blob: ba8309a3e01beede11dc5adc01e26dc72fda15b5 [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 lokef6fb8f102011-08-19 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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
92#ifdef CONFIG_INET
93#include <net/inet_common.h>
94#endif
95
Pavel Emelyanov2787b042012-08-13 05:49:39 +000096#include "internal.h"
97
Linus Torvalds1da177e2005-04-16 15:20:36 -070098/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 Assumptions:
100 - if device has no dev->hard_header routine, it adds and removes ll header
101 inside itself. In this case ll header is invisible outside of device,
102 but higher levels still should reserve dev->hard_header_len.
103 Some devices are enough clever to reallocate skb, when header
104 will not fit to reserved space (tunnel), another ones are silly
105 (PPP).
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
108
109On receive:
110-----------
111
112Incoming, dev->hard_header!=NULL
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700113 mac_header -> ll header
114 data -> data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115
116Outgoing, dev->hard_header!=NULL
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700117 mac_header -> ll header
118 data -> ll header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
120Incoming, dev->hard_header==NULL
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
YOSHIFUJI Hideakidb0c58f2007-07-19 10:44:35 +0900123 assymetry between rx and tx paths.
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700124 data -> data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
126Outgoing, dev->hard_header==NULL
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700127 mac_header -> data. ll header is still not built!
128 data -> data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130Resume
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
132
133
134On transmit:
135------------
136
137dev->hard_header != NULL
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700138 mac_header -> ll header
139 data -> ll header
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
141dev->hard_header == NULL (ll header is added by device, we cannot control it)
Arnaldo Carvalho de Melob0e380b2007-04-10 21:21:55 -0700142 mac_header -> data
143 data -> data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
147 */
148
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149/* Private packet socket structures. */
150
Eric W. Biederman0fb375f2005-09-21 00:11:37 -0700151/* identical to struct packet_mreq except it has
152 * a longer address field.
153 */
Eric Dumazet40d4e3d2009-07-21 21:57:59 +0000154struct packet_mreq_max {
Eric W. Biederman0fb375f2005-09-21 00:11:37 -0700155 int mr_ifindex;
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159};
David S. Millera2efcfa2007-05-29 13:12:50 -0700160
Daniel Borkmann184f4892013-04-16 01:57:46 +0000161union tpacket_uhdr {
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
165 void *raw;
166};
167
chetan lokef6fb8f102011-08-19 10:18:16 +0000168static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
Johann Baudy69e3c752009-05-18 22:11:22 -0700169 int closing, int tx_ring);
170
chetan lokef6fb8f102011-08-19 10:18:16 +0000171#define V3_ALIGNMENT (8)
172
chetan lokebc59ba32011-08-25 10:43:30 +0000173#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
chetan lokef6fb8f102011-08-19 10:18:16 +0000174
175#define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
177
chetan lokef6fb8f102011-08-19 10:18:16 +0000178#define PGV_FROM_VMALLOC 1
Johann Baudy69e3c752009-05-18 22:11:22 -0700179
chetan lokef6fb8f102011-08-19 10:18:16 +0000180#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186#define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
187
Johann Baudy69e3c752009-05-18 22:11:22 -0700188struct packet_sock;
189static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +0000190static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
chetan lokef6fb8f102011-08-19 10:18:16 +0000193static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
195 int status);
196static void packet_increment_head(struct packet_ring_buffer *buff);
chetan lokebc59ba32011-08-25 10:43:30 +0000197static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
chetan lokef6fb8f102011-08-19 10:18:16 +0000200 struct packet_sock *);
chetan lokebc59ba32011-08-25 10:43:30 +0000201static void prb_retire_current_block(struct tpacket_kbdq_core *,
chetan lokef6fb8f102011-08-19 10:18:16 +0000202 struct packet_sock *, unsigned int status);
chetan lokebc59ba32011-08-25 10:43:30 +0000203static int prb_queue_frozen(struct tpacket_kbdq_core *);
204static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
chetan lokef6fb8f102011-08-19 10:18:16 +0000206static void prb_retire_rx_blk_timer_expired(unsigned long);
chetan lokebc59ba32011-08-25 10:43:30 +0000207static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216static void packet_flush_mclist(struct sock *sk);
217
Herbert Xuffbc6112007-02-04 23:33:10 -0800218struct packet_skb_cb {
219 unsigned int origlen;
220 union {
221 struct sockaddr_pkt pkt;
222 struct sockaddr_ll ll;
223 } sa;
224};
225
226#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
Herbert Xu8dc41942007-02-04 23:31:32 -0800227
chetan lokebc59ba32011-08-25 10:43:30 +0000228#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
chetan lokef6fb8f102011-08-19 10:18:16 +0000229#define GET_PBLOCK_DESC(x, bid) \
chetan lokebc59ba32011-08-25 10:43:30 +0000230 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
chetan lokef6fb8f102011-08-19 10:18:16 +0000231#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
chetan lokebc59ba32011-08-25 10:43:30 +0000232 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
chetan lokef6fb8f102011-08-19 10:18:16 +0000233#define GET_NEXT_PRB_BLK_NUM(x) \
234 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
235 ((x)->kactive_blk_num+1) : 0)
236
David S. Millerdc99f602011-07-05 01:45:05 -0700237static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
238static void __fanout_link(struct sock *sk, struct packet_sock *po);
239
David S. Millerce06b032011-07-04 01:44:29 -0700240/* register_prot_hook must be invoked with the po->bind_lock held,
241 * or from a context in which asynchronous accesses to the packet
242 * socket is not possible (packet_create()).
243 */
244static void register_prot_hook(struct sock *sk)
245{
246 struct packet_sock *po = pkt_sk(sk);
247 if (!po->running) {
David S. Millerdc99f602011-07-05 01:45:05 -0700248 if (po->fanout)
249 __fanout_link(sk, po);
250 else
251 dev_add_pack(&po->prot_hook);
David S. Millerce06b032011-07-04 01:44:29 -0700252 sock_hold(sk);
253 po->running = 1;
254 }
255}
256
257/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
258 * held. If the sync parameter is true, we will temporarily drop
259 * the po->bind_lock and do a synchronize_net to make sure no
260 * asynchronous packet processing paths still refer to the elements
261 * of po->prot_hook. If the sync parameter is false, it is the
262 * callers responsibility to take care of this.
263 */
264static void __unregister_prot_hook(struct sock *sk, bool sync)
265{
266 struct packet_sock *po = pkt_sk(sk);
267
268 po->running = 0;
David S. Millerdc99f602011-07-05 01:45:05 -0700269 if (po->fanout)
270 __fanout_unlink(sk, po);
271 else
272 __dev_remove_pack(&po->prot_hook);
David S. Millerce06b032011-07-04 01:44:29 -0700273 __sock_put(sk);
274
275 if (sync) {
276 spin_unlock(&po->bind_lock);
277 synchronize_net();
278 spin_lock(&po->bind_lock);
279 }
280}
281
282static void unregister_prot_hook(struct sock *sk, bool sync)
283{
284 struct packet_sock *po = pkt_sk(sk);
285
286 if (po->running)
287 __unregister_prot_hook(sk, sync);
288}
289
Changli Gaof6dafa92010-12-07 04:26:16 +0000290static inline __pure struct page *pgv_to_page(void *addr)
Changli Gao0af55bb2010-12-01 02:52:20 +0000291{
292 if (is_vmalloc_addr(addr))
293 return vmalloc_to_page(addr);
294 return virt_to_page(addr);
295}
296
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700297static void __packet_set_status(struct packet_sock *po, void *frame, int status)
298{
Daniel Borkmann184f4892013-04-16 01:57:46 +0000299 union tpacket_uhdr h;
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700300
301 h.raw = frame;
302 switch (po->tp_version) {
303 case TPACKET_V1:
304 h.h1->tp_status = status;
Changli Gao0af55bb2010-12-01 02:52:20 +0000305 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700306 break;
307 case TPACKET_V2:
308 h.h2->tp_status = status;
Changli Gao0af55bb2010-12-01 02:52:20 +0000309 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700310 break;
chetan lokef6fb8f102011-08-19 10:18:16 +0000311 case TPACKET_V3:
Johann Baudy69e3c752009-05-18 22:11:22 -0700312 default:
chetan lokef6fb8f102011-08-19 10:18:16 +0000313 WARN(1, "TPACKET version not supported.\n");
Johann Baudy69e3c752009-05-18 22:11:22 -0700314 BUG();
315 }
316
317 smp_wmb();
318}
319
320static int __packet_get_status(struct packet_sock *po, void *frame)
321{
Daniel Borkmann184f4892013-04-16 01:57:46 +0000322 union tpacket_uhdr h;
Johann Baudy69e3c752009-05-18 22:11:22 -0700323
324 smp_rmb();
325
326 h.raw = frame;
327 switch (po->tp_version) {
328 case TPACKET_V1:
Changli Gao0af55bb2010-12-01 02:52:20 +0000329 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
Johann Baudy69e3c752009-05-18 22:11:22 -0700330 return h.h1->tp_status;
331 case TPACKET_V2:
Changli Gao0af55bb2010-12-01 02:52:20 +0000332 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
Johann Baudy69e3c752009-05-18 22:11:22 -0700333 return h.h2->tp_status;
chetan lokef6fb8f102011-08-19 10:18:16 +0000334 case TPACKET_V3:
Johann Baudy69e3c752009-05-18 22:11:22 -0700335 default:
chetan lokef6fb8f102011-08-19 10:18:16 +0000336 WARN(1, "TPACKET version not supported.\n");
Johann Baudy69e3c752009-05-18 22:11:22 -0700337 BUG();
338 return 0;
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700339 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340}
Johann Baudy69e3c752009-05-18 22:11:22 -0700341
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000342static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
343 unsigned int flags)
Daniel Borkmann7a513842013-04-23 00:39:29 +0000344{
345 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
346
347 if (shhwtstamps) {
348 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
349 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000350 return TP_STATUS_TS_SYS_HARDWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000351 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
352 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000353 return TP_STATUS_TS_RAW_HARDWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000354 }
355
356 if (ktime_to_timespec_cond(skb->tstamp, ts))
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000357 return TP_STATUS_TS_SOFTWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000358
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000359 return 0;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000360}
361
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000362static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
363 struct sk_buff *skb)
Willem de Bruijn2e313962013-04-23 00:39:28 +0000364{
365 union tpacket_uhdr h;
366 struct timespec ts;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000367 __u32 ts_status;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000368
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000369 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
370 return 0;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000371
372 h.raw = frame;
373 switch (po->tp_version) {
374 case TPACKET_V1:
375 h.h1->tp_sec = ts.tv_sec;
376 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
377 break;
378 case TPACKET_V2:
379 h.h2->tp_sec = ts.tv_sec;
380 h.h2->tp_nsec = ts.tv_nsec;
381 break;
382 case TPACKET_V3:
383 default:
384 WARN(1, "TPACKET version not supported.\n");
385 BUG();
386 }
387
388 /* one flush is safe, as both fields always lie on the same cacheline */
389 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
390 smp_wmb();
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000391
392 return ts_status;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000393}
394
Johann Baudy69e3c752009-05-18 22:11:22 -0700395static void *packet_lookup_frame(struct packet_sock *po,
396 struct packet_ring_buffer *rb,
397 unsigned int position,
398 int status)
399{
400 unsigned int pg_vec_pos, frame_offset;
Daniel Borkmann184f4892013-04-16 01:57:46 +0000401 union tpacket_uhdr h;
Johann Baudy69e3c752009-05-18 22:11:22 -0700402
403 pg_vec_pos = position / rb->frames_per_block;
404 frame_offset = position % rb->frames_per_block;
405
Neil Horman0e3125c2010-11-16 10:26:47 -0800406 h.raw = rb->pg_vec[pg_vec_pos].buffer +
407 (frame_offset * rb->frame_size);
Johann Baudy69e3c752009-05-18 22:11:22 -0700408
409 if (status != __packet_get_status(po, h.raw))
410 return NULL;
411
412 return h.raw;
413}
414
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000415static void *packet_current_frame(struct packet_sock *po,
Johann Baudy69e3c752009-05-18 22:11:22 -0700416 struct packet_ring_buffer *rb,
417 int status)
418{
419 return packet_lookup_frame(po, rb, rb->head, status);
420}
421
chetan lokebc59ba32011-08-25 10:43:30 +0000422static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000423{
424 del_timer_sync(&pkc->retire_blk_timer);
425}
426
427static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
428 int tx_ring,
429 struct sk_buff_head *rb_queue)
430{
chetan lokebc59ba32011-08-25 10:43:30 +0000431 struct tpacket_kbdq_core *pkc;
chetan lokef6fb8f102011-08-19 10:18:16 +0000432
433 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
434
435 spin_lock(&rb_queue->lock);
436 pkc->delete_blk_timer = 1;
437 spin_unlock(&rb_queue->lock);
438
439 prb_del_retire_blk_timer(pkc);
440}
441
442static void prb_init_blk_timer(struct packet_sock *po,
chetan lokebc59ba32011-08-25 10:43:30 +0000443 struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000444 void (*func) (unsigned long))
445{
446 init_timer(&pkc->retire_blk_timer);
447 pkc->retire_blk_timer.data = (long)po;
448 pkc->retire_blk_timer.function = func;
449 pkc->retire_blk_timer.expires = jiffies;
450}
451
452static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
453{
chetan lokebc59ba32011-08-25 10:43:30 +0000454 struct tpacket_kbdq_core *pkc;
chetan lokef6fb8f102011-08-19 10:18:16 +0000455
456 if (tx_ring)
457 BUG();
458
459 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
460 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
461}
462
463static int prb_calc_retire_blk_tmo(struct packet_sock *po,
464 int blk_size_in_bytes)
465{
466 struct net_device *dev;
467 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000468 struct ethtool_cmd ecmd;
469 int err;
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000470 u32 speed;
chetan lokef6fb8f102011-08-19 10:18:16 +0000471
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000472 rtnl_lock();
473 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
474 if (unlikely(!dev)) {
475 rtnl_unlock();
chetan lokef6fb8f102011-08-19 10:18:16 +0000476 return DEFAULT_PRB_RETIRE_TOV;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000477 }
478 err = __ethtool_get_settings(dev, &ecmd);
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000479 speed = ethtool_cmd_speed(&ecmd);
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000480 rtnl_unlock();
481 if (!err) {
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000482 /*
483 * If the link speed is so slow you don't really
484 * need to worry about perf anyways
485 */
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000486 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000487 return DEFAULT_PRB_RETIRE_TOV;
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000488 } else {
489 msec = 1;
490 div = speed / 1000;
chetan lokef6fb8f102011-08-19 10:18:16 +0000491 }
492 }
493
494 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
495
496 if (div)
497 mbits /= div;
498
499 tmo = mbits * msec;
500
501 if (div)
502 return tmo+1;
503 return tmo;
504}
505
chetan lokebc59ba32011-08-25 10:43:30 +0000506static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
chetan lokef6fb8f102011-08-19 10:18:16 +0000507 union tpacket_req_u *req_u)
508{
509 p1->feature_req_word = req_u->req3.tp_feature_req_word;
510}
511
512static void init_prb_bdqc(struct packet_sock *po,
513 struct packet_ring_buffer *rb,
514 struct pgv *pg_vec,
515 union tpacket_req_u *req_u, int tx_ring)
516{
chetan lokebc59ba32011-08-25 10:43:30 +0000517 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
518 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000519
520 memset(p1, 0x0, sizeof(*p1));
521
522 p1->knxt_seq_num = 1;
523 p1->pkbdq = pg_vec;
chetan lokebc59ba32011-08-25 10:43:30 +0000524 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
Joe Perchese3192692012-06-03 17:41:40 +0000525 p1->pkblk_start = pg_vec[0].buffer;
chetan lokef6fb8f102011-08-19 10:18:16 +0000526 p1->kblk_size = req_u->req3.tp_block_size;
527 p1->knum_blocks = req_u->req3.tp_block_nr;
528 p1->hdrlen = po->tp_hdrlen;
529 p1->version = po->tp_version;
530 p1->last_kactive_blk_num = 0;
531 po->stats_u.stats3.tp_freeze_q_cnt = 0;
532 if (req_u->req3.tp_retire_blk_tov)
533 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
534 else
535 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
536 req_u->req3.tp_block_size);
537 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
538 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
539
540 prb_init_ft_ops(p1, req_u);
541 prb_setup_retire_blk_timer(po, tx_ring);
542 prb_open_block(p1, pbd);
543}
544
545/* Do NOT update the last_blk_num first.
546 * Assumes sk_buff_head lock is held.
547 */
chetan lokebc59ba32011-08-25 10:43:30 +0000548static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000549{
550 mod_timer(&pkc->retire_blk_timer,
551 jiffies + pkc->tov_in_jiffies);
552 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
553}
554
555/*
556 * Timer logic:
557 * 1) We refresh the timer only when we open a block.
558 * By doing this we don't waste cycles refreshing the timer
559 * on packet-by-packet basis.
560 *
561 * With a 1MB block-size, on a 1Gbps line, it will take
562 * i) ~8 ms to fill a block + ii) memcpy etc.
563 * In this cut we are not accounting for the memcpy time.
564 *
565 * So, if the user sets the 'tmo' to 10ms then the timer
566 * will never fire while the block is still getting filled
567 * (which is what we want). However, the user could choose
568 * to close a block early and that's fine.
569 *
570 * But when the timer does fire, we check whether or not to refresh it.
571 * Since the tmo granularity is in msecs, it is not too expensive
572 * to refresh the timer, lets say every '8' msecs.
573 * Either the user can set the 'tmo' or we can derive it based on
574 * a) line-speed and b) block-size.
575 * prb_calc_retire_blk_tmo() calculates the tmo.
576 *
577 */
578static void prb_retire_rx_blk_timer_expired(unsigned long data)
579{
580 struct packet_sock *po = (struct packet_sock *)data;
chetan lokebc59ba32011-08-25 10:43:30 +0000581 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
chetan lokef6fb8f102011-08-19 10:18:16 +0000582 unsigned int frozen;
chetan lokebc59ba32011-08-25 10:43:30 +0000583 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000584
585 spin_lock(&po->sk.sk_receive_queue.lock);
586
587 frozen = prb_queue_frozen(pkc);
588 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
589
590 if (unlikely(pkc->delete_blk_timer))
591 goto out;
592
593 /* We only need to plug the race when the block is partially filled.
594 * tpacket_rcv:
595 * lock(); increment BLOCK_NUM_PKTS; unlock()
596 * copy_bits() is in progress ...
597 * timer fires on other cpu:
598 * we can't retire the current block because copy_bits
599 * is in progress.
600 *
601 */
602 if (BLOCK_NUM_PKTS(pbd)) {
603 while (atomic_read(&pkc->blk_fill_in_prog)) {
604 /* Waiting for skb_copy_bits to finish... */
605 cpu_relax();
606 }
607 }
608
609 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
610 if (!frozen) {
611 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
612 if (!prb_dispatch_next_block(pkc, po))
613 goto refresh_timer;
614 else
615 goto out;
616 } else {
617 /* Case 1. Queue was frozen because user-space was
618 * lagging behind.
619 */
620 if (prb_curr_blk_in_use(pkc, pbd)) {
621 /*
622 * Ok, user-space is still behind.
623 * So just refresh the timer.
624 */
625 goto refresh_timer;
626 } else {
627 /* Case 2. queue was frozen,user-space caught up,
628 * now the link went idle && the timer fired.
629 * We don't have a block to close.So we open this
630 * block and restart the timer.
631 * opening a block thaws the queue,restarts timer
632 * Thawing/timer-refresh is a side effect.
633 */
634 prb_open_block(pkc, pbd);
635 goto out;
636 }
637 }
638 }
639
640refresh_timer:
641 _prb_refresh_rx_retire_blk_timer(pkc);
642
643out:
644 spin_unlock(&po->sk.sk_receive_queue.lock);
645}
646
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000647static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
chetan lokebc59ba32011-08-25 10:43:30 +0000648 struct tpacket_block_desc *pbd1, __u32 status)
chetan lokef6fb8f102011-08-19 10:18:16 +0000649{
650 /* Flush everything minus the block header */
651
652#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
653 u8 *start, *end;
654
655 start = (u8 *)pbd1;
656
657 /* Skip the block header(we know header WILL fit in 4K) */
658 start += PAGE_SIZE;
659
660 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
661 for (; start < end; start += PAGE_SIZE)
662 flush_dcache_page(pgv_to_page(start));
663
664 smp_wmb();
665#endif
666
667 /* Now update the block status. */
668
669 BLOCK_STATUS(pbd1) = status;
670
671 /* Flush the block header */
672
673#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
674 start = (u8 *)pbd1;
675 flush_dcache_page(pgv_to_page(start));
676
677 smp_wmb();
678#endif
679}
680
681/*
682 * Side effect:
683 *
684 * 1) flush the block
685 * 2) Increment active_blk_num
686 *
687 * Note:We DONT refresh the timer on purpose.
688 * Because almost always the next block will be opened.
689 */
chetan lokebc59ba32011-08-25 10:43:30 +0000690static void prb_close_block(struct tpacket_kbdq_core *pkc1,
691 struct tpacket_block_desc *pbd1,
chetan lokef6fb8f102011-08-19 10:18:16 +0000692 struct packet_sock *po, unsigned int stat)
693{
694 __u32 status = TP_STATUS_USER | stat;
695
696 struct tpacket3_hdr *last_pkt;
chetan lokebc59ba32011-08-25 10:43:30 +0000697 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
chetan lokef6fb8f102011-08-19 10:18:16 +0000698
699 if (po->stats.tp_drops)
700 status |= TP_STATUS_LOSING;
701
702 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
703 last_pkt->tp_next_offset = 0;
704
705 /* Get the ts of the last pkt */
706 if (BLOCK_NUM_PKTS(pbd1)) {
707 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
708 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
709 } else {
710 /* Ok, we tmo'd - so get the current time */
711 struct timespec ts;
712 getnstimeofday(&ts);
713 h1->ts_last_pkt.ts_sec = ts.tv_sec;
714 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
715 }
716
717 smp_wmb();
718
719 /* Flush the block */
720 prb_flush_block(pkc1, pbd1, status);
721
722 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
723}
724
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000725static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000726{
727 pkc->reset_pending_on_curr_blk = 0;
728}
729
730/*
731 * Side effect of opening a block:
732 *
733 * 1) prb_queue is thawed.
734 * 2) retire_blk_timer is refreshed.
735 *
736 */
chetan lokebc59ba32011-08-25 10:43:30 +0000737static void prb_open_block(struct tpacket_kbdq_core *pkc1,
738 struct tpacket_block_desc *pbd1)
chetan lokef6fb8f102011-08-19 10:18:16 +0000739{
740 struct timespec ts;
chetan lokebc59ba32011-08-25 10:43:30 +0000741 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
chetan lokef6fb8f102011-08-19 10:18:16 +0000742
743 smp_rmb();
744
745 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
746
747 /* We could have just memset this but we will lose the
748 * flexibility of making the priv area sticky
749 */
750 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
751 BLOCK_NUM_PKTS(pbd1) = 0;
752 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
753 getnstimeofday(&ts);
754 h1->ts_first_pkt.ts_sec = ts.tv_sec;
755 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
756 pkc1->pkblk_start = (char *)pbd1;
Joe Perchese3192692012-06-03 17:41:40 +0000757 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
chetan lokef6fb8f102011-08-19 10:18:16 +0000758 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
759 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
760 pbd1->version = pkc1->version;
761 pkc1->prev = pkc1->nxt_offset;
762 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
763 prb_thaw_queue(pkc1);
764 _prb_refresh_rx_retire_blk_timer(pkc1);
765
766 smp_wmb();
767
768 return;
769 }
770
771 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
772 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
773 dump_stack();
774 BUG();
775}
776
777/*
778 * Queue freeze logic:
779 * 1) Assume tp_block_nr = 8 blocks.
780 * 2) At time 't0', user opens Rx ring.
781 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
782 * 4) user-space is either sleeping or processing block '0'.
783 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
784 * it will close block-7,loop around and try to fill block '0'.
785 * call-flow:
786 * __packet_lookup_frame_in_block
787 * prb_retire_current_block()
788 * prb_dispatch_next_block()
789 * |->(BLOCK_STATUS == USER) evaluates to true
790 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
791 * 6) Now there are two cases:
792 * 6.1) Link goes idle right after the queue is frozen.
793 * But remember, the last open_block() refreshed the timer.
794 * When this timer expires,it will refresh itself so that we can
795 * re-open block-0 in near future.
796 * 6.2) Link is busy and keeps on receiving packets. This is a simple
797 * case and __packet_lookup_frame_in_block will check if block-0
798 * is free and can now be re-used.
799 */
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000800static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000801 struct packet_sock *po)
802{
803 pkc->reset_pending_on_curr_blk = 1;
804 po->stats_u.stats3.tp_freeze_q_cnt++;
805}
806
807#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
808
809/*
810 * If the next block is free then we will dispatch it
811 * and return a good offset.
812 * Else, we will freeze the queue.
813 * So, caller must check the return value.
814 */
chetan lokebc59ba32011-08-25 10:43:30 +0000815static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000816 struct packet_sock *po)
817{
chetan lokebc59ba32011-08-25 10:43:30 +0000818 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000819
820 smp_rmb();
821
822 /* 1. Get current block num */
823 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
824
825 /* 2. If this block is currently in_use then freeze the queue */
826 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
827 prb_freeze_queue(pkc, po);
828 return NULL;
829 }
830
831 /*
832 * 3.
833 * open this block and return the offset where the first packet
834 * needs to get stored.
835 */
836 prb_open_block(pkc, pbd);
837 return (void *)pkc->nxt_offset;
838}
839
chetan lokebc59ba32011-08-25 10:43:30 +0000840static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000841 struct packet_sock *po, unsigned int status)
842{
chetan lokebc59ba32011-08-25 10:43:30 +0000843 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
chetan lokef6fb8f102011-08-19 10:18:16 +0000844
845 /* retire/close the current block */
846 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
847 /*
848 * Plug the case where copy_bits() is in progress on
849 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
850 * have space to copy the pkt in the current block and
851 * called prb_retire_current_block()
852 *
853 * We don't need to worry about the TMO case because
854 * the timer-handler already handled this case.
855 */
856 if (!(status & TP_STATUS_BLK_TMO)) {
857 while (atomic_read(&pkc->blk_fill_in_prog)) {
858 /* Waiting for skb_copy_bits to finish... */
859 cpu_relax();
860 }
861 }
862 prb_close_block(pkc, pbd, po, status);
863 return;
864 }
865
866 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
867 dump_stack();
868 BUG();
869}
870
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000871static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
chetan lokebc59ba32011-08-25 10:43:30 +0000872 struct tpacket_block_desc *pbd)
chetan lokef6fb8f102011-08-19 10:18:16 +0000873{
874 return TP_STATUS_USER & BLOCK_STATUS(pbd);
875}
876
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000877static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000878{
879 return pkc->reset_pending_on_curr_blk;
880}
881
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000882static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
chetan lokef6fb8f102011-08-19 10:18:16 +0000883{
chetan lokebc59ba32011-08-25 10:43:30 +0000884 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
chetan lokef6fb8f102011-08-19 10:18:16 +0000885 atomic_dec(&pkc->blk_fill_in_prog);
886}
887
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000888static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000889 struct tpacket3_hdr *ppd)
890{
891 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
892}
893
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000894static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000895 struct tpacket3_hdr *ppd)
896{
897 ppd->hv1.tp_rxhash = 0;
898}
899
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000900static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000901 struct tpacket3_hdr *ppd)
902{
903 if (vlan_tx_tag_present(pkc->skb)) {
904 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
905 ppd->tp_status = TP_STATUS_VLAN_VALID;
906 } else {
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +0000907 ppd->hv1.tp_vlan_tci = 0;
908 ppd->tp_status = TP_STATUS_AVAILABLE;
chetan lokef6fb8f102011-08-19 10:18:16 +0000909 }
910}
911
chetan lokebc59ba32011-08-25 10:43:30 +0000912static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000913 struct tpacket3_hdr *ppd)
914{
915 prb_fill_vlan_info(pkc, ppd);
916
917 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
918 prb_fill_rxhash(pkc, ppd);
919 else
920 prb_clear_rxhash(pkc, ppd);
921}
922
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000923static void prb_fill_curr_block(char *curr,
chetan lokebc59ba32011-08-25 10:43:30 +0000924 struct tpacket_kbdq_core *pkc,
925 struct tpacket_block_desc *pbd,
chetan lokef6fb8f102011-08-19 10:18:16 +0000926 unsigned int len)
927{
928 struct tpacket3_hdr *ppd;
929
930 ppd = (struct tpacket3_hdr *)curr;
931 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
932 pkc->prev = curr;
933 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
934 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
935 BLOCK_NUM_PKTS(pbd) += 1;
936 atomic_inc(&pkc->blk_fill_in_prog);
937 prb_run_all_ft_ops(pkc, ppd);
938}
939
940/* Assumes caller has the sk->rx_queue.lock */
941static void *__packet_lookup_frame_in_block(struct packet_sock *po,
942 struct sk_buff *skb,
943 int status,
944 unsigned int len
945 )
946{
chetan lokebc59ba32011-08-25 10:43:30 +0000947 struct tpacket_kbdq_core *pkc;
948 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000949 char *curr, *end;
950
Joe Perchese3192692012-06-03 17:41:40 +0000951 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f102011-08-19 10:18:16 +0000952 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
953
954 /* Queue is frozen when user space is lagging behind */
955 if (prb_queue_frozen(pkc)) {
956 /*
957 * Check if that last block which caused the queue to freeze,
958 * is still in_use by user-space.
959 */
960 if (prb_curr_blk_in_use(pkc, pbd)) {
961 /* Can't record this packet */
962 return NULL;
963 } else {
964 /*
965 * Ok, the block was released by user-space.
966 * Now let's open that block.
967 * opening a block also thaws the queue.
968 * Thawing is a side effect.
969 */
970 prb_open_block(pkc, pbd);
971 }
972 }
973
974 smp_mb();
975 curr = pkc->nxt_offset;
976 pkc->skb = skb;
Joe Perchese3192692012-06-03 17:41:40 +0000977 end = (char *)pbd + pkc->kblk_size;
chetan lokef6fb8f102011-08-19 10:18:16 +0000978
979 /* first try the current block */
980 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
981 prb_fill_curr_block(curr, pkc, pbd, len);
982 return (void *)curr;
983 }
984
985 /* Ok, close the current block */
986 prb_retire_current_block(pkc, po, 0);
987
988 /* Now, try to dispatch the next block */
989 curr = (char *)prb_dispatch_next_block(pkc, po);
990 if (curr) {
991 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
992 prb_fill_curr_block(curr, pkc, pbd, len);
993 return (void *)curr;
994 }
995
996 /*
997 * No free blocks are available.user_space hasn't caught up yet.
998 * Queue was just frozen and now this packet will get dropped.
999 */
1000 return NULL;
1001}
1002
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001003static void *packet_current_rx_frame(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001004 struct sk_buff *skb,
1005 int status, unsigned int len)
1006{
1007 char *curr = NULL;
1008 switch (po->tp_version) {
1009 case TPACKET_V1:
1010 case TPACKET_V2:
1011 curr = packet_lookup_frame(po, &po->rx_ring,
1012 po->rx_ring.head, status);
1013 return curr;
1014 case TPACKET_V3:
1015 return __packet_lookup_frame_in_block(po, skb, status, len);
1016 default:
1017 WARN(1, "TPACKET version not supported\n");
1018 BUG();
Ying Xue99aa3472012-08-06 16:27:10 +00001019 return NULL;
chetan lokef6fb8f102011-08-19 10:18:16 +00001020 }
1021}
1022
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001023static void *prb_lookup_block(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001024 struct packet_ring_buffer *rb,
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001025 unsigned int idx,
chetan lokef6fb8f102011-08-19 10:18:16 +00001026 int status)
1027{
chetan lokebc59ba32011-08-25 10:43:30 +00001028 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001029 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
chetan lokef6fb8f102011-08-19 10:18:16 +00001030
1031 if (status != BLOCK_STATUS(pbd))
1032 return NULL;
1033 return pbd;
1034}
1035
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001036static int prb_previous_blk_num(struct packet_ring_buffer *rb)
chetan lokef6fb8f102011-08-19 10:18:16 +00001037{
1038 unsigned int prev;
1039 if (rb->prb_bdqc.kactive_blk_num)
1040 prev = rb->prb_bdqc.kactive_blk_num-1;
1041 else
1042 prev = rb->prb_bdqc.knum_blocks-1;
1043 return prev;
1044}
1045
1046/* Assumes caller has held the rx_queue.lock */
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001047static void *__prb_previous_block(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001048 struct packet_ring_buffer *rb,
1049 int status)
1050{
1051 unsigned int previous = prb_previous_blk_num(rb);
1052 return prb_lookup_block(po, rb, previous, status);
1053}
1054
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001055static void *packet_previous_rx_frame(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001056 struct packet_ring_buffer *rb,
1057 int status)
1058{
1059 if (po->tp_version <= TPACKET_V2)
1060 return packet_previous_frame(po, rb, status);
1061
1062 return __prb_previous_block(po, rb, status);
1063}
1064
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001065static void packet_increment_rx_head(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001066 struct packet_ring_buffer *rb)
1067{
1068 switch (po->tp_version) {
1069 case TPACKET_V1:
1070 case TPACKET_V2:
1071 return packet_increment_head(rb);
1072 case TPACKET_V3:
1073 default:
1074 WARN(1, "TPACKET version not supported.\n");
1075 BUG();
1076 return;
1077 }
1078}
1079
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001080static void *packet_previous_frame(struct packet_sock *po,
Johann Baudy69e3c752009-05-18 22:11:22 -07001081 struct packet_ring_buffer *rb,
1082 int status)
1083{
1084 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1085 return packet_lookup_frame(po, rb, previous, status);
1086}
1087
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001088static void packet_increment_head(struct packet_ring_buffer *buff)
Johann Baudy69e3c752009-05-18 22:11:22 -07001089{
1090 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1091}
1092
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001093static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1094{
1095 struct sock *sk = &po->sk;
1096 bool has_room;
1097
1098 if (po->prot_hook.func != tpacket_rcv)
1099 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1100 <= sk->sk_rcvbuf;
1101
1102 spin_lock(&sk->sk_receive_queue.lock);
1103 if (po->tp_version == TPACKET_V3)
1104 has_room = prb_lookup_block(po, &po->rx_ring,
1105 po->rx_ring.prb_bdqc.kactive_blk_num,
1106 TP_STATUS_KERNEL);
1107 else
1108 has_room = packet_lookup_frame(po, &po->rx_ring,
1109 po->rx_ring.head,
1110 TP_STATUS_KERNEL);
1111 spin_unlock(&sk->sk_receive_queue.lock);
1112
1113 return has_room;
1114}
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116static void packet_sock_destruct(struct sock *sk)
1117{
Richard Cochraned85b562010-04-07 22:41:28 +00001118 skb_queue_purge(&sk->sk_error_queue);
1119
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001120 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1121 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
1123 if (!sock_flag(sk, SOCK_DEAD)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001124 pr_err("Attempt to release alive packet socket: %p\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return;
1126 }
1127
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08001128 sk_refcnt_debug_dec(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129}
1130
David S. Millerdc99f602011-07-05 01:45:05 -07001131static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1132{
1133 int x = atomic_read(&f->rr_cur) + 1;
1134
1135 if (x >= num)
1136 x = 0;
1137
1138 return x;
1139}
1140
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001141static unsigned int fanout_demux_hash(struct packet_fanout *f,
1142 struct sk_buff *skb,
1143 unsigned int num)
David S. Millerdc99f602011-07-05 01:45:05 -07001144{
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001145 return (((u64)skb->rxhash) * num) >> 32;
David S. Millerdc99f602011-07-05 01:45:05 -07001146}
1147
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001148static unsigned int fanout_demux_lb(struct packet_fanout *f,
1149 struct sk_buff *skb,
1150 unsigned int num)
David S. Millerdc99f602011-07-05 01:45:05 -07001151{
1152 int cur, old;
1153
1154 cur = atomic_read(&f->rr_cur);
1155 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1156 fanout_rr_next(f, num))) != cur)
1157 cur = old;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001158 return cur;
David S. Millerdc99f602011-07-05 01:45:05 -07001159}
1160
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001161static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1162 struct sk_buff *skb,
1163 unsigned int num)
David S. Miller95ec3eb2011-07-06 01:56:38 -07001164{
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001165 return smp_processor_id() % num;
1166}
David S. Miller95ec3eb2011-07-06 01:56:38 -07001167
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001168static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1169 struct sk_buff *skb,
1170 unsigned int idx, unsigned int skip,
1171 unsigned int num)
1172{
1173 unsigned int i, j;
1174
1175 i = j = min_t(int, f->next[idx], num - 1);
1176 do {
1177 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1178 if (i != j)
1179 f->next[idx] = i;
1180 return i;
1181 }
1182 if (++i == num)
1183 i = 0;
1184 } while (i != j);
1185
1186 return idx;
1187}
1188
1189static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1190{
1191 return f->flags & (flag >> 8);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001192}
1193
David S. Miller95ec3eb2011-07-06 01:56:38 -07001194static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1195 struct packet_type *pt, struct net_device *orig_dev)
David S. Millerdc99f602011-07-05 01:45:05 -07001196{
1197 struct packet_fanout *f = pt->af_packet_priv;
1198 unsigned int num = f->num_members;
1199 struct packet_sock *po;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001200 unsigned int idx;
David S. Millerdc99f602011-07-05 01:45:05 -07001201
1202 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1203 !num) {
1204 kfree_skb(skb);
1205 return 0;
1206 }
1207
David S. Miller95ec3eb2011-07-06 01:56:38 -07001208 switch (f->type) {
1209 case PACKET_FANOUT_HASH:
1210 default:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001211 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
Eric Dumazetbc416d92011-10-06 10:28:31 +00001212 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001213 if (!skb)
1214 return 0;
1215 }
1216 skb_get_rxhash(skb);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001217 idx = fanout_demux_hash(f, skb, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001218 break;
1219 case PACKET_FANOUT_LB:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001220 idx = fanout_demux_lb(f, skb, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001221 break;
1222 case PACKET_FANOUT_CPU:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001223 idx = fanout_demux_cpu(f, skb, num);
1224 break;
1225 case PACKET_FANOUT_ROLLOVER:
1226 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001227 break;
David S. Miller7736d332011-07-05 01:43:20 -07001228 }
1229
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001230 po = pkt_sk(f->arr[idx]);
1231 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1232 unlikely(!packet_rcv_has_room(po, skb))) {
1233 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1234 po = pkt_sk(f->arr[idx]);
1235 }
David S. Millerdc99f602011-07-05 01:45:05 -07001236
1237 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1238}
1239
Pavel Emelyanovfff33212012-08-16 05:36:48 +00001240DEFINE_MUTEX(fanout_mutex);
1241EXPORT_SYMBOL_GPL(fanout_mutex);
David S. Millerdc99f602011-07-05 01:45:05 -07001242static LIST_HEAD(fanout_list);
1243
1244static void __fanout_link(struct sock *sk, struct packet_sock *po)
1245{
1246 struct packet_fanout *f = po->fanout;
1247
1248 spin_lock(&f->lock);
1249 f->arr[f->num_members] = sk;
1250 smp_wmb();
1251 f->num_members++;
1252 spin_unlock(&f->lock);
1253}
1254
1255static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1256{
1257 struct packet_fanout *f = po->fanout;
1258 int i;
1259
1260 spin_lock(&f->lock);
1261 for (i = 0; i < f->num_members; i++) {
1262 if (f->arr[i] == sk)
1263 break;
1264 }
1265 BUG_ON(i >= f->num_members);
1266 f->arr[i] = f->arr[f->num_members - 1];
1267 f->num_members--;
1268 spin_unlock(&f->lock);
1269}
1270
Fengguang Wua0dfb262012-08-23 19:51:21 +08001271static bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
Eric Leblondc0de08d2012-08-16 22:02:58 +00001272{
1273 if (ptype->af_packet_priv == (void*)((struct packet_sock *)sk)->fanout)
1274 return true;
1275
1276 return false;
1277}
1278
David S. Miller7736d332011-07-05 01:43:20 -07001279static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
David S. Millerdc99f602011-07-05 01:45:05 -07001280{
1281 struct packet_sock *po = pkt_sk(sk);
1282 struct packet_fanout *f, *match;
David S. Miller7736d332011-07-05 01:43:20 -07001283 u8 type = type_flags & 0xff;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001284 u8 flags = type_flags >> 8;
David S. Millerdc99f602011-07-05 01:45:05 -07001285 int err;
1286
1287 switch (type) {
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001288 case PACKET_FANOUT_ROLLOVER:
1289 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1290 return -EINVAL;
David S. Millerdc99f602011-07-05 01:45:05 -07001291 case PACKET_FANOUT_HASH:
1292 case PACKET_FANOUT_LB:
David S. Miller95ec3eb2011-07-06 01:56:38 -07001293 case PACKET_FANOUT_CPU:
David S. Millerdc99f602011-07-05 01:45:05 -07001294 break;
1295 default:
1296 return -EINVAL;
1297 }
1298
1299 if (!po->running)
1300 return -EINVAL;
1301
1302 if (po->fanout)
1303 return -EALREADY;
1304
1305 mutex_lock(&fanout_mutex);
1306 match = NULL;
1307 list_for_each_entry(f, &fanout_list, list) {
1308 if (f->id == id &&
1309 read_pnet(&f->net) == sock_net(sk)) {
1310 match = f;
1311 break;
1312 }
1313 }
Eric Dumazetafe62c62011-07-07 06:41:29 -07001314 err = -EINVAL;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001315 if (match && match->flags != flags)
Eric Dumazetafe62c62011-07-07 06:41:29 -07001316 goto out;
David S. Millerdc99f602011-07-05 01:45:05 -07001317 if (!match) {
Eric Dumazetafe62c62011-07-07 06:41:29 -07001318 err = -ENOMEM;
David S. Millerdc99f602011-07-05 01:45:05 -07001319 match = kzalloc(sizeof(*match), GFP_KERNEL);
Eric Dumazetafe62c62011-07-07 06:41:29 -07001320 if (!match)
1321 goto out;
1322 write_pnet(&match->net, sock_net(sk));
1323 match->id = id;
1324 match->type = type;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001325 match->flags = flags;
Eric Dumazetafe62c62011-07-07 06:41:29 -07001326 atomic_set(&match->rr_cur, 0);
1327 INIT_LIST_HEAD(&match->list);
1328 spin_lock_init(&match->lock);
1329 atomic_set(&match->sk_ref, 0);
1330 match->prot_hook.type = po->prot_hook.type;
1331 match->prot_hook.dev = po->prot_hook.dev;
1332 match->prot_hook.func = packet_rcv_fanout;
1333 match->prot_hook.af_packet_priv = match;
Eric Leblondc0de08d2012-08-16 22:02:58 +00001334 match->prot_hook.id_match = match_fanout_group;
Eric Dumazetafe62c62011-07-07 06:41:29 -07001335 dev_add_pack(&match->prot_hook);
1336 list_add(&match->list, &fanout_list);
1337 }
1338 err = -EINVAL;
1339 if (match->type == type &&
1340 match->prot_hook.type == po->prot_hook.type &&
1341 match->prot_hook.dev == po->prot_hook.dev) {
1342 err = -ENOSPC;
1343 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1344 __dev_remove_pack(&po->prot_hook);
1345 po->fanout = match;
1346 atomic_inc(&match->sk_ref);
1347 __fanout_link(sk, po);
1348 err = 0;
David S. Millerdc99f602011-07-05 01:45:05 -07001349 }
1350 }
Eric Dumazetafe62c62011-07-07 06:41:29 -07001351out:
David S. Millerdc99f602011-07-05 01:45:05 -07001352 mutex_unlock(&fanout_mutex);
1353 return err;
1354}
1355
1356static void fanout_release(struct sock *sk)
1357{
1358 struct packet_sock *po = pkt_sk(sk);
1359 struct packet_fanout *f;
1360
1361 f = po->fanout;
1362 if (!f)
1363 return;
1364
Pavel Emelyanovfff33212012-08-16 05:36:48 +00001365 mutex_lock(&fanout_mutex);
David S. Millerdc99f602011-07-05 01:45:05 -07001366 po->fanout = NULL;
1367
David S. Millerdc99f602011-07-05 01:45:05 -07001368 if (atomic_dec_and_test(&f->sk_ref)) {
1369 list_del(&f->list);
1370 dev_remove_pack(&f->prot_hook);
1371 kfree(f);
1372 }
1373 mutex_unlock(&fanout_mutex);
1374}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08001376static const struct proto_ops packet_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08001378static const struct proto_ops packet_ops_spkt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001380static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1381 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382{
1383 struct sock *sk;
1384 struct sockaddr_pkt *spkt;
1385
1386 /*
1387 * When we registered the protocol we saved the socket in the data
1388 * field for just this event.
1389 */
1390
1391 sk = pt->af_packet_priv;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001392
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 /*
1394 * Yank back the headers [hope the device set this
1395 * right or kerboom...]
1396 *
1397 * Incoming packets have ll header pulled,
1398 * push it back.
1399 *
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001400 * For outgoing ones skb->data == skb_mac_header(skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 * so that this procedure is noop.
1402 */
1403
1404 if (skb->pkt_type == PACKET_LOOPBACK)
1405 goto out;
1406
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001407 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08001408 goto out;
1409
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001410 skb = skb_share_check(skb, GFP_ATOMIC);
1411 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 goto oom;
1413
1414 /* drop any routing info */
Eric Dumazetadf30902009-06-02 05:19:30 +00001415 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Phil Oester84531c22005-07-12 11:57:52 -07001417 /* drop conntrack reference */
1418 nf_reset(skb);
1419
Herbert Xuffbc6112007-02-04 23:33:10 -08001420 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001422 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 /*
1425 * The SOCK_PACKET socket receives _all_ frames.
1426 */
1427
1428 spkt->spkt_family = dev->type;
1429 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1430 spkt->spkt_protocol = skb->protocol;
1431
1432 /*
1433 * Charge the memory to the socket. This is done specifically
1434 * to prevent sockets using all the memory up.
1435 */
1436
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001437 if (sock_queue_rcv_skb(sk, skb) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 return 0;
1439
1440out:
1441 kfree_skb(skb);
1442oom:
1443 return 0;
1444}
1445
1446
1447/*
1448 * Output a raw packet to a device layer. This bypasses all the other
1449 * protocol layers and you must therefore supply it with a complete frame
1450 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001451
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1453 struct msghdr *msg, size_t len)
1454{
1455 struct sock *sk = sock->sk;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001456 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001457 struct sk_buff *skb = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 struct net_device *dev;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001459 __be16 proto = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 int err;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001461 int extra_len = 0;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001464 * Get and verify the address.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 */
1466
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001467 if (saddr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 if (msg->msg_namelen < sizeof(struct sockaddr))
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001469 return -EINVAL;
1470 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1471 proto = saddr->spkt_protocol;
1472 } else
1473 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
1475 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001476 * Find the device first to size check it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 */
1478
danborkmann@iogearbox.netde74e922012-06-10 08:59:28 +00001479 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001480retry:
Eric Dumazet654d1f82009-11-02 10:43:32 +01001481 rcu_read_lock();
1482 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 err = -ENODEV;
1484 if (dev == NULL)
1485 goto out_unlock;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001486
David S. Millerd5e76b02007-01-25 19:30:36 -08001487 err = -ENETDOWN;
1488 if (!(dev->flags & IFF_UP))
1489 goto out_unlock;
1490
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 /*
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001492 * You may not queue a frame bigger than the mtu. This is the lowest level
1493 * raw protocol and you must do your own fragmentation at this level.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001495
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001496 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1497 if (!netif_supports_nofcs(dev)) {
1498 err = -EPROTONOSUPPORT;
1499 goto out_unlock;
1500 }
1501 extra_len = 4; /* We're doing our own CRC */
1502 }
1503
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 err = -EMSGSIZE;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001505 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 goto out_unlock;
1507
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001508 if (!skb) {
1509 size_t reserved = LL_RESERVED_SPACE(dev);
Herbert Xu4ce40912011-11-18 02:20:05 +00001510 int tlen = dev->needed_tailroom;
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001511 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001513 rcu_read_unlock();
Herbert Xu4ce40912011-11-18 02:20:05 +00001514 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001515 if (skb == NULL)
1516 return -ENOBUFS;
1517 /* FIXME: Save some space for broken drivers that write a hard
1518 * header at transmission time by themselves. PPP is the notable
1519 * one here. This should really be fixed at the driver level.
1520 */
1521 skb_reserve(skb, reserved);
1522 skb_reset_network_header(skb);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001523
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001524 /* Try to align data part correctly */
1525 if (hhlen) {
1526 skb->data -= hhlen;
1527 skb->tail -= hhlen;
1528 if (len < hhlen)
1529 skb_reset_network_header(skb);
1530 }
1531 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1532 if (err)
1533 goto out_free;
1534 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 }
1536
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001537 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
Ben Greear57f89bf2011-02-11 09:35:18 +00001538 /* Earlier code assumed this would be a VLAN pkt,
1539 * double-check this now that we have the actual
1540 * packet in hand.
1541 */
1542 struct ethhdr *ehdr;
1543 skb_reset_mac_header(skb);
1544 ehdr = eth_hdr(skb);
1545 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1546 err = -EMSGSIZE;
1547 goto out_unlock;
1548 }
1549 }
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 skb->protocol = proto;
1552 skb->dev = dev;
1553 skb->priority = sk->sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00001554 skb->mark = sk->sk_mark;
Daniel Borkmannbf84a012013-04-14 08:08:13 +00001555
1556 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001558 if (unlikely(extra_len == 4))
1559 skb->no_fcs = 1;
1560
Jason Wang40893fd2013-03-26 23:11:22 +00001561 skb_probe_transport_header(skb, 0);
Jason Wangc1aad272013-03-25 20:19:57 +00001562
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 dev_queue_xmit(skb);
Eric Dumazet654d1f82009-11-02 10:43:32 +01001564 rcu_read_unlock();
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001565 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567out_unlock:
Eric Dumazet654d1f82009-11-02 10:43:32 +01001568 rcu_read_unlock();
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001569out_free:
1570 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 return err;
1572}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001574static unsigned int run_filter(const struct sk_buff *skb,
Eric Dumazet62ab0812010-12-06 20:50:09 +00001575 const struct sock *sk,
David S. Millerdbcb5852007-01-24 15:21:02 -08001576 unsigned int res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577{
1578 struct sk_filter *filter;
1579
Eric Dumazet80f8f102011-01-18 07:46:52 +00001580 rcu_read_lock();
1581 filter = rcu_dereference(sk->sk_filter);
David S. Millerdbcb5852007-01-24 15:21:02 -08001582 if (filter != NULL)
Eric Dumazet0a148422011-04-20 09:27:32 +00001583 res = SK_RUN_FILTER(filter, skb);
Eric Dumazet80f8f102011-01-18 07:46:52 +00001584 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585
David S. Millerdbcb5852007-01-24 15:21:02 -08001586 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587}
1588
1589/*
Eric Dumazet62ab0812010-12-06 20:50:09 +00001590 * This function makes lazy skb cloning in hope that most of packets
1591 * are discarded by BPF.
1592 *
1593 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1594 * and skb->cb are mangled. It works because (and until) packets
1595 * falling here are owned by current CPU. Output packets are cloned
1596 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1597 * sequencially, so that if we return skb to original state on exit,
1598 * we will not harm anyone.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 */
1600
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001601static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1602 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603{
1604 struct sock *sk;
1605 struct sockaddr_ll *sll;
1606 struct packet_sock *po;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001607 u8 *skb_head = skb->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 int skb_len = skb->len;
David S. Millerdbcb5852007-01-24 15:21:02 -08001609 unsigned int snaplen, res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610
1611 if (skb->pkt_type == PACKET_LOOPBACK)
1612 goto drop;
1613
1614 sk = pt->af_packet_priv;
1615 po = pkt_sk(sk);
1616
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001617 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08001618 goto drop;
1619
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 skb->dev = dev;
1621
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -07001622 if (dev->header_ops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 /* The device has an explicit notion of ll header,
Eric Dumazet62ab0812010-12-06 20:50:09 +00001624 * exported to higher levels.
1625 *
1626 * Otherwise, the device hides details of its frame
1627 * structure, so that corresponding packet head is
1628 * never delivered to user.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 */
1630 if (sk->sk_type != SOCK_DGRAM)
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001631 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 else if (skb->pkt_type == PACKET_OUTGOING) {
1633 /* Special case: outgoing packets have ll header at head */
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001634 skb_pull(skb, skb_network_offset(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 }
1636 }
1637
1638 snaplen = skb->len;
1639
David S. Millerdbcb5852007-01-24 15:21:02 -08001640 res = run_filter(skb, sk, snaplen);
1641 if (!res)
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07001642 goto drop_n_restore;
David S. Millerdbcb5852007-01-24 15:21:02 -08001643 if (snaplen > res)
1644 snaplen = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
Eric Dumazet0fd7bac2011-12-21 07:11:44 +00001646 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 goto drop_n_acct;
1648
1649 if (skb_shared(skb)) {
1650 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1651 if (nskb == NULL)
1652 goto drop_n_acct;
1653
1654 if (skb_head != skb->data) {
1655 skb->data = skb_head;
1656 skb->len = skb_len;
1657 }
Eric Dumazetabc4e4f2012-04-19 02:24:42 +00001658 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 skb = nskb;
1660 }
1661
Herbert Xuffbc6112007-02-04 23:33:10 -08001662 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1663 sizeof(skb->cb));
1664
1665 sll = &PACKET_SKB_CB(skb)->sa.ll;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 sll->sll_family = AF_PACKET;
1667 sll->sll_hatype = dev->type;
1668 sll->sll_protocol = skb->protocol;
1669 sll->sll_pkttype = skb->pkt_type;
Peter P Waskiewicz Jr8032b462007-11-10 22:03:25 -08001670 if (unlikely(po->origdev))
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07001671 sll->sll_ifindex = orig_dev->ifindex;
1672 else
1673 sll->sll_ifindex = dev->ifindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Stephen Hemmingerb95cce32007-09-26 22:13:38 -07001675 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676
Herbert Xuffbc6112007-02-04 23:33:10 -08001677 PACKET_SKB_CB(skb)->origlen = skb->len;
Herbert Xu8dc41942007-02-04 23:31:32 -08001678
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 if (pskb_trim(skb, snaplen))
1680 goto drop_n_acct;
1681
1682 skb_set_owner_r(skb, sk);
1683 skb->dev = NULL;
Eric Dumazetadf30902009-06-02 05:19:30 +00001684 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
Phil Oester84531c22005-07-12 11:57:52 -07001686 /* drop conntrack reference */
1687 nf_reset(skb);
1688
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 spin_lock(&sk->sk_receive_queue.lock);
1690 po->stats.tp_packets++;
Neil Horman3b885782009-10-12 13:26:31 -07001691 skb->dropcount = atomic_read(&sk->sk_drops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 __skb_queue_tail(&sk->sk_receive_queue, skb);
1693 spin_unlock(&sk->sk_receive_queue.lock);
1694 sk->sk_data_ready(sk, skb->len);
1695 return 0;
1696
1697drop_n_acct:
Willem de Bruijn7091fbd2011-09-30 10:38:28 +00001698 spin_lock(&sk->sk_receive_queue.lock);
1699 po->stats.tp_drops++;
1700 atomic_inc(&sk->sk_drops);
1701 spin_unlock(&sk->sk_receive_queue.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
1703drop_n_restore:
1704 if (skb_head != skb->data && skb_shared(skb)) {
1705 skb->data = skb_head;
1706 skb->len = skb_len;
1707 }
1708drop:
Neil Hormanead2ceb2009-03-11 09:49:55 +00001709 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 return 0;
1711}
1712
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001713static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1714 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715{
1716 struct sock *sk;
1717 struct packet_sock *po;
1718 struct sockaddr_ll *sll;
Daniel Borkmann184f4892013-04-16 01:57:46 +00001719 union tpacket_uhdr h;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001720 u8 *skb_head = skb->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 int skb_len = skb->len;
David S. Millerdbcb5852007-01-24 15:21:02 -08001722 unsigned int snaplen, res;
chetan lokef6fb8f102011-08-19 10:18:16 +00001723 unsigned long status = TP_STATUS_USER;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001724 unsigned short macoff, netoff, hdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 struct sk_buff *copy_skb = NULL;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001726 struct timespec ts;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001727 __u32 ts_status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
1729 if (skb->pkt_type == PACKET_LOOPBACK)
1730 goto drop;
1731
1732 sk = pt->af_packet_priv;
1733 po = pkt_sk(sk);
1734
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001735 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08001736 goto drop;
1737
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -07001738 if (dev->header_ops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 if (sk->sk_type != SOCK_DGRAM)
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001740 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 else if (skb->pkt_type == PACKET_OUTGOING) {
1742 /* Special case: outgoing packets have ll header at head */
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001743 skb_pull(skb, skb_network_offset(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 }
1745 }
1746
Herbert Xu8dc41942007-02-04 23:31:32 -08001747 if (skb->ip_summed == CHECKSUM_PARTIAL)
1748 status |= TP_STATUS_CSUMNOTREADY;
1749
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 snaplen = skb->len;
1751
David S. Millerdbcb5852007-01-24 15:21:02 -08001752 res = run_filter(skb, sk, snaplen);
1753 if (!res)
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07001754 goto drop_n_restore;
David S. Millerdbcb5852007-01-24 15:21:02 -08001755 if (snaplen > res)
1756 snaplen = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
1758 if (sk->sk_type == SOCK_DGRAM) {
Patrick McHardy8913336a2008-07-18 18:05:19 -07001759 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1760 po->tp_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 } else {
Eric Dumazet95c96172012-04-15 05:58:06 +00001762 unsigned int maclen = skb_network_offset(skb);
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001763 netoff = TPACKET_ALIGN(po->tp_hdrlen +
Patrick McHardy8913336a2008-07-18 18:05:19 -07001764 (maclen < 16 ? 16 : maclen)) +
1765 po->tp_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 macoff = netoff - maclen;
1767 }
chetan lokef6fb8f102011-08-19 10:18:16 +00001768 if (po->tp_version <= TPACKET_V2) {
1769 if (macoff + snaplen > po->rx_ring.frame_size) {
1770 if (po->copy_thresh &&
Eric Dumazet0fd7bac2011-12-21 07:11:44 +00001771 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
chetan lokef6fb8f102011-08-19 10:18:16 +00001772 if (skb_shared(skb)) {
1773 copy_skb = skb_clone(skb, GFP_ATOMIC);
1774 } else {
1775 copy_skb = skb_get(skb);
1776 skb_head = skb->data;
1777 }
1778 if (copy_skb)
1779 skb_set_owner_r(copy_skb, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 }
chetan lokef6fb8f102011-08-19 10:18:16 +00001781 snaplen = po->rx_ring.frame_size - macoff;
1782 if ((int)snaplen < 0)
1783 snaplen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 spin_lock(&sk->sk_receive_queue.lock);
chetan lokef6fb8f102011-08-19 10:18:16 +00001787 h.raw = packet_current_rx_frame(po, skb,
1788 TP_STATUS_KERNEL, (macoff+snaplen));
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001789 if (!h.raw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 goto ring_is_full;
chetan lokef6fb8f102011-08-19 10:18:16 +00001791 if (po->tp_version <= TPACKET_V2) {
1792 packet_increment_rx_head(po, &po->rx_ring);
1793 /*
1794 * LOSING will be reported till you read the stats,
1795 * because it's COR - Clear On Read.
1796 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1797 * at packet level.
1798 */
1799 if (po->stats.tp_drops)
1800 status |= TP_STATUS_LOSING;
1801 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 po->stats.tp_packets++;
1803 if (copy_skb) {
1804 status |= TP_STATUS_COPY;
1805 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 spin_unlock(&sk->sk_receive_queue.lock);
1808
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001809 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001810
1811 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
Daniel Borkmann7a513842013-04-23 00:39:29 +00001812 getnstimeofday(&ts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001814 status |= ts_status;
1815
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001816 switch (po->tp_version) {
1817 case TPACKET_V1:
1818 h.h1->tp_len = skb->len;
1819 h.h1->tp_snaplen = snaplen;
1820 h.h1->tp_mac = macoff;
1821 h.h1->tp_net = netoff;
Daniel Borkmann4b457bd2013-04-16 01:29:11 +00001822 h.h1->tp_sec = ts.tv_sec;
1823 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001824 hdrlen = sizeof(*h.h1);
1825 break;
1826 case TPACKET_V2:
1827 h.h2->tp_len = skb->len;
1828 h.h2->tp_snaplen = snaplen;
1829 h.h2->tp_mac = macoff;
1830 h.h2->tp_net = netoff;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001831 h.h2->tp_sec = ts.tv_sec;
1832 h.h2->tp_nsec = ts.tv_nsec;
Ben Greeara3bcc232011-06-01 06:49:10 +00001833 if (vlan_tx_tag_present(skb)) {
1834 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1835 status |= TP_STATUS_VLAN_VALID;
1836 } else {
1837 h.h2->tp_vlan_tci = 0;
1838 }
Eric Dumazet13fcb7b2011-06-06 22:42:06 -07001839 h.h2->tp_padding = 0;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001840 hdrlen = sizeof(*h.h2);
1841 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00001842 case TPACKET_V3:
1843 /* tp_nxt_offset,vlan are already populated above.
1844 * So DONT clear those fields here
1845 */
1846 h.h3->tp_status |= status;
1847 h.h3->tp_len = skb->len;
1848 h.h3->tp_snaplen = snaplen;
1849 h.h3->tp_mac = macoff;
1850 h.h3->tp_net = netoff;
chetan lokef6fb8f102011-08-19 10:18:16 +00001851 h.h3->tp_sec = ts.tv_sec;
1852 h.h3->tp_nsec = ts.tv_nsec;
1853 hdrlen = sizeof(*h.h3);
1854 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001855 default:
1856 BUG();
1857 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001859 sll = h.raw + TPACKET_ALIGN(hdrlen);
Stephen Hemmingerb95cce32007-09-26 22:13:38 -07001860 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 sll->sll_family = AF_PACKET;
1862 sll->sll_hatype = dev->type;
1863 sll->sll_protocol = skb->protocol;
1864 sll->sll_pkttype = skb->pkt_type;
Peter P Waskiewicz Jr8032b462007-11-10 22:03:25 -08001865 if (unlikely(po->origdev))
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07001866 sll->sll_ifindex = orig_dev->ifindex;
1867 else
1868 sll->sll_ifindex = dev->ifindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869
Ralf Baechlee16aa202006-12-07 00:11:33 -08001870 smp_mb();
Changli Gaof6dafa92010-12-07 04:26:16 +00001871#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 {
Changli Gao0af55bb2010-12-01 02:52:20 +00001873 u8 *start, *end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874
chetan lokef6fb8f102011-08-19 10:18:16 +00001875 if (po->tp_version <= TPACKET_V2) {
1876 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1877 + macoff + snaplen);
1878 for (start = h.raw; start < end; start += PAGE_SIZE)
1879 flush_dcache_page(pgv_to_page(start));
1880 }
Chetan Lokecc9f01b2011-07-14 08:36:33 -07001881 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 }
Changli Gaof6dafa92010-12-07 04:26:16 +00001883#endif
chetan lokef6fb8f102011-08-19 10:18:16 +00001884 if (po->tp_version <= TPACKET_V2)
1885 __packet_set_status(po, h.raw, status);
1886 else
1887 prb_clear_blk_fill_status(&po->rx_ring);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888
1889 sk->sk_data_ready(sk, 0);
1890
1891drop_n_restore:
1892 if (skb_head != skb->data && skb_shared(skb)) {
1893 skb->data = skb_head;
1894 skb->len = skb_len;
1895 }
1896drop:
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001897 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 return 0;
1899
1900ring_is_full:
1901 po->stats.tp_drops++;
1902 spin_unlock(&sk->sk_receive_queue.lock);
1903
1904 sk->sk_data_ready(sk, 0);
Wei Yongjunacb5d752009-02-25 00:36:42 +00001905 kfree_skb(copy_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 goto drop_n_restore;
1907}
1908
Johann Baudy69e3c752009-05-18 22:11:22 -07001909static void tpacket_destruct_skb(struct sk_buff *skb)
1910{
1911 struct packet_sock *po = pkt_sk(skb->sk);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001912 void *ph;
Johann Baudy69e3c752009-05-18 22:11:22 -07001913
Johann Baudy69e3c752009-05-18 22:11:22 -07001914 if (likely(po->tx_ring.pg_vec)) {
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001915 __u32 ts;
1916
Johann Baudy69e3c752009-05-18 22:11:22 -07001917 ph = skb_shinfo(skb)->destructor_arg;
Johann Baudy69e3c752009-05-18 22:11:22 -07001918 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1919 atomic_dec(&po->tx_ring.pending);
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001920
1921 ts = __packet_set_timestamp(po, ph, skb);
1922 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
Johann Baudy69e3c752009-05-18 22:11:22 -07001923 }
1924
1925 sock_wfree(skb);
1926}
1927
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001928static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1929 void *frame, struct net_device *dev, int size_max,
Herbert Xuae641942011-11-18 02:20:04 +00001930 __be16 proto, unsigned char *addr, int hlen)
Johann Baudy69e3c752009-05-18 22:11:22 -07001931{
Daniel Borkmann184f4892013-04-16 01:57:46 +00001932 union tpacket_uhdr ph;
Johann Baudy69e3c752009-05-18 22:11:22 -07001933 int to_write, offset, len, tp_len, nr_frags, len_max;
1934 struct socket *sock = po->sk.sk_socket;
1935 struct page *page;
1936 void *data;
1937 int err;
1938
1939 ph.raw = frame;
1940
1941 skb->protocol = proto;
1942 skb->dev = dev;
1943 skb->priority = po->sk.sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00001944 skb->mark = po->sk.sk_mark;
Willem de Bruijn2e313962013-04-23 00:39:28 +00001945 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
Johann Baudy69e3c752009-05-18 22:11:22 -07001946 skb_shinfo(skb)->destructor_arg = ph.raw;
1947
1948 switch (po->tp_version) {
1949 case TPACKET_V2:
1950 tp_len = ph.h2->tp_len;
1951 break;
1952 default:
1953 tp_len = ph.h1->tp_len;
1954 break;
1955 }
1956 if (unlikely(tp_len > size_max)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001957 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
Johann Baudy69e3c752009-05-18 22:11:22 -07001958 return -EMSGSIZE;
1959 }
1960
Herbert Xuae641942011-11-18 02:20:04 +00001961 skb_reserve(skb, hlen);
Johann Baudy69e3c752009-05-18 22:11:22 -07001962 skb_reset_network_header(skb);
Jason Wang40893fd2013-03-26 23:11:22 +00001963 skb_probe_transport_header(skb, 0);
Jason Wangc1aad272013-03-25 20:19:57 +00001964
Paul Chavent5920cd3a2012-11-06 23:10:47 +00001965 if (po->tp_tx_has_off) {
1966 int off_min, off_max, off;
1967 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
1968 off_max = po->tx_ring.frame_size - tp_len;
1969 if (sock->type == SOCK_DGRAM) {
1970 switch (po->tp_version) {
1971 case TPACKET_V2:
1972 off = ph.h2->tp_net;
1973 break;
1974 default:
1975 off = ph.h1->tp_net;
1976 break;
1977 }
1978 } else {
1979 switch (po->tp_version) {
1980 case TPACKET_V2:
1981 off = ph.h2->tp_mac;
1982 break;
1983 default:
1984 off = ph.h1->tp_mac;
1985 break;
1986 }
1987 }
1988 if (unlikely((off < off_min) || (off_max < off)))
1989 return -EINVAL;
1990 data = ph.raw + off;
1991 } else {
1992 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1993 }
Johann Baudy69e3c752009-05-18 22:11:22 -07001994 to_write = tp_len;
1995
1996 if (sock->type == SOCK_DGRAM) {
1997 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1998 NULL, tp_len);
1999 if (unlikely(err < 0))
2000 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002001 } else if (dev->hard_header_len) {
Johann Baudy69e3c752009-05-18 22:11:22 -07002002 /* net device doesn't like empty head */
2003 if (unlikely(tp_len <= dev->hard_header_len)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002004 pr_err("packet size is too short (%d < %d)\n",
2005 tp_len, dev->hard_header_len);
Johann Baudy69e3c752009-05-18 22:11:22 -07002006 return -EINVAL;
2007 }
2008
2009 skb_push(skb, dev->hard_header_len);
2010 err = skb_store_bits(skb, 0, data,
2011 dev->hard_header_len);
2012 if (unlikely(err))
2013 return err;
2014
2015 data += dev->hard_header_len;
2016 to_write -= dev->hard_header_len;
2017 }
2018
Johann Baudy69e3c752009-05-18 22:11:22 -07002019 offset = offset_in_page(data);
2020 len_max = PAGE_SIZE - offset;
2021 len = ((to_write > len_max) ? len_max : to_write);
2022
2023 skb->data_len = to_write;
2024 skb->len += to_write;
2025 skb->truesize += to_write;
2026 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2027
2028 while (likely(to_write)) {
2029 nr_frags = skb_shinfo(skb)->nr_frags;
2030
2031 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002032 pr_err("Packet exceed the number of skb frags(%lu)\n",
2033 MAX_SKB_FRAGS);
Johann Baudy69e3c752009-05-18 22:11:22 -07002034 return -EFAULT;
2035 }
2036
Changli Gao0af55bb2010-12-01 02:52:20 +00002037 page = pgv_to_page(data);
2038 data += len;
Johann Baudy69e3c752009-05-18 22:11:22 -07002039 flush_dcache_page(page);
2040 get_page(page);
Changli Gao0af55bb2010-12-01 02:52:20 +00002041 skb_fill_page_desc(skb, nr_frags, page, offset, len);
Johann Baudy69e3c752009-05-18 22:11:22 -07002042 to_write -= len;
2043 offset = 0;
2044 len_max = PAGE_SIZE;
2045 len = ((to_write > len_max) ? len_max : to_write);
2046 }
2047
2048 return tp_len;
2049}
2050
2051static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2052{
Johann Baudy69e3c752009-05-18 22:11:22 -07002053 struct sk_buff *skb;
2054 struct net_device *dev;
2055 __be16 proto;
Ben Greear827d9782011-06-01 07:18:53 +00002056 bool need_rls_dev = false;
2057 int err, reserve = 0;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002058 void *ph;
2059 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
Johann Baudy69e3c752009-05-18 22:11:22 -07002060 int tp_len, size_max;
2061 unsigned char *addr;
2062 int len_sum = 0;
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +00002063 int status = TP_STATUS_AVAILABLE;
Herbert Xuae641942011-11-18 02:20:04 +00002064 int hlen, tlen;
Johann Baudy69e3c752009-05-18 22:11:22 -07002065
Johann Baudy69e3c752009-05-18 22:11:22 -07002066 mutex_lock(&po->pg_vec_lock);
2067
Johann Baudy69e3c752009-05-18 22:11:22 -07002068 if (saddr == NULL) {
Ben Greear827d9782011-06-01 07:18:53 +00002069 dev = po->prot_hook.dev;
Johann Baudy69e3c752009-05-18 22:11:22 -07002070 proto = po->num;
2071 addr = NULL;
2072 } else {
2073 err = -EINVAL;
2074 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2075 goto out;
2076 if (msg->msg_namelen < (saddr->sll_halen
2077 + offsetof(struct sockaddr_ll,
2078 sll_addr)))
2079 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07002080 proto = saddr->sll_protocol;
2081 addr = saddr->sll_addr;
Ben Greear827d9782011-06-01 07:18:53 +00002082 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2083 need_rls_dev = true;
Johann Baudy69e3c752009-05-18 22:11:22 -07002084 }
2085
Johann Baudy69e3c752009-05-18 22:11:22 -07002086 err = -ENXIO;
2087 if (unlikely(dev == NULL))
2088 goto out;
2089
2090 reserve = dev->hard_header_len;
2091
2092 err = -ENETDOWN;
2093 if (unlikely(!(dev->flags & IFF_UP)))
2094 goto out_put;
2095
2096 size_max = po->tx_ring.frame_size
Gabor Gombasb5dd8842009-10-29 03:19:11 -07002097 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
Johann Baudy69e3c752009-05-18 22:11:22 -07002098
2099 if (size_max > dev->mtu + reserve)
2100 size_max = dev->mtu + reserve;
2101
2102 do {
2103 ph = packet_current_frame(po, &po->tx_ring,
2104 TP_STATUS_SEND_REQUEST);
2105
2106 if (unlikely(ph == NULL)) {
2107 schedule();
2108 continue;
2109 }
2110
2111 status = TP_STATUS_SEND_REQUEST;
Herbert Xuae641942011-11-18 02:20:04 +00002112 hlen = LL_RESERVED_SPACE(dev);
2113 tlen = dev->needed_tailroom;
Johann Baudy69e3c752009-05-18 22:11:22 -07002114 skb = sock_alloc_send_skb(&po->sk,
Herbert Xuae641942011-11-18 02:20:04 +00002115 hlen + tlen + sizeof(struct sockaddr_ll),
Johann Baudy69e3c752009-05-18 22:11:22 -07002116 0, &err);
2117
2118 if (unlikely(skb == NULL))
2119 goto out_status;
2120
2121 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
Herbert Xuae641942011-11-18 02:20:04 +00002122 addr, hlen);
Johann Baudy69e3c752009-05-18 22:11:22 -07002123
2124 if (unlikely(tp_len < 0)) {
2125 if (po->tp_loss) {
2126 __packet_set_status(po, ph,
2127 TP_STATUS_AVAILABLE);
2128 packet_increment_head(&po->tx_ring);
2129 kfree_skb(skb);
2130 continue;
2131 } else {
2132 status = TP_STATUS_WRONG_FORMAT;
2133 err = tp_len;
2134 goto out_status;
2135 }
2136 }
2137
2138 skb->destructor = tpacket_destruct_skb;
2139 __packet_set_status(po, ph, TP_STATUS_SENDING);
2140 atomic_inc(&po->tx_ring.pending);
2141
2142 status = TP_STATUS_SEND_REQUEST;
2143 err = dev_queue_xmit(skb);
Jarek Poplawskieb70df12010-01-10 22:04:19 +00002144 if (unlikely(err > 0)) {
2145 err = net_xmit_errno(err);
2146 if (err && __packet_get_status(po, ph) ==
2147 TP_STATUS_AVAILABLE) {
2148 /* skb was destructed already */
2149 skb = NULL;
2150 goto out_status;
2151 }
2152 /*
2153 * skb was dropped but not destructed yet;
2154 * let's treat it like congestion or err < 0
2155 */
2156 err = 0;
2157 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002158 packet_increment_head(&po->tx_ring);
2159 len_sum += tp_len;
Joe Perchesf64f9e72009-11-29 16:55:45 -08002160 } while (likely((ph != NULL) ||
2161 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2162 (atomic_read(&po->tx_ring.pending))))
2163 );
Johann Baudy69e3c752009-05-18 22:11:22 -07002164
2165 err = len_sum;
2166 goto out_put;
2167
Johann Baudy69e3c752009-05-18 22:11:22 -07002168out_status:
2169 __packet_set_status(po, ph, status);
2170 kfree_skb(skb);
2171out_put:
Ben Greear827d9782011-06-01 07:18:53 +00002172 if (need_rls_dev)
2173 dev_put(dev);
Johann Baudy69e3c752009-05-18 22:11:22 -07002174out:
2175 mutex_unlock(&po->pg_vec_lock);
2176 return err;
2177}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
Olof Johanssoneea49cc92011-11-02 11:00:49 +00002179static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2180 size_t reserve, size_t len,
2181 size_t linear, int noblock,
2182 int *err)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002183{
2184 struct sk_buff *skb;
2185
2186 /* Under a page? Don't bother with paged skb. */
2187 if (prepad + len < PAGE_SIZE || !linear)
2188 linear = len;
2189
2190 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2191 err);
2192 if (!skb)
2193 return NULL;
2194
2195 skb_reserve(skb, reserve);
2196 skb_put(skb, linear);
2197 skb->data_len = len - linear;
2198 skb->len += len - linear;
2199
2200 return skb;
2201}
2202
Johann Baudy69e3c752009-05-18 22:11:22 -07002203static int packet_snd(struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 struct msghdr *msg, size_t len)
2205{
2206 struct sock *sk = sock->sk;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002207 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 struct sk_buff *skb;
2209 struct net_device *dev;
Al Viro0e11c912006-11-08 00:26:29 -08002210 __be16 proto;
Ben Greear827d9782011-06-01 07:18:53 +00002211 bool need_rls_dev = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 unsigned char *addr;
Ben Greear827d9782011-06-01 07:18:53 +00002213 int err, reserve = 0;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002214 struct virtio_net_hdr vnet_hdr = { 0 };
2215 int offset = 0;
2216 int vnet_hdr_len;
2217 struct packet_sock *po = pkt_sk(sk);
2218 unsigned short gso_type = 0;
Herbert Xuae641942011-11-18 02:20:04 +00002219 int hlen, tlen;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002220 int extra_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
2222 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002223 * Get and verify the address.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002225
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 if (saddr == NULL) {
Ben Greear827d9782011-06-01 07:18:53 +00002227 dev = po->prot_hook.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 proto = po->num;
2229 addr = NULL;
2230 } else {
2231 err = -EINVAL;
2232 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2233 goto out;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002234 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2235 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 proto = saddr->sll_protocol;
2237 addr = saddr->sll_addr;
Ben Greear827d9782011-06-01 07:18:53 +00002238 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2239 need_rls_dev = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 }
2241
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 err = -ENXIO;
2243 if (dev == NULL)
2244 goto out_unlock;
2245 if (sock->type == SOCK_RAW)
2246 reserve = dev->hard_header_len;
2247
David S. Millerd5e76b02007-01-25 19:30:36 -08002248 err = -ENETDOWN;
2249 if (!(dev->flags & IFF_UP))
2250 goto out_unlock;
2251
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002252 if (po->has_vnet_hdr) {
2253 vnet_hdr_len = sizeof(vnet_hdr);
2254
2255 err = -EINVAL;
2256 if (len < vnet_hdr_len)
2257 goto out_unlock;
2258
2259 len -= vnet_hdr_len;
2260
2261 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2262 vnet_hdr_len);
2263 if (err < 0)
2264 goto out_unlock;
2265
2266 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2267 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2268 vnet_hdr.hdr_len))
2269 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2270 vnet_hdr.csum_offset + 2;
2271
2272 err = -EINVAL;
2273 if (vnet_hdr.hdr_len > len)
2274 goto out_unlock;
2275
2276 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2277 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2278 case VIRTIO_NET_HDR_GSO_TCPV4:
2279 gso_type = SKB_GSO_TCPV4;
2280 break;
2281 case VIRTIO_NET_HDR_GSO_TCPV6:
2282 gso_type = SKB_GSO_TCPV6;
2283 break;
2284 case VIRTIO_NET_HDR_GSO_UDP:
2285 gso_type = SKB_GSO_UDP;
2286 break;
2287 default:
2288 goto out_unlock;
2289 }
2290
2291 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2292 gso_type |= SKB_GSO_TCP_ECN;
2293
2294 if (vnet_hdr.gso_size == 0)
2295 goto out_unlock;
2296
2297 }
2298 }
2299
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002300 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2301 if (!netif_supports_nofcs(dev)) {
2302 err = -EPROTONOSUPPORT;
2303 goto out_unlock;
2304 }
2305 extra_len = 4; /* We're doing our own CRC */
2306 }
2307
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 err = -EMSGSIZE;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002309 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 goto out_unlock;
2311
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002312 err = -ENOBUFS;
Herbert Xuae641942011-11-18 02:20:04 +00002313 hlen = LL_RESERVED_SPACE(dev);
2314 tlen = dev->needed_tailroom;
2315 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002316 msg->msg_flags & MSG_DONTWAIT, &err);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002317 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318 goto out_unlock;
2319
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002320 skb_set_network_header(skb, reserve);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
Stephen Hemminger0c4e8582007-10-09 01:36:32 -07002322 err = -EINVAL;
2323 if (sock->type == SOCK_DGRAM &&
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002324 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
Stephen Hemminger0c4e8582007-10-09 01:36:32 -07002325 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326
2327 /* Returns -EFAULT on error */
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002328 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 if (err)
2330 goto out_free;
Daniel Borkmannbf84a012013-04-14 08:08:13 +00002331
2332 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002334 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
Ben Greear57f89bf2011-02-11 09:35:18 +00002335 /* Earlier code assumed this would be a VLAN pkt,
2336 * double-check this now that we have the actual
2337 * packet in hand.
2338 */
2339 struct ethhdr *ehdr;
2340 skb_reset_mac_header(skb);
2341 ehdr = eth_hdr(skb);
2342 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2343 err = -EMSGSIZE;
2344 goto out_free;
2345 }
2346 }
2347
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 skb->protocol = proto;
2349 skb->dev = dev;
2350 skb->priority = sk->sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00002351 skb->mark = sk->sk_mark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002353 if (po->has_vnet_hdr) {
2354 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2355 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2356 vnet_hdr.csum_offset)) {
2357 err = -EINVAL;
2358 goto out_free;
2359 }
2360 }
2361
2362 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2363 skb_shinfo(skb)->gso_type = gso_type;
2364
2365 /* Header must be checked, and gso_segs computed. */
2366 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2367 skb_shinfo(skb)->gso_segs = 0;
2368
2369 len += vnet_hdr_len;
2370 }
2371
Jason Wang40893fd2013-03-26 23:11:22 +00002372 skb_probe_transport_header(skb, reserve);
Jason Wangc1aad272013-03-25 20:19:57 +00002373
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002374 if (unlikely(extra_len == 4))
2375 skb->no_fcs = 1;
2376
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 /*
2378 * Now send it
2379 */
2380
2381 err = dev_queue_xmit(skb);
2382 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2383 goto out_unlock;
2384
Ben Greear827d9782011-06-01 07:18:53 +00002385 if (need_rls_dev)
2386 dev_put(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002388 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
2390out_free:
2391 kfree_skb(skb);
2392out_unlock:
Ben Greear827d9782011-06-01 07:18:53 +00002393 if (dev && need_rls_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 dev_put(dev);
2395out:
2396 return err;
2397}
2398
Johann Baudy69e3c752009-05-18 22:11:22 -07002399static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2400 struct msghdr *msg, size_t len)
2401{
Johann Baudy69e3c752009-05-18 22:11:22 -07002402 struct sock *sk = sock->sk;
2403 struct packet_sock *po = pkt_sk(sk);
2404 if (po->tx_ring.pg_vec)
2405 return tpacket_snd(po, msg);
2406 else
Johann Baudy69e3c752009-05-18 22:11:22 -07002407 return packet_snd(sock, msg, len);
2408}
2409
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410/*
2411 * Close a PACKET socket. This is fairly simple. We immediately go
2412 * to 'closed' state and remove our protocol entry in the device list.
2413 */
2414
2415static int packet_release(struct socket *sock)
2416{
2417 struct sock *sk = sock->sk;
2418 struct packet_sock *po;
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08002419 struct net *net;
chetan lokef6fb8f102011-08-19 10:18:16 +00002420 union tpacket_req_u req_u;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
2422 if (!sk)
2423 return 0;
2424
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002425 net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 po = pkt_sk(sk);
2427
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002428 mutex_lock(&net->packet.sklist_lock);
stephen hemminger808f5112010-02-22 07:57:18 +00002429 sk_del_node_init_rcu(sk);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002430 mutex_unlock(&net->packet.sklist_lock);
2431
2432 preempt_disable();
Eric Dumazet920de802008-11-24 00:09:29 -08002433 sock_prot_inuse_add(net, sk->sk_prot, -1);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002434 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435
stephen hemminger808f5112010-02-22 07:57:18 +00002436 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07002437 unregister_prot_hook(sk, false);
Ben Greear160ff182011-06-01 07:18:52 +00002438 if (po->prot_hook.dev) {
2439 dev_put(po->prot_hook.dev);
2440 po->prot_hook.dev = NULL;
2441 }
stephen hemminger808f5112010-02-22 07:57:18 +00002442 spin_unlock(&po->bind_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 packet_flush_mclist(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445
Phil Sutter9665d5d2013-02-01 07:21:41 +00002446 if (po->rx_ring.pg_vec) {
2447 memset(&req_u, 0, sizeof(req_u));
chetan lokef6fb8f102011-08-19 10:18:16 +00002448 packet_set_ring(sk, &req_u, 1, 0);
Phil Sutter9665d5d2013-02-01 07:21:41 +00002449 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002450
Phil Sutter9665d5d2013-02-01 07:21:41 +00002451 if (po->tx_ring.pg_vec) {
2452 memset(&req_u, 0, sizeof(req_u));
chetan lokef6fb8f102011-08-19 10:18:16 +00002453 packet_set_ring(sk, &req_u, 1, 1);
Phil Sutter9665d5d2013-02-01 07:21:41 +00002454 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455
David S. Millerdc99f602011-07-05 01:45:05 -07002456 fanout_release(sk);
2457
stephen hemminger808f5112010-02-22 07:57:18 +00002458 synchronize_net();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 /*
2460 * Now the socket is dead. No more input will appear.
2461 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 sock_orphan(sk);
2463 sock->sk = NULL;
2464
2465 /* Purge queues */
2466
2467 skb_queue_purge(&sk->sk_receive_queue);
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08002468 sk_refcnt_debug_release(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469
2470 sock_put(sk);
2471 return 0;
2472}
2473
2474/*
2475 * Attach a packet hook.
2476 */
2477
Al Viro0e11c912006-11-08 00:26:29 -08002478static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479{
2480 struct packet_sock *po = pkt_sk(sk);
David S. Millerdc99f602011-07-05 01:45:05 -07002481
Wei Yongjunaef950b2011-12-27 22:32:41 -05002482 if (po->fanout) {
2483 if (dev)
2484 dev_put(dev);
2485
David S. Millerdc99f602011-07-05 01:45:05 -07002486 return -EINVAL;
Wei Yongjunaef950b2011-12-27 22:32:41 -05002487 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488
2489 lock_sock(sk);
2490
2491 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07002492 unregister_prot_hook(sk, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 po->num = protocol;
2494 po->prot_hook.type = protocol;
Ben Greear160ff182011-06-01 07:18:52 +00002495 if (po->prot_hook.dev)
2496 dev_put(po->prot_hook.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 po->prot_hook.dev = dev;
2498
2499 po->ifindex = dev ? dev->ifindex : 0;
2500
2501 if (protocol == 0)
2502 goto out_unlock;
2503
Urs Thuermannbe85d4a2007-11-12 21:05:20 -08002504 if (!dev || (dev->flags & IFF_UP)) {
David S. Millerce06b032011-07-04 01:44:29 -07002505 register_prot_hook(sk);
Urs Thuermannbe85d4a2007-11-12 21:05:20 -08002506 } else {
2507 sk->sk_err = ENETDOWN;
2508 if (!sock_flag(sk, SOCK_DEAD))
2509 sk->sk_error_report(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 }
2511
2512out_unlock:
2513 spin_unlock(&po->bind_lock);
2514 release_sock(sk);
2515 return 0;
2516}
2517
2518/*
2519 * Bind a packet socket to a device
2520 */
2521
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002522static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2523 int addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002525 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 char name[15];
2527 struct net_device *dev;
2528 int err = -ENODEV;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002529
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530 /*
2531 * Check legality
2532 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002533
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08002534 if (addr_len != sizeof(struct sockaddr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002536 strlcpy(name, uaddr->sa_data, sizeof(name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002538 dev = dev_get_by_name(sock_net(sk), name);
Ben Greear160ff182011-06-01 07:18:52 +00002539 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541 return err;
2542}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543
2544static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2545{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002546 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2547 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002548 struct net_device *dev = NULL;
2549 int err;
2550
2551
2552 /*
2553 * Check legality
2554 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002555
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 if (addr_len < sizeof(struct sockaddr_ll))
2557 return -EINVAL;
2558 if (sll->sll_family != AF_PACKET)
2559 return -EINVAL;
2560
2561 if (sll->sll_ifindex) {
2562 err = -ENODEV;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002563 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 if (dev == NULL)
2565 goto out;
2566 }
2567 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568
2569out:
2570 return err;
2571}
2572
2573static struct proto packet_proto = {
2574 .name = "PACKET",
2575 .owner = THIS_MODULE,
2576 .obj_size = sizeof(struct packet_sock),
2577};
2578
2579/*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002580 * Create a packet of type SOCK_PACKET.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 */
2582
Eric Paris3f378b62009-11-05 22:18:14 -08002583static int packet_create(struct net *net, struct socket *sock, int protocol,
2584 int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585{
2586 struct sock *sk;
2587 struct packet_sock *po;
Al Viro0e11c912006-11-08 00:26:29 -08002588 __be16 proto = (__force __be16)protocol; /* weird, but documented */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 int err;
2590
Eric W. Biedermandf008c92012-11-16 03:03:07 +00002591 if (!ns_capable(net->user_ns, CAP_NET_RAW))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 return -EPERM;
David S. Millerbe020972007-05-29 13:16:31 -07002593 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2594 sock->type != SOCK_PACKET)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 return -ESOCKTNOSUPPORT;
2596
2597 sock->state = SS_UNCONNECTED;
2598
2599 err = -ENOBUFS;
Pavel Emelyanov6257ff22007-11-01 00:39:31 -07002600 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 if (sk == NULL)
2602 goto out;
2603
2604 sock->ops = &packet_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 if (sock->type == SOCK_PACKET)
2606 sock->ops = &packet_ops_spkt;
David S. Millerbe020972007-05-29 13:16:31 -07002607
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608 sock_init_data(sock, sk);
2609
2610 po = pkt_sk(sk);
2611 sk->sk_family = PF_PACKET;
Al Viro0e11c912006-11-08 00:26:29 -08002612 po->num = proto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613
2614 sk->sk_destruct = packet_sock_destruct;
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08002615 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616
2617 /*
2618 * Attach a protocol block
2619 */
2620
2621 spin_lock_init(&po->bind_lock);
Herbert Xu905db442009-01-30 14:12:06 -08002622 mutex_init(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 po->prot_hook.func = packet_rcv;
David S. Millerbe020972007-05-29 13:16:31 -07002624
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625 if (sock->type == SOCK_PACKET)
2626 po->prot_hook.func = packet_rcv_spkt;
David S. Millerbe020972007-05-29 13:16:31 -07002627
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 po->prot_hook.af_packet_priv = sk;
2629
Al Viro0e11c912006-11-08 00:26:29 -08002630 if (proto) {
2631 po->prot_hook.type = proto;
David S. Millerce06b032011-07-04 01:44:29 -07002632 register_prot_hook(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002633 }
2634
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002635 mutex_lock(&net->packet.sklist_lock);
stephen hemminger808f5112010-02-22 07:57:18 +00002636 sk_add_node_rcu(sk, &net->packet.sklist);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002637 mutex_unlock(&net->packet.sklist_lock);
2638
2639 preempt_disable();
Eric Dumazet36804532008-11-19 14:25:35 -08002640 sock_prot_inuse_add(net, &packet_proto, 1);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002641 preempt_enable();
stephen hemminger808f5112010-02-22 07:57:18 +00002642
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002643 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644out:
2645 return err;
2646}
2647
Richard Cochraned85b562010-04-07 22:41:28 +00002648static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2649{
2650 struct sock_exterr_skb *serr;
2651 struct sk_buff *skb, *skb2;
2652 int copied, err;
2653
2654 err = -EAGAIN;
2655 skb = skb_dequeue(&sk->sk_error_queue);
2656 if (skb == NULL)
2657 goto out;
2658
2659 copied = skb->len;
2660 if (copied > len) {
2661 msg->msg_flags |= MSG_TRUNC;
2662 copied = len;
2663 }
2664 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2665 if (err)
2666 goto out_free_skb;
2667
2668 sock_recv_timestamp(msg, sk, skb);
2669
2670 serr = SKB_EXT_ERR(skb);
2671 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2672 sizeof(serr->ee), &serr->ee);
2673
2674 msg->msg_flags |= MSG_ERRQUEUE;
2675 err = copied;
2676
2677 /* Reset and regenerate socket error */
2678 spin_lock_bh(&sk->sk_error_queue.lock);
2679 sk->sk_err = 0;
2680 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2681 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2682 spin_unlock_bh(&sk->sk_error_queue.lock);
2683 sk->sk_error_report(sk);
2684 } else
2685 spin_unlock_bh(&sk->sk_error_queue.lock);
2686
2687out_free_skb:
2688 kfree_skb(skb);
2689out:
2690 return err;
2691}
2692
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693/*
2694 * Pull a packet from our receive queue and hand it to the user.
2695 * If necessary we block.
2696 */
2697
2698static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2699 struct msghdr *msg, size_t len, int flags)
2700{
2701 struct sock *sk = sock->sk;
2702 struct sk_buff *skb;
2703 int copied, err;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002704 struct sockaddr_ll *sll;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002705 int vnet_hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
2707 err = -EINVAL;
Richard Cochraned85b562010-04-07 22:41:28 +00002708 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 goto out;
2710
2711#if 0
2712 /* What error should we return now? EUNATTACH? */
2713 if (pkt_sk(sk)->ifindex < 0)
2714 return -ENODEV;
2715#endif
2716
Richard Cochraned85b562010-04-07 22:41:28 +00002717 if (flags & MSG_ERRQUEUE) {
2718 err = packet_recv_error(sk, msg, len);
2719 goto out;
2720 }
2721
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 * Call the generic datagram receiver. This handles all sorts
2724 * of horrible races and re-entrancy so we can forget about it
2725 * in the protocol layers.
2726 *
2727 * Now it will return ENETDOWN, if device have just gone down,
2728 * but then it will block.
2729 */
2730
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002731 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732
2733 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002734 * An error occurred so return it. Because skb_recv_datagram()
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 * handles the blocking we don't see and worry about blocking
2736 * retries.
2737 */
2738
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08002739 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 goto out;
2741
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002742 if (pkt_sk(sk)->has_vnet_hdr) {
2743 struct virtio_net_hdr vnet_hdr = { 0 };
2744
2745 err = -EINVAL;
2746 vnet_hdr_len = sizeof(vnet_hdr);
Mariusz Kozlowski1f18b712010-11-08 11:58:45 +00002747 if (len < vnet_hdr_len)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002748 goto out_free;
2749
Mariusz Kozlowski1f18b712010-11-08 11:58:45 +00002750 len -= vnet_hdr_len;
2751
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002752 if (skb_is_gso(skb)) {
2753 struct skb_shared_info *sinfo = skb_shinfo(skb);
2754
2755 /* This is a hint as to how much should be linear. */
2756 vnet_hdr.hdr_len = skb_headlen(skb);
2757 vnet_hdr.gso_size = sinfo->gso_size;
2758 if (sinfo->gso_type & SKB_GSO_TCPV4)
2759 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2760 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2761 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2762 else if (sinfo->gso_type & SKB_GSO_UDP)
2763 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2764 else if (sinfo->gso_type & SKB_GSO_FCOE)
2765 goto out_free;
2766 else
2767 BUG();
2768 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2769 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2770 } else
2771 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2772
2773 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2774 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Michał Mirosław55508d62010-12-14 15:24:08 +00002775 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002776 vnet_hdr.csum_offset = skb->csum_offset;
Jason Wang10a8d942011-06-10 00:56:17 +00002777 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2778 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002779 } /* else everything is zero */
2780
2781 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2782 vnet_hdr_len);
2783 if (err < 0)
2784 goto out_free;
2785 }
2786
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 /*
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002788 * If the address length field is there to be filled in, we fill
2789 * it in now.
2790 */
2791
Herbert Xuffbc6112007-02-04 23:33:10 -08002792 sll = &PACKET_SKB_CB(skb)->sa.ll;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002793 if (sock->type == SOCK_PACKET)
2794 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2795 else
2796 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2797
2798 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 * You lose any data beyond the buffer you gave. If it worries a
2800 * user program they can ask the device for its MTU anyway.
2801 */
2802
2803 copied = skb->len;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002804 if (copied > len) {
2805 copied = len;
2806 msg->msg_flags |= MSG_TRUNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807 }
2808
2809 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2810 if (err)
2811 goto out_free;
2812
Neil Horman3b885782009-10-12 13:26:31 -07002813 sock_recv_ts_and_drops(msg, sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814
2815 if (msg->msg_name)
Herbert Xuffbc6112007-02-04 23:33:10 -08002816 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2817 msg->msg_namelen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002818
Herbert Xu8dc41942007-02-04 23:31:32 -08002819 if (pkt_sk(sk)->auxdata) {
Herbert Xuffbc6112007-02-04 23:33:10 -08002820 struct tpacket_auxdata aux;
2821
2822 aux.tp_status = TP_STATUS_USER;
2823 if (skb->ip_summed == CHECKSUM_PARTIAL)
2824 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2825 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2826 aux.tp_snaplen = skb->len;
2827 aux.tp_mac = 0;
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03002828 aux.tp_net = skb_network_offset(skb);
Ben Greeara3bcc232011-06-01 06:49:10 +00002829 if (vlan_tx_tag_present(skb)) {
2830 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2831 aux.tp_status |= TP_STATUS_VLAN_VALID;
2832 } else {
2833 aux.tp_vlan_tci = 0;
2834 }
Eric Dumazet13fcb7b2011-06-06 22:42:06 -07002835 aux.tp_padding = 0;
Herbert Xuffbc6112007-02-04 23:33:10 -08002836 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
Herbert Xu8dc41942007-02-04 23:31:32 -08002837 }
2838
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 /*
2840 * Free or return the buffer as appropriate. Again this
2841 * hides all the races and re-entrancy issues from us.
2842 */
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002843 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844
2845out_free:
2846 skb_free_datagram(sk, skb);
2847out:
2848 return err;
2849}
2850
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2852 int *uaddr_len, int peer)
2853{
2854 struct net_device *dev;
2855 struct sock *sk = sock->sk;
2856
2857 if (peer)
2858 return -EOPNOTSUPP;
2859
2860 uaddr->sa_family = AF_PACKET;
Eric Dumazet654d1f82009-11-02 10:43:32 +01002861 rcu_read_lock();
2862 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2863 if (dev)
Vasiliy Kulikov67286642010-11-10 12:09:10 -08002864 strncpy(uaddr->sa_data, dev->name, 14);
Eric Dumazet654d1f82009-11-02 10:43:32 +01002865 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 memset(uaddr->sa_data, 0, 14);
Eric Dumazet654d1f82009-11-02 10:43:32 +01002867 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868 *uaddr_len = sizeof(*uaddr);
2869
2870 return 0;
2871}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872
2873static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2874 int *uaddr_len, int peer)
2875{
2876 struct net_device *dev;
2877 struct sock *sk = sock->sk;
2878 struct packet_sock *po = pkt_sk(sk);
Cyrill Gorcunov13cfa972009-11-08 05:51:19 +00002879 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002880
2881 if (peer)
2882 return -EOPNOTSUPP;
2883
2884 sll->sll_family = AF_PACKET;
2885 sll->sll_ifindex = po->ifindex;
2886 sll->sll_protocol = po->num;
Vasiliy Kulikov67286642010-11-10 12:09:10 -08002887 sll->sll_pkttype = 0;
Eric Dumazet654d1f82009-11-02 10:43:32 +01002888 rcu_read_lock();
2889 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890 if (dev) {
2891 sll->sll_hatype = dev->type;
2892 sll->sll_halen = dev->addr_len;
2893 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 } else {
2895 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2896 sll->sll_halen = 0;
2897 }
Eric Dumazet654d1f82009-11-02 10:43:32 +01002898 rcu_read_unlock();
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002899 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
2901 return 0;
2902}
2903
Wang Chen2aeb0b82008-07-14 20:49:46 -07002904static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2905 int what)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906{
2907 switch (i->type) {
2908 case PACKET_MR_MULTICAST:
Jiri Pirko11625632010-03-02 20:40:01 +00002909 if (i->alen != dev->addr_len)
2910 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 if (what > 0)
Jiri Pirko22bedad32010-04-01 21:22:57 +00002912 return dev_mc_add(dev, i->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 else
Jiri Pirko22bedad32010-04-01 21:22:57 +00002914 return dev_mc_del(dev, i->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915 break;
2916 case PACKET_MR_PROMISC:
Wang Chen2aeb0b82008-07-14 20:49:46 -07002917 return dev_set_promiscuity(dev, what);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002918 break;
2919 case PACKET_MR_ALLMULTI:
Wang Chen2aeb0b82008-07-14 20:49:46 -07002920 return dev_set_allmulti(dev, what);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 break;
Eric W. Biedermand95ed922009-05-19 18:27:17 +00002922 case PACKET_MR_UNICAST:
Jiri Pirko11625632010-03-02 20:40:01 +00002923 if (i->alen != dev->addr_len)
2924 return -EINVAL;
Eric W. Biedermand95ed922009-05-19 18:27:17 +00002925 if (what > 0)
Jiri Pirkoa748ee22010-04-01 21:22:09 +00002926 return dev_uc_add(dev, i->addr);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00002927 else
Jiri Pirkoa748ee22010-04-01 21:22:09 +00002928 return dev_uc_del(dev, i->addr);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00002929 break;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002930 default:
2931 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 }
Wang Chen2aeb0b82008-07-14 20:49:46 -07002933 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934}
2935
2936static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2937{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002938 for ( ; i; i = i->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 if (i->ifindex == dev->ifindex)
2940 packet_dev_mc(dev, i, what);
2941 }
2942}
2943
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002944static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002945{
2946 struct packet_sock *po = pkt_sk(sk);
2947 struct packet_mclist *ml, *i;
2948 struct net_device *dev;
2949 int err;
2950
2951 rtnl_lock();
2952
2953 err = -ENODEV;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002954 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 if (!dev)
2956 goto done;
2957
2958 err = -EINVAL;
Jiri Pirko11625632010-03-02 20:40:01 +00002959 if (mreq->mr_alen > dev->addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002960 goto done;
2961
2962 err = -ENOBUFS;
Kris Katterjohn8b3a7002006-01-11 15:56:43 -08002963 i = kmalloc(sizeof(*i), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 if (i == NULL)
2965 goto done;
2966
2967 err = 0;
2968 for (ml = po->mclist; ml; ml = ml->next) {
2969 if (ml->ifindex == mreq->mr_ifindex &&
2970 ml->type == mreq->mr_type &&
2971 ml->alen == mreq->mr_alen &&
2972 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2973 ml->count++;
2974 /* Free the new element ... */
2975 kfree(i);
2976 goto done;
2977 }
2978 }
2979
2980 i->type = mreq->mr_type;
2981 i->ifindex = mreq->mr_ifindex;
2982 i->alen = mreq->mr_alen;
2983 memcpy(i->addr, mreq->mr_address, i->alen);
2984 i->count = 1;
2985 i->next = po->mclist;
2986 po->mclist = i;
Wang Chen2aeb0b82008-07-14 20:49:46 -07002987 err = packet_dev_mc(dev, i, 1);
2988 if (err) {
2989 po->mclist = i->next;
2990 kfree(i);
2991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
2993done:
2994 rtnl_unlock();
2995 return err;
2996}
2997
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002998static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999{
3000 struct packet_mclist *ml, **mlp;
3001
3002 rtnl_lock();
3003
3004 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3005 if (ml->ifindex == mreq->mr_ifindex &&
3006 ml->type == mreq->mr_type &&
3007 ml->alen == mreq->mr_alen &&
3008 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3009 if (--ml->count == 0) {
3010 struct net_device *dev;
3011 *mlp = ml->next;
Eric Dumazetad959e72009-10-16 06:38:46 +00003012 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3013 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 packet_dev_mc(dev, ml, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 kfree(ml);
3016 }
3017 rtnl_unlock();
3018 return 0;
3019 }
3020 }
3021 rtnl_unlock();
3022 return -EADDRNOTAVAIL;
3023}
3024
3025static void packet_flush_mclist(struct sock *sk)
3026{
3027 struct packet_sock *po = pkt_sk(sk);
3028 struct packet_mclist *ml;
3029
3030 if (!po->mclist)
3031 return;
3032
3033 rtnl_lock();
3034 while ((ml = po->mclist) != NULL) {
3035 struct net_device *dev;
3036
3037 po->mclist = ml->next;
Eric Dumazetad959e72009-10-16 06:38:46 +00003038 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3039 if (dev != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 packet_dev_mc(dev, ml, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 kfree(ml);
3042 }
3043 rtnl_unlock();
3044}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045
3046static int
David S. Millerb7058842009-09-30 16:12:20 -07003047packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048{
3049 struct sock *sk = sock->sk;
Herbert Xu8dc41942007-02-04 23:31:32 -08003050 struct packet_sock *po = pkt_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051 int ret;
3052
3053 if (level != SOL_PACKET)
3054 return -ENOPROTOOPT;
3055
Johann Baudy69e3c752009-05-18 22:11:22 -07003056 switch (optname) {
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003057 case PACKET_ADD_MEMBERSHIP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058 case PACKET_DROP_MEMBERSHIP:
3059 {
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003060 struct packet_mreq_max mreq;
3061 int len = optlen;
3062 memset(&mreq, 0, sizeof(mreq));
3063 if (len < sizeof(struct packet_mreq))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 return -EINVAL;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003065 if (len > sizeof(mreq))
3066 len = sizeof(mreq);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003067 if (copy_from_user(&mreq, optval, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 return -EFAULT;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003069 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3070 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 if (optname == PACKET_ADD_MEMBERSHIP)
3072 ret = packet_mc_add(sk, &mreq);
3073 else
3074 ret = packet_mc_drop(sk, &mreq);
3075 return ret;
3076 }
David S. Millera2efcfa2007-05-29 13:12:50 -07003077
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078 case PACKET_RX_RING:
Johann Baudy69e3c752009-05-18 22:11:22 -07003079 case PACKET_TX_RING:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080 {
chetan lokef6fb8f102011-08-19 10:18:16 +00003081 union tpacket_req_u req_u;
3082 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
chetan lokef6fb8f102011-08-19 10:18:16 +00003084 switch (po->tp_version) {
3085 case TPACKET_V1:
3086 case TPACKET_V2:
3087 len = sizeof(req_u.req);
3088 break;
3089 case TPACKET_V3:
3090 default:
3091 len = sizeof(req_u.req3);
3092 break;
3093 }
3094 if (optlen < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 return -EINVAL;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003096 if (pkt_sk(sk)->has_vnet_hdr)
3097 return -EINVAL;
chetan lokef6fb8f102011-08-19 10:18:16 +00003098 if (copy_from_user(&req_u.req, optval, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 return -EFAULT;
chetan lokef6fb8f102011-08-19 10:18:16 +00003100 return packet_set_ring(sk, &req_u, 0,
3101 optname == PACKET_TX_RING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 }
3103 case PACKET_COPY_THRESH:
3104 {
3105 int val;
3106
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003107 if (optlen != sizeof(val))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003108 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003109 if (copy_from_user(&val, optval, sizeof(val)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110 return -EFAULT;
3111
3112 pkt_sk(sk)->copy_thresh = val;
3113 return 0;
3114 }
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003115 case PACKET_VERSION:
3116 {
3117 int val;
3118
3119 if (optlen != sizeof(val))
3120 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07003121 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003122 return -EBUSY;
3123 if (copy_from_user(&val, optval, sizeof(val)))
3124 return -EFAULT;
3125 switch (val) {
3126 case TPACKET_V1:
3127 case TPACKET_V2:
chetan lokef6fb8f102011-08-19 10:18:16 +00003128 case TPACKET_V3:
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003129 po->tp_version = val;
3130 return 0;
3131 default:
3132 return -EINVAL;
3133 }
3134 }
Patrick McHardy8913336a2008-07-18 18:05:19 -07003135 case PACKET_RESERVE:
3136 {
3137 unsigned int val;
3138
3139 if (optlen != sizeof(val))
3140 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07003141 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
Patrick McHardy8913336a2008-07-18 18:05:19 -07003142 return -EBUSY;
3143 if (copy_from_user(&val, optval, sizeof(val)))
3144 return -EFAULT;
3145 po->tp_reserve = val;
3146 return 0;
3147 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003148 case PACKET_LOSS:
3149 {
3150 unsigned int val;
3151
3152 if (optlen != sizeof(val))
3153 return -EINVAL;
3154 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3155 return -EBUSY;
3156 if (copy_from_user(&val, optval, sizeof(val)))
3157 return -EFAULT;
3158 po->tp_loss = !!val;
3159 return 0;
3160 }
Herbert Xu8dc41942007-02-04 23:31:32 -08003161 case PACKET_AUXDATA:
3162 {
3163 int val;
3164
3165 if (optlen < sizeof(val))
3166 return -EINVAL;
3167 if (copy_from_user(&val, optval, sizeof(val)))
3168 return -EFAULT;
3169
3170 po->auxdata = !!val;
3171 return 0;
3172 }
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003173 case PACKET_ORIGDEV:
3174 {
3175 int val;
3176
3177 if (optlen < sizeof(val))
3178 return -EINVAL;
3179 if (copy_from_user(&val, optval, sizeof(val)))
3180 return -EFAULT;
3181
3182 po->origdev = !!val;
3183 return 0;
3184 }
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003185 case PACKET_VNET_HDR:
3186 {
3187 int val;
3188
3189 if (sock->type != SOCK_RAW)
3190 return -EINVAL;
3191 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3192 return -EBUSY;
3193 if (optlen < sizeof(val))
3194 return -EINVAL;
3195 if (copy_from_user(&val, optval, sizeof(val)))
3196 return -EFAULT;
3197
3198 po->has_vnet_hdr = !!val;
3199 return 0;
3200 }
Scott McMillan614f60f2010-06-02 05:53:56 -07003201 case PACKET_TIMESTAMP:
3202 {
3203 int val;
3204
3205 if (optlen != sizeof(val))
3206 return -EINVAL;
3207 if (copy_from_user(&val, optval, sizeof(val)))
3208 return -EFAULT;
3209
3210 po->tp_tstamp = val;
3211 return 0;
3212 }
David S. Millerdc99f602011-07-05 01:45:05 -07003213 case PACKET_FANOUT:
3214 {
3215 int val;
3216
3217 if (optlen != sizeof(val))
3218 return -EINVAL;
3219 if (copy_from_user(&val, optval, sizeof(val)))
3220 return -EFAULT;
3221
3222 return fanout_add(sk, val & 0xffff, val >> 16);
3223 }
Paul Chavent5920cd3a2012-11-06 23:10:47 +00003224 case PACKET_TX_HAS_OFF:
3225 {
3226 unsigned int val;
3227
3228 if (optlen != sizeof(val))
3229 return -EINVAL;
3230 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3231 return -EBUSY;
3232 if (copy_from_user(&val, optval, sizeof(val)))
3233 return -EFAULT;
3234 po->tp_tx_has_off = !!val;
3235 return 0;
3236 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003237 default:
3238 return -ENOPROTOOPT;
3239 }
3240}
3241
3242static int packet_getsockopt(struct socket *sock, int level, int optname,
3243 char __user *optval, int __user *optlen)
3244{
3245 int len;
Eric Dumazetc06fff62012-04-19 21:56:11 +00003246 int val, lv = sizeof(val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003247 struct sock *sk = sock->sk;
3248 struct packet_sock *po = pkt_sk(sk);
Eric Dumazetc06fff62012-04-19 21:56:11 +00003249 void *data = &val;
Herbert Xu8dc41942007-02-04 23:31:32 -08003250 struct tpacket_stats st;
chetan lokef6fb8f102011-08-19 10:18:16 +00003251 union tpacket_stats_u st_u;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003252
3253 if (level != SOL_PACKET)
3254 return -ENOPROTOOPT;
3255
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003256 if (get_user(len, optlen))
3257 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258
3259 if (len < 0)
3260 return -EINVAL;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003261
Johann Baudy69e3c752009-05-18 22:11:22 -07003262 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 case PACKET_STATISTICS:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 spin_lock_bh(&sk->sk_receive_queue.lock);
chetan lokef6fb8f102011-08-19 10:18:16 +00003265 if (po->tp_version == TPACKET_V3) {
Eric Dumazetc06fff62012-04-19 21:56:11 +00003266 lv = sizeof(struct tpacket_stats_v3);
chetan lokef6fb8f102011-08-19 10:18:16 +00003267 memcpy(&st_u.stats3, &po->stats,
Eric Dumazetc06fff62012-04-19 21:56:11 +00003268 sizeof(struct tpacket_stats));
chetan lokef6fb8f102011-08-19 10:18:16 +00003269 st_u.stats3.tp_freeze_q_cnt =
Eric Dumazetc06fff62012-04-19 21:56:11 +00003270 po->stats_u.stats3.tp_freeze_q_cnt;
chetan lokef6fb8f102011-08-19 10:18:16 +00003271 st_u.stats3.tp_packets += po->stats.tp_drops;
3272 data = &st_u.stats3;
3273 } else {
Eric Dumazetc06fff62012-04-19 21:56:11 +00003274 lv = sizeof(struct tpacket_stats);
chetan lokef6fb8f102011-08-19 10:18:16 +00003275 st = po->stats;
3276 st.tp_packets += st.tp_drops;
3277 data = &st;
3278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003279 memset(&po->stats, 0, sizeof(st));
3280 spin_unlock_bh(&sk->sk_receive_queue.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 break;
Herbert Xu8dc41942007-02-04 23:31:32 -08003282 case PACKET_AUXDATA:
Herbert Xu8dc41942007-02-04 23:31:32 -08003283 val = po->auxdata;
Herbert Xu8dc41942007-02-04 23:31:32 -08003284 break;
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003285 case PACKET_ORIGDEV:
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003286 val = po->origdev;
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003287 break;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003288 case PACKET_VNET_HDR:
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003289 val = po->has_vnet_hdr;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003290 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003291 case PACKET_VERSION:
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003292 val = po->tp_version;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003293 break;
3294 case PACKET_HDRLEN:
3295 if (len > sizeof(int))
3296 len = sizeof(int);
3297 if (copy_from_user(&val, optval, len))
3298 return -EFAULT;
3299 switch (val) {
3300 case TPACKET_V1:
3301 val = sizeof(struct tpacket_hdr);
3302 break;
3303 case TPACKET_V2:
3304 val = sizeof(struct tpacket2_hdr);
3305 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00003306 case TPACKET_V3:
3307 val = sizeof(struct tpacket3_hdr);
3308 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003309 default:
3310 return -EINVAL;
3311 }
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003312 break;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003313 case PACKET_RESERVE:
Patrick McHardy8913336a2008-07-18 18:05:19 -07003314 val = po->tp_reserve;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003315 break;
Johann Baudy69e3c752009-05-18 22:11:22 -07003316 case PACKET_LOSS:
Johann Baudy69e3c752009-05-18 22:11:22 -07003317 val = po->tp_loss;
Johann Baudy69e3c752009-05-18 22:11:22 -07003318 break;
Scott McMillan614f60f2010-06-02 05:53:56 -07003319 case PACKET_TIMESTAMP:
Scott McMillan614f60f2010-06-02 05:53:56 -07003320 val = po->tp_tstamp;
Scott McMillan614f60f2010-06-02 05:53:56 -07003321 break;
David S. Millerdc99f602011-07-05 01:45:05 -07003322 case PACKET_FANOUT:
David S. Millerdc99f602011-07-05 01:45:05 -07003323 val = (po->fanout ?
3324 ((u32)po->fanout->id |
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00003325 ((u32)po->fanout->type << 16) |
3326 ((u32)po->fanout->flags << 24)) :
David S. Millerdc99f602011-07-05 01:45:05 -07003327 0);
David S. Millerdc99f602011-07-05 01:45:05 -07003328 break;
Paul Chavent5920cd3a2012-11-06 23:10:47 +00003329 case PACKET_TX_HAS_OFF:
3330 val = po->tp_tx_has_off;
3331 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003332 default:
3333 return -ENOPROTOOPT;
3334 }
3335
Eric Dumazetc06fff62012-04-19 21:56:11 +00003336 if (len > lv)
3337 len = lv;
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003338 if (put_user(len, optlen))
3339 return -EFAULT;
Herbert Xu8dc41942007-02-04 23:31:32 -08003340 if (copy_to_user(optval, data, len))
3341 return -EFAULT;
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003342 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343}
3344
3345
3346static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3347{
3348 struct sock *sk;
Jason Lunzad930652007-02-20 23:19:54 -08003349 struct net_device *dev = data;
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09003350 struct net *net = dev_net(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351
stephen hemminger808f5112010-02-22 07:57:18 +00003352 rcu_read_lock();
Sasha Levinb67bfe02013-02-27 17:06:00 -08003353 sk_for_each_rcu(sk, &net->packet.sklist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003354 struct packet_sock *po = pkt_sk(sk);
3355
3356 switch (msg) {
3357 case NETDEV_UNREGISTER:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358 if (po->mclist)
3359 packet_dev_mclist(dev, po->mclist, -1);
David S. Millera2efcfa2007-05-29 13:12:50 -07003360 /* fallthrough */
3361
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 case NETDEV_DOWN:
3363 if (dev->ifindex == po->ifindex) {
3364 spin_lock(&po->bind_lock);
3365 if (po->running) {
David S. Millerce06b032011-07-04 01:44:29 -07003366 __unregister_prot_hook(sk, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003367 sk->sk_err = ENETDOWN;
3368 if (!sock_flag(sk, SOCK_DEAD))
3369 sk->sk_error_report(sk);
3370 }
3371 if (msg == NETDEV_UNREGISTER) {
3372 po->ifindex = -1;
Ben Greear160ff182011-06-01 07:18:52 +00003373 if (po->prot_hook.dev)
3374 dev_put(po->prot_hook.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 po->prot_hook.dev = NULL;
3376 }
3377 spin_unlock(&po->bind_lock);
3378 }
3379 break;
3380 case NETDEV_UP:
stephen hemminger808f5112010-02-22 07:57:18 +00003381 if (dev->ifindex == po->ifindex) {
3382 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07003383 if (po->num)
3384 register_prot_hook(sk);
stephen hemminger808f5112010-02-22 07:57:18 +00003385 spin_unlock(&po->bind_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003387 break;
3388 }
3389 }
stephen hemminger808f5112010-02-22 07:57:18 +00003390 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 return NOTIFY_DONE;
3392}
3393
3394
3395static int packet_ioctl(struct socket *sock, unsigned int cmd,
3396 unsigned long arg)
3397{
3398 struct sock *sk = sock->sk;
3399
Johann Baudy69e3c752009-05-18 22:11:22 -07003400 switch (cmd) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003401 case SIOCOUTQ:
3402 {
3403 int amount = sk_wmem_alloc_get(sk);
Eric Dumazet31e6d362009-06-17 19:05:41 -07003404
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003405 return put_user(amount, (int __user *)arg);
3406 }
3407 case SIOCINQ:
3408 {
3409 struct sk_buff *skb;
3410 int amount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003411
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003412 spin_lock_bh(&sk->sk_receive_queue.lock);
3413 skb = skb_peek(&sk->sk_receive_queue);
3414 if (skb)
3415 amount = skb->len;
3416 spin_unlock_bh(&sk->sk_receive_queue.lock);
3417 return put_user(amount, (int __user *)arg);
3418 }
3419 case SIOCGSTAMP:
3420 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3421 case SIOCGSTAMPNS:
3422 return sock_get_timestampns(sk, (struct timespec __user *)arg);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003423
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424#ifdef CONFIG_INET
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003425 case SIOCADDRT:
3426 case SIOCDELRT:
3427 case SIOCDARP:
3428 case SIOCGARP:
3429 case SIOCSARP:
3430 case SIOCGIFADDR:
3431 case SIOCSIFADDR:
3432 case SIOCGIFBRDADDR:
3433 case SIOCSIFBRDADDR:
3434 case SIOCGIFNETMASK:
3435 case SIOCSIFNETMASK:
3436 case SIOCGIFDSTADDR:
3437 case SIOCSIFDSTADDR:
3438 case SIOCSIFFLAGS:
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003439 return inet_dgram_ops.ioctl(sock, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003440#endif
3441
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003442 default:
3443 return -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444 }
3445 return 0;
3446}
3447
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003448static unsigned int packet_poll(struct file *file, struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449 poll_table *wait)
3450{
3451 struct sock *sk = sock->sk;
3452 struct packet_sock *po = pkt_sk(sk);
3453 unsigned int mask = datagram_poll(file, sock, wait);
3454
3455 spin_lock_bh(&sk->sk_receive_queue.lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003456 if (po->rx_ring.pg_vec) {
chetan lokef6fb8f102011-08-19 10:18:16 +00003457 if (!packet_previous_rx_frame(po, &po->rx_ring,
3458 TP_STATUS_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003459 mask |= POLLIN | POLLRDNORM;
3460 }
3461 spin_unlock_bh(&sk->sk_receive_queue.lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003462 spin_lock_bh(&sk->sk_write_queue.lock);
3463 if (po->tx_ring.pg_vec) {
3464 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3465 mask |= POLLOUT | POLLWRNORM;
3466 }
3467 spin_unlock_bh(&sk->sk_write_queue.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003468 return mask;
3469}
3470
3471
3472/* Dirty? Well, I still did not learn better way to account
3473 * for user mmaps.
3474 */
3475
3476static void packet_mm_open(struct vm_area_struct *vma)
3477{
3478 struct file *file = vma->vm_file;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003479 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003480 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003481
Linus Torvalds1da177e2005-04-16 15:20:36 -07003482 if (sk)
3483 atomic_inc(&pkt_sk(sk)->mapped);
3484}
3485
3486static void packet_mm_close(struct vm_area_struct *vma)
3487{
3488 struct file *file = vma->vm_file;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003489 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003490 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003491
Linus Torvalds1da177e2005-04-16 15:20:36 -07003492 if (sk)
3493 atomic_dec(&pkt_sk(sk)->mapped);
3494}
3495
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003496static const struct vm_operations_struct packet_mmap_ops = {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003497 .open = packet_mm_open,
3498 .close = packet_mm_close,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499};
3500
Neil Horman0e3125c2010-11-16 10:26:47 -08003501static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3502 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503{
3504 int i;
3505
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003506 for (i = 0; i < len; i++) {
Neil Horman0e3125c2010-11-16 10:26:47 -08003507 if (likely(pg_vec[i].buffer)) {
Changli Gaoc56b4d92010-12-01 02:52:57 +00003508 if (is_vmalloc_addr(pg_vec[i].buffer))
Neil Horman0e3125c2010-11-16 10:26:47 -08003509 vfree(pg_vec[i].buffer);
3510 else
3511 free_pages((unsigned long)pg_vec[i].buffer,
3512 order);
3513 pg_vec[i].buffer = NULL;
3514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003515 }
3516 kfree(pg_vec);
3517}
3518
Olof Johanssoneea49cc92011-11-02 11:00:49 +00003519static char *alloc_one_pg_vec_page(unsigned long order)
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003520{
Neil Horman0e3125c2010-11-16 10:26:47 -08003521 char *buffer = NULL;
3522 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3523 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
Eric Dumazet719bfea2009-04-15 03:39:52 -07003524
Neil Horman0e3125c2010-11-16 10:26:47 -08003525 buffer = (char *) __get_free_pages(gfp_flags, order);
3526
3527 if (buffer)
3528 return buffer;
3529
3530 /*
3531 * __get_free_pages failed, fall back to vmalloc
3532 */
Eric Dumazetbbce5a52010-11-20 07:31:54 +00003533 buffer = vzalloc((1 << order) * PAGE_SIZE);
Neil Horman0e3125c2010-11-16 10:26:47 -08003534
3535 if (buffer)
3536 return buffer;
3537
3538 /*
3539 * vmalloc failed, lets dig into swap here
3540 */
Neil Horman0e3125c2010-11-16 10:26:47 -08003541 gfp_flags &= ~__GFP_NORETRY;
3542 buffer = (char *)__get_free_pages(gfp_flags, order);
3543 if (buffer)
3544 return buffer;
3545
3546 /*
3547 * complete and utter failure
3548 */
3549 return NULL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003550}
3551
Neil Horman0e3125c2010-11-16 10:26:47 -08003552static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003553{
3554 unsigned int block_nr = req->tp_block_nr;
Neil Horman0e3125c2010-11-16 10:26:47 -08003555 struct pgv *pg_vec;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003556 int i;
3557
Neil Horman0e3125c2010-11-16 10:26:47 -08003558 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003559 if (unlikely(!pg_vec))
3560 goto out;
3561
3562 for (i = 0; i < block_nr; i++) {
Changli Gaoc56b4d92010-12-01 02:52:57 +00003563 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
Neil Horman0e3125c2010-11-16 10:26:47 -08003564 if (unlikely(!pg_vec[i].buffer))
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003565 goto out_free_pgvec;
3566 }
3567
3568out:
3569 return pg_vec;
3570
3571out_free_pgvec:
3572 free_pg_vec(pg_vec, order, block_nr);
3573 pg_vec = NULL;
3574 goto out;
3575}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003576
chetan lokef6fb8f102011-08-19 10:18:16 +00003577static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
Johann Baudy69e3c752009-05-18 22:11:22 -07003578 int closing, int tx_ring)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003579{
Neil Horman0e3125c2010-11-16 10:26:47 -08003580 struct pgv *pg_vec = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581 struct packet_sock *po = pkt_sk(sk);
Al Viro0e11c912006-11-08 00:26:29 -08003582 int was_running, order = 0;
Johann Baudy69e3c752009-05-18 22:11:22 -07003583 struct packet_ring_buffer *rb;
3584 struct sk_buff_head *rb_queue;
Al Viro0e11c912006-11-08 00:26:29 -08003585 __be16 num;
chetan lokef6fb8f102011-08-19 10:18:16 +00003586 int err = -EINVAL;
3587 /* Added to avoid minimal code churn */
3588 struct tpacket_req *req = &req_u->req;
3589
3590 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3591 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3592 WARN(1, "Tx-ring is not supported.\n");
3593 goto out;
3594 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003595
3596 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3597 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3598
3599 err = -EBUSY;
3600 if (!closing) {
3601 if (atomic_read(&po->mapped))
3602 goto out;
3603 if (atomic_read(&rb->pending))
3604 goto out;
3605 }
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003606
Linus Torvalds1da177e2005-04-16 15:20:36 -07003607 if (req->tp_block_nr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003608 /* Sanity tests and some calculations */
Johann Baudy69e3c752009-05-18 22:11:22 -07003609 err = -EBUSY;
3610 if (unlikely(rb->pg_vec))
3611 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003612
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003613 switch (po->tp_version) {
3614 case TPACKET_V1:
3615 po->tp_hdrlen = TPACKET_HDRLEN;
3616 break;
3617 case TPACKET_V2:
3618 po->tp_hdrlen = TPACKET2_HDRLEN;
3619 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00003620 case TPACKET_V3:
3621 po->tp_hdrlen = TPACKET3_HDRLEN;
3622 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003623 }
3624
Johann Baudy69e3c752009-05-18 22:11:22 -07003625 err = -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003626 if (unlikely((int)req->tp_block_size <= 0))
Johann Baudy69e3c752009-05-18 22:11:22 -07003627 goto out;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003628 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
Johann Baudy69e3c752009-05-18 22:11:22 -07003629 goto out;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003630 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
Johann Baudy69e3c752009-05-18 22:11:22 -07003631 po->tp_reserve))
3632 goto out;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003633 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
Johann Baudy69e3c752009-05-18 22:11:22 -07003634 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003635
Johann Baudy69e3c752009-05-18 22:11:22 -07003636 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3637 if (unlikely(rb->frames_per_block <= 0))
3638 goto out;
3639 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3640 req->tp_frame_nr))
3641 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003642
3643 err = -ENOMEM;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003644 order = get_order(req->tp_block_size);
3645 pg_vec = alloc_pg_vec(req, order);
3646 if (unlikely(!pg_vec))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003647 goto out;
chetan lokef6fb8f102011-08-19 10:18:16 +00003648 switch (po->tp_version) {
3649 case TPACKET_V3:
3650 /* Transmit path is not supported. We checked
3651 * it above but just being paranoid
3652 */
3653 if (!tx_ring)
3654 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3655 break;
3656 default:
3657 break;
3658 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003659 }
3660 /* Done */
3661 else {
3662 err = -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003663 if (unlikely(req->tp_frame_nr))
Johann Baudy69e3c752009-05-18 22:11:22 -07003664 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003665 }
3666
3667 lock_sock(sk);
3668
3669 /* Detach socket from network */
3670 spin_lock(&po->bind_lock);
3671 was_running = po->running;
3672 num = po->num;
3673 if (was_running) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003674 po->num = 0;
David S. Millerce06b032011-07-04 01:44:29 -07003675 __unregister_prot_hook(sk, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003676 }
3677 spin_unlock(&po->bind_lock);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003678
Linus Torvalds1da177e2005-04-16 15:20:36 -07003679 synchronize_net();
3680
3681 err = -EBUSY;
Herbert Xu905db442009-01-30 14:12:06 -08003682 mutex_lock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003683 if (closing || atomic_read(&po->mapped) == 0) {
3684 err = 0;
Johann Baudy69e3c752009-05-18 22:11:22 -07003685 spin_lock_bh(&rb_queue->lock);
Changli Gaoc053fd92010-12-10 16:02:20 -08003686 swap(rb->pg_vec, pg_vec);
Johann Baudy69e3c752009-05-18 22:11:22 -07003687 rb->frame_max = (req->tp_frame_nr - 1);
3688 rb->head = 0;
3689 rb->frame_size = req->tp_frame_size;
3690 spin_unlock_bh(&rb_queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003691
Changli Gaoc053fd92010-12-10 16:02:20 -08003692 swap(rb->pg_vec_order, order);
3693 swap(rb->pg_vec_len, req->tp_block_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694
Johann Baudy69e3c752009-05-18 22:11:22 -07003695 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3696 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3697 tpacket_rcv : packet_rcv;
3698 skb_queue_purge(rb_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003699 if (atomic_read(&po->mapped))
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003700 pr_err("packet_mmap: vma is busy: %d\n",
3701 atomic_read(&po->mapped));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702 }
Herbert Xu905db442009-01-30 14:12:06 -08003703 mutex_unlock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704
3705 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07003706 if (was_running) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003707 po->num = num;
David S. Millerce06b032011-07-04 01:44:29 -07003708 register_prot_hook(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003709 }
3710 spin_unlock(&po->bind_lock);
chetan lokef6fb8f102011-08-19 10:18:16 +00003711 if (closing && (po->tp_version > TPACKET_V2)) {
3712 /* Because we don't support block-based V3 on tx-ring */
3713 if (!tx_ring)
3714 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3715 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003716 release_sock(sk);
3717
Linus Torvalds1da177e2005-04-16 15:20:36 -07003718 if (pg_vec)
3719 free_pg_vec(pg_vec, order, req->tp_block_nr);
3720out:
3721 return err;
3722}
3723
Johann Baudy69e3c752009-05-18 22:11:22 -07003724static int packet_mmap(struct file *file, struct socket *sock,
3725 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003726{
3727 struct sock *sk = sock->sk;
3728 struct packet_sock *po = pkt_sk(sk);
Johann Baudy69e3c752009-05-18 22:11:22 -07003729 unsigned long size, expected_size;
3730 struct packet_ring_buffer *rb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003731 unsigned long start;
3732 int err = -EINVAL;
3733 int i;
3734
3735 if (vma->vm_pgoff)
3736 return -EINVAL;
3737
Herbert Xu905db442009-01-30 14:12:06 -08003738 mutex_lock(&po->pg_vec_lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003739
3740 expected_size = 0;
3741 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3742 if (rb->pg_vec) {
3743 expected_size += rb->pg_vec_len
3744 * rb->pg_vec_pages
3745 * PAGE_SIZE;
3746 }
3747 }
3748
3749 if (expected_size == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003750 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07003751
3752 size = vma->vm_end - vma->vm_start;
3753 if (size != expected_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003754 goto out;
3755
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 start = vma->vm_start;
Johann Baudy69e3c752009-05-18 22:11:22 -07003757 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3758 if (rb->pg_vec == NULL)
3759 continue;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003760
Johann Baudy69e3c752009-05-18 22:11:22 -07003761 for (i = 0; i < rb->pg_vec_len; i++) {
Neil Horman0e3125c2010-11-16 10:26:47 -08003762 struct page *page;
3763 void *kaddr = rb->pg_vec[i].buffer;
Johann Baudy69e3c752009-05-18 22:11:22 -07003764 int pg_num;
3765
Changli Gaoc56b4d92010-12-01 02:52:57 +00003766 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3767 page = pgv_to_page(kaddr);
Johann Baudy69e3c752009-05-18 22:11:22 -07003768 err = vm_insert_page(vma, start, page);
3769 if (unlikely(err))
3770 goto out;
3771 start += PAGE_SIZE;
Neil Horman0e3125c2010-11-16 10:26:47 -08003772 kaddr += PAGE_SIZE;
Johann Baudy69e3c752009-05-18 22:11:22 -07003773 }
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003774 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003776
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003777 atomic_inc(&po->mapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778 vma->vm_ops = &packet_mmap_ops;
3779 err = 0;
3780
3781out:
Herbert Xu905db442009-01-30 14:12:06 -08003782 mutex_unlock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783 return err;
3784}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08003786static const struct proto_ops packet_ops_spkt = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003787 .family = PF_PACKET,
3788 .owner = THIS_MODULE,
3789 .release = packet_release,
3790 .bind = packet_bind_spkt,
3791 .connect = sock_no_connect,
3792 .socketpair = sock_no_socketpair,
3793 .accept = sock_no_accept,
3794 .getname = packet_getname_spkt,
3795 .poll = datagram_poll,
3796 .ioctl = packet_ioctl,
3797 .listen = sock_no_listen,
3798 .shutdown = sock_no_shutdown,
3799 .setsockopt = sock_no_setsockopt,
3800 .getsockopt = sock_no_getsockopt,
3801 .sendmsg = packet_sendmsg_spkt,
3802 .recvmsg = packet_recvmsg,
3803 .mmap = sock_no_mmap,
3804 .sendpage = sock_no_sendpage,
3805};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08003807static const struct proto_ops packet_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003808 .family = PF_PACKET,
3809 .owner = THIS_MODULE,
3810 .release = packet_release,
3811 .bind = packet_bind,
3812 .connect = sock_no_connect,
3813 .socketpair = sock_no_socketpair,
3814 .accept = sock_no_accept,
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003815 .getname = packet_getname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816 .poll = packet_poll,
3817 .ioctl = packet_ioctl,
3818 .listen = sock_no_listen,
3819 .shutdown = sock_no_shutdown,
3820 .setsockopt = packet_setsockopt,
3821 .getsockopt = packet_getsockopt,
3822 .sendmsg = packet_sendmsg,
3823 .recvmsg = packet_recvmsg,
3824 .mmap = packet_mmap,
3825 .sendpage = sock_no_sendpage,
3826};
3827
Stephen Hemmingerec1b4cf2009-10-05 05:58:39 +00003828static const struct net_proto_family packet_family_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003829 .family = PF_PACKET,
3830 .create = packet_create,
3831 .owner = THIS_MODULE,
3832};
3833
3834static struct notifier_block packet_netdev_notifier = {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003835 .notifier_call = packet_notifier,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836};
3837
3838#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003839
3840static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
stephen hemminger808f5112010-02-22 07:57:18 +00003841 __acquires(RCU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003842{
Denis V. Luneve372c412007-11-19 22:31:54 -08003843 struct net *net = seq_file_net(seq);
stephen hemminger808f5112010-02-22 07:57:18 +00003844
3845 rcu_read_lock();
3846 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003847}
3848
3849static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3850{
Herbert Xu1bf40952007-12-16 14:04:02 -08003851 struct net *net = seq_file_net(seq);
stephen hemminger808f5112010-02-22 07:57:18 +00003852 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853}
3854
3855static void packet_seq_stop(struct seq_file *seq, void *v)
stephen hemminger808f5112010-02-22 07:57:18 +00003856 __releases(RCU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003857{
stephen hemminger808f5112010-02-22 07:57:18 +00003858 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003859}
3860
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003861static int packet_seq_show(struct seq_file *seq, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003862{
3863 if (v == SEQ_START_TOKEN)
3864 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3865 else {
Li Zefanb7ceabd2010-02-08 23:19:29 +00003866 struct sock *s = sk_entry(v);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003867 const struct packet_sock *po = pkt_sk(s);
3868
3869 seq_printf(seq,
Dan Rosenberg71338aa2011-05-23 12:17:35 +00003870 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871 s,
3872 atomic_read(&s->sk_refcnt),
3873 s->sk_type,
3874 ntohs(po->num),
3875 po->ifindex,
3876 po->running,
3877 atomic_read(&s->sk_rmem_alloc),
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06003878 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003879 sock_i_ino(s));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 }
3881
3882 return 0;
3883}
3884
Philippe De Muyter56b3d972007-07-10 23:07:31 -07003885static const struct seq_operations packet_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003886 .start = packet_seq_start,
3887 .next = packet_seq_next,
3888 .stop = packet_seq_stop,
3889 .show = packet_seq_show,
3890};
3891
3892static int packet_seq_open(struct inode *inode, struct file *file)
3893{
Denis V. Luneve372c412007-11-19 22:31:54 -08003894 return seq_open_net(inode, file, &packet_seq_ops,
3895 sizeof(struct seq_net_private));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896}
3897
Arjan van de Venda7071d2007-02-12 00:55:36 -08003898static const struct file_operations packet_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899 .owner = THIS_MODULE,
3900 .open = packet_seq_open,
3901 .read = seq_read,
3902 .llseek = seq_lseek,
Denis V. Luneve372c412007-11-19 22:31:54 -08003903 .release = seq_release_net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904};
3905
3906#endif
3907
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00003908static int __net_init packet_net_init(struct net *net)
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08003909{
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00003910 mutex_init(&net->packet.sklist_lock);
Denis V. Lunev2aaef4e2007-12-11 04:19:54 -08003911 INIT_HLIST_HEAD(&net->packet.sklist);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08003912
Gao fengd4beaa62013-02-18 01:34:54 +00003913 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08003914 return -ENOMEM;
3915
3916 return 0;
3917}
3918
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00003919static void __net_exit packet_net_exit(struct net *net)
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08003920{
Gao fengece31ff2013-02-18 01:34:56 +00003921 remove_proc_entry("packet", net->proc_net);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08003922}
3923
3924static struct pernet_operations packet_net_ops = {
3925 .init = packet_net_init,
3926 .exit = packet_net_exit,
3927};
3928
3929
Linus Torvalds1da177e2005-04-16 15:20:36 -07003930static void __exit packet_exit(void)
3931{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003932 unregister_netdevice_notifier(&packet_netdev_notifier);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08003933 unregister_pernet_subsys(&packet_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934 sock_unregister(PF_PACKET);
3935 proto_unregister(&packet_proto);
3936}
3937
3938static int __init packet_init(void)
3939{
3940 int rc = proto_register(&packet_proto, 0);
3941
3942 if (rc != 0)
3943 goto out;
3944
3945 sock_register(&packet_family_ops);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08003946 register_pernet_subsys(&packet_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947 register_netdevice_notifier(&packet_netdev_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948out:
3949 return rc;
3950}
3951
3952module_init(packet_init);
3953module_exit(packet_exit);
3954MODULE_LICENSE("GPL");
3955MODULE_ALIAS_NETPROTO(PF_PACKET);