blob: 404c9735aee90e5ebd6a7d3506762ec1358eadaa [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 {
Herbert Xuffbc6112007-02-04 23:33:10 -0800219 union {
220 struct sockaddr_pkt pkt;
Eyal Birger2472d762015-03-01 14:58:28 +0200221 union {
222 /* Trick: alias skb original length with
223 * ll.sll_family and ll.protocol in order
224 * to save room.
225 */
226 unsigned int origlen;
227 struct sockaddr_ll ll;
228 };
Herbert Xuffbc6112007-02-04 23:33:10 -0800229 } sa;
230};
231
232#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
Herbert Xu8dc41942007-02-04 23:31:32 -0800233
chetan lokebc59ba32011-08-25 10:43:30 +0000234#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
chetan lokef6fb8f102011-08-19 10:18:16 +0000235#define GET_PBLOCK_DESC(x, bid) \
chetan lokebc59ba32011-08-25 10:43:30 +0000236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
chetan lokef6fb8f102011-08-19 10:18:16 +0000237#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
chetan lokebc59ba32011-08-25 10:43:30 +0000238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
chetan lokef6fb8f102011-08-19 10:18:16 +0000239#define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
242
David S. Millerdc99f602011-07-05 01:45:05 -0700243static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244static void __fanout_link(struct sock *sk, struct packet_sock *po);
245
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100246static int packet_direct_xmit(struct sk_buff *skb)
247{
248 struct net_device *dev = skb->dev;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100249 netdev_features_t features;
250 struct netdev_queue *txq;
Daniel Borkmann43279502014-03-27 16:38:30 +0100251 int ret = NETDEV_TX_BUSY;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100252
253 if (unlikely(!netif_running(dev) ||
Daniel Borkmann43279502014-03-27 16:38:30 +0100254 !netif_carrier_ok(dev)))
255 goto drop;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100256
257 features = netif_skb_features(skb);
258 if (skb_needs_linearize(skb, features) &&
Daniel Borkmann43279502014-03-27 16:38:30 +0100259 __skb_linearize(skb))
260 goto drop;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100261
Daniel Borkmann10c51b56232014-08-27 11:11:27 +0200262 txq = skb_get_tx_queue(dev, skb);
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100263
Daniel Borkmann43279502014-03-27 16:38:30 +0100264 local_bh_disable();
Daniel Borkmannd346a3f2013-12-06 11:36:17 +0100265
Daniel Borkmann43279502014-03-27 16:38:30 +0100266 HARD_TX_LOCK(dev, txq, smp_processor_id());
David S. Miller10b3ad82014-08-29 21:07:24 -0700267 if (!netif_xmit_frozen_or_drv_stopped(txq))
David S. Millerfa2dbdc2014-08-29 21:55:22 -0700268 ret = netdev_start_xmit(skb, dev, txq, false);
Daniel Borkmann43279502014-03-27 16:38:30 +0100269 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
Michael S. Tsirkin6e580402014-11-24 13:32:16 +0200387static inline struct page * __pure 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
Willem de Bruijn68a360e2014-07-25 18:01:31 -0400444 if (shhwtstamps &&
445 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
446 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
447 return TP_STATUS_TS_RAW_HARDWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000448
449 if (ktime_to_timespec_cond(skb->tstamp, ts))
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000450 return TP_STATUS_TS_SOFTWARE;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000451
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000452 return 0;
Daniel Borkmann7a513842013-04-23 00:39:29 +0000453}
454
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000455static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
456 struct sk_buff *skb)
Willem de Bruijn2e313962013-04-23 00:39:28 +0000457{
458 union tpacket_uhdr h;
459 struct timespec ts;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000460 __u32 ts_status;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000461
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000462 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
463 return 0;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000464
465 h.raw = frame;
466 switch (po->tp_version) {
467 case TPACKET_V1:
468 h.h1->tp_sec = ts.tv_sec;
469 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
470 break;
471 case TPACKET_V2:
472 h.h2->tp_sec = ts.tv_sec;
473 h.h2->tp_nsec = ts.tv_nsec;
474 break;
475 case TPACKET_V3:
476 default:
477 WARN(1, "TPACKET version not supported.\n");
478 BUG();
479 }
480
481 /* one flush is safe, as both fields always lie on the same cacheline */
482 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
483 smp_wmb();
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +0000484
485 return ts_status;
Willem de Bruijn2e313962013-04-23 00:39:28 +0000486}
487
Johann Baudy69e3c752009-05-18 22:11:22 -0700488static void *packet_lookup_frame(struct packet_sock *po,
489 struct packet_ring_buffer *rb,
490 unsigned int position,
491 int status)
492{
493 unsigned int pg_vec_pos, frame_offset;
Daniel Borkmann184f4892013-04-16 01:57:46 +0000494 union tpacket_uhdr h;
Johann Baudy69e3c752009-05-18 22:11:22 -0700495
496 pg_vec_pos = position / rb->frames_per_block;
497 frame_offset = position % rb->frames_per_block;
498
Neil Horman0e3125c2010-11-16 10:26:47 -0800499 h.raw = rb->pg_vec[pg_vec_pos].buffer +
500 (frame_offset * rb->frame_size);
Johann Baudy69e3c752009-05-18 22:11:22 -0700501
502 if (status != __packet_get_status(po, h.raw))
503 return NULL;
504
505 return h.raw;
506}
507
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000508static void *packet_current_frame(struct packet_sock *po,
Johann Baudy69e3c752009-05-18 22:11:22 -0700509 struct packet_ring_buffer *rb,
510 int status)
511{
512 return packet_lookup_frame(po, rb, rb->head, status);
513}
514
chetan lokebc59ba32011-08-25 10:43:30 +0000515static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000516{
517 del_timer_sync(&pkc->retire_blk_timer);
518}
519
520static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
521 int tx_ring,
522 struct sk_buff_head *rb_queue)
523{
chetan lokebc59ba32011-08-25 10:43:30 +0000524 struct tpacket_kbdq_core *pkc;
chetan lokef6fb8f102011-08-19 10:18:16 +0000525
Duan Jiong22781a52013-12-06 13:29:36 +0800526 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
527 GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f102011-08-19 10:18:16 +0000528
Veaceslav Falicoec6f8092013-11-29 09:53:23 +0100529 spin_lock_bh(&rb_queue->lock);
chetan lokef6fb8f102011-08-19 10:18:16 +0000530 pkc->delete_blk_timer = 1;
Veaceslav Falicoec6f8092013-11-29 09:53:23 +0100531 spin_unlock_bh(&rb_queue->lock);
chetan lokef6fb8f102011-08-19 10:18:16 +0000532
533 prb_del_retire_blk_timer(pkc);
534}
535
536static void prb_init_blk_timer(struct packet_sock *po,
chetan lokebc59ba32011-08-25 10:43:30 +0000537 struct tpacket_kbdq_core *pkc,
chetan lokef6fb8f102011-08-19 10:18:16 +0000538 void (*func) (unsigned long))
539{
540 init_timer(&pkc->retire_blk_timer);
541 pkc->retire_blk_timer.data = (long)po;
542 pkc->retire_blk_timer.function = func;
543 pkc->retire_blk_timer.expires = jiffies;
544}
545
546static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
547{
chetan lokebc59ba32011-08-25 10:43:30 +0000548 struct tpacket_kbdq_core *pkc;
chetan lokef6fb8f102011-08-19 10:18:16 +0000549
550 if (tx_ring)
551 BUG();
552
Duan Jiong22781a52013-12-06 13:29:36 +0800553 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
554 GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f102011-08-19 10:18:16 +0000555 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
556}
557
558static int prb_calc_retire_blk_tmo(struct packet_sock *po,
559 int blk_size_in_bytes)
560{
561 struct net_device *dev;
562 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000563 struct ethtool_cmd ecmd;
564 int err;
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000565 u32 speed;
chetan lokef6fb8f102011-08-19 10:18:16 +0000566
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000567 rtnl_lock();
568 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
569 if (unlikely(!dev)) {
570 rtnl_unlock();
chetan lokef6fb8f102011-08-19 10:18:16 +0000571 return DEFAULT_PRB_RETIRE_TOV;
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000572 }
573 err = __ethtool_get_settings(dev, &ecmd);
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000574 speed = ethtool_cmd_speed(&ecmd);
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000575 rtnl_unlock();
576 if (!err) {
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000577 /*
578 * If the link speed is so slow you don't really
579 * need to worry about perf anyways
580 */
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000581 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
Jiri Pirko4bc71cb2011-09-03 03:34:30 +0000582 return DEFAULT_PRB_RETIRE_TOV;
parav.pandit@emulex.come440cf22012-06-27 03:56:12 +0000583 } else {
584 msec = 1;
585 div = speed / 1000;
chetan lokef6fb8f102011-08-19 10:18:16 +0000586 }
587 }
588
589 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
590
591 if (div)
592 mbits /= div;
593
594 tmo = mbits * msec;
595
596 if (div)
597 return tmo+1;
598 return tmo;
599}
600
chetan lokebc59ba32011-08-25 10:43:30 +0000601static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
chetan lokef6fb8f102011-08-19 10:18:16 +0000602 union tpacket_req_u *req_u)
603{
604 p1->feature_req_word = req_u->req3.tp_feature_req_word;
605}
606
607static void init_prb_bdqc(struct packet_sock *po,
608 struct packet_ring_buffer *rb,
609 struct pgv *pg_vec,
610 union tpacket_req_u *req_u, int tx_ring)
611{
Duan Jiong22781a52013-12-06 13:29:36 +0800612 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
chetan lokebc59ba32011-08-25 10:43:30 +0000613 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000614
615 memset(p1, 0x0, sizeof(*p1));
616
617 p1->knxt_seq_num = 1;
618 p1->pkbdq = pg_vec;
chetan lokebc59ba32011-08-25 10:43:30 +0000619 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
Joe Perchese3192692012-06-03 17:41:40 +0000620 p1->pkblk_start = pg_vec[0].buffer;
chetan lokef6fb8f102011-08-19 10:18:16 +0000621 p1->kblk_size = req_u->req3.tp_block_size;
622 p1->knum_blocks = req_u->req3.tp_block_nr;
623 p1->hdrlen = po->tp_hdrlen;
624 p1->version = po->tp_version;
625 p1->last_kactive_blk_num = 0;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +0000626 po->stats.stats3.tp_freeze_q_cnt = 0;
chetan lokef6fb8f102011-08-19 10:18:16 +0000627 if (req_u->req3.tp_retire_blk_tov)
628 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
629 else
630 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
631 req_u->req3.tp_block_size);
632 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
633 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
634
Eric Dumazetdc808112014-08-15 09:16:04 -0700635 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
chetan lokef6fb8f102011-08-19 10:18:16 +0000636 prb_init_ft_ops(p1, req_u);
637 prb_setup_retire_blk_timer(po, tx_ring);
638 prb_open_block(p1, pbd);
639}
640
641/* Do NOT update the last_blk_num first.
642 * Assumes sk_buff_head lock is held.
643 */
chetan lokebc59ba32011-08-25 10:43:30 +0000644static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
chetan lokef6fb8f102011-08-19 10:18:16 +0000645{
646 mod_timer(&pkc->retire_blk_timer,
647 jiffies + pkc->tov_in_jiffies);
648 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
649}
650
651/*
652 * Timer logic:
653 * 1) We refresh the timer only when we open a block.
654 * By doing this we don't waste cycles refreshing the timer
655 * on packet-by-packet basis.
656 *
657 * With a 1MB block-size, on a 1Gbps line, it will take
658 * i) ~8 ms to fill a block + ii) memcpy etc.
659 * In this cut we are not accounting for the memcpy time.
660 *
661 * So, if the user sets the 'tmo' to 10ms then the timer
662 * will never fire while the block is still getting filled
663 * (which is what we want). However, the user could choose
664 * to close a block early and that's fine.
665 *
666 * But when the timer does fire, we check whether or not to refresh it.
667 * Since the tmo granularity is in msecs, it is not too expensive
668 * to refresh the timer, lets say every '8' msecs.
669 * Either the user can set the 'tmo' or we can derive it based on
670 * a) line-speed and b) block-size.
671 * prb_calc_retire_blk_tmo() calculates the tmo.
672 *
673 */
674static void prb_retire_rx_blk_timer_expired(unsigned long data)
675{
676 struct packet_sock *po = (struct packet_sock *)data;
Duan Jiong22781a52013-12-06 13:29:36 +0800677 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
chetan lokef6fb8f102011-08-19 10:18:16 +0000678 unsigned int frozen;
chetan lokebc59ba32011-08-25 10:43:30 +0000679 struct tpacket_block_desc *pbd;
chetan lokef6fb8f102011-08-19 10:18:16 +0000680
681 spin_lock(&po->sk.sk_receive_queue.lock);
682
683 frozen = prb_queue_frozen(pkc);
684 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
685
686 if (unlikely(pkc->delete_blk_timer))
687 goto out;
688
689 /* We only need to plug the race when the block is partially filled.
690 * tpacket_rcv:
691 * lock(); increment BLOCK_NUM_PKTS; unlock()
692 * copy_bits() is in progress ...
693 * timer fires on other cpu:
694 * we can't retire the current block because copy_bits
695 * is in progress.
696 *
697 */
698 if (BLOCK_NUM_PKTS(pbd)) {
699 while (atomic_read(&pkc->blk_fill_in_prog)) {
700 /* Waiting for skb_copy_bits to finish... */
701 cpu_relax();
702 }
703 }
704
705 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
706 if (!frozen) {
707 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
708 if (!prb_dispatch_next_block(pkc, po))
709 goto refresh_timer;
710 else
711 goto out;
712 } else {
713 /* Case 1. Queue was frozen because user-space was
714 * lagging behind.
715 */
716 if (prb_curr_blk_in_use(pkc, pbd)) {
717 /*
718 * Ok, user-space is still behind.
719 * So just refresh the timer.
720 */
721 goto refresh_timer;
722 } else {
723 /* Case 2. queue was frozen,user-space caught up,
724 * now the link went idle && the timer fired.
725 * We don't have a block to close.So we open this
726 * block and restart the timer.
727 * opening a block thaws the queue,restarts timer
728 * Thawing/timer-refresh is a side effect.
729 */
730 prb_open_block(pkc, pbd);
731 goto out;
732 }
733 }
734 }
735
736refresh_timer:
737 _prb_refresh_rx_retire_blk_timer(pkc);
738
739out:
740 spin_unlock(&po->sk.sk_receive_queue.lock);
741}
742
Olof Johanssoneea49cc92011-11-02 11:00:49 +0000743static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
chetan lokebc59ba32011-08-25 10:43:30 +0000744 struct tpacket_block_desc *pbd1, __u32 status)
chetan lokef6fb8f102011-08-19 10:18:16 +0000745{
746 /* Flush everything minus the block header */
747
748#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
749 u8 *start, *end;
750
751 start = (u8 *)pbd1;
752
753 /* Skip the block header(we know header WILL fit in 4K) */
754 start += PAGE_SIZE;
755
756 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
757 for (; start < end; start += PAGE_SIZE)
758 flush_dcache_page(pgv_to_page(start));
759
760 smp_wmb();
761#endif
762
763 /* Now update the block status. */
764
765 BLOCK_STATUS(pbd1) = status;
766
767 /* Flush the block header */
768
769#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
770 start = (u8 *)pbd1;
771 flush_dcache_page(pgv_to_page(start));
772
773 smp_wmb();
774#endif
775}
776
777/*
778 * Side effect:
779 *
780 * 1) flush the block
781 * 2) Increment active_blk_num
782 *
783 * Note:We DONT refresh the timer on purpose.
784 * Because almost always the next block will be opened.
785 */
chetan lokebc59ba32011-08-25 10:43:30 +0000786static void prb_close_block(struct tpacket_kbdq_core *pkc1,
787 struct tpacket_block_desc *pbd1,
chetan lokef6fb8f102011-08-19 10:18:16 +0000788 struct packet_sock *po, unsigned int stat)
789{
790 __u32 status = TP_STATUS_USER | stat;
791
792 struct tpacket3_hdr *last_pkt;
chetan lokebc59ba32011-08-25 10:43:30 +0000793 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
Dan Collinsda413ee2014-12-19 16:49:25 +1300794 struct sock *sk = &po->sk;
chetan lokef6fb8f102011-08-19 10:18:16 +0000795
Daniel Borkmannee80fbf2013-04-19 06:12:29 +0000796 if (po->stats.stats3.tp_drops)
chetan lokef6fb8f102011-08-19 10:18:16 +0000797 status |= TP_STATUS_LOSING;
798
799 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
800 last_pkt->tp_next_offset = 0;
801
802 /* Get the ts of the last pkt */
803 if (BLOCK_NUM_PKTS(pbd1)) {
804 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
805 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
806 } else {
807 /* Ok, we tmo'd - so get the current time */
808 struct timespec ts;
809 getnstimeofday(&ts);
810 h1->ts_last_pkt.ts_sec = ts.tv_sec;
811 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
812 }
813
814 smp_wmb();
815
816 /* Flush the block */
817 prb_flush_block(pkc1, pbd1, status);
818
Dan Collinsda413ee2014-12-19 16:49:25 +1300819 sk->sk_data_ready(sk);
820
chetan lokef6fb8f102011-08-19 10:18:16 +0000821 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{
Jiri Pirkodf8a39d2015-01-13 17:13:44 +0100995 if (skb_vlan_tag_present(pkc->skb)) {
996 ppd->hv1.tp_vlan_tci = skb_vlan_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
Ying Xue1b784142015-03-02 15:37:48 +08001609static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1610 size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611{
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 }
Al Viro6ce8e9c2014-04-06 21:25:44 -04001688 err = memcpy_from_msg(skb_put(skb, len), msg, len);
Eric Dumazet1a35ca82009-12-15 05:47:03 +00001689 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
Eyal Birgerb4772ef2015-03-01 14:58:29 +02001819 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
Herbert Xuffbc6112007-02-04 23:33:10 -08001820
1821 sll = &PACKET_SKB_CB(skb)->sa.ll;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 sll->sll_hatype = dev->type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 sll->sll_pkttype = skb->pkt_type;
Peter P Waskiewicz Jr8032b462007-11-10 22:03:25 -08001824 if (unlikely(po->origdev))
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07001825 sll->sll_ifindex = orig_dev->ifindex;
1826 else
1827 sll->sll_ifindex = dev->ifindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
Stephen Hemmingerb95cce32007-09-26 22:13:38 -07001829 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830
Eyal Birger2472d762015-03-01 14:58:28 +02001831 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1832 * Use their space for storing the original skb length.
1833 */
1834 PACKET_SKB_CB(skb)->sa.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++;
Eyal Birger3bc3b962015-03-01 14:58:30 +02001848 sock_skb_set_dropcount(sk, skb);
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 }
Eric Dumazetdc808112014-08-15 09:16:04 -07001949 } else if (unlikely(macoff + snaplen >
1950 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1951 u32 nval;
1952
1953 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1954 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1955 snaplen, nval, macoff);
1956 snaplen = nval;
1957 if (unlikely((int)snaplen < 0)) {
1958 snaplen = 0;
1959 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1960 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 spin_lock(&sk->sk_receive_queue.lock);
chetan lokef6fb8f102011-08-19 10:18:16 +00001963 h.raw = packet_current_rx_frame(po, skb,
1964 TP_STATUS_KERNEL, (macoff+snaplen));
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001965 if (!h.raw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 goto ring_is_full;
chetan lokef6fb8f102011-08-19 10:18:16 +00001967 if (po->tp_version <= TPACKET_V2) {
1968 packet_increment_rx_head(po, &po->rx_ring);
1969 /*
1970 * LOSING will be reported till you read the stats,
1971 * because it's COR - Clear On Read.
1972 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1973 * at packet level.
1974 */
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00001975 if (po->stats.stats1.tp_drops)
chetan lokef6fb8f102011-08-19 10:18:16 +00001976 status |= TP_STATUS_LOSING;
1977 }
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00001978 po->stats.stats1.tp_packets++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 if (copy_skb) {
1980 status |= TP_STATUS_COPY;
1981 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1982 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 spin_unlock(&sk->sk_receive_queue.lock);
1984
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001985 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001986
1987 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
Daniel Borkmann7a513842013-04-23 00:39:29 +00001988 getnstimeofday(&ts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00001990 status |= ts_status;
1991
Patrick McHardybbd6ef82008-07-14 22:50:15 -07001992 switch (po->tp_version) {
1993 case TPACKET_V1:
1994 h.h1->tp_len = skb->len;
1995 h.h1->tp_snaplen = snaplen;
1996 h.h1->tp_mac = macoff;
1997 h.h1->tp_net = netoff;
Daniel Borkmann4b457bd2013-04-16 01:29:11 +00001998 h.h1->tp_sec = ts.tv_sec;
1999 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002000 hdrlen = sizeof(*h.h1);
2001 break;
2002 case TPACKET_V2:
2003 h.h2->tp_len = skb->len;
2004 h.h2->tp_snaplen = snaplen;
2005 h.h2->tp_mac = macoff;
2006 h.h2->tp_net = netoff;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002007 h.h2->tp_sec = ts.tv_sec;
2008 h.h2->tp_nsec = ts.tv_nsec;
Jiri Pirkodf8a39d2015-01-13 17:13:44 +01002009 if (skb_vlan_tag_present(skb)) {
2010 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09002011 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2012 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
Ben Greeara3bcc232011-06-01 06:49:10 +00002013 } else {
2014 h.h2->tp_vlan_tci = 0;
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09002015 h.h2->tp_vlan_tpid = 0;
Ben Greeara3bcc232011-06-01 06:49:10 +00002016 }
Atzm Watanabee4d26f42013-12-17 22:53:36 +09002017 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002018 hdrlen = sizeof(*h.h2);
2019 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00002020 case TPACKET_V3:
2021 /* tp_nxt_offset,vlan are already populated above.
2022 * So DONT clear those fields here
2023 */
2024 h.h3->tp_status |= status;
2025 h.h3->tp_len = skb->len;
2026 h.h3->tp_snaplen = snaplen;
2027 h.h3->tp_mac = macoff;
2028 h.h3->tp_net = netoff;
chetan lokef6fb8f102011-08-19 10:18:16 +00002029 h.h3->tp_sec = ts.tv_sec;
2030 h.h3->tp_nsec = ts.tv_nsec;
Atzm Watanabee4d26f42013-12-17 22:53:36 +09002031 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
chetan lokef6fb8f102011-08-19 10:18:16 +00002032 hdrlen = sizeof(*h.h3);
2033 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002034 default:
2035 BUG();
2036 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
Patrick McHardybbd6ef82008-07-14 22:50:15 -07002038 sll = h.raw + TPACKET_ALIGN(hdrlen);
Stephen Hemmingerb95cce32007-09-26 22:13:38 -07002039 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 sll->sll_family = AF_PACKET;
2041 sll->sll_hatype = dev->type;
2042 sll->sll_protocol = skb->protocol;
2043 sll->sll_pkttype = skb->pkt_type;
Peter P Waskiewicz Jr8032b462007-11-10 22:03:25 -08002044 if (unlikely(po->origdev))
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07002045 sll->sll_ifindex = orig_dev->ifindex;
2046 else
2047 sll->sll_ifindex = dev->ifindex;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048
Ralf Baechlee16aa202006-12-07 00:11:33 -08002049 smp_mb();
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002050
Changli Gaof6dafa92010-12-07 04:26:16 +00002051#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002052 if (po->tp_version <= TPACKET_V2) {
Changli Gao0af55bb2010-12-01 02:52:20 +00002053 u8 *start, *end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002055 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2056 macoff + snaplen);
2057
2058 for (start = h.raw; start < end; start += PAGE_SIZE)
2059 flush_dcache_page(pgv_to_page(start));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 }
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002061 smp_wmb();
Changli Gaof6dafa92010-12-07 04:26:16 +00002062#endif
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002063
Dan Collinsda413ee2014-12-19 16:49:25 +13002064 if (po->tp_version <= TPACKET_V2) {
chetan lokef6fb8f102011-08-19 10:18:16 +00002065 __packet_set_status(po, h.raw, status);
Dan Collinsda413ee2014-12-19 16:49:25 +13002066 sk->sk_data_ready(sk);
2067 } else {
chetan lokef6fb8f102011-08-19 10:18:16 +00002068 prb_clear_blk_fill_status(&po->rx_ring);
Dan Collinsda413ee2014-12-19 16:49:25 +13002069 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070
2071drop_n_restore:
2072 if (skb_head != skb->data && skb_shared(skb)) {
2073 skb->data = skb_head;
2074 skb->len = skb_len;
2075 }
2076drop:
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002077 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 return 0;
2079
2080ring_is_full:
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00002081 po->stats.stats1.tp_drops++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 spin_unlock(&sk->sk_receive_queue.lock);
2083
David S. Miller676d2362014-04-11 16:15:36 -04002084 sk->sk_data_ready(sk);
Wei Yongjunacb5d752009-02-25 00:36:42 +00002085 kfree_skb(copy_skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 goto drop_n_restore;
2087}
2088
Johann Baudy69e3c752009-05-18 22:11:22 -07002089static void tpacket_destruct_skb(struct sk_buff *skb)
2090{
2091 struct packet_sock *po = pkt_sk(skb->sk);
Johann Baudy69e3c752009-05-18 22:11:22 -07002092
Johann Baudy69e3c752009-05-18 22:11:22 -07002093 if (likely(po->tx_ring.pg_vec)) {
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01002094 void *ph;
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002095 __u32 ts;
2096
Johann Baudy69e3c752009-05-18 22:11:22 -07002097 ph = skb_shinfo(skb)->destructor_arg;
Daniel Borkmannb0138402014-01-15 16:25:36 +01002098 packet_dec_pending(&po->tx_ring);
Daniel Borkmannb9c32fb2013-04-23 00:39:31 +00002099
2100 ts = __packet_set_timestamp(po, ph, skb);
2101 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
Johann Baudy69e3c752009-05-18 22:11:22 -07002102 }
2103
2104 sock_wfree(skb);
2105}
2106
Willem de Bruijn9c707762014-11-19 13:10:16 -05002107static bool ll_header_truncated(const struct net_device *dev, int len)
2108{
2109 /* net device doesn't like empty head */
2110 if (unlikely(len <= dev->hard_header_len)) {
Willem de Bruijneee2f042015-01-08 11:29:18 -05002111 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
Willem de Bruijn9c707762014-11-19 13:10:16 -05002112 current->comm, len, dev->hard_header_len);
2113 return true;
2114 }
2115
2116 return false;
2117}
2118
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002119static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2120 void *frame, struct net_device *dev, int size_max,
Herbert Xuae641942011-11-18 02:20:04 +00002121 __be16 proto, unsigned char *addr, int hlen)
Johann Baudy69e3c752009-05-18 22:11:22 -07002122{
Daniel Borkmann184f4892013-04-16 01:57:46 +00002123 union tpacket_uhdr ph;
David S. Miller09effa62013-08-07 17:11:00 -07002124 int to_write, offset, len, tp_len, nr_frags, len_max;
Johann Baudy69e3c752009-05-18 22:11:22 -07002125 struct socket *sock = po->sk.sk_socket;
2126 struct page *page;
2127 void *data;
2128 int err;
2129
2130 ph.raw = frame;
2131
2132 skb->protocol = proto;
2133 skb->dev = dev;
2134 skb->priority = po->sk.sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00002135 skb->mark = po->sk.sk_mark;
Willem de Bruijn2e313962013-04-23 00:39:28 +00002136 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
Johann Baudy69e3c752009-05-18 22:11:22 -07002137 skb_shinfo(skb)->destructor_arg = ph.raw;
2138
2139 switch (po->tp_version) {
2140 case TPACKET_V2:
2141 tp_len = ph.h2->tp_len;
2142 break;
2143 default:
2144 tp_len = ph.h1->tp_len;
2145 break;
2146 }
David S. Miller09effa62013-08-07 17:11:00 -07002147 if (unlikely(tp_len > size_max)) {
2148 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2149 return -EMSGSIZE;
2150 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002151
Herbert Xuae641942011-11-18 02:20:04 +00002152 skb_reserve(skb, hlen);
Johann Baudy69e3c752009-05-18 22:11:22 -07002153 skb_reset_network_header(skb);
Jason Wangc1aad272013-03-25 20:19:57 +00002154
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002155 if (!packet_use_direct_xmit(po))
2156 skb_probe_transport_header(skb, 0);
2157 if (unlikely(po->tp_tx_has_off)) {
Paul Chavent5920cd3a2012-11-06 23:10:47 +00002158 int off_min, off_max, off;
2159 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2160 off_max = po->tx_ring.frame_size - tp_len;
2161 if (sock->type == SOCK_DGRAM) {
2162 switch (po->tp_version) {
2163 case TPACKET_V2:
2164 off = ph.h2->tp_net;
2165 break;
2166 default:
2167 off = ph.h1->tp_net;
2168 break;
2169 }
2170 } else {
2171 switch (po->tp_version) {
2172 case TPACKET_V2:
2173 off = ph.h2->tp_mac;
2174 break;
2175 default:
2176 off = ph.h1->tp_mac;
2177 break;
2178 }
2179 }
2180 if (unlikely((off < off_min) || (off_max < off)))
2181 return -EINVAL;
2182 data = ph.raw + off;
2183 } else {
2184 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2185 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002186 to_write = tp_len;
2187
2188 if (sock->type == SOCK_DGRAM) {
2189 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2190 NULL, tp_len);
2191 if (unlikely(err < 0))
2192 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002193 } else if (dev->hard_header_len) {
Willem de Bruijn9c707762014-11-19 13:10:16 -05002194 if (ll_header_truncated(dev, tp_len))
Johann Baudy69e3c752009-05-18 22:11:22 -07002195 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07002196
2197 skb_push(skb, dev->hard_header_len);
2198 err = skb_store_bits(skb, 0, data,
2199 dev->hard_header_len);
2200 if (unlikely(err))
2201 return err;
2202
2203 data += dev->hard_header_len;
2204 to_write -= dev->hard_header_len;
2205 }
2206
Johann Baudy69e3c752009-05-18 22:11:22 -07002207 offset = offset_in_page(data);
2208 len_max = PAGE_SIZE - offset;
2209 len = ((to_write > len_max) ? len_max : to_write);
2210
2211 skb->data_len = to_write;
2212 skb->len += to_write;
2213 skb->truesize += to_write;
2214 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2215
2216 while (likely(to_write)) {
2217 nr_frags = skb_shinfo(skb)->nr_frags;
2218
2219 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002220 pr_err("Packet exceed the number of skb frags(%lu)\n",
2221 MAX_SKB_FRAGS);
Johann Baudy69e3c752009-05-18 22:11:22 -07002222 return -EFAULT;
2223 }
2224
Changli Gao0af55bb2010-12-01 02:52:20 +00002225 page = pgv_to_page(data);
2226 data += len;
Johann Baudy69e3c752009-05-18 22:11:22 -07002227 flush_dcache_page(page);
2228 get_page(page);
Changli Gao0af55bb2010-12-01 02:52:20 +00002229 skb_fill_page_desc(skb, nr_frags, page, offset, len);
Johann Baudy69e3c752009-05-18 22:11:22 -07002230 to_write -= len;
2231 offset = 0;
2232 len_max = PAGE_SIZE;
2233 len = ((to_write > len_max) ? len_max : to_write);
2234 }
2235
2236 return tp_len;
2237}
2238
2239static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2240{
Johann Baudy69e3c752009-05-18 22:11:22 -07002241 struct sk_buff *skb;
2242 struct net_device *dev;
2243 __be16 proto;
David S. Miller09effa62013-08-07 17:11:00 -07002244 int err, reserve = 0;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002245 void *ph;
Steffen Hurrle342dfc32014-01-17 22:53:15 +01002246 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002247 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
Johann Baudy69e3c752009-05-18 22:11:22 -07002248 int tp_len, size_max;
2249 unsigned char *addr;
2250 int len_sum = 0;
danborkmann@iogearbox.net9e670302012-08-20 03:34:03 +00002251 int status = TP_STATUS_AVAILABLE;
Herbert Xuae641942011-11-18 02:20:04 +00002252 int hlen, tlen;
Johann Baudy69e3c752009-05-18 22:11:22 -07002253
Johann Baudy69e3c752009-05-18 22:11:22 -07002254 mutex_lock(&po->pg_vec_lock);
2255
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002256 if (likely(saddr == NULL)) {
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002257 dev = packet_cached_dev_get(po);
Johann Baudy69e3c752009-05-18 22:11:22 -07002258 proto = po->num;
2259 addr = NULL;
2260 } else {
2261 err = -EINVAL;
2262 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2263 goto out;
2264 if (msg->msg_namelen < (saddr->sll_halen
2265 + offsetof(struct sockaddr_ll,
2266 sll_addr)))
2267 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07002268 proto = saddr->sll_protocol;
2269 addr = saddr->sll_addr;
Ben Greear827d9782011-06-01 07:18:53 +00002270 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
Johann Baudy69e3c752009-05-18 22:11:22 -07002271 }
2272
Johann Baudy69e3c752009-05-18 22:11:22 -07002273 err = -ENXIO;
2274 if (unlikely(dev == NULL))
2275 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07002276 err = -ENETDOWN;
2277 if (unlikely(!(dev->flags & IFF_UP)))
2278 goto out_put;
2279
Daniel Borkmann52f14542014-02-28 02:22:06 +01002280 reserve = dev->hard_header_len + VLAN_HLEN;
Johann Baudy69e3c752009-05-18 22:11:22 -07002281 size_max = po->tx_ring.frame_size
Gabor Gombasb5dd8842009-10-29 03:19:11 -07002282 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
Johann Baudy69e3c752009-05-18 22:11:22 -07002283
David S. Miller09effa62013-08-07 17:11:00 -07002284 if (size_max > dev->mtu + reserve)
2285 size_max = dev->mtu + reserve;
2286
Johann Baudy69e3c752009-05-18 22:11:22 -07002287 do {
2288 ph = packet_current_frame(po, &po->tx_ring,
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002289 TP_STATUS_SEND_REQUEST);
Johann Baudy69e3c752009-05-18 22:11:22 -07002290 if (unlikely(ph == NULL)) {
Daniel Borkmann87a2fd22014-01-15 16:25:35 +01002291 if (need_wait && need_resched())
2292 schedule();
Johann Baudy69e3c752009-05-18 22:11:22 -07002293 continue;
2294 }
2295
2296 status = TP_STATUS_SEND_REQUEST;
Herbert Xuae641942011-11-18 02:20:04 +00002297 hlen = LL_RESERVED_SPACE(dev);
2298 tlen = dev->needed_tailroom;
Johann Baudy69e3c752009-05-18 22:11:22 -07002299 skb = sock_alloc_send_skb(&po->sk,
Herbert Xuae641942011-11-18 02:20:04 +00002300 hlen + tlen + sizeof(struct sockaddr_ll),
Johann Baudy69e3c752009-05-18 22:11:22 -07002301 0, &err);
2302
2303 if (unlikely(skb == NULL))
2304 goto out_status;
2305
2306 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
Daniel Borkmann52f14542014-02-28 02:22:06 +01002307 addr, hlen);
2308 if (tp_len > dev->mtu + dev->hard_header_len) {
2309 struct ethhdr *ehdr;
2310 /* Earlier code assumed this would be a VLAN pkt,
2311 * double-check this now that we have the actual
2312 * packet in hand.
2313 */
Johann Baudy69e3c752009-05-18 22:11:22 -07002314
Daniel Borkmann52f14542014-02-28 02:22:06 +01002315 skb_reset_mac_header(skb);
2316 ehdr = eth_hdr(skb);
2317 if (ehdr->h_proto != htons(ETH_P_8021Q))
2318 tp_len = -EMSGSIZE;
2319 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002320 if (unlikely(tp_len < 0)) {
2321 if (po->tp_loss) {
2322 __packet_set_status(po, ph,
2323 TP_STATUS_AVAILABLE);
2324 packet_increment_head(&po->tx_ring);
2325 kfree_skb(skb);
2326 continue;
2327 } else {
2328 status = TP_STATUS_WRONG_FORMAT;
2329 err = tp_len;
2330 goto out_status;
2331 }
2332 }
2333
Daniel Borkmann0fd5d572014-02-16 15:55:22 +01002334 packet_pick_tx_queue(dev, skb);
2335
Johann Baudy69e3c752009-05-18 22:11:22 -07002336 skb->destructor = tpacket_destruct_skb;
2337 __packet_set_status(po, ph, TP_STATUS_SENDING);
Daniel Borkmannb0138402014-01-15 16:25:36 +01002338 packet_inc_pending(&po->tx_ring);
Johann Baudy69e3c752009-05-18 22:11:22 -07002339
2340 status = TP_STATUS_SEND_REQUEST;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002341 err = po->xmit(skb);
Jarek Poplawskieb70df12010-01-10 22:04:19 +00002342 if (unlikely(err > 0)) {
2343 err = net_xmit_errno(err);
2344 if (err && __packet_get_status(po, ph) ==
2345 TP_STATUS_AVAILABLE) {
2346 /* skb was destructed already */
2347 skb = NULL;
2348 goto out_status;
2349 }
2350 /*
2351 * skb was dropped but not destructed yet;
2352 * let's treat it like congestion or err < 0
2353 */
2354 err = 0;
2355 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002356 packet_increment_head(&po->tx_ring);
2357 len_sum += tp_len;
Daniel Borkmannb0138402014-01-15 16:25:36 +01002358 } while (likely((ph != NULL) ||
2359 /* Note: packet_read_pending() might be slow if we have
2360 * to call it as it's per_cpu variable, but in fast-path
2361 * we already short-circuit the loop with the first
2362 * condition, and luckily don't have to go that path
2363 * anyway.
2364 */
2365 (need_wait && packet_read_pending(&po->tx_ring))));
Johann Baudy69e3c752009-05-18 22:11:22 -07002366
2367 err = len_sum;
2368 goto out_put;
2369
Johann Baudy69e3c752009-05-18 22:11:22 -07002370out_status:
2371 __packet_set_status(po, ph, status);
2372 kfree_skb(skb);
2373out_put:
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002374 dev_put(dev);
Johann Baudy69e3c752009-05-18 22:11:22 -07002375out:
2376 mutex_unlock(&po->pg_vec_lock);
2377 return err;
2378}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
Olof Johanssoneea49cc92011-11-02 11:00:49 +00002380static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2381 size_t reserve, size_t len,
2382 size_t linear, int noblock,
2383 int *err)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002384{
2385 struct sk_buff *skb;
2386
2387 /* Under a page? Don't bother with paged skb. */
2388 if (prepad + len < PAGE_SIZE || !linear)
2389 linear = len;
2390
2391 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
Eric Dumazet28d64272013-08-08 14:38:47 -07002392 err, 0);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002393 if (!skb)
2394 return NULL;
2395
2396 skb_reserve(skb, reserve);
2397 skb_put(skb, linear);
2398 skb->data_len = len - linear;
2399 skb->len += len - linear;
2400
2401 return skb;
2402}
2403
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002404static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405{
2406 struct sock *sk = sock->sk;
Steffen Hurrle342dfc32014-01-17 22:53:15 +01002407 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408 struct sk_buff *skb;
2409 struct net_device *dev;
Al Viro0e11c912006-11-08 00:26:29 -08002410 __be16 proto;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 unsigned char *addr;
Ben Greear827d9782011-06-01 07:18:53 +00002412 int err, reserve = 0;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002413 struct virtio_net_hdr vnet_hdr = { 0 };
2414 int offset = 0;
2415 int vnet_hdr_len;
2416 struct packet_sock *po = pkt_sk(sk);
2417 unsigned short gso_type = 0;
Herbert Xuae641942011-11-18 02:20:04 +00002418 int hlen, tlen;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002419 int extra_len = 0;
Al Viro8feb2fb2014-11-06 01:10:59 -05002420 ssize_t n;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421
2422 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002423 * Get and verify the address.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002425
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002426 if (likely(saddr == NULL)) {
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002427 dev = packet_cached_dev_get(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 proto = po->num;
2429 addr = NULL;
2430 } else {
2431 err = -EINVAL;
2432 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2433 goto out;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002434 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2435 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 proto = saddr->sll_protocol;
2437 addr = saddr->sll_addr;
Ben Greear827d9782011-06-01 07:18:53 +00002438 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 }
2440
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 err = -ENXIO;
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002442 if (unlikely(dev == NULL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 goto out_unlock;
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002444 err = -ENETDOWN;
2445 if (unlikely(!(dev->flags & IFF_UP)))
2446 goto out_unlock;
2447
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 if (sock->type == SOCK_RAW)
2449 reserve = dev->hard_header_len;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002450 if (po->has_vnet_hdr) {
2451 vnet_hdr_len = sizeof(vnet_hdr);
2452
2453 err = -EINVAL;
2454 if (len < vnet_hdr_len)
2455 goto out_unlock;
2456
2457 len -= vnet_hdr_len;
2458
Al Viro8feb2fb2014-11-06 01:10:59 -05002459 err = -EFAULT;
Al Viroc0371da2014-11-24 10:42:55 -05002460 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
Al Viro8feb2fb2014-11-06 01:10:59 -05002461 if (n != vnet_hdr_len)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002462 goto out_unlock;
2463
2464 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002465 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2466 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2467 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2468 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2469 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2470 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002471
2472 err = -EINVAL;
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002473 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002474 goto out_unlock;
2475
2476 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2477 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2478 case VIRTIO_NET_HDR_GSO_TCPV4:
2479 gso_type = SKB_GSO_TCPV4;
2480 break;
2481 case VIRTIO_NET_HDR_GSO_TCPV6:
2482 gso_type = SKB_GSO_TCPV6;
2483 break;
2484 case VIRTIO_NET_HDR_GSO_UDP:
2485 gso_type = SKB_GSO_UDP;
2486 break;
2487 default:
2488 goto out_unlock;
2489 }
2490
2491 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2492 gso_type |= SKB_GSO_TCP_ECN;
2493
2494 if (vnet_hdr.gso_size == 0)
2495 goto out_unlock;
2496
2497 }
2498 }
2499
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002500 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2501 if (!netif_supports_nofcs(dev)) {
2502 err = -EPROTONOSUPPORT;
2503 goto out_unlock;
2504 }
2505 extra_len = 4; /* We're doing our own CRC */
2506 }
2507
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 err = -EMSGSIZE;
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002509 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 goto out_unlock;
2511
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002512 err = -ENOBUFS;
Herbert Xuae641942011-11-18 02:20:04 +00002513 hlen = LL_RESERVED_SPACE(dev);
2514 tlen = dev->needed_tailroom;
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002515 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2516 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002517 msg->msg_flags & MSG_DONTWAIT, &err);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002518 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 goto out_unlock;
2520
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002521 skb_set_network_header(skb, reserve);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522
Stephen Hemminger0c4e8582007-10-09 01:36:32 -07002523 err = -EINVAL;
Willem de Bruijn9c707762014-11-19 13:10:16 -05002524 if (sock->type == SOCK_DGRAM) {
2525 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
Christoph Jaeger46d2cfb2015-01-11 13:01:16 -05002526 if (unlikely(offset < 0))
Willem de Bruijn9c707762014-11-19 13:10:16 -05002527 goto out_free;
2528 } else {
2529 if (ll_header_truncated(dev, len))
2530 goto out_free;
2531 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532
2533 /* Returns -EFAULT on error */
Al Viroc0371da2014-11-24 10:42:55 -05002534 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 if (err)
2536 goto out_free;
Daniel Borkmannbf84a012013-04-14 08:08:13 +00002537
2538 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002540 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
David S. Miller09effa62013-08-07 17:11:00 -07002541 /* Earlier code assumed this would be a VLAN pkt,
2542 * double-check this now that we have the actual
2543 * packet in hand.
2544 */
2545 struct ethhdr *ehdr;
2546 skb_reset_mac_header(skb);
2547 ehdr = eth_hdr(skb);
2548 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2549 err = -EMSGSIZE;
2550 goto out_free;
2551 }
Ben Greear57f89bf2011-02-11 09:35:18 +00002552 }
2553
David S. Miller09effa62013-08-07 17:11:00 -07002554 skb->protocol = proto;
2555 skb->dev = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 skb->priority = sk->sk_priority;
Eric Dumazet2d37a182009-10-01 19:14:46 +00002557 skb->mark = sk->sk_mark;
Daniel Borkmann0fd5d572014-02-16 15:55:22 +01002558
2559 packet_pick_tx_queue(dev, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002561 if (po->has_vnet_hdr) {
2562 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002563 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2564 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2565 if (!skb_partial_csum_set(skb, s, o)) {
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002566 err = -EINVAL;
2567 goto out_free;
2568 }
2569 }
2570
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002571 skb_shinfo(skb)->gso_size =
2572 __virtio16_to_cpu(false, vnet_hdr.gso_size);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002573 skb_shinfo(skb)->gso_type = gso_type;
2574
2575 /* Header must be checked, and gso_segs computed. */
2576 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2577 skb_shinfo(skb)->gso_segs = 0;
2578
2579 len += vnet_hdr_len;
2580 }
2581
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002582 if (!packet_use_direct_xmit(po))
2583 skb_probe_transport_header(skb, reserve);
Ben Greear3bdc0eb2012-02-11 15:39:30 +00002584 if (unlikely(extra_len == 4))
2585 skb->no_fcs = 1;
2586
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002587 err = po->xmit(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2589 goto out_unlock;
2590
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002591 dev_put(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002593 return len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594
2595out_free:
2596 kfree_skb(skb);
2597out_unlock:
Daniel Borkmanne40526c2013-11-21 16:50:58 +01002598 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 dev_put(dev);
2600out:
2601 return err;
2602}
2603
Ying Xue1b784142015-03-02 15:37:48 +08002604static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
Johann Baudy69e3c752009-05-18 22:11:22 -07002605{
Johann Baudy69e3c752009-05-18 22:11:22 -07002606 struct sock *sk = sock->sk;
2607 struct packet_sock *po = pkt_sk(sk);
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002608
Johann Baudy69e3c752009-05-18 22:11:22 -07002609 if (po->tx_ring.pg_vec)
2610 return tpacket_snd(po, msg);
2611 else
Johann Baudy69e3c752009-05-18 22:11:22 -07002612 return packet_snd(sock, msg, len);
2613}
2614
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615/*
2616 * Close a PACKET socket. This is fairly simple. We immediately go
2617 * to 'closed' state and remove our protocol entry in the device list.
2618 */
2619
2620static int packet_release(struct socket *sock)
2621{
2622 struct sock *sk = sock->sk;
2623 struct packet_sock *po;
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08002624 struct net *net;
chetan lokef6fb8f102011-08-19 10:18:16 +00002625 union tpacket_req_u req_u;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
2627 if (!sk)
2628 return 0;
2629
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002630 net = sock_net(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 po = pkt_sk(sk);
2632
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002633 mutex_lock(&net->packet.sklist_lock);
stephen hemminger808f5112010-02-22 07:57:18 +00002634 sk_del_node_init_rcu(sk);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002635 mutex_unlock(&net->packet.sklist_lock);
2636
2637 preempt_disable();
Eric Dumazet920de802008-11-24 00:09:29 -08002638 sock_prot_inuse_add(net, sk->sk_prot, -1);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002639 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640
stephen hemminger808f5112010-02-22 07:57:18 +00002641 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07002642 unregister_prot_hook(sk, false);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002643 packet_cached_dev_reset(po);
2644
Ben Greear160ff182011-06-01 07:18:52 +00002645 if (po->prot_hook.dev) {
2646 dev_put(po->prot_hook.dev);
2647 po->prot_hook.dev = NULL;
2648 }
stephen hemminger808f5112010-02-22 07:57:18 +00002649 spin_unlock(&po->bind_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 packet_flush_mclist(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652
Phil Sutter9665d5d2013-02-01 07:21:41 +00002653 if (po->rx_ring.pg_vec) {
2654 memset(&req_u, 0, sizeof(req_u));
chetan lokef6fb8f102011-08-19 10:18:16 +00002655 packet_set_ring(sk, &req_u, 1, 0);
Phil Sutter9665d5d2013-02-01 07:21:41 +00002656 }
Johann Baudy69e3c752009-05-18 22:11:22 -07002657
Phil Sutter9665d5d2013-02-01 07:21:41 +00002658 if (po->tx_ring.pg_vec) {
2659 memset(&req_u, 0, sizeof(req_u));
chetan lokef6fb8f102011-08-19 10:18:16 +00002660 packet_set_ring(sk, &req_u, 1, 1);
Phil Sutter9665d5d2013-02-01 07:21:41 +00002661 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662
David S. Millerdc99f602011-07-05 01:45:05 -07002663 fanout_release(sk);
2664
stephen hemminger808f5112010-02-22 07:57:18 +00002665 synchronize_net();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 /*
2667 * Now the socket is dead. No more input will appear.
2668 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669 sock_orphan(sk);
2670 sock->sk = NULL;
2671
2672 /* Purge queues */
2673
2674 skb_queue_purge(&sk->sk_receive_queue);
Daniel Borkmannb0138402014-01-15 16:25:36 +01002675 packet_free_pending(po);
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08002676 sk_refcnt_debug_release(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677
2678 sock_put(sk);
2679 return 0;
2680}
2681
2682/*
2683 * Attach a packet hook.
2684 */
2685
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002686static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687{
2688 struct packet_sock *po = pkt_sk(sk);
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002689 const struct net_device *dev_curr;
2690 __be16 proto_curr;
2691 bool need_rehook;
David S. Millerdc99f602011-07-05 01:45:05 -07002692
Wei Yongjunaef950b2011-12-27 22:32:41 -05002693 if (po->fanout) {
2694 if (dev)
2695 dev_put(dev);
2696
David S. Millerdc99f602011-07-05 01:45:05 -07002697 return -EINVAL;
Wei Yongjunaef950b2011-12-27 22:32:41 -05002698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
2700 lock_sock(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 spin_lock(&po->bind_lock);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002702
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002703 proto_curr = po->prot_hook.type;
2704 dev_curr = po->prot_hook.dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002706 need_rehook = proto_curr != proto || dev_curr != dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002708 if (need_rehook) {
2709 unregister_prot_hook(sk, true);
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002710
Daniel Borkmann902fefb2014-01-15 16:25:34 +01002711 po->num = proto;
2712 po->prot_hook.type = proto;
2713
2714 if (po->prot_hook.dev)
2715 dev_put(po->prot_hook.dev);
2716
2717 po->prot_hook.dev = dev;
2718
2719 po->ifindex = dev ? dev->ifindex : 0;
2720 packet_cached_dev_assign(po, dev);
2721 }
2722
2723 if (proto == 0 || !need_rehook)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 goto out_unlock;
2725
Urs Thuermannbe85d4a2007-11-12 21:05:20 -08002726 if (!dev || (dev->flags & IFF_UP)) {
David S. Millerce06b032011-07-04 01:44:29 -07002727 register_prot_hook(sk);
Urs Thuermannbe85d4a2007-11-12 21:05:20 -08002728 } else {
2729 sk->sk_err = ENETDOWN;
2730 if (!sock_flag(sk, SOCK_DEAD))
2731 sk->sk_error_report(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 }
2733
2734out_unlock:
2735 spin_unlock(&po->bind_lock);
2736 release_sock(sk);
2737 return 0;
2738}
2739
2740/*
2741 * Bind a packet socket to a device
2742 */
2743
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002744static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2745 int addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002747 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 char name[15];
2749 struct net_device *dev;
2750 int err = -ENODEV;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002751
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 /*
2753 * Check legality
2754 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002755
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08002756 if (addr_len != sizeof(struct sockaddr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002758 strlcpy(name, uaddr->sa_data, sizeof(name));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002760 dev = dev_get_by_name(sock_net(sk), name);
Ben Greear160ff182011-06-01 07:18:52 +00002761 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 return err;
2764}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765
2766static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2767{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002768 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2769 struct sock *sk = sock->sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 struct net_device *dev = NULL;
2771 int err;
2772
2773
2774 /*
2775 * Check legality
2776 */
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002777
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 if (addr_len < sizeof(struct sockaddr_ll))
2779 return -EINVAL;
2780 if (sll->sll_family != AF_PACKET)
2781 return -EINVAL;
2782
2783 if (sll->sll_ifindex) {
2784 err = -ENODEV;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09002785 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786 if (dev == NULL)
2787 goto out;
2788 }
2789 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790
2791out:
2792 return err;
2793}
2794
2795static struct proto packet_proto = {
2796 .name = "PACKET",
2797 .owner = THIS_MODULE,
2798 .obj_size = sizeof(struct packet_sock),
2799};
2800
2801/*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002802 * Create a packet of type SOCK_PACKET.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 */
2804
Eric Paris3f378b62009-11-05 22:18:14 -08002805static int packet_create(struct net *net, struct socket *sock, int protocol,
2806 int kern)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807{
2808 struct sock *sk;
2809 struct packet_sock *po;
Al Viro0e11c912006-11-08 00:26:29 -08002810 __be16 proto = (__force __be16)protocol; /* weird, but documented */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 int err;
2812
Eric W. Biedermandf008c92012-11-16 03:03:07 +00002813 if (!ns_capable(net->user_ns, CAP_NET_RAW))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 return -EPERM;
David S. Millerbe020972007-05-29 13:16:31 -07002815 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2816 sock->type != SOCK_PACKET)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 return -ESOCKTNOSUPPORT;
2818
2819 sock->state = SS_UNCONNECTED;
2820
2821 err = -ENOBUFS;
Pavel Emelyanov6257ff22007-11-01 00:39:31 -07002822 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002823 if (sk == NULL)
2824 goto out;
2825
2826 sock->ops = &packet_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 if (sock->type == SOCK_PACKET)
2828 sock->ops = &packet_ops_spkt;
David S. Millerbe020972007-05-29 13:16:31 -07002829
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 sock_init_data(sock, sk);
2831
2832 po = pkt_sk(sk);
2833 sk->sk_family = PF_PACKET;
Al Viro0e11c912006-11-08 00:26:29 -08002834 po->num = proto;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01002835 po->xmit = dev_queue_xmit;
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002836
Daniel Borkmannb0138402014-01-15 16:25:36 +01002837 err = packet_alloc_pending(po);
2838 if (err)
2839 goto out2;
2840
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01002841 packet_cached_dev_reset(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002842
2843 sk->sk_destruct = packet_sock_destruct;
Pavel Emelyanov17ab56a2007-11-10 21:38:48 -08002844 sk_refcnt_debug_inc(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845
2846 /*
2847 * Attach a protocol block
2848 */
2849
2850 spin_lock_init(&po->bind_lock);
Herbert Xu905db442009-01-30 14:12:06 -08002851 mutex_init(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 po->prot_hook.func = packet_rcv;
David S. Millerbe020972007-05-29 13:16:31 -07002853
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 if (sock->type == SOCK_PACKET)
2855 po->prot_hook.func = packet_rcv_spkt;
David S. Millerbe020972007-05-29 13:16:31 -07002856
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 po->prot_hook.af_packet_priv = sk;
2858
Al Viro0e11c912006-11-08 00:26:29 -08002859 if (proto) {
2860 po->prot_hook.type = proto;
David S. Millerce06b032011-07-04 01:44:29 -07002861 register_prot_hook(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862 }
2863
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002864 mutex_lock(&net->packet.sklist_lock);
stephen hemminger808f5112010-02-22 07:57:18 +00002865 sk_add_node_rcu(sk, &net->packet.sklist);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002866 mutex_unlock(&net->packet.sklist_lock);
2867
2868 preempt_disable();
Eric Dumazet36804532008-11-19 14:25:35 -08002869 sock_prot_inuse_add(net, &packet_proto, 1);
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00002870 preempt_enable();
stephen hemminger808f5112010-02-22 07:57:18 +00002871
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002872 return 0;
Daniel Borkmannb0138402014-01-15 16:25:36 +01002873out2:
2874 sk_free(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875out:
2876 return err;
2877}
2878
2879/*
2880 * Pull a packet from our receive queue and hand it to the user.
2881 * If necessary we block.
2882 */
2883
Ying Xue1b784142015-03-02 15:37:48 +08002884static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2885 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002886{
2887 struct sock *sk = sock->sk;
2888 struct sk_buff *skb;
2889 int copied, err;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002890 int vnet_hdr_len = 0;
Eyal Birger2472d762015-03-01 14:58:28 +02002891 unsigned int origlen = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892
2893 err = -EINVAL;
Richard Cochraned85b562010-04-07 22:41:28 +00002894 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895 goto out;
2896
2897#if 0
2898 /* What error should we return now? EUNATTACH? */
2899 if (pkt_sk(sk)->ifindex < 0)
2900 return -ENODEV;
2901#endif
2902
Richard Cochraned85b562010-04-07 22:41:28 +00002903 if (flags & MSG_ERRQUEUE) {
Richard Cochrancb820f82013-07-19 19:40:09 +02002904 err = sock_recv_errqueue(sk, msg, len,
2905 SOL_PACKET, PACKET_TX_TIMESTAMP);
Richard Cochraned85b562010-04-07 22:41:28 +00002906 goto out;
2907 }
2908
Linus Torvalds1da177e2005-04-16 15:20:36 -07002909 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 * Call the generic datagram receiver. This handles all sorts
2911 * of horrible races and re-entrancy so we can forget about it
2912 * in the protocol layers.
2913 *
2914 * Now it will return ENETDOWN, if device have just gone down,
2915 * but then it will block.
2916 */
2917
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002918 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002919
2920 /*
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09002921 * An error occurred so return it. Because skb_recv_datagram()
Linus Torvalds1da177e2005-04-16 15:20:36 -07002922 * handles the blocking we don't see and worry about blocking
2923 * retries.
2924 */
2925
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08002926 if (skb == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 goto out;
2928
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002929 if (pkt_sk(sk)->has_vnet_hdr) {
2930 struct virtio_net_hdr vnet_hdr = { 0 };
2931
2932 err = -EINVAL;
2933 vnet_hdr_len = sizeof(vnet_hdr);
Mariusz Kozlowski1f18b712010-11-08 11:58:45 +00002934 if (len < vnet_hdr_len)
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002935 goto out_free;
2936
Mariusz Kozlowski1f18b712010-11-08 11:58:45 +00002937 len -= vnet_hdr_len;
2938
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002939 if (skb_is_gso(skb)) {
2940 struct skb_shared_info *sinfo = skb_shinfo(skb);
2941
2942 /* This is a hint as to how much should be linear. */
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002943 vnet_hdr.hdr_len =
2944 __cpu_to_virtio16(false, skb_headlen(skb));
2945 vnet_hdr.gso_size =
2946 __cpu_to_virtio16(false, sinfo->gso_size);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002947 if (sinfo->gso_type & SKB_GSO_TCPV4)
2948 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2949 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2950 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2951 else if (sinfo->gso_type & SKB_GSO_UDP)
2952 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2953 else if (sinfo->gso_type & SKB_GSO_FCOE)
2954 goto out_free;
2955 else
2956 BUG();
2957 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2958 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2959 } else
2960 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2961
2962 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2963 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
Michael S. Tsirkindc9e5152014-11-23 18:16:19 +02002964 vnet_hdr.csum_start = __cpu_to_virtio16(false,
2965 skb_checksum_start_offset(skb));
2966 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
2967 skb->csum_offset);
Jason Wang10a8d942011-06-10 00:56:17 +00002968 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2969 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002970 } /* else everything is zero */
2971
Al Viro7eab8d92014-04-06 21:51:23 -04002972 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08002973 if (err < 0)
2974 goto out_free;
2975 }
2976
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01002977 /* You lose any data beyond the buffer you gave. If it worries
2978 * a user program they can ask the device for its MTU
2979 * anyway.
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07002980 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981 copied = skb->len;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00002982 if (copied > len) {
2983 copied = len;
2984 msg->msg_flags |= MSG_TRUNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 }
2986
David S. Miller51f3d022014-11-05 16:46:40 -05002987 err = skb_copy_datagram_msg(skb, 0, msg, copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002988 if (err)
2989 goto out_free;
2990
Eyal Birger2472d762015-03-01 14:58:28 +02002991 if (sock->type != SOCK_PACKET) {
2992 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2993
2994 /* Original length was stored in sockaddr_ll fields */
2995 origlen = PACKET_SKB_CB(skb)->sa.origlen;
2996 sll->sll_family = AF_PACKET;
2997 sll->sll_protocol = skb->protocol;
2998 }
2999
Neil Horman3b885782009-10-12 13:26:31 -07003000 sock_recv_ts_and_drops(msg, sk, skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003002 if (msg->msg_name) {
3003 /* If the address length field is there to be filled
3004 * in, we fill it in now.
3005 */
3006 if (sock->type == SOCK_PACKET) {
Steffen Hurrle342dfc32014-01-17 22:53:15 +01003007 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003008 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3009 } else {
3010 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
Eyal Birger2472d762015-03-01 14:58:28 +02003011
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003012 msg->msg_namelen = sll->sll_halen +
3013 offsetof(struct sockaddr_ll, sll_addr);
3014 }
Herbert Xuffbc6112007-02-04 23:33:10 -08003015 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3016 msg->msg_namelen);
Hannes Frederic Sowaf3d33422013-11-21 03:14:22 +01003017 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018
Herbert Xu8dc41942007-02-04 23:31:32 -08003019 if (pkt_sk(sk)->auxdata) {
Herbert Xuffbc6112007-02-04 23:33:10 -08003020 struct tpacket_auxdata aux;
3021
3022 aux.tp_status = TP_STATUS_USER;
3023 if (skb->ip_summed == CHECKSUM_PARTIAL)
3024 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
Eyal Birger2472d762015-03-01 14:58:28 +02003025 aux.tp_len = origlen;
Herbert Xuffbc6112007-02-04 23:33:10 -08003026 aux.tp_snaplen = skb->len;
3027 aux.tp_mac = 0;
Arnaldo Carvalho de Melobbe735e2007-03-10 22:16:10 -03003028 aux.tp_net = skb_network_offset(skb);
Jiri Pirkodf8a39d2015-01-13 17:13:44 +01003029 if (skb_vlan_tag_present(skb)) {
3030 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09003031 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3032 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
Ben Greeara3bcc232011-06-01 06:49:10 +00003033 } else {
3034 aux.tp_vlan_tci = 0;
Atzm Watanabea0cdfcf2013-12-17 22:53:40 +09003035 aux.tp_vlan_tpid = 0;
Ben Greeara3bcc232011-06-01 06:49:10 +00003036 }
Herbert Xuffbc6112007-02-04 23:33:10 -08003037 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
Herbert Xu8dc41942007-02-04 23:31:32 -08003038 }
3039
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 /*
3041 * Free or return the buffer as appropriate. Again this
3042 * hides all the races and re-entrancy issues from us.
3043 */
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003044 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045
3046out_free:
3047 skb_free_datagram(sk, skb);
3048out:
3049 return err;
3050}
3051
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3053 int *uaddr_len, int peer)
3054{
3055 struct net_device *dev;
3056 struct sock *sk = sock->sk;
3057
3058 if (peer)
3059 return -EOPNOTSUPP;
3060
3061 uaddr->sa_family = AF_PACKET;
Daniel Borkmann2dc85bf2013-06-12 16:02:27 +02003062 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
Eric Dumazet654d1f82009-11-02 10:43:32 +01003063 rcu_read_lock();
3064 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3065 if (dev)
Daniel Borkmann2dc85bf2013-06-12 16:02:27 +02003066 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
Eric Dumazet654d1f82009-11-02 10:43:32 +01003067 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 *uaddr_len = sizeof(*uaddr);
3069
3070 return 0;
3071}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003072
3073static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3074 int *uaddr_len, int peer)
3075{
3076 struct net_device *dev;
3077 struct sock *sk = sock->sk;
3078 struct packet_sock *po = pkt_sk(sk);
Cyrill Gorcunov13cfa972009-11-08 05:51:19 +00003079 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
3081 if (peer)
3082 return -EOPNOTSUPP;
3083
3084 sll->sll_family = AF_PACKET;
3085 sll->sll_ifindex = po->ifindex;
3086 sll->sll_protocol = po->num;
Vasiliy Kulikov67286642010-11-10 12:09:10 -08003087 sll->sll_pkttype = 0;
Eric Dumazet654d1f82009-11-02 10:43:32 +01003088 rcu_read_lock();
3089 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 if (dev) {
3091 sll->sll_hatype = dev->type;
3092 sll->sll_halen = dev->addr_len;
3093 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003094 } else {
3095 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3096 sll->sll_halen = 0;
3097 }
Eric Dumazet654d1f82009-11-02 10:43:32 +01003098 rcu_read_unlock();
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003099 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100
3101 return 0;
3102}
3103
Wang Chen2aeb0b82008-07-14 20:49:46 -07003104static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3105 int what)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106{
3107 switch (i->type) {
3108 case PACKET_MR_MULTICAST:
Jiri Pirko11625632010-03-02 20:40:01 +00003109 if (i->alen != dev->addr_len)
3110 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 if (what > 0)
Jiri Pirko22bedad32010-04-01 21:22:57 +00003112 return dev_mc_add(dev, i->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003113 else
Jiri Pirko22bedad32010-04-01 21:22:57 +00003114 return dev_mc_del(dev, i->addr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 break;
3116 case PACKET_MR_PROMISC:
Wang Chen2aeb0b82008-07-14 20:49:46 -07003117 return dev_set_promiscuity(dev, what);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 case PACKET_MR_ALLMULTI:
Wang Chen2aeb0b82008-07-14 20:49:46 -07003119 return dev_set_allmulti(dev, what);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003120 case PACKET_MR_UNICAST:
Jiri Pirko11625632010-03-02 20:40:01 +00003121 if (i->alen != dev->addr_len)
3122 return -EINVAL;
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003123 if (what > 0)
Jiri Pirkoa748ee22010-04-01 21:22:09 +00003124 return dev_uc_add(dev, i->addr);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003125 else
Jiri Pirkoa748ee22010-04-01 21:22:09 +00003126 return dev_uc_del(dev, i->addr);
Eric W. Biedermand95ed922009-05-19 18:27:17 +00003127 break;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003128 default:
3129 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 }
Wang Chen2aeb0b82008-07-14 20:49:46 -07003131 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132}
3133
3134static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3135{
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003136 for ( ; i; i = i->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137 if (i->ifindex == dev->ifindex)
3138 packet_dev_mc(dev, i, what);
3139 }
3140}
3141
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003142static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003143{
3144 struct packet_sock *po = pkt_sk(sk);
3145 struct packet_mclist *ml, *i;
3146 struct net_device *dev;
3147 int err;
3148
3149 rtnl_lock();
3150
3151 err = -ENODEV;
YOSHIFUJI Hideaki3b1e0a62008-03-26 02:26:21 +09003152 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 if (!dev)
3154 goto done;
3155
3156 err = -EINVAL;
Jiri Pirko11625632010-03-02 20:40:01 +00003157 if (mreq->mr_alen > dev->addr_len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 goto done;
3159
3160 err = -ENOBUFS;
Kris Katterjohn8b3a7002006-01-11 15:56:43 -08003161 i = kmalloc(sizeof(*i), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 if (i == NULL)
3163 goto done;
3164
3165 err = 0;
3166 for (ml = po->mclist; ml; ml = ml->next) {
3167 if (ml->ifindex == mreq->mr_ifindex &&
3168 ml->type == mreq->mr_type &&
3169 ml->alen == mreq->mr_alen &&
3170 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3171 ml->count++;
3172 /* Free the new element ... */
3173 kfree(i);
3174 goto done;
3175 }
3176 }
3177
3178 i->type = mreq->mr_type;
3179 i->ifindex = mreq->mr_ifindex;
3180 i->alen = mreq->mr_alen;
3181 memcpy(i->addr, mreq->mr_address, i->alen);
3182 i->count = 1;
3183 i->next = po->mclist;
3184 po->mclist = i;
Wang Chen2aeb0b82008-07-14 20:49:46 -07003185 err = packet_dev_mc(dev, i, 1);
3186 if (err) {
3187 po->mclist = i->next;
3188 kfree(i);
3189 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003190
3191done:
3192 rtnl_unlock();
3193 return err;
3194}
3195
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003196static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197{
3198 struct packet_mclist *ml, **mlp;
3199
3200 rtnl_lock();
3201
3202 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3203 if (ml->ifindex == mreq->mr_ifindex &&
3204 ml->type == mreq->mr_type &&
3205 ml->alen == mreq->mr_alen &&
3206 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3207 if (--ml->count == 0) {
3208 struct net_device *dev;
3209 *mlp = ml->next;
Eric Dumazetad959e72009-10-16 06:38:46 +00003210 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3211 if (dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 packet_dev_mc(dev, ml, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003213 kfree(ml);
3214 }
3215 rtnl_unlock();
3216 return 0;
3217 }
3218 }
3219 rtnl_unlock();
3220 return -EADDRNOTAVAIL;
3221}
3222
3223static void packet_flush_mclist(struct sock *sk)
3224{
3225 struct packet_sock *po = pkt_sk(sk);
3226 struct packet_mclist *ml;
3227
3228 if (!po->mclist)
3229 return;
3230
3231 rtnl_lock();
3232 while ((ml = po->mclist) != NULL) {
3233 struct net_device *dev;
3234
3235 po->mclist = ml->next;
Eric Dumazetad959e72009-10-16 06:38:46 +00003236 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3237 if (dev != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238 packet_dev_mc(dev, ml, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 kfree(ml);
3240 }
3241 rtnl_unlock();
3242}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243
3244static int
David S. Millerb7058842009-09-30 16:12:20 -07003245packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246{
3247 struct sock *sk = sock->sk;
Herbert Xu8dc41942007-02-04 23:31:32 -08003248 struct packet_sock *po = pkt_sk(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 int ret;
3250
3251 if (level != SOL_PACKET)
3252 return -ENOPROTOOPT;
3253
Johann Baudy69e3c752009-05-18 22:11:22 -07003254 switch (optname) {
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003255 case PACKET_ADD_MEMBERSHIP:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 case PACKET_DROP_MEMBERSHIP:
3257 {
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003258 struct packet_mreq_max mreq;
3259 int len = optlen;
3260 memset(&mreq, 0, sizeof(mreq));
3261 if (len < sizeof(struct packet_mreq))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 return -EINVAL;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003263 if (len > sizeof(mreq))
3264 len = sizeof(mreq);
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003265 if (copy_from_user(&mreq, optval, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 return -EFAULT;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07003267 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3268 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269 if (optname == PACKET_ADD_MEMBERSHIP)
3270 ret = packet_mc_add(sk, &mreq);
3271 else
3272 ret = packet_mc_drop(sk, &mreq);
3273 return ret;
3274 }
David S. Millera2efcfa2007-05-29 13:12:50 -07003275
Linus Torvalds1da177e2005-04-16 15:20:36 -07003276 case PACKET_RX_RING:
Johann Baudy69e3c752009-05-18 22:11:22 -07003277 case PACKET_TX_RING:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278 {
chetan lokef6fb8f102011-08-19 10:18:16 +00003279 union tpacket_req_u req_u;
3280 int len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281
chetan lokef6fb8f102011-08-19 10:18:16 +00003282 switch (po->tp_version) {
3283 case TPACKET_V1:
3284 case TPACKET_V2:
3285 len = sizeof(req_u.req);
3286 break;
3287 case TPACKET_V3:
3288 default:
3289 len = sizeof(req_u.req3);
3290 break;
3291 }
3292 if (optlen < len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003293 return -EINVAL;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003294 if (pkt_sk(sk)->has_vnet_hdr)
3295 return -EINVAL;
chetan lokef6fb8f102011-08-19 10:18:16 +00003296 if (copy_from_user(&req_u.req, optval, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297 return -EFAULT;
chetan lokef6fb8f102011-08-19 10:18:16 +00003298 return packet_set_ring(sk, &req_u, 0,
3299 optname == PACKET_TX_RING);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 }
3301 case PACKET_COPY_THRESH:
3302 {
3303 int val;
3304
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003305 if (optlen != sizeof(val))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306 return -EINVAL;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003307 if (copy_from_user(&val, optval, sizeof(val)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 return -EFAULT;
3309
3310 pkt_sk(sk)->copy_thresh = val;
3311 return 0;
3312 }
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003313 case PACKET_VERSION:
3314 {
3315 int val;
3316
3317 if (optlen != sizeof(val))
3318 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07003319 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003320 return -EBUSY;
3321 if (copy_from_user(&val, optval, sizeof(val)))
3322 return -EFAULT;
3323 switch (val) {
3324 case TPACKET_V1:
3325 case TPACKET_V2:
chetan lokef6fb8f102011-08-19 10:18:16 +00003326 case TPACKET_V3:
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003327 po->tp_version = val;
3328 return 0;
3329 default:
3330 return -EINVAL;
3331 }
3332 }
Patrick McHardy8913336a2008-07-18 18:05:19 -07003333 case PACKET_RESERVE:
3334 {
3335 unsigned int val;
3336
3337 if (optlen != sizeof(val))
3338 return -EINVAL;
Johann Baudy69e3c752009-05-18 22:11:22 -07003339 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
Patrick McHardy8913336a2008-07-18 18:05:19 -07003340 return -EBUSY;
3341 if (copy_from_user(&val, optval, sizeof(val)))
3342 return -EFAULT;
3343 po->tp_reserve = val;
3344 return 0;
3345 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003346 case PACKET_LOSS:
3347 {
3348 unsigned int val;
3349
3350 if (optlen != sizeof(val))
3351 return -EINVAL;
3352 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3353 return -EBUSY;
3354 if (copy_from_user(&val, optval, sizeof(val)))
3355 return -EFAULT;
3356 po->tp_loss = !!val;
3357 return 0;
3358 }
Herbert Xu8dc41942007-02-04 23:31:32 -08003359 case PACKET_AUXDATA:
3360 {
3361 int val;
3362
3363 if (optlen < sizeof(val))
3364 return -EINVAL;
3365 if (copy_from_user(&val, optval, sizeof(val)))
3366 return -EFAULT;
3367
3368 po->auxdata = !!val;
3369 return 0;
3370 }
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003371 case PACKET_ORIGDEV:
3372 {
3373 int val;
3374
3375 if (optlen < sizeof(val))
3376 return -EINVAL;
3377 if (copy_from_user(&val, optval, sizeof(val)))
3378 return -EFAULT;
3379
3380 po->origdev = !!val;
3381 return 0;
3382 }
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003383 case PACKET_VNET_HDR:
3384 {
3385 int val;
3386
3387 if (sock->type != SOCK_RAW)
3388 return -EINVAL;
3389 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3390 return -EBUSY;
3391 if (optlen < sizeof(val))
3392 return -EINVAL;
3393 if (copy_from_user(&val, optval, sizeof(val)))
3394 return -EFAULT;
3395
3396 po->has_vnet_hdr = !!val;
3397 return 0;
3398 }
Scott McMillan614f60f2010-06-02 05:53:56 -07003399 case PACKET_TIMESTAMP:
3400 {
3401 int val;
3402
3403 if (optlen != sizeof(val))
3404 return -EINVAL;
3405 if (copy_from_user(&val, optval, sizeof(val)))
3406 return -EFAULT;
3407
3408 po->tp_tstamp = val;
3409 return 0;
3410 }
David S. Millerdc99f602011-07-05 01:45:05 -07003411 case PACKET_FANOUT:
3412 {
3413 int val;
3414
3415 if (optlen != sizeof(val))
3416 return -EINVAL;
3417 if (copy_from_user(&val, optval, sizeof(val)))
3418 return -EFAULT;
3419
3420 return fanout_add(sk, val & 0xffff, val >> 16);
3421 }
Paul Chavent5920cd3a2012-11-06 23:10:47 +00003422 case PACKET_TX_HAS_OFF:
3423 {
3424 unsigned int val;
3425
3426 if (optlen != sizeof(val))
3427 return -EINVAL;
3428 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3429 return -EBUSY;
3430 if (copy_from_user(&val, optval, sizeof(val)))
3431 return -EFAULT;
3432 po->tp_tx_has_off = !!val;
3433 return 0;
3434 }
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01003435 case PACKET_QDISC_BYPASS:
3436 {
3437 int val;
3438
3439 if (optlen != sizeof(val))
3440 return -EINVAL;
3441 if (copy_from_user(&val, optval, sizeof(val)))
3442 return -EFAULT;
3443
3444 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3445 return 0;
3446 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003447 default:
3448 return -ENOPROTOOPT;
3449 }
3450}
3451
3452static int packet_getsockopt(struct socket *sock, int level, int optname,
3453 char __user *optval, int __user *optlen)
3454{
3455 int len;
Eric Dumazetc06fff62012-04-19 21:56:11 +00003456 int val, lv = sizeof(val);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457 struct sock *sk = sock->sk;
3458 struct packet_sock *po = pkt_sk(sk);
Eric Dumazetc06fff62012-04-19 21:56:11 +00003459 void *data = &val;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003460 union tpacket_stats_u st;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461
3462 if (level != SOL_PACKET)
3463 return -ENOPROTOOPT;
3464
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003465 if (get_user(len, optlen))
3466 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003467
3468 if (len < 0)
3469 return -EINVAL;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003470
Johann Baudy69e3c752009-05-18 22:11:22 -07003471 switch (optname) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472 case PACKET_STATISTICS:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003473 spin_lock_bh(&sk->sk_receive_queue.lock);
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003474 memcpy(&st, &po->stats, sizeof(st));
3475 memset(&po->stats, 0, sizeof(po->stats));
3476 spin_unlock_bh(&sk->sk_receive_queue.lock);
3477
chetan lokef6fb8f102011-08-19 10:18:16 +00003478 if (po->tp_version == TPACKET_V3) {
Eric Dumazetc06fff62012-04-19 21:56:11 +00003479 lv = sizeof(struct tpacket_stats_v3);
Willem de Bruijn8bcdeaf2013-08-19 16:40:22 -04003480 st.stats3.tp_packets += st.stats3.tp_drops;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003481 data = &st.stats3;
chetan lokef6fb8f102011-08-19 10:18:16 +00003482 } else {
Eric Dumazetc06fff62012-04-19 21:56:11 +00003483 lv = sizeof(struct tpacket_stats);
Willem de Bruijn8bcdeaf2013-08-19 16:40:22 -04003484 st.stats1.tp_packets += st.stats1.tp_drops;
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003485 data = &st.stats1;
chetan lokef6fb8f102011-08-19 10:18:16 +00003486 }
Daniel Borkmannee80fbf2013-04-19 06:12:29 +00003487
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488 break;
Herbert Xu8dc41942007-02-04 23:31:32 -08003489 case PACKET_AUXDATA:
Herbert Xu8dc41942007-02-04 23:31:32 -08003490 val = po->auxdata;
Herbert Xu8dc41942007-02-04 23:31:32 -08003491 break;
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003492 case PACKET_ORIGDEV:
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003493 val = po->origdev;
Peter P. Waskiewicz Jr80feaac2007-04-20 16:05:39 -07003494 break;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003495 case PACKET_VNET_HDR:
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003496 val = po->has_vnet_hdr;
Sridhar Samudralabfd5f4a2010-02-04 20:24:10 -08003497 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003498 case PACKET_VERSION:
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003499 val = po->tp_version;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003500 break;
3501 case PACKET_HDRLEN:
3502 if (len > sizeof(int))
3503 len = sizeof(int);
3504 if (copy_from_user(&val, optval, len))
3505 return -EFAULT;
3506 switch (val) {
3507 case TPACKET_V1:
3508 val = sizeof(struct tpacket_hdr);
3509 break;
3510 case TPACKET_V2:
3511 val = sizeof(struct tpacket2_hdr);
3512 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00003513 case TPACKET_V3:
3514 val = sizeof(struct tpacket3_hdr);
3515 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003516 default:
3517 return -EINVAL;
3518 }
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003519 break;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003520 case PACKET_RESERVE:
Patrick McHardy8913336a2008-07-18 18:05:19 -07003521 val = po->tp_reserve;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003522 break;
Johann Baudy69e3c752009-05-18 22:11:22 -07003523 case PACKET_LOSS:
Johann Baudy69e3c752009-05-18 22:11:22 -07003524 val = po->tp_loss;
Johann Baudy69e3c752009-05-18 22:11:22 -07003525 break;
Scott McMillan614f60f2010-06-02 05:53:56 -07003526 case PACKET_TIMESTAMP:
Scott McMillan614f60f2010-06-02 05:53:56 -07003527 val = po->tp_tstamp;
Scott McMillan614f60f2010-06-02 05:53:56 -07003528 break;
David S. Millerdc99f602011-07-05 01:45:05 -07003529 case PACKET_FANOUT:
David S. Millerdc99f602011-07-05 01:45:05 -07003530 val = (po->fanout ?
3531 ((u32)po->fanout->id |
Willem de Bruijn77f65eb2013-03-19 10:18:11 +00003532 ((u32)po->fanout->type << 16) |
3533 ((u32)po->fanout->flags << 24)) :
David S. Millerdc99f602011-07-05 01:45:05 -07003534 0);
David S. Millerdc99f602011-07-05 01:45:05 -07003535 break;
Paul Chavent5920cd3a2012-11-06 23:10:47 +00003536 case PACKET_TX_HAS_OFF:
3537 val = po->tp_tx_has_off;
3538 break;
Daniel Borkmannd346a3f2013-12-06 11:36:17 +01003539 case PACKET_QDISC_BYPASS:
3540 val = packet_use_direct_xmit(po);
3541 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003542 default:
3543 return -ENOPROTOOPT;
3544 }
3545
Eric Dumazetc06fff62012-04-19 21:56:11 +00003546 if (len > lv)
3547 len = lv;
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003548 if (put_user(len, optlen))
3549 return -EFAULT;
Herbert Xu8dc41942007-02-04 23:31:32 -08003550 if (copy_to_user(optval, data, len))
3551 return -EFAULT;
Kris Katterjohn8ae55f02006-01-23 16:28:02 -08003552 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003553}
3554
3555
Jiri Pirko351638e2013-05-28 01:30:21 +00003556static int packet_notifier(struct notifier_block *this,
3557 unsigned long msg, void *ptr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003558{
3559 struct sock *sk;
Jiri Pirko351638e2013-05-28 01:30:21 +00003560 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
YOSHIFUJI Hideakic346dca2008-03-25 21:47:49 +09003561 struct net *net = dev_net(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003562
stephen hemminger808f5112010-02-22 07:57:18 +00003563 rcu_read_lock();
Sasha Levinb67bfe02013-02-27 17:06:00 -08003564 sk_for_each_rcu(sk, &net->packet.sklist) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565 struct packet_sock *po = pkt_sk(sk);
3566
3567 switch (msg) {
3568 case NETDEV_UNREGISTER:
Linus Torvalds1da177e2005-04-16 15:20:36 -07003569 if (po->mclist)
3570 packet_dev_mclist(dev, po->mclist, -1);
David S. Millera2efcfa2007-05-29 13:12:50 -07003571 /* fallthrough */
3572
Linus Torvalds1da177e2005-04-16 15:20:36 -07003573 case NETDEV_DOWN:
3574 if (dev->ifindex == po->ifindex) {
3575 spin_lock(&po->bind_lock);
3576 if (po->running) {
David S. Millerce06b032011-07-04 01:44:29 -07003577 __unregister_prot_hook(sk, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003578 sk->sk_err = ENETDOWN;
3579 if (!sock_flag(sk, SOCK_DEAD))
3580 sk->sk_error_report(sk);
3581 }
3582 if (msg == NETDEV_UNREGISTER) {
Daniel Borkmann66e56cd2013-12-06 11:36:15 +01003583 packet_cached_dev_reset(po);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003584 po->ifindex = -1;
Ben Greear160ff182011-06-01 07:18:52 +00003585 if (po->prot_hook.dev)
3586 dev_put(po->prot_hook.dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003587 po->prot_hook.dev = NULL;
3588 }
3589 spin_unlock(&po->bind_lock);
3590 }
3591 break;
3592 case NETDEV_UP:
stephen hemminger808f5112010-02-22 07:57:18 +00003593 if (dev->ifindex == po->ifindex) {
3594 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07003595 if (po->num)
3596 register_prot_hook(sk);
stephen hemminger808f5112010-02-22 07:57:18 +00003597 spin_unlock(&po->bind_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003598 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003599 break;
3600 }
3601 }
stephen hemminger808f5112010-02-22 07:57:18 +00003602 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003603 return NOTIFY_DONE;
3604}
3605
3606
3607static int packet_ioctl(struct socket *sock, unsigned int cmd,
3608 unsigned long arg)
3609{
3610 struct sock *sk = sock->sk;
3611
Johann Baudy69e3c752009-05-18 22:11:22 -07003612 switch (cmd) {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003613 case SIOCOUTQ:
3614 {
3615 int amount = sk_wmem_alloc_get(sk);
Eric Dumazet31e6d362009-06-17 19:05:41 -07003616
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003617 return put_user(amount, (int __user *)arg);
3618 }
3619 case SIOCINQ:
3620 {
3621 struct sk_buff *skb;
3622 int amount = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003623
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003624 spin_lock_bh(&sk->sk_receive_queue.lock);
3625 skb = skb_peek(&sk->sk_receive_queue);
3626 if (skb)
3627 amount = skb->len;
3628 spin_unlock_bh(&sk->sk_receive_queue.lock);
3629 return put_user(amount, (int __user *)arg);
3630 }
3631 case SIOCGSTAMP:
3632 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3633 case SIOCGSTAMPNS:
3634 return sock_get_timestampns(sk, (struct timespec __user *)arg);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003635
Linus Torvalds1da177e2005-04-16 15:20:36 -07003636#ifdef CONFIG_INET
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003637 case SIOCADDRT:
3638 case SIOCDELRT:
3639 case SIOCDARP:
3640 case SIOCGARP:
3641 case SIOCSARP:
3642 case SIOCGIFADDR:
3643 case SIOCSIFADDR:
3644 case SIOCGIFBRDADDR:
3645 case SIOCSIFBRDADDR:
3646 case SIOCGIFNETMASK:
3647 case SIOCSIFNETMASK:
3648 case SIOCGIFDSTADDR:
3649 case SIOCSIFDSTADDR:
3650 case SIOCSIFFLAGS:
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003651 return inet_dgram_ops.ioctl(sock, cmd, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003652#endif
3653
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003654 default:
3655 return -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003656 }
3657 return 0;
3658}
3659
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003660static unsigned int packet_poll(struct file *file, struct socket *sock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003661 poll_table *wait)
3662{
3663 struct sock *sk = sock->sk;
3664 struct packet_sock *po = pkt_sk(sk);
3665 unsigned int mask = datagram_poll(file, sock, wait);
3666
3667 spin_lock_bh(&sk->sk_receive_queue.lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003668 if (po->rx_ring.pg_vec) {
chetan lokef6fb8f102011-08-19 10:18:16 +00003669 if (!packet_previous_rx_frame(po, &po->rx_ring,
3670 TP_STATUS_KERNEL))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003671 mask |= POLLIN | POLLRDNORM;
3672 }
3673 spin_unlock_bh(&sk->sk_receive_queue.lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003674 spin_lock_bh(&sk->sk_write_queue.lock);
3675 if (po->tx_ring.pg_vec) {
3676 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3677 mask |= POLLOUT | POLLWRNORM;
3678 }
3679 spin_unlock_bh(&sk->sk_write_queue.lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003680 return mask;
3681}
3682
3683
3684/* Dirty? Well, I still did not learn better way to account
3685 * for user mmaps.
3686 */
3687
3688static void packet_mm_open(struct vm_area_struct *vma)
3689{
3690 struct file *file = vma->vm_file;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003691 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003692 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003693
Linus Torvalds1da177e2005-04-16 15:20:36 -07003694 if (sk)
3695 atomic_inc(&pkt_sk(sk)->mapped);
3696}
3697
3698static void packet_mm_close(struct vm_area_struct *vma)
3699{
3700 struct file *file = vma->vm_file;
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003701 struct socket *sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003702 struct sock *sk = sock->sk;
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003703
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704 if (sk)
3705 atomic_dec(&pkt_sk(sk)->mapped);
3706}
3707
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04003708static const struct vm_operations_struct packet_mmap_ops = {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003709 .open = packet_mm_open,
3710 .close = packet_mm_close,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003711};
3712
Neil Horman0e3125c2010-11-16 10:26:47 -08003713static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3714 unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003715{
3716 int i;
3717
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003718 for (i = 0; i < len; i++) {
Neil Horman0e3125c2010-11-16 10:26:47 -08003719 if (likely(pg_vec[i].buffer)) {
Changli Gaoc56b4d92010-12-01 02:52:57 +00003720 if (is_vmalloc_addr(pg_vec[i].buffer))
Neil Horman0e3125c2010-11-16 10:26:47 -08003721 vfree(pg_vec[i].buffer);
3722 else
3723 free_pages((unsigned long)pg_vec[i].buffer,
3724 order);
3725 pg_vec[i].buffer = NULL;
3726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003727 }
3728 kfree(pg_vec);
3729}
3730
Olof Johanssoneea49cc92011-11-02 11:00:49 +00003731static char *alloc_one_pg_vec_page(unsigned long order)
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003732{
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003733 char *buffer;
Neil Horman0e3125c2010-11-16 10:26:47 -08003734 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3735 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
Eric Dumazet719bfea2009-04-15 03:39:52 -07003736
Neil Horman0e3125c2010-11-16 10:26:47 -08003737 buffer = (char *) __get_free_pages(gfp_flags, order);
Neil Horman0e3125c2010-11-16 10:26:47 -08003738 if (buffer)
3739 return buffer;
3740
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003741 /* __get_free_pages failed, fall back to vmalloc */
Eric Dumazetbbce5a52010-11-20 07:31:54 +00003742 buffer = vzalloc((1 << order) * PAGE_SIZE);
Neil Horman0e3125c2010-11-16 10:26:47 -08003743 if (buffer)
3744 return buffer;
3745
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003746 /* vmalloc failed, lets dig into swap here */
Neil Horman0e3125c2010-11-16 10:26:47 -08003747 gfp_flags &= ~__GFP_NORETRY;
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003748 buffer = (char *) __get_free_pages(gfp_flags, order);
Neil Horman0e3125c2010-11-16 10:26:47 -08003749 if (buffer)
3750 return buffer;
3751
Daniel Borkmannf0d4eb22014-01-19 11:46:53 +01003752 /* complete and utter failure */
Neil Horman0e3125c2010-11-16 10:26:47 -08003753 return NULL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003754}
3755
Neil Horman0e3125c2010-11-16 10:26:47 -08003756static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003757{
3758 unsigned int block_nr = req->tp_block_nr;
Neil Horman0e3125c2010-11-16 10:26:47 -08003759 struct pgv *pg_vec;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003760 int i;
3761
Neil Horman0e3125c2010-11-16 10:26:47 -08003762 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003763 if (unlikely(!pg_vec))
3764 goto out;
3765
3766 for (i = 0; i < block_nr; i++) {
Changli Gaoc56b4d92010-12-01 02:52:57 +00003767 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
Neil Horman0e3125c2010-11-16 10:26:47 -08003768 if (unlikely(!pg_vec[i].buffer))
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003769 goto out_free_pgvec;
3770 }
3771
3772out:
3773 return pg_vec;
3774
3775out_free_pgvec:
3776 free_pg_vec(pg_vec, order, block_nr);
3777 pg_vec = NULL;
3778 goto out;
3779}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003780
chetan lokef6fb8f102011-08-19 10:18:16 +00003781static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
Johann Baudy69e3c752009-05-18 22:11:22 -07003782 int closing, int tx_ring)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783{
Neil Horman0e3125c2010-11-16 10:26:47 -08003784 struct pgv *pg_vec = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003785 struct packet_sock *po = pkt_sk(sk);
Al Viro0e11c912006-11-08 00:26:29 -08003786 int was_running, order = 0;
Johann Baudy69e3c752009-05-18 22:11:22 -07003787 struct packet_ring_buffer *rb;
3788 struct sk_buff_head *rb_queue;
Al Viro0e11c912006-11-08 00:26:29 -08003789 __be16 num;
chetan lokef6fb8f102011-08-19 10:18:16 +00003790 int err = -EINVAL;
3791 /* Added to avoid minimal code churn */
3792 struct tpacket_req *req = &req_u->req;
3793
3794 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3795 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3796 WARN(1, "Tx-ring is not supported.\n");
3797 goto out;
3798 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003799
3800 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3801 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3802
3803 err = -EBUSY;
3804 if (!closing) {
3805 if (atomic_read(&po->mapped))
3806 goto out;
Daniel Borkmannb0138402014-01-15 16:25:36 +01003807 if (packet_read_pending(rb))
Johann Baudy69e3c752009-05-18 22:11:22 -07003808 goto out;
3809 }
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003810
Linus Torvalds1da177e2005-04-16 15:20:36 -07003811 if (req->tp_block_nr) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812 /* Sanity tests and some calculations */
Johann Baudy69e3c752009-05-18 22:11:22 -07003813 err = -EBUSY;
3814 if (unlikely(rb->pg_vec))
3815 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003816
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003817 switch (po->tp_version) {
3818 case TPACKET_V1:
3819 po->tp_hdrlen = TPACKET_HDRLEN;
3820 break;
3821 case TPACKET_V2:
3822 po->tp_hdrlen = TPACKET2_HDRLEN;
3823 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00003824 case TPACKET_V3:
3825 po->tp_hdrlen = TPACKET3_HDRLEN;
3826 break;
Patrick McHardybbd6ef82008-07-14 22:50:15 -07003827 }
3828
Johann Baudy69e3c752009-05-18 22:11:22 -07003829 err = -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003830 if (unlikely((int)req->tp_block_size <= 0))
Johann Baudy69e3c752009-05-18 22:11:22 -07003831 goto out;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003832 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
Johann Baudy69e3c752009-05-18 22:11:22 -07003833 goto out;
Eric Dumazetdc808112014-08-15 09:16:04 -07003834 if (po->tp_version >= TPACKET_V3 &&
3835 (int)(req->tp_block_size -
3836 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3837 goto out;
Patrick McHardy8913336a2008-07-18 18:05:19 -07003838 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
Johann Baudy69e3c752009-05-18 22:11:22 -07003839 po->tp_reserve))
3840 goto out;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003841 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
Johann Baudy69e3c752009-05-18 22:11:22 -07003842 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843
Johann Baudy69e3c752009-05-18 22:11:22 -07003844 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3845 if (unlikely(rb->frames_per_block <= 0))
3846 goto out;
3847 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3848 req->tp_frame_nr))
3849 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003850
3851 err = -ENOMEM;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003852 order = get_order(req->tp_block_size);
3853 pg_vec = alloc_pg_vec(req, order);
3854 if (unlikely(!pg_vec))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003855 goto out;
chetan lokef6fb8f102011-08-19 10:18:16 +00003856 switch (po->tp_version) {
3857 case TPACKET_V3:
3858 /* Transmit path is not supported. We checked
3859 * it above but just being paranoid
3860 */
3861 if (!tx_ring)
3862 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
Dan Carpenterd7cf0c32014-02-18 15:20:51 +03003863 break;
chetan lokef6fb8f102011-08-19 10:18:16 +00003864 default:
3865 break;
3866 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003867 }
3868 /* Done */
3869 else {
3870 err = -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003871 if (unlikely(req->tp_frame_nr))
Johann Baudy69e3c752009-05-18 22:11:22 -07003872 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003873 }
3874
3875 lock_sock(sk);
3876
3877 /* Detach socket from network */
3878 spin_lock(&po->bind_lock);
3879 was_running = po->running;
3880 num = po->num;
3881 if (was_running) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003882 po->num = 0;
David S. Millerce06b032011-07-04 01:44:29 -07003883 __unregister_prot_hook(sk, false);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003884 }
3885 spin_unlock(&po->bind_lock);
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09003886
Linus Torvalds1da177e2005-04-16 15:20:36 -07003887 synchronize_net();
3888
3889 err = -EBUSY;
Herbert Xu905db442009-01-30 14:12:06 -08003890 mutex_lock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003891 if (closing || atomic_read(&po->mapped) == 0) {
3892 err = 0;
Johann Baudy69e3c752009-05-18 22:11:22 -07003893 spin_lock_bh(&rb_queue->lock);
Changli Gaoc053fd92010-12-10 16:02:20 -08003894 swap(rb->pg_vec, pg_vec);
Johann Baudy69e3c752009-05-18 22:11:22 -07003895 rb->frame_max = (req->tp_frame_nr - 1);
3896 rb->head = 0;
3897 rb->frame_size = req->tp_frame_size;
3898 spin_unlock_bh(&rb_queue->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003899
Changli Gaoc053fd92010-12-10 16:02:20 -08003900 swap(rb->pg_vec_order, order);
3901 swap(rb->pg_vec_len, req->tp_block_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902
Johann Baudy69e3c752009-05-18 22:11:22 -07003903 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3904 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3905 tpacket_rcv : packet_rcv;
3906 skb_queue_purge(rb_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003907 if (atomic_read(&po->mapped))
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00003908 pr_err("packet_mmap: vma is busy: %d\n",
3909 atomic_read(&po->mapped));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910 }
Herbert Xu905db442009-01-30 14:12:06 -08003911 mutex_unlock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003912
3913 spin_lock(&po->bind_lock);
David S. Millerce06b032011-07-04 01:44:29 -07003914 if (was_running) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915 po->num = num;
David S. Millerce06b032011-07-04 01:44:29 -07003916 register_prot_hook(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003917 }
3918 spin_unlock(&po->bind_lock);
chetan lokef6fb8f102011-08-19 10:18:16 +00003919 if (closing && (po->tp_version > TPACKET_V2)) {
3920 /* Because we don't support block-based V3 on tx-ring */
3921 if (!tx_ring)
3922 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3923 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003924 release_sock(sk);
3925
Linus Torvalds1da177e2005-04-16 15:20:36 -07003926 if (pg_vec)
3927 free_pg_vec(pg_vec, order, req->tp_block_nr);
3928out:
3929 return err;
3930}
3931
Johann Baudy69e3c752009-05-18 22:11:22 -07003932static int packet_mmap(struct file *file, struct socket *sock,
3933 struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934{
3935 struct sock *sk = sock->sk;
3936 struct packet_sock *po = pkt_sk(sk);
Johann Baudy69e3c752009-05-18 22:11:22 -07003937 unsigned long size, expected_size;
3938 struct packet_ring_buffer *rb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003939 unsigned long start;
3940 int err = -EINVAL;
3941 int i;
3942
3943 if (vma->vm_pgoff)
3944 return -EINVAL;
3945
Herbert Xu905db442009-01-30 14:12:06 -08003946 mutex_lock(&po->pg_vec_lock);
Johann Baudy69e3c752009-05-18 22:11:22 -07003947
3948 expected_size = 0;
3949 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3950 if (rb->pg_vec) {
3951 expected_size += rb->pg_vec_len
3952 * rb->pg_vec_pages
3953 * PAGE_SIZE;
3954 }
3955 }
3956
3957 if (expected_size == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003958 goto out;
Johann Baudy69e3c752009-05-18 22:11:22 -07003959
3960 size = vma->vm_end - vma->vm_start;
3961 if (size != expected_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003962 goto out;
3963
Linus Torvalds1da177e2005-04-16 15:20:36 -07003964 start = vma->vm_start;
Johann Baudy69e3c752009-05-18 22:11:22 -07003965 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3966 if (rb->pg_vec == NULL)
3967 continue;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003968
Johann Baudy69e3c752009-05-18 22:11:22 -07003969 for (i = 0; i < rb->pg_vec_len; i++) {
Neil Horman0e3125c2010-11-16 10:26:47 -08003970 struct page *page;
3971 void *kaddr = rb->pg_vec[i].buffer;
Johann Baudy69e3c752009-05-18 22:11:22 -07003972 int pg_num;
3973
Changli Gaoc56b4d92010-12-01 02:52:57 +00003974 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3975 page = pgv_to_page(kaddr);
Johann Baudy69e3c752009-05-18 22:11:22 -07003976 err = vm_insert_page(vma, start, page);
3977 if (unlikely(err))
3978 goto out;
3979 start += PAGE_SIZE;
Neil Horman0e3125c2010-11-16 10:26:47 -08003980 kaddr += PAGE_SIZE;
Johann Baudy69e3c752009-05-18 22:11:22 -07003981 }
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003982 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003983 }
Johann Baudy69e3c752009-05-18 22:11:22 -07003984
David S. Miller4ebf0ae2005-12-06 16:38:35 -08003985 atomic_inc(&po->mapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003986 vma->vm_ops = &packet_mmap_ops;
3987 err = 0;
3988
3989out:
Herbert Xu905db442009-01-30 14:12:06 -08003990 mutex_unlock(&po->pg_vec_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003991 return err;
3992}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003993
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08003994static const struct proto_ops packet_ops_spkt = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003995 .family = PF_PACKET,
3996 .owner = THIS_MODULE,
3997 .release = packet_release,
3998 .bind = packet_bind_spkt,
3999 .connect = sock_no_connect,
4000 .socketpair = sock_no_socketpair,
4001 .accept = sock_no_accept,
4002 .getname = packet_getname_spkt,
4003 .poll = datagram_poll,
4004 .ioctl = packet_ioctl,
4005 .listen = sock_no_listen,
4006 .shutdown = sock_no_shutdown,
4007 .setsockopt = sock_no_setsockopt,
4008 .getsockopt = sock_no_getsockopt,
4009 .sendmsg = packet_sendmsg_spkt,
4010 .recvmsg = packet_recvmsg,
4011 .mmap = sock_no_mmap,
4012 .sendpage = sock_no_sendpage,
4013};
Linus Torvalds1da177e2005-04-16 15:20:36 -07004014
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08004015static const struct proto_ops packet_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004016 .family = PF_PACKET,
4017 .owner = THIS_MODULE,
4018 .release = packet_release,
4019 .bind = packet_bind,
4020 .connect = sock_no_connect,
4021 .socketpair = sock_no_socketpair,
4022 .accept = sock_no_accept,
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09004023 .getname = packet_getname,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004024 .poll = packet_poll,
4025 .ioctl = packet_ioctl,
4026 .listen = sock_no_listen,
4027 .shutdown = sock_no_shutdown,
4028 .setsockopt = packet_setsockopt,
4029 .getsockopt = packet_getsockopt,
4030 .sendmsg = packet_sendmsg,
4031 .recvmsg = packet_recvmsg,
4032 .mmap = packet_mmap,
4033 .sendpage = sock_no_sendpage,
4034};
4035
Stephen Hemmingerec1b4cf2009-10-05 05:58:39 +00004036static const struct net_proto_family packet_family_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004037 .family = PF_PACKET,
4038 .create = packet_create,
4039 .owner = THIS_MODULE,
4040};
4041
4042static struct notifier_block packet_netdev_notifier = {
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00004043 .notifier_call = packet_notifier,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004044};
4045
4046#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -07004047
4048static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
stephen hemminger808f5112010-02-22 07:57:18 +00004049 __acquires(RCU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004050{
Denis V. Luneve372c412007-11-19 22:31:54 -08004051 struct net *net = seq_file_net(seq);
stephen hemminger808f5112010-02-22 07:57:18 +00004052
4053 rcu_read_lock();
4054 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004055}
4056
4057static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4058{
Herbert Xu1bf40952007-12-16 14:04:02 -08004059 struct net *net = seq_file_net(seq);
stephen hemminger808f5112010-02-22 07:57:18 +00004060 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004061}
4062
4063static void packet_seq_stop(struct seq_file *seq, void *v)
stephen hemminger808f5112010-02-22 07:57:18 +00004064 __releases(RCU)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004065{
stephen hemminger808f5112010-02-22 07:57:18 +00004066 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07004067}
4068
YOSHIFUJI Hideaki1ce4f282007-02-09 23:25:10 +09004069static int packet_seq_show(struct seq_file *seq, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004070{
4071 if (v == SEQ_START_TOKEN)
4072 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4073 else {
Li Zefanb7ceabd2010-02-08 23:19:29 +00004074 struct sock *s = sk_entry(v);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004075 const struct packet_sock *po = pkt_sk(s);
4076
4077 seq_printf(seq,
Dan Rosenberg71338aa2011-05-23 12:17:35 +00004078 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07004079 s,
4080 atomic_read(&s->sk_refcnt),
4081 s->sk_type,
4082 ntohs(po->num),
4083 po->ifindex,
4084 po->running,
4085 atomic_read(&s->sk_rmem_alloc),
Eric W. Biedermana7cb5a42012-05-24 01:10:10 -06004086 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
Eric Dumazet40d4e3d2009-07-21 21:57:59 +00004087 sock_i_ino(s));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004088 }
4089
4090 return 0;
4091}
4092
Philippe De Muyter56b3d972007-07-10 23:07:31 -07004093static const struct seq_operations packet_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004094 .start = packet_seq_start,
4095 .next = packet_seq_next,
4096 .stop = packet_seq_stop,
4097 .show = packet_seq_show,
4098};
4099
4100static int packet_seq_open(struct inode *inode, struct file *file)
4101{
Denis V. Luneve372c412007-11-19 22:31:54 -08004102 return seq_open_net(inode, file, &packet_seq_ops,
4103 sizeof(struct seq_net_private));
Linus Torvalds1da177e2005-04-16 15:20:36 -07004104}
4105
Arjan van de Venda7071d2007-02-12 00:55:36 -08004106static const struct file_operations packet_seq_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07004107 .owner = THIS_MODULE,
4108 .open = packet_seq_open,
4109 .read = seq_read,
4110 .llseek = seq_lseek,
Denis V. Luneve372c412007-11-19 22:31:54 -08004111 .release = seq_release_net,
Linus Torvalds1da177e2005-04-16 15:20:36 -07004112};
4113
4114#endif
4115
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004116static int __net_init packet_net_init(struct net *net)
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004117{
Pavel Emelyanov0fa7fa92012-08-21 01:06:47 +00004118 mutex_init(&net->packet.sklist_lock);
Denis V. Lunev2aaef4e2007-12-11 04:19:54 -08004119 INIT_HLIST_HEAD(&net->packet.sklist);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004120
Gao fengd4beaa62013-02-18 01:34:54 +00004121 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004122 return -ENOMEM;
4123
4124 return 0;
4125}
4126
Alexey Dobriyan2c8c1e72010-01-17 03:35:32 +00004127static void __net_exit packet_net_exit(struct net *net)
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004128{
Gao fengece31ff2013-02-18 01:34:56 +00004129 remove_proc_entry("packet", net->proc_net);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004130}
4131
4132static struct pernet_operations packet_net_ops = {
4133 .init = packet_net_init,
4134 .exit = packet_net_exit,
4135};
4136
4137
Linus Torvalds1da177e2005-04-16 15:20:36 -07004138static void __exit packet_exit(void)
4139{
Linus Torvalds1da177e2005-04-16 15:20:36 -07004140 unregister_netdevice_notifier(&packet_netdev_notifier);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004141 unregister_pernet_subsys(&packet_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142 sock_unregister(PF_PACKET);
4143 proto_unregister(&packet_proto);
4144}
4145
4146static int __init packet_init(void)
4147{
4148 int rc = proto_register(&packet_proto, 0);
4149
4150 if (rc != 0)
4151 goto out;
4152
4153 sock_register(&packet_family_ops);
Denis V. Lunevd12d01d2007-11-19 22:28:35 -08004154 register_pernet_subsys(&packet_net_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004155 register_netdevice_notifier(&packet_netdev_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004156out:
4157 return rc;
4158}
4159
4160module_init(packet_init);
4161module_exit(packet_exit);
4162MODULE_LICENSE("GPL");
4163MODULE_ALIAS_NETPROTO(PF_PACKET);