.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 */
1998 if (sk->sk_type != SOCK_DGRAM)
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001999 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 else if (skb->pkt_type == PACKET_OUTGOING) {
2001 /* Special case: outgoing packets have ll header at head */
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03002002 skb_pull(skb, skb_network_offset(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 }
2004 }
2005
2006 snaplen = skb->len;
2007
David S. Millerdbcb5852007-01-24 15:21:02 -08002008 res = run_filter(skb, sk, snaplen);
2009 if (!res)
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07002010 goto drop_n_restore;
David S. Millerdbcb5852007-01-24 15:21:02 -08002011 if (snaplen > res)
2012 snaplen = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
Eric Dumazet0fd7bac2011-12-21 07:11:44 +00002014 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 goto drop_n_acct;
2016
2017 if (skb_shared(skb)) {
2018 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2019 if (nskb == NULL)
2020 goto drop_n_acct;
2021
2022 if (skb_head != skb->data) {
2023 skb->data = skb_head;
2024 skb->len = skb_len;
2025 }
Eric Dumazetabc4e4f2012-04-19 02:24:42 +00002026 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 skb = nskb;
2028 }
2029
Eyal Birgerb4772ef2015-03-01 14:58:29 +02002030 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
Herbert Xuffbc6112007-02-04 23:33:10 -08002031
2032 sll = &PACKET_SKB_CB(skb)->sa.ll;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 sll->sll_hatype = dev->type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 sll->sll_pkttype = skb->pkt_type;
Peter P Waskiewicz Jr8032b462007-11-10 22:03:25 -08002035 if (unlikely(po->origdev))
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07002036 sll->sll_ifindex = orig_dev->ifindex;
2037 else
2038 sll->sll_ifindex = dev->ifindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039
Stephen Hemmingerb95cce32007-09-26 22:13:38 -07002040 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Eyal Birger2472d762015-03-01 14:58:28 +02002042 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2043 * Use their space for storing the original skb length.
2044 */
2045 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
Herbert Xu8dc41942007-02-04 23:31:32 -08002046
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 if (pskb_trim(skb, snaplen))
2048 goto drop_n_acct;
2049
2050 skb_set_owner_r(skb, sk);
2051 skb->dev = NULL;
Eric Dumazetadf30902009-06-02 05:19:30 +00002052 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Phil Oester84531c22005-07-12 11:57:52 -07002054 /* drop conntrack reference */
2055 nf_reset(skb);
2056
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 spin_lock(&sk->sk_receive_queue.lock);
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00002058 po->stats.stats1.tp_packets++;
Eyal Birger3bc3b962015-03-01 14:58:30 +02002059 sock_skb_set_dropcount(sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 __skb_queue_tail(&sk->sk_receive_queue, skb);
2061 spin_unlock(&sk->sk_receive_queue.lock);
David S. Miller676d2362014-04-11 16:15:36 -04002062 sk->sk_data_ready(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 return 0;
2064
2065drop_n_acct:
Willem de Bruijn7091fbd2011-09-30 10:38:28 +00002066 spin_lock(&sk->sk_receive_queue.lock);
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00002067 po->stats.stats1.tp_drops++;
Willem de Bruijn7091fbd2011-09-30 10:38:28 +00002068 atomic_inc(&sk->sk_drops);
2069 spin_unlock(&sk->sk_receive_queue.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071drop_n_restore:
2072 if (skb_head != skb->data && skb_shared(skb)) {
2073 skb->data = skb_head;
2074 skb->len = skb_len;
2075 }
2076drop:
Neil Hormanead2ceb2009-03-11 09:49:55 +00002077 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 return 0;
2079}
2080
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002081static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2082 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
2084 struct sock *sk;
2085 struct packet_sock *po;
2086 struct sockaddr_ll *sll;
Daniel Borkmann184f4892013-04-16 01:57:46 +00002087 union tpacket_uhdr h;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002088 u8 *skb_head = skb->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 int skb_len = skb->len;
David S. Millerdbcb5852007-01-24 15:21:02 -08002090 unsigned int snaplen, res;
chetan lokef6fb8f12011-08-19 10:18:16 +00002091 unsigned long status = TP_STATUS_USER;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002092 unsigned short macoff, netoff, hdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 struct sk_buff *copy_skb = NULL;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002094 struct timespec ts;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002095 __u32 ts_status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096
Atzm Watanabe51846352013-12-17 22:53:32 +09002097 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2098 * We may add members to them until current aligned size without forcing
2099 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2100 */
2101 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2102 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2103
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 if (skb->pkt_type == PACKET_LOOPBACK)
2105 goto drop;
2106
2107 sk = pt->af_packet_priv;
2108 po = pkt_sk(sk);
2109
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08002110 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08002111 goto drop;
2112
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -07002113 if (dev->header_ops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 if (sk->sk_type != SOCK_DGRAM)
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07002115 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 else if (skb->pkt_type == PACKET_OUTGOING) {
2117 /* Special case: outgoing packets have ll header at head */
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03002118 skb_pull(skb, skb_network_offset(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 }
2120 }
2121
2122 snaplen = skb->len;
2123
David S. Millerdbcb5852007-01-24 15:21:02 -08002124 res = run_filter(skb, sk, snaplen);
2125 if (!res)
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07002126 goto drop_n_restore;
Alexander Drozdov68c2e5d2015-03-23 09:11:12 +03002127
2128 if (skb->ip_summed == CHECKSUM_PARTIAL)
2129 status |= TP_STATUS_CSUMNOTREADY;
Alexander Drozdov682f0482015-03-23 09:11:13 +03002130 else if (skb->pkt_type != PACKET_OUTGOING &&
2131 (skb->ip_summed == CHECKSUM_COMPLETE ||
2132 skb_csum_unnecessary(skb)))
2133 status |= TP_STATUS_CSUM_VALID;
Alexander Drozdov68c2e5d2015-03-23 09:11:12 +03002134
David S. Millerdbcb5852007-01-24 15:21:02 -08002135 if (snaplen > res)
2136 snaplen = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137
2138 if (sk->sk_type == SOCK_DGRAM) {
Patrick McHardy8913336a2008-07-18 18:05:19 -07002139 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2140 po->tp_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 } else {
Eric Dumazet95c96172012-04-15 05:58:06 +00002142 unsigned int maclen = skb_network_offset(skb);
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002143 netoff = TPACKET_ALIGN(po->tp_hdrlen +
Patrick McHardy8913336a2008-07-18 18:05:19 -07002144 (maclen < 16 ? 16 : maclen)) +
2145 po->tp_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 macoff = netoff - maclen;
2147 }
chetan lokef6fb8f12011-08-19 10:18:16 +00002148 if (po->tp_version <= TPACKET_V2) {
2149 if (macoff + snaplen > po->rx_ring.frame_size) {
2150 if (po->copy_thresh &&
Eric Dumazet0fd7bac2011-12-21 07:11:44 +00002151 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
chetan lokef6fb8f12011-08-19 10:18:16 +00002152 if (skb_shared(skb)) {
2153 copy_skb = skb_clone(skb, GFP_ATOMIC);
2154 } else {
2155 copy_skb = skb_get(skb);
2156 skb_head = skb->data;
2157 }
2158 if (copy_skb)
2159 skb_set_owner_r(copy_skb, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 }
chetan lokef6fb8f12011-08-19 10:18:16 +00002161 snaplen = po->rx_ring.frame_size - macoff;
2162 if ((int)snaplen < 0)
2163 snaplen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 }
Eric Dumazetdc808112014-08-15 09:16:04 -07002165 } else if (unlikely(macoff + snaplen >
2166 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2167 u32 nval;
2168
2169 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2170 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2171 snaplen, nval, macoff);
2172 snaplen = nval;
2173 if (unlikely((int)snaplen < 0)) {
2174 snaplen = 0;
2175 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2176 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 spin_lock(&sk->sk_receive_queue.lock);
chetan lokef6fb8f12011-08-19 10:18:16 +00002179 h.raw = packet_current_rx_frame(po, skb,
2180 TP_STATUS_KERNEL, (macoff+snaplen));
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002181 if (!h.raw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 goto ring_is_full;
chetan lokef6fb8f12011-08-19 10:18:16 +00002183 if (po->tp_version <= TPACKET_V2) {
2184 packet_increment_rx_head(po, &po->rx_ring);
2185 /*
2186 * LOSING will be reported till you read the stats,
2187 * because it's COR - Clear On Read.
2188 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2189 * at packet level.
2190 */
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00002191 if (po->stats.stats1.tp_drops)
chetan lokef6fb8f12011-08-19 10:18:16 +00002192 status |= TP_STATUS_LOSING;
2193 }
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00002194 po->stats.stats1.tp_packets++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 if (copy_skb) {
2196 status |= TP_STATUS_COPY;
2197 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2198 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 spin_unlock(&sk->sk_receive_queue.lock);
2200
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002201 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002202
2203 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
Daniel Borkmann7a513842013-04-23 00:39:29 +00002204 getnstimeofday(&ts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002206 status |= ts_status;
2207
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002208 switch (po->tp_version) {
2209 case TPACKET_V1:
2210 h.h1->tp_len = skb->len;
2211 h.h1->tp_snaplen = snaplen;
2212 h.h1->tp_mac = macoff;
2213 h.h1->tp_net = netoff;
Daniel Borkmann4b457bd2013-04-16 01:29:11 +00002214 h.h1->tp_sec = ts.tv_sec;
2215 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002216 hdrlen = sizeof(*h.h1);
2217 break;
2218 case TPACKET_V2:
2219 h.h2->tp_len = skb->len;
2220 h.h2->tp_snaplen = snaplen;
2221 h.h2->tp_mac = macoff;
2222 h.h2->tp_net = netoff;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002223 h.h2->tp_sec = ts.tv_sec;
2224 h.h2->tp_nsec = ts.tv_nsec;
Jiri Pirkodf8a39de2015-01-13 17:13:44 +01002225 if (skb_vlan_tag_present(skb)) {
2226 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09002227 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2228 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
Ben Greeara3bcc232011-06-01 06:49:10 +00002229 } else {
2230 h.h2->tp_vlan_tci = 0;
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09002231 h.h2->tp_vlan_tpid = 0;
Ben Greeara3bcc232011-06-01 06:49:10 +00002232 }
Atzm Watanabee4d26f42013-12-17 22:53:36 +09002233 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002234 hdrlen = sizeof(*h.h2);
2235 break;
chetan lokef6fb8f12011-08-19 10:18:16 +00002236 case TPACKET_V3:
2237 /* tp_nxt_offset,vlan are already populated above.
2238 * So DONT clear those fields here
2239 */
2240 h.h3->tp_status |= status;
2241 h.h3->tp_len = skb->len;
2242 h.h3->tp_snaplen = snaplen;
2243 h.h3->tp_mac = macoff;
2244 h.h3->tp_net = netoff;
chetan lokef6fb8f12011-08-19 10:18:16 +00002245 h.h3->tp_sec = ts.tv_sec;
2246 h.h3->tp_nsec = ts.tv_nsec;
Atzm Watanabee4d26f42013-12-17 22:53:36 +09002247 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
chetan lokef6fb8f12011-08-19 10:18:16 +00002248 hdrlen = sizeof(*h.h3);
2249 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002250 default:
2251 BUG();
2252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002254 sll = h.raw + TPACKET_ALIGN(hdrlen);
Stephen Hemmingerb95cce32007-09-26 22:13:38 -07002255 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 sll->sll_family = AF_PACKET;
2257 sll->sll_hatype = dev->type;
2258 sll->sll_protocol = skb->protocol;
2259 sll->sll_pkttype = skb->pkt_type;
Peter P Waskiewicz Jr8032b462007-11-10 22:03:25 -08002260 if (unlikely(po->origdev))
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07002261 sll->sll_ifindex = orig_dev->ifindex;
2262 else
2263 sll->sll_ifindex = dev->ifindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
Ralf Baechlee16aa202006-12-07 00:11:33 -08002265 smp_mb();
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002266
Changli Gaof6dafa92010-12-07 04:26:16 +00002267#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002268 if (po->tp_version <= TPACKET_V2) {
Changli Gao0af55bb2010-12-01 02:52:20 +00002269 u8 *start, *end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002271 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2272 macoff + snaplen);
2273
2274 for (start = h.raw; start < end; start += PAGE_SIZE)
2275 flush_dcache_page(pgv_to_page(start));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 }
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002277 smp_wmb();
Changli Gaof6dafa92010-12-07 04:26:16 +00002278#endif
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002279
Dan Collinsda413ee2014-12-19 16:49:25 +13002280 if (po->tp_version <= TPACKET_V2) {
chetan lokef6fb8f12011-08-19 10:18:16 +00002281 __packet_set_status(po, h.raw, status);
Dan Collinsda413ee2014-12-19 16:49:25 +13002282 sk->sk_data_ready(sk);
2283 } else {
chetan lokef6fb8f12011-08-19 10:18:16 +00002284 prb_clear_blk_fill_status(&po->rx_ring);
Dan Collinsda413ee2014-12-19 16:49:25 +13002285 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286
2287drop_n_restore:
2288 if (skb_head != skb->data && skb_shared(skb)) {
2289 skb->data = skb_head;
2290 skb->len = skb_len;
2291 }
2292drop:
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002293 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 return 0;
2295
2296ring_is_full:
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00002297 po->stats.stats1.tp_drops++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 spin_unlock(&sk->sk_receive_queue.lock);
2299
David S. Miller676d2362014-04-11 16:15:36 -04002300 sk->sk_data_ready(sk);
Wei Yongjunacb5d752009-02-25 00:36:42 +00002301 kfree_skb(copy_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 goto drop_n_restore;
2303}
2304
Johann Baudy69e3c752009-05-18 22:11:22 -07002305static void tpacket_destruct_skb(struct sk_buff *skb)
2306{
2307 struct packet_sock *po = pkt_sk(skb->sk);
Johann Baudy69e3c752009-05-18 22:11:22 -07002308
Johann Baudy69e3c752009-05-18 22:11:22 -07002309 if (likely(po->tx_ring.pg_vec)) {
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002310 void *ph;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002311 __u32 ts;
2312
Johann Baudy69e3c752009-05-18 22:11:22 -07002313 ph = skb_shinfo(skb)->destructor_arg;
Daniel Borkmannb0138402014-01-15 16:25:36 +01002314 packet_dec_pending(&po->tx_ring);
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002315
2316 ts = __packet_set_timestamp(po, ph, skb);
2317 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
Johann Baudy69e3c752009-05-18 22:11:22 -07002318 }
2319
2320 sock_wfree(skb);
2321}
2322
Willem de Bruijn9c707762014-11-19 13:10:16 -05002323static bool ll_header_truncated(const struct net_device *dev, int len)
2324{
2325 /* net device doesn't like empty head */
2326 if (unlikely(len <= dev->hard_header_len)) {
Willem de Bruijneee2f042015-01-08 11:29:18 -05002327 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
Willem de Bruijn9c707762014-11-19 13:10:16 -05002328 current->comm, len, dev->hard_header_len);
2329 return true;
2330 }
2331
2332 return false;
2333}
2334
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002335static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2336 void *frame, struct net_device *dev, int size_max,
Herbert Xuae641942011-11-18 02:20:04 +00002337 __be16 proto, unsigned char *addr, int hlen)
Johann Baudy69e3c752009-05-18 22:11:22 -07002338{
Daniel Borkmann184f4892013-04-16 01:57:46 +00002339 union tpacket_uhdr ph;
David S. Miller09effa62013-08-07 17:11:00 -07002340 int to_write, offset, len, tp_len, nr_frags, len_max;
Johann Baudy69e3c752009-05-18 22:11:22 -07002341 struct socket *sock = po->sk.sk_socket;
2342 struct page *page;
2343 void *data;
2344 int err;
2345
2346 ph.raw = frame;
2347
2348 skb->protocol = proto;
2349 skb->dev = dev;
2350 skb->priority = po->sk.sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00002351 skb->mark = po->sk.sk_mark;
Willem de Bruijn2e313962013-04-23 00:39:28 +00002352 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
Johann Baudy69e3c752009-05-18 22:11:22 -07002353 skb_shinfo(skb)->destructor_arg = ph.raw;
2354
2355 switch (po->tp_version) {
2356 case TPACKET_V2:
2357 tp_len = ph.h2->tp_len;
2358 break;
2359 default:
2360 tp_len = ph.h1->tp_len;
2361 break;
2362 }
David S. Miller09effa62013-08-07 17:11:00 -07002363 if (unlikely(tp_len > size_max)) {
2364 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2365 return -EMSGSIZE;
2366 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002367
Herbert Xuae641942011-11-18 02:20:04 +00002368 skb_reserve(skb, hlen);
Johann Baudy69e3c752009-05-18 22:11:22 -07002369 skb_reset_network_header(skb);
Jason Wangc1aad272013-03-25 20:19:57 +00002370
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002371 if (!packet_use_direct_xmit(po))
2372 skb_probe_transport_header(skb, 0);
2373 if (unlikely(po->tp_tx_has_off)) {
Paul Chavent5920cd3a2012-11-06 23:10:47 +00002374 int off_min, off_max, off;
2375 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2376 off_max = po->tx_ring.frame_size - tp_len;
2377 if (sock->type == SOCK_DGRAM) {
2378 switch (po->tp_version) {
2379 case TPACKET_V2:
2380 off = ph.h2->tp_net;
2381 break;
2382 default:
2383 off = ph.h1->tp_net;
2384 break;
2385 }
2386 } else {
2387 switch (po->tp_version) {
2388 case TPACKET_V2:
2389 off = ph.h2->tp_mac;
2390 break;
2391 default:
2392 off = ph.h1->tp_mac;
2393 break;
2394 }
2395 }
2396 if (unlikely((off < off_min) || (off_max < off)))
2397 return -EINVAL;
2398 data = ph.raw + off;
2399 } else {
2400 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2401 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002402 to_write = tp_len;
2403
2404 if (sock->type == SOCK_DGRAM) {
2405 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2406 NULL, tp_len);
2407 if (unlikely(err < 0))
2408 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002409 } else if (dev->hard_header_len) {
Willem de Bruijn9c707762014-11-19 13:10:16 -05002410 if (ll_header_truncated(dev, tp_len))
Johann Baudy69e3c752009-05-18 22:11:22 -07002411 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07002412
2413 skb_push(skb, dev->hard_header_len);
2414 err = skb_store_bits(skb, 0, data,
2415 dev->hard_header_len);
2416 if (unlikely(err))
2417 return err;
2418
2419 data += dev->hard_header_len;
2420 to_write -= dev->hard_header_len;
2421 }
2422
Johann Baudy69e3c752009-05-18 22:11:22 -07002423 offset = offset_in_page(data);
2424 len_max = PAGE_SIZE - offset;
2425 len = ((to_write > len_max) ? len_max : to_write);
2426
2427 skb->data_len = to_write;
2428 skb->len += to_write;
2429 skb->truesize += to_write;
2430 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2431
2432 while (likely(to_write)) {
2433 nr_frags = skb_shinfo(skb)->nr_frags;
2434
2435 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002436 pr_err("Packet exceed the number of skb frags(%lu)\n",
2437 MAX_SKB_FRAGS);
Johann Baudy69e3c752009-05-18 22:11:22 -07002438 return -EFAULT;
2439 }
2440
Changli Gao0af55bb2010-12-01 02:52:20 +00002441 page = pgv_to_page(data);
2442 data += len;
Johann Baudy69e3c752009-05-18 22:11:22 -07002443 flush_dcache_page(page);
2444 get_page(page);
Changli Gao0af55bb2010-12-01 02:52:20 +00002445 skb_fill_page_desc(skb, nr_frags, page, offset, len);
Johann Baudy69e3c752009-05-18 22:11:22 -07002446 to_write -= len;
2447 offset = 0;
2448 len_max = PAGE_SIZE;
2449 len = ((to_write > len_max) ? len_max : to_write);
2450 }
2451
2452 return tp_len;
2453}
2454
2455static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2456{
Johann Baudy69e3c752009-05-18 22:11:22 -07002457 struct sk_buff *skb;
2458 struct net_device *dev;
2459 __be16 proto;
David S. Miller09effa62013-08-07 17:11:00 -07002460 int err, reserve = 0;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002461 void *ph;
Steffen Hurrle342dfc32014-01-17 22:53:15 +01002462 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002463 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
Johann Baudy69e3c752009-05-18 22:11:22 -07002464 int tp_len, size_max;
2465 unsigned char *addr;
2466 int len_sum = 0;
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +00002467 int status = TP_STATUS_AVAILABLE;
Herbert Xuae641942011-11-18 02:20:04 +00002468 int hlen, tlen;
Johann Baudy69e3c752009-05-18 22:11:22 -07002469
Johann Baudy69e3c752009-05-18 22:11:22 -07002470 mutex_lock(&po->pg_vec_lock);
2471
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002472 if (likely(saddr == NULL)) {
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002473 dev = packet_cached_dev_get(po);
Johann Baudy69e3c752009-05-18 22:11:22 -07002474 proto = po->num;
2475 addr = NULL;
2476 } else {
2477 err = -EINVAL;
2478 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2479 goto out;
2480 if (msg->msg_namelen < (saddr->sll_halen
2481 + offsetof(struct sockaddr_ll,
2482 sll_addr)))
2483 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07002484 proto = saddr->sll_protocol;
2485 addr = saddr->sll_addr;
Ben Greear827d9782011-06-01 07:18:53 +00002486 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
Johann Baudy69e3c752009-05-18 22:11:22 -07002487 }
2488
Johann Baudy69e3c752009-05-18 22:11:22 -07002489 err = -ENXIO;
2490 if (unlikely(dev == NULL))
2491 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07002492 err = -ENETDOWN;
2493 if (unlikely(!(dev->flags & IFF_UP)))
2494 goto out_put;
2495
Daniel Borkmann52f14542014-02-28 02:22:06 +01002496 reserve = dev->hard_header_len + VLAN_HLEN;
Johann Baudy69e3c752009-05-18 22:11:22 -07002497 size_max = po->tx_ring.frame_size
Gabor Gombasb5dd8842009-10-29 03:19:11 -07002498 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
Johann Baudy69e3c752009-05-18 22:11:22 -07002499
David S. Miller09effa62013-08-07 17:11:00 -07002500 if (size_max > dev->mtu + reserve)
2501 size_max = dev->mtu + reserve;
2502
Johann Baudy69e3c752009-05-18 22:11:22 -07002503 do {
2504 ph = packet_current_frame(po, &po->tx_ring,
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002505 TP_STATUS_SEND_REQUEST);
Johann Baudy69e3c752009-05-18 22:11:22 -07002506 if (unlikely(ph == NULL)) {
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002507 if (need_wait && need_resched())
2508 schedule();
Johann Baudy69e3c752009-05-18 22:11:22 -07002509 continue;
2510 }
2511
2512 status = TP_STATUS_SEND_REQUEST;
Herbert Xuae641942011-11-18 02:20:04 +00002513 hlen = LL_RESERVED_SPACE(dev);
2514 tlen = dev->needed_tailroom;
Johann Baudy69e3c752009-05-18 22:11:22 -07002515 skb = sock_alloc_send_skb(&po->sk,
Herbert Xuae641942011-11-18 02:20:04 +00002516 hlen + tlen + sizeof(struct sockaddr_ll),
Kretschmer, Mathiasfbf33a22015-05-08 15:44:37 +02002517 !need_wait, &err);
Johann Baudy69e3c752009-05-18 22:11:22 -07002518
Kretschmer, Mathiasfbf33a22015-05-08 15:44:37 +02002519 if (unlikely(skb == NULL)) {
2520 /* we assume the socket was initially writeable ... */
2521 if (likely(len_sum > 0))
2522 err = len_sum;
Johann Baudy69e3c752009-05-18 22:11:22 -07002523 goto out_status;
Kretschmer, Mathiasfbf33a22015-05-08 15:44:37 +02002524 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002525 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
Daniel Borkmann52f14542014-02-28 02:22:06 +01002526 addr, hlen);
Alexander Drozdovdbd46ab2015-07-28 13:57:01 +03002527 if (likely(tp_len >= 0) &&
2528 tp_len > dev->mtu + dev->hard_header_len) {
Daniel Borkmann52f14542014-02-28 02:22:06 +01002529 struct ethhdr *ehdr;
2530 /* Earlier code assumed this would be a VLAN pkt,
2531 * double-check this now that we have the actual
2532 * packet in hand.
2533 */
Johann Baudy69e3c752009-05-18 22:11:22 -07002534
Daniel Borkmann52f14542014-02-28 02:22:06 +01002535 skb_reset_mac_header(skb);
2536 ehdr = eth_hdr(skb);
2537 if (ehdr->h_proto != htons(ETH_P_8021Q))
2538 tp_len = -EMSGSIZE;
2539 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002540 if (unlikely(tp_len < 0)) {
2541 if (po->tp_loss) {
2542 __packet_set_status(po, ph,
2543 TP_STATUS_AVAILABLE);
2544 packet_increment_head(&po->tx_ring);
2545 kfree_skb(skb);
2546 continue;
2547 } else {
2548 status = TP_STATUS_WRONG_FORMAT;
2549 err = tp_len;
2550 goto out_status;
2551 }
2552 }
2553
Daniel Borkmann0fd5d572014-02-16 15:55:22 +01002554 packet_pick_tx_queue(dev, skb);
2555
Johann Baudy69e3c752009-05-18 22:11:22 -07002556 skb->destructor = tpacket_destruct_skb;
2557 __packet_set_status(po, ph, TP_STATUS_SENDING);
Daniel Borkmannb0138402014-01-15 16:25:36 +01002558 packet_inc_pending(&po->tx_ring);
Johann Baudy69e3c752009-05-18 22:11:22 -07002559
2560 status = TP_STATUS_SEND_REQUEST;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002561 err = po->xmit(skb);
Jarek Poplawskieb70df12010-01-10 22:04:19 +00002562 if (unlikely(err > 0)) {
2563 err = net_xmit_errno(err);
2564 if (err && __packet_get_status(po, ph) ==
2565 TP_STATUS_AVAILABLE) {
2566 /* skb was destructed already */
2567 skb = NULL;
2568 goto out_status;
2569 }
2570 /*
2571 * skb was dropped but not destructed yet;
2572 * let's treat it like congestion or err < 0
2573 */
2574 err = 0;
2575 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002576 packet_increment_head(&po->tx_ring);
2577 len_sum += tp_len;
Daniel Borkmannb0138402014-01-15 16:25:36 +01002578 } while (likely((ph != NULL) ||
2579 /* Note: packet_read_pending() might be slow if we have
2580 * to call it as it's per_cpu variable, but in fast-path
2581 * we already short-circuit the loop with the first
2582 * condition, and luckily don't have to go that path
2583 * anyway.
2584 */
2585 (need_wait && packet_read_pending(&po->tx_ring))));
Johann Baudy69e3c752009-05-18 22:11:22 -07002586
2587 err = len_sum;
2588 goto out_put;
2589
Johann Baudy69e3c752009-05-18 22:11:22 -07002590out_status:
2591 __packet_set_status(po, ph, status);
2592 kfree_skb(skb);
2593out_put:
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002594 dev_put(dev);
Johann Baudy69e3c752009-05-18 22:11:22 -07002595out:
2596 mutex_unlock(&po->pg_vec_lock);
2597 return err;
2598}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599
Olof Johanssoneea49cc92011-11-02 11:00:49 +00002600static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2601 size_t reserve, size_t len,
2602 size_t linear, int noblock,
2603 int *err)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002604{
2605 struct sk_buff *skb;
2606
2607 /* Under a page? Don't bother with paged skb. */
2608 if (prepad + len < PAGE_SIZE || !linear)
2609 linear = len;
2610
2611 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
Eric Dumazet28d64272013-08-08 14:38:47 -07002612 err, 0);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002613 if (!skb)
2614 return NULL;
2615
2616 skb_reserve(skb, reserve);
2617 skb_put(skb, linear);
2618 skb->data_len = len - linear;
2619 skb->len += len - linear;
2620
2621 return skb;
2622}
2623
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002624static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625{
2626 struct sock *sk = sock->sk;
Steffen Hurrle342dfc32014-01-17 22:53:15 +01002627 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 struct sk_buff *skb;
2629 struct net_device *dev;
Al Viro0e11c912006-11-08 00:26:29 -08002630 __be16 proto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 unsigned char *addr;
Ben Greear827d9782011-06-01 07:18:53 +00002632 int err, reserve = 0;
Edward Jeec7d39e32015-10-08 14:56:49 -07002633 struct sockcm_cookie sockc;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002634 struct virtio_net_hdr vnet_hdr = { 0 };
2635 int offset = 0;
2636 int vnet_hdr_len;
2637 struct packet_sock *po = pkt_sk(sk);
2638 unsigned short gso_type = 0;
Herbert Xuae641942011-11-18 02:20:04 +00002639 int hlen, tlen;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002640 int extra_len = 0;
Al Viro8feb2fb2014-11-06 01:10:59 -05002641 ssize_t n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
2643 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002644 * Get and verify the address.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002646
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002647 if (likely(saddr == NULL)) {
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002648 dev = packet_cached_dev_get(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 proto = po->num;
2650 addr = NULL;
2651 } else {
2652 err = -EINVAL;
2653 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2654 goto out;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002655 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2656 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 proto = saddr->sll_protocol;
2658 addr = saddr->sll_addr;
Ben Greear827d9782011-06-01 07:18:53 +00002659 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 }
2661
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662 err = -ENXIO;
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002663 if (unlikely(dev == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 goto out_unlock;
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002665 err = -ENETDOWN;
2666 if (unlikely(!(dev->flags & IFF_UP)))
2667 goto out_unlock;
2668
Edward Jeec7d39e32015-10-08 14:56:49 -07002669 sockc.mark = sk->sk_mark;
2670 if (msg->msg_controllen) {
2671 err = sock_cmsg_send(sk, msg, &sockc);
2672 if (unlikely(err))
2673 goto out_unlock;
2674 }
2675
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 if (sock->type == SOCK_RAW)
2677 reserve = dev->hard_header_len;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002678 if (po->has_vnet_hdr) {
2679 vnet_hdr_len = sizeof(vnet_hdr);
2680
2681 err = -EINVAL;
2682 if (len < vnet_hdr_len)
2683 goto out_unlock;
2684
2685 len -= vnet_hdr_len;
2686
Al Viro8feb2fb2014-11-06 01:10:59 -05002687 err = -EFAULT;
Al Viroc0371da2014-11-24 10:42:55 -05002688 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
Al Viro8feb2fb2014-11-06 01:10:59 -05002689 if (n != vnet_hdr_len)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002690 goto out_unlock;
2691
2692 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
David Woodhoused3869ef2015-09-23 19:45:08 +01002693 (__virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2694 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2 >
2695 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len)))
2696 vnet_hdr.hdr_len = __cpu_to_virtio16(vio_le(),
2697 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start) +
2698 __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset) + 2);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002699
2700 err = -EINVAL;
David Woodhoused3869ef2015-09-23 19:45:08 +01002701 if (__virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len) > len)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002702 goto out_unlock;
2703
2704 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2705 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2706 case VIRTIO_NET_HDR_GSO_TCPV4:
2707 gso_type = SKB_GSO_TCPV4;
2708 break;
2709 case VIRTIO_NET_HDR_GSO_TCPV6:
2710 gso_type = SKB_GSO_TCPV6;
2711 break;
2712 case VIRTIO_NET_HDR_GSO_UDP:
2713 gso_type = SKB_GSO_UDP;
2714 break;
2715 default:
2716 goto out_unlock;
2717 }
2718
2719 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2720 gso_type |= SKB_GSO_TCP_ECN;
2721
2722 if (vnet_hdr.gso_size == 0)
2723 goto out_unlock;
2724
2725 }
2726 }
2727
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002728 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2729 if (!netif_supports_nofcs(dev)) {
2730 err = -EPROTONOSUPPORT;
2731 goto out_unlock;
2732 }
2733 extra_len = 4; /* We're doing our own CRC */
2734 }
2735
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 err = -EMSGSIZE;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002737 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 goto out_unlock;
2739
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002740 err = -ENOBUFS;
Herbert Xuae641942011-11-18 02:20:04 +00002741 hlen = LL_RESERVED_SPACE(dev);
2742 tlen = dev->needed_tailroom;
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002743 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
David Woodhoused3869ef2015-09-23 19:45:08 +01002744 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002745 msg->msg_flags & MSG_DONTWAIT, &err);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002746 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 goto out_unlock;
2748
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002749 skb_set_network_header(skb, reserve);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750
Stephen Hemminger0c4e8582007-10-09 01:36:32 -07002751 err = -EINVAL;
Willem de Bruijn9c707762014-11-19 13:10:16 -05002752 if (sock->type == SOCK_DGRAM) {
2753 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
Christoph Jaeger46d2cfb2015-01-11 13:01:16 -05002754 if (unlikely(offset < 0))
Willem de Bruijn9c707762014-11-19 13:10:16 -05002755 goto out_free;
2756 } else {
2757 if (ll_header_truncated(dev, len))
2758 goto out_free;
2759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760
2761 /* Returns -EFAULT on error */
Al Viroc0371da2014-11-24 10:42:55 -05002762 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 if (err)
2764 goto out_free;
Daniel Borkmannbf84a0102013-04-14 08:08:13 +00002765
2766 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002768 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
David S. Miller09effa62013-08-07 17:11:00 -07002769 /* Earlier code assumed this would be a VLAN pkt,
2770 * double-check this now that we have the actual
2771 * packet in hand.
2772 */
2773 struct ethhdr *ehdr;
2774 skb_reset_mac_header(skb);
2775 ehdr = eth_hdr(skb);
2776 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2777 err = -EMSGSIZE;
2778 goto out_free;
2779 }
Ben Greear57f89bf2011-02-11 09:35:18 +00002780 }
2781
David S. Miller09effa62013-08-07 17:11:00 -07002782 skb->protocol = proto;
2783 skb->dev = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 skb->priority = sk->sk_priority;
Edward Jeec7d39e32015-10-08 14:56:49 -07002785 skb->mark = sockc.mark;
Daniel Borkmann0fd5d572014-02-16 15:55:22 +01002786
2787 packet_pick_tx_queue(dev, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002789 if (po->has_vnet_hdr) {
2790 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
David Woodhoused3869ef2015-09-23 19:45:08 +01002791 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_start);
2792 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr.csum_offset);
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002793 if (!skb_partial_csum_set(skb, s, o)) {
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002794 err = -EINVAL;
2795 goto out_free;
2796 }
2797 }
2798
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002799 skb_shinfo(skb)->gso_size =
David Woodhoused3869ef2015-09-23 19:45:08 +01002800 __virtio16_to_cpu(vio_le(), vnet_hdr.gso_size);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002801 skb_shinfo(skb)->gso_type = gso_type;
2802
2803 /* Header must be checked, and gso_segs computed. */
2804 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2805 skb_shinfo(skb)->gso_segs = 0;
2806
2807 len += vnet_hdr_len;
2808 }
2809
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002810 if (!packet_use_direct_xmit(po))
2811 skb_probe_transport_header(skb, reserve);
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002812 if (unlikely(extra_len == 4))
2813 skb->no_fcs = 1;
2814
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002815 err = po->xmit(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2817 goto out_unlock;
2818
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002819 dev_put(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002821 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822
2823out_free:
2824 kfree_skb(skb);
2825out_unlock:
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002826 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 dev_put(dev);
2828out:
2829 return err;
2830}
2831
Ying Xue1b784142015-03-02 15:37:48 +08002832static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
Johann Baudy69e3c752009-05-18 22:11:22 -07002833{
Johann Baudy69e3c752009-05-18 22:11:22 -07002834 struct sock *sk = sock->sk;
2835 struct packet_sock *po = pkt_sk(sk);
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002836
Johann Baudy69e3c752009-05-18 22:11:22 -07002837 if (po->tx_ring.pg_vec)
2838 return tpacket_snd(po, msg);
2839 else
Johann Baudy69e3c752009-05-18 22:11:22 -07002840 return packet_snd(sock, msg, len);
2841}
2842
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843/*
2844 * Close a PACKET socket. This is fairly simple. We immediately go
2845 * to 'closed' state and remove our protocol entry in the device list.
2846 */
2847
2848static int packet_release(struct socket *sock)
2849{
2850 struct sock *sk = sock->sk;
2851 struct packet_sock *po;
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08002852 struct net *net;
chetan lokef6fb8f12011-08-19 10:18:16 +00002853 union tpacket_req_u req_u;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854
2855 if (!sk)
2856 return 0;
2857
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002858 net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002859 po = pkt_sk(sk);
2860
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002861 mutex_lock(&net->packet.sklist_lock);
stephen hemminger808f5112010-02-22 07:57:18 +00002862 sk_del_node_init_rcu(sk);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002863 mutex_unlock(&net->packet.sklist_lock);
2864
2865 preempt_disable();
Eric Dumazet920de802008-11-24 00:09:29 -08002866 sock_prot_inuse_add(net, sk->sk_prot, -1);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002867 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868
stephen hemminger808f5112010-02-22 07:57:18 +00002869 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07002870 unregister_prot_hook(sk, false);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002871 packet_cached_dev_reset(po);
2872
Ben Greear160ff182011-06-01 07:18:52 +00002873 if (po->prot_hook.dev) {
2874 dev_put(po->prot_hook.dev);
2875 po->prot_hook.dev = NULL;
2876 }
stephen hemminger808f5112010-02-22 07:57:18 +00002877 spin_unlock(&po->bind_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 packet_flush_mclist(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880
Phil Sutter9665d5d2013-02-01 07:21:41 +00002881 if (po->rx_ring.pg_vec) {
2882 memset(&req_u, 0, sizeof(req_u));
chetan lokef6fb8f12011-08-19 10:18:16 +00002883 packet_set_ring(sk, &req_u, 1, 0);
Phil Sutter9665d5d2013-02-01 07:21:41 +00002884 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002885
Phil Sutter9665d5d2013-02-01 07:21:41 +00002886 if (po->tx_ring.pg_vec) {
2887 memset(&req_u, 0, sizeof(req_u));
chetan lokef6fb8f12011-08-19 10:18:16 +00002888 packet_set_ring(sk, &req_u, 1, 1);
Phil Sutter9665d5d2013-02-01 07:21:41 +00002889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890
David S. Millerdc99f602011-07-05 01:45:05 -07002891 fanout_release(sk);
2892
stephen hemminger808f5112010-02-22 07:57:18 +00002893 synchronize_net();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 /*
2895 * Now the socket is dead. No more input will appear.
2896 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 sock_orphan(sk);
2898 sock->sk = NULL;
2899
2900 /* Purge queues */
2901
2902 skb_queue_purge(&sk->sk_receive_queue);
Daniel Borkmannb0138402014-01-15 16:25:36 +01002903 packet_free_pending(po);
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08002904 sk_refcnt_debug_release(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002905
2906 sock_put(sk);
2907 return 0;
2908}
2909
2910/*
2911 * Attach a packet hook.
2912 */
2913
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002914static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2915 __be16 proto)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916{
2917 struct packet_sock *po = pkt_sk(sk);
Lars Westerhoff158cd4a2015-07-28 01:32:21 +03002918 struct net_device *dev_curr;
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002919 __be16 proto_curr;
2920 bool need_rehook;
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002921 struct net_device *dev = NULL;
2922 int ret = 0;
2923 bool unlisted = false;
David S. Millerdc99f602011-07-05 01:45:05 -07002924
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002925 if (po->fanout)
David S. Millerdc99f602011-07-05 01:45:05 -07002926 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927
2928 lock_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002929 spin_lock(&po->bind_lock);
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002930 rcu_read_lock();
2931
2932 if (name) {
2933 dev = dev_get_by_name_rcu(sock_net(sk), name);
2934 if (!dev) {
2935 ret = -ENODEV;
2936 goto out_unlock;
2937 }
2938 } else if (ifindex) {
2939 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2940 if (!dev) {
2941 ret = -ENODEV;
2942 goto out_unlock;
2943 }
2944 }
2945
2946 if (dev)
2947 dev_hold(dev);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002948
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002949 proto_curr = po->prot_hook.type;
2950 dev_curr = po->prot_hook.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002952 need_rehook = proto_curr != proto || dev_curr != dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002954 if (need_rehook) {
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002955 if (po->running) {
2956 rcu_read_unlock();
2957 __unregister_prot_hook(sk, true);
2958 rcu_read_lock();
2959 dev_curr = po->prot_hook.dev;
2960 if (dev)
2961 unlisted = !dev_get_by_index_rcu(sock_net(sk),
2962 dev->ifindex);
2963 }
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002964
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002965 po->num = proto;
2966 po->prot_hook.type = proto;
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002967
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002968 if (unlikely(unlisted)) {
2969 dev_put(dev);
2970 po->prot_hook.dev = NULL;
2971 po->ifindex = -1;
2972 packet_cached_dev_reset(po);
2973 } else {
2974 po->prot_hook.dev = dev;
2975 po->ifindex = dev ? dev->ifindex : 0;
2976 packet_cached_dev_assign(po, dev);
2977 }
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002978 }
Lars Westerhoff158cd4a2015-07-28 01:32:21 +03002979 if (dev_curr)
2980 dev_put(dev_curr);
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002981
2982 if (proto == 0 || !need_rehook)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 goto out_unlock;
2984
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002985 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
David S. Millerce06b032011-07-04 01:44:29 -07002986 register_prot_hook(sk);
Urs Thuermannbe85d4a2007-11-12 21:05:20 -08002987 } else {
2988 sk->sk_err = ENETDOWN;
2989 if (!sock_flag(sk, SOCK_DEAD))
2990 sk->sk_error_report(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 }
2992
2993out_unlock:
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002994 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995 spin_unlock(&po->bind_lock);
2996 release_sock(sk);
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08002997 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998}
2999
3000/*
3001 * Bind a packet socket to a device
3002 */
3003
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003004static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3005 int addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003007 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 char name[15];
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003009
Linus Torvalds1da177e2005-04-16 15:20:36 -07003010 /*
3011 * Check legality
3012 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003013
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003014 if (addr_len != sizeof(struct sockaddr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003016 strlcpy(name, uaddr->sa_data, sizeof(name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08003018 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003020
3021static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3022{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003023 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3024 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025
3026 /*
3027 * Check legality
3028 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003029
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 if (addr_len < sizeof(struct sockaddr_ll))
3031 return -EINVAL;
3032 if (sll->sll_family != AF_PACKET)
3033 return -EINVAL;
3034
Francesco Ruggeri30f7ea12015-11-05 08:16:14 -08003035 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3036 sll->sll_protocol ? : pkt_sk(sk)->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037}
3038
3039static struct proto packet_proto = {
3040 .name = "PACKET",
3041 .owner = THIS_MODULE,
3042 .obj_size = sizeof(struct packet_sock),
3043};
3044
3045/*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003046 * Create a packet of type SOCK_PACKET.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 */
3048
Eric Paris3f378b682009-11-05 22:18:14 -08003049static int packet_create(struct net *net, struct socket *sock, int protocol,
3050 int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051{
3052 struct sock *sk;
3053 struct packet_sock *po;
Al Viro0e11c912006-11-08 00:26:29 -08003054 __be16 proto = (__force __be16)protocol; /* weird, but documented */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003055 int err;
3056
Eric W. Biedermandf008c92012-11-16 03:03:07 +00003057 if (!ns_capable(net->user_ns, CAP_NET_RAW))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 return -EPERM;
David S. Millerbe020972007-05-29 13:16:31 -07003059 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3060 sock->type != SOCK_PACKET)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061 return -ESOCKTNOSUPPORT;
3062
3063 sock->state = SS_UNCONNECTED;
3064
3065 err = -ENOBUFS;
Eric W. Biederman11aa9c22015-05-08 21:09:13 -05003066 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 if (sk == NULL)
3068 goto out;
3069
3070 sock->ops = &packet_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 if (sock->type == SOCK_PACKET)
3072 sock->ops = &packet_ops_spkt;
David S. Millerbe020972007-05-29 13:16:31 -07003073
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 sock_init_data(sock, sk);
3075
3076 po = pkt_sk(sk);
3077 sk->sk_family = PF_PACKET;
Al Viro0e11c912006-11-08 00:26:29 -08003078 po->num = proto;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01003079 po->xmit = dev_queue_xmit;
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01003080
Daniel Borkmannb0138402014-01-15 16:25:36 +01003081 err = packet_alloc_pending(po);
3082 if (err)
3083 goto out2;
3084
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01003085 packet_cached_dev_reset(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086
3087 sk->sk_destruct = packet_sock_destruct;
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08003088 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089
3090 /*
3091 * Attach a protocol block
3092 */
3093
3094 spin_lock_init(&po->bind_lock);
Herbert Xu905db442009-01-30 14:12:06 -08003095 mutex_init(&po->pg_vec_lock);
Willem de Bruijn0648ab72015-05-12 11:56:46 -04003096 po->rollover = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003097 po->prot_hook.func = packet_rcv;
David S. Millerbe020972007-05-29 13:16:31 -07003098
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 if (sock->type == SOCK_PACKET)
3100 po->prot_hook.func = packet_rcv_spkt;
David S. Millerbe020972007-05-29 13:16:31 -07003101
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 po->prot_hook.af_packet_priv = sk;
3103
Al Viro0e11c912006-11-08 00:26:29 -08003104 if (proto) {
3105 po->prot_hook.type = proto;
David S. Millerce06b032011-07-04 01:44:29 -07003106 register_prot_hook(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003107 }
3108
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00003109 mutex_lock(&net->packet.sklist_lock);
stephen hemminger808f5112010-02-22 07:57:18 +00003110 sk_add_node_rcu(sk, &net->packet.sklist);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00003111 mutex_unlock(&net->packet.sklist_lock);
3112
3113 preempt_disable();
Eric Dumazet36804532008-11-19 14:25:35 -08003114 sock_prot_inuse_add(net, &packet_proto, 1);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00003115 preempt_enable();
stephen hemminger808f5112010-02-22 07:57:18 +00003116
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003117 return 0;
Daniel Borkmannb0138402014-01-15 16:25:36 +01003118out2:
3119 sk_free(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120out:
3121 return err;
3122}
3123
3124/*
3125 * Pull a packet from our receive queue and hand it to the user.
3126 * If necessary we block.
3127 */
3128
Ying Xue1b784142015-03-02 15:37:48 +08003129static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3130 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131{
3132 struct sock *sk = sock->sk;
3133 struct sk_buff *skb;
3134 int copied, err;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003135 int vnet_hdr_len = 0;
Eyal Birger2472d762015-03-01 14:58:28 +02003136 unsigned int origlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137
3138 err = -EINVAL;
Richard Cochraned85b562010-04-07 22:41:28 +00003139 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 goto out;
3141
3142#if 0
3143 /* What error should we return now? EUNATTACH? */
3144 if (pkt_sk(sk)->ifindex < 0)
3145 return -ENODEV;
3146#endif
3147
Richard Cochraned85b562010-04-07 22:41:28 +00003148 if (flags & MSG_ERRQUEUE) {
Richard Cochrancb820f82013-07-19 19:40:09 +02003149 err = sock_recv_errqueue(sk, msg, len,
3150 SOL_PACKET, PACKET_TX_TIMESTAMP);
Richard Cochraned85b562010-04-07 22:41:28 +00003151 goto out;
3152 }
3153
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 * Call the generic datagram receiver. This handles all sorts
3156 * of horrible races and re-entrancy so we can forget about it
3157 * in the protocol layers.
3158 *
3159 * Now it will return ENETDOWN, if device have just gone down,
3160 * but then it will block.
3161 */
3162
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003163 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164
3165 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003166 * An error occurred so return it. Because skb_recv_datagram()
Linus Torvalds1da177e2005-04-16 15:20:36 -07003167 * handles the blocking we don't see and worry about blocking
3168 * retries.
3169 */
3170
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003171 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003172 goto out;
3173
Willem de Bruijn2ccdbaa2015-05-12 11:56:48 -04003174 if (pkt_sk(sk)->pressure)
3175 packet_rcv_has_room(pkt_sk(sk), NULL);
3176
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003177 if (pkt_sk(sk)->has_vnet_hdr) {
3178 struct virtio_net_hdr vnet_hdr = { 0 };
3179
3180 err = -EINVAL;
3181 vnet_hdr_len = sizeof(vnet_hdr);
Mariusz Kozlowski1f18b712010-11-08 11:58:45 +00003182 if (len < vnet_hdr_len)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003183 goto out_free;
3184
Mariusz Kozlowski1f18b712010-11-08 11:58:45 +00003185 len -= vnet_hdr_len;
3186
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003187 if (skb_is_gso(skb)) {
3188 struct skb_shared_info *sinfo = skb_shinfo(skb);
3189
3190 /* This is a hint as to how much should be linear. */
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02003191 vnet_hdr.hdr_len =
David Woodhoused3869ef2015-09-23 19:45:08 +01003192 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02003193 vnet_hdr.gso_size =
David Woodhoused3869ef2015-09-23 19:45:08 +01003194 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003195 if (sinfo->gso_type & SKB_GSO_TCPV4)
3196 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3197 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3198 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3199 else if (sinfo->gso_type & SKB_GSO_UDP)
3200 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3201 else if (sinfo->gso_type & SKB_GSO_FCOE)
3202 goto out_free;
3203 else
3204 BUG();
3205 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3206 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3207 } else
3208 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3209
3210 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3211 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
David Woodhoused3869ef2015-09-23 19:45:08 +01003212 vnet_hdr.csum_start = __cpu_to_virtio16(vio_le(),
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02003213 skb_checksum_start_offset(skb));
David Woodhoused3869ef2015-09-23 19:45:08 +01003214 vnet_hdr.csum_offset = __cpu_to_virtio16(vio_le(),
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02003215 skb->csum_offset);
Jason Wang10a8d942011-06-10 00:56:17 +00003216 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3217 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003218 } /* else everything is zero */
3219
Al Viro7eab8d92014-04-06 21:51:23 -04003220 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003221 if (err < 0)
3222 goto out_free;
3223 }
3224
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003225 /* You lose any data beyond the buffer you gave. If it worries
3226 * a user program they can ask the device for its MTU
3227 * anyway.
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003228 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003229 copied = skb->len;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003230 if (copied > len) {
3231 copied = len;
3232 msg->msg_flags |= MSG_TRUNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233 }
3234
David S. Miller51f3d022014-11-05 16:46:40 -05003235 err = skb_copy_datagram_msg(skb, 0, msg, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236 if (err)
3237 goto out_free;
3238
Eyal Birger2472d762015-03-01 14:58:28 +02003239 if (sock->type != SOCK_PACKET) {
3240 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3241
3242 /* Original length was stored in sockaddr_ll fields */
3243 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3244 sll->sll_family = AF_PACKET;
3245 sll->sll_protocol = skb->protocol;
3246 }
3247
Neil Horman3b885782009-10-12 13:26:31 -07003248 sock_recv_ts_and_drops(msg, sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003250 if (msg->msg_name) {
3251 /* If the address length field is there to be filled
3252 * in, we fill it in now.
3253 */
3254 if (sock->type == SOCK_PACKET) {
Steffen Hurrle342dfc32014-01-17 22:53:15 +01003255 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003256 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3257 } else {
3258 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
Eyal Birger2472d762015-03-01 14:58:28 +02003259
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003260 msg->msg_namelen = sll->sll_halen +
3261 offsetof(struct sockaddr_ll, sll_addr);
3262 }
Herbert Xuffbc6112007-02-04 23:33:10 -08003263 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3264 msg->msg_namelen);
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003265 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266
Herbert Xu8dc41942007-02-04 23:31:32 -08003267 if (pkt_sk(sk)->auxdata) {
Herbert Xuffbc6112007-02-04 23:33:10 -08003268 struct tpacket_auxdata aux;
3269
3270 aux.tp_status = TP_STATUS_USER;
3271 if (skb->ip_summed == CHECKSUM_PARTIAL)
3272 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
Alexander Drozdov682f0482015-03-23 09:11:13 +03003273 else if (skb->pkt_type != PACKET_OUTGOING &&
3274 (skb->ip_summed == CHECKSUM_COMPLETE ||
3275 skb_csum_unnecessary(skb)))
3276 aux.tp_status |= TP_STATUS_CSUM_VALID;
3277
Eyal Birger2472d762015-03-01 14:58:28 +02003278 aux.tp_len = origlen;
Herbert Xuffbc6112007-02-04 23:33:10 -08003279 aux.tp_snaplen = skb->len;
3280 aux.tp_mac = 0;
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03003281 aux.tp_net = skb_network_offset(skb);
Jiri Pirkodf8a39de2015-01-13 17:13:44 +01003282 if (skb_vlan_tag_present(skb)) {
3283 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09003284 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3285 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
Ben Greeara3bcc232011-06-01 06:49:10 +00003286 } else {
3287 aux.tp_vlan_tci = 0;
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09003288 aux.tp_vlan_tpid = 0;
Ben Greeara3bcc232011-06-01 06:49:10 +00003289 }
Herbert Xuffbc6112007-02-04 23:33:10 -08003290 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
Herbert Xu8dc41942007-02-04 23:31:32 -08003291 }
3292
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 /*
3294 * Free or return the buffer as appropriate. Again this
3295 * hides all the races and re-entrancy issues from us.
3296 */
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003297 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298
3299out_free:
3300 skb_free_datagram(sk, skb);
3301out:
3302 return err;
3303}
3304
Linus Torvalds1da177e2005-04-16 15:20:36 -07003305static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3306 int *uaddr_len, int peer)
3307{
3308 struct net_device *dev;
3309 struct sock *sk = sock->sk;
3310
3311 if (peer)
3312 return -EOPNOTSUPP;
3313
3314 uaddr->sa_family = AF_PACKET;
Daniel Borkmann2dc85bf2013-06-12 16:02:27 +02003315 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
Eric Dumazet654d1f82009-11-02 10:43:32 +01003316 rcu_read_lock();
3317 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3318 if (dev)
Daniel Borkmann2dc85bf2013-06-12 16:02:27 +02003319 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
Eric Dumazet654d1f82009-11-02 10:43:32 +01003320 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321 *uaddr_len = sizeof(*uaddr);
3322
3323 return 0;
3324}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325
3326static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3327 int *uaddr_len, int peer)
3328{
3329 struct net_device *dev;
3330 struct sock *sk = sock->sk;
3331 struct packet_sock *po = pkt_sk(sk);
Cyrill Gorcunov13cfa972009-11-08 05:51:19 +00003332 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003333
3334 if (peer)
3335 return -EOPNOTSUPP;
3336
3337 sll->sll_family = AF_PACKET;
3338 sll->sll_ifindex = po->ifindex;
3339 sll->sll_protocol = po->num;
Vasiliy Kulikov67286642010-11-10 12:09:10 -08003340 sll->sll_pkttype = 0;
Eric Dumazet654d1f82009-11-02 10:43:32 +01003341 rcu_read_lock();
3342 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343 if (dev) {
3344 sll->sll_hatype = dev->type;
3345 sll->sll_halen = dev->addr_len;
3346 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347 } else {
3348 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3349 sll->sll_halen = 0;
3350 }
Eric Dumazet654d1f82009-11-02 10:43:32 +01003351 rcu_read_unlock();
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003352 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353
3354 return 0;
3355}
3356
Wang Chen2aeb0b82008-07-14 20:49:46 -07003357static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3358 int what)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359{
3360 switch (i->type) {
3361 case PACKET_MR_MULTICAST:
Jiri Pirko11625632010-03-02 20:40:01 +00003362 if (i->alen != dev->addr_len)
3363 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 if (what > 0)
Jiri Pirko22bedad32010-04-01 21:22:57 +00003365 return dev_mc_add(dev, i->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366 else
Jiri Pirko22bedad32010-04-01 21:22:57 +00003367 return dev_mc_del(dev, i->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368 break;
3369 case PACKET_MR_PROMISC:
Wang Chen2aeb0b82008-07-14 20:49:46 -07003370 return dev_set_promiscuity(dev, what);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371 case PACKET_MR_ALLMULTI:
Wang Chen2aeb0b82008-07-14 20:49:46 -07003372 return dev_set_allmulti(dev, what);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003373 case PACKET_MR_UNICAST:
Jiri Pirko11625632010-03-02 20:40:01 +00003374 if (i->alen != dev->addr_len)
3375 return -EINVAL;
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003376 if (what > 0)
Jiri Pirkoa748ee22010-04-01 21:22:09 +00003377 return dev_uc_add(dev, i->addr);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003378 else
Jiri Pirkoa748ee22010-04-01 21:22:09 +00003379 return dev_uc_del(dev, i->addr);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003380 break;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003381 default:
3382 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383 }
Wang Chen2aeb0b82008-07-14 20:49:46 -07003384 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385}
3386
Francesco Ruggeri82f17092015-03-09 11:51:04 -07003387static void packet_dev_mclist_delete(struct net_device *dev,
3388 struct packet_mclist **mlp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389{
Francesco Ruggeri82f17092015-03-09 11:51:04 -07003390 struct packet_mclist *ml;
3391
3392 while ((ml = *mlp) != NULL) {
3393 if (ml->ifindex == dev->ifindex) {
3394 packet_dev_mc(dev, ml, -1);
3395 *mlp = ml->next;
3396 kfree(ml);
3397 } else
3398 mlp = &ml->next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003399 }
3400}
3401
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003402static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003403{
3404 struct packet_sock *po = pkt_sk(sk);
3405 struct packet_mclist *ml, *i;
3406 struct net_device *dev;
3407 int err;
3408
3409 rtnl_lock();
3410
3411 err = -ENODEV;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09003412 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003413 if (!dev)
3414 goto done;
3415
3416 err = -EINVAL;
Jiri Pirko11625632010-03-02 20:40:01 +00003417 if (mreq->mr_alen > dev->addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418 goto done;
3419
3420 err = -ENOBUFS;
Kris Katterjohn8b3a7002006-01-11 15:56:43 -08003421 i = kmalloc(sizeof(*i), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003422 if (i == NULL)
3423 goto done;
3424
3425 err = 0;
3426 for (ml = po->mclist; ml; ml = ml->next) {
3427 if (ml->ifindex == mreq->mr_ifindex &&
3428 ml->type == mreq->mr_type &&
3429 ml->alen == mreq->mr_alen &&
3430 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3431 ml->count++;
3432 /* Free the new element ... */
3433 kfree(i);
3434 goto done;
3435 }
3436 }
3437
3438 i->type = mreq->mr_type;
3439 i->ifindex = mreq->mr_ifindex;
3440 i->alen = mreq->mr_alen;
3441 memcpy(i->addr, mreq->mr_address, i->alen);
3442 i->count = 1;
3443 i->next = po->mclist;
3444 po->mclist = i;
Wang Chen2aeb0b82008-07-14 20:49:46 -07003445 err = packet_dev_mc(dev, i, 1);
3446 if (err) {
3447 po->mclist = i->next;
3448 kfree(i);
3449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003450
3451done:
3452 rtnl_unlock();
3453 return err;
3454}
3455
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003456static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457{
3458 struct packet_mclist *ml, **mlp;
3459
3460 rtnl_lock();
3461
3462 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3463 if (ml->ifindex == mreq->mr_ifindex &&
3464 ml->type == mreq->mr_type &&
3465 ml->alen == mreq->mr_alen &&
3466 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3467 if (--ml->count == 0) {
3468 struct net_device *dev;
3469 *mlp = ml->next;
Eric Dumazetad959e72009-10-16 06:38:46 +00003470 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3471 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472 packet_dev_mc(dev, ml, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473 kfree(ml);
3474 }
Francesco Ruggeri82f17092015-03-09 11:51:04 -07003475 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003476 }
3477 }
3478 rtnl_unlock();
Francesco Ruggeri82f17092015-03-09 11:51:04 -07003479 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480}
3481
3482static void packet_flush_mclist(struct sock *sk)
3483{
3484 struct packet_sock *po = pkt_sk(sk);
3485 struct packet_mclist *ml;
3486
3487 if (!po->mclist)
3488 return;
3489
3490 rtnl_lock();
3491 while ((ml = po->mclist) != NULL) {
3492 struct net_device *dev;
3493
3494 po->mclist = ml->next;
Eric Dumazetad959e72009-10-16 06:38:46 +00003495 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3496 if (dev != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 packet_dev_mc(dev, ml, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003498 kfree(ml);
3499 }
3500 rtnl_unlock();
3501}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003502
3503static int
David S. Millerb7058842009-09-30 16:12:20 -07003504packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505{
3506 struct sock *sk = sock->sk;
Herbert Xu8dc41942007-02-04 23:31:32 -08003507 struct packet_sock *po = pkt_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003508 int ret;
3509
3510 if (level != SOL_PACKET)
3511 return -ENOPROTOOPT;
3512
Johann Baudy69e3c752009-05-18 22:11:22 -07003513 switch (optname) {
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003514 case PACKET_ADD_MEMBERSHIP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 case PACKET_DROP_MEMBERSHIP:
3516 {
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003517 struct packet_mreq_max mreq;
3518 int len = optlen;
3519 memset(&mreq, 0, sizeof(mreq));
3520 if (len < sizeof(struct packet_mreq))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003521 return -EINVAL;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003522 if (len > sizeof(mreq))
3523 len = sizeof(mreq);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003524 if (copy_from_user(&mreq, optval, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525 return -EFAULT;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003526 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3527 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003528 if (optname == PACKET_ADD_MEMBERSHIP)
3529 ret = packet_mc_add(sk, &mreq);
3530 else
3531 ret = packet_mc_drop(sk, &mreq);
3532 return ret;
3533 }
David S. Millera2efcfa2007-05-29 13:12:50 -07003534
Linus Torvalds1da177e2005-04-16 15:20:36 -07003535 case PACKET_RX_RING:
Johann Baudy69e3c752009-05-18 22:11:22 -07003536 case PACKET_TX_RING:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 {
chetan lokef6fb8f12011-08-19 10:18:16 +00003538 union tpacket_req_u req_u;
3539 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540
chetan lokef6fb8f12011-08-19 10:18:16 +00003541 switch (po->tp_version) {
3542 case TPACKET_V1:
3543 case TPACKET_V2:
3544 len = sizeof(req_u.req);
3545 break;
3546 case TPACKET_V3:
3547 default:
3548 len = sizeof(req_u.req3);
3549 break;
3550 }
3551 if (optlen < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003552 return -EINVAL;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003553 if (pkt_sk(sk)->has_vnet_hdr)
3554 return -EINVAL;
chetan lokef6fb8f12011-08-19 10:18:16 +00003555 if (copy_from_user(&req_u.req, optval, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556 return -EFAULT;
chetan lokef6fb8f12011-08-19 10:18:16 +00003557 return packet_set_ring(sk, &req_u, 0,
3558 optname == PACKET_TX_RING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003559 }
3560 case PACKET_COPY_THRESH:
3561 {
3562 int val;
3563
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003564 if (optlen != sizeof(val))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003566 if (copy_from_user(&val, optval, sizeof(val)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003567 return -EFAULT;
3568
3569 pkt_sk(sk)->copy_thresh = val;
3570 return 0;
3571 }
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003572 case PACKET_VERSION:
3573 {
3574 int val;
3575
3576 if (optlen != sizeof(val))
3577 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07003578 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003579 return -EBUSY;
3580 if (copy_from_user(&val, optval, sizeof(val)))
3581 return -EFAULT;
3582 switch (val) {
3583 case TPACKET_V1:
3584 case TPACKET_V2:
chetan lokef6fb8f12011-08-19 10:18:16 +00003585 case TPACKET_V3:
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003586 po->tp_version = val;
3587 return 0;
3588 default:
3589 return -EINVAL;
3590 }
3591 }
Patrick McHardy8913336a2008-07-18 18:05:19 -07003592 case PACKET_RESERVE:
3593 {
3594 unsigned int val;
3595
3596 if (optlen != sizeof(val))
3597 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07003598 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
Patrick McHardy8913336a2008-07-18 18:05:19 -07003599 return -EBUSY;
3600 if (copy_from_user(&val, optval, sizeof(val)))
3601 return -EFAULT;
3602 po->tp_reserve = val;
3603 return 0;
3604 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003605 case PACKET_LOSS:
3606 {
3607 unsigned int val;
3608
3609 if (optlen != sizeof(val))
3610 return -EINVAL;
3611 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3612 return -EBUSY;
3613 if (copy_from_user(&val, optval, sizeof(val)))
3614 return -EFAULT;
3615 po->tp_loss = !!val;
3616 return 0;
3617 }
Herbert Xu8dc41942007-02-04 23:31:32 -08003618 case PACKET_AUXDATA:
3619 {
3620 int val;
3621
3622 if (optlen < sizeof(val))
3623 return -EINVAL;
3624 if (copy_from_user(&val, optval, sizeof(val)))
3625 return -EFAULT;
3626
3627 po->auxdata = !!val;
3628 return 0;
3629 }
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003630 case PACKET_ORIGDEV:
3631 {
3632 int val;
3633
3634 if (optlen < sizeof(val))
3635 return -EINVAL;
3636 if (copy_from_user(&val, optval, sizeof(val)))
3637 return -EFAULT;
3638
3639 po->origdev = !!val;
3640 return 0;
3641 }
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003642 case PACKET_VNET_HDR:
3643 {
3644 int val;
3645
3646 if (sock->type != SOCK_RAW)
3647 return -EINVAL;
3648 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3649 return -EBUSY;
3650 if (optlen < sizeof(val))
3651 return -EINVAL;
3652 if (copy_from_user(&val, optval, sizeof(val)))
3653 return -EFAULT;
3654
3655 po->has_vnet_hdr = !!val;
3656 return 0;
3657 }
Scott McMillan614f60f2010-06-02 05:53:56 -07003658 case PACKET_TIMESTAMP:
3659 {
3660 int val;
3661
3662 if (optlen != sizeof(val))
3663 return -EINVAL;
3664 if (copy_from_user(&val, optval, sizeof(val)))
3665 return -EFAULT;
3666
3667 po->tp_tstamp = val;
3668 return 0;
3669 }
David S. Millerdc99f602011-07-05 01:45:05 -07003670 case PACKET_FANOUT:
3671 {
3672 int val;
3673
3674 if (optlen != sizeof(val))
3675 return -EINVAL;
3676 if (copy_from_user(&val, optval, sizeof(val)))
3677 return -EFAULT;
3678
3679 return fanout_add(sk, val & 0xffff, val >> 16);
3680 }
Willem de Bruijn47dceb82015-08-14 22:31:34 -04003681 case PACKET_FANOUT_DATA:
3682 {
3683 if (!po->fanout)
3684 return -EINVAL;
3685
3686 return fanout_set_data(po, optval, optlen);
3687 }
Paul Chavent5920cd3a2012-11-06 23:10:47 +00003688 case PACKET_TX_HAS_OFF:
3689 {
3690 unsigned int val;
3691
3692 if (optlen != sizeof(val))
3693 return -EINVAL;
3694 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3695 return -EBUSY;
3696 if (copy_from_user(&val, optval, sizeof(val)))
3697 return -EFAULT;
3698 po->tp_tx_has_off = !!val;
3699 return 0;
3700 }
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01003701 case PACKET_QDISC_BYPASS:
3702 {
3703 int val;
3704
3705 if (optlen != sizeof(val))
3706 return -EINVAL;
3707 if (copy_from_user(&val, optval, sizeof(val)))
3708 return -EFAULT;
3709
3710 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3711 return 0;
3712 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003713 default:
3714 return -ENOPROTOOPT;
3715 }
3716}
3717
3718static int packet_getsockopt(struct socket *sock, int level, int optname,
3719 char __user *optval, int __user *optlen)
3720{
3721 int len;
Eric Dumazetc06fff62012-04-19 21:56:11 +00003722 int val, lv = sizeof(val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003723 struct sock *sk = sock->sk;
3724 struct packet_sock *po = pkt_sk(sk);
Eric Dumazetc06fff62012-04-19 21:56:11 +00003725 void *data = &val;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003726 union tpacket_stats_u st;
Willem de Bruijna9b63912015-05-12 11:56:50 -04003727 struct tpacket_rollover_stats rstats;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003728
3729 if (level != SOL_PACKET)
3730 return -ENOPROTOOPT;
3731
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003732 if (get_user(len, optlen))
3733 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003734
3735 if (len < 0)
3736 return -EINVAL;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003737
Johann Baudy69e3c752009-05-18 22:11:22 -07003738 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739 case PACKET_STATISTICS:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003740 spin_lock_bh(&sk->sk_receive_queue.lock);
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003741 memcpy(&st, &po->stats, sizeof(st));
3742 memset(&po->stats, 0, sizeof(po->stats));
3743 spin_unlock_bh(&sk->sk_receive_queue.lock);
3744
chetan lokef6fb8f12011-08-19 10:18:16 +00003745 if (po->tp_version == TPACKET_V3) {
Eric Dumazetc06fff62012-04-19 21:56:11 +00003746 lv = sizeof(struct tpacket_stats_v3);
Willem de Bruijn8bcdeaf2013-08-19 16:40:22 -04003747 st.stats3.tp_packets += st.stats3.tp_drops;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003748 data = &st.stats3;
chetan lokef6fb8f12011-08-19 10:18:16 +00003749 } else {
Eric Dumazetc06fff62012-04-19 21:56:11 +00003750 lv = sizeof(struct tpacket_stats);
Willem de Bruijn8bcdeaf2013-08-19 16:40:22 -04003751 st.stats1.tp_packets += st.stats1.tp_drops;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003752 data = &st.stats1;
chetan lokef6fb8f12011-08-19 10:18:16 +00003753 }
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003754
Linus Torvalds1da177e2005-04-16 15:20:36 -07003755 break;
Herbert Xu8dc41942007-02-04 23:31:32 -08003756 case PACKET_AUXDATA:
Herbert Xu8dc41942007-02-04 23:31:32 -08003757 val = po->auxdata;
Herbert Xu8dc41942007-02-04 23:31:32 -08003758 break;
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003759 case PACKET_ORIGDEV:
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003760 val = po->origdev;
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003761 break;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003762 case PACKET_VNET_HDR:
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003763 val = po->has_vnet_hdr;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003764 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003765 case PACKET_VERSION:
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003766 val = po->tp_version;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003767 break;
3768 case PACKET_HDRLEN:
3769 if (len > sizeof(int))
3770 len = sizeof(int);
3771 if (copy_from_user(&val, optval, len))
3772 return -EFAULT;
3773 switch (val) {
3774 case TPACKET_V1:
3775 val = sizeof(struct tpacket_hdr);
3776 break;
3777 case TPACKET_V2:
3778 val = sizeof(struct tpacket2_hdr);
3779 break;
chetan lokef6fb8f12011-08-19 10:18:16 +00003780 case TPACKET_V3:
3781 val = sizeof(struct tpacket3_hdr);
3782 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003783 default:
3784 return -EINVAL;
3785 }
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003786 break;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003787 case PACKET_RESERVE:
Patrick McHardy8913336a2008-07-18 18:05:19 -07003788 val = po->tp_reserve;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003789 break;
Johann Baudy69e3c752009-05-18 22:11:22 -07003790 case PACKET_LOSS:
Johann Baudy69e3c752009-05-18 22:11:22 -07003791 val = po->tp_loss;
Johann Baudy69e3c752009-05-18 22:11:22 -07003792 break;
Scott McMillan614f60f2010-06-02 05:53:56 -07003793 case PACKET_TIMESTAMP:
Scott McMillan614f60f2010-06-02 05:53:56 -07003794 val = po->tp_tstamp;
Scott McMillan614f60f2010-06-02 05:53:56 -07003795 break;
David S. Millerdc99f602011-07-05 01:45:05 -07003796 case PACKET_FANOUT:
David S. Millerdc99f602011-07-05 01:45:05 -07003797 val = (po->fanout ?
3798 ((u32)po->fanout->id |
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00003799 ((u32)po->fanout->type << 16) |
3800 ((u32)po->fanout->flags << 24)) :
David S. Millerdc99f602011-07-05 01:45:05 -07003801 0);
David S. Millerdc99f602011-07-05 01:45:05 -07003802 break;
Willem de Bruijna9b63912015-05-12 11:56:50 -04003803 case PACKET_ROLLOVER_STATS:
3804 if (!po->rollover)
3805 return -EINVAL;
3806 rstats.tp_all = atomic_long_read(&po->rollover->num);
3807 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3808 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3809 data = &rstats;
3810 lv = sizeof(rstats);
3811 break;
Paul Chavent5920cd3a2012-11-06 23:10:47 +00003812 case PACKET_TX_HAS_OFF:
3813 val = po->tp_tx_has_off;
3814 break;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01003815 case PACKET_QDISC_BYPASS:
3816 val = packet_use_direct_xmit(po);
3817 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003818 default:
3819 return -ENOPROTOOPT;
3820 }
3821
Eric Dumazetc06fff62012-04-19 21:56:11 +00003822 if (len > lv)
3823 len = lv;
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003824 if (put_user(len, optlen))
3825 return -EFAULT;
Herbert Xu8dc41942007-02-04 23:31:32 -08003826 if (copy_to_user(optval, data, len))
3827 return -EFAULT;
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003828 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829}
3830
3831
Jiri Pirko351638e2013-05-28 01:30:21 +00003832static int packet_notifier(struct notifier_block *this,
3833 unsigned long msg, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003834{
3835 struct sock *sk;
Jiri Pirko351638e2013-05-28 01:30:21 +00003836 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09003837 struct net *net = dev_net(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838
stephen hemminger808f5112010-02-22 07:57:18 +00003839 rcu_read_lock();
Sasha Levinb67bfe02013-02-27 17:06:00 -08003840 sk_for_each_rcu(sk, &net->packet.sklist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841 struct packet_sock *po = pkt_sk(sk);
3842
3843 switch (msg) {
3844 case NETDEV_UNREGISTER:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845 if (po->mclist)
Francesco Ruggeri82f17092015-03-09 11:51:04 -07003846 packet_dev_mclist_delete(dev, &po->mclist);
David S. Millera2efcfa2007-05-29 13:12:50 -07003847 /* fallthrough */
3848
Linus Torvalds1da177e2005-04-16 15:20:36 -07003849 case NETDEV_DOWN:
3850 if (dev->ifindex == po->ifindex) {
3851 spin_lock(&po->bind_lock);
3852 if (po->running) {
David S. Millerce06b032011-07-04 01:44:29 -07003853 __unregister_prot_hook(sk, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003854 sk->sk_err = ENETDOWN;
3855 if (!sock_flag(sk, SOCK_DEAD))
3856 sk->sk_error_report(sk);
3857 }
3858 if (msg == NETDEV_UNREGISTER) {
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01003859 packet_cached_dev_reset(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003860 po->ifindex = -1;
Ben Greear160ff182011-06-01 07:18:52 +00003861 if (po->prot_hook.dev)
3862 dev_put(po->prot_hook.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003863 po->prot_hook.dev = NULL;
3864 }
3865 spin_unlock(&po->bind_lock);
3866 }
3867 break;
3868 case NETDEV_UP:
stephen hemminger808f5112010-02-22 07:57:18 +00003869 if (dev->ifindex == po->ifindex) {
3870 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07003871 if (po->num)
3872 register_prot_hook(sk);
stephen hemminger808f5112010-02-22 07:57:18 +00003873 spin_unlock(&po->bind_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875 break;
3876 }
3877 }
stephen hemminger808f5112010-02-22 07:57:18 +00003878 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003879 return NOTIFY_DONE;
3880}
3881
3882
3883static int packet_ioctl(struct socket *sock, unsigned int cmd,
3884 unsigned long arg)
3885{
3886 struct sock *sk = sock->sk;
3887
Johann Baudy69e3c752009-05-18 22:11:22 -07003888 switch (cmd) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003889 case SIOCOUTQ:
3890 {
3891 int amount = sk_wmem_alloc_get(sk);
Eric Dumazet31e6d362009-06-17 19:05:41 -07003892
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003893 return put_user(amount, (int __user *)arg);
3894 }
3895 case SIOCINQ:
3896 {
3897 struct sk_buff *skb;
3898 int amount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003900 spin_lock_bh(&sk->sk_receive_queue.lock);
3901 skb = skb_peek(&sk->sk_receive_queue);
3902 if (skb)
3903 amount = skb->len;
3904 spin_unlock_bh(&sk->sk_receive_queue.lock);
3905 return put_user(amount, (int __user *)arg);
3906 }
3907 case SIOCGSTAMP:
3908 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3909 case SIOCGSTAMPNS:
3910 return sock_get_timestampns(sk, (struct timespec __user *)arg);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003911
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912#ifdef CONFIG_INET
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003913 case SIOCADDRT:
3914 case SIOCDELRT:
3915 case SIOCDARP:
3916 case SIOCGARP:
3917 case SIOCSARP:
3918 case SIOCGIFADDR:
3919 case SIOCSIFADDR:
3920 case SIOCGIFBRDADDR:
3921 case SIOCSIFBRDADDR:
3922 case SIOCGIFNETMASK:
3923 case SIOCSIFNETMASK:
3924 case SIOCGIFDSTADDR:
3925 case SIOCSIFDSTADDR:
3926 case SIOCSIFFLAGS:
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003927 return inet_dgram_ops.ioctl(sock, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928#endif
3929
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003930 default:
3931 return -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 }
3933 return 0;
3934}
3935
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003936static unsigned int packet_poll(struct file *file, struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937 poll_table *wait)
3938{
3939 struct sock *sk = sock->sk;
3940 struct packet_sock *po = pkt_sk(sk);
3941 unsigned int mask = datagram_poll(file, sock, wait);
3942
3943 spin_lock_bh(&sk->sk_receive_queue.lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003944 if (po->rx_ring.pg_vec) {
chetan lokef6fb8f12011-08-19 10:18:16 +00003945 if (!packet_previous_rx_frame(po, &po->rx_ring,
3946 TP_STATUS_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947 mask |= POLLIN | POLLRDNORM;
3948 }
Willem de Bruijn2ccdbaa2015-05-12 11:56:48 -04003949 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
Willem de Bruijn54d7c012015-05-14 15:25:02 -04003950 po->pressure = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003951 spin_unlock_bh(&sk->sk_receive_queue.lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003952 spin_lock_bh(&sk->sk_write_queue.lock);
3953 if (po->tx_ring.pg_vec) {
3954 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3955 mask |= POLLOUT | POLLWRNORM;
3956 }
3957 spin_unlock_bh(&sk->sk_write_queue.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958 return mask;
3959}
3960
3961
3962/* Dirty? Well, I still did not learn better way to account
3963 * for user mmaps.
3964 */
3965
3966static void packet_mm_open(struct vm_area_struct *vma)
3967{
3968 struct file *file = vma->vm_file;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003969 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003971
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972 if (sk)
3973 atomic_inc(&pkt_sk(sk)->mapped);
3974}
3975
3976static void packet_mm_close(struct vm_area_struct *vma)
3977{
3978 struct file *file = vma->vm_file;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003979 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003980 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003981
Linus Torvalds1da177e2005-04-16 15:20:36 -07003982 if (sk)
3983 atomic_dec(&pkt_sk(sk)->mapped);
3984}
3985
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04003986static const struct vm_operations_struct packet_mmap_ops = {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003987 .open = packet_mm_open,
3988 .close = packet_mm_close,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003989};
3990
Neil Horman0e3125c2010-11-16 10:26:47 -08003991static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3992 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993{
3994 int i;
3995
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003996 for (i = 0; i < len; i++) {
Neil Horman0e3125c2010-11-16 10:26:47 -08003997 if (likely(pg_vec[i].buffer)) {
Changli Gaoc56b4d92010-12-01 02:52:57 +00003998 if (is_vmalloc_addr(pg_vec[i].buffer))
Neil Horman0e3125c2010-11-16 10:26:47 -08003999 vfree(pg_vec[i].buffer);
4000 else
4001 free_pages((unsigned long)pg_vec[i].buffer,
4002 order);
4003 pg_vec[i].buffer = NULL;
4004 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004005 }
4006 kfree(pg_vec);
4007}
4008
Olof Johanssoneea49cc92011-11-02 11:00:49 +00004009static char *alloc_one_pg_vec_page(unsigned long order)
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004010{
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01004011 char *buffer;
Neil Horman0e3125c2010-11-16 10:26:47 -08004012 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4013 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
Eric Dumazet719bfea2009-04-15 03:39:52 -07004014
Neil Horman0e3125c2010-11-16 10:26:47 -08004015 buffer = (char *) __get_free_pages(gfp_flags, order);
Neil Horman0e3125c2010-11-16 10:26:47 -08004016 if (buffer)
4017 return buffer;
4018
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01004019 /* __get_free_pages failed, fall back to vmalloc */
Eric Dumazetbbce5a52010-11-20 07:31:54 +00004020 buffer = vzalloc((1 << order) * PAGE_SIZE);
Neil Horman0e3125c2010-11-16 10:26:47 -08004021 if (buffer)
4022 return buffer;
4023
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01004024 /* vmalloc failed, lets dig into swap here */
Neil Horman0e3125c2010-11-16 10:26:47 -08004025 gfp_flags &= ~__GFP_NORETRY;
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01004026 buffer = (char *) __get_free_pages(gfp_flags, order);
Neil Horman0e3125c2010-11-16 10:26:47 -08004027 if (buffer)
4028 return buffer;
4029
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01004030 /* complete and utter failure */
Neil Horman0e3125c2010-11-16 10:26:47 -08004031 return NULL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004032}
4033
Neil Horman0e3125c2010-11-16 10:26:47 -08004034static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004035{
4036 unsigned int block_nr = req->tp_block_nr;
Neil Horman0e3125c2010-11-16 10:26:47 -08004037 struct pgv *pg_vec;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004038 int i;
4039
Neil Horman0e3125c2010-11-16 10:26:47 -08004040 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004041 if (unlikely(!pg_vec))
4042 goto out;
4043
4044 for (i = 0; i < block_nr; i++) {
Changli Gaoc56b4d92010-12-01 02:52:57 +00004045 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
Neil Horman0e3125c2010-11-16 10:26:47 -08004046 if (unlikely(!pg_vec[i].buffer))
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004047 goto out_free_pgvec;
4048 }
4049
4050out:
4051 return pg_vec;
4052
4053out_free_pgvec:
4054 free_pg_vec(pg_vec, order, block_nr);
4055 pg_vec = NULL;
4056 goto out;
4057}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058
chetan lokef6fb8f12011-08-19 10:18:16 +00004059static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
Johann Baudy69e3c752009-05-18 22:11:22 -07004060 int closing, int tx_ring)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061{
Neil Horman0e3125c2010-11-16 10:26:47 -08004062 struct pgv *pg_vec = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004063 struct packet_sock *po = pkt_sk(sk);
Al Viro0e11c912006-11-08 00:26:29 -08004064 int was_running, order = 0;
Johann Baudy69e3c752009-05-18 22:11:22 -07004065 struct packet_ring_buffer *rb;
4066 struct sk_buff_head *rb_queue;
Al Viro0e11c912006-11-08 00:26:29 -08004067 __be16 num;
chetan lokef6fb8f12011-08-19 10:18:16 +00004068 int err = -EINVAL;
4069 /* Added to avoid minimal code churn */
4070 struct tpacket_req *req = &req_u->req;
4071
4072 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4073 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4074 WARN(1, "Tx-ring is not supported.\n");
4075 goto out;
4076 }
Johann Baudy69e3c752009-05-18 22:11:22 -07004077
4078 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4079 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4080
4081 err = -EBUSY;
4082 if (!closing) {
4083 if (atomic_read(&po->mapped))
4084 goto out;
Daniel Borkmannb0138402014-01-15 16:25:36 +01004085 if (packet_read_pending(rb))
Johann Baudy69e3c752009-05-18 22:11:22 -07004086 goto out;
4087 }
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09004088
Linus Torvalds1da177e2005-04-16 15:20:36 -07004089 if (req->tp_block_nr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004090 /* Sanity tests and some calculations */
Johann Baudy69e3c752009-05-18 22:11:22 -07004091 err = -EBUSY;
4092 if (unlikely(rb->pg_vec))
4093 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094
Patrick McHardybbd6ef82008-07-14 22:50:15 -07004095 switch (po->tp_version) {
4096 case TPACKET_V1:
4097 po->tp_hdrlen = TPACKET_HDRLEN;
4098 break;
4099 case TPACKET_V2:
4100 po->tp_hdrlen = TPACKET2_HDRLEN;
4101 break;
chetan lokef6fb8f12011-08-19 10:18:16 +00004102 case TPACKET_V3:
4103 po->tp_hdrlen = TPACKET3_HDRLEN;
4104 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07004105 }
4106
Johann Baudy69e3c752009-05-18 22:11:22 -07004107 err = -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004108 if (unlikely((int)req->tp_block_size <= 0))
Johann Baudy69e3c752009-05-18 22:11:22 -07004109 goto out;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004110 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
Johann Baudy69e3c752009-05-18 22:11:22 -07004111 goto out;
Eric Dumazetdc808112014-08-15 09:16:04 -07004112 if (po->tp_version >= TPACKET_V3 &&
4113 (int)(req->tp_block_size -
4114 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4115 goto out;
Patrick McHardy8913336a2008-07-18 18:05:19 -07004116 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
Johann Baudy69e3c752009-05-18 22:11:22 -07004117 po->tp_reserve))
4118 goto out;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004119 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
Johann Baudy69e3c752009-05-18 22:11:22 -07004120 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121
Johann Baudy69e3c752009-05-18 22:11:22 -07004122 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
4123 if (unlikely(rb->frames_per_block <= 0))
4124 goto out;
4125 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4126 req->tp_frame_nr))
4127 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004128
4129 err = -ENOMEM;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004130 order = get_order(req->tp_block_size);
4131 pg_vec = alloc_pg_vec(req, order);
4132 if (unlikely(!pg_vec))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004133 goto out;
chetan lokef6fb8f12011-08-19 10:18:16 +00004134 switch (po->tp_version) {
4135 case TPACKET_V3:
4136 /* Transmit path is not supported. We checked
4137 * it above but just being paranoid
4138 */
4139 if (!tx_ring)
Maninder Singhe8e85cc2015-06-22 12:39:16 +05304140 init_prb_bdqc(po, rb, pg_vec, req_u);
Dan Carpenterd7cf0c32014-02-18 15:20:51 +03004141 break;
chetan lokef6fb8f12011-08-19 10:18:16 +00004142 default:
4143 break;
4144 }
Johann Baudy69e3c752009-05-18 22:11:22 -07004145 }
4146 /* Done */
4147 else {
4148 err = -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004149 if (unlikely(req->tp_frame_nr))
Johann Baudy69e3c752009-05-18 22:11:22 -07004150 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004151 }
4152
4153 lock_sock(sk);
4154
4155 /* Detach socket from network */
4156 spin_lock(&po->bind_lock);
4157 was_running = po->running;
4158 num = po->num;
4159 if (was_running) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004160 po->num = 0;
David S. Millerce06b032011-07-04 01:44:29 -07004161 __unregister_prot_hook(sk, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004162 }
4163 spin_unlock(&po->bind_lock);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09004164
Linus Torvalds1da177e2005-04-16 15:20:36 -07004165 synchronize_net();
4166
4167 err = -EBUSY;
Herbert Xu905db442009-01-30 14:12:06 -08004168 mutex_lock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004169 if (closing || atomic_read(&po->mapped) == 0) {
4170 err = 0;
Johann Baudy69e3c752009-05-18 22:11:22 -07004171 spin_lock_bh(&rb_queue->lock);
Changli Gaoc053fd92010-12-10 16:02:20 -08004172 swap(rb->pg_vec, pg_vec);
Johann Baudy69e3c752009-05-18 22:11:22 -07004173 rb->frame_max = (req->tp_frame_nr - 1);
4174 rb->head = 0;
4175 rb->frame_size = req->tp_frame_size;
4176 spin_unlock_bh(&rb_queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004177
Changli Gaoc053fd92010-12-10 16:02:20 -08004178 swap(rb->pg_vec_order, order);
4179 swap(rb->pg_vec_len, req->tp_block_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004180
Johann Baudy69e3c752009-05-18 22:11:22 -07004181 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4182 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4183 tpacket_rcv : packet_rcv;
4184 skb_queue_purge(rb_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004185 if (atomic_read(&po->mapped))
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00004186 pr_err("packet_mmap: vma is busy: %d\n",
4187 atomic_read(&po->mapped));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004188 }
Herbert Xu905db442009-01-30 14:12:06 -08004189 mutex_unlock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004190
4191 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07004192 if (was_running) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004193 po->num = num;
David S. Millerce06b032011-07-04 01:44:29 -07004194 register_prot_hook(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004195 }
4196 spin_unlock(&po->bind_lock);
chetan lokef6fb8f12011-08-19 10:18:16 +00004197 if (closing && (po->tp_version > TPACKET_V2)) {
4198 /* Because we don't support block-based V3 on tx-ring */
4199 if (!tx_ring)
Tobias Klauser73d0fcf2015-07-28 14:21:26 +02004200 prb_shutdown_retire_blk_timer(po, rb_queue);
chetan lokef6fb8f12011-08-19 10:18:16 +00004201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004202 release_sock(sk);
4203
Linus Torvalds1da177e2005-04-16 15:20:36 -07004204 if (pg_vec)
4205 free_pg_vec(pg_vec, order, req->tp_block_nr);
4206out:
4207 return err;
4208}
4209
Johann Baudy69e3c752009-05-18 22:11:22 -07004210static int packet_mmap(struct file *file, struct socket *sock,
4211 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004212{
4213 struct sock *sk = sock->sk;
4214 struct packet_sock *po = pkt_sk(sk);
Johann Baudy69e3c752009-05-18 22:11:22 -07004215 unsigned long size, expected_size;
4216 struct packet_ring_buffer *rb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217 unsigned long start;
4218 int err = -EINVAL;
4219 int i;
4220
4221 if (vma->vm_pgoff)
4222 return -EINVAL;
4223
Herbert Xu905db442009-01-30 14:12:06 -08004224 mutex_lock(&po->pg_vec_lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07004225
4226 expected_size = 0;
4227 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4228 if (rb->pg_vec) {
4229 expected_size += rb->pg_vec_len
4230 * rb->pg_vec_pages
4231 * PAGE_SIZE;
4232 }
4233 }
4234
4235 if (expected_size == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004236 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07004237
4238 size = vma->vm_end - vma->vm_start;
4239 if (size != expected_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004240 goto out;
4241
Linus Torvalds1da177e2005-04-16 15:20:36 -07004242 start = vma->vm_start;
Johann Baudy69e3c752009-05-18 22:11:22 -07004243 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4244 if (rb->pg_vec == NULL)
4245 continue;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004246
Johann Baudy69e3c752009-05-18 22:11:22 -07004247 for (i = 0; i < rb->pg_vec_len; i++) {
Neil Horman0e3125c2010-11-16 10:26:47 -08004248 struct page *page;
4249 void *kaddr = rb->pg_vec[i].buffer;
Johann Baudy69e3c752009-05-18 22:11:22 -07004250 int pg_num;
4251
Changli Gaoc56b4d92010-12-01 02:52:57 +00004252 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4253 page = pgv_to_page(kaddr);
Johann Baudy69e3c752009-05-18 22:11:22 -07004254 err = vm_insert_page(vma, start, page);
4255 if (unlikely(err))
4256 goto out;
4257 start += PAGE_SIZE;
Neil Horman0e3125c2010-11-16 10:26:47 -08004258 kaddr += PAGE_SIZE;
Johann Baudy69e3c752009-05-18 22:11:22 -07004259 }
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004260 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004261 }
Johann Baudy69e3c752009-05-18 22:11:22 -07004262
David S. Miller4ebf0ae2005-12-06 16:38:35 -08004263 atomic_inc(&po->mapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004264 vma->vm_ops = &packet_mmap_ops;
4265 err = 0;
4266
4267out:
Herbert Xu905db442009-01-30 14:12:06 -08004268 mutex_unlock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004269 return err;
4270}
Linus Torvalds1da177e2005-04-16 15:20:36 -07004271
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08004272static const struct proto_ops packet_ops_spkt = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004273 .family = PF_PACKET,
4274 .owner = THIS_MODULE,
4275 .release = packet_release,
4276 .bind = packet_bind_spkt,
4277 .connect = sock_no_connect,
4278 .socketpair = sock_no_socketpair,
4279 .accept = sock_no_accept,
4280 .getname = packet_getname_spkt,
4281 .poll = datagram_poll,
4282 .ioctl = packet_ioctl,
4283 .listen = sock_no_listen,
4284 .shutdown = sock_no_shutdown,
4285 .setsockopt = sock_no_setsockopt,
4286 .getsockopt = sock_no_getsockopt,
4287 .sendmsg = packet_sendmsg_spkt,
4288 .recvmsg = packet_recvmsg,
4289 .mmap = sock_no_mmap,
4290 .sendpage = sock_no_sendpage,
4291};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004292
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08004293static const struct proto_ops packet_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004294 .family = PF_PACKET,
4295 .owner = THIS_MODULE,
4296 .release = packet_release,
4297 .bind = packet_bind,
4298 .connect = sock_no_connect,
4299 .socketpair = sock_no_socketpair,
4300 .accept = sock_no_accept,
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09004301 .getname = packet_getname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004302 .poll = packet_poll,
4303 .ioctl = packet_ioctl,
4304 .listen = sock_no_listen,
4305 .shutdown = sock_no_shutdown,
4306 .setsockopt = packet_setsockopt,
4307 .getsockopt = packet_getsockopt,
4308 .sendmsg = packet_sendmsg,
4309 .recvmsg = packet_recvmsg,
4310 .mmap = packet_mmap,
4311 .sendpage = sock_no_sendpage,
4312};
4313
Stephen Hemmingerec1b4cf2009-10-05 05:58:39 +00004314static const struct net_proto_family packet_family_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004315 .family = PF_PACKET,
4316 .create = packet_create,
4317 .owner = THIS_MODULE,
4318};
4319
4320static struct notifier_block packet_netdev_notifier = {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00004321 .notifier_call = packet_notifier,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004322};
4323
4324#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07004325
4326static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
stephen hemminger808f5112010-02-22 07:57:18 +00004327 __acquires(RCU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004328{
Denis V. Luneve372c412007-11-19 22:31:54 -08004329 struct net *net = seq_file_net(seq);
stephen hemminger808f5112010-02-22 07:57:18 +00004330
4331 rcu_read_lock();
4332 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004333}
4334
4335static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4336{
Herbert Xu1bf40952007-12-16 14:04:02 -08004337 struct net *net = seq_file_net(seq);
stephen hemminger808f5112010-02-22 07:57:18 +00004338 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004339}
4340
4341static void packet_seq_stop(struct seq_file *seq, void *v)
stephen hemminger808f5112010-02-22 07:57:18 +00004342 __releases(RCU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004343{
stephen hemminger808f5112010-02-22 07:57:18 +00004344 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004345}
4346
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09004347static int packet_seq_show(struct seq_file *seq, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004348{
4349 if (v == SEQ_START_TOKEN)
4350 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4351 else {
Li Zefanb7ceabd2010-02-08 23:19:29 +00004352 struct sock *s = sk_entry(v);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004353 const struct packet_sock *po = pkt_sk(s);
4354
4355 seq_printf(seq,
Dan Rosenberg71338aa2011-05-23 12:17:35 +00004356 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004357 s,
4358 atomic_read(&s->sk_refcnt),
4359 s->sk_type,
4360 ntohs(po->num),
4361 po->ifindex,
4362 po->running,
4363 atomic_read(&s->sk_rmem_alloc),
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06004364 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00004365 sock_i_ino(s));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004366 }
4367
4368 return 0;
4369}
4370
Philippe De Muyter56b3d972007-07-10 23:07:31 -07004371static const struct seq_operations packet_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004372 .start = packet_seq_start,
4373 .next = packet_seq_next,
4374 .stop = packet_seq_stop,
4375 .show = packet_seq_show,
4376};
4377
4378static int packet_seq_open(struct inode *inode, struct file *file)
4379{
Denis V. Luneve372c412007-11-19 22:31:54 -08004380 return seq_open_net(inode, file, &packet_seq_ops,
4381 sizeof(struct seq_net_private));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004382}
4383
Arjan van de Venda7071d2007-02-12 00:55:36 -08004384static const struct file_operations packet_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004385 .owner = THIS_MODULE,
4386 .open = packet_seq_open,
4387 .read = seq_read,
4388 .llseek = seq_lseek,
Denis V. Luneve372c412007-11-19 22:31:54 -08004389 .release = seq_release_net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004390};
4391
4392#endif
4393
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004394static int __net_init packet_net_init(struct net *net)
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004395{
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00004396 mutex_init(&net->packet.sklist_lock);
Denis V. Lunev2aaef4e2007-12-11 04:19:54 -08004397 INIT_HLIST_HEAD(&net->packet.sklist);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004398
Gao fengd4beaa62013-02-18 01:34:54 +00004399 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004400 return -ENOMEM;
4401
4402 return 0;
4403}
4404
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004405static void __net_exit packet_net_exit(struct net *net)
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004406{
Gao fengece31ff2013-02-18 01:34:56 +00004407 remove_proc_entry("packet", net->proc_net);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004408}
4409
4410static struct pernet_operations packet_net_ops = {
4411 .init = packet_net_init,
4412 .exit = packet_net_exit,
4413};
4414
4415
Linus Torvalds1da177e2005-04-16 15:20:36 -07004416static void __exit packet_exit(void)
4417{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004418 unregister_netdevice_notifier(&packet_netdev_notifier);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004419 unregister_pernet_subsys(&packet_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004420 sock_unregister(PF_PACKET);
4421 proto_unregister(&packet_proto);
4422}
4423
4424static int __init packet_init(void)
4425{
4426 int rc = proto_register(&packet_proto, 0);
4427
4428 if (rc != 0)
4429 goto out;
4430
4431 sock_register(&packet_family_ops);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004432 register_pernet_subsys(&packet_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004433 register_netdevice_notifier(&packet_netdev_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004434out:
4435 return rc;
4436}
4437
4438module_init(packet_init);
4439module_exit(packet_exit);
4440MODULE_LICENSE("GPL");
4441MODULE_ALIAS_NETPROTO(PF_PACKET);