blob: b85c67ccb797197abf51596ac5f3044131aa97d8 [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>
Daniel Borkmannb0138402014-01-15 16:25:36 +010091#include <linux/percpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070092#ifdef CONFIG_INET
93#include <net/inet_common.h>
94#endif
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
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100240static int packet_direct_xmit(struct sk_buff *skb)
241{
242 struct net_device *dev = skb->dev;
243 const struct net_device_ops *ops = dev->netdev_ops;
244 netdev_features_t features;
245 struct netdev_queue *txq;
Daniel Borkmann43279502014-03-27 16:38:30 +0100246 int ret = NETDEV_TX_BUSY;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100247 u16 queue_map;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100248
249 if (unlikely(!netif_running(dev) ||
Daniel Borkmann43279502014-03-27 16:38:30 +0100250 !netif_carrier_ok(dev)))
251 goto drop;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100252
253 features = netif_skb_features(skb);
254 if (skb_needs_linearize(skb, features) &&
Daniel Borkmann43279502014-03-27 16:38:30 +0100255 __skb_linearize(skb))
256 goto drop;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100257
258 queue_map = skb_get_queue_mapping(skb);
259 txq = netdev_get_tx_queue(dev, queue_map);
260
Daniel Borkmann43279502014-03-27 16:38:30 +0100261 local_bh_disable();
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100262
Daniel Borkmann43279502014-03-27 16:38:30 +0100263 HARD_TX_LOCK(dev, txq, smp_processor_id());
Daniel Borkmann8e2f1a62014-04-02 20:52:57 +0200264 if (!netif_xmit_frozen_or_drv_stopped(txq)) {
Daniel Borkmann43279502014-03-27 16:38:30 +0100265 ret = ops->ndo_start_xmit(skb, dev);
266 if (ret == NETDEV_TX_OK)
267 txq_trans_update(txq);
268 }
269 HARD_TX_UNLOCK(dev, txq);
270
271 local_bh_enable();
272
273 if (!dev_xmit_complete(ret))
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100274 kfree_skb(skb);
Daniel Borkmann43279502014-03-27 16:38:30 +0100275
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100276 return ret;
Daniel Borkmann43279502014-03-27 16:38:30 +0100277drop:
Daniel Borkmann0f97ede2014-04-02 20:52:56 +0200278 atomic_long_inc(&dev->tx_dropped);
Daniel Borkmann43279502014-03-27 16:38:30 +0100279 kfree_skb(skb);
280 return NET_XMIT_DROP;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100281}
282
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100283static struct net_device *packet_cached_dev_get(struct packet_sock *po)
284{
285 struct net_device *dev;
286
287 rcu_read_lock();
288 dev = rcu_dereference(po->cached_dev);
289 if (likely(dev))
290 dev_hold(dev);
291 rcu_read_unlock();
292
293 return dev;
294}
295
296static void packet_cached_dev_assign(struct packet_sock *po,
297 struct net_device *dev)
298{
299 rcu_assign_pointer(po->cached_dev, dev);
300}
301
302static void packet_cached_dev_reset(struct packet_sock *po)
303{
304 RCU_INIT_POINTER(po->cached_dev, NULL);
305}
306
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100307static bool packet_use_direct_xmit(const struct packet_sock *po)
308{
309 return po->xmit == packet_direct_xmit;
310}
311
Daniel Borkmann0fd5d572014-02-16 15:55:22 +0100312static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100313{
Li Zhong1cbac012013-12-12 22:39:55 +0100314 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100315}
316
Daniel Borkmann0fd5d572014-02-16 15:55:22 +0100317static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
318{
319 const struct net_device_ops *ops = dev->netdev_ops;
320 u16 queue_index;
321
322 if (ops->ndo_select_queue) {
323 queue_index = ops->ndo_select_queue(dev, skb, NULL,
324 __packet_pick_tx_queue);
325 queue_index = netdev_cap_txqueue(dev, queue_index);
326 } else {
327 queue_index = __packet_pick_tx_queue(dev, skb);
328 }
329
330 skb_set_queue_mapping(skb, queue_index);
331}
332
David S. Millerce06b032011-07-04 01:44:29 -0700333/* register_prot_hook must be invoked with the po->bind_lock held,
334 * or from a context in which asynchronous accesses to the packet
335 * socket is not possible (packet_create()).
336 */
337static void register_prot_hook(struct sock *sk)
338{
339 struct packet_sock *po = pkt_sk(sk);
Daniel Borkmanne40526c2013-11-21 16:50:58 +0100340
David S. Millerce06b032011-07-04 01:44:29 -0700341 if (!po->running) {
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100342 if (po->fanout)
David S. Millerdc99f602011-07-05 01:45:05 -0700343 __fanout_link(sk, po);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100344 else
David S. Millerdc99f602011-07-05 01:45:05 -0700345 dev_add_pack(&po->prot_hook);
Daniel Borkmanne40526c2013-11-21 16:50:58 +0100346
David S. Millerce06b032011-07-04 01:44:29 -0700347 sock_hold(sk);
348 po->running = 1;
349 }
350}
351
352/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
353 * held. If the sync parameter is true, we will temporarily drop
354 * the po->bind_lock and do a synchronize_net to make sure no
355 * asynchronous packet processing paths still refer to the elements
356 * of po->prot_hook. If the sync parameter is false, it is the
357 * callers responsibility to take care of this.
358 */
359static void __unregister_prot_hook(struct sock *sk, bool sync)
360{
361 struct packet_sock *po = pkt_sk(sk);
362
363 po->running = 0;
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100364
365 if (po->fanout)
David S. Millerdc99f602011-07-05 01:45:05 -0700366 __fanout_unlink(sk, po);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +0100367 else
David S. Millerdc99f602011-07-05 01:45:05 -0700368 __dev_remove_pack(&po->prot_hook);
Daniel Borkmanne40526c2013-11-21 16:50:58 +0100369
David S. Millerce06b032011-07-04 01:44:29 -0700370 __sock_put(sk);
371
372 if (sync) {
373 spin_unlock(&po->bind_lock);
374 synchronize_net();
375 spin_lock(&po->bind_lock);
376 }
377}
378
379static void unregister_prot_hook(struct sock *sk, bool sync)
380{
381 struct packet_sock *po = pkt_sk(sk);
382
383 if (po->running)
384 __unregister_prot_hook(sk, sync);
385}
386
Changli Gaof6dafa92010-12-07 04:26:16 +0000387static inline __pure struct page *pgv_to_page(void *addr)
Changli Gao0af55bb2010-12-01 02:52:20 +0000388{
389 if (is_vmalloc_addr(addr))
390 return vmalloc_to_page(addr);
391 return virt_to_page(addr);
392}
393
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700394static void __packet_set_status(struct packet_sock *po, void *frame, int status)
395{
Daniel Borkmann184f4892013-04-16 01:57:46 +0000396 union tpacket_uhdr h;
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700397
398 h.raw = frame;
399 switch (po->tp_version) {
400 case TPACKET_V1:
401 h.h1->tp_status = status;
Changli Gao0af55bb2010-12-01 02:52:20 +0000402 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700403 break;
404 case TPACKET_V2:
405 h.h2->tp_status = status;
Changli Gao0af55bb2010-12-01 02:52:20 +0000406 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700407 break;
chetan lokef6fb8f102011-08-19 10:18:16 +0000408 case TPACKET_V3:
Johann Baudy69e3c752009-05-18 22:11:22 -0700409 default:
chetan lokef6fb8f102011-08-19 10:18:16 +0000410 WARN(1, "TPACKET version not supported.\n");
Johann Baudy69e3c752009-05-18 22:11:22 -0700411 BUG();
412 }
413
414 smp_wmb();
415}
416
417static int __packet_get_status(struct packet_sock *po, void *frame)
418{
Daniel Borkmann184f4892013-04-16 01:57:46 +0000419 union tpacket_uhdr h;
Johann Baudy69e3c752009-05-18 22:11:22 -0700420
421 smp_rmb();
422
423 h.raw = frame;
424 switch (po->tp_version) {
425 case TPACKET_V1:
Changli Gao0af55bb2010-12-01 02:52:20 +0000426 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
Johann Baudy69e3c752009-05-18 22:11:22 -0700427 return h.h1->tp_status;
428 case TPACKET_V2:
Changli Gao0af55bb2010-12-01 02:52:20 +0000429 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
Johann Baudy69e3c752009-05-18 22:11:22 -0700430 return h.h2->tp_status;
chetan lokef6fb8f102011-08-19 10:18:16 +0000431 case TPACKET_V3:
Johann Baudy69e3c752009-05-18 22:11:22 -0700432 default:
chetan lokef6fb8f102011-08-19 10:18:16 +0000433 WARN(1, "TPACKET version not supported.\n");
Johann Baudy69e3c752009-05-18 22:11:22 -0700434 BUG();
435 return 0;
Patrick McHardybbd6ef82008-07-14 22:50:15 -0700436 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
Johann Baudy69e3c752009-05-18 22:11:22 -0700438
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000439static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
440 unsigned int flags)
Daniel Borkmann7a513842013-04-23 00:39:29 +0000441{
442 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
443
444 if (shhwtstamps) {
445 if ((flags & SOF_TIMESTAMPING_SYS_HARDWARE) &&
446 ktime_to_timespec_cond(shhwtstamps->syststamp, ts))
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000447 return TP_STATUS_TS_SYS_HARDWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000448 if ((flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000450 return TP_STATUS_TS_RAW_HARDWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000451 }
452
453 if (ktime_to_timespec_cond(skb->tstamp, ts))
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000454 return TP_STATUS_TS_SOFTWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000455
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000456 return 0;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000457}
458
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000459static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
460 struct sk_buff *skb)
Willem de Bruijn2e313962013-04-23 00:39:28 +0000461{
462 union tpacket_uhdr h;
463 struct timespec ts;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000464 __u32 ts_status;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000465
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000466 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
467 return 0;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000468
469 h.raw = frame;
470 switch (po->tp_version) {
471 case TPACKET_V1:
472 h.h1->tp_sec = ts.tv_sec;
473 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
474 break;
475 case TPACKET_V2:
476 h.h2->tp_sec = ts.tv_sec;
477 h.h2->tp_nsec = ts.tv_nsec;
478 break;
479 case TPACKET_V3:
480 default:
481 WARN(1, "TPACKET version not supported.\n");
482 BUG();
483 }
484
485 /* one flush is safe, as both fields always lie on the same cacheline */
486 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
487 smp_wmb();
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000488
489 return ts_status;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000490}
491
Johann Baudy69e3c752009-05-18 22:11:22 -0700492static void *packet_lookup_frame(struct packet_sock *po,
493 struct packet_ring_buffer *rb,
494 unsigned int position,
495 int status)
496{
497 unsigned int pg_vec_pos, frame_offset;
Daniel Borkmann184f4892013-04-16 01:57:46 +0000498 union tpacket_uhdr h;
Johann Baudy69e3c752009-05-18 22:11:22 -0700499
500 pg_vec_pos = position / rb->frames_per_block;
501 frame_offset = position % rb->frames_per_block;
502
Neil Horman0e3125c2010-11-16 10:26:47 -0800503 h.raw = rb->pg_vec[pg_vec_pos].buffer +
504 (frame_offset * rb->frame_size);
Johann Baudy69e3c752009-05-18 22:11:22 -0700505
506 if (status != __packet_get_status(po, h.raw))
507 return NULL;
508
509 return h.raw;
510}
511
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000512static void *packet_current_frame(struct packet_sock *po,
Johann Baudy69e3c752009-05-18 22:11:22 -0700513 struct packet_ring_buffer *rb,
514 int status)
515{
516 return packet_lookup_frame(po, rb, rb->head, status);
517}
518
chetan lokebc59ba32011-08-25 10:43:30 +0000519static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000520{
521 del_timer_sync(&pkc->retire_blk_timer);
522}
523
524static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
525 int tx_ring,
526 struct sk_buff_head *rb_queue)
527{
chetan lokebc59ba32011-08-25 10:43:30 +0000528 struct tpacket_kbdq_core *pkc;
chetan lokef6fb8f102011-08-19 10:18:16 +0000529
Duan Jiong22781a52013-12-06 13:29:36 +0800530 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
531 GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f102011-08-19 10:18:16 +0000532
Veaceslav Falicoec6f8092013-11-29 09:53:23 +0100533 spin_lock_bh(&rb_queue->lock);
chetan lokef6fb8f102011-08-19 10:18:16 +0000534 pkc->delete_blk_timer = 1;
Veaceslav Falicoec6f8092013-11-29 09:53:23 +0100535 spin_unlock_bh(&rb_queue->lock);
chetan lokef6fb8f102011-08-19 10:18:16 +0000536
537 prb_del_retire_blk_timer(pkc);
538}
539
540static void prb_init_blk_timer(struct packet_sock *po,
chetan lokebc59ba32011-08-25 10:43:30 +0000541 struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000542 void (*func) (unsigned long))
543{
544 init_timer(&pkc->retire_blk_timer);
545 pkc->retire_blk_timer.data = (long)po;
546 pkc->retire_blk_timer.function = func;
547 pkc->retire_blk_timer.expires = jiffies;
548}
549
550static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
551{
chetan lokebc59ba32011-08-25 10:43:30 +0000552 struct tpacket_kbdq_core *pkc;
chetan lokef6fb8f102011-08-19 10:18:16 +0000553
554 if (tx_ring)
555 BUG();
556
Duan Jiong22781a52013-12-06 13:29:36 +0800557 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
558 GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f102011-08-19 10:18:16 +0000559 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
560}
561
562static int prb_calc_retire_blk_tmo(struct packet_sock *po,
563 int blk_size_in_bytes)
564{
565 struct net_device *dev;
566 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000567 struct ethtool_cmd ecmd;
568 int err;
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000569 u32 speed;
chetan lokef6fb8f102011-08-19 10:18:16 +0000570
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000571 rtnl_lock();
572 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
573 if (unlikely(!dev)) {
574 rtnl_unlock();
chetan lokef6fb8f102011-08-19 10:18:16 +0000575 return DEFAULT_PRB_RETIRE_TOV;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000576 }
577 err = __ethtool_get_settings(dev, &ecmd);
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000578 speed = ethtool_cmd_speed(&ecmd);
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000579 rtnl_unlock();
580 if (!err) {
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000581 /*
582 * If the link speed is so slow you don't really
583 * need to worry about perf anyways
584 */
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000585 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000586 return DEFAULT_PRB_RETIRE_TOV;
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000587 } else {
588 msec = 1;
589 div = speed / 1000;
chetan lokef6fb8f102011-08-19 10:18:16 +0000590 }
591 }
592
593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
594
595 if (div)
596 mbits /= div;
597
598 tmo = mbits * msec;
599
600 if (div)
601 return tmo+1;
602 return tmo;
603}
604
chetan lokebc59ba32011-08-25 10:43:30 +0000605static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
chetan lokef6fb8f102011-08-19 10:18:16 +0000606 union tpacket_req_u *req_u)
607{
608 p1->feature_req_word = req_u->req3.tp_feature_req_word;
609}
610
611static void init_prb_bdqc(struct packet_sock *po,
612 struct packet_ring_buffer *rb,
613 struct pgv *pg_vec,
614 union tpacket_req_u *req_u, int tx_ring)
615{
Duan Jiong22781a52013-12-06 13:29:36 +0800616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
chetan lokebc59ba32011-08-25 10:43:30 +0000617 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000618
619 memset(p1, 0x0, sizeof(*p1));
620
621 p1->knxt_seq_num = 1;
622 p1->pkbdq = pg_vec;
chetan lokebc59ba32011-08-25 10:43:30 +0000623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
Joe Perchese3192692012-06-03 17:41:40 +0000624 p1->pkblk_start = pg_vec[0].buffer;
chetan lokef6fb8f102011-08-19 10:18:16 +0000625 p1->kblk_size = req_u->req3.tp_block_size;
626 p1->knum_blocks = req_u->req3.tp_block_nr;
627 p1->hdrlen = po->tp_hdrlen;
628 p1->version = po->tp_version;
629 p1->last_kactive_blk_num = 0;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +0000630 po->stats.stats3.tp_freeze_q_cnt = 0;
chetan lokef6fb8f102011-08-19 10:18:16 +0000631 if (req_u->req3.tp_retire_blk_tov)
632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
633 else
634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 req_u->req3.tp_block_size);
636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
638
639 prb_init_ft_ops(p1, req_u);
640 prb_setup_retire_blk_timer(po, tx_ring);
641 prb_open_block(p1, pbd);
642}
643
644/* Do NOT update the last_blk_num first.
645 * Assumes sk_buff_head lock is held.
646 */
chetan lokebc59ba32011-08-25 10:43:30 +0000647static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000648{
649 mod_timer(&pkc->retire_blk_timer,
650 jiffies + pkc->tov_in_jiffies);
651 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
652}
653
654/*
655 * Timer logic:
656 * 1) We refresh the timer only when we open a block.
657 * By doing this we don't waste cycles refreshing the timer
658 * on packet-by-packet basis.
659 *
660 * With a 1MB block-size, on a 1Gbps line, it will take
661 * i) ~8 ms to fill a block + ii) memcpy etc.
662 * In this cut we are not accounting for the memcpy time.
663 *
664 * So, if the user sets the 'tmo' to 10ms then the timer
665 * will never fire while the block is still getting filled
666 * (which is what we want). However, the user could choose
667 * to close a block early and that's fine.
668 *
669 * But when the timer does fire, we check whether or not to refresh it.
670 * Since the tmo granularity is in msecs, it is not too expensive
671 * to refresh the timer, lets say every '8' msecs.
672 * Either the user can set the 'tmo' or we can derive it based on
673 * a) line-speed and b) block-size.
674 * prb_calc_retire_blk_tmo() calculates the tmo.
675 *
676 */
677static void prb_retire_rx_blk_timer_expired(unsigned long data)
678{
679 struct packet_sock *po = (struct packet_sock *)data;
Duan Jiong22781a52013-12-06 13:29:36 +0800680 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f102011-08-19 10:18:16 +0000681 unsigned int frozen;
chetan lokebc59ba32011-08-25 10:43:30 +0000682 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000683
684 spin_lock(&po->sk.sk_receive_queue.lock);
685
686 frozen = prb_queue_frozen(pkc);
687 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
688
689 if (unlikely(pkc->delete_blk_timer))
690 goto out;
691
692 /* We only need to plug the race when the block is partially filled.
693 * tpacket_rcv:
694 * lock(); increment BLOCK_NUM_PKTS; unlock()
695 * copy_bits() is in progress ...
696 * timer fires on other cpu:
697 * we can't retire the current block because copy_bits
698 * is in progress.
699 *
700 */
701 if (BLOCK_NUM_PKTS(pbd)) {
702 while (atomic_read(&pkc->blk_fill_in_prog)) {
703 /* Waiting for skb_copy_bits to finish... */
704 cpu_relax();
705 }
706 }
707
708 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
709 if (!frozen) {
710 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
711 if (!prb_dispatch_next_block(pkc, po))
712 goto refresh_timer;
713 else
714 goto out;
715 } else {
716 /* Case 1. Queue was frozen because user-space was
717 * lagging behind.
718 */
719 if (prb_curr_blk_in_use(pkc, pbd)) {
720 /*
721 * Ok, user-space is still behind.
722 * So just refresh the timer.
723 */
724 goto refresh_timer;
725 } else {
726 /* Case 2. queue was frozen,user-space caught up,
727 * now the link went idle && the timer fired.
728 * We don't have a block to close.So we open this
729 * block and restart the timer.
730 * opening a block thaws the queue,restarts timer
731 * Thawing/timer-refresh is a side effect.
732 */
733 prb_open_block(pkc, pbd);
734 goto out;
735 }
736 }
737 }
738
739refresh_timer:
740 _prb_refresh_rx_retire_blk_timer(pkc);
741
742out:
743 spin_unlock(&po->sk.sk_receive_queue.lock);
744}
745
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000746static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
chetan lokebc59ba32011-08-25 10:43:30 +0000747 struct tpacket_block_desc *pbd1, __u32 status)
chetan lokef6fb8f102011-08-19 10:18:16 +0000748{
749 /* Flush everything minus the block header */
750
751#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
752 u8 *start, *end;
753
754 start = (u8 *)pbd1;
755
756 /* Skip the block header(we know header WILL fit in 4K) */
757 start += PAGE_SIZE;
758
759 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
760 for (; start < end; start += PAGE_SIZE)
761 flush_dcache_page(pgv_to_page(start));
762
763 smp_wmb();
764#endif
765
766 /* Now update the block status. */
767
768 BLOCK_STATUS(pbd1) = status;
769
770 /* Flush the block header */
771
772#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
773 start = (u8 *)pbd1;
774 flush_dcache_page(pgv_to_page(start));
775
776 smp_wmb();
777#endif
778}
779
780/*
781 * Side effect:
782 *
783 * 1) flush the block
784 * 2) Increment active_blk_num
785 *
786 * Note:We DONT refresh the timer on purpose.
787 * Because almost always the next block will be opened.
788 */
chetan lokebc59ba32011-08-25 10:43:30 +0000789static void prb_close_block(struct tpacket_kbdq_core *pkc1,
790 struct tpacket_block_desc *pbd1,
chetan lokef6fb8f102011-08-19 10:18:16 +0000791 struct packet_sock *po, unsigned int stat)
792{
793 __u32 status = TP_STATUS_USER | stat;
794
795 struct tpacket3_hdr *last_pkt;
chetan lokebc59ba32011-08-25 10:43:30 +0000796 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
chetan lokef6fb8f102011-08-19 10:18:16 +0000797
Daniel Borkmannee80fbf2013-04-19 06:12:29 +0000798 if (po->stats.stats3.tp_drops)
chetan lokef6fb8f102011-08-19 10:18:16 +0000799 status |= TP_STATUS_LOSING;
800
801 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
802 last_pkt->tp_next_offset = 0;
803
804 /* Get the ts of the last pkt */
805 if (BLOCK_NUM_PKTS(pbd1)) {
806 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
807 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
808 } else {
809 /* Ok, we tmo'd - so get the current time */
810 struct timespec ts;
811 getnstimeofday(&ts);
812 h1->ts_last_pkt.ts_sec = ts.tv_sec;
813 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
814 }
815
816 smp_wmb();
817
818 /* Flush the block */
819 prb_flush_block(pkc1, pbd1, status);
820
821 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
822}
823
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000824static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000825{
826 pkc->reset_pending_on_curr_blk = 0;
827}
828
829/*
830 * Side effect of opening a block:
831 *
832 * 1) prb_queue is thawed.
833 * 2) retire_blk_timer is refreshed.
834 *
835 */
chetan lokebc59ba32011-08-25 10:43:30 +0000836static void prb_open_block(struct tpacket_kbdq_core *pkc1,
837 struct tpacket_block_desc *pbd1)
chetan lokef6fb8f102011-08-19 10:18:16 +0000838{
839 struct timespec ts;
chetan lokebc59ba32011-08-25 10:43:30 +0000840 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
chetan lokef6fb8f102011-08-19 10:18:16 +0000841
842 smp_rmb();
843
Daniel Borkmann8da30562013-05-03 02:57:00 +0000844 /* We could have just memset this but we will lose the
845 * flexibility of making the priv area sticky
846 */
chetan lokef6fb8f102011-08-19 10:18:16 +0000847
Daniel Borkmann8da30562013-05-03 02:57:00 +0000848 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
849 BLOCK_NUM_PKTS(pbd1) = 0;
850 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
chetan lokef6fb8f102011-08-19 10:18:16 +0000851
Daniel Borkmann8da30562013-05-03 02:57:00 +0000852 getnstimeofday(&ts);
chetan lokef6fb8f102011-08-19 10:18:16 +0000853
Daniel Borkmann8da30562013-05-03 02:57:00 +0000854 h1->ts_first_pkt.ts_sec = ts.tv_sec;
855 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
chetan lokef6fb8f102011-08-19 10:18:16 +0000856
Daniel Borkmann8da30562013-05-03 02:57:00 +0000857 pkc1->pkblk_start = (char *)pbd1;
858 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
859
860 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
861 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
862
863 pbd1->version = pkc1->version;
864 pkc1->prev = pkc1->nxt_offset;
865 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
866
867 prb_thaw_queue(pkc1);
868 _prb_refresh_rx_retire_blk_timer(pkc1);
869
870 smp_wmb();
chetan lokef6fb8f102011-08-19 10:18:16 +0000871}
872
873/*
874 * Queue freeze logic:
875 * 1) Assume tp_block_nr = 8 blocks.
876 * 2) At time 't0', user opens Rx ring.
877 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
878 * 4) user-space is either sleeping or processing block '0'.
879 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
880 * it will close block-7,loop around and try to fill block '0'.
881 * call-flow:
882 * __packet_lookup_frame_in_block
883 * prb_retire_current_block()
884 * prb_dispatch_next_block()
885 * |->(BLOCK_STATUS == USER) evaluates to true
886 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
887 * 6) Now there are two cases:
888 * 6.1) Link goes idle right after the queue is frozen.
889 * But remember, the last open_block() refreshed the timer.
890 * When this timer expires,it will refresh itself so that we can
891 * re-open block-0 in near future.
892 * 6.2) Link is busy and keeps on receiving packets. This is a simple
893 * case and __packet_lookup_frame_in_block will check if block-0
894 * is free and can now be re-used.
895 */
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000896static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000897 struct packet_sock *po)
898{
899 pkc->reset_pending_on_curr_blk = 1;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +0000900 po->stats.stats3.tp_freeze_q_cnt++;
chetan lokef6fb8f102011-08-19 10:18:16 +0000901}
902
903#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
904
905/*
906 * If the next block is free then we will dispatch it
907 * and return a good offset.
908 * Else, we will freeze the queue.
909 * So, caller must check the return value.
910 */
chetan lokebc59ba32011-08-25 10:43:30 +0000911static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000912 struct packet_sock *po)
913{
chetan lokebc59ba32011-08-25 10:43:30 +0000914 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000915
916 smp_rmb();
917
918 /* 1. Get current block num */
919 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
920
921 /* 2. If this block is currently in_use then freeze the queue */
922 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
923 prb_freeze_queue(pkc, po);
924 return NULL;
925 }
926
927 /*
928 * 3.
929 * open this block and return the offset where the first packet
930 * needs to get stored.
931 */
932 prb_open_block(pkc, pbd);
933 return (void *)pkc->nxt_offset;
934}
935
chetan lokebc59ba32011-08-25 10:43:30 +0000936static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000937 struct packet_sock *po, unsigned int status)
938{
chetan lokebc59ba32011-08-25 10:43:30 +0000939 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
chetan lokef6fb8f102011-08-19 10:18:16 +0000940
941 /* retire/close the current block */
942 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
943 /*
944 * Plug the case where copy_bits() is in progress on
945 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
946 * have space to copy the pkt in the current block and
947 * called prb_retire_current_block()
948 *
949 * We don't need to worry about the TMO case because
950 * the timer-handler already handled this case.
951 */
952 if (!(status & TP_STATUS_BLK_TMO)) {
953 while (atomic_read(&pkc->blk_fill_in_prog)) {
954 /* Waiting for skb_copy_bits to finish... */
955 cpu_relax();
956 }
957 }
958 prb_close_block(pkc, pbd, po, status);
959 return;
960 }
chetan lokef6fb8f102011-08-19 10:18:16 +0000961}
962
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000963static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
chetan lokebc59ba32011-08-25 10:43:30 +0000964 struct tpacket_block_desc *pbd)
chetan lokef6fb8f102011-08-19 10:18:16 +0000965{
966 return TP_STATUS_USER & BLOCK_STATUS(pbd);
967}
968
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000969static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000970{
971 return pkc->reset_pending_on_curr_blk;
972}
973
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000974static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
chetan lokef6fb8f102011-08-19 10:18:16 +0000975{
chetan lokebc59ba32011-08-25 10:43:30 +0000976 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
chetan lokef6fb8f102011-08-19 10:18:16 +0000977 atomic_dec(&pkc->blk_fill_in_prog);
978}
979
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000980static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000981 struct tpacket3_hdr *ppd)
982{
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800983 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
chetan lokef6fb8f102011-08-19 10:18:16 +0000984}
985
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000986static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000987 struct tpacket3_hdr *ppd)
988{
989 ppd->hv1.tp_rxhash = 0;
990}
991
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000992static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000993 struct tpacket3_hdr *ppd)
994{
995 if (vlan_tx_tag_present(pkc->skb)) {
996 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +0900997 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
998 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
chetan lokef6fb8f102011-08-19 10:18:16 +0000999 } else {
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +00001000 ppd->hv1.tp_vlan_tci = 0;
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09001001 ppd->hv1.tp_vlan_tpid = 0;
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +00001002 ppd->tp_status = TP_STATUS_AVAILABLE;
chetan lokef6fb8f102011-08-19 10:18:16 +00001003 }
1004}
1005
chetan lokebc59ba32011-08-25 10:43:30 +00001006static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +00001007 struct tpacket3_hdr *ppd)
1008{
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09001009 ppd->hv1.tp_padding = 0;
chetan lokef6fb8f102011-08-19 10:18:16 +00001010 prb_fill_vlan_info(pkc, ppd);
1011
1012 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1013 prb_fill_rxhash(pkc, ppd);
1014 else
1015 prb_clear_rxhash(pkc, ppd);
1016}
1017
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001018static void prb_fill_curr_block(char *curr,
chetan lokebc59ba32011-08-25 10:43:30 +00001019 struct tpacket_kbdq_core *pkc,
1020 struct tpacket_block_desc *pbd,
chetan lokef6fb8f102011-08-19 10:18:16 +00001021 unsigned int len)
1022{
1023 struct tpacket3_hdr *ppd;
1024
1025 ppd = (struct tpacket3_hdr *)curr;
1026 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1027 pkc->prev = curr;
1028 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1029 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1030 BLOCK_NUM_PKTS(pbd) += 1;
1031 atomic_inc(&pkc->blk_fill_in_prog);
1032 prb_run_all_ft_ops(pkc, ppd);
1033}
1034
1035/* Assumes caller has the sk->rx_queue.lock */
1036static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1037 struct sk_buff *skb,
1038 int status,
1039 unsigned int len
1040 )
1041{
chetan lokebc59ba32011-08-25 10:43:30 +00001042 struct tpacket_kbdq_core *pkc;
1043 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +00001044 char *curr, *end;
1045
Joe Perchese3192692012-06-03 17:41:40 +00001046 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f102011-08-19 10:18:16 +00001047 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1048
1049 /* Queue is frozen when user space is lagging behind */
1050 if (prb_queue_frozen(pkc)) {
1051 /*
1052 * Check if that last block which caused the queue to freeze,
1053 * is still in_use by user-space.
1054 */
1055 if (prb_curr_blk_in_use(pkc, pbd)) {
1056 /* Can't record this packet */
1057 return NULL;
1058 } else {
1059 /*
1060 * Ok, the block was released by user-space.
1061 * Now let's open that block.
1062 * opening a block also thaws the queue.
1063 * Thawing is a side effect.
1064 */
1065 prb_open_block(pkc, pbd);
1066 }
1067 }
1068
1069 smp_mb();
1070 curr = pkc->nxt_offset;
1071 pkc->skb = skb;
Joe Perchese3192692012-06-03 17:41:40 +00001072 end = (char *)pbd + pkc->kblk_size;
chetan lokef6fb8f102011-08-19 10:18:16 +00001073
1074 /* first try the current block */
1075 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1076 prb_fill_curr_block(curr, pkc, pbd, len);
1077 return (void *)curr;
1078 }
1079
1080 /* Ok, close the current block */
1081 prb_retire_current_block(pkc, po, 0);
1082
1083 /* Now, try to dispatch the next block */
1084 curr = (char *)prb_dispatch_next_block(pkc, po);
1085 if (curr) {
1086 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1087 prb_fill_curr_block(curr, pkc, pbd, len);
1088 return (void *)curr;
1089 }
1090
1091 /*
1092 * No free blocks are available.user_space hasn't caught up yet.
1093 * Queue was just frozen and now this packet will get dropped.
1094 */
1095 return NULL;
1096}
1097
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001098static void *packet_current_rx_frame(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001099 struct sk_buff *skb,
1100 int status, unsigned int len)
1101{
1102 char *curr = NULL;
1103 switch (po->tp_version) {
1104 case TPACKET_V1:
1105 case TPACKET_V2:
1106 curr = packet_lookup_frame(po, &po->rx_ring,
1107 po->rx_ring.head, status);
1108 return curr;
1109 case TPACKET_V3:
1110 return __packet_lookup_frame_in_block(po, skb, status, len);
1111 default:
1112 WARN(1, "TPACKET version not supported\n");
1113 BUG();
Ying Xue99aa3472012-08-06 16:27:10 +00001114 return NULL;
chetan lokef6fb8f102011-08-19 10:18:16 +00001115 }
1116}
1117
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001118static void *prb_lookup_block(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001119 struct packet_ring_buffer *rb,
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001120 unsigned int idx,
chetan lokef6fb8f102011-08-19 10:18:16 +00001121 int status)
1122{
chetan lokebc59ba32011-08-25 10:43:30 +00001123 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001124 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
chetan lokef6fb8f102011-08-19 10:18:16 +00001125
1126 if (status != BLOCK_STATUS(pbd))
1127 return NULL;
1128 return pbd;
1129}
1130
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001131static int prb_previous_blk_num(struct packet_ring_buffer *rb)
chetan lokef6fb8f102011-08-19 10:18:16 +00001132{
1133 unsigned int prev;
1134 if (rb->prb_bdqc.kactive_blk_num)
1135 prev = rb->prb_bdqc.kactive_blk_num-1;
1136 else
1137 prev = rb->prb_bdqc.knum_blocks-1;
1138 return prev;
1139}
1140
1141/* Assumes caller has held the rx_queue.lock */
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001142static void *__prb_previous_block(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001143 struct packet_ring_buffer *rb,
1144 int status)
1145{
1146 unsigned int previous = prb_previous_blk_num(rb);
1147 return prb_lookup_block(po, rb, previous, status);
1148}
1149
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001150static void *packet_previous_rx_frame(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001151 struct packet_ring_buffer *rb,
1152 int status)
1153{
1154 if (po->tp_version <= TPACKET_V2)
1155 return packet_previous_frame(po, rb, status);
1156
1157 return __prb_previous_block(po, rb, status);
1158}
1159
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001160static void packet_increment_rx_head(struct packet_sock *po,
chetan lokef6fb8f102011-08-19 10:18:16 +00001161 struct packet_ring_buffer *rb)
1162{
1163 switch (po->tp_version) {
1164 case TPACKET_V1:
1165 case TPACKET_V2:
1166 return packet_increment_head(rb);
1167 case TPACKET_V3:
1168 default:
1169 WARN(1, "TPACKET version not supported.\n");
1170 BUG();
1171 return;
1172 }
1173}
1174
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001175static void *packet_previous_frame(struct packet_sock *po,
Johann Baudy69e3c752009-05-18 22:11:22 -07001176 struct packet_ring_buffer *rb,
1177 int status)
1178{
1179 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1180 return packet_lookup_frame(po, rb, previous, status);
1181}
1182
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001183static void packet_increment_head(struct packet_ring_buffer *buff)
Johann Baudy69e3c752009-05-18 22:11:22 -07001184{
1185 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1186}
1187
Daniel Borkmannb0138402014-01-15 16:25:36 +01001188static void packet_inc_pending(struct packet_ring_buffer *rb)
1189{
1190 this_cpu_inc(*rb->pending_refcnt);
1191}
1192
1193static void packet_dec_pending(struct packet_ring_buffer *rb)
1194{
1195 this_cpu_dec(*rb->pending_refcnt);
1196}
1197
1198static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1199{
1200 unsigned int refcnt = 0;
1201 int cpu;
1202
1203 /* We don't use pending refcount in rx_ring. */
1204 if (rb->pending_refcnt == NULL)
1205 return 0;
1206
1207 for_each_possible_cpu(cpu)
1208 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1209
1210 return refcnt;
1211}
1212
1213static int packet_alloc_pending(struct packet_sock *po)
1214{
1215 po->rx_ring.pending_refcnt = NULL;
1216
1217 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1218 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1219 return -ENOBUFS;
1220
1221 return 0;
1222}
1223
1224static void packet_free_pending(struct packet_sock *po)
1225{
1226 free_percpu(po->tx_ring.pending_refcnt);
1227}
1228
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001229static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1230{
1231 struct sock *sk = &po->sk;
1232 bool has_room;
1233
1234 if (po->prot_hook.func != tpacket_rcv)
1235 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1236 <= sk->sk_rcvbuf;
1237
1238 spin_lock(&sk->sk_receive_queue.lock);
1239 if (po->tp_version == TPACKET_V3)
1240 has_room = prb_lookup_block(po, &po->rx_ring,
1241 po->rx_ring.prb_bdqc.kactive_blk_num,
1242 TP_STATUS_KERNEL);
1243 else
1244 has_room = packet_lookup_frame(po, &po->rx_ring,
1245 po->rx_ring.head,
1246 TP_STATUS_KERNEL);
1247 spin_unlock(&sk->sk_receive_queue.lock);
1248
1249 return has_room;
1250}
1251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252static void packet_sock_destruct(struct sock *sk)
1253{
Richard Cochraned85b562010-04-07 22:41:28 +00001254 skb_queue_purge(&sk->sk_error_queue);
1255
Ilpo Järvinen547b7922008-07-25 21:43:18 -07001256 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1257 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
1259 if (!sock_flag(sk, SOCK_DEAD)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001260 pr_err("Attempt to release alive packet socket: %p\n", sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 return;
1262 }
1263
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08001264 sk_refcnt_debug_dec(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265}
1266
David S. Millerdc99f602011-07-05 01:45:05 -07001267static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1268{
1269 int x = atomic_read(&f->rr_cur) + 1;
1270
1271 if (x >= num)
1272 x = 0;
1273
1274 return x;
1275}
1276
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001277static unsigned int fanout_demux_hash(struct packet_fanout *f,
1278 struct sk_buff *skb,
1279 unsigned int num)
David S. Millerdc99f602011-07-05 01:45:05 -07001280{
Tom Herbert61b905d2014-03-24 15:34:47 -07001281 return reciprocal_scale(skb_get_hash(skb), num);
David S. Millerdc99f602011-07-05 01:45:05 -07001282}
1283
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001284static unsigned int fanout_demux_lb(struct packet_fanout *f,
1285 struct sk_buff *skb,
1286 unsigned int num)
David S. Millerdc99f602011-07-05 01:45:05 -07001287{
1288 int cur, old;
1289
1290 cur = atomic_read(&f->rr_cur);
1291 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1292 fanout_rr_next(f, num))) != cur)
1293 cur = old;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001294 return cur;
David S. Millerdc99f602011-07-05 01:45:05 -07001295}
1296
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001297static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1298 struct sk_buff *skb,
1299 unsigned int num)
David S. Miller95ec3eb2011-07-06 01:56:38 -07001300{
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001301 return smp_processor_id() % num;
1302}
David S. Miller95ec3eb2011-07-06 01:56:38 -07001303
Daniel Borkmann5df0ddf2013-08-28 22:13:09 +02001304static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1305 struct sk_buff *skb,
1306 unsigned int num)
1307{
Daniel Borkmannf337db62014-01-22 02:29:39 +01001308 return prandom_u32_max(num);
Daniel Borkmann5df0ddf2013-08-28 22:13:09 +02001309}
1310
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001311static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1312 struct sk_buff *skb,
1313 unsigned int idx, unsigned int skip,
1314 unsigned int num)
1315{
1316 unsigned int i, j;
1317
1318 i = j = min_t(int, f->next[idx], num - 1);
1319 do {
1320 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1321 if (i != j)
1322 f->next[idx] = i;
1323 return i;
1324 }
1325 if (++i == num)
1326 i = 0;
1327 } while (i != j);
1328
1329 return idx;
1330}
1331
Neil Horman2d360972014-01-22 16:01:44 -05001332static unsigned int fanout_demux_qm(struct packet_fanout *f,
1333 struct sk_buff *skb,
1334 unsigned int num)
1335{
1336 return skb_get_queue_mapping(skb) % num;
1337}
1338
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001339static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1340{
1341 return f->flags & (flag >> 8);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001342}
1343
David S. Miller95ec3eb2011-07-06 01:56:38 -07001344static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1345 struct packet_type *pt, struct net_device *orig_dev)
David S. Millerdc99f602011-07-05 01:45:05 -07001346{
1347 struct packet_fanout *f = pt->af_packet_priv;
1348 unsigned int num = f->num_members;
1349 struct packet_sock *po;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001350 unsigned int idx;
David S. Millerdc99f602011-07-05 01:45:05 -07001351
1352 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1353 !num) {
1354 kfree_skb(skb);
1355 return 0;
1356 }
1357
David S. Miller95ec3eb2011-07-06 01:56:38 -07001358 switch (f->type) {
1359 case PACKET_FANOUT_HASH:
1360 default:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001361 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
Eric Dumazetbc416d92011-10-06 10:28:31 +00001362 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001363 if (!skb)
1364 return 0;
1365 }
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001366 idx = fanout_demux_hash(f, skb, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001367 break;
1368 case PACKET_FANOUT_LB:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001369 idx = fanout_demux_lb(f, skb, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001370 break;
1371 case PACKET_FANOUT_CPU:
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001372 idx = fanout_demux_cpu(f, skb, num);
1373 break;
Daniel Borkmann5df0ddf2013-08-28 22:13:09 +02001374 case PACKET_FANOUT_RND:
1375 idx = fanout_demux_rnd(f, skb, num);
1376 break;
Neil Horman2d360972014-01-22 16:01:44 -05001377 case PACKET_FANOUT_QM:
1378 idx = fanout_demux_qm(f, skb, num);
1379 break;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001380 case PACKET_FANOUT_ROLLOVER:
1381 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
David S. Miller95ec3eb2011-07-06 01:56:38 -07001382 break;
David S. Miller7736d332011-07-05 01:43:20 -07001383 }
1384
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001385 po = pkt_sk(f->arr[idx]);
1386 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1387 unlikely(!packet_rcv_has_room(po, skb))) {
1388 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1389 po = pkt_sk(f->arr[idx]);
1390 }
David S. Millerdc99f602011-07-05 01:45:05 -07001391
1392 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1393}
1394
Pavel Emelyanovfff33212012-08-16 05:36:48 +00001395DEFINE_MUTEX(fanout_mutex);
1396EXPORT_SYMBOL_GPL(fanout_mutex);
David S. Millerdc99f602011-07-05 01:45:05 -07001397static LIST_HEAD(fanout_list);
1398
1399static void __fanout_link(struct sock *sk, struct packet_sock *po)
1400{
1401 struct packet_fanout *f = po->fanout;
1402
1403 spin_lock(&f->lock);
1404 f->arr[f->num_members] = sk;
1405 smp_wmb();
1406 f->num_members++;
1407 spin_unlock(&f->lock);
1408}
1409
1410static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1411{
1412 struct packet_fanout *f = po->fanout;
1413 int i;
1414
1415 spin_lock(&f->lock);
1416 for (i = 0; i < f->num_members; i++) {
1417 if (f->arr[i] == sk)
1418 break;
1419 }
1420 BUG_ON(i >= f->num_members);
1421 f->arr[i] = f->arr[f->num_members - 1];
1422 f->num_members--;
1423 spin_unlock(&f->lock);
1424}
1425
Weilong Chend4dd8ae2013-12-23 11:31:38 +08001426static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
Eric Leblondc0de08d2012-08-16 22:02:58 +00001427{
Weilong Chend4dd8ae2013-12-23 11:31:38 +08001428 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
Eric Leblondc0de08d2012-08-16 22:02:58 +00001429 return true;
1430
1431 return false;
1432}
1433
David S. Miller7736d332011-07-05 01:43:20 -07001434static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
David S. Millerdc99f602011-07-05 01:45:05 -07001435{
1436 struct packet_sock *po = pkt_sk(sk);
1437 struct packet_fanout *f, *match;
David S. Miller7736d332011-07-05 01:43:20 -07001438 u8 type = type_flags & 0xff;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001439 u8 flags = type_flags >> 8;
David S. Millerdc99f602011-07-05 01:45:05 -07001440 int err;
1441
1442 switch (type) {
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001443 case PACKET_FANOUT_ROLLOVER:
1444 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1445 return -EINVAL;
David S. Millerdc99f602011-07-05 01:45:05 -07001446 case PACKET_FANOUT_HASH:
1447 case PACKET_FANOUT_LB:
David S. Miller95ec3eb2011-07-06 01:56:38 -07001448 case PACKET_FANOUT_CPU:
Daniel Borkmann5df0ddf2013-08-28 22:13:09 +02001449 case PACKET_FANOUT_RND:
Neil Horman2d360972014-01-22 16:01:44 -05001450 case PACKET_FANOUT_QM:
David S. Millerdc99f602011-07-05 01:45:05 -07001451 break;
1452 default:
1453 return -EINVAL;
1454 }
1455
1456 if (!po->running)
1457 return -EINVAL;
1458
1459 if (po->fanout)
1460 return -EALREADY;
1461
1462 mutex_lock(&fanout_mutex);
1463 match = NULL;
1464 list_for_each_entry(f, &fanout_list, list) {
1465 if (f->id == id &&
1466 read_pnet(&f->net) == sock_net(sk)) {
1467 match = f;
1468 break;
1469 }
1470 }
Eric Dumazetafe62c62011-07-07 06:41:29 -07001471 err = -EINVAL;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001472 if (match && match->flags != flags)
Eric Dumazetafe62c62011-07-07 06:41:29 -07001473 goto out;
David S. Millerdc99f602011-07-05 01:45:05 -07001474 if (!match) {
Eric Dumazetafe62c62011-07-07 06:41:29 -07001475 err = -ENOMEM;
David S. Millerdc99f602011-07-05 01:45:05 -07001476 match = kzalloc(sizeof(*match), GFP_KERNEL);
Eric Dumazetafe62c62011-07-07 06:41:29 -07001477 if (!match)
1478 goto out;
1479 write_pnet(&match->net, sock_net(sk));
1480 match->id = id;
1481 match->type = type;
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00001482 match->flags = flags;
Eric Dumazetafe62c62011-07-07 06:41:29 -07001483 atomic_set(&match->rr_cur, 0);
1484 INIT_LIST_HEAD(&match->list);
1485 spin_lock_init(&match->lock);
1486 atomic_set(&match->sk_ref, 0);
1487 match->prot_hook.type = po->prot_hook.type;
1488 match->prot_hook.dev = po->prot_hook.dev;
1489 match->prot_hook.func = packet_rcv_fanout;
1490 match->prot_hook.af_packet_priv = match;
Eric Leblondc0de08d2012-08-16 22:02:58 +00001491 match->prot_hook.id_match = match_fanout_group;
Eric Dumazetafe62c62011-07-07 06:41:29 -07001492 dev_add_pack(&match->prot_hook);
1493 list_add(&match->list, &fanout_list);
1494 }
1495 err = -EINVAL;
1496 if (match->type == type &&
1497 match->prot_hook.type == po->prot_hook.type &&
1498 match->prot_hook.dev == po->prot_hook.dev) {
1499 err = -ENOSPC;
1500 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1501 __dev_remove_pack(&po->prot_hook);
1502 po->fanout = match;
1503 atomic_inc(&match->sk_ref);
1504 __fanout_link(sk, po);
1505 err = 0;
David S. Millerdc99f602011-07-05 01:45:05 -07001506 }
1507 }
Eric Dumazetafe62c62011-07-07 06:41:29 -07001508out:
David S. Millerdc99f602011-07-05 01:45:05 -07001509 mutex_unlock(&fanout_mutex);
1510 return err;
1511}
1512
1513static void fanout_release(struct sock *sk)
1514{
1515 struct packet_sock *po = pkt_sk(sk);
1516 struct packet_fanout *f;
1517
1518 f = po->fanout;
1519 if (!f)
1520 return;
1521
Pavel Emelyanovfff33212012-08-16 05:36:48 +00001522 mutex_lock(&fanout_mutex);
David S. Millerdc99f602011-07-05 01:45:05 -07001523 po->fanout = NULL;
1524
David S. Millerdc99f602011-07-05 01:45:05 -07001525 if (atomic_dec_and_test(&f->sk_ref)) {
1526 list_del(&f->list);
1527 dev_remove_pack(&f->prot_hook);
1528 kfree(f);
1529 }
1530 mutex_unlock(&fanout_mutex);
1531}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08001533static const struct proto_ops packet_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08001535static const struct proto_ops packet_ops_spkt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001537static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1538 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539{
1540 struct sock *sk;
1541 struct sockaddr_pkt *spkt;
1542
1543 /*
1544 * When we registered the protocol we saved the socket in the data
1545 * field for just this event.
1546 */
1547
1548 sk = pt->af_packet_priv;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001549
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 /*
1551 * Yank back the headers [hope the device set this
1552 * right or kerboom...]
1553 *
1554 * Incoming packets have ll header pulled,
1555 * push it back.
1556 *
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001557 * For outgoing ones skb->data == skb_mac_header(skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 * so that this procedure is noop.
1559 */
1560
1561 if (skb->pkt_type == PACKET_LOOPBACK)
1562 goto out;
1563
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001564 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08001565 goto out;
1566
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001567 skb = skb_share_check(skb, GFP_ATOMIC);
1568 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 goto oom;
1570
1571 /* drop any routing info */
Eric Dumazetadf30902009-06-02 05:19:30 +00001572 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573
Phil Oester84531c22005-07-12 11:57:52 -07001574 /* drop conntrack reference */
1575 nf_reset(skb);
1576
Herbert Xuffbc6112007-02-04 23:33:10 -08001577 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001579 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580
1581 /*
1582 * The SOCK_PACKET socket receives _all_ frames.
1583 */
1584
1585 spkt->spkt_family = dev->type;
1586 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1587 spkt->spkt_protocol = skb->protocol;
1588
1589 /*
1590 * Charge the memory to the socket. This is done specifically
1591 * to prevent sockets using all the memory up.
1592 */
1593
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001594 if (sock_queue_rcv_skb(sk, skb) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 return 0;
1596
1597out:
1598 kfree_skb(skb);
1599oom:
1600 return 0;
1601}
1602
1603
1604/*
1605 * Output a raw packet to a device layer. This bypasses all the other
1606 * protocol layers and you must therefore supply it with a complete frame
1607 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001608
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1610 struct msghdr *msg, size_t len)
1611{
1612 struct sock *sk = sock->sk;
Steffen Hurrle342dfc32014-01-17 22:53:15 +01001613 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001614 struct sk_buff *skb = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 struct net_device *dev;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001616 __be16 proto = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 int err;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001618 int extra_len = 0;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001619
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001621 * Get and verify the address.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 */
1623
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001624 if (saddr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 if (msg->msg_namelen < sizeof(struct sockaddr))
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001626 return -EINVAL;
1627 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1628 proto = saddr->spkt_protocol;
1629 } else
1630 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001633 * Find the device first to size check it
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 */
1635
danborkmann@iogearbox.netde74e922012-06-10 08:59:28 +00001636 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001637retry:
Eric Dumazet654d1f82009-11-02 10:43:32 +01001638 rcu_read_lock();
1639 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 err = -ENODEV;
1641 if (dev == NULL)
1642 goto out_unlock;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001643
David S. Millerd5e76b02007-01-25 19:30:36 -08001644 err = -ENETDOWN;
1645 if (!(dev->flags & IFF_UP))
1646 goto out_unlock;
1647
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 /*
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001649 * You may not queue a frame bigger than the mtu. This is the lowest level
1650 * raw protocol and you must do your own fragmentation at this level.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001652
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001653 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1654 if (!netif_supports_nofcs(dev)) {
1655 err = -EPROTONOSUPPORT;
1656 goto out_unlock;
1657 }
1658 extra_len = 4; /* We're doing our own CRC */
1659 }
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 err = -EMSGSIZE;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001662 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 goto out_unlock;
1664
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001665 if (!skb) {
1666 size_t reserved = LL_RESERVED_SPACE(dev);
Herbert Xu4ce40912011-11-18 02:20:05 +00001667 int tlen = dev->needed_tailroom;
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001668 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001670 rcu_read_unlock();
Herbert Xu4ce40912011-11-18 02:20:05 +00001671 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001672 if (skb == NULL)
1673 return -ENOBUFS;
1674 /* FIXME: Save some space for broken drivers that write a hard
1675 * header at transmission time by themselves. PPP is the notable
1676 * one here. This should really be fixed at the driver level.
1677 */
1678 skb_reserve(skb, reserved);
1679 skb_reset_network_header(skb);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09001680
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001681 /* Try to align data part correctly */
1682 if (hhlen) {
1683 skb->data -= hhlen;
1684 skb->tail -= hhlen;
1685 if (len < hhlen)
1686 skb_reset_network_header(skb);
1687 }
1688 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1689 if (err)
1690 goto out_free;
1691 goto retry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 }
1693
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001694 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
Ben Greear57f89bf2011-02-11 09:35:18 +00001695 /* Earlier code assumed this would be a VLAN pkt,
1696 * double-check this now that we have the actual
1697 * packet in hand.
1698 */
1699 struct ethhdr *ehdr;
1700 skb_reset_mac_header(skb);
1701 ehdr = eth_hdr(skb);
1702 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1703 err = -EMSGSIZE;
1704 goto out_unlock;
1705 }
1706 }
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 skb->protocol = proto;
1709 skb->dev = dev;
1710 skb->priority = sk->sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00001711 skb->mark = sk->sk_mark;
Daniel Borkmannbf84a012013-04-14 08:08:13 +00001712
1713 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714
Ben Greear3bdc0eb2012-02-11 15:39:30 +00001715 if (unlikely(extra_len == 4))
1716 skb->no_fcs = 1;
1717
Jason Wang40893fd2013-03-26 23:11:22 +00001718 skb_probe_transport_header(skb, 0);
Jason Wangc1aad272013-03-25 20:19:57 +00001719
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 dev_queue_xmit(skb);
Eric Dumazet654d1f82009-11-02 10:43:32 +01001721 rcu_read_unlock();
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001722 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724out_unlock:
Eric Dumazet654d1f82009-11-02 10:43:32 +01001725 rcu_read_unlock();
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001726out_free:
1727 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 return err;
1729}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
Olof Johanssoneea49cc92011-11-02 11:00:49 +00001731static unsigned int run_filter(const struct sk_buff *skb,
Eric Dumazet62ab0812010-12-06 20:50:09 +00001732 const struct sock *sk,
David S. Millerdbcb5852007-01-24 15:21:02 -08001733 unsigned int res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
1735 struct sk_filter *filter;
1736
Eric Dumazet80f8f102011-01-18 07:46:52 +00001737 rcu_read_lock();
1738 filter = rcu_dereference(sk->sk_filter);
David S. Millerdbcb5852007-01-24 15:21:02 -08001739 if (filter != NULL)
Eric Dumazet0a148422011-04-20 09:27:32 +00001740 res = SK_RUN_FILTER(filter, skb);
Eric Dumazet80f8f102011-01-18 07:46:52 +00001741 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
David S. Millerdbcb5852007-01-24 15:21:02 -08001743 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744}
1745
1746/*
Eric Dumazet62ab0812010-12-06 20:50:09 +00001747 * This function makes lazy skb cloning in hope that most of packets
1748 * are discarded by BPF.
1749 *
1750 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1751 * and skb->cb are mangled. It works because (and until) packets
1752 * falling here are owned by current CPU. Output packets are cloned
1753 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1754 * sequencially, so that if we return skb to original state on exit,
1755 * we will not harm anyone.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 */
1757
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001758static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1759 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
1761 struct sock *sk;
1762 struct sockaddr_ll *sll;
1763 struct packet_sock *po;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001764 u8 *skb_head = skb->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 int skb_len = skb->len;
David S. Millerdbcb5852007-01-24 15:21:02 -08001766 unsigned int snaplen, res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
1768 if (skb->pkt_type == PACKET_LOOPBACK)
1769 goto drop;
1770
1771 sk = pt->af_packet_priv;
1772 po = pkt_sk(sk);
1773
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001774 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08001775 goto drop;
1776
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 skb->dev = dev;
1778
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -07001779 if (dev->header_ops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 /* The device has an explicit notion of ll header,
Eric Dumazet62ab0812010-12-06 20:50:09 +00001781 * exported to higher levels.
1782 *
1783 * Otherwise, the device hides details of its frame
1784 * structure, so that corresponding packet head is
1785 * never delivered to user.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 */
1787 if (sk->sk_type != SOCK_DGRAM)
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001788 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 else if (skb->pkt_type == PACKET_OUTGOING) {
1790 /* Special case: outgoing packets have ll header at head */
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001791 skb_pull(skb, skb_network_offset(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 }
1793 }
1794
1795 snaplen = skb->len;
1796
David S. Millerdbcb5852007-01-24 15:21:02 -08001797 res = run_filter(skb, sk, snaplen);
1798 if (!res)
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07001799 goto drop_n_restore;
David S. Millerdbcb5852007-01-24 15:21:02 -08001800 if (snaplen > res)
1801 snaplen = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
Eric Dumazet0fd7bac2011-12-21 07:11:44 +00001803 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 goto drop_n_acct;
1805
1806 if (skb_shared(skb)) {
1807 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1808 if (nskb == NULL)
1809 goto drop_n_acct;
1810
1811 if (skb_head != skb->data) {
1812 skb->data = skb_head;
1813 skb->len = skb_len;
1814 }
Eric Dumazetabc4e4f2012-04-19 02:24:42 +00001815 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 skb = nskb;
1817 }
1818
Herbert Xuffbc6112007-02-04 23:33:10 -08001819 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1820 sizeof(skb->cb));
1821
1822 sll = &PACKET_SKB_CB(skb)->sa.ll;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 sll->sll_family = AF_PACKET;
1824 sll->sll_hatype = dev->type;
1825 sll->sll_protocol = skb->protocol;
1826 sll->sll_pkttype = skb->pkt_type;
Peter P Waskiewicz Jr8032b462007-11-10 22:03:25 -08001827 if (unlikely(po->origdev))
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07001828 sll->sll_ifindex = orig_dev->ifindex;
1829 else
1830 sll->sll_ifindex = dev->ifindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831
Stephen Hemmingerb95cce32007-09-26 22:13:38 -07001832 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Herbert Xuffbc6112007-02-04 23:33:10 -08001834 PACKET_SKB_CB(skb)->origlen = skb->len;
Herbert Xu8dc41942007-02-04 23:31:32 -08001835
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 if (pskb_trim(skb, snaplen))
1837 goto drop_n_acct;
1838
1839 skb_set_owner_r(skb, sk);
1840 skb->dev = NULL;
Eric Dumazetadf30902009-06-02 05:19:30 +00001841 skb_dst_drop(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842
Phil Oester84531c22005-07-12 11:57:52 -07001843 /* drop conntrack reference */
1844 nf_reset(skb);
1845
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 spin_lock(&sk->sk_receive_queue.lock);
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00001847 po->stats.stats1.tp_packets++;
Neil Horman3b885782009-10-12 13:26:31 -07001848 skb->dropcount = atomic_read(&sk->sk_drops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 __skb_queue_tail(&sk->sk_receive_queue, skb);
1850 spin_unlock(&sk->sk_receive_queue.lock);
David S. Miller676d2362014-04-11 16:15:36 -04001851 sk->sk_data_ready(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 return 0;
1853
1854drop_n_acct:
Willem de Bruijn7091fbd2011-09-30 10:38:28 +00001855 spin_lock(&sk->sk_receive_queue.lock);
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00001856 po->stats.stats1.tp_drops++;
Willem de Bruijn7091fbd2011-09-30 10:38:28 +00001857 atomic_inc(&sk->sk_drops);
1858 spin_unlock(&sk->sk_receive_queue.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859
1860drop_n_restore:
1861 if (skb_head != skb->data && skb_shared(skb)) {
1862 skb->data = skb_head;
1863 skb->len = skb_len;
1864 }
1865drop:
Neil Hormanead2ceb2009-03-11 09:49:55 +00001866 consume_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 return 0;
1868}
1869
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001870static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1871 struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
1873 struct sock *sk;
1874 struct packet_sock *po;
1875 struct sockaddr_ll *sll;
Daniel Borkmann184f4892013-04-16 01:57:46 +00001876 union tpacket_uhdr h;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00001877 u8 *skb_head = skb->data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 int skb_len = skb->len;
David S. Millerdbcb5852007-01-24 15:21:02 -08001879 unsigned int snaplen, res;
chetan lokef6fb8f102011-08-19 10:18:16 +00001880 unsigned long status = TP_STATUS_USER;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001881 unsigned short macoff, netoff, hdrlen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 struct sk_buff *copy_skb = NULL;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001883 struct timespec ts;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001884 __u32 ts_status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
Atzm Watanabe51846352013-12-17 22:53:32 +09001886 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1887 * We may add members to them until current aligned size without forcing
1888 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1889 */
1890 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1891 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1892
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893 if (skb->pkt_type == PACKET_LOOPBACK)
1894 goto drop;
1895
1896 sk = pt->af_packet_priv;
1897 po = pkt_sk(sk);
1898
Octavian Purdila09ad9bc2009-11-25 15:14:13 -08001899 if (!net_eq(dev_net(dev), sock_net(sk)))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08001900 goto drop;
1901
Stephen Hemminger3b04ddd2007-10-09 01:40:57 -07001902 if (dev->header_ops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 if (sk->sk_type != SOCK_DGRAM)
Arnaldo Carvalho de Melo98e399f2007-03-19 15:33:04 -07001904 skb_push(skb, skb->data - skb_mac_header(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905 else if (skb->pkt_type == PACKET_OUTGOING) {
1906 /* Special case: outgoing packets have ll header at head */
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03001907 skb_pull(skb, skb_network_offset(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 }
1909 }
1910
Herbert Xu8dc41942007-02-04 23:31:32 -08001911 if (skb->ip_summed == CHECKSUM_PARTIAL)
1912 status |= TP_STATUS_CSUMNOTREADY;
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 snaplen = skb->len;
1915
David S. Millerdbcb5852007-01-24 15:21:02 -08001916 res = run_filter(skb, sk, snaplen);
1917 if (!res)
Dmitry Mishinfda9ef52006-08-31 15:28:39 -07001918 goto drop_n_restore;
David S. Millerdbcb5852007-01-24 15:21:02 -08001919 if (snaplen > res)
1920 snaplen = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
1922 if (sk->sk_type == SOCK_DGRAM) {
Patrick McHardy8913336a2008-07-18 18:05:19 -07001923 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1924 po->tp_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 } else {
Eric Dumazet95c96172012-04-15 05:58:06 +00001926 unsigned int maclen = skb_network_offset(skb);
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001927 netoff = TPACKET_ALIGN(po->tp_hdrlen +
Patrick McHardy8913336a2008-07-18 18:05:19 -07001928 (maclen < 16 ? 16 : maclen)) +
1929 po->tp_reserve;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 macoff = netoff - maclen;
1931 }
chetan lokef6fb8f102011-08-19 10:18:16 +00001932 if (po->tp_version <= TPACKET_V2) {
1933 if (macoff + snaplen > po->rx_ring.frame_size) {
1934 if (po->copy_thresh &&
Eric Dumazet0fd7bac2011-12-21 07:11:44 +00001935 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
chetan lokef6fb8f102011-08-19 10:18:16 +00001936 if (skb_shared(skb)) {
1937 copy_skb = skb_clone(skb, GFP_ATOMIC);
1938 } else {
1939 copy_skb = skb_get(skb);
1940 skb_head = skb->data;
1941 }
1942 if (copy_skb)
1943 skb_set_owner_r(copy_skb, sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 }
chetan lokef6fb8f102011-08-19 10:18:16 +00001945 snaplen = po->rx_ring.frame_size - macoff;
1946 if ((int)snaplen < 0)
1947 snaplen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 spin_lock(&sk->sk_receive_queue.lock);
chetan lokef6fb8f102011-08-19 10:18:16 +00001951 h.raw = packet_current_rx_frame(po, skb,
1952 TP_STATUS_KERNEL, (macoff+snaplen));
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001953 if (!h.raw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 goto ring_is_full;
chetan lokef6fb8f102011-08-19 10:18:16 +00001955 if (po->tp_version <= TPACKET_V2) {
1956 packet_increment_rx_head(po, &po->rx_ring);
1957 /*
1958 * LOSING will be reported till you read the stats,
1959 * because it's COR - Clear On Read.
1960 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1961 * at packet level.
1962 */
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00001963 if (po->stats.stats1.tp_drops)
chetan lokef6fb8f102011-08-19 10:18:16 +00001964 status |= TP_STATUS_LOSING;
1965 }
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00001966 po->stats.stats1.tp_packets++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 if (copy_skb) {
1968 status |= TP_STATUS_COPY;
1969 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 spin_unlock(&sk->sk_receive_queue.lock);
1972
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001973 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001974
1975 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
Daniel Borkmann7a513842013-04-23 00:39:29 +00001976 getnstimeofday(&ts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001978 status |= ts_status;
1979
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001980 switch (po->tp_version) {
1981 case TPACKET_V1:
1982 h.h1->tp_len = skb->len;
1983 h.h1->tp_snaplen = snaplen;
1984 h.h1->tp_mac = macoff;
1985 h.h1->tp_net = netoff;
Daniel Borkmann4b457bd2013-04-16 01:29:11 +00001986 h.h1->tp_sec = ts.tv_sec;
1987 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001988 hdrlen = sizeof(*h.h1);
1989 break;
1990 case TPACKET_V2:
1991 h.h2->tp_len = skb->len;
1992 h.h2->tp_snaplen = snaplen;
1993 h.h2->tp_mac = macoff;
1994 h.h2->tp_net = netoff;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001995 h.h2->tp_sec = ts.tv_sec;
1996 h.h2->tp_nsec = ts.tv_nsec;
Ben Greeara3bcc232011-06-01 06:49:10 +00001997 if (vlan_tx_tag_present(skb)) {
1998 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09001999 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2000 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
Ben Greeara3bcc232011-06-01 06:49:10 +00002001 } else {
2002 h.h2->tp_vlan_tci = 0;
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09002003 h.h2->tp_vlan_tpid = 0;
Ben Greeara3bcc232011-06-01 06:49:10 +00002004 }
Atzm Watanabee4d26f42013-12-17 22:53:36 +09002005 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002006 hdrlen = sizeof(*h.h2);
2007 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00002008 case TPACKET_V3:
2009 /* tp_nxt_offset,vlan are already populated above.
2010 * So DONT clear those fields here
2011 */
2012 h.h3->tp_status |= status;
2013 h.h3->tp_len = skb->len;
2014 h.h3->tp_snaplen = snaplen;
2015 h.h3->tp_mac = macoff;
2016 h.h3->tp_net = netoff;
chetan lokef6fb8f102011-08-19 10:18:16 +00002017 h.h3->tp_sec = ts.tv_sec;
2018 h.h3->tp_nsec = ts.tv_nsec;
Atzm Watanabee4d26f42013-12-17 22:53:36 +09002019 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
chetan lokef6fb8f102011-08-19 10:18:16 +00002020 hdrlen = sizeof(*h.h3);
2021 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002022 default:
2023 BUG();
2024 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002026 sll = h.raw + TPACKET_ALIGN(hdrlen);
Stephen Hemmingerb95cce32007-09-26 22:13:38 -07002027 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 sll->sll_family = AF_PACKET;
2029 sll->sll_hatype = dev->type;
2030 sll->sll_protocol = skb->protocol;
2031 sll->sll_pkttype = skb->pkt_type;
Peter P Waskiewicz Jr8032b462007-11-10 22:03:25 -08002032 if (unlikely(po->origdev))
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07002033 sll->sll_ifindex = orig_dev->ifindex;
2034 else
2035 sll->sll_ifindex = dev->ifindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
Ralf Baechlee16aa202006-12-07 00:11:33 -08002037 smp_mb();
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002038
Changli Gaof6dafa92010-12-07 04:26:16 +00002039#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002040 if (po->tp_version <= TPACKET_V2) {
Changli Gao0af55bb2010-12-01 02:52:20 +00002041 u8 *start, *end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002043 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2044 macoff + snaplen);
2045
2046 for (start = h.raw; start < end; start += PAGE_SIZE)
2047 flush_dcache_page(pgv_to_page(start));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 }
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002049 smp_wmb();
Changli Gaof6dafa92010-12-07 04:26:16 +00002050#endif
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002051
chetan lokef6fb8f102011-08-19 10:18:16 +00002052 if (po->tp_version <= TPACKET_V2)
2053 __packet_set_status(po, h.raw, status);
2054 else
2055 prb_clear_blk_fill_status(&po->rx_ring);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
David S. Miller676d2362014-04-11 16:15:36 -04002057 sk->sk_data_ready(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
2059drop_n_restore:
2060 if (skb_head != skb->data && skb_shared(skb)) {
2061 skb->data = skb_head;
2062 skb->len = skb_len;
2063 }
2064drop:
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002065 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 return 0;
2067
2068ring_is_full:
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00002069 po->stats.stats1.tp_drops++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 spin_unlock(&sk->sk_receive_queue.lock);
2071
David S. Miller676d2362014-04-11 16:15:36 -04002072 sk->sk_data_ready(sk);
Wei Yongjunacb5d752009-02-25 00:36:42 +00002073 kfree_skb(copy_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 goto drop_n_restore;
2075}
2076
Johann Baudy69e3c752009-05-18 22:11:22 -07002077static void tpacket_destruct_skb(struct sk_buff *skb)
2078{
2079 struct packet_sock *po = pkt_sk(skb->sk);
Johann Baudy69e3c752009-05-18 22:11:22 -07002080
Johann Baudy69e3c752009-05-18 22:11:22 -07002081 if (likely(po->tx_ring.pg_vec)) {
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002082 void *ph;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002083 __u32 ts;
2084
Johann Baudy69e3c752009-05-18 22:11:22 -07002085 ph = skb_shinfo(skb)->destructor_arg;
Daniel Borkmannb0138402014-01-15 16:25:36 +01002086 packet_dec_pending(&po->tx_ring);
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002087
2088 ts = __packet_set_timestamp(po, ph, skb);
2089 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
Johann Baudy69e3c752009-05-18 22:11:22 -07002090 }
2091
2092 sock_wfree(skb);
2093}
2094
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002095static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2096 void *frame, struct net_device *dev, int size_max,
Herbert Xuae641942011-11-18 02:20:04 +00002097 __be16 proto, unsigned char *addr, int hlen)
Johann Baudy69e3c752009-05-18 22:11:22 -07002098{
Daniel Borkmann184f4892013-04-16 01:57:46 +00002099 union tpacket_uhdr ph;
David S. Miller09effa62013-08-07 17:11:00 -07002100 int to_write, offset, len, tp_len, nr_frags, len_max;
Johann Baudy69e3c752009-05-18 22:11:22 -07002101 struct socket *sock = po->sk.sk_socket;
2102 struct page *page;
2103 void *data;
2104 int err;
2105
2106 ph.raw = frame;
2107
2108 skb->protocol = proto;
2109 skb->dev = dev;
2110 skb->priority = po->sk.sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00002111 skb->mark = po->sk.sk_mark;
Willem de Bruijn2e313962013-04-23 00:39:28 +00002112 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
Johann Baudy69e3c752009-05-18 22:11:22 -07002113 skb_shinfo(skb)->destructor_arg = ph.raw;
2114
2115 switch (po->tp_version) {
2116 case TPACKET_V2:
2117 tp_len = ph.h2->tp_len;
2118 break;
2119 default:
2120 tp_len = ph.h1->tp_len;
2121 break;
2122 }
David S. Miller09effa62013-08-07 17:11:00 -07002123 if (unlikely(tp_len > size_max)) {
2124 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2125 return -EMSGSIZE;
2126 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002127
Herbert Xuae641942011-11-18 02:20:04 +00002128 skb_reserve(skb, hlen);
Johann Baudy69e3c752009-05-18 22:11:22 -07002129 skb_reset_network_header(skb);
Jason Wangc1aad272013-03-25 20:19:57 +00002130
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002131 if (!packet_use_direct_xmit(po))
2132 skb_probe_transport_header(skb, 0);
2133 if (unlikely(po->tp_tx_has_off)) {
Paul Chavent5920cd3a2012-11-06 23:10:47 +00002134 int off_min, off_max, off;
2135 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2136 off_max = po->tx_ring.frame_size - tp_len;
2137 if (sock->type == SOCK_DGRAM) {
2138 switch (po->tp_version) {
2139 case TPACKET_V2:
2140 off = ph.h2->tp_net;
2141 break;
2142 default:
2143 off = ph.h1->tp_net;
2144 break;
2145 }
2146 } else {
2147 switch (po->tp_version) {
2148 case TPACKET_V2:
2149 off = ph.h2->tp_mac;
2150 break;
2151 default:
2152 off = ph.h1->tp_mac;
2153 break;
2154 }
2155 }
2156 if (unlikely((off < off_min) || (off_max < off)))
2157 return -EINVAL;
2158 data = ph.raw + off;
2159 } else {
2160 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2161 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002162 to_write = tp_len;
2163
2164 if (sock->type == SOCK_DGRAM) {
2165 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2166 NULL, tp_len);
2167 if (unlikely(err < 0))
2168 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002169 } else if (dev->hard_header_len) {
Johann Baudy69e3c752009-05-18 22:11:22 -07002170 /* net device doesn't like empty head */
2171 if (unlikely(tp_len <= dev->hard_header_len)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002172 pr_err("packet size is too short (%d < %d)\n",
2173 tp_len, dev->hard_header_len);
Johann Baudy69e3c752009-05-18 22:11:22 -07002174 return -EINVAL;
2175 }
2176
2177 skb_push(skb, dev->hard_header_len);
2178 err = skb_store_bits(skb, 0, data,
2179 dev->hard_header_len);
2180 if (unlikely(err))
2181 return err;
2182
2183 data += dev->hard_header_len;
2184 to_write -= dev->hard_header_len;
2185 }
2186
Johann Baudy69e3c752009-05-18 22:11:22 -07002187 offset = offset_in_page(data);
2188 len_max = PAGE_SIZE - offset;
2189 len = ((to_write > len_max) ? len_max : to_write);
2190
2191 skb->data_len = to_write;
2192 skb->len += to_write;
2193 skb->truesize += to_write;
2194 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2195
2196 while (likely(to_write)) {
2197 nr_frags = skb_shinfo(skb)->nr_frags;
2198
2199 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002200 pr_err("Packet exceed the number of skb frags(%lu)\n",
2201 MAX_SKB_FRAGS);
Johann Baudy69e3c752009-05-18 22:11:22 -07002202 return -EFAULT;
2203 }
2204
Changli Gao0af55bb2010-12-01 02:52:20 +00002205 page = pgv_to_page(data);
2206 data += len;
Johann Baudy69e3c752009-05-18 22:11:22 -07002207 flush_dcache_page(page);
2208 get_page(page);
Changli Gao0af55bb2010-12-01 02:52:20 +00002209 skb_fill_page_desc(skb, nr_frags, page, offset, len);
Johann Baudy69e3c752009-05-18 22:11:22 -07002210 to_write -= len;
2211 offset = 0;
2212 len_max = PAGE_SIZE;
2213 len = ((to_write > len_max) ? len_max : to_write);
2214 }
2215
2216 return tp_len;
2217}
2218
2219static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2220{
Johann Baudy69e3c752009-05-18 22:11:22 -07002221 struct sk_buff *skb;
2222 struct net_device *dev;
2223 __be16 proto;
David S. Miller09effa62013-08-07 17:11:00 -07002224 int err, reserve = 0;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002225 void *ph;
Steffen Hurrle342dfc32014-01-17 22:53:15 +01002226 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002227 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
Johann Baudy69e3c752009-05-18 22:11:22 -07002228 int tp_len, size_max;
2229 unsigned char *addr;
2230 int len_sum = 0;
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +00002231 int status = TP_STATUS_AVAILABLE;
Herbert Xuae641942011-11-18 02:20:04 +00002232 int hlen, tlen;
Johann Baudy69e3c752009-05-18 22:11:22 -07002233
Johann Baudy69e3c752009-05-18 22:11:22 -07002234 mutex_lock(&po->pg_vec_lock);
2235
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002236 if (likely(saddr == NULL)) {
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002237 dev = packet_cached_dev_get(po);
Johann Baudy69e3c752009-05-18 22:11:22 -07002238 proto = po->num;
2239 addr = NULL;
2240 } else {
2241 err = -EINVAL;
2242 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2243 goto out;
2244 if (msg->msg_namelen < (saddr->sll_halen
2245 + offsetof(struct sockaddr_ll,
2246 sll_addr)))
2247 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07002248 proto = saddr->sll_protocol;
2249 addr = saddr->sll_addr;
Ben Greear827d9782011-06-01 07:18:53 +00002250 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
Johann Baudy69e3c752009-05-18 22:11:22 -07002251 }
2252
Johann Baudy69e3c752009-05-18 22:11:22 -07002253 err = -ENXIO;
2254 if (unlikely(dev == NULL))
2255 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07002256 err = -ENETDOWN;
2257 if (unlikely(!(dev->flags & IFF_UP)))
2258 goto out_put;
2259
Daniel Borkmann52f14542014-02-28 02:22:06 +01002260 reserve = dev->hard_header_len + VLAN_HLEN;
Johann Baudy69e3c752009-05-18 22:11:22 -07002261 size_max = po->tx_ring.frame_size
Gabor Gombasb5dd8842009-10-29 03:19:11 -07002262 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
Johann Baudy69e3c752009-05-18 22:11:22 -07002263
David S. Miller09effa62013-08-07 17:11:00 -07002264 if (size_max > dev->mtu + reserve)
2265 size_max = dev->mtu + reserve;
2266
Johann Baudy69e3c752009-05-18 22:11:22 -07002267 do {
2268 ph = packet_current_frame(po, &po->tx_ring,
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002269 TP_STATUS_SEND_REQUEST);
Johann Baudy69e3c752009-05-18 22:11:22 -07002270 if (unlikely(ph == NULL)) {
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002271 if (need_wait && need_resched())
2272 schedule();
Johann Baudy69e3c752009-05-18 22:11:22 -07002273 continue;
2274 }
2275
2276 status = TP_STATUS_SEND_REQUEST;
Herbert Xuae641942011-11-18 02:20:04 +00002277 hlen = LL_RESERVED_SPACE(dev);
2278 tlen = dev->needed_tailroom;
Johann Baudy69e3c752009-05-18 22:11:22 -07002279 skb = sock_alloc_send_skb(&po->sk,
Herbert Xuae641942011-11-18 02:20:04 +00002280 hlen + tlen + sizeof(struct sockaddr_ll),
Johann Baudy69e3c752009-05-18 22:11:22 -07002281 0, &err);
2282
2283 if (unlikely(skb == NULL))
2284 goto out_status;
2285
2286 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
Daniel Borkmann52f14542014-02-28 02:22:06 +01002287 addr, hlen);
2288 if (tp_len > dev->mtu + dev->hard_header_len) {
2289 struct ethhdr *ehdr;
2290 /* Earlier code assumed this would be a VLAN pkt,
2291 * double-check this now that we have the actual
2292 * packet in hand.
2293 */
Johann Baudy69e3c752009-05-18 22:11:22 -07002294
Daniel Borkmann52f14542014-02-28 02:22:06 +01002295 skb_reset_mac_header(skb);
2296 ehdr = eth_hdr(skb);
2297 if (ehdr->h_proto != htons(ETH_P_8021Q))
2298 tp_len = -EMSGSIZE;
2299 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002300 if (unlikely(tp_len < 0)) {
2301 if (po->tp_loss) {
2302 __packet_set_status(po, ph,
2303 TP_STATUS_AVAILABLE);
2304 packet_increment_head(&po->tx_ring);
2305 kfree_skb(skb);
2306 continue;
2307 } else {
2308 status = TP_STATUS_WRONG_FORMAT;
2309 err = tp_len;
2310 goto out_status;
2311 }
2312 }
2313
Daniel Borkmann0fd5d572014-02-16 15:55:22 +01002314 packet_pick_tx_queue(dev, skb);
2315
Johann Baudy69e3c752009-05-18 22:11:22 -07002316 skb->destructor = tpacket_destruct_skb;
2317 __packet_set_status(po, ph, TP_STATUS_SENDING);
Daniel Borkmannb0138402014-01-15 16:25:36 +01002318 packet_inc_pending(&po->tx_ring);
Johann Baudy69e3c752009-05-18 22:11:22 -07002319
2320 status = TP_STATUS_SEND_REQUEST;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002321 err = po->xmit(skb);
Jarek Poplawskieb70df12010-01-10 22:04:19 +00002322 if (unlikely(err > 0)) {
2323 err = net_xmit_errno(err);
2324 if (err && __packet_get_status(po, ph) ==
2325 TP_STATUS_AVAILABLE) {
2326 /* skb was destructed already */
2327 skb = NULL;
2328 goto out_status;
2329 }
2330 /*
2331 * skb was dropped but not destructed yet;
2332 * let's treat it like congestion or err < 0
2333 */
2334 err = 0;
2335 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002336 packet_increment_head(&po->tx_ring);
2337 len_sum += tp_len;
Daniel Borkmannb0138402014-01-15 16:25:36 +01002338 } while (likely((ph != NULL) ||
2339 /* Note: packet_read_pending() might be slow if we have
2340 * to call it as it's per_cpu variable, but in fast-path
2341 * we already short-circuit the loop with the first
2342 * condition, and luckily don't have to go that path
2343 * anyway.
2344 */
2345 (need_wait && packet_read_pending(&po->tx_ring))));
Johann Baudy69e3c752009-05-18 22:11:22 -07002346
2347 err = len_sum;
2348 goto out_put;
2349
Johann Baudy69e3c752009-05-18 22:11:22 -07002350out_status:
2351 __packet_set_status(po, ph, status);
2352 kfree_skb(skb);
2353out_put:
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002354 dev_put(dev);
Johann Baudy69e3c752009-05-18 22:11:22 -07002355out:
2356 mutex_unlock(&po->pg_vec_lock);
2357 return err;
2358}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359
Olof Johanssoneea49cc92011-11-02 11:00:49 +00002360static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2361 size_t reserve, size_t len,
2362 size_t linear, int noblock,
2363 int *err)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002364{
2365 struct sk_buff *skb;
2366
2367 /* Under a page? Don't bother with paged skb. */
2368 if (prepad + len < PAGE_SIZE || !linear)
2369 linear = len;
2370
2371 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
Eric Dumazet28d64272013-08-08 14:38:47 -07002372 err, 0);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002373 if (!skb)
2374 return NULL;
2375
2376 skb_reserve(skb, reserve);
2377 skb_put(skb, linear);
2378 skb->data_len = len - linear;
2379 skb->len += len - linear;
2380
2381 return skb;
2382}
2383
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002384static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385{
2386 struct sock *sk = sock->sk;
Steffen Hurrle342dfc32014-01-17 22:53:15 +01002387 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388 struct sk_buff *skb;
2389 struct net_device *dev;
Al Viro0e11c912006-11-08 00:26:29 -08002390 __be16 proto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 unsigned char *addr;
Ben Greear827d9782011-06-01 07:18:53 +00002392 int err, reserve = 0;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002393 struct virtio_net_hdr vnet_hdr = { 0 };
2394 int offset = 0;
2395 int vnet_hdr_len;
2396 struct packet_sock *po = pkt_sk(sk);
2397 unsigned short gso_type = 0;
Herbert Xuae641942011-11-18 02:20:04 +00002398 int hlen, tlen;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002399 int extra_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002400
2401 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002402 * Get and verify the address.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002404
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002405 if (likely(saddr == NULL)) {
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002406 dev = packet_cached_dev_get(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 proto = po->num;
2408 addr = NULL;
2409 } else {
2410 err = -EINVAL;
2411 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2412 goto out;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002413 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2414 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 proto = saddr->sll_protocol;
2416 addr = saddr->sll_addr;
Ben Greear827d9782011-06-01 07:18:53 +00002417 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 }
2419
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 err = -ENXIO;
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002421 if (unlikely(dev == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 goto out_unlock;
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002423 err = -ENETDOWN;
2424 if (unlikely(!(dev->flags & IFF_UP)))
2425 goto out_unlock;
2426
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 if (sock->type == SOCK_RAW)
2428 reserve = dev->hard_header_len;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002429 if (po->has_vnet_hdr) {
2430 vnet_hdr_len = sizeof(vnet_hdr);
2431
2432 err = -EINVAL;
2433 if (len < vnet_hdr_len)
2434 goto out_unlock;
2435
2436 len -= vnet_hdr_len;
2437
2438 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2439 vnet_hdr_len);
2440 if (err < 0)
2441 goto out_unlock;
2442
2443 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2444 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2445 vnet_hdr.hdr_len))
2446 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2447 vnet_hdr.csum_offset + 2;
2448
2449 err = -EINVAL;
2450 if (vnet_hdr.hdr_len > len)
2451 goto out_unlock;
2452
2453 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2454 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2455 case VIRTIO_NET_HDR_GSO_TCPV4:
2456 gso_type = SKB_GSO_TCPV4;
2457 break;
2458 case VIRTIO_NET_HDR_GSO_TCPV6:
2459 gso_type = SKB_GSO_TCPV6;
2460 break;
2461 case VIRTIO_NET_HDR_GSO_UDP:
2462 gso_type = SKB_GSO_UDP;
2463 break;
2464 default:
2465 goto out_unlock;
2466 }
2467
2468 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2469 gso_type |= SKB_GSO_TCP_ECN;
2470
2471 if (vnet_hdr.gso_size == 0)
2472 goto out_unlock;
2473
2474 }
2475 }
2476
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002477 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2478 if (!netif_supports_nofcs(dev)) {
2479 err = -EPROTONOSUPPORT;
2480 goto out_unlock;
2481 }
2482 extra_len = 4; /* We're doing our own CRC */
2483 }
2484
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 err = -EMSGSIZE;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002486 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 goto out_unlock;
2488
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002489 err = -ENOBUFS;
Herbert Xuae641942011-11-18 02:20:04 +00002490 hlen = LL_RESERVED_SPACE(dev);
2491 tlen = dev->needed_tailroom;
2492 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002493 msg->msg_flags & MSG_DONTWAIT, &err);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002494 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 goto out_unlock;
2496
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002497 skb_set_network_header(skb, reserve);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498
Stephen Hemminger0c4e8582007-10-09 01:36:32 -07002499 err = -EINVAL;
2500 if (sock->type == SOCK_DGRAM &&
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002501 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
Stephen Hemminger0c4e8582007-10-09 01:36:32 -07002502 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503
2504 /* Returns -EFAULT on error */
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002505 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 if (err)
2507 goto out_free;
Daniel Borkmannbf84a012013-04-14 08:08:13 +00002508
2509 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002511 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
David S. Miller09effa62013-08-07 17:11:00 -07002512 /* Earlier code assumed this would be a VLAN pkt,
2513 * double-check this now that we have the actual
2514 * packet in hand.
2515 */
2516 struct ethhdr *ehdr;
2517 skb_reset_mac_header(skb);
2518 ehdr = eth_hdr(skb);
2519 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2520 err = -EMSGSIZE;
2521 goto out_free;
2522 }
Ben Greear57f89bf2011-02-11 09:35:18 +00002523 }
2524
David S. Miller09effa62013-08-07 17:11:00 -07002525 skb->protocol = proto;
2526 skb->dev = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 skb->priority = sk->sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00002528 skb->mark = sk->sk_mark;
Daniel Borkmann0fd5d572014-02-16 15:55:22 +01002529
2530 packet_pick_tx_queue(dev, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002532 if (po->has_vnet_hdr) {
2533 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2534 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2535 vnet_hdr.csum_offset)) {
2536 err = -EINVAL;
2537 goto out_free;
2538 }
2539 }
2540
2541 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2542 skb_shinfo(skb)->gso_type = gso_type;
2543
2544 /* Header must be checked, and gso_segs computed. */
2545 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2546 skb_shinfo(skb)->gso_segs = 0;
2547
2548 len += vnet_hdr_len;
2549 }
2550
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002551 if (!packet_use_direct_xmit(po))
2552 skb_probe_transport_header(skb, reserve);
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002553 if (unlikely(extra_len == 4))
2554 skb->no_fcs = 1;
2555
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002556 err = po->xmit(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2558 goto out_unlock;
2559
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002560 dev_put(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002562 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563
2564out_free:
2565 kfree_skb(skb);
2566out_unlock:
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002567 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 dev_put(dev);
2569out:
2570 return err;
2571}
2572
Johann Baudy69e3c752009-05-18 22:11:22 -07002573static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2574 struct msghdr *msg, size_t len)
2575{
Johann Baudy69e3c752009-05-18 22:11:22 -07002576 struct sock *sk = sock->sk;
2577 struct packet_sock *po = pkt_sk(sk);
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002578
Johann Baudy69e3c752009-05-18 22:11:22 -07002579 if (po->tx_ring.pg_vec)
2580 return tpacket_snd(po, msg);
2581 else
Johann Baudy69e3c752009-05-18 22:11:22 -07002582 return packet_snd(sock, msg, len);
2583}
2584
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585/*
2586 * Close a PACKET socket. This is fairly simple. We immediately go
2587 * to 'closed' state and remove our protocol entry in the device list.
2588 */
2589
2590static int packet_release(struct socket *sock)
2591{
2592 struct sock *sk = sock->sk;
2593 struct packet_sock *po;
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08002594 struct net *net;
chetan lokef6fb8f102011-08-19 10:18:16 +00002595 union tpacket_req_u req_u;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596
2597 if (!sk)
2598 return 0;
2599
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002600 net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 po = pkt_sk(sk);
2602
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002603 mutex_lock(&net->packet.sklist_lock);
stephen hemminger808f5112010-02-22 07:57:18 +00002604 sk_del_node_init_rcu(sk);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002605 mutex_unlock(&net->packet.sklist_lock);
2606
2607 preempt_disable();
Eric Dumazet920de802008-11-24 00:09:29 -08002608 sock_prot_inuse_add(net, sk->sk_prot, -1);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002609 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610
stephen hemminger808f5112010-02-22 07:57:18 +00002611 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07002612 unregister_prot_hook(sk, false);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002613 packet_cached_dev_reset(po);
2614
Ben Greear160ff182011-06-01 07:18:52 +00002615 if (po->prot_hook.dev) {
2616 dev_put(po->prot_hook.dev);
2617 po->prot_hook.dev = NULL;
2618 }
stephen hemminger808f5112010-02-22 07:57:18 +00002619 spin_unlock(&po->bind_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 packet_flush_mclist(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
Phil Sutter9665d5d2013-02-01 07:21:41 +00002623 if (po->rx_ring.pg_vec) {
2624 memset(&req_u, 0, sizeof(req_u));
chetan lokef6fb8f102011-08-19 10:18:16 +00002625 packet_set_ring(sk, &req_u, 1, 0);
Phil Sutter9665d5d2013-02-01 07:21:41 +00002626 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002627
Phil Sutter9665d5d2013-02-01 07:21:41 +00002628 if (po->tx_ring.pg_vec) {
2629 memset(&req_u, 0, sizeof(req_u));
chetan lokef6fb8f102011-08-19 10:18:16 +00002630 packet_set_ring(sk, &req_u, 1, 1);
Phil Sutter9665d5d2013-02-01 07:21:41 +00002631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632
David S. Millerdc99f602011-07-05 01:45:05 -07002633 fanout_release(sk);
2634
stephen hemminger808f5112010-02-22 07:57:18 +00002635 synchronize_net();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 /*
2637 * Now the socket is dead. No more input will appear.
2638 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 sock_orphan(sk);
2640 sock->sk = NULL;
2641
2642 /* Purge queues */
2643
2644 skb_queue_purge(&sk->sk_receive_queue);
Daniel Borkmannb0138402014-01-15 16:25:36 +01002645 packet_free_pending(po);
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08002646 sk_refcnt_debug_release(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647
2648 sock_put(sk);
2649 return 0;
2650}
2651
2652/*
2653 * Attach a packet hook.
2654 */
2655
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002656static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657{
2658 struct packet_sock *po = pkt_sk(sk);
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002659 const struct net_device *dev_curr;
2660 __be16 proto_curr;
2661 bool need_rehook;
David S. Millerdc99f602011-07-05 01:45:05 -07002662
Wei Yongjunaef950b2011-12-27 22:32:41 -05002663 if (po->fanout) {
2664 if (dev)
2665 dev_put(dev);
2666
David S. Millerdc99f602011-07-05 01:45:05 -07002667 return -EINVAL;
Wei Yongjunaef950b2011-12-27 22:32:41 -05002668 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
2670 lock_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 spin_lock(&po->bind_lock);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002672
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002673 proto_curr = po->prot_hook.type;
2674 dev_curr = po->prot_hook.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002676 need_rehook = proto_curr != proto || dev_curr != dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002678 if (need_rehook) {
2679 unregister_prot_hook(sk, true);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002680
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002681 po->num = proto;
2682 po->prot_hook.type = proto;
2683
2684 if (po->prot_hook.dev)
2685 dev_put(po->prot_hook.dev);
2686
2687 po->prot_hook.dev = dev;
2688
2689 po->ifindex = dev ? dev->ifindex : 0;
2690 packet_cached_dev_assign(po, dev);
2691 }
2692
2693 if (proto == 0 || !need_rehook)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 goto out_unlock;
2695
Urs Thuermannbe85d4a2007-11-12 21:05:20 -08002696 if (!dev || (dev->flags & IFF_UP)) {
David S. Millerce06b032011-07-04 01:44:29 -07002697 register_prot_hook(sk);
Urs Thuermannbe85d4a2007-11-12 21:05:20 -08002698 } else {
2699 sk->sk_err = ENETDOWN;
2700 if (!sock_flag(sk, SOCK_DEAD))
2701 sk->sk_error_report(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 }
2703
2704out_unlock:
2705 spin_unlock(&po->bind_lock);
2706 release_sock(sk);
2707 return 0;
2708}
2709
2710/*
2711 * Bind a packet socket to a device
2712 */
2713
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002714static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2715 int addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002717 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 char name[15];
2719 struct net_device *dev;
2720 int err = -ENODEV;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002721
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 /*
2723 * Check legality
2724 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002725
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08002726 if (addr_len != sizeof(struct sockaddr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002728 strlcpy(name, uaddr->sa_data, sizeof(name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002730 dev = dev_get_by_name(sock_net(sk), name);
Ben Greear160ff182011-06-01 07:18:52 +00002731 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 return err;
2734}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
2736static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2737{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002738 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2739 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740 struct net_device *dev = NULL;
2741 int err;
2742
2743
2744 /*
2745 * Check legality
2746 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002747
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 if (addr_len < sizeof(struct sockaddr_ll))
2749 return -EINVAL;
2750 if (sll->sll_family != AF_PACKET)
2751 return -EINVAL;
2752
2753 if (sll->sll_ifindex) {
2754 err = -ENODEV;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002755 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756 if (dev == NULL)
2757 goto out;
2758 }
2759 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760
2761out:
2762 return err;
2763}
2764
2765static struct proto packet_proto = {
2766 .name = "PACKET",
2767 .owner = THIS_MODULE,
2768 .obj_size = sizeof(struct packet_sock),
2769};
2770
2771/*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002772 * Create a packet of type SOCK_PACKET.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773 */
2774
Eric Paris3f378b62009-11-05 22:18:14 -08002775static int packet_create(struct net *net, struct socket *sock, int protocol,
2776 int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777{
2778 struct sock *sk;
2779 struct packet_sock *po;
Al Viro0e11c912006-11-08 00:26:29 -08002780 __be16 proto = (__force __be16)protocol; /* weird, but documented */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 int err;
2782
Eric W. Biedermandf008c92012-11-16 03:03:07 +00002783 if (!ns_capable(net->user_ns, CAP_NET_RAW))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 return -EPERM;
David S. Millerbe020972007-05-29 13:16:31 -07002785 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2786 sock->type != SOCK_PACKET)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 return -ESOCKTNOSUPPORT;
2788
2789 sock->state = SS_UNCONNECTED;
2790
2791 err = -ENOBUFS;
Pavel Emelyanov6257ff22007-11-01 00:39:31 -07002792 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793 if (sk == NULL)
2794 goto out;
2795
2796 sock->ops = &packet_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 if (sock->type == SOCK_PACKET)
2798 sock->ops = &packet_ops_spkt;
David S. Millerbe020972007-05-29 13:16:31 -07002799
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 sock_init_data(sock, sk);
2801
2802 po = pkt_sk(sk);
2803 sk->sk_family = PF_PACKET;
Al Viro0e11c912006-11-08 00:26:29 -08002804 po->num = proto;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002805 po->xmit = dev_queue_xmit;
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002806
Daniel Borkmannb0138402014-01-15 16:25:36 +01002807 err = packet_alloc_pending(po);
2808 if (err)
2809 goto out2;
2810
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002811 packet_cached_dev_reset(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812
2813 sk->sk_destruct = packet_sock_destruct;
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08002814 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815
2816 /*
2817 * Attach a protocol block
2818 */
2819
2820 spin_lock_init(&po->bind_lock);
Herbert Xu905db442009-01-30 14:12:06 -08002821 mutex_init(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 po->prot_hook.func = packet_rcv;
David S. Millerbe020972007-05-29 13:16:31 -07002823
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 if (sock->type == SOCK_PACKET)
2825 po->prot_hook.func = packet_rcv_spkt;
David S. Millerbe020972007-05-29 13:16:31 -07002826
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 po->prot_hook.af_packet_priv = sk;
2828
Al Viro0e11c912006-11-08 00:26:29 -08002829 if (proto) {
2830 po->prot_hook.type = proto;
David S. Millerce06b032011-07-04 01:44:29 -07002831 register_prot_hook(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 }
2833
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002834 mutex_lock(&net->packet.sklist_lock);
stephen hemminger808f5112010-02-22 07:57:18 +00002835 sk_add_node_rcu(sk, &net->packet.sklist);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002836 mutex_unlock(&net->packet.sklist_lock);
2837
2838 preempt_disable();
Eric Dumazet36804532008-11-19 14:25:35 -08002839 sock_prot_inuse_add(net, &packet_proto, 1);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002840 preempt_enable();
stephen hemminger808f5112010-02-22 07:57:18 +00002841
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002842 return 0;
Daniel Borkmannb0138402014-01-15 16:25:36 +01002843out2:
2844 sk_free(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845out:
2846 return err;
2847}
2848
2849/*
2850 * Pull a packet from our receive queue and hand it to the user.
2851 * If necessary we block.
2852 */
2853
2854static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2855 struct msghdr *msg, size_t len, int flags)
2856{
2857 struct sock *sk = sock->sk;
2858 struct sk_buff *skb;
2859 int copied, err;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002860 int vnet_hdr_len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861
2862 err = -EINVAL;
Richard Cochraned85b562010-04-07 22:41:28 +00002863 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864 goto out;
2865
2866#if 0
2867 /* What error should we return now? EUNATTACH? */
2868 if (pkt_sk(sk)->ifindex < 0)
2869 return -ENODEV;
2870#endif
2871
Richard Cochraned85b562010-04-07 22:41:28 +00002872 if (flags & MSG_ERRQUEUE) {
Richard Cochrancb820f82013-07-19 19:40:09 +02002873 err = sock_recv_errqueue(sk, msg, len,
2874 SOL_PACKET, PACKET_TX_TIMESTAMP);
Richard Cochraned85b562010-04-07 22:41:28 +00002875 goto out;
2876 }
2877
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879 * Call the generic datagram receiver. This handles all sorts
2880 * of horrible races and re-entrancy so we can forget about it
2881 * in the protocol layers.
2882 *
2883 * Now it will return ENETDOWN, if device have just gone down,
2884 * but then it will block.
2885 */
2886
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002887 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888
2889 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002890 * An error occurred so return it. Because skb_recv_datagram()
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 * handles the blocking we don't see and worry about blocking
2892 * retries.
2893 */
2894
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08002895 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002896 goto out;
2897
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002898 if (pkt_sk(sk)->has_vnet_hdr) {
2899 struct virtio_net_hdr vnet_hdr = { 0 };
2900
2901 err = -EINVAL;
2902 vnet_hdr_len = sizeof(vnet_hdr);
Mariusz Kozlowski1f18b712010-11-08 11:58:45 +00002903 if (len < vnet_hdr_len)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002904 goto out_free;
2905
Mariusz Kozlowski1f18b712010-11-08 11:58:45 +00002906 len -= vnet_hdr_len;
2907
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002908 if (skb_is_gso(skb)) {
2909 struct skb_shared_info *sinfo = skb_shinfo(skb);
2910
2911 /* This is a hint as to how much should be linear. */
2912 vnet_hdr.hdr_len = skb_headlen(skb);
2913 vnet_hdr.gso_size = sinfo->gso_size;
2914 if (sinfo->gso_type & SKB_GSO_TCPV4)
2915 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2916 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2917 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2918 else if (sinfo->gso_type & SKB_GSO_UDP)
2919 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2920 else if (sinfo->gso_type & SKB_GSO_FCOE)
2921 goto out_free;
2922 else
2923 BUG();
2924 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2925 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2926 } else
2927 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2928
2929 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2930 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Michał Mirosław55508d62010-12-14 15:24:08 +00002931 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002932 vnet_hdr.csum_offset = skb->csum_offset;
Jason Wang10a8d942011-06-10 00:56:17 +00002933 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2934 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002935 } /* else everything is zero */
2936
2937 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2938 vnet_hdr_len);
2939 if (err < 0)
2940 goto out_free;
2941 }
2942
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01002943 /* You lose any data beyond the buffer you gave. If it worries
2944 * a user program they can ask the device for its MTU
2945 * anyway.
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002946 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 copied = skb->len;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002948 if (copied > len) {
2949 copied = len;
2950 msg->msg_flags |= MSG_TRUNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951 }
2952
2953 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2954 if (err)
2955 goto out_free;
2956
Neil Horman3b885782009-10-12 13:26:31 -07002957 sock_recv_ts_and_drops(msg, sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01002959 if (msg->msg_name) {
2960 /* If the address length field is there to be filled
2961 * in, we fill it in now.
2962 */
2963 if (sock->type == SOCK_PACKET) {
Steffen Hurrle342dfc32014-01-17 22:53:15 +01002964 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01002965 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2966 } else {
2967 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2968 msg->msg_namelen = sll->sll_halen +
2969 offsetof(struct sockaddr_ll, sll_addr);
2970 }
Herbert Xuffbc6112007-02-04 23:33:10 -08002971 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2972 msg->msg_namelen);
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01002973 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002974
Herbert Xu8dc41942007-02-04 23:31:32 -08002975 if (pkt_sk(sk)->auxdata) {
Herbert Xuffbc6112007-02-04 23:33:10 -08002976 struct tpacket_auxdata aux;
2977
2978 aux.tp_status = TP_STATUS_USER;
2979 if (skb->ip_summed == CHECKSUM_PARTIAL)
2980 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2981 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2982 aux.tp_snaplen = skb->len;
2983 aux.tp_mac = 0;
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03002984 aux.tp_net = skb_network_offset(skb);
Ben Greeara3bcc232011-06-01 06:49:10 +00002985 if (vlan_tx_tag_present(skb)) {
2986 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09002987 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
2988 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
Ben Greeara3bcc232011-06-01 06:49:10 +00002989 } else {
2990 aux.tp_vlan_tci = 0;
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09002991 aux.tp_vlan_tpid = 0;
Ben Greeara3bcc232011-06-01 06:49:10 +00002992 }
Herbert Xuffbc6112007-02-04 23:33:10 -08002993 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
Herbert Xu8dc41942007-02-04 23:31:32 -08002994 }
2995
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 /*
2997 * Free or return the buffer as appropriate. Again this
2998 * hides all the races and re-entrancy issues from us.
2999 */
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003000 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001
3002out_free:
3003 skb_free_datagram(sk, skb);
3004out:
3005 return err;
3006}
3007
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3009 int *uaddr_len, int peer)
3010{
3011 struct net_device *dev;
3012 struct sock *sk = sock->sk;
3013
3014 if (peer)
3015 return -EOPNOTSUPP;
3016
3017 uaddr->sa_family = AF_PACKET;
Daniel Borkmann2dc85bf2013-06-12 16:02:27 +02003018 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
Eric Dumazet654d1f82009-11-02 10:43:32 +01003019 rcu_read_lock();
3020 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3021 if (dev)
Daniel Borkmann2dc85bf2013-06-12 16:02:27 +02003022 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
Eric Dumazet654d1f82009-11-02 10:43:32 +01003023 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 *uaddr_len = sizeof(*uaddr);
3025
3026 return 0;
3027}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003028
3029static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3030 int *uaddr_len, int peer)
3031{
3032 struct net_device *dev;
3033 struct sock *sk = sock->sk;
3034 struct packet_sock *po = pkt_sk(sk);
Cyrill Gorcunov13cfa972009-11-08 05:51:19 +00003035 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036
3037 if (peer)
3038 return -EOPNOTSUPP;
3039
3040 sll->sll_family = AF_PACKET;
3041 sll->sll_ifindex = po->ifindex;
3042 sll->sll_protocol = po->num;
Vasiliy Kulikov67286642010-11-10 12:09:10 -08003043 sll->sll_pkttype = 0;
Eric Dumazet654d1f82009-11-02 10:43:32 +01003044 rcu_read_lock();
3045 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046 if (dev) {
3047 sll->sll_hatype = dev->type;
3048 sll->sll_halen = dev->addr_len;
3049 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 } else {
3051 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3052 sll->sll_halen = 0;
3053 }
Eric Dumazet654d1f82009-11-02 10:43:32 +01003054 rcu_read_unlock();
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003055 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056
3057 return 0;
3058}
3059
Wang Chen2aeb0b82008-07-14 20:49:46 -07003060static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3061 int what)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003062{
3063 switch (i->type) {
3064 case PACKET_MR_MULTICAST:
Jiri Pirko11625632010-03-02 20:40:01 +00003065 if (i->alen != dev->addr_len)
3066 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067 if (what > 0)
Jiri Pirko22bedad32010-04-01 21:22:57 +00003068 return dev_mc_add(dev, i->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 else
Jiri Pirko22bedad32010-04-01 21:22:57 +00003070 return dev_mc_del(dev, i->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 break;
3072 case PACKET_MR_PROMISC:
Wang Chen2aeb0b82008-07-14 20:49:46 -07003073 return dev_set_promiscuity(dev, what);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 break;
3075 case PACKET_MR_ALLMULTI:
Wang Chen2aeb0b82008-07-14 20:49:46 -07003076 return dev_set_allmulti(dev, what);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077 break;
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003078 case PACKET_MR_UNICAST:
Jiri Pirko11625632010-03-02 20:40:01 +00003079 if (i->alen != dev->addr_len)
3080 return -EINVAL;
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003081 if (what > 0)
Jiri Pirkoa748ee22010-04-01 21:22:09 +00003082 return dev_uc_add(dev, i->addr);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003083 else
Jiri Pirkoa748ee22010-04-01 21:22:09 +00003084 return dev_uc_del(dev, i->addr);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003085 break;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003086 default:
3087 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088 }
Wang Chen2aeb0b82008-07-14 20:49:46 -07003089 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090}
3091
3092static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3093{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003094 for ( ; i; i = i->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 if (i->ifindex == dev->ifindex)
3096 packet_dev_mc(dev, i, what);
3097 }
3098}
3099
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003100static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101{
3102 struct packet_sock *po = pkt_sk(sk);
3103 struct packet_mclist *ml, *i;
3104 struct net_device *dev;
3105 int err;
3106
3107 rtnl_lock();
3108
3109 err = -ENODEV;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09003110 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 if (!dev)
3112 goto done;
3113
3114 err = -EINVAL;
Jiri Pirko11625632010-03-02 20:40:01 +00003115 if (mreq->mr_alen > dev->addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116 goto done;
3117
3118 err = -ENOBUFS;
Kris Katterjohn8b3a7002006-01-11 15:56:43 -08003119 i = kmalloc(sizeof(*i), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 if (i == NULL)
3121 goto done;
3122
3123 err = 0;
3124 for (ml = po->mclist; ml; ml = ml->next) {
3125 if (ml->ifindex == mreq->mr_ifindex &&
3126 ml->type == mreq->mr_type &&
3127 ml->alen == mreq->mr_alen &&
3128 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3129 ml->count++;
3130 /* Free the new element ... */
3131 kfree(i);
3132 goto done;
3133 }
3134 }
3135
3136 i->type = mreq->mr_type;
3137 i->ifindex = mreq->mr_ifindex;
3138 i->alen = mreq->mr_alen;
3139 memcpy(i->addr, mreq->mr_address, i->alen);
3140 i->count = 1;
3141 i->next = po->mclist;
3142 po->mclist = i;
Wang Chen2aeb0b82008-07-14 20:49:46 -07003143 err = packet_dev_mc(dev, i, 1);
3144 if (err) {
3145 po->mclist = i->next;
3146 kfree(i);
3147 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148
3149done:
3150 rtnl_unlock();
3151 return err;
3152}
3153
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003154static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155{
3156 struct packet_mclist *ml, **mlp;
3157
3158 rtnl_lock();
3159
3160 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3161 if (ml->ifindex == mreq->mr_ifindex &&
3162 ml->type == mreq->mr_type &&
3163 ml->alen == mreq->mr_alen &&
3164 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3165 if (--ml->count == 0) {
3166 struct net_device *dev;
3167 *mlp = ml->next;
Eric Dumazetad959e72009-10-16 06:38:46 +00003168 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3169 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170 packet_dev_mc(dev, ml, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 kfree(ml);
3172 }
3173 rtnl_unlock();
3174 return 0;
3175 }
3176 }
3177 rtnl_unlock();
3178 return -EADDRNOTAVAIL;
3179}
3180
3181static void packet_flush_mclist(struct sock *sk)
3182{
3183 struct packet_sock *po = pkt_sk(sk);
3184 struct packet_mclist *ml;
3185
3186 if (!po->mclist)
3187 return;
3188
3189 rtnl_lock();
3190 while ((ml = po->mclist) != NULL) {
3191 struct net_device *dev;
3192
3193 po->mclist = ml->next;
Eric Dumazetad959e72009-10-16 06:38:46 +00003194 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3195 if (dev != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 packet_dev_mc(dev, ml, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 kfree(ml);
3198 }
3199 rtnl_unlock();
3200}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201
3202static int
David S. Millerb7058842009-09-30 16:12:20 -07003203packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204{
3205 struct sock *sk = sock->sk;
Herbert Xu8dc41942007-02-04 23:31:32 -08003206 struct packet_sock *po = pkt_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 int ret;
3208
3209 if (level != SOL_PACKET)
3210 return -ENOPROTOOPT;
3211
Johann Baudy69e3c752009-05-18 22:11:22 -07003212 switch (optname) {
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003213 case PACKET_ADD_MEMBERSHIP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 case PACKET_DROP_MEMBERSHIP:
3215 {
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003216 struct packet_mreq_max mreq;
3217 int len = optlen;
3218 memset(&mreq, 0, sizeof(mreq));
3219 if (len < sizeof(struct packet_mreq))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003220 return -EINVAL;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003221 if (len > sizeof(mreq))
3222 len = sizeof(mreq);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003223 if (copy_from_user(&mreq, optval, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 return -EFAULT;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003225 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3226 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227 if (optname == PACKET_ADD_MEMBERSHIP)
3228 ret = packet_mc_add(sk, &mreq);
3229 else
3230 ret = packet_mc_drop(sk, &mreq);
3231 return ret;
3232 }
David S. Millera2efcfa2007-05-29 13:12:50 -07003233
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234 case PACKET_RX_RING:
Johann Baudy69e3c752009-05-18 22:11:22 -07003235 case PACKET_TX_RING:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003236 {
chetan lokef6fb8f102011-08-19 10:18:16 +00003237 union tpacket_req_u req_u;
3238 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239
chetan lokef6fb8f102011-08-19 10:18:16 +00003240 switch (po->tp_version) {
3241 case TPACKET_V1:
3242 case TPACKET_V2:
3243 len = sizeof(req_u.req);
3244 break;
3245 case TPACKET_V3:
3246 default:
3247 len = sizeof(req_u.req3);
3248 break;
3249 }
3250 if (optlen < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251 return -EINVAL;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003252 if (pkt_sk(sk)->has_vnet_hdr)
3253 return -EINVAL;
chetan lokef6fb8f102011-08-19 10:18:16 +00003254 if (copy_from_user(&req_u.req, optval, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 return -EFAULT;
chetan lokef6fb8f102011-08-19 10:18:16 +00003256 return packet_set_ring(sk, &req_u, 0,
3257 optname == PACKET_TX_RING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 }
3259 case PACKET_COPY_THRESH:
3260 {
3261 int val;
3262
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003263 if (optlen != sizeof(val))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003265 if (copy_from_user(&val, optval, sizeof(val)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 return -EFAULT;
3267
3268 pkt_sk(sk)->copy_thresh = val;
3269 return 0;
3270 }
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003271 case PACKET_VERSION:
3272 {
3273 int val;
3274
3275 if (optlen != sizeof(val))
3276 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07003277 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003278 return -EBUSY;
3279 if (copy_from_user(&val, optval, sizeof(val)))
3280 return -EFAULT;
3281 switch (val) {
3282 case TPACKET_V1:
3283 case TPACKET_V2:
chetan lokef6fb8f102011-08-19 10:18:16 +00003284 case TPACKET_V3:
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003285 po->tp_version = val;
3286 return 0;
3287 default:
3288 return -EINVAL;
3289 }
3290 }
Patrick McHardy8913336a2008-07-18 18:05:19 -07003291 case PACKET_RESERVE:
3292 {
3293 unsigned int val;
3294
3295 if (optlen != sizeof(val))
3296 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07003297 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
Patrick McHardy8913336a2008-07-18 18:05:19 -07003298 return -EBUSY;
3299 if (copy_from_user(&val, optval, sizeof(val)))
3300 return -EFAULT;
3301 po->tp_reserve = val;
3302 return 0;
3303 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003304 case PACKET_LOSS:
3305 {
3306 unsigned int val;
3307
3308 if (optlen != sizeof(val))
3309 return -EINVAL;
3310 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3311 return -EBUSY;
3312 if (copy_from_user(&val, optval, sizeof(val)))
3313 return -EFAULT;
3314 po->tp_loss = !!val;
3315 return 0;
3316 }
Herbert Xu8dc41942007-02-04 23:31:32 -08003317 case PACKET_AUXDATA:
3318 {
3319 int val;
3320
3321 if (optlen < sizeof(val))
3322 return -EINVAL;
3323 if (copy_from_user(&val, optval, sizeof(val)))
3324 return -EFAULT;
3325
3326 po->auxdata = !!val;
3327 return 0;
3328 }
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003329 case PACKET_ORIGDEV:
3330 {
3331 int val;
3332
3333 if (optlen < sizeof(val))
3334 return -EINVAL;
3335 if (copy_from_user(&val, optval, sizeof(val)))
3336 return -EFAULT;
3337
3338 po->origdev = !!val;
3339 return 0;
3340 }
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003341 case PACKET_VNET_HDR:
3342 {
3343 int val;
3344
3345 if (sock->type != SOCK_RAW)
3346 return -EINVAL;
3347 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3348 return -EBUSY;
3349 if (optlen < sizeof(val))
3350 return -EINVAL;
3351 if (copy_from_user(&val, optval, sizeof(val)))
3352 return -EFAULT;
3353
3354 po->has_vnet_hdr = !!val;
3355 return 0;
3356 }
Scott McMillan614f60f2010-06-02 05:53:56 -07003357 case PACKET_TIMESTAMP:
3358 {
3359 int val;
3360
3361 if (optlen != sizeof(val))
3362 return -EINVAL;
3363 if (copy_from_user(&val, optval, sizeof(val)))
3364 return -EFAULT;
3365
3366 po->tp_tstamp = val;
3367 return 0;
3368 }
David S. Millerdc99f602011-07-05 01:45:05 -07003369 case PACKET_FANOUT:
3370 {
3371 int val;
3372
3373 if (optlen != sizeof(val))
3374 return -EINVAL;
3375 if (copy_from_user(&val, optval, sizeof(val)))
3376 return -EFAULT;
3377
3378 return fanout_add(sk, val & 0xffff, val >> 16);
3379 }
Paul Chavent5920cd3a2012-11-06 23:10:47 +00003380 case PACKET_TX_HAS_OFF:
3381 {
3382 unsigned int val;
3383
3384 if (optlen != sizeof(val))
3385 return -EINVAL;
3386 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3387 return -EBUSY;
3388 if (copy_from_user(&val, optval, sizeof(val)))
3389 return -EFAULT;
3390 po->tp_tx_has_off = !!val;
3391 return 0;
3392 }
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01003393 case PACKET_QDISC_BYPASS:
3394 {
3395 int val;
3396
3397 if (optlen != sizeof(val))
3398 return -EINVAL;
3399 if (copy_from_user(&val, optval, sizeof(val)))
3400 return -EFAULT;
3401
3402 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3403 return 0;
3404 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405 default:
3406 return -ENOPROTOOPT;
3407 }
3408}
3409
3410static int packet_getsockopt(struct socket *sock, int level, int optname,
3411 char __user *optval, int __user *optlen)
3412{
3413 int len;
Eric Dumazetc06fff62012-04-19 21:56:11 +00003414 int val, lv = sizeof(val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003415 struct sock *sk = sock->sk;
3416 struct packet_sock *po = pkt_sk(sk);
Eric Dumazetc06fff62012-04-19 21:56:11 +00003417 void *data = &val;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003418 union tpacket_stats_u st;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419
3420 if (level != SOL_PACKET)
3421 return -ENOPROTOOPT;
3422
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003423 if (get_user(len, optlen))
3424 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003425
3426 if (len < 0)
3427 return -EINVAL;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003428
Johann Baudy69e3c752009-05-18 22:11:22 -07003429 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 case PACKET_STATISTICS:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003431 spin_lock_bh(&sk->sk_receive_queue.lock);
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003432 memcpy(&st, &po->stats, sizeof(st));
3433 memset(&po->stats, 0, sizeof(po->stats));
3434 spin_unlock_bh(&sk->sk_receive_queue.lock);
3435
chetan lokef6fb8f102011-08-19 10:18:16 +00003436 if (po->tp_version == TPACKET_V3) {
Eric Dumazetc06fff62012-04-19 21:56:11 +00003437 lv = sizeof(struct tpacket_stats_v3);
Willem de Bruijn8bcdeaf2013-08-19 16:40:22 -04003438 st.stats3.tp_packets += st.stats3.tp_drops;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003439 data = &st.stats3;
chetan lokef6fb8f102011-08-19 10:18:16 +00003440 } else {
Eric Dumazetc06fff62012-04-19 21:56:11 +00003441 lv = sizeof(struct tpacket_stats);
Willem de Bruijn8bcdeaf2013-08-19 16:40:22 -04003442 st.stats1.tp_packets += st.stats1.tp_drops;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003443 data = &st.stats1;
chetan lokef6fb8f102011-08-19 10:18:16 +00003444 }
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003445
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446 break;
Herbert Xu8dc41942007-02-04 23:31:32 -08003447 case PACKET_AUXDATA:
Herbert Xu8dc41942007-02-04 23:31:32 -08003448 val = po->auxdata;
Herbert Xu8dc41942007-02-04 23:31:32 -08003449 break;
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003450 case PACKET_ORIGDEV:
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003451 val = po->origdev;
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003452 break;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003453 case PACKET_VNET_HDR:
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003454 val = po->has_vnet_hdr;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003455 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003456 case PACKET_VERSION:
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003457 val = po->tp_version;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003458 break;
3459 case PACKET_HDRLEN:
3460 if (len > sizeof(int))
3461 len = sizeof(int);
3462 if (copy_from_user(&val, optval, len))
3463 return -EFAULT;
3464 switch (val) {
3465 case TPACKET_V1:
3466 val = sizeof(struct tpacket_hdr);
3467 break;
3468 case TPACKET_V2:
3469 val = sizeof(struct tpacket2_hdr);
3470 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00003471 case TPACKET_V3:
3472 val = sizeof(struct tpacket3_hdr);
3473 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003474 default:
3475 return -EINVAL;
3476 }
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003477 break;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003478 case PACKET_RESERVE:
Patrick McHardy8913336a2008-07-18 18:05:19 -07003479 val = po->tp_reserve;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003480 break;
Johann Baudy69e3c752009-05-18 22:11:22 -07003481 case PACKET_LOSS:
Johann Baudy69e3c752009-05-18 22:11:22 -07003482 val = po->tp_loss;
Johann Baudy69e3c752009-05-18 22:11:22 -07003483 break;
Scott McMillan614f60f2010-06-02 05:53:56 -07003484 case PACKET_TIMESTAMP:
Scott McMillan614f60f2010-06-02 05:53:56 -07003485 val = po->tp_tstamp;
Scott McMillan614f60f2010-06-02 05:53:56 -07003486 break;
David S. Millerdc99f602011-07-05 01:45:05 -07003487 case PACKET_FANOUT:
David S. Millerdc99f602011-07-05 01:45:05 -07003488 val = (po->fanout ?
3489 ((u32)po->fanout->id |
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00003490 ((u32)po->fanout->type << 16) |
3491 ((u32)po->fanout->flags << 24)) :
David S. Millerdc99f602011-07-05 01:45:05 -07003492 0);
David S. Millerdc99f602011-07-05 01:45:05 -07003493 break;
Paul Chavent5920cd3a2012-11-06 23:10:47 +00003494 case PACKET_TX_HAS_OFF:
3495 val = po->tp_tx_has_off;
3496 break;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01003497 case PACKET_QDISC_BYPASS:
3498 val = packet_use_direct_xmit(po);
3499 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500 default:
3501 return -ENOPROTOOPT;
3502 }
3503
Eric Dumazetc06fff62012-04-19 21:56:11 +00003504 if (len > lv)
3505 len = lv;
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003506 if (put_user(len, optlen))
3507 return -EFAULT;
Herbert Xu8dc41942007-02-04 23:31:32 -08003508 if (copy_to_user(optval, data, len))
3509 return -EFAULT;
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003510 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003511}
3512
3513
Jiri Pirko351638e2013-05-28 01:30:21 +00003514static int packet_notifier(struct notifier_block *this,
3515 unsigned long msg, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516{
3517 struct sock *sk;
Jiri Pirko351638e2013-05-28 01:30:21 +00003518 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09003519 struct net *net = dev_net(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003520
stephen hemminger808f5112010-02-22 07:57:18 +00003521 rcu_read_lock();
Sasha Levinb67bfe02013-02-27 17:06:00 -08003522 sk_for_each_rcu(sk, &net->packet.sklist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003523 struct packet_sock *po = pkt_sk(sk);
3524
3525 switch (msg) {
3526 case NETDEV_UNREGISTER:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003527 if (po->mclist)
3528 packet_dev_mclist(dev, po->mclist, -1);
David S. Millera2efcfa2007-05-29 13:12:50 -07003529 /* fallthrough */
3530
Linus Torvalds1da177e2005-04-16 15:20:36 -07003531 case NETDEV_DOWN:
3532 if (dev->ifindex == po->ifindex) {
3533 spin_lock(&po->bind_lock);
3534 if (po->running) {
David S. Millerce06b032011-07-04 01:44:29 -07003535 __unregister_prot_hook(sk, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003536 sk->sk_err = ENETDOWN;
3537 if (!sock_flag(sk, SOCK_DEAD))
3538 sk->sk_error_report(sk);
3539 }
3540 if (msg == NETDEV_UNREGISTER) {
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01003541 packet_cached_dev_reset(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542 po->ifindex = -1;
Ben Greear160ff182011-06-01 07:18:52 +00003543 if (po->prot_hook.dev)
3544 dev_put(po->prot_hook.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003545 po->prot_hook.dev = NULL;
3546 }
3547 spin_unlock(&po->bind_lock);
3548 }
3549 break;
3550 case NETDEV_UP:
stephen hemminger808f5112010-02-22 07:57:18 +00003551 if (dev->ifindex == po->ifindex) {
3552 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07003553 if (po->num)
3554 register_prot_hook(sk);
stephen hemminger808f5112010-02-22 07:57:18 +00003555 spin_unlock(&po->bind_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003556 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003557 break;
3558 }
3559 }
stephen hemminger808f5112010-02-22 07:57:18 +00003560 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003561 return NOTIFY_DONE;
3562}
3563
3564
3565static int packet_ioctl(struct socket *sock, unsigned int cmd,
3566 unsigned long arg)
3567{
3568 struct sock *sk = sock->sk;
3569
Johann Baudy69e3c752009-05-18 22:11:22 -07003570 switch (cmd) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003571 case SIOCOUTQ:
3572 {
3573 int amount = sk_wmem_alloc_get(sk);
Eric Dumazet31e6d362009-06-17 19:05:41 -07003574
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003575 return put_user(amount, (int __user *)arg);
3576 }
3577 case SIOCINQ:
3578 {
3579 struct sk_buff *skb;
3580 int amount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003581
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003582 spin_lock_bh(&sk->sk_receive_queue.lock);
3583 skb = skb_peek(&sk->sk_receive_queue);
3584 if (skb)
3585 amount = skb->len;
3586 spin_unlock_bh(&sk->sk_receive_queue.lock);
3587 return put_user(amount, (int __user *)arg);
3588 }
3589 case SIOCGSTAMP:
3590 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3591 case SIOCGSTAMPNS:
3592 return sock_get_timestampns(sk, (struct timespec __user *)arg);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003593
Linus Torvalds1da177e2005-04-16 15:20:36 -07003594#ifdef CONFIG_INET
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003595 case SIOCADDRT:
3596 case SIOCDELRT:
3597 case SIOCDARP:
3598 case SIOCGARP:
3599 case SIOCSARP:
3600 case SIOCGIFADDR:
3601 case SIOCSIFADDR:
3602 case SIOCGIFBRDADDR:
3603 case SIOCSIFBRDADDR:
3604 case SIOCGIFNETMASK:
3605 case SIOCSIFNETMASK:
3606 case SIOCGIFDSTADDR:
3607 case SIOCSIFDSTADDR:
3608 case SIOCSIFFLAGS:
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003609 return inet_dgram_ops.ioctl(sock, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003610#endif
3611
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003612 default:
3613 return -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003614 }
3615 return 0;
3616}
3617
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003618static unsigned int packet_poll(struct file *file, struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003619 poll_table *wait)
3620{
3621 struct sock *sk = sock->sk;
3622 struct packet_sock *po = pkt_sk(sk);
3623 unsigned int mask = datagram_poll(file, sock, wait);
3624
3625 spin_lock_bh(&sk->sk_receive_queue.lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003626 if (po->rx_ring.pg_vec) {
chetan lokef6fb8f102011-08-19 10:18:16 +00003627 if (!packet_previous_rx_frame(po, &po->rx_ring,
3628 TP_STATUS_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003629 mask |= POLLIN | POLLRDNORM;
3630 }
3631 spin_unlock_bh(&sk->sk_receive_queue.lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003632 spin_lock_bh(&sk->sk_write_queue.lock);
3633 if (po->tx_ring.pg_vec) {
3634 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3635 mask |= POLLOUT | POLLWRNORM;
3636 }
3637 spin_unlock_bh(&sk->sk_write_queue.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003638 return mask;
3639}
3640
3641
3642/* Dirty? Well, I still did not learn better way to account
3643 * for user mmaps.
3644 */
3645
3646static void packet_mm_open(struct vm_area_struct *vma)
3647{
3648 struct file *file = vma->vm_file;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003649 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003650 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003651
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652 if (sk)
3653 atomic_inc(&pkt_sk(sk)->mapped);
3654}
3655
3656static void packet_mm_close(struct vm_area_struct *vma)
3657{
3658 struct file *file = vma->vm_file;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003659 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003660 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003661
Linus Torvalds1da177e2005-04-16 15:20:36 -07003662 if (sk)
3663 atomic_dec(&pkt_sk(sk)->mapped);
3664}
3665
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003666static const struct vm_operations_struct packet_mmap_ops = {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003667 .open = packet_mm_open,
3668 .close = packet_mm_close,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003669};
3670
Neil Horman0e3125c2010-11-16 10:26:47 -08003671static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3672 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003673{
3674 int i;
3675
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003676 for (i = 0; i < len; i++) {
Neil Horman0e3125c2010-11-16 10:26:47 -08003677 if (likely(pg_vec[i].buffer)) {
Changli Gaoc56b4d92010-12-01 02:52:57 +00003678 if (is_vmalloc_addr(pg_vec[i].buffer))
Neil Horman0e3125c2010-11-16 10:26:47 -08003679 vfree(pg_vec[i].buffer);
3680 else
3681 free_pages((unsigned long)pg_vec[i].buffer,
3682 order);
3683 pg_vec[i].buffer = NULL;
3684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003685 }
3686 kfree(pg_vec);
3687}
3688
Olof Johanssoneea49cc92011-11-02 11:00:49 +00003689static char *alloc_one_pg_vec_page(unsigned long order)
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003690{
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003691 char *buffer;
Neil Horman0e3125c2010-11-16 10:26:47 -08003692 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3693 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
Eric Dumazet719bfea2009-04-15 03:39:52 -07003694
Neil Horman0e3125c2010-11-16 10:26:47 -08003695 buffer = (char *) __get_free_pages(gfp_flags, order);
Neil Horman0e3125c2010-11-16 10:26:47 -08003696 if (buffer)
3697 return buffer;
3698
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003699 /* __get_free_pages failed, fall back to vmalloc */
Eric Dumazetbbce5a52010-11-20 07:31:54 +00003700 buffer = vzalloc((1 << order) * PAGE_SIZE);
Neil Horman0e3125c2010-11-16 10:26:47 -08003701 if (buffer)
3702 return buffer;
3703
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003704 /* vmalloc failed, lets dig into swap here */
Neil Horman0e3125c2010-11-16 10:26:47 -08003705 gfp_flags &= ~__GFP_NORETRY;
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003706 buffer = (char *) __get_free_pages(gfp_flags, order);
Neil Horman0e3125c2010-11-16 10:26:47 -08003707 if (buffer)
3708 return buffer;
3709
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003710 /* complete and utter failure */
Neil Horman0e3125c2010-11-16 10:26:47 -08003711 return NULL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003712}
3713
Neil Horman0e3125c2010-11-16 10:26:47 -08003714static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003715{
3716 unsigned int block_nr = req->tp_block_nr;
Neil Horman0e3125c2010-11-16 10:26:47 -08003717 struct pgv *pg_vec;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003718 int i;
3719
Neil Horman0e3125c2010-11-16 10:26:47 -08003720 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003721 if (unlikely(!pg_vec))
3722 goto out;
3723
3724 for (i = 0; i < block_nr; i++) {
Changli Gaoc56b4d92010-12-01 02:52:57 +00003725 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
Neil Horman0e3125c2010-11-16 10:26:47 -08003726 if (unlikely(!pg_vec[i].buffer))
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003727 goto out_free_pgvec;
3728 }
3729
3730out:
3731 return pg_vec;
3732
3733out_free_pgvec:
3734 free_pg_vec(pg_vec, order, block_nr);
3735 pg_vec = NULL;
3736 goto out;
3737}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003738
chetan lokef6fb8f102011-08-19 10:18:16 +00003739static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
Johann Baudy69e3c752009-05-18 22:11:22 -07003740 int closing, int tx_ring)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003741{
Neil Horman0e3125c2010-11-16 10:26:47 -08003742 struct pgv *pg_vec = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003743 struct packet_sock *po = pkt_sk(sk);
Al Viro0e11c912006-11-08 00:26:29 -08003744 int was_running, order = 0;
Johann Baudy69e3c752009-05-18 22:11:22 -07003745 struct packet_ring_buffer *rb;
3746 struct sk_buff_head *rb_queue;
Al Viro0e11c912006-11-08 00:26:29 -08003747 __be16 num;
chetan lokef6fb8f102011-08-19 10:18:16 +00003748 int err = -EINVAL;
3749 /* Added to avoid minimal code churn */
3750 struct tpacket_req *req = &req_u->req;
3751
3752 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3753 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3754 WARN(1, "Tx-ring is not supported.\n");
3755 goto out;
3756 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003757
3758 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3759 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3760
3761 err = -EBUSY;
3762 if (!closing) {
3763 if (atomic_read(&po->mapped))
3764 goto out;
Daniel Borkmannb0138402014-01-15 16:25:36 +01003765 if (packet_read_pending(rb))
Johann Baudy69e3c752009-05-18 22:11:22 -07003766 goto out;
3767 }
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003768
Linus Torvalds1da177e2005-04-16 15:20:36 -07003769 if (req->tp_block_nr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003770 /* Sanity tests and some calculations */
Johann Baudy69e3c752009-05-18 22:11:22 -07003771 err = -EBUSY;
3772 if (unlikely(rb->pg_vec))
3773 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003774
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003775 switch (po->tp_version) {
3776 case TPACKET_V1:
3777 po->tp_hdrlen = TPACKET_HDRLEN;
3778 break;
3779 case TPACKET_V2:
3780 po->tp_hdrlen = TPACKET2_HDRLEN;
3781 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00003782 case TPACKET_V3:
3783 po->tp_hdrlen = TPACKET3_HDRLEN;
3784 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003785 }
3786
Johann Baudy69e3c752009-05-18 22:11:22 -07003787 err = -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003788 if (unlikely((int)req->tp_block_size <= 0))
Johann Baudy69e3c752009-05-18 22:11:22 -07003789 goto out;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003790 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
Johann Baudy69e3c752009-05-18 22:11:22 -07003791 goto out;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003792 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
Johann Baudy69e3c752009-05-18 22:11:22 -07003793 po->tp_reserve))
3794 goto out;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003795 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
Johann Baudy69e3c752009-05-18 22:11:22 -07003796 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003797
Johann Baudy69e3c752009-05-18 22:11:22 -07003798 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3799 if (unlikely(rb->frames_per_block <= 0))
3800 goto out;
3801 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3802 req->tp_frame_nr))
3803 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804
3805 err = -ENOMEM;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003806 order = get_order(req->tp_block_size);
3807 pg_vec = alloc_pg_vec(req, order);
3808 if (unlikely(!pg_vec))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809 goto out;
chetan lokef6fb8f102011-08-19 10:18:16 +00003810 switch (po->tp_version) {
3811 case TPACKET_V3:
3812 /* Transmit path is not supported. We checked
3813 * it above but just being paranoid
3814 */
3815 if (!tx_ring)
3816 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
Dan Carpenterd7cf0c32014-02-18 15:20:51 +03003817 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00003818 default:
3819 break;
3820 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003821 }
3822 /* Done */
3823 else {
3824 err = -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003825 if (unlikely(req->tp_frame_nr))
Johann Baudy69e3c752009-05-18 22:11:22 -07003826 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003827 }
3828
3829 lock_sock(sk);
3830
3831 /* Detach socket from network */
3832 spin_lock(&po->bind_lock);
3833 was_running = po->running;
3834 num = po->num;
3835 if (was_running) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003836 po->num = 0;
David S. Millerce06b032011-07-04 01:44:29 -07003837 __unregister_prot_hook(sk, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838 }
3839 spin_unlock(&po->bind_lock);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003840
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841 synchronize_net();
3842
3843 err = -EBUSY;
Herbert Xu905db442009-01-30 14:12:06 -08003844 mutex_lock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003845 if (closing || atomic_read(&po->mapped) == 0) {
3846 err = 0;
Johann Baudy69e3c752009-05-18 22:11:22 -07003847 spin_lock_bh(&rb_queue->lock);
Changli Gaoc053fd92010-12-10 16:02:20 -08003848 swap(rb->pg_vec, pg_vec);
Johann Baudy69e3c752009-05-18 22:11:22 -07003849 rb->frame_max = (req->tp_frame_nr - 1);
3850 rb->head = 0;
3851 rb->frame_size = req->tp_frame_size;
3852 spin_unlock_bh(&rb_queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003853
Changli Gaoc053fd92010-12-10 16:02:20 -08003854 swap(rb->pg_vec_order, order);
3855 swap(rb->pg_vec_len, req->tp_block_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003856
Johann Baudy69e3c752009-05-18 22:11:22 -07003857 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3858 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3859 tpacket_rcv : packet_rcv;
3860 skb_queue_purge(rb_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861 if (atomic_read(&po->mapped))
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003862 pr_err("packet_mmap: vma is busy: %d\n",
3863 atomic_read(&po->mapped));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864 }
Herbert Xu905db442009-01-30 14:12:06 -08003865 mutex_unlock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003866
3867 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07003868 if (was_running) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003869 po->num = num;
David S. Millerce06b032011-07-04 01:44:29 -07003870 register_prot_hook(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003871 }
3872 spin_unlock(&po->bind_lock);
chetan lokef6fb8f102011-08-19 10:18:16 +00003873 if (closing && (po->tp_version > TPACKET_V2)) {
3874 /* Because we don't support block-based V3 on tx-ring */
3875 if (!tx_ring)
3876 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3877 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003878 release_sock(sk);
3879
Linus Torvalds1da177e2005-04-16 15:20:36 -07003880 if (pg_vec)
3881 free_pg_vec(pg_vec, order, req->tp_block_nr);
3882out:
3883 return err;
3884}
3885
Johann Baudy69e3c752009-05-18 22:11:22 -07003886static int packet_mmap(struct file *file, struct socket *sock,
3887 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003888{
3889 struct sock *sk = sock->sk;
3890 struct packet_sock *po = pkt_sk(sk);
Johann Baudy69e3c752009-05-18 22:11:22 -07003891 unsigned long size, expected_size;
3892 struct packet_ring_buffer *rb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893 unsigned long start;
3894 int err = -EINVAL;
3895 int i;
3896
3897 if (vma->vm_pgoff)
3898 return -EINVAL;
3899
Herbert Xu905db442009-01-30 14:12:06 -08003900 mutex_lock(&po->pg_vec_lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003901
3902 expected_size = 0;
3903 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3904 if (rb->pg_vec) {
3905 expected_size += rb->pg_vec_len
3906 * rb->pg_vec_pages
3907 * PAGE_SIZE;
3908 }
3909 }
3910
3911 if (expected_size == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07003913
3914 size = vma->vm_end - vma->vm_start;
3915 if (size != expected_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003916 goto out;
3917
Linus Torvalds1da177e2005-04-16 15:20:36 -07003918 start = vma->vm_start;
Johann Baudy69e3c752009-05-18 22:11:22 -07003919 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3920 if (rb->pg_vec == NULL)
3921 continue;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003922
Johann Baudy69e3c752009-05-18 22:11:22 -07003923 for (i = 0; i < rb->pg_vec_len; i++) {
Neil Horman0e3125c2010-11-16 10:26:47 -08003924 struct page *page;
3925 void *kaddr = rb->pg_vec[i].buffer;
Johann Baudy69e3c752009-05-18 22:11:22 -07003926 int pg_num;
3927
Changli Gaoc56b4d92010-12-01 02:52:57 +00003928 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3929 page = pgv_to_page(kaddr);
Johann Baudy69e3c752009-05-18 22:11:22 -07003930 err = vm_insert_page(vma, start, page);
3931 if (unlikely(err))
3932 goto out;
3933 start += PAGE_SIZE;
Neil Horman0e3125c2010-11-16 10:26:47 -08003934 kaddr += PAGE_SIZE;
Johann Baudy69e3c752009-05-18 22:11:22 -07003935 }
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003936 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003937 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003938
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003939 atomic_inc(&po->mapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003940 vma->vm_ops = &packet_mmap_ops;
3941 err = 0;
3942
3943out:
Herbert Xu905db442009-01-30 14:12:06 -08003944 mutex_unlock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003945 return err;
3946}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003947
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08003948static const struct proto_ops packet_ops_spkt = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003949 .family = PF_PACKET,
3950 .owner = THIS_MODULE,
3951 .release = packet_release,
3952 .bind = packet_bind_spkt,
3953 .connect = sock_no_connect,
3954 .socketpair = sock_no_socketpair,
3955 .accept = sock_no_accept,
3956 .getname = packet_getname_spkt,
3957 .poll = datagram_poll,
3958 .ioctl = packet_ioctl,
3959 .listen = sock_no_listen,
3960 .shutdown = sock_no_shutdown,
3961 .setsockopt = sock_no_setsockopt,
3962 .getsockopt = sock_no_getsockopt,
3963 .sendmsg = packet_sendmsg_spkt,
3964 .recvmsg = packet_recvmsg,
3965 .mmap = sock_no_mmap,
3966 .sendpage = sock_no_sendpage,
3967};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003968
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08003969static const struct proto_ops packet_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970 .family = PF_PACKET,
3971 .owner = THIS_MODULE,
3972 .release = packet_release,
3973 .bind = packet_bind,
3974 .connect = sock_no_connect,
3975 .socketpair = sock_no_socketpair,
3976 .accept = sock_no_accept,
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003977 .getname = packet_getname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003978 .poll = packet_poll,
3979 .ioctl = packet_ioctl,
3980 .listen = sock_no_listen,
3981 .shutdown = sock_no_shutdown,
3982 .setsockopt = packet_setsockopt,
3983 .getsockopt = packet_getsockopt,
3984 .sendmsg = packet_sendmsg,
3985 .recvmsg = packet_recvmsg,
3986 .mmap = packet_mmap,
3987 .sendpage = sock_no_sendpage,
3988};
3989
Stephen Hemmingerec1b4cf2009-10-05 05:58:39 +00003990static const struct net_proto_family packet_family_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 .family = PF_PACKET,
3992 .create = packet_create,
3993 .owner = THIS_MODULE,
3994};
3995
3996static struct notifier_block packet_netdev_notifier = {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003997 .notifier_call = packet_notifier,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003998};
3999
4000#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07004001
4002static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
stephen hemminger808f5112010-02-22 07:57:18 +00004003 __acquires(RCU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004004{
Denis V. Luneve372c412007-11-19 22:31:54 -08004005 struct net *net = seq_file_net(seq);
stephen hemminger808f5112010-02-22 07:57:18 +00004006
4007 rcu_read_lock();
4008 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004009}
4010
4011static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4012{
Herbert Xu1bf40952007-12-16 14:04:02 -08004013 struct net *net = seq_file_net(seq);
stephen hemminger808f5112010-02-22 07:57:18 +00004014 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004015}
4016
4017static void packet_seq_stop(struct seq_file *seq, void *v)
stephen hemminger808f5112010-02-22 07:57:18 +00004018 __releases(RCU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004019{
stephen hemminger808f5112010-02-22 07:57:18 +00004020 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004021}
4022
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09004023static int packet_seq_show(struct seq_file *seq, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024{
4025 if (v == SEQ_START_TOKEN)
4026 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4027 else {
Li Zefanb7ceabd2010-02-08 23:19:29 +00004028 struct sock *s = sk_entry(v);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004029 const struct packet_sock *po = pkt_sk(s);
4030
4031 seq_printf(seq,
Dan Rosenberg71338aa2011-05-23 12:17:35 +00004032 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004033 s,
4034 atomic_read(&s->sk_refcnt),
4035 s->sk_type,
4036 ntohs(po->num),
4037 po->ifindex,
4038 po->running,
4039 atomic_read(&s->sk_rmem_alloc),
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06004040 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00004041 sock_i_ino(s));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004042 }
4043
4044 return 0;
4045}
4046
Philippe De Muyter56b3d972007-07-10 23:07:31 -07004047static const struct seq_operations packet_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004048 .start = packet_seq_start,
4049 .next = packet_seq_next,
4050 .stop = packet_seq_stop,
4051 .show = packet_seq_show,
4052};
4053
4054static int packet_seq_open(struct inode *inode, struct file *file)
4055{
Denis V. Luneve372c412007-11-19 22:31:54 -08004056 return seq_open_net(inode, file, &packet_seq_ops,
4057 sizeof(struct seq_net_private));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004058}
4059
Arjan van de Venda7071d2007-02-12 00:55:36 -08004060static const struct file_operations packet_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061 .owner = THIS_MODULE,
4062 .open = packet_seq_open,
4063 .read = seq_read,
4064 .llseek = seq_lseek,
Denis V. Luneve372c412007-11-19 22:31:54 -08004065 .release = seq_release_net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004066};
4067
4068#endif
4069
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004070static int __net_init packet_net_init(struct net *net)
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004071{
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00004072 mutex_init(&net->packet.sklist_lock);
Denis V. Lunev2aaef4e2007-12-11 04:19:54 -08004073 INIT_HLIST_HEAD(&net->packet.sklist);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004074
Gao fengd4beaa62013-02-18 01:34:54 +00004075 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004076 return -ENOMEM;
4077
4078 return 0;
4079}
4080
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004081static void __net_exit packet_net_exit(struct net *net)
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004082{
Gao fengece31ff2013-02-18 01:34:56 +00004083 remove_proc_entry("packet", net->proc_net);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004084}
4085
4086static struct pernet_operations packet_net_ops = {
4087 .init = packet_net_init,
4088 .exit = packet_net_exit,
4089};
4090
4091
Linus Torvalds1da177e2005-04-16 15:20:36 -07004092static void __exit packet_exit(void)
4093{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 unregister_netdevice_notifier(&packet_netdev_notifier);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004095 unregister_pernet_subsys(&packet_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004096 sock_unregister(PF_PACKET);
4097 proto_unregister(&packet_proto);
4098}
4099
4100static int __init packet_init(void)
4101{
4102 int rc = proto_register(&packet_proto, 0);
4103
4104 if (rc != 0)
4105 goto out;
4106
4107 sock_register(&packet_family_ops);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004108 register_pernet_subsys(&packet_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004109 register_netdevice_notifier(&packet_netdev_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004110out:
4111 return rc;
4112}
4113
4114module_init(packet_init);
4115module_exit(packet_exit);
4116MODULE_LICENSE("GPL");
4117MODULE_ALIAS_NETPROTO(PF_PACKET);