]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - net/packet/af_packet.c
projects / mirror_ubuntu-zesty-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'gue-next'
[mirror_ubuntu-zesty-kernel.git] / net / packet / af_packet.c
CommitLineData
1da177e4
LT		 1/*
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 *
02c30a84 8 * Authors: Ross Biro
1da177e4
LT		 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
11 *
1ce4f28b 12 * Fixes:
1da177e4
LT		 13 * 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.
1ce4f28b 35 * Ulises Alonso : Frame number limit removal and
1da177e4 36 * packet_set_ring memory leak.
0fb375fb
EB		 37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
1ce4f28b 40 * byte arrays at the end of sockaddr_ll
0fb375fb 41 * and packet_mreq.
69e3c75f 42 * Johann Baudy : Added TX RING.
f6fb8f10 43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
44 * layer.
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
46 *
1da177e4
LT		 47 *
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 */
1ce4f28b 54
1da177e4 55#include <linux/types.h>
1da177e4 56#include <linux/mm.h>
4fc268d2 57#include <linux/capability.h>
1da177e4
LT		 58#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>
ffbc6111 65#include <linux/kernel.h>
1da177e4 66#include <linux/kmod.h>
5a0e3ad6 67#include <linux/slab.h>
0e3125c7 68#include <linux/vmalloc.h>
457c4cbc 69#include <net/net_namespace.h>
1da177e4
LT		 70#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>
1da177e4
LT		 76#include <asm/uaccess.h>
77#include <asm/ioctls.h>
78#include <asm/page.h>
a1f8e7f7 79#include <asm/cacheflush.h>
1da177e4
LT		 80#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>
905db440 86#include <linux/mutex.h>
05423b24 87#include <linux/if_vlan.h>
bfd5f4a3 88#include <linux/virtio_net.h>
ed85b565 89#include <linux/errqueue.h>
614f60fa 90#include <linux/net_tstamp.h>
b0138408 91#include <linux/percpu.h>
1da177e4
LT		 92#ifdef CONFIG_INET
93#include <net/inet_common.h>
94#endif
95
2787b04b
PE		 96#include "internal.h"
97
1da177e4
LT		 98/*
99 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
b0e380b1
ACM		 113 mac_header -> ll header
114 data -> data
1da177e4
LT		 115
116Outgoing, dev->hard_header!=NULL
b0e380b1
ACM		 117 mac_header -> ll header
118 data -> ll header
1da177e4
LT		 119
120Incoming, dev->hard_header==NULL
b0e380b1
ACM		 121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
db0c58f9 123 assymetry between rx and tx paths.
b0e380b1 124 data -> data
1da177e4
LT		 125
126Outgoing, dev->hard_header==NULL
b0e380b1
ACM		 127 mac_header -> data. ll header is still not built!
128 data -> data
1da177e4
LT		 129
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
b0e380b1
ACM		 138 mac_header -> ll header
139 data -> ll header
1da177e4
LT		 140
141dev->hard_header == NULL (ll header is added by device, we cannot control it)
b0e380b1
ACM		 142 mac_header -> data
143 data -> data
1da177e4
LT		 144
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
147 */
148
1da177e4
LT		 149/* Private packet socket structures. */
150
0fb375fb
EB		 151/* identical to struct packet_mreq except it has
152 * a longer address field.
153 */
40d4e3df 154struct packet_mreq_max {
0fb375fb
EB		 155 int mr_ifindex;
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
1da177e4 159};
a2efcfa0 160
184f489e
DB		 161union tpacket_uhdr {
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
165 void *raw;
166};
167
f6fb8f10 168static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
69e3c75f
JB		 169 int closing, int tx_ring);
170
f6fb8f10 171#define V3_ALIGNMENT (8)
172
bc59ba39 173#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
f6fb8f10 174
175#define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
177
f6fb8f10 178#define PGV_FROM_VMALLOC 1
69e3c75f 179
f6fb8f10 180#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
69e3c75f
JB		 188struct packet_sock;
189static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
77f65ebd
WB		 190static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
1da177e4 192
f6fb8f10 193static 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);
bc59ba39 197static 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 *,
f6fb8f10 200 struct packet_sock *);
bc59ba39 201static void prb_retire_current_block(struct tpacket_kbdq_core *,
f6fb8f10 202 struct packet_sock *, unsigned int status);
bc59ba39 203static int prb_queue_frozen(struct tpacket_kbdq_core *);
204static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
f6fb8f10 206static void prb_retire_rx_blk_timer_expired(unsigned long);
bc59ba39 207static 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 *);
1da177e4
LT		 216static void packet_flush_mclist(struct sock *sk);
217
ffbc6111
HX		 218struct packet_skb_cb {
219 unsigned int origlen;
220 union {
221 struct sockaddr_pkt pkt;
222 struct sockaddr_ll ll;
223 } sa;
224};
225
226#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
8dc41944 227
bc59ba39 228#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
f6fb8f10 229#define GET_PBLOCK_DESC(x, bid) \
bc59ba39 230 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
f6fb8f10 231#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
bc59ba39 232 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
f6fb8f10 233#define GET_NEXT_PRB_BLK_NUM(x) \
234 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
235 ((x)->kactive_blk_num+1) : 0)
236
dc99f600
DM		 237static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
238static void __fanout_link(struct sock *sk, struct packet_sock *po);
239
d346a3fa
DB		 240static int packet_direct_xmit(struct sk_buff *skb)
241{
242 struct net_device *dev = skb->dev;
d346a3fa
DB		 243 netdev_features_t features;
244 struct netdev_queue *txq;
43279500 245 int ret = NETDEV_TX_BUSY;
d346a3fa
DB		 246
247 if (unlikely(!netif_running(dev) ||
43279500
DB		 248 !netif_carrier_ok(dev)))
249 goto drop;
d346a3fa
DB		 250
251 features = netif_skb_features(skb);
252 if (skb_needs_linearize(skb, features) &&
43279500
DB		 253 __skb_linearize(skb))
254 goto drop;
d346a3fa 255
10c51b56 256 txq = skb_get_tx_queue(dev, skb);
d346a3fa 257
43279500
DB		 258 local_bh_disable();
259
260 HARD_TX_LOCK(dev, txq, smp_processor_id());
10b3ad8c 261 if (!netif_xmit_frozen_or_drv_stopped(txq))
fa2dbdc2 262 ret = netdev_start_xmit(skb, dev, txq, false);
43279500 263 HARD_TX_UNLOCK(dev, txq);
d346a3fa 264
43279500
DB		 265 local_bh_enable();
266
267 if (!dev_xmit_complete(ret))
d346a3fa 268 kfree_skb(skb);
43279500 269
d346a3fa 270 return ret;
43279500 271drop:
0f97ede4 272 atomic_long_inc(&dev->tx_dropped);
43279500
DB		 273 kfree_skb(skb);
274 return NET_XMIT_DROP;
d346a3fa
DB		 275}
276
66e56cd4
DB		 277static struct net_device *packet_cached_dev_get(struct packet_sock *po)
278{
279 struct net_device *dev;
280
281 rcu_read_lock();
282 dev = rcu_dereference(po->cached_dev);
283 if (likely(dev))
284 dev_hold(dev);
285 rcu_read_unlock();
286
287 return dev;
288}
289
290static void packet_cached_dev_assign(struct packet_sock *po,
291 struct net_device *dev)
292{
293 rcu_assign_pointer(po->cached_dev, dev);
294}
295
296static void packet_cached_dev_reset(struct packet_sock *po)
297{
298 RCU_INIT_POINTER(po->cached_dev, NULL);
299}
300
d346a3fa
DB		 301static bool packet_use_direct_xmit(const struct packet_sock *po)
302{
303 return po->xmit == packet_direct_xmit;
304}
305
0fd5d57b 306static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
d346a3fa 307{
1cbac010 308 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
d346a3fa
DB		 309}
310
0fd5d57b
DB		 311static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
312{
313 const struct net_device_ops *ops = dev->netdev_ops;
314 u16 queue_index;
315
316 if (ops->ndo_select_queue) {
317 queue_index = ops->ndo_select_queue(dev, skb, NULL,
318 __packet_pick_tx_queue);
319 queue_index = netdev_cap_txqueue(dev, queue_index);
320 } else {
321 queue_index = __packet_pick_tx_queue(dev, skb);
322 }
323
324 skb_set_queue_mapping(skb, queue_index);
325}
326
ce06b03e
DM		 327/* register_prot_hook must be invoked with the po->bind_lock held,
328 * or from a context in which asynchronous accesses to the packet
329 * socket is not possible (packet_create()).
330 */
331static void register_prot_hook(struct sock *sk)
332{
333 struct packet_sock *po = pkt_sk(sk);
e40526cb 334
ce06b03e 335 if (!po->running) {
66e56cd4 336 if (po->fanout)
dc99f600 337 __fanout_link(sk, po);
66e56cd4 338 else
dc99f600 339 dev_add_pack(&po->prot_hook);
e40526cb 340
ce06b03e
DM		 341 sock_hold(sk);
342 po->running = 1;
343 }
344}
345
346/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
347 * held. If the sync parameter is true, we will temporarily drop
348 * the po->bind_lock and do a synchronize_net to make sure no
349 * asynchronous packet processing paths still refer to the elements
350 * of po->prot_hook. If the sync parameter is false, it is the
351 * callers responsibility to take care of this.
352 */
353static void __unregister_prot_hook(struct sock *sk, bool sync)
354{
355 struct packet_sock *po = pkt_sk(sk);
356
357 po->running = 0;
66e56cd4
DB		 358
359 if (po->fanout)
dc99f600 360 __fanout_unlink(sk, po);
66e56cd4 361 else
dc99f600 362 __dev_remove_pack(&po->prot_hook);
e40526cb 363
ce06b03e
DM		 364 __sock_put(sk);
365
366 if (sync) {
367 spin_unlock(&po->bind_lock);
368 synchronize_net();
369 spin_lock(&po->bind_lock);
370 }
371}
372
373static void unregister_prot_hook(struct sock *sk, bool sync)
374{
375 struct packet_sock *po = pkt_sk(sk);
376
377 if (po->running)
378 __unregister_prot_hook(sk, sync);
379}
380
f6dafa95 381static inline __pure struct page *pgv_to_page(void *addr)
0af55bb5
CG		 382{
383 if (is_vmalloc_addr(addr))
384 return vmalloc_to_page(addr);
385 return virt_to_page(addr);
386}
387
69e3c75f 388static void __packet_set_status(struct packet_sock *po, void *frame, int status)
1da177e4 389{
184f489e 390 union tpacket_uhdr h;
1da177e4 391
69e3c75f 392 h.raw = frame;
bbd6ef87
PM		 393 switch (po->tp_version) {
394 case TPACKET_V1:
69e3c75f 395 h.h1->tp_status = status;
0af55bb5 396 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
bbd6ef87
PM		 397 break;
398 case TPACKET_V2:
69e3c75f 399 h.h2->tp_status = status;
0af55bb5 400 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
bbd6ef87 401 break;
f6fb8f10 402 case TPACKET_V3:
69e3c75f 403 default:
f6fb8f10 404 WARN(1, "TPACKET version not supported.\n");
69e3c75f 405 BUG();
bbd6ef87 406 }
69e3c75f
JB		 407
408 smp_wmb();
bbd6ef87
PM		 409}
410
69e3c75f 411static int __packet_get_status(struct packet_sock *po, void *frame)
bbd6ef87 412{
184f489e 413 union tpacket_uhdr h;
bbd6ef87 414
69e3c75f
JB		 415 smp_rmb();
416
bbd6ef87
PM		 417 h.raw = frame;
418 switch (po->tp_version) {
419 case TPACKET_V1:
0af55bb5 420 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
69e3c75f 421 return h.h1->tp_status;
bbd6ef87 422 case TPACKET_V2:
0af55bb5 423 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
69e3c75f 424 return h.h2->tp_status;
f6fb8f10 425 case TPACKET_V3:
69e3c75f 426 default:
f6fb8f10 427 WARN(1, "TPACKET version not supported.\n");
69e3c75f
JB		 428 BUG();
429 return 0;
bbd6ef87 430 }
1da177e4 431}
69e3c75f 432
b9c32fb2
DB		 433static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
434 unsigned int flags)
7a51384c
DB		 435{
436 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
437
68a360e8
WB		 438 if (shhwtstamps &&
439 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
440 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
441 return TP_STATUS_TS_RAW_HARDWARE;
7a51384c
DB		 442
443 if (ktime_to_timespec_cond(skb->tstamp, ts))
b9c32fb2 444 return TP_STATUS_TS_SOFTWARE;
7a51384c 445
b9c32fb2 446 return 0;
7a51384c
DB		 447}
448
b9c32fb2
DB		 449static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
450 struct sk_buff *skb)
2e31396f
WB		 451{
452 union tpacket_uhdr h;
453 struct timespec ts;
b9c32fb2 454 __u32 ts_status;
2e31396f 455
b9c32fb2
DB		 456 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
457 return 0;
2e31396f
WB		 458
459 h.raw = frame;
460 switch (po->tp_version) {
461 case TPACKET_V1:
462 h.h1->tp_sec = ts.tv_sec;
463 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
464 break;
465 case TPACKET_V2:
466 h.h2->tp_sec = ts.tv_sec;
467 h.h2->tp_nsec = ts.tv_nsec;
468 break;
469 case TPACKET_V3:
470 default:
471 WARN(1, "TPACKET version not supported.\n");
472 BUG();
473 }
474
475 /* one flush is safe, as both fields always lie on the same cacheline */
476 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
477 smp_wmb();
b9c32fb2
DB		 478
479 return ts_status;
2e31396f
WB		 480}
481
69e3c75f
JB		 482static void *packet_lookup_frame(struct packet_sock *po,
483 struct packet_ring_buffer *rb,
484 unsigned int position,
485 int status)
486{
487 unsigned int pg_vec_pos, frame_offset;
184f489e 488 union tpacket_uhdr h;
69e3c75f
JB		 489
490 pg_vec_pos = position / rb->frames_per_block;
491 frame_offset = position % rb->frames_per_block;
492
0e3125c7
NH		 493 h.raw = rb->pg_vec[pg_vec_pos].buffer +
494 (frame_offset * rb->frame_size);
69e3c75f
JB		 495
496 if (status != __packet_get_status(po, h.raw))
497 return NULL;
498
499 return h.raw;
500}
501
eea49cc9 502static void *packet_current_frame(struct packet_sock *po,
69e3c75f
JB		 503 struct packet_ring_buffer *rb,
504 int status)
505{
506 return packet_lookup_frame(po, rb, rb->head, status);
507}
508
bc59ba39 509static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
f6fb8f10 510{
511 del_timer_sync(&pkc->retire_blk_timer);
512}
513
514static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
515 int tx_ring,
516 struct sk_buff_head *rb_queue)
517{
bc59ba39 518 struct tpacket_kbdq_core *pkc;
f6fb8f10 519
22781a5b
DJ		 520 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
521 GET_PBDQC_FROM_RB(&po->rx_ring);
f6fb8f10 522
ec6f809f 523 spin_lock_bh(&rb_queue->lock);
f6fb8f10 524 pkc->delete_blk_timer = 1;
ec6f809f 525 spin_unlock_bh(&rb_queue->lock);
f6fb8f10 526
527 prb_del_retire_blk_timer(pkc);
528}
529
530static void prb_init_blk_timer(struct packet_sock *po,
bc59ba39 531 struct tpacket_kbdq_core *pkc,
f6fb8f10 532 void (*func) (unsigned long))
533{
534 init_timer(&pkc->retire_blk_timer);
535 pkc->retire_blk_timer.data = (long)po;
536 pkc->retire_blk_timer.function = func;
537 pkc->retire_blk_timer.expires = jiffies;
538}
539
540static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
541{
bc59ba39 542 struct tpacket_kbdq_core *pkc;
f6fb8f10 543
544 if (tx_ring)
545 BUG();
546
22781a5b
DJ		 547 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
548 GET_PBDQC_FROM_RB(&po->rx_ring);
f6fb8f10 549 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
550}
551
552static int prb_calc_retire_blk_tmo(struct packet_sock *po,
553 int blk_size_in_bytes)
554{
555 struct net_device *dev;
556 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
4bc71cb9
JP		 557 struct ethtool_cmd ecmd;
558 int err;
e440cf2c 559 u32 speed;
f6fb8f10 560
4bc71cb9
JP		 561 rtnl_lock();
562 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
563 if (unlikely(!dev)) {
564 rtnl_unlock();
f6fb8f10 565 return DEFAULT_PRB_RETIRE_TOV;
4bc71cb9
JP		 566 }
567 err = __ethtool_get_settings(dev, &ecmd);
e440cf2c 568 speed = ethtool_cmd_speed(&ecmd);
4bc71cb9
JP		 569 rtnl_unlock();
570 if (!err) {
4bc71cb9
JP		 571 /*
572 * If the link speed is so slow you don't really
573 * need to worry about perf anyways
574 */
e440cf2c 575 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
4bc71cb9 576 return DEFAULT_PRB_RETIRE_TOV;
e440cf2c 577 } else {
578 msec = 1;
579 div = speed / 1000;
f6fb8f10 580 }
581 }
582
583 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
584
585 if (div)
586 mbits /= div;
587
588 tmo = mbits * msec;
589
590 if (div)
591 return tmo+1;
592 return tmo;
593}
594
bc59ba39 595static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
f6fb8f10 596 union tpacket_req_u *req_u)
597{
598 p1->feature_req_word = req_u->req3.tp_feature_req_word;
599}
600
601static void init_prb_bdqc(struct packet_sock *po,
602 struct packet_ring_buffer *rb,
603 struct pgv *pg_vec,
604 union tpacket_req_u *req_u, int tx_ring)
605{
22781a5b 606 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
bc59ba39 607 struct tpacket_block_desc *pbd;
f6fb8f10 608
609 memset(p1, 0x0, sizeof(*p1));
610
611 p1->knxt_seq_num = 1;
612 p1->pkbdq = pg_vec;
bc59ba39 613 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
e3192690 614 p1->pkblk_start = pg_vec[0].buffer;
f6fb8f10 615 p1->kblk_size = req_u->req3.tp_block_size;
616 p1->knum_blocks = req_u->req3.tp_block_nr;
617 p1->hdrlen = po->tp_hdrlen;
618 p1->version = po->tp_version;
619 p1->last_kactive_blk_num = 0;
ee80fbf3 620 po->stats.stats3.tp_freeze_q_cnt = 0;
f6fb8f10 621 if (req_u->req3.tp_retire_blk_tov)
622 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
623 else
624 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
625 req_u->req3.tp_block_size);
626 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
627 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
628
dc808110 629 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
f6fb8f10 630 prb_init_ft_ops(p1, req_u);
631 prb_setup_retire_blk_timer(po, tx_ring);
632 prb_open_block(p1, pbd);
633}
634
635/* Do NOT update the last_blk_num first.
636 * Assumes sk_buff_head lock is held.
637 */
bc59ba39 638static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
f6fb8f10 639{
640 mod_timer(&pkc->retire_blk_timer,
641 jiffies + pkc->tov_in_jiffies);
642 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
643}
644
645/*
646 * Timer logic:
647 * 1) We refresh the timer only when we open a block.
648 * By doing this we don't waste cycles refreshing the timer
649 * on packet-by-packet basis.
650 *
651 * With a 1MB block-size, on a 1Gbps line, it will take
652 * i) ~8 ms to fill a block + ii) memcpy etc.
653 * In this cut we are not accounting for the memcpy time.
654 *
655 * So, if the user sets the 'tmo' to 10ms then the timer
656 * will never fire while the block is still getting filled
657 * (which is what we want). However, the user could choose
658 * to close a block early and that's fine.
659 *
660 * But when the timer does fire, we check whether or not to refresh it.
661 * Since the tmo granularity is in msecs, it is not too expensive
662 * to refresh the timer, lets say every '8' msecs.
663 * Either the user can set the 'tmo' or we can derive it based on
664 * a) line-speed and b) block-size.
665 * prb_calc_retire_blk_tmo() calculates the tmo.
666 *
667 */
668static void prb_retire_rx_blk_timer_expired(unsigned long data)
669{
670 struct packet_sock *po = (struct packet_sock *)data;
22781a5b 671 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
f6fb8f10 672 unsigned int frozen;
bc59ba39 673 struct tpacket_block_desc *pbd;
f6fb8f10 674
675 spin_lock(&po->sk.sk_receive_queue.lock);
676
677 frozen = prb_queue_frozen(pkc);
678 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
679
680 if (unlikely(pkc->delete_blk_timer))
681 goto out;
682
683 /* We only need to plug the race when the block is partially filled.
684 * tpacket_rcv:
685 * lock(); increment BLOCK_NUM_PKTS; unlock()
686 * copy_bits() is in progress ...
687 * timer fires on other cpu:
688 * we can't retire the current block because copy_bits
689 * is in progress.
690 *
691 */
692 if (BLOCK_NUM_PKTS(pbd)) {
693 while (atomic_read(&pkc->blk_fill_in_prog)) {
694 /* Waiting for skb_copy_bits to finish... */
695 cpu_relax();
696 }
697 }
698
699 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
700 if (!frozen) {
701 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
702 if (!prb_dispatch_next_block(pkc, po))
703 goto refresh_timer;
704 else
705 goto out;
706 } else {
707 /* Case 1. Queue was frozen because user-space was
708 * lagging behind.
709 */
710 if (prb_curr_blk_in_use(pkc, pbd)) {
711 /*
712 * Ok, user-space is still behind.
713 * So just refresh the timer.
714 */
715 goto refresh_timer;
716 } else {
717 /* Case 2. queue was frozen,user-space caught up,
718 * now the link went idle && the timer fired.
719 * We don't have a block to close.So we open this
720 * block and restart the timer.
721 * opening a block thaws the queue,restarts timer
722 * Thawing/timer-refresh is a side effect.
723 */
724 prb_open_block(pkc, pbd);
725 goto out;
726 }
727 }
728 }
729
730refresh_timer:
731 _prb_refresh_rx_retire_blk_timer(pkc);
732
733out:
734 spin_unlock(&po->sk.sk_receive_queue.lock);
735}
736
eea49cc9 737static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
bc59ba39 738 struct tpacket_block_desc *pbd1, __u32 status)
f6fb8f10 739{
740 /* Flush everything minus the block header */
741
742#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
743 u8 *start, *end;
744
745 start = (u8 *)pbd1;
746
747 /* Skip the block header(we know header WILL fit in 4K) */
748 start += PAGE_SIZE;
749
750 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
751 for (; start < end; start += PAGE_SIZE)
752 flush_dcache_page(pgv_to_page(start));
753
754 smp_wmb();
755#endif
756
757 /* Now update the block status. */
758
759 BLOCK_STATUS(pbd1) = status;
760
761 /* Flush the block header */
762
763#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
764 start = (u8 *)pbd1;
765 flush_dcache_page(pgv_to_page(start));
766
767 smp_wmb();
768#endif
769}
770
771/*
772 * Side effect:
773 *
774 * 1) flush the block
775 * 2) Increment active_blk_num
776 *
777 * Note:We DONT refresh the timer on purpose.
778 * Because almost always the next block will be opened.
779 */
bc59ba39 780static void prb_close_block(struct tpacket_kbdq_core *pkc1,
781 struct tpacket_block_desc *pbd1,
f6fb8f10 782 struct packet_sock *po, unsigned int stat)
783{
784 __u32 status = TP_STATUS_USER | stat;
785
786 struct tpacket3_hdr *last_pkt;
bc59ba39 787 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
f6fb8f10 788
ee80fbf3 789 if (po->stats.stats3.tp_drops)
f6fb8f10 790 status |= TP_STATUS_LOSING;
791
792 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
793 last_pkt->tp_next_offset = 0;
794
795 /* Get the ts of the last pkt */
796 if (BLOCK_NUM_PKTS(pbd1)) {
797 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
798 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
799 } else {
800 /* Ok, we tmo'd - so get the current time */
801 struct timespec ts;
802 getnstimeofday(&ts);
803 h1->ts_last_pkt.ts_sec = ts.tv_sec;
804 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
805 }
806
807 smp_wmb();
808
809 /* Flush the block */
810 prb_flush_block(pkc1, pbd1, status);
811
812 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
813}
814
eea49cc9 815static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
f6fb8f10 816{
817 pkc->reset_pending_on_curr_blk = 0;
818}
819
820/*
821 * Side effect of opening a block:
822 *
823 * 1) prb_queue is thawed.
824 * 2) retire_blk_timer is refreshed.
825 *
826 */
bc59ba39 827static void prb_open_block(struct tpacket_kbdq_core *pkc1,
828 struct tpacket_block_desc *pbd1)
f6fb8f10 829{
830 struct timespec ts;
bc59ba39 831 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
f6fb8f10 832
833 smp_rmb();
834
8da3056c
DB		 835 /* We could have just memset this but we will lose the
836 * flexibility of making the priv area sticky
837 */
f6fb8f10 838
8da3056c
DB		 839 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
840 BLOCK_NUM_PKTS(pbd1) = 0;
841 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
f6fb8f10 842
8da3056c
DB		 843 getnstimeofday(&ts);
844
845 h1->ts_first_pkt.ts_sec = ts.tv_sec;
846 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
f6fb8f10 847
8da3056c
DB		 848 pkc1->pkblk_start = (char *)pbd1;
849 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
850
851 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
852 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
853
854 pbd1->version = pkc1->version;
855 pkc1->prev = pkc1->nxt_offset;
856 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
857
858 prb_thaw_queue(pkc1);
859 _prb_refresh_rx_retire_blk_timer(pkc1);
860
861 smp_wmb();
f6fb8f10 862}
863
864/*
865 * Queue freeze logic:
866 * 1) Assume tp_block_nr = 8 blocks.
867 * 2) At time 't0', user opens Rx ring.
868 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
869 * 4) user-space is either sleeping or processing block '0'.
870 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
871 * it will close block-7,loop around and try to fill block '0'.
872 * call-flow:
873 * __packet_lookup_frame_in_block
874 * prb_retire_current_block()
875 * prb_dispatch_next_block()
876 * |->(BLOCK_STATUS == USER) evaluates to true
877 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
878 * 6) Now there are two cases:
879 * 6.1) Link goes idle right after the queue is frozen.
880 * But remember, the last open_block() refreshed the timer.
881 * When this timer expires,it will refresh itself so that we can
882 * re-open block-0 in near future.
883 * 6.2) Link is busy and keeps on receiving packets. This is a simple
884 * case and __packet_lookup_frame_in_block will check if block-0
885 * is free and can now be re-used.
886 */
eea49cc9 887static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
f6fb8f10 888 struct packet_sock *po)
889{
890 pkc->reset_pending_on_curr_blk = 1;
ee80fbf3 891 po->stats.stats3.tp_freeze_q_cnt++;
f6fb8f10 892}
893
894#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
895
896/*
897 * If the next block is free then we will dispatch it
898 * and return a good offset.
899 * Else, we will freeze the queue.
900 * So, caller must check the return value.
901 */
bc59ba39 902static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
f6fb8f10 903 struct packet_sock *po)
904{
bc59ba39 905 struct tpacket_block_desc *pbd;
f6fb8f10 906
907 smp_rmb();
908
909 /* 1. Get current block num */
910 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
911
912 /* 2. If this block is currently in_use then freeze the queue */
913 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
914 prb_freeze_queue(pkc, po);
915 return NULL;
916 }
917
918 /*
919 * 3.
920 * open this block and return the offset where the first packet
921 * needs to get stored.
922 */
923 prb_open_block(pkc, pbd);
924 return (void *)pkc->nxt_offset;
925}
926
bc59ba39 927static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
f6fb8f10 928 struct packet_sock *po, unsigned int status)
929{
bc59ba39 930 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
f6fb8f10 931
932 /* retire/close the current block */
933 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
934 /*
935 * Plug the case where copy_bits() is in progress on
936 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
937 * have space to copy the pkt in the current block and
938 * called prb_retire_current_block()
939 *
940 * We don't need to worry about the TMO case because
941 * the timer-handler already handled this case.
942 */
943 if (!(status & TP_STATUS_BLK_TMO)) {
944 while (atomic_read(&pkc->blk_fill_in_prog)) {
945 /* Waiting for skb_copy_bits to finish... */
946 cpu_relax();
947 }
948 }
949 prb_close_block(pkc, pbd, po, status);
950 return;
951 }
f6fb8f10 952}
953
eea49cc9 954static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
bc59ba39 955 struct tpacket_block_desc *pbd)
f6fb8f10 956{
957 return TP_STATUS_USER & BLOCK_STATUS(pbd);
958}
959
eea49cc9 960static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
f6fb8f10 961{
962 return pkc->reset_pending_on_curr_blk;
963}
964
eea49cc9 965static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
f6fb8f10 966{
bc59ba39 967 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
f6fb8f10 968 atomic_dec(&pkc->blk_fill_in_prog);
969}
970
eea49cc9 971static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
f6fb8f10 972 struct tpacket3_hdr *ppd)
973{
3958afa1 974 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
f6fb8f10 975}
976
eea49cc9 977static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
f6fb8f10 978 struct tpacket3_hdr *ppd)
979{
980 ppd->hv1.tp_rxhash = 0;
981}
982
eea49cc9 983static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
f6fb8f10 984 struct tpacket3_hdr *ppd)
985{
986 if (vlan_tx_tag_present(pkc->skb)) {
987 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
a0cdfcf3
AW		 988 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
989 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
f6fb8f10 990 } else {
9e67030a 991 ppd->hv1.tp_vlan_tci = 0;
a0cdfcf3 992 ppd->hv1.tp_vlan_tpid = 0;
9e67030a 993 ppd->tp_status = TP_STATUS_AVAILABLE;
f6fb8f10 994 }
995}
996
bc59ba39 997static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
f6fb8f10 998 struct tpacket3_hdr *ppd)
999{
a0cdfcf3 1000 ppd->hv1.tp_padding = 0;
f6fb8f10 1001 prb_fill_vlan_info(pkc, ppd);
1002
1003 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1004 prb_fill_rxhash(pkc, ppd);
1005 else
1006 prb_clear_rxhash(pkc, ppd);
1007}
1008
eea49cc9 1009static void prb_fill_curr_block(char *curr,
bc59ba39 1010 struct tpacket_kbdq_core *pkc,
1011 struct tpacket_block_desc *pbd,
f6fb8f10 1012 unsigned int len)
1013{
1014 struct tpacket3_hdr *ppd;
1015
1016 ppd = (struct tpacket3_hdr *)curr;
1017 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1018 pkc->prev = curr;
1019 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1020 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1021 BLOCK_NUM_PKTS(pbd) += 1;
1022 atomic_inc(&pkc->blk_fill_in_prog);
1023 prb_run_all_ft_ops(pkc, ppd);
1024}
1025
1026/* Assumes caller has the sk->rx_queue.lock */
1027static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1028 struct sk_buff *skb,
1029 int status,
1030 unsigned int len
1031 )
1032{
bc59ba39 1033 struct tpacket_kbdq_core *pkc;
1034 struct tpacket_block_desc *pbd;
f6fb8f10 1035 char *curr, *end;
1036
e3192690 1037 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
f6fb8f10 1038 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1039
1040 /* Queue is frozen when user space is lagging behind */
1041 if (prb_queue_frozen(pkc)) {
1042 /*
1043 * Check if that last block which caused the queue to freeze,
1044 * is still in_use by user-space.
1045 */
1046 if (prb_curr_blk_in_use(pkc, pbd)) {
1047 /* Can't record this packet */
1048 return NULL;
1049 } else {
1050 /*
1051 * Ok, the block was released by user-space.
1052 * Now let's open that block.
1053 * opening a block also thaws the queue.
1054 * Thawing is a side effect.
1055 */
1056 prb_open_block(pkc, pbd);
1057 }
1058 }
1059
1060 smp_mb();
1061 curr = pkc->nxt_offset;
1062 pkc->skb = skb;
e3192690 1063 end = (char *)pbd + pkc->kblk_size;
f6fb8f10 1064
1065 /* first try the current block */
1066 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1067 prb_fill_curr_block(curr, pkc, pbd, len);
1068 return (void *)curr;
1069 }
1070
1071 /* Ok, close the current block */
1072 prb_retire_current_block(pkc, po, 0);
1073
1074 /* Now, try to dispatch the next block */
1075 curr = (char *)prb_dispatch_next_block(pkc, po);
1076 if (curr) {
1077 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1078 prb_fill_curr_block(curr, pkc, pbd, len);
1079 return (void *)curr;
1080 }
1081
1082 /*
1083 * No free blocks are available.user_space hasn't caught up yet.
1084 * Queue was just frozen and now this packet will get dropped.
1085 */
1086 return NULL;
1087}
1088
eea49cc9 1089static void *packet_current_rx_frame(struct packet_sock *po,
f6fb8f10 1090 struct sk_buff *skb,
1091 int status, unsigned int len)
1092{
1093 char *curr = NULL;
1094 switch (po->tp_version) {
1095 case TPACKET_V1:
1096 case TPACKET_V2:
1097 curr = packet_lookup_frame(po, &po->rx_ring,
1098 po->rx_ring.head, status);
1099 return curr;
1100 case TPACKET_V3:
1101 return __packet_lookup_frame_in_block(po, skb, status, len);
1102 default:
1103 WARN(1, "TPACKET version not supported\n");
1104 BUG();
99aa3473 1105 return NULL;
f6fb8f10 1106 }
1107}
1108
eea49cc9 1109static void *prb_lookup_block(struct packet_sock *po,
f6fb8f10 1110 struct packet_ring_buffer *rb,
77f65ebd 1111 unsigned int idx,
f6fb8f10 1112 int status)
1113{
bc59ba39 1114 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
77f65ebd 1115 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
f6fb8f10 1116
1117 if (status != BLOCK_STATUS(pbd))
1118 return NULL;
1119 return pbd;
1120}
1121
eea49cc9 1122static int prb_previous_blk_num(struct packet_ring_buffer *rb)
f6fb8f10 1123{
1124 unsigned int prev;
1125 if (rb->prb_bdqc.kactive_blk_num)
1126 prev = rb->prb_bdqc.kactive_blk_num-1;
1127 else
1128 prev = rb->prb_bdqc.knum_blocks-1;
1129 return prev;
1130}
1131
1132/* Assumes caller has held the rx_queue.lock */
eea49cc9 1133static void *__prb_previous_block(struct packet_sock *po,
f6fb8f10 1134 struct packet_ring_buffer *rb,
1135 int status)
1136{
1137 unsigned int previous = prb_previous_blk_num(rb);
1138 return prb_lookup_block(po, rb, previous, status);
1139}
1140
eea49cc9 1141static void *packet_previous_rx_frame(struct packet_sock *po,
f6fb8f10 1142 struct packet_ring_buffer *rb,
1143 int status)
1144{
1145 if (po->tp_version <= TPACKET_V2)
1146 return packet_previous_frame(po, rb, status);
1147
1148 return __prb_previous_block(po, rb, status);
1149}
1150
eea49cc9 1151static void packet_increment_rx_head(struct packet_sock *po,
f6fb8f10 1152 struct packet_ring_buffer *rb)
1153{
1154 switch (po->tp_version) {
1155 case TPACKET_V1:
1156 case TPACKET_V2:
1157 return packet_increment_head(rb);
1158 case TPACKET_V3:
1159 default:
1160 WARN(1, "TPACKET version not supported.\n");
1161 BUG();
1162 return;
1163 }
1164}
1165
eea49cc9 1166static void *packet_previous_frame(struct packet_sock *po,
69e3c75f
JB		 1167 struct packet_ring_buffer *rb,
1168 int status)
1169{
1170 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1171 return packet_lookup_frame(po, rb, previous, status);
1172}
1173
eea49cc9 1174static void packet_increment_head(struct packet_ring_buffer *buff)
69e3c75f
JB		 1175{
1176 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1177}
1178
b0138408
DB		 1179static void packet_inc_pending(struct packet_ring_buffer *rb)
1180{
1181 this_cpu_inc(*rb->pending_refcnt);
1182}
1183
1184static void packet_dec_pending(struct packet_ring_buffer *rb)
1185{
1186 this_cpu_dec(*rb->pending_refcnt);
1187}
1188
1189static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1190{
1191 unsigned int refcnt = 0;
1192 int cpu;
1193
1194 /* We don't use pending refcount in rx_ring. */
1195 if (rb->pending_refcnt == NULL)
1196 return 0;
1197
1198 for_each_possible_cpu(cpu)
1199 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1200
1201 return refcnt;
1202}
1203
1204static int packet_alloc_pending(struct packet_sock *po)
1205{
1206 po->rx_ring.pending_refcnt = NULL;
1207
1208 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1209 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1210 return -ENOBUFS;
1211
1212 return 0;
1213}
1214
1215static void packet_free_pending(struct packet_sock *po)
1216{
1217 free_percpu(po->tx_ring.pending_refcnt);
1218}
1219
77f65ebd
WB		 1220static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1221{
1222 struct sock *sk = &po->sk;
1223 bool has_room;
1224
1225 if (po->prot_hook.func != tpacket_rcv)
1226 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1227 <= sk->sk_rcvbuf;
1228
1229 spin_lock(&sk->sk_receive_queue.lock);
1230 if (po->tp_version == TPACKET_V3)
1231 has_room = prb_lookup_block(po, &po->rx_ring,
1232 po->rx_ring.prb_bdqc.kactive_blk_num,
1233 TP_STATUS_KERNEL);
1234 else
1235 has_room = packet_lookup_frame(po, &po->rx_ring,
1236 po->rx_ring.head,
1237 TP_STATUS_KERNEL);
1238 spin_unlock(&sk->sk_receive_queue.lock);
1239
1240 return has_room;
1241}
1242
1da177e4
LT		 1243static void packet_sock_destruct(struct sock *sk)
1244{
ed85b565
RC		 1245 skb_queue_purge(&sk->sk_error_queue);
1246
547b792c
IJ		 1247 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1248 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1da177e4
LT		 1249
1250 if (!sock_flag(sk, SOCK_DEAD)) {
40d4e3df 1251 pr_err("Attempt to release alive packet socket: %p\n", sk);
1da177e4
LT		 1252 return;
1253 }
1254
17ab56a2 1255 sk_refcnt_debug_dec(sk);
1da177e4
LT		 1256}
1257
dc99f600
DM		 1258static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1259{
1260 int x = atomic_read(&f->rr_cur) + 1;
1261
1262 if (x >= num)
1263 x = 0;
1264
1265 return x;
1266}
1267
77f65ebd
WB		 1268static unsigned int fanout_demux_hash(struct packet_fanout *f,
1269 struct sk_buff *skb,
1270 unsigned int num)
dc99f600 1271{
61b905da 1272 return reciprocal_scale(skb_get_hash(skb), num);
dc99f600
DM		 1273}
1274
77f65ebd
WB		 1275static unsigned int fanout_demux_lb(struct packet_fanout *f,
1276 struct sk_buff *skb,
1277 unsigned int num)
dc99f600
DM		 1278{
1279 int cur, old;
1280
1281 cur = atomic_read(&f->rr_cur);
1282 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1283 fanout_rr_next(f, num))) != cur)
1284 cur = old;
77f65ebd
WB		 1285 return cur;
1286}
1287
1288static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1289 struct sk_buff *skb,
1290 unsigned int num)
1291{
1292 return smp_processor_id() % num;
dc99f600
DM		 1293}
1294
5df0ddfb
DB		 1295static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1296 struct sk_buff *skb,
1297 unsigned int num)
1298{
f337db64 1299 return prandom_u32_max(num);
5df0ddfb
DB		 1300}
1301
77f65ebd
WB		 1302static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1303 struct sk_buff *skb,
1304 unsigned int idx, unsigned int skip,
1305 unsigned int num)
95ec3eb4 1306{
77f65ebd 1307 unsigned int i, j;
95ec3eb4 1308
77f65ebd
WB		 1309 i = j = min_t(int, f->next[idx], num - 1);
1310 do {
1311 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1312 if (i != j)
1313 f->next[idx] = i;
1314 return i;
1315 }
1316 if (++i == num)
1317 i = 0;
1318 } while (i != j);
1319
1320 return idx;
1321}
1322
2d36097d
NH		 1323static unsigned int fanout_demux_qm(struct packet_fanout *f,
1324 struct sk_buff *skb,
1325 unsigned int num)
1326{
1327 return skb_get_queue_mapping(skb) % num;
1328}
1329
77f65ebd
WB		 1330static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1331{
1332 return f->flags & (flag >> 8);
95ec3eb4
DM		 1333}
1334
95ec3eb4
DM		 1335static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1336 struct packet_type *pt, struct net_device *orig_dev)
dc99f600
DM		 1337{
1338 struct packet_fanout *f = pt->af_packet_priv;
1339 unsigned int num = f->num_members;
1340 struct packet_sock *po;
77f65ebd 1341 unsigned int idx;
dc99f600
DM		 1342
1343 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1344 !num) {
1345 kfree_skb(skb);
1346 return 0;
1347 }
1348
95ec3eb4
DM		 1349 switch (f->type) {
1350 case PACKET_FANOUT_HASH:
1351 default:
77f65ebd 1352 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
bc416d97 1353 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
95ec3eb4
DM		 1354 if (!skb)
1355 return 0;
1356 }
77f65ebd 1357 idx = fanout_demux_hash(f, skb, num);
95ec3eb4
DM		 1358 break;
1359 case PACKET_FANOUT_LB:
77f65ebd 1360 idx = fanout_demux_lb(f, skb, num);
95ec3eb4
DM		 1361 break;
1362 case PACKET_FANOUT_CPU:
77f65ebd
WB		 1363 idx = fanout_demux_cpu(f, skb, num);
1364 break;
5df0ddfb
DB		 1365 case PACKET_FANOUT_RND:
1366 idx = fanout_demux_rnd(f, skb, num);
1367 break;
2d36097d
NH		 1368 case PACKET_FANOUT_QM:
1369 idx = fanout_demux_qm(f, skb, num);
1370 break;
77f65ebd
WB		 1371 case PACKET_FANOUT_ROLLOVER:
1372 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
95ec3eb4 1373 break;
dc99f600
DM		 1374 }
1375
77f65ebd
WB		 1376 po = pkt_sk(f->arr[idx]);
1377 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1378 unlikely(!packet_rcv_has_room(po, skb))) {
1379 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1380 po = pkt_sk(f->arr[idx]);
1381 }
dc99f600
DM		 1382
1383 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1384}
1385
fff3321d
PE		 1386DEFINE_MUTEX(fanout_mutex);
1387EXPORT_SYMBOL_GPL(fanout_mutex);
dc99f600
DM		 1388static LIST_HEAD(fanout_list);
1389
1390static void __fanout_link(struct sock *sk, struct packet_sock *po)
1391{
1392 struct packet_fanout *f = po->fanout;
1393
1394 spin_lock(&f->lock);
1395 f->arr[f->num_members] = sk;
1396 smp_wmb();
1397 f->num_members++;
1398 spin_unlock(&f->lock);
1399}
1400
1401static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1402{
1403 struct packet_fanout *f = po->fanout;
1404 int i;
1405
1406 spin_lock(&f->lock);
1407 for (i = 0; i < f->num_members; i++) {
1408 if (f->arr[i] == sk)
1409 break;
1410 }
1411 BUG_ON(i >= f->num_members);
1412 f->arr[i] = f->arr[f->num_members - 1];
1413 f->num_members--;
1414 spin_unlock(&f->lock);
1415}
1416
d4dd8aee 1417static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
c0de08d0 1418{
d4dd8aee 1419 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
c0de08d0
EL		 1420 return true;
1421
1422 return false;
1423}
1424
7736d33f 1425static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
dc99f600
DM		 1426{
1427 struct packet_sock *po = pkt_sk(sk);
1428 struct packet_fanout *f, *match;
7736d33f 1429 u8 type = type_flags & 0xff;
77f65ebd 1430 u8 flags = type_flags >> 8;
dc99f600
DM		 1431 int err;
1432
1433 switch (type) {
77f65ebd
WB		 1434 case PACKET_FANOUT_ROLLOVER:
1435 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1436 return -EINVAL;
dc99f600
DM		 1437 case PACKET_FANOUT_HASH:
1438 case PACKET_FANOUT_LB:
95ec3eb4 1439 case PACKET_FANOUT_CPU:
5df0ddfb 1440 case PACKET_FANOUT_RND:
2d36097d 1441 case PACKET_FANOUT_QM:
dc99f600
DM		 1442 break;
1443 default:
1444 return -EINVAL;
1445 }
1446
1447 if (!po->running)
1448 return -EINVAL;
1449
1450 if (po->fanout)
1451 return -EALREADY;
1452
1453 mutex_lock(&fanout_mutex);
1454 match = NULL;
1455 list_for_each_entry(f, &fanout_list, list) {
1456 if (f->id == id &&
1457 read_pnet(&f->net) == sock_net(sk)) {
1458 match = f;
1459 break;
1460 }
1461 }
afe62c68 1462 err = -EINVAL;
77f65ebd 1463 if (match && match->flags != flags)
afe62c68 1464 goto out;
dc99f600 1465 if (!match) {
afe62c68 1466 err = -ENOMEM;
dc99f600 1467 match = kzalloc(sizeof(*match), GFP_KERNEL);
afe62c68
ED		 1468 if (!match)
1469 goto out;
1470 write_pnet(&match->net, sock_net(sk));
1471 match->id = id;
1472 match->type = type;
77f65ebd 1473 match->flags = flags;
afe62c68
ED		 1474 atomic_set(&match->rr_cur, 0);
1475 INIT_LIST_HEAD(&match->list);
1476 spin_lock_init(&match->lock);
1477 atomic_set(&match->sk_ref, 0);
1478 match->prot_hook.type = po->prot_hook.type;
1479 match->prot_hook.dev = po->prot_hook.dev;
1480 match->prot_hook.func = packet_rcv_fanout;
1481 match->prot_hook.af_packet_priv = match;
c0de08d0 1482 match->prot_hook.id_match = match_fanout_group;
afe62c68
ED		 1483 dev_add_pack(&match->prot_hook);
1484 list_add(&match->list, &fanout_list);
dc99f600 1485 }
afe62c68
ED		 1486 err = -EINVAL;
1487 if (match->type == type &&
1488 match->prot_hook.type == po->prot_hook.type &&
1489 match->prot_hook.dev == po->prot_hook.dev) {
1490 err = -ENOSPC;
1491 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1492 __dev_remove_pack(&po->prot_hook);
1493 po->fanout = match;
1494 atomic_inc(&match->sk_ref);
1495 __fanout_link(sk, po);
1496 err = 0;
dc99f600
DM		 1497 }
1498 }
afe62c68 1499out:
dc99f600
DM		 1500 mutex_unlock(&fanout_mutex);
1501 return err;
1502}
1503
1504static void fanout_release(struct sock *sk)
1505{
1506 struct packet_sock *po = pkt_sk(sk);
1507 struct packet_fanout *f;
1508
1509 f = po->fanout;
1510 if (!f)
1511 return;
1512
fff3321d 1513 mutex_lock(&fanout_mutex);
dc99f600
DM		 1514 po->fanout = NULL;
1515
dc99f600
DM		 1516 if (atomic_dec_and_test(&f->sk_ref)) {
1517 list_del(&f->list);
1518 dev_remove_pack(&f->prot_hook);
1519 kfree(f);
1520 }
1521 mutex_unlock(&fanout_mutex);
1522}
1da177e4 1523
90ddc4f0 1524static const struct proto_ops packet_ops;
1da177e4 1525
90ddc4f0 1526static const struct proto_ops packet_ops_spkt;
1da177e4 1527
40d4e3df
ED		 1528static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1529 struct packet_type *pt, struct net_device *orig_dev)
1da177e4
LT		 1530{
1531 struct sock *sk;
1532 struct sockaddr_pkt *spkt;
1533
1534 /*
1535 * When we registered the protocol we saved the socket in the data
1536 * field for just this event.
1537 */
1538
1539 sk = pt->af_packet_priv;
1ce4f28b 1540
1da177e4
LT		 1541 /*
1542 * Yank back the headers [hope the device set this
1543 * right or kerboom...]
1544 *
1545 * Incoming packets have ll header pulled,
1546 * push it back.
1547 *
98e399f8 1548 * For outgoing ones skb->data == skb_mac_header(skb)
1da177e4
LT		 1549 * so that this procedure is noop.
1550 */
1551
1552 if (skb->pkt_type == PACKET_LOOPBACK)
1553 goto out;
1554
09ad9bc7 1555 if (!net_eq(dev_net(dev), sock_net(sk)))
d12d01d6
DL		 1556 goto out;
1557
40d4e3df
ED		 1558 skb = skb_share_check(skb, GFP_ATOMIC);
1559 if (skb == NULL)
1da177e4
LT		 1560 goto oom;
1561
1562 /* drop any routing info */
adf30907 1563 skb_dst_drop(skb);
1da177e4 1564
84531c24
PO		 1565 /* drop conntrack reference */
1566 nf_reset(skb);
1567
ffbc6111 1568 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1da177e4 1569
98e399f8 1570 skb_push(skb, skb->data - skb_mac_header(skb));
1da177e4
LT		 1571
1572 /*
1573 * The SOCK_PACKET socket receives _all_ frames.
1574 */
1575
1576 spkt->spkt_family = dev->type;
1577 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1578 spkt->spkt_protocol = skb->protocol;
1579
1580 /*
1581 * Charge the memory to the socket. This is done specifically
1582 * to prevent sockets using all the memory up.
1583 */
1584
40d4e3df 1585 if (sock_queue_rcv_skb(sk, skb) == 0)
1da177e4
LT		 1586 return 0;
1587
1588out:
1589 kfree_skb(skb);
1590oom:
1591 return 0;
1592}
1593
1594
1595/*
1596 * Output a raw packet to a device layer. This bypasses all the other
1597 * protocol layers and you must therefore supply it with a complete frame
1598 */
1ce4f28b 1599
1da177e4
LT		 1600static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1601 struct msghdr *msg, size_t len)
1602{
1603 struct sock *sk = sock->sk;
342dfc30 1604 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1a35ca80 1605 struct sk_buff *skb = NULL;
1da177e4 1606 struct net_device *dev;
40d4e3df 1607 __be16 proto = 0;
1da177e4 1608 int err;
3bdc0eba 1609 int extra_len = 0;
1ce4f28b 1610
1da177e4 1611 /*
1ce4f28b 1612 * Get and verify the address.
1da177e4
LT		 1613 */
1614
40d4e3df 1615 if (saddr) {
1da177e4 1616 if (msg->msg_namelen < sizeof(struct sockaddr))
40d4e3df
ED		 1617 return -EINVAL;
1618 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1619 proto = saddr->spkt_protocol;
1620 } else
1621 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1da177e4
LT		 1622
1623 /*
1ce4f28b 1624 * Find the device first to size check it
1da177e4
LT		 1625 */
1626
de74e92a 1627 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1a35ca80 1628retry:
654d1f8a
ED		 1629 rcu_read_lock();
1630 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1da177e4
LT		 1631 err = -ENODEV;
1632 if (dev == NULL)
1633 goto out_unlock;
1ce4f28b 1634
d5e76b0a
DM		 1635 err = -ENETDOWN;
1636 if (!(dev->flags & IFF_UP))
1637 goto out_unlock;
1638
1da177e4 1639 /*
40d4e3df
ED		 1640 * You may not queue a frame bigger than the mtu. This is the lowest level
1641 * raw protocol and you must do your own fragmentation at this level.
1da177e4 1642 */
1ce4f28b 1643
3bdc0eba
BG		 1644 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1645 if (!netif_supports_nofcs(dev)) {
1646 err = -EPROTONOSUPPORT;
1647 goto out_unlock;
1648 }
1649 extra_len = 4; /* We're doing our own CRC */
1650 }
1651
1da177e4 1652 err = -EMSGSIZE;
3bdc0eba 1653 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1da177e4
LT		 1654 goto out_unlock;
1655
1a35ca80
ED		 1656 if (!skb) {
1657 size_t reserved = LL_RESERVED_SPACE(dev);
4ce40912 1658 int tlen = dev->needed_tailroom;
1a35ca80
ED		 1659 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1660
1661 rcu_read_unlock();
4ce40912 1662 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1a35ca80
ED		 1663 if (skb == NULL)
1664 return -ENOBUFS;
1665 /* FIXME: Save some space for broken drivers that write a hard
1666 * header at transmission time by themselves. PPP is the notable
1667 * one here. This should really be fixed at the driver level.
1668 */
1669 skb_reserve(skb, reserved);
1670 skb_reset_network_header(skb);
1671
1672 /* Try to align data part correctly */
1673 if (hhlen) {
1674 skb->data -= hhlen;
1675 skb->tail -= hhlen;
1676 if (len < hhlen)
1677 skb_reset_network_header(skb);
1678 }
1679 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1680 if (err)
1681 goto out_free;
1682 goto retry;
1da177e4
LT		 1683 }
1684
3bdc0eba 1685 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
57f89bfa
BG		 1686 /* Earlier code assumed this would be a VLAN pkt,
1687 * double-check this now that we have the actual
1688 * packet in hand.
1689 */
1690 struct ethhdr *ehdr;
1691 skb_reset_mac_header(skb);
1692 ehdr = eth_hdr(skb);
1693 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1694 err = -EMSGSIZE;
1695 goto out_unlock;
1696 }
1697 }
1a35ca80 1698
1da177e4
LT		 1699 skb->protocol = proto;
1700 skb->dev = dev;
1701 skb->priority = sk->sk_priority;
2d37a186 1702 skb->mark = sk->sk_mark;
bf84a010
DB		 1703
1704 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1da177e4 1705
3bdc0eba
BG		 1706 if (unlikely(extra_len == 4))
1707 skb->no_fcs = 1;
1708
40893fd0 1709 skb_probe_transport_header(skb, 0);
c1aad275 1710
1da177e4 1711 dev_queue_xmit(skb);
654d1f8a 1712 rcu_read_unlock();
40d4e3df 1713 return len;
1da177e4 1714
1da177e4 1715out_unlock:
654d1f8a 1716 rcu_read_unlock();
1a35ca80
ED		 1717out_free:
1718 kfree_skb(skb);
1da177e4
LT		 1719 return err;
1720}
1da177e4 1721
eea49cc9 1722static unsigned int run_filter(const struct sk_buff *skb,
62ab0812 1723 const struct sock *sk,
dbcb5855 1724 unsigned int res)
1da177e4
LT		 1725{
1726 struct sk_filter *filter;
fda9ef5d 1727
80f8f102
ED		 1728 rcu_read_lock();
1729 filter = rcu_dereference(sk->sk_filter);
dbcb5855 1730 if (filter != NULL)
0a14842f 1731 res = SK_RUN_FILTER(filter, skb);
80f8f102 1732 rcu_read_unlock();
1da177e4 1733
dbcb5855 1734 return res;
1da177e4
LT		 1735}
1736
1737/*
62ab0812
ED		 1738 * This function makes lazy skb cloning in hope that most of packets
1739 * are discarded by BPF.
1740 *
1741 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1742 * and skb->cb are mangled. It works because (and until) packets
1743 * falling here are owned by current CPU. Output packets are cloned
1744 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1745 * sequencially, so that if we return skb to original state on exit,
1746 * we will not harm anyone.
1da177e4
LT		 1747 */
1748
40d4e3df
ED		 1749static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1750 struct packet_type *pt, struct net_device *orig_dev)
1da177e4
LT		 1751{
1752 struct sock *sk;
1753 struct sockaddr_ll *sll;
1754 struct packet_sock *po;
40d4e3df 1755 u8 *skb_head = skb->data;
1da177e4 1756 int skb_len = skb->len;
dbcb5855 1757 unsigned int snaplen, res;
1da177e4
LT		 1758
1759 if (skb->pkt_type == PACKET_LOOPBACK)
1760 goto drop;
1761
1762 sk = pt->af_packet_priv;
1763 po = pkt_sk(sk);
1764
09ad9bc7 1765 if (!net_eq(dev_net(dev), sock_net(sk)))
d12d01d6
DL		 1766 goto drop;
1767
1da177e4
LT		 1768 skb->dev = dev;
1769
3b04ddde 1770 if (dev->header_ops) {
1da177e4 1771 /* The device has an explicit notion of ll header,
62ab0812
ED		 1772 * exported to higher levels.
1773 *
1774 * Otherwise, the device hides details of its frame
1775 * structure, so that corresponding packet head is
1776 * never delivered to user.
1da177e4
LT		 1777 */
1778 if (sk->sk_type != SOCK_DGRAM)
98e399f8 1779 skb_push(skb, skb->data - skb_mac_header(skb));
1da177e4
LT		 1780 else if (skb->pkt_type == PACKET_OUTGOING) {
1781 /* Special case: outgoing packets have ll header at head */
bbe735e4 1782 skb_pull(skb, skb_network_offset(skb));
1da177e4
LT		 1783 }
1784 }
1785
1786 snaplen = skb->len;
1787
dbcb5855
DM		 1788 res = run_filter(skb, sk, snaplen);
1789 if (!res)
fda9ef5d 1790 goto drop_n_restore;
dbcb5855
DM		 1791 if (snaplen > res)
1792 snaplen = res;
1da177e4 1793
0fd7bac6 1794 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1da177e4
LT		 1795 goto drop_n_acct;
1796
1797 if (skb_shared(skb)) {
1798 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1799 if (nskb == NULL)
1800 goto drop_n_acct;
1801
1802 if (skb_head != skb->data) {
1803 skb->data = skb_head;
1804 skb->len = skb_len;
1805 }
abc4e4fa 1806 consume_skb(skb);
1da177e4
LT		 1807 skb = nskb;
1808 }
1809
ffbc6111
HX		 1810 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1811 sizeof(skb->cb));
1812
1813 sll = &PACKET_SKB_CB(skb)->sa.ll;
1da177e4
LT		 1814 sll->sll_family = AF_PACKET;
1815 sll->sll_hatype = dev->type;
1816 sll->sll_protocol = skb->protocol;
1817 sll->sll_pkttype = skb->pkt_type;
8032b464 1818 if (unlikely(po->origdev))
80feaacb
PWJ		 1819 sll->sll_ifindex = orig_dev->ifindex;
1820 else
1821 sll->sll_ifindex = dev->ifindex;
1da177e4 1822
b95cce35 1823 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1da177e4 1824
ffbc6111 1825 PACKET_SKB_CB(skb)->origlen = skb->len;
8dc41944 1826
1da177e4
LT		 1827 if (pskb_trim(skb, snaplen))
1828 goto drop_n_acct;
1829
1830 skb_set_owner_r(skb, sk);
1831 skb->dev = NULL;
adf30907 1832 skb_dst_drop(skb);
1da177e4 1833
84531c24
PO		 1834 /* drop conntrack reference */
1835 nf_reset(skb);
1836
1da177e4 1837 spin_lock(&sk->sk_receive_queue.lock);
ee80fbf3 1838 po->stats.stats1.tp_packets++;
3b885787 1839 skb->dropcount = atomic_read(&sk->sk_drops);
1da177e4
LT		 1840 __skb_queue_tail(&sk->sk_receive_queue, skb);
1841 spin_unlock(&sk->sk_receive_queue.lock);
676d2369 1842 sk->sk_data_ready(sk);
1da177e4
LT		 1843 return 0;
1844
1845drop_n_acct:
7091fbd8 1846 spin_lock(&sk->sk_receive_queue.lock);
ee80fbf3 1847 po->stats.stats1.tp_drops++;
7091fbd8
WB		 1848 atomic_inc(&sk->sk_drops);
1849 spin_unlock(&sk->sk_receive_queue.lock);
1da177e4
LT		 1850
1851drop_n_restore:
1852 if (skb_head != skb->data && skb_shared(skb)) {
1853 skb->data = skb_head;
1854 skb->len = skb_len;
1855 }
1856drop:
ead2ceb0 1857 consume_skb(skb);
1da177e4
LT		 1858 return 0;
1859}
1860
40d4e3df
ED		 1861static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1862 struct packet_type *pt, struct net_device *orig_dev)
1da177e4
LT		 1863{
1864 struct sock *sk;
1865 struct packet_sock *po;
1866 struct sockaddr_ll *sll;
184f489e 1867 union tpacket_uhdr h;
40d4e3df 1868 u8 *skb_head = skb->data;
1da177e4 1869 int skb_len = skb->len;
dbcb5855 1870 unsigned int snaplen, res;
f6fb8f10 1871 unsigned long status = TP_STATUS_USER;
bbd6ef87 1872 unsigned short macoff, netoff, hdrlen;
1da177e4 1873 struct sk_buff *copy_skb = NULL;
bbd6ef87 1874 struct timespec ts;
b9c32fb2 1875 __u32 ts_status;
1da177e4 1876
51846355
AW		 1877 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1878 * We may add members to them until current aligned size without forcing
1879 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1880 */
1881 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1882 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1883
1da177e4
LT		 1884 if (skb->pkt_type == PACKET_LOOPBACK)
1885 goto drop;
1886
1887 sk = pt->af_packet_priv;
1888 po = pkt_sk(sk);
1889
09ad9bc7 1890 if (!net_eq(dev_net(dev), sock_net(sk)))
d12d01d6
DL		 1891 goto drop;
1892
3b04ddde 1893 if (dev->header_ops) {
1da177e4 1894 if (sk->sk_type != SOCK_DGRAM)
98e399f8 1895 skb_push(skb, skb->data - skb_mac_header(skb));
1da177e4
LT		 1896 else if (skb->pkt_type == PACKET_OUTGOING) {
1897 /* Special case: outgoing packets have ll header at head */
bbe735e4 1898 skb_pull(skb, skb_network_offset(skb));
1da177e4
LT		 1899 }
1900 }
1901
8dc41944
HX		 1902 if (skb->ip_summed == CHECKSUM_PARTIAL)
1903 status |= TP_STATUS_CSUMNOTREADY;
1904
1da177e4
LT		 1905 snaplen = skb->len;
1906
dbcb5855
DM		 1907 res = run_filter(skb, sk, snaplen);
1908 if (!res)
fda9ef5d 1909 goto drop_n_restore;
dbcb5855
DM		 1910 if (snaplen > res)
1911 snaplen = res;
1da177e4
LT		 1912
1913 if (sk->sk_type == SOCK_DGRAM) {
8913336a
PM		 1914 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1915 po->tp_reserve;
1da177e4 1916 } else {
95c96174 1917 unsigned int maclen = skb_network_offset(skb);
bbd6ef87 1918 netoff = TPACKET_ALIGN(po->tp_hdrlen +
8913336a
PM		 1919 (maclen < 16 ? 16 : maclen)) +
1920 po->tp_reserve;
1da177e4
LT		 1921 macoff = netoff - maclen;
1922 }
f6fb8f10 1923 if (po->tp_version <= TPACKET_V2) {
1924 if (macoff + snaplen > po->rx_ring.frame_size) {
1925 if (po->copy_thresh &&
0fd7bac6 1926 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
f6fb8f10 1927 if (skb_shared(skb)) {
1928 copy_skb = skb_clone(skb, GFP_ATOMIC);
1929 } else {
1930 copy_skb = skb_get(skb);
1931 skb_head = skb->data;
1932 }
1933 if (copy_skb)
1934 skb_set_owner_r(copy_skb, sk);
1da177e4 1935 }
f6fb8f10 1936 snaplen = po->rx_ring.frame_size - macoff;
1937 if ((int)snaplen < 0)
1938 snaplen = 0;
1da177e4 1939 }
dc808110
ED		 1940 } else if (unlikely(macoff + snaplen >
1941 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1942 u32 nval;
1943
1944 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1945 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1946 snaplen, nval, macoff);
1947 snaplen = nval;
1948 if (unlikely((int)snaplen < 0)) {
1949 snaplen = 0;
1950 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1951 }
1da177e4 1952 }
1da177e4 1953 spin_lock(&sk->sk_receive_queue.lock);
f6fb8f10 1954 h.raw = packet_current_rx_frame(po, skb,
1955 TP_STATUS_KERNEL, (macoff+snaplen));
bbd6ef87 1956 if (!h.raw)
1da177e4 1957 goto ring_is_full;
f6fb8f10 1958 if (po->tp_version <= TPACKET_V2) {
1959 packet_increment_rx_head(po, &po->rx_ring);
1960 /*
1961 * LOSING will be reported till you read the stats,
1962 * because it's COR - Clear On Read.
1963 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1964 * at packet level.
1965 */
ee80fbf3 1966 if (po->stats.stats1.tp_drops)
f6fb8f10 1967 status |= TP_STATUS_LOSING;
1968 }
ee80fbf3 1969 po->stats.stats1.tp_packets++;
1da177e4
LT		 1970 if (copy_skb) {
1971 status |= TP_STATUS_COPY;
1972 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1973 }
1da177e4
LT		 1974 spin_unlock(&sk->sk_receive_queue.lock);
1975
bbd6ef87 1976 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
b9c32fb2
DB		 1977
1978 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
7a51384c 1979 getnstimeofday(&ts);
1da177e4 1980
b9c32fb2
DB		 1981 status |= ts_status;
1982
bbd6ef87
PM		 1983 switch (po->tp_version) {
1984 case TPACKET_V1:
1985 h.h1->tp_len = skb->len;
1986 h.h1->tp_snaplen = snaplen;
1987 h.h1->tp_mac = macoff;
1988 h.h1->tp_net = netoff;
4b457bdf
DB		 1989 h.h1->tp_sec = ts.tv_sec;
1990 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
bbd6ef87
PM		 1991 hdrlen = sizeof(*h.h1);
1992 break;
1993 case TPACKET_V2:
1994 h.h2->tp_len = skb->len;
1995 h.h2->tp_snaplen = snaplen;
1996 h.h2->tp_mac = macoff;
1997 h.h2->tp_net = netoff;
bbd6ef87
PM		 1998 h.h2->tp_sec = ts.tv_sec;
1999 h.h2->tp_nsec = ts.tv_nsec;
a3bcc23e
BG		 2000 if (vlan_tx_tag_present(skb)) {
2001 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
a0cdfcf3
AW		 2002 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2003 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
a3bcc23e
BG		 2004 } else {
2005 h.h2->tp_vlan_tci = 0;
a0cdfcf3 2006 h.h2->tp_vlan_tpid = 0;
a3bcc23e 2007 }
e4d26f4b 2008 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
bbd6ef87
PM		 2009 hdrlen = sizeof(*h.h2);
2010 break;
f6fb8f10 2011 case TPACKET_V3:
2012 /* tp_nxt_offset,vlan are already populated above.
2013 * So DONT clear those fields here
2014 */
2015 h.h3->tp_status |= status;
2016 h.h3->tp_len = skb->len;
2017 h.h3->tp_snaplen = snaplen;
2018 h.h3->tp_mac = macoff;
2019 h.h3->tp_net = netoff;
f6fb8f10 2020 h.h3->tp_sec = ts.tv_sec;
2021 h.h3->tp_nsec = ts.tv_nsec;
e4d26f4b 2022 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
f6fb8f10 2023 hdrlen = sizeof(*h.h3);
2024 break;
bbd6ef87
PM		 2025 default:
2026 BUG();
2027 }
1da177e4 2028
bbd6ef87 2029 sll = h.raw + TPACKET_ALIGN(hdrlen);
b95cce35 2030 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1da177e4
LT		 2031 sll->sll_family = AF_PACKET;
2032 sll->sll_hatype = dev->type;
2033 sll->sll_protocol = skb->protocol;
2034 sll->sll_pkttype = skb->pkt_type;
8032b464 2035 if (unlikely(po->origdev))
80feaacb
PWJ		 2036 sll->sll_ifindex = orig_dev->ifindex;
2037 else
2038 sll->sll_ifindex = dev->ifindex;
1da177e4 2039
e16aa207 2040 smp_mb();
f0d4eb29 2041
f6dafa95 2042#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
f0d4eb29 2043 if (po->tp_version <= TPACKET_V2) {
0af55bb5
CG		 2044 u8 *start, *end;
2045
f0d4eb29
DB		 2046 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2047 macoff + snaplen);
2048
2049 for (start = h.raw; start < end; start += PAGE_SIZE)
2050 flush_dcache_page(pgv_to_page(start));
1da177e4 2051 }
f0d4eb29 2052 smp_wmb();
f6dafa95 2053#endif
f0d4eb29 2054
f6fb8f10 2055 if (po->tp_version <= TPACKET_V2)
2056 __packet_set_status(po, h.raw, status);
2057 else
2058 prb_clear_blk_fill_status(&po->rx_ring);
1da177e4 2059
676d2369 2060 sk->sk_data_ready(sk);
1da177e4
LT		 2061
2062drop_n_restore:
2063 if (skb_head != skb->data && skb_shared(skb)) {
2064 skb->data = skb_head;
2065 skb->len = skb_len;
2066 }
2067drop:
1ce4f28b 2068 kfree_skb(skb);
1da177e4
LT		 2069 return 0;
2070
2071ring_is_full:
ee80fbf3 2072 po->stats.stats1.tp_drops++;
1da177e4
LT		 2073 spin_unlock(&sk->sk_receive_queue.lock);
2074
676d2369 2075 sk->sk_data_ready(sk);
acb5d75b 2076 kfree_skb(copy_skb);
1da177e4
LT		 2077 goto drop_n_restore;
2078}
2079
69e3c75f
JB		 2080static void tpacket_destruct_skb(struct sk_buff *skb)
2081{
2082 struct packet_sock *po = pkt_sk(skb->sk);
1da177e4 2083
69e3c75f 2084 if (likely(po->tx_ring.pg_vec)) {
f0d4eb29 2085 void *ph;
b9c32fb2
DB		 2086 __u32 ts;
2087
69e3c75f 2088 ph = skb_shinfo(skb)->destructor_arg;
b0138408 2089 packet_dec_pending(&po->tx_ring);
b9c32fb2
DB		 2090
2091 ts = __packet_set_timestamp(po, ph, skb);
2092 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
69e3c75f
JB		 2093 }
2094
2095 sock_wfree(skb);
2096}
2097
40d4e3df
ED		 2098static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2099 void *frame, struct net_device *dev, int size_max,
ae641949 2100 __be16 proto, unsigned char *addr, int hlen)
69e3c75f 2101{
184f489e 2102 union tpacket_uhdr ph;
09effa67 2103 int to_write, offset, len, tp_len, nr_frags, len_max;
69e3c75f
JB		 2104 struct socket *sock = po->sk.sk_socket;
2105 struct page *page;
2106 void *data;
2107 int err;
2108
2109 ph.raw = frame;
2110
2111 skb->protocol = proto;
2112 skb->dev = dev;
2113 skb->priority = po->sk.sk_priority;
2d37a186 2114 skb->mark = po->sk.sk_mark;
2e31396f 2115 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
69e3c75f
JB		 2116 skb_shinfo(skb)->destructor_arg = ph.raw;
2117
2118 switch (po->tp_version) {
2119 case TPACKET_V2:
2120 tp_len = ph.h2->tp_len;
2121 break;
2122 default:
2123 tp_len = ph.h1->tp_len;
2124 break;
2125 }
09effa67
DM		 2126 if (unlikely(tp_len > size_max)) {
2127 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2128 return -EMSGSIZE;
2129 }
69e3c75f 2130
ae641949 2131 skb_reserve(skb, hlen);
69e3c75f 2132 skb_reset_network_header(skb);
c1aad275 2133
d346a3fa
DB		 2134 if (!packet_use_direct_xmit(po))
2135 skb_probe_transport_header(skb, 0);
2136 if (unlikely(po->tp_tx_has_off)) {
5920cd3a
PC		 2137 int off_min, off_max, off;
2138 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2139 off_max = po->tx_ring.frame_size - tp_len;
2140 if (sock->type == SOCK_DGRAM) {
2141 switch (po->tp_version) {
2142 case TPACKET_V2:
2143 off = ph.h2->tp_net;
2144 break;
2145 default:
2146 off = ph.h1->tp_net;
2147 break;
2148 }
2149 } else {
2150 switch (po->tp_version) {
2151 case TPACKET_V2:
2152 off = ph.h2->tp_mac;
2153 break;
2154 default:
2155 off = ph.h1->tp_mac;
2156 break;
2157 }
2158 }
2159 if (unlikely((off < off_min) || (off_max < off)))
2160 return -EINVAL;
2161 data = ph.raw + off;
2162 } else {
2163 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2164 }
69e3c75f
JB		 2165 to_write = tp_len;
2166
2167 if (sock->type == SOCK_DGRAM) {
2168 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2169 NULL, tp_len);
2170 if (unlikely(err < 0))
2171 return -EINVAL;
40d4e3df 2172 } else if (dev->hard_header_len) {
69e3c75f
JB		 2173 /* net device doesn't like empty head */
2174 if (unlikely(tp_len <= dev->hard_header_len)) {
40d4e3df
ED		 2175 pr_err("packet size is too short (%d < %d)\n",
2176 tp_len, dev->hard_header_len);
69e3c75f
JB		 2177 return -EINVAL;
2178 }
2179
2180 skb_push(skb, dev->hard_header_len);
2181 err = skb_store_bits(skb, 0, data,
2182 dev->hard_header_len);
2183 if (unlikely(err))
2184 return err;
2185
2186 data += dev->hard_header_len;
2187 to_write -= dev->hard_header_len;
2188 }
2189
69e3c75f
JB		 2190 offset = offset_in_page(data);
2191 len_max = PAGE_SIZE - offset;
2192 len = ((to_write > len_max) ? len_max : to_write);
2193
2194 skb->data_len = to_write;
2195 skb->len += to_write;
2196 skb->truesize += to_write;
2197 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2198
2199 while (likely(to_write)) {
2200 nr_frags = skb_shinfo(skb)->nr_frags;
2201
2202 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
40d4e3df
ED		 2203 pr_err("Packet exceed the number of skb frags(%lu)\n",
2204 MAX_SKB_FRAGS);
69e3c75f
JB		 2205 return -EFAULT;
2206 }
2207
0af55bb5
CG		 2208 page = pgv_to_page(data);
2209 data += len;
69e3c75f
JB		 2210 flush_dcache_page(page);
2211 get_page(page);
0af55bb5 2212 skb_fill_page_desc(skb, nr_frags, page, offset, len);
69e3c75f
JB		 2213 to_write -= len;
2214 offset = 0;
2215 len_max = PAGE_SIZE;
2216 len = ((to_write > len_max) ? len_max : to_write);
2217 }
2218
2219 return tp_len;
2220}
2221
2222static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2223{
69e3c75f
JB		 2224 struct sk_buff *skb;
2225 struct net_device *dev;
2226 __be16 proto;
09effa67 2227 int err, reserve = 0;
40d4e3df 2228 void *ph;
342dfc30 2229 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
87a2fd28 2230 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
69e3c75f
JB		 2231 int tp_len, size_max;
2232 unsigned char *addr;
2233 int len_sum = 0;
9e67030a 2234 int status = TP_STATUS_AVAILABLE;
ae641949 2235 int hlen, tlen;
69e3c75f 2236
69e3c75f
JB		 2237 mutex_lock(&po->pg_vec_lock);
2238
66e56cd4 2239 if (likely(saddr == NULL)) {
e40526cb 2240 dev = packet_cached_dev_get(po);
69e3c75f
JB		 2241 proto = po->num;
2242 addr = NULL;
2243 } else {
2244 err = -EINVAL;
2245 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2246 goto out;
2247 if (msg->msg_namelen < (saddr->sll_halen
2248 + offsetof(struct sockaddr_ll,
2249 sll_addr)))
2250 goto out;
69e3c75f
JB		 2251 proto = saddr->sll_protocol;
2252 addr = saddr->sll_addr;
827d9780 2253 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
69e3c75f
JB		 2254 }
2255
69e3c75f
JB		 2256 err = -ENXIO;
2257 if (unlikely(dev == NULL))
2258 goto out;
69e3c75f
JB		 2259 err = -ENETDOWN;
2260 if (unlikely(!(dev->flags & IFF_UP)))
2261 goto out_put;
2262
52f1454f 2263 reserve = dev->hard_header_len + VLAN_HLEN;
69e3c75f 2264 size_max = po->tx_ring.frame_size
b5dd884e 2265 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
69e3c75f 2266
09effa67
DM		 2267 if (size_max > dev->mtu + reserve)
2268 size_max = dev->mtu + reserve;
2269
69e3c75f
JB		 2270 do {
2271 ph = packet_current_frame(po, &po->tx_ring,
87a2fd28 2272 TP_STATUS_SEND_REQUEST);
69e3c75f 2273 if (unlikely(ph == NULL)) {
87a2fd28
DB		 2274 if (need_wait && need_resched())
2275 schedule();
69e3c75f
JB		 2276 continue;
2277 }
2278
2279 status = TP_STATUS_SEND_REQUEST;
ae641949
HX		 2280 hlen = LL_RESERVED_SPACE(dev);
2281 tlen = dev->needed_tailroom;
69e3c75f 2282 skb = sock_alloc_send_skb(&po->sk,
ae641949 2283 hlen + tlen + sizeof(struct sockaddr_ll),
69e3c75f
JB		 2284 0, &err);
2285
2286 if (unlikely(skb == NULL))
2287 goto out_status;
2288
2289 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
52f1454f
DB		 2290 addr, hlen);
2291 if (tp_len > dev->mtu + dev->hard_header_len) {
2292 struct ethhdr *ehdr;
2293 /* Earlier code assumed this would be a VLAN pkt,
2294 * double-check this now that we have the actual
2295 * packet in hand.
2296 */
69e3c75f 2297
52f1454f
DB		 2298 skb_reset_mac_header(skb);
2299 ehdr = eth_hdr(skb);
2300 if (ehdr->h_proto != htons(ETH_P_8021Q))
2301 tp_len = -EMSGSIZE;
2302 }
69e3c75f
JB		 2303 if (unlikely(tp_len < 0)) {
2304 if (po->tp_loss) {
2305 __packet_set_status(po, ph,
2306 TP_STATUS_AVAILABLE);
2307 packet_increment_head(&po->tx_ring);
2308 kfree_skb(skb);
2309 continue;
2310 } else {
2311 status = TP_STATUS_WRONG_FORMAT;
2312 err = tp_len;
2313 goto out_status;
2314 }
2315 }
2316
0fd5d57b
DB		 2317 packet_pick_tx_queue(dev, skb);
2318
69e3c75f
JB		 2319 skb->destructor = tpacket_destruct_skb;
2320 __packet_set_status(po, ph, TP_STATUS_SENDING);
b0138408 2321 packet_inc_pending(&po->tx_ring);
69e3c75f
JB		 2322
2323 status = TP_STATUS_SEND_REQUEST;
d346a3fa 2324 err = po->xmit(skb);
eb70df13
JP		 2325 if (unlikely(err > 0)) {
2326 err = net_xmit_errno(err);
2327 if (err && __packet_get_status(po, ph) ==
2328 TP_STATUS_AVAILABLE) {
2329 /* skb was destructed already */
2330 skb = NULL;
2331 goto out_status;
2332 }
2333 /*
2334 * skb was dropped but not destructed yet;
2335 * let's treat it like congestion or err < 0
2336 */
2337 err = 0;
2338 }
69e3c75f
JB		 2339 packet_increment_head(&po->tx_ring);
2340 len_sum += tp_len;
b0138408
DB		 2341 } while (likely((ph != NULL) ||
2342 /* Note: packet_read_pending() might be slow if we have
2343 * to call it as it's per_cpu variable, but in fast-path
2344 * we already short-circuit the loop with the first
2345 * condition, and luckily don't have to go that path
2346 * anyway.
2347 */
2348 (need_wait && packet_read_pending(&po->tx_ring))));
69e3c75f
JB		 2349
2350 err = len_sum;
2351 goto out_put;
2352
69e3c75f
JB		 2353out_status:
2354 __packet_set_status(po, ph, status);
2355 kfree_skb(skb);
2356out_put:
e40526cb 2357 dev_put(dev);
69e3c75f
JB		 2358out:
2359 mutex_unlock(&po->pg_vec_lock);
2360 return err;
2361}
69e3c75f 2362
eea49cc9
OJ		 2363static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2364 size_t reserve, size_t len,
2365 size_t linear, int noblock,
2366 int *err)
bfd5f4a3
SS		 2367{
2368 struct sk_buff *skb;
2369
2370 /* Under a page? Don't bother with paged skb. */
2371 if (prepad + len < PAGE_SIZE || !linear)
2372 linear = len;
2373
2374 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
28d64271 2375 err, 0);
bfd5f4a3
SS		 2376 if (!skb)
2377 return NULL;
2378
2379 skb_reserve(skb, reserve);
2380 skb_put(skb, linear);
2381 skb->data_len = len - linear;
2382 skb->len += len - linear;
2383
2384 return skb;
2385}
2386
d346a3fa 2387static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
1da177e4
LT		 2388{
2389 struct sock *sk = sock->sk;
342dfc30 2390 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
1da177e4
LT		 2391 struct sk_buff *skb;
2392 struct net_device *dev;
0e11c91e 2393 __be16 proto;
1da177e4 2394 unsigned char *addr;
827d9780 2395 int err, reserve = 0;
bfd5f4a3
SS		 2396 struct virtio_net_hdr vnet_hdr = { 0 };
2397 int offset = 0;
2398 int vnet_hdr_len;
2399 struct packet_sock *po = pkt_sk(sk);
2400 unsigned short gso_type = 0;
ae641949 2401 int hlen, tlen;
3bdc0eba 2402 int extra_len = 0;
1da177e4
LT		 2403
2404 /*
1ce4f28b 2405 * Get and verify the address.
1da177e4 2406 */
1ce4f28b 2407
66e56cd4 2408 if (likely(saddr == NULL)) {
e40526cb 2409 dev = packet_cached_dev_get(po);
1da177e4
LT		 2410 proto = po->num;
2411 addr = NULL;
2412 } else {
2413 err = -EINVAL;
2414 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2415 goto out;
0fb375fb
EB		 2416 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2417 goto out;
1da177e4
LT		 2418 proto = saddr->sll_protocol;
2419 addr = saddr->sll_addr;
827d9780 2420 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
1da177e4
LT		 2421 }
2422
1da177e4 2423 err = -ENXIO;
e40526cb 2424 if (unlikely(dev == NULL))
1da177e4 2425 goto out_unlock;
d5e76b0a 2426 err = -ENETDOWN;
e40526cb 2427 if (unlikely(!(dev->flags & IFF_UP)))
d5e76b0a
DM		 2428 goto out_unlock;
2429
e40526cb
DB		 2430 if (sock->type == SOCK_RAW)
2431 reserve = dev->hard_header_len;
bfd5f4a3
SS		 2432 if (po->has_vnet_hdr) {
2433 vnet_hdr_len = sizeof(vnet_hdr);
2434
2435 err = -EINVAL;
2436 if (len < vnet_hdr_len)
2437 goto out_unlock;
2438
2439 len -= vnet_hdr_len;
2440
2441 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2442 vnet_hdr_len);
2443 if (err < 0)
2444 goto out_unlock;
2445
2446 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2447 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2448 vnet_hdr.hdr_len))
2449 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2450 vnet_hdr.csum_offset + 2;
2451
2452 err = -EINVAL;
2453 if (vnet_hdr.hdr_len > len)
2454 goto out_unlock;
2455
2456 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2457 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2458 case VIRTIO_NET_HDR_GSO_TCPV4:
2459 gso_type = SKB_GSO_TCPV4;
2460 break;
2461 case VIRTIO_NET_HDR_GSO_TCPV6:
2462 gso_type = SKB_GSO_TCPV6;
2463 break;
2464 case VIRTIO_NET_HDR_GSO_UDP:
2465 gso_type = SKB_GSO_UDP;
2466 break;
2467 default:
2468 goto out_unlock;
2469 }
2470
2471 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2472 gso_type |= SKB_GSO_TCP_ECN;
2473
2474 if (vnet_hdr.gso_size == 0)
2475 goto out_unlock;
2476
2477 }
2478 }
2479
3bdc0eba
BG		 2480 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2481 if (!netif_supports_nofcs(dev)) {
2482 err = -EPROTONOSUPPORT;
2483 goto out_unlock;
2484 }
2485 extra_len = 4; /* We're doing our own CRC */
2486 }
2487
1da177e4 2488 err = -EMSGSIZE;
3bdc0eba 2489 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
1da177e4
LT		 2490 goto out_unlock;
2491
bfd5f4a3 2492 err = -ENOBUFS;
ae641949
HX		 2493 hlen = LL_RESERVED_SPACE(dev);
2494 tlen = dev->needed_tailroom;
2495 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
bfd5f4a3 2496 msg->msg_flags & MSG_DONTWAIT, &err);
40d4e3df 2497 if (skb == NULL)
1da177e4
LT		 2498 goto out_unlock;
2499
bfd5f4a3 2500 skb_set_network_header(skb, reserve);
1da177e4 2501
0c4e8581
SH		 2502 err = -EINVAL;
2503 if (sock->type == SOCK_DGRAM &&
bfd5f4a3 2504 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
0c4e8581 2505 goto out_free;
1da177e4
LT		 2506
2507 /* Returns -EFAULT on error */
bfd5f4a3 2508 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1da177e4
LT		 2509 if (err)
2510 goto out_free;
bf84a010
DB		 2511
2512 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1da177e4 2513
3bdc0eba 2514 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
09effa67
DM		 2515 /* Earlier code assumed this would be a VLAN pkt,
2516 * double-check this now that we have the actual
2517 * packet in hand.
2518 */
2519 struct ethhdr *ehdr;
2520 skb_reset_mac_header(skb);
2521 ehdr = eth_hdr(skb);
2522 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2523 err = -EMSGSIZE;
2524 goto out_free;
2525 }
57f89bfa
BG		 2526 }
2527
09effa67
DM		 2528 skb->protocol = proto;
2529 skb->dev = dev;
1da177e4 2530 skb->priority = sk->sk_priority;
2d37a186 2531 skb->mark = sk->sk_mark;
0fd5d57b
DB		 2532
2533 packet_pick_tx_queue(dev, skb);
1da177e4 2534
bfd5f4a3
SS		 2535 if (po->has_vnet_hdr) {
2536 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2537 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2538 vnet_hdr.csum_offset)) {
2539 err = -EINVAL;
2540 goto out_free;
2541 }
2542 }
2543
2544 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2545 skb_shinfo(skb)->gso_type = gso_type;
2546
2547 /* Header must be checked, and gso_segs computed. */
2548 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2549 skb_shinfo(skb)->gso_segs = 0;
2550
2551 len += vnet_hdr_len;
2552 }
2553
d346a3fa
DB		 2554 if (!packet_use_direct_xmit(po))
2555 skb_probe_transport_header(skb, reserve);
3bdc0eba
BG		 2556 if (unlikely(extra_len == 4))
2557 skb->no_fcs = 1;
2558
d346a3fa 2559 err = po->xmit(skb);
1da177e4
LT		 2560 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2561 goto out_unlock;
2562
e40526cb 2563 dev_put(dev);
1da177e4 2564
40d4e3df 2565 return len;
1da177e4
LT		 2566
2567out_free:
2568 kfree_skb(skb);
2569out_unlock:
e40526cb 2570 if (dev)
1da177e4
LT		 2571 dev_put(dev);
2572out:
2573 return err;
2574}
2575
69e3c75f
JB		 2576static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2577 struct msghdr *msg, size_t len)
2578{
69e3c75f
JB		 2579 struct sock *sk = sock->sk;
2580 struct packet_sock *po = pkt_sk(sk);
d346a3fa 2581
69e3c75f
JB		 2582 if (po->tx_ring.pg_vec)
2583 return tpacket_snd(po, msg);
2584 else
69e3c75f
JB		 2585 return packet_snd(sock, msg, len);
2586}
2587
1da177e4
LT		 2588/*
2589 * Close a PACKET socket. This is fairly simple. We immediately go
2590 * to 'closed' state and remove our protocol entry in the device list.
2591 */
2592
2593static int packet_release(struct socket *sock)
2594{
2595 struct sock *sk = sock->sk;
2596 struct packet_sock *po;
d12d01d6 2597 struct net *net;
f6fb8f10 2598 union tpacket_req_u req_u;
1da177e4
LT		 2599
2600 if (!sk)
2601 return 0;
2602
3b1e0a65 2603 net = sock_net(sk);
1da177e4
LT		 2604 po = pkt_sk(sk);
2605
0fa7fa98 2606 mutex_lock(&net->packet.sklist_lock);
808f5114 2607 sk_del_node_init_rcu(sk);
0fa7fa98
PE		 2608 mutex_unlock(&net->packet.sklist_lock);
2609
2610 preempt_disable();
920de804 2611 sock_prot_inuse_add(net, sk->sk_prot, -1);
0fa7fa98 2612 preempt_enable();
1da177e4 2613
808f5114 2614 spin_lock(&po->bind_lock);
ce06b03e 2615 unregister_prot_hook(sk, false);
66e56cd4
DB		 2616 packet_cached_dev_reset(po);
2617
160ff18a
BG		 2618 if (po->prot_hook.dev) {
2619 dev_put(po->prot_hook.dev);
2620 po->prot_hook.dev = NULL;
2621 }
808f5114 2622 spin_unlock(&po->bind_lock);
1da177e4 2623
1da177e4 2624 packet_flush_mclist(sk);
1da177e4 2625
9665d5d6
PS		 2626 if (po->rx_ring.pg_vec) {
2627 memset(&req_u, 0, sizeof(req_u));
f6fb8f10 2628 packet_set_ring(sk, &req_u, 1, 0);
9665d5d6 2629 }
69e3c75f 2630
9665d5d6
PS		 2631 if (po->tx_ring.pg_vec) {
2632 memset(&req_u, 0, sizeof(req_u));
f6fb8f10 2633 packet_set_ring(sk, &req_u, 1, 1);
9665d5d6 2634 }
1da177e4 2635
dc99f600
DM		 2636 fanout_release(sk);
2637
808f5114 2638 synchronize_net();
1da177e4
LT		 2639 /*
2640 * Now the socket is dead. No more input will appear.
2641 */
1da177e4
LT		 2642 sock_orphan(sk);
2643 sock->sk = NULL;
2644
2645 /* Purge queues */
2646
2647 skb_queue_purge(&sk->sk_receive_queue);
b0138408 2648 packet_free_pending(po);
17ab56a2 2649 sk_refcnt_debug_release(sk);
1da177e4
LT		 2650
2651 sock_put(sk);
2652 return 0;
2653}
2654
2655/*
2656 * Attach a packet hook.
2657 */
2658
902fefb8 2659static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
1da177e4
LT		 2660{
2661 struct packet_sock *po = pkt_sk(sk);
902fefb8
DB		 2662 const struct net_device *dev_curr;
2663 __be16 proto_curr;
2664 bool need_rehook;
dc99f600 2665
aef950b4
WY		 2666 if (po->fanout) {
2667 if (dev)
2668 dev_put(dev);
2669
dc99f600 2670 return -EINVAL;
aef950b4 2671 }
1da177e4
LT		 2672
2673 lock_sock(sk);
1da177e4 2674 spin_lock(&po->bind_lock);
66e56cd4 2675
902fefb8
DB		 2676 proto_curr = po->prot_hook.type;
2677 dev_curr = po->prot_hook.dev;
2678
2679 need_rehook = proto_curr != proto || dev_curr != dev;
2680
2681 if (need_rehook) {
2682 unregister_prot_hook(sk, true);
1da177e4 2683
902fefb8
DB		 2684 po->num = proto;
2685 po->prot_hook.type = proto;
1da177e4 2686
902fefb8
DB		 2687 if (po->prot_hook.dev)
2688 dev_put(po->prot_hook.dev);
2689
2690 po->prot_hook.dev = dev;
2691
2692 po->ifindex = dev ? dev->ifindex : 0;
2693 packet_cached_dev_assign(po, dev);
2694 }
66e56cd4 2695
902fefb8 2696 if (proto == 0 || !need_rehook)
1da177e4
LT		 2697 goto out_unlock;
2698
be85d4ad 2699 if (!dev || (dev->flags & IFF_UP)) {
ce06b03e 2700 register_prot_hook(sk);
be85d4ad
UT		 2701 } else {
2702 sk->sk_err = ENETDOWN;
2703 if (!sock_flag(sk, SOCK_DEAD))
2704 sk->sk_error_report(sk);
1da177e4
LT		 2705 }
2706
2707out_unlock:
2708 spin_unlock(&po->bind_lock);
2709 release_sock(sk);
2710 return 0;
2711}
2712
2713/*
2714 * Bind a packet socket to a device
2715 */
2716
40d4e3df
ED		 2717static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2718 int addr_len)
1da177e4 2719{
40d4e3df 2720 struct sock *sk = sock->sk;
1da177e4
LT		 2721 char name[15];
2722 struct net_device *dev;
2723 int err = -ENODEV;
1ce4f28b 2724
1da177e4
LT		 2725 /*
2726 * Check legality
2727 */
1ce4f28b 2728
8ae55f04 2729 if (addr_len != sizeof(struct sockaddr))
1da177e4 2730 return -EINVAL;
40d4e3df 2731 strlcpy(name, uaddr->sa_data, sizeof(name));
1da177e4 2732
3b1e0a65 2733 dev = dev_get_by_name(sock_net(sk), name);
160ff18a 2734 if (dev)
1da177e4 2735 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1da177e4
LT		 2736 return err;
2737}
1da177e4
LT		 2738
2739static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2740{
40d4e3df
ED		 2741 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2742 struct sock *sk = sock->sk;
1da177e4
LT		 2743 struct net_device *dev = NULL;
2744 int err;
2745
2746
2747 /*
2748 * Check legality
2749 */
1ce4f28b 2750
1da177e4
LT		 2751 if (addr_len < sizeof(struct sockaddr_ll))
2752 return -EINVAL;
2753 if (sll->sll_family != AF_PACKET)
2754 return -EINVAL;
2755
2756 if (sll->sll_ifindex) {
2757 err = -ENODEV;
3b1e0a65 2758 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1da177e4
LT		 2759 if (dev == NULL)
2760 goto out;
2761 }
2762 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1da177e4
LT		 2763
2764out:
2765 return err;
2766}
2767
2768static struct proto packet_proto = {
2769 .name = "PACKET",
2770 .owner = THIS_MODULE,
2771 .obj_size = sizeof(struct packet_sock),
2772};
2773
2774/*
1ce4f28b 2775 * Create a packet of type SOCK_PACKET.
1da177e4
LT		 2776 */
2777
3f378b68
EP		 2778static int packet_create(struct net *net, struct socket *sock, int protocol,
2779 int kern)
1da177e4
LT		 2780{
2781 struct sock *sk;
2782 struct packet_sock *po;
0e11c91e 2783 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1da177e4
LT		 2784 int err;
2785
df008c91 2786 if (!ns_capable(net->user_ns, CAP_NET_RAW))
1da177e4 2787 return -EPERM;
be02097c
DM		 2788 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2789 sock->type != SOCK_PACKET)
1da177e4
LT		 2790 return -ESOCKTNOSUPPORT;
2791
2792 sock->state = SS_UNCONNECTED;
2793
2794 err = -ENOBUFS;
6257ff21 2795 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1da177e4
LT		 2796 if (sk == NULL)
2797 goto out;
2798
2799 sock->ops = &packet_ops;
1da177e4
LT		 2800 if (sock->type == SOCK_PACKET)
2801 sock->ops = &packet_ops_spkt;
be02097c 2802
1da177e4
LT		 2803 sock_init_data(sock, sk);
2804
2805 po = pkt_sk(sk);
2806 sk->sk_family = PF_PACKET;
0e11c91e 2807 po->num = proto;
d346a3fa 2808 po->xmit = dev_queue_xmit;
66e56cd4 2809
b0138408
DB		 2810 err = packet_alloc_pending(po);
2811 if (err)
2812 goto out2;
2813
66e56cd4 2814 packet_cached_dev_reset(po);
1da177e4
LT		 2815
2816 sk->sk_destruct = packet_sock_destruct;
17ab56a2 2817 sk_refcnt_debug_inc(sk);
1da177e4
LT		 2818
2819 /*
2820 * Attach a protocol block
2821 */
2822
2823 spin_lock_init(&po->bind_lock);
905db440 2824 mutex_init(&po->pg_vec_lock);
1da177e4 2825 po->prot_hook.func = packet_rcv;
be02097c 2826
1da177e4
LT		 2827 if (sock->type == SOCK_PACKET)
2828 po->prot_hook.func = packet_rcv_spkt;
be02097c 2829
1da177e4
LT		 2830 po->prot_hook.af_packet_priv = sk;
2831
0e11c91e
AV		 2832 if (proto) {
2833 po->prot_hook.type = proto;
ce06b03e 2834 register_prot_hook(sk);
1da177e4
LT		 2835 }
2836
0fa7fa98 2837 mutex_lock(&net->packet.sklist_lock);
808f5114 2838 sk_add_node_rcu(sk, &net->packet.sklist);
0fa7fa98
PE		 2839 mutex_unlock(&net->packet.sklist_lock);
2840
2841 preempt_disable();
3680453c 2842 sock_prot_inuse_add(net, &packet_proto, 1);
0fa7fa98 2843 preempt_enable();
808f5114 2844
40d4e3df 2845 return 0;
b0138408
DB		 2846out2:
2847 sk_free(sk);
1da177e4
LT		 2848out:
2849 return err;
2850}
2851
2852/*
2853 * Pull a packet from our receive queue and hand it to the user.
2854 * If necessary we block.
2855 */
2856
2857static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2858 struct msghdr *msg, size_t len, int flags)
2859{
2860 struct sock *sk = sock->sk;
2861 struct sk_buff *skb;
2862 int copied, err;
bfd5f4a3 2863 int vnet_hdr_len = 0;
1da177e4
LT		 2864
2865 err = -EINVAL;
ed85b565 2866 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1da177e4
LT		 2867 goto out;
2868
2869#if 0
2870 /* What error should we return now? EUNATTACH? */
2871 if (pkt_sk(sk)->ifindex < 0)
2872 return -ENODEV;
2873#endif
2874
ed85b565 2875 if (flags & MSG_ERRQUEUE) {
cb820f8e
RC		 2876 err = sock_recv_errqueue(sk, msg, len,
2877 SOL_PACKET, PACKET_TX_TIMESTAMP);
ed85b565
RC		 2878 goto out;
2879 }
2880
1da177e4
LT		 2881 /*
2882 * Call the generic datagram receiver. This handles all sorts
2883 * of horrible races and re-entrancy so we can forget about it
2884 * in the protocol layers.
2885 *
2886 * Now it will return ENETDOWN, if device have just gone down,
2887 * but then it will block.
2888 */
2889
40d4e3df 2890 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1da177e4
LT		 2891
2892 /*
1ce4f28b 2893 * An error occurred so return it. Because skb_recv_datagram()
1da177e4
LT		 2894 * handles the blocking we don't see and worry about blocking
2895 * retries.
2896 */
2897
8ae55f04 2898 if (skb == NULL)
1da177e4
LT		 2899 goto out;
2900
bfd5f4a3
SS		 2901 if (pkt_sk(sk)->has_vnet_hdr) {
2902 struct virtio_net_hdr vnet_hdr = { 0 };
2903
2904 err = -EINVAL;
2905 vnet_hdr_len = sizeof(vnet_hdr);
1f18b717 2906 if (len < vnet_hdr_len)
bfd5f4a3
SS		 2907 goto out_free;
2908
1f18b717
MK		 2909 len -= vnet_hdr_len;
2910
bfd5f4a3
SS		 2911 if (skb_is_gso(skb)) {
2912 struct skb_shared_info *sinfo = skb_shinfo(skb);
2913
2914 /* This is a hint as to how much should be linear. */
2915 vnet_hdr.hdr_len = skb_headlen(skb);
2916 vnet_hdr.gso_size = sinfo->gso_size;
2917 if (sinfo->gso_type & SKB_GSO_TCPV4)
2918 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2919 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2920 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2921 else if (sinfo->gso_type & SKB_GSO_UDP)
2922 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2923 else if (sinfo->gso_type & SKB_GSO_FCOE)
2924 goto out_free;
2925 else
2926 BUG();
2927 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2928 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2929 } else
2930 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2931
2932 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2933 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
55508d60 2934 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
bfd5f4a3 2935 vnet_hdr.csum_offset = skb->csum_offset;
10a8d94a
JW		 2936 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2937 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
bfd5f4a3
SS		 2938 } /* else everything is zero */
2939
2940 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2941 vnet_hdr_len);
2942 if (err < 0)
2943 goto out_free;
2944 }
2945
f3d33426
HFS		 2946 /* You lose any data beyond the buffer you gave. If it worries
2947 * a user program they can ask the device for its MTU
2948 * anyway.
1da177e4 2949 */
1da177e4 2950 copied = skb->len;
40d4e3df
ED		 2951 if (copied > len) {
2952 copied = len;
2953 msg->msg_flags |= MSG_TRUNC;
1da177e4
LT		 2954 }
2955
2956 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2957 if (err)
2958 goto out_free;
2959
3b885787 2960 sock_recv_ts_and_drops(msg, sk, skb);
1da177e4 2961
f3d33426
HFS		 2962 if (msg->msg_name) {
2963 /* If the address length field is there to be filled
2964 * in, we fill it in now.
2965 */
2966 if (sock->type == SOCK_PACKET) {
342dfc30 2967 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
f3d33426
HFS		 2968 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2969 } else {
2970 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2971 msg->msg_namelen = sll->sll_halen +
2972 offsetof(struct sockaddr_ll, sll_addr);
2973 }
ffbc6111
HX		 2974 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2975 msg->msg_namelen);
f3d33426 2976 }
1da177e4 2977
8dc41944 2978 if (pkt_sk(sk)->auxdata) {
ffbc6111
HX		 2979 struct tpacket_auxdata aux;
2980
2981 aux.tp_status = TP_STATUS_USER;
2982 if (skb->ip_summed == CHECKSUM_PARTIAL)
2983 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2984 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2985 aux.tp_snaplen = skb->len;
2986 aux.tp_mac = 0;
bbe735e4 2987 aux.tp_net = skb_network_offset(skb);
a3bcc23e
BG		 2988 if (vlan_tx_tag_present(skb)) {
2989 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
a0cdfcf3
AW		 2990 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
2991 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
a3bcc23e
BG		 2992 } else {
2993 aux.tp_vlan_tci = 0;
a0cdfcf3 2994 aux.tp_vlan_tpid = 0;
a3bcc23e 2995 }
ffbc6111 2996 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
8dc41944
HX		 2997 }
2998
1da177e4
LT		 2999 /*
3000 * Free or return the buffer as appropriate. Again this
3001 * hides all the races and re-entrancy issues from us.
3002 */
bfd5f4a3 3003 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1da177e4
LT		 3004
3005out_free:
3006 skb_free_datagram(sk, skb);
3007out:
3008 return err;
3009}
3010
1da177e4
LT		 3011static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3012 int *uaddr_len, int peer)
3013{
3014 struct net_device *dev;
3015 struct sock *sk = sock->sk;
3016
3017 if (peer)
3018 return -EOPNOTSUPP;
3019
3020 uaddr->sa_family = AF_PACKET;
2dc85bf3 3021 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
654d1f8a
ED		 3022 rcu_read_lock();
3023 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3024 if (dev)
2dc85bf3 3025 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
654d1f8a 3026 rcu_read_unlock();
1da177e4
LT		 3027 *uaddr_len = sizeof(*uaddr);
3028
3029 return 0;
3030}
1da177e4
LT		 3031
3032static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3033 int *uaddr_len, int peer)
3034{
3035 struct net_device *dev;
3036 struct sock *sk = sock->sk;
3037 struct packet_sock *po = pkt_sk(sk);
13cfa97b 3038 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1da177e4
LT		 3039
3040 if (peer)
3041 return -EOPNOTSUPP;
3042
3043 sll->sll_family = AF_PACKET;
3044 sll->sll_ifindex = po->ifindex;
3045 sll->sll_protocol = po->num;
67286640 3046 sll->sll_pkttype = 0;
654d1f8a
ED		 3047 rcu_read_lock();
3048 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1da177e4
LT		 3049 if (dev) {
3050 sll->sll_hatype = dev->type;
3051 sll->sll_halen = dev->addr_len;
3052 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1da177e4
LT		 3053 } else {
3054 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3055 sll->sll_halen = 0;
3056 }
654d1f8a 3057 rcu_read_unlock();
0fb375fb 3058 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1da177e4
LT		 3059
3060 return 0;
3061}
3062
2aeb0b88
WC		 3063static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3064 int what)
1da177e4
LT		 3065{
3066 switch (i->type) {
3067 case PACKET_MR_MULTICAST:
1162563f
JP		 3068 if (i->alen != dev->addr_len)
3069 return -EINVAL;
1da177e4 3070 if (what > 0)
22bedad3 3071 return dev_mc_add(dev, i->addr);
1da177e4 3072 else
22bedad3 3073 return dev_mc_del(dev, i->addr);
1da177e4
LT		 3074 break;
3075 case PACKET_MR_PROMISC:
2aeb0b88 3076 return dev_set_promiscuity(dev, what);
1da177e4 3077 case PACKET_MR_ALLMULTI:
2aeb0b88 3078 return dev_set_allmulti(dev, what);
d95ed927 3079 case PACKET_MR_UNICAST:
1162563f
JP		 3080 if (i->alen != dev->addr_len)
3081 return -EINVAL;
d95ed927 3082 if (what > 0)
a748ee24 3083 return dev_uc_add(dev, i->addr);
d95ed927 3084 else
a748ee24 3085 return dev_uc_del(dev, i->addr);
d95ed927 3086 break;
40d4e3df
ED		 3087 default:
3088 break;
1da177e4 3089 }
2aeb0b88 3090 return 0;
1da177e4
LT		 3091}
3092
3093static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3094{
40d4e3df 3095 for ( ; i; i = i->next) {
1da177e4
LT		 3096 if (i->ifindex == dev->ifindex)
3097 packet_dev_mc(dev, i, what);
3098 }
3099}
3100
0fb375fb 3101static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1da177e4
LT		 3102{
3103 struct packet_sock *po = pkt_sk(sk);
3104 struct packet_mclist *ml, *i;
3105 struct net_device *dev;
3106 int err;
3107
3108 rtnl_lock();
3109
3110 err = -ENODEV;
3b1e0a65 3111 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1da177e4
LT		 3112 if (!dev)
3113 goto done;
3114
3115 err = -EINVAL;
1162563f 3116 if (mreq->mr_alen > dev->addr_len)
1da177e4
LT		 3117 goto done;
3118
3119 err = -ENOBUFS;
8b3a7005 3120 i = kmalloc(sizeof(*i), GFP_KERNEL);
1da177e4
LT		 3121 if (i == NULL)
3122 goto done;
3123
3124 err = 0;
3125 for (ml = po->mclist; ml; ml = ml->next) {
3126 if (ml->ifindex == mreq->mr_ifindex &&
3127 ml->type == mreq->mr_type &&
3128 ml->alen == mreq->mr_alen &&
3129 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3130 ml->count++;
3131 /* Free the new element ... */
3132 kfree(i);
3133 goto done;
3134 }
3135 }
3136
3137 i->type = mreq->mr_type;
3138 i->ifindex = mreq->mr_ifindex;
3139 i->alen = mreq->mr_alen;
3140 memcpy(i->addr, mreq->mr_address, i->alen);
3141 i->count = 1;
3142 i->next = po->mclist;
3143 po->mclist = i;
2aeb0b88
WC		 3144 err = packet_dev_mc(dev, i, 1);
3145 if (err) {
3146 po->mclist = i->next;
3147 kfree(i);
3148 }
1da177e4
LT		 3149
3150done:
3151 rtnl_unlock();
3152 return err;
3153}
3154
0fb375fb 3155static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1da177e4
LT		 3156{
3157 struct packet_mclist *ml, **mlp;
3158
3159 rtnl_lock();
3160
3161 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3162 if (ml->ifindex == mreq->mr_ifindex &&
3163 ml->type == mreq->mr_type &&
3164 ml->alen == mreq->mr_alen &&
3165 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3166 if (--ml->count == 0) {
3167 struct net_device *dev;
3168 *mlp = ml->next;
ad959e76
ED		 3169 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3170 if (dev)
1da177e4 3171 packet_dev_mc(dev, ml, -1);
1da177e4
LT		 3172 kfree(ml);
3173 }
3174 rtnl_unlock();
3175 return 0;
3176 }
3177 }
3178 rtnl_unlock();
3179 return -EADDRNOTAVAIL;
3180}
3181
3182static void packet_flush_mclist(struct sock *sk)
3183{
3184 struct packet_sock *po = pkt_sk(sk);
3185 struct packet_mclist *ml;
3186
3187 if (!po->mclist)
3188 return;
3189
3190 rtnl_lock();
3191 while ((ml = po->mclist) != NULL) {
3192 struct net_device *dev;
3193
3194 po->mclist = ml->next;
ad959e76
ED		 3195 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3196 if (dev != NULL)
1da177e4 3197 packet_dev_mc(dev, ml, -1);
1da177e4
LT		 3198 kfree(ml);
3199 }
3200 rtnl_unlock();
3201}
1da177e4
LT		 3202
3203static int
b7058842 3204packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1da177e4
LT		 3205{
3206 struct sock *sk = sock->sk;
8dc41944 3207 struct packet_sock *po = pkt_sk(sk);
1da177e4
LT		 3208 int ret;
3209
3210 if (level != SOL_PACKET)
3211 return -ENOPROTOOPT;
3212
69e3c75f 3213 switch (optname) {
1ce4f28b 3214 case PACKET_ADD_MEMBERSHIP:
1da177e4
LT		 3215 case PACKET_DROP_MEMBERSHIP:
3216 {
0fb375fb
EB		 3217 struct packet_mreq_max mreq;
3218 int len = optlen;
3219 memset(&mreq, 0, sizeof(mreq));
3220 if (len < sizeof(struct packet_mreq))
1da177e4 3221 return -EINVAL;
0fb375fb
EB		 3222 if (len > sizeof(mreq))
3223 len = sizeof(mreq);
40d4e3df 3224 if (copy_from_user(&mreq, optval, len))
1da177e4 3225 return -EFAULT;
0fb375fb
EB		 3226 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3227 return -EINVAL;
1da177e4
LT		 3228 if (optname == PACKET_ADD_MEMBERSHIP)
3229 ret = packet_mc_add(sk, &mreq);
3230 else
3231 ret = packet_mc_drop(sk, &mreq);
3232 return ret;
3233 }
a2efcfa0 3234
1da177e4 3235 case PACKET_RX_RING:
69e3c75f 3236 case PACKET_TX_RING:
1da177e4 3237 {
f6fb8f10 3238 union tpacket_req_u req_u;
3239 int len;
1da177e4 3240
f6fb8f10 3241 switch (po->tp_version) {
3242 case TPACKET_V1:
3243 case TPACKET_V2:
3244 len = sizeof(req_u.req);
3245 break;
3246 case TPACKET_V3:
3247 default:
3248 len = sizeof(req_u.req3);
3249 break;
3250 }
3251 if (optlen < len)
1da177e4 3252 return -EINVAL;
bfd5f4a3
SS		 3253 if (pkt_sk(sk)->has_vnet_hdr)
3254 return -EINVAL;
f6fb8f10 3255 if (copy_from_user(&req_u.req, optval, len))
1da177e4 3256 return -EFAULT;
f6fb8f10 3257 return packet_set_ring(sk, &req_u, 0,
3258 optname == PACKET_TX_RING);
1da177e4
LT		 3259 }
3260 case PACKET_COPY_THRESH:
3261 {
3262 int val;
3263
40d4e3df 3264 if (optlen != sizeof(val))
1da177e4 3265 return -EINVAL;
40d4e3df 3266 if (copy_from_user(&val, optval, sizeof(val)))
1da177e4
LT		 3267 return -EFAULT;
3268
3269 pkt_sk(sk)->copy_thresh = val;
3270 return 0;
3271 }
bbd6ef87
PM		 3272 case PACKET_VERSION:
3273 {
3274 int val;
3275
3276 if (optlen != sizeof(val))
3277 return -EINVAL;
69e3c75f 3278 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
bbd6ef87
PM		 3279 return -EBUSY;
3280 if (copy_from_user(&val, optval, sizeof(val)))
3281 return -EFAULT;
3282 switch (val) {
3283 case TPACKET_V1:
3284 case TPACKET_V2:
f6fb8f10 3285 case TPACKET_V3:
bbd6ef87
PM		 3286 po->tp_version = val;
3287 return 0;
3288 default:
3289 return -EINVAL;
3290 }
3291 }
8913336a
PM		 3292 case PACKET_RESERVE:
3293 {
3294 unsigned int val;
3295
3296 if (optlen != sizeof(val))
3297 return -EINVAL;
69e3c75f 3298 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
8913336a
PM		 3299 return -EBUSY;
3300 if (copy_from_user(&val, optval, sizeof(val)))
3301 return -EFAULT;
3302 po->tp_reserve = val;
3303 return 0;
3304 }
69e3c75f
JB		 3305 case PACKET_LOSS:
3306 {
3307 unsigned int val;
3308
3309 if (optlen != sizeof(val))
3310 return -EINVAL;
3311 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3312 return -EBUSY;
3313 if (copy_from_user(&val, optval, sizeof(val)))
3314 return -EFAULT;
3315 po->tp_loss = !!val;
3316 return 0;
3317 }
8dc41944
HX		 3318 case PACKET_AUXDATA:
3319 {
3320 int val;
3321
3322 if (optlen < sizeof(val))
3323 return -EINVAL;
3324 if (copy_from_user(&val, optval, sizeof(val)))
3325 return -EFAULT;
3326
3327 po->auxdata = !!val;
3328 return 0;
3329 }
80feaacb
PWJ		 3330 case PACKET_ORIGDEV:
3331 {
3332 int val;
3333
3334 if (optlen < sizeof(val))
3335 return -EINVAL;
3336 if (copy_from_user(&val, optval, sizeof(val)))
3337 return -EFAULT;
3338
3339 po->origdev = !!val;
3340 return 0;
3341 }
bfd5f4a3
SS		 3342 case PACKET_VNET_HDR:
3343 {
3344 int val;
3345
3346 if (sock->type != SOCK_RAW)
3347 return -EINVAL;
3348 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3349 return -EBUSY;
3350 if (optlen < sizeof(val))
3351 return -EINVAL;
3352 if (copy_from_user(&val, optval, sizeof(val)))
3353 return -EFAULT;
3354
3355 po->has_vnet_hdr = !!val;
3356 return 0;
3357 }
614f60fa
SM		 3358 case PACKET_TIMESTAMP:
3359 {
3360 int val;
3361
3362 if (optlen != sizeof(val))
3363 return -EINVAL;
3364 if (copy_from_user(&val, optval, sizeof(val)))
3365 return -EFAULT;
3366
3367 po->tp_tstamp = val;
3368 return 0;
3369 }
dc99f600
DM		 3370 case PACKET_FANOUT:
3371 {
3372 int val;
3373
3374 if (optlen != sizeof(val))
3375 return -EINVAL;
3376 if (copy_from_user(&val, optval, sizeof(val)))
3377 return -EFAULT;
3378
3379 return fanout_add(sk, val & 0xffff, val >> 16);
3380 }
5920cd3a
PC		 3381 case PACKET_TX_HAS_OFF:
3382 {
3383 unsigned int val;
3384
3385 if (optlen != sizeof(val))
3386 return -EINVAL;
3387 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3388 return -EBUSY;
3389 if (copy_from_user(&val, optval, sizeof(val)))
3390 return -EFAULT;
3391 po->tp_tx_has_off = !!val;
3392 return 0;
3393 }
d346a3fa
DB		 3394 case PACKET_QDISC_BYPASS:
3395 {
3396 int val;
3397
3398 if (optlen != sizeof(val))
3399 return -EINVAL;
3400 if (copy_from_user(&val, optval, sizeof(val)))
3401 return -EFAULT;
3402
3403 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3404 return 0;
3405 }
1da177e4
LT		 3406 default:
3407 return -ENOPROTOOPT;
3408 }
3409}
3410
3411static int packet_getsockopt(struct socket *sock, int level, int optname,
3412 char __user *optval, int __user *optlen)
3413{
3414 int len;
c06fff6e 3415 int val, lv = sizeof(val);
1da177e4
LT		 3416 struct sock *sk = sock->sk;
3417 struct packet_sock *po = pkt_sk(sk);
c06fff6e 3418 void *data = &val;
ee80fbf3 3419 union tpacket_stats_u st;
1da177e4
LT		 3420
3421 if (level != SOL_PACKET)
3422 return -ENOPROTOOPT;
3423
8ae55f04
KK		 3424 if (get_user(len, optlen))
3425 return -EFAULT;
1da177e4
LT		 3426
3427 if (len < 0)
3428 return -EINVAL;
1ce4f28b 3429
69e3c75f 3430 switch (optname) {
1da177e4 3431 case PACKET_STATISTICS:
1da177e4 3432 spin_lock_bh(&sk->sk_receive_queue.lock);
ee80fbf3
DB		 3433 memcpy(&st, &po->stats, sizeof(st));
3434 memset(&po->stats, 0, sizeof(po->stats));
3435 spin_unlock_bh(&sk->sk_receive_queue.lock);
3436
f6fb8f10 3437 if (po->tp_version == TPACKET_V3) {
c06fff6e 3438 lv = sizeof(struct tpacket_stats_v3);
8bcdeaff 3439 st.stats3.tp_packets += st.stats3.tp_drops;
ee80fbf3 3440 data = &st.stats3;
f6fb8f10 3441 } else {
c06fff6e 3442 lv = sizeof(struct tpacket_stats);
8bcdeaff 3443 st.stats1.tp_packets += st.stats1.tp_drops;
ee80fbf3 3444 data = &st.stats1;
f6fb8f10 3445 }
ee80fbf3 3446
8dc41944
HX		 3447 break;
3448 case PACKET_AUXDATA:
8dc41944 3449 val = po->auxdata;
80feaacb
PWJ		 3450 break;
3451 case PACKET_ORIGDEV:
80feaacb 3452 val = po->origdev;
bfd5f4a3
SS		 3453 break;
3454 case PACKET_VNET_HDR:
bfd5f4a3 3455 val = po->has_vnet_hdr;
1da177e4 3456 break;
bbd6ef87 3457 case PACKET_VERSION:
bbd6ef87 3458 val = po->tp_version;
bbd6ef87
PM		 3459 break;
3460 case PACKET_HDRLEN:
3461 if (len > sizeof(int))
3462 len = sizeof(int);
3463 if (copy_from_user(&val, optval, len))
3464 return -EFAULT;
3465 switch (val) {
3466 case TPACKET_V1:
3467 val = sizeof(struct tpacket_hdr);
3468 break;
3469 case TPACKET_V2:
3470 val = sizeof(struct tpacket2_hdr);
3471 break;
f6fb8f10 3472 case TPACKET_V3:
3473 val = sizeof(struct tpacket3_hdr);
3474 break;
bbd6ef87
PM		 3475 default:
3476 return -EINVAL;
3477 }
bbd6ef87 3478 break;
8913336a 3479 case PACKET_RESERVE:
8913336a 3480 val = po->tp_reserve;
8913336a 3481 break;
69e3c75f 3482 case PACKET_LOSS:
69e3c75f 3483 val = po->tp_loss;
69e3c75f 3484 break;
614f60fa 3485 case PACKET_TIMESTAMP:
614f60fa 3486 val = po->tp_tstamp;
614f60fa 3487 break;
dc99f600 3488 case PACKET_FANOUT:
dc99f600
DM		 3489 val = (po->fanout ?
3490 ((u32)po->fanout->id |
77f65ebd
WB		 3491 ((u32)po->fanout->type << 16) |
3492 ((u32)po->fanout->flags << 24)) :
dc99f600 3493 0);
dc99f600 3494 break;
5920cd3a
PC		 3495 case PACKET_TX_HAS_OFF:
3496 val = po->tp_tx_has_off;
3497 break;
d346a3fa
DB		 3498 case PACKET_QDISC_BYPASS:
3499 val = packet_use_direct_xmit(po);
3500 break;
1da177e4
LT		 3501 default:
3502 return -ENOPROTOOPT;
3503 }
3504
c06fff6e
ED		 3505 if (len > lv)
3506 len = lv;
8ae55f04
KK		 3507 if (put_user(len, optlen))
3508 return -EFAULT;
8dc41944
HX		 3509 if (copy_to_user(optval, data, len))
3510 return -EFAULT;
8ae55f04 3511 return 0;
1da177e4
LT		 3512}
3513
3514
351638e7
JP		 3515static int packet_notifier(struct notifier_block *this,
3516 unsigned long msg, void *ptr)
1da177e4
LT		 3517{
3518 struct sock *sk;
351638e7 3519 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
c346dca1 3520 struct net *net = dev_net(dev);
1da177e4 3521
808f5114 3522 rcu_read_lock();
b67bfe0d 3523 sk_for_each_rcu(sk, &net->packet.sklist) {
1da177e4
LT		 3524 struct packet_sock *po = pkt_sk(sk);
3525
3526 switch (msg) {
3527 case NETDEV_UNREGISTER:
1da177e4
LT		 3528 if (po->mclist)
3529 packet_dev_mclist(dev, po->mclist, -1);
a2efcfa0
DM		 3530 /* fallthrough */
3531
1da177e4
LT		 3532 case NETDEV_DOWN:
3533 if (dev->ifindex == po->ifindex) {
3534 spin_lock(&po->bind_lock);
3535 if (po->running) {
ce06b03e 3536 __unregister_prot_hook(sk, false);
1da177e4
LT		 3537 sk->sk_err = ENETDOWN;
3538 if (!sock_flag(sk, SOCK_DEAD))
3539 sk->sk_error_report(sk);
3540 }
3541 if (msg == NETDEV_UNREGISTER) {
66e56cd4 3542 packet_cached_dev_reset(po);
1da177e4 3543 po->ifindex = -1;
160ff18a
BG		 3544 if (po->prot_hook.dev)
3545 dev_put(po->prot_hook.dev);
1da177e4
LT		 3546 po->prot_hook.dev = NULL;
3547 }
3548 spin_unlock(&po->bind_lock);
3549 }
3550 break;
3551 case NETDEV_UP:
808f5114 3552 if (dev->ifindex == po->ifindex) {
3553 spin_lock(&po->bind_lock);
ce06b03e
DM		 3554 if (po->num)
3555 register_prot_hook(sk);
808f5114 3556 spin_unlock(&po->bind_lock);
1da177e4 3557 }
1da177e4
LT		 3558 break;
3559 }
3560 }
808f5114 3561 rcu_read_unlock();
1da177e4
LT		 3562 return NOTIFY_DONE;
3563}
3564
3565
3566static int packet_ioctl(struct socket *sock, unsigned int cmd,
3567 unsigned long arg)
3568{
3569 struct sock *sk = sock->sk;
3570
69e3c75f 3571 switch (cmd) {
40d4e3df
ED		 3572 case SIOCOUTQ:
3573 {
3574 int amount = sk_wmem_alloc_get(sk);
31e6d363 3575
40d4e3df
ED		 3576 return put_user(amount, (int __user *)arg);
3577 }
3578 case SIOCINQ:
3579 {
3580 struct sk_buff *skb;
3581 int amount = 0;
3582
3583 spin_lock_bh(&sk->sk_receive_queue.lock);
3584 skb = skb_peek(&sk->sk_receive_queue);
3585 if (skb)
3586 amount = skb->len;
3587 spin_unlock_bh(&sk->sk_receive_queue.lock);
3588 return put_user(amount, (int __user *)arg);
3589 }
3590 case SIOCGSTAMP:
3591 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3592 case SIOCGSTAMPNS:
3593 return sock_get_timestampns(sk, (struct timespec __user *)arg);
1ce4f28b 3594
1da177e4 3595#ifdef CONFIG_INET
40d4e3df
ED		 3596 case SIOCADDRT:
3597 case SIOCDELRT:
3598 case SIOCDARP:
3599 case SIOCGARP:
3600 case SIOCSARP:
3601 case SIOCGIFADDR:
3602 case SIOCSIFADDR:
3603 case SIOCGIFBRDADDR:
3604 case SIOCSIFBRDADDR:
3605 case SIOCGIFNETMASK:
3606 case SIOCSIFNETMASK:
3607 case SIOCGIFDSTADDR:
3608 case SIOCSIFDSTADDR:
3609 case SIOCSIFFLAGS:
40d4e3df 3610 return inet_dgram_ops.ioctl(sock, cmd, arg);
1da177e4
LT		 3611#endif
3612
40d4e3df
ED		 3613 default:
3614 return -ENOIOCTLCMD;
1da177e4
LT		 3615 }
3616 return 0;
3617}
3618
40d4e3df 3619static unsigned int packet_poll(struct file *file, struct socket *sock,
1da177e4
LT		 3620 poll_table *wait)
3621{
3622 struct sock *sk = sock->sk;
3623 struct packet_sock *po = pkt_sk(sk);
3624 unsigned int mask = datagram_poll(file, sock, wait);
3625
3626 spin_lock_bh(&sk->sk_receive_queue.lock);
69e3c75f 3627 if (po->rx_ring.pg_vec) {
f6fb8f10 3628 if (!packet_previous_rx_frame(po, &po->rx_ring,
3629 TP_STATUS_KERNEL))
1da177e4
LT		 3630 mask |= POLLIN | POLLRDNORM;
3631 }
3632 spin_unlock_bh(&sk->sk_receive_queue.lock);
69e3c75f
JB		 3633 spin_lock_bh(&sk->sk_write_queue.lock);
3634 if (po->tx_ring.pg_vec) {
3635 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3636 mask |= POLLOUT | POLLWRNORM;
3637 }
3638 spin_unlock_bh(&sk->sk_write_queue.lock);
1da177e4
LT		 3639 return mask;
3640}
3641
3642
3643/* Dirty? Well, I still did not learn better way to account
3644 * for user mmaps.
3645 */
3646
3647static void packet_mm_open(struct vm_area_struct *vma)
3648{
3649 struct file *file = vma->vm_file;
40d4e3df 3650 struct socket *sock = file->private_data;
1da177e4 3651 struct sock *sk = sock->sk;
1ce4f28b 3652
1da177e4
LT		 3653 if (sk)
3654 atomic_inc(&pkt_sk(sk)->mapped);
3655}
3656
3657static void packet_mm_close(struct vm_area_struct *vma)
3658{
3659 struct file *file = vma->vm_file;
40d4e3df 3660 struct socket *sock = file->private_data;
1da177e4 3661 struct sock *sk = sock->sk;
1ce4f28b 3662
1da177e4
LT		 3663 if (sk)
3664 atomic_dec(&pkt_sk(sk)->mapped);
3665}
3666
f0f37e2f 3667static const struct vm_operations_struct packet_mmap_ops = {
40d4e3df
ED		 3668 .open = packet_mm_open,
3669 .close = packet_mm_close,
1da177e4
LT		 3670};
3671
0e3125c7
NH		 3672static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3673 unsigned int len)
1da177e4
LT		 3674{
3675 int i;
3676
4ebf0ae2 3677 for (i = 0; i < len; i++) {
0e3125c7 3678 if (likely(pg_vec[i].buffer)) {
c56b4d90 3679 if (is_vmalloc_addr(pg_vec[i].buffer))
0e3125c7
NH		 3680 vfree(pg_vec[i].buffer);
3681 else
3682 free_pages((unsigned long)pg_vec[i].buffer,
3683 order);
3684 pg_vec[i].buffer = NULL;
3685 }
1da177e4
LT		 3686 }
3687 kfree(pg_vec);
3688}
3689
eea49cc9 3690static char *alloc_one_pg_vec_page(unsigned long order)
4ebf0ae2 3691{
f0d4eb29 3692 char *buffer;
0e3125c7
NH		 3693 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3694 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3695
3696 buffer = (char *) __get_free_pages(gfp_flags, order);
0e3125c7
NH		 3697 if (buffer)
3698 return buffer;
3699
f0d4eb29 3700 /* __get_free_pages failed, fall back to vmalloc */
bbce5a59 3701 buffer = vzalloc((1 << order) * PAGE_SIZE);
0e3125c7
NH		 3702 if (buffer)
3703 return buffer;
3704
f0d4eb29 3705 /* vmalloc failed, lets dig into swap here */
0e3125c7 3706 gfp_flags &= ~__GFP_NORETRY;
f0d4eb29 3707 buffer = (char *) __get_free_pages(gfp_flags, order);
0e3125c7
NH		 3708 if (buffer)
3709 return buffer;
3710
f0d4eb29 3711 /* complete and utter failure */
0e3125c7 3712 return NULL;
4ebf0ae2
DM		 3713}
3714
0e3125c7 3715static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4ebf0ae2
DM		 3716{
3717 unsigned int block_nr = req->tp_block_nr;
0e3125c7 3718 struct pgv *pg_vec;
4ebf0ae2
DM		 3719 int i;
3720
0e3125c7 3721 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4ebf0ae2
DM		 3722 if (unlikely(!pg_vec))
3723 goto out;
3724
3725 for (i = 0; i < block_nr; i++) {
c56b4d90 3726 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
0e3125c7 3727 if (unlikely(!pg_vec[i].buffer))
4ebf0ae2
DM		 3728 goto out_free_pgvec;
3729 }
3730
3731out:
3732 return pg_vec;
3733
3734out_free_pgvec:
3735 free_pg_vec(pg_vec, order, block_nr);
3736 pg_vec = NULL;
3737 goto out;
3738}
1da177e4 3739
f6fb8f10 3740static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
69e3c75f 3741 int closing, int tx_ring)
1da177e4 3742{
0e3125c7 3743 struct pgv *pg_vec = NULL;
1da177e4 3744 struct packet_sock *po = pkt_sk(sk);
0e11c91e 3745 int was_running, order = 0;
69e3c75f
JB		 3746 struct packet_ring_buffer *rb;
3747 struct sk_buff_head *rb_queue;
0e11c91e 3748 __be16 num;
f6fb8f10 3749 int err = -EINVAL;
3750 /* Added to avoid minimal code churn */
3751 struct tpacket_req *req = &req_u->req;
3752
3753 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3754 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3755 WARN(1, "Tx-ring is not supported.\n");
3756 goto out;
3757 }
1ce4f28b 3758
69e3c75f
JB		 3759 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3760 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
1da177e4 3761
69e3c75f
JB		 3762 err = -EBUSY;
3763 if (!closing) {
3764 if (atomic_read(&po->mapped))
3765 goto out;
b0138408 3766 if (packet_read_pending(rb))
69e3c75f
JB		 3767 goto out;
3768 }
1da177e4 3769
69e3c75f
JB		 3770 if (req->tp_block_nr) {
3771 /* Sanity tests and some calculations */
3772 err = -EBUSY;
3773 if (unlikely(rb->pg_vec))
3774 goto out;
1da177e4 3775
bbd6ef87
PM		 3776 switch (po->tp_version) {
3777 case TPACKET_V1:
3778 po->tp_hdrlen = TPACKET_HDRLEN;
3779 break;
3780 case TPACKET_V2:
3781 po->tp_hdrlen = TPACKET2_HDRLEN;
3782 break;
f6fb8f10 3783 case TPACKET_V3:
3784 po->tp_hdrlen = TPACKET3_HDRLEN;
3785 break;
bbd6ef87
PM		 3786 }
3787
69e3c75f 3788 err = -EINVAL;
4ebf0ae2 3789 if (unlikely((int)req->tp_block_size <= 0))
69e3c75f 3790 goto out;
4ebf0ae2 3791 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
69e3c75f 3792 goto out;
dc808110
ED		 3793 if (po->tp_version >= TPACKET_V3 &&
3794 (int)(req->tp_block_size -
3795 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3796 goto out;
8913336a 3797 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
69e3c75f
JB		 3798 po->tp_reserve))
3799 goto out;
4ebf0ae2 3800 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
69e3c75f 3801 goto out;
1da177e4 3802
69e3c75f
JB		 3803 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3804 if (unlikely(rb->frames_per_block <= 0))
3805 goto out;
3806 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3807 req->tp_frame_nr))
3808 goto out;
1da177e4
LT		 3809
3810 err = -ENOMEM;
4ebf0ae2
DM		 3811 order = get_order(req->tp_block_size);
3812 pg_vec = alloc_pg_vec(req, order);
3813 if (unlikely(!pg_vec))
1da177e4 3814 goto out;
f6fb8f10 3815 switch (po->tp_version) {
3816 case TPACKET_V3:
3817 /* Transmit path is not supported. We checked
3818 * it above but just being paranoid
3819 */
3820 if (!tx_ring)
3821 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
d7cf0c34 3822 break;
f6fb8f10 3823 default:
3824 break;
3825 }
69e3c75f
JB		 3826 }
3827 /* Done */
3828 else {
3829 err = -EINVAL;
4ebf0ae2 3830 if (unlikely(req->tp_frame_nr))
69e3c75f 3831 goto out;
1da177e4
LT		 3832 }
3833
3834 lock_sock(sk);
3835
3836 /* Detach socket from network */
3837 spin_lock(&po->bind_lock);
3838 was_running = po->running;
3839 num = po->num;
3840 if (was_running) {
1da177e4 3841 po->num = 0;
ce06b03e 3842 __unregister_prot_hook(sk, false);
1da177e4
LT		 3843 }
3844 spin_unlock(&po->bind_lock);
1ce4f28b 3845
1da177e4
LT		 3846 synchronize_net();
3847
3848 err = -EBUSY;
905db440 3849 mutex_lock(&po->pg_vec_lock);
1da177e4
LT		 3850 if (closing || atomic_read(&po->mapped) == 0) {
3851 err = 0;
69e3c75f 3852 spin_lock_bh(&rb_queue->lock);
c053fd96 3853 swap(rb->pg_vec, pg_vec);
69e3c75f
JB		 3854 rb->frame_max = (req->tp_frame_nr - 1);
3855 rb->head = 0;
3856 rb->frame_size = req->tp_frame_size;
3857 spin_unlock_bh(&rb_queue->lock);
3858
c053fd96
CG		 3859 swap(rb->pg_vec_order, order);
3860 swap(rb->pg_vec_len, req->tp_block_nr);
69e3c75f
JB		 3861
3862 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3863 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3864 tpacket_rcv : packet_rcv;
3865 skb_queue_purge(rb_queue);
1da177e4 3866 if (atomic_read(&po->mapped))
40d4e3df
ED		 3867 pr_err("packet_mmap: vma is busy: %d\n",
3868 atomic_read(&po->mapped));
1da177e4 3869 }
905db440 3870 mutex_unlock(&po->pg_vec_lock);
1da177e4
LT		 3871
3872 spin_lock(&po->bind_lock);
ce06b03e 3873 if (was_running) {
1da177e4 3874 po->num = num;
ce06b03e 3875 register_prot_hook(sk);
1da177e4
LT		 3876 }
3877 spin_unlock(&po->bind_lock);
f6fb8f10 3878 if (closing && (po->tp_version > TPACKET_V2)) {
3879 /* Because we don't support block-based V3 on tx-ring */
3880 if (!tx_ring)
3881 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3882 }
1da177e4
LT		 3883 release_sock(sk);
3884
1da177e4
LT		 3885 if (pg_vec)
3886 free_pg_vec(pg_vec, order, req->tp_block_nr);
3887out:
3888 return err;
3889}
3890
69e3c75f
JB		 3891static int packet_mmap(struct file *file, struct socket *sock,
3892 struct vm_area_struct *vma)
1da177e4
LT		 3893{
3894 struct sock *sk = sock->sk;
3895 struct packet_sock *po = pkt_sk(sk);
69e3c75f
JB		 3896 unsigned long size, expected_size;
3897 struct packet_ring_buffer *rb;
1da177e4
LT		 3898 unsigned long start;
3899 int err = -EINVAL;
3900 int i;
3901
3902 if (vma->vm_pgoff)
3903 return -EINVAL;
3904
905db440 3905 mutex_lock(&po->pg_vec_lock);
69e3c75f
JB		 3906
3907 expected_size = 0;
3908 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3909 if (rb->pg_vec) {
3910 expected_size += rb->pg_vec_len
3911 * rb->pg_vec_pages
3912 * PAGE_SIZE;
3913 }
3914 }
3915
3916 if (expected_size == 0)
1da177e4 3917 goto out;
69e3c75f
JB		 3918
3919 size = vma->vm_end - vma->vm_start;
3920 if (size != expected_size)
1da177e4
LT		 3921 goto out;
3922
1da177e4 3923 start = vma->vm_start;
69e3c75f
JB		 3924 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3925 if (rb->pg_vec == NULL)
3926 continue;
3927
3928 for (i = 0; i < rb->pg_vec_len; i++) {
0e3125c7
NH		 3929 struct page *page;
3930 void *kaddr = rb->pg_vec[i].buffer;
69e3c75f
JB		 3931 int pg_num;
3932
c56b4d90
CG		 3933 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3934 page = pgv_to_page(kaddr);
69e3c75f
JB		 3935 err = vm_insert_page(vma, start, page);
3936 if (unlikely(err))
3937 goto out;
3938 start += PAGE_SIZE;
0e3125c7 3939 kaddr += PAGE_SIZE;
69e3c75f 3940 }
4ebf0ae2 3941 }
1da177e4 3942 }
69e3c75f 3943
4ebf0ae2 3944 atomic_inc(&po->mapped);
1da177e4
LT		 3945 vma->vm_ops = &packet_mmap_ops;
3946 err = 0;
3947
3948out:
905db440 3949 mutex_unlock(&po->pg_vec_lock);
1da177e4
LT		 3950 return err;
3951}
1da177e4 3952
90ddc4f0 3953static const struct proto_ops packet_ops_spkt = {
1da177e4
LT		 3954 .family = PF_PACKET,
3955 .owner = THIS_MODULE,
3956 .release = packet_release,
3957 .bind = packet_bind_spkt,
3958 .connect = sock_no_connect,
3959 .socketpair = sock_no_socketpair,
3960 .accept = sock_no_accept,
3961 .getname = packet_getname_spkt,
3962 .poll = datagram_poll,
3963 .ioctl = packet_ioctl,
3964 .listen = sock_no_listen,
3965 .shutdown = sock_no_shutdown,
3966 .setsockopt = sock_no_setsockopt,
3967 .getsockopt = sock_no_getsockopt,
3968 .sendmsg = packet_sendmsg_spkt,
3969 .recvmsg = packet_recvmsg,
3970 .mmap = sock_no_mmap,
3971 .sendpage = sock_no_sendpage,
3972};
1da177e4 3973
90ddc4f0 3974static const struct proto_ops packet_ops = {
1da177e4
LT		 3975 .family = PF_PACKET,
3976 .owner = THIS_MODULE,
3977 .release = packet_release,
3978 .bind = packet_bind,
3979 .connect = sock_no_connect,
3980 .socketpair = sock_no_socketpair,
3981 .accept = sock_no_accept,
1ce4f28b 3982 .getname = packet_getname,
1da177e4
LT		 3983 .poll = packet_poll,
3984 .ioctl = packet_ioctl,
3985 .listen = sock_no_listen,
3986 .shutdown = sock_no_shutdown,
3987 .setsockopt = packet_setsockopt,
3988 .getsockopt = packet_getsockopt,
3989 .sendmsg = packet_sendmsg,
3990 .recvmsg = packet_recvmsg,
3991 .mmap = packet_mmap,
3992 .sendpage = sock_no_sendpage,
3993};
3994
ec1b4cf7 3995static const struct net_proto_family packet_family_ops = {
1da177e4
LT		 3996 .family = PF_PACKET,
3997 .create = packet_create,
3998 .owner = THIS_MODULE,
3999};
4000
4001static struct notifier_block packet_netdev_notifier = {
40d4e3df 4002 .notifier_call = packet_notifier,
1da177e4
LT		 4003};
4004
4005#ifdef CONFIG_PROC_FS
1da177e4
LT		 4006
4007static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
808f5114 4008 __acquires(RCU)
1da177e4 4009{
e372c414 4010 struct net *net = seq_file_net(seq);
808f5114 4011
4012 rcu_read_lock();
4013 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
1da177e4
LT		 4014}
4015
4016static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4017{
1bf40954 4018 struct net *net = seq_file_net(seq);
808f5114 4019 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
1da177e4
LT		 4020}
4021
4022static void packet_seq_stop(struct seq_file *seq, void *v)
808f5114 4023 __releases(RCU)
1da177e4 4024{
808f5114 4025 rcu_read_unlock();
1da177e4
LT		 4026}
4027
1ce4f28b 4028static int packet_seq_show(struct seq_file *seq, void *v)
1da177e4
LT		 4029{
4030 if (v == SEQ_START_TOKEN)
4031 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4032 else {
b7ceabd9 4033 struct sock *s = sk_entry(v);
1da177e4
LT		 4034 const struct packet_sock *po = pkt_sk(s);
4035
4036 seq_printf(seq,
71338aa7 4037 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
1da177e4
LT		 4038 s,
4039 atomic_read(&s->sk_refcnt),
4040 s->sk_type,
4041 ntohs(po->num),
4042 po->ifindex,
4043 po->running,
4044 atomic_read(&s->sk_rmem_alloc),
a7cb5a49 4045 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
40d4e3df 4046 sock_i_ino(s));
1da177e4
LT		 4047 }
4048
4049 return 0;
4050}
4051
56b3d975 4052static const struct seq_operations packet_seq_ops = {
1da177e4
LT		 4053 .start = packet_seq_start,
4054 .next = packet_seq_next,
4055 .stop = packet_seq_stop,
4056 .show = packet_seq_show,
4057};
4058
4059static int packet_seq_open(struct inode *inode, struct file *file)
4060{
e372c414
DL		 4061 return seq_open_net(inode, file, &packet_seq_ops,
4062 sizeof(struct seq_net_private));
1da177e4
LT		 4063}
4064
da7071d7 4065static const struct file_operations packet_seq_fops = {
1da177e4
LT		 4066 .owner = THIS_MODULE,
4067 .open = packet_seq_open,
4068 .read = seq_read,
4069 .llseek = seq_lseek,
e372c414 4070 .release = seq_release_net,
1da177e4
LT		 4071};
4072
4073#endif
4074
2c8c1e72 4075static int __net_init packet_net_init(struct net *net)
d12d01d6 4076{
0fa7fa98 4077 mutex_init(&net->packet.sklist_lock);
2aaef4e4 4078 INIT_HLIST_HEAD(&net->packet.sklist);
d12d01d6 4079
d4beaa66 4080 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
d12d01d6
DL		 4081 return -ENOMEM;
4082
4083 return 0;
4084}
4085
2c8c1e72 4086static void __net_exit packet_net_exit(struct net *net)
d12d01d6 4087{
ece31ffd 4088 remove_proc_entry("packet", net->proc_net);
d12d01d6
DL		 4089}
4090
4091static struct pernet_operations packet_net_ops = {
4092 .init = packet_net_init,
4093 .exit = packet_net_exit,
4094};
4095
4096
1da177e4
LT		 4097static void __exit packet_exit(void)
4098{
1da177e4 4099 unregister_netdevice_notifier(&packet_netdev_notifier);
d12d01d6 4100 unregister_pernet_subsys(&packet_net_ops);
1da177e4
LT		 4101 sock_unregister(PF_PACKET);
4102 proto_unregister(&packet_proto);
4103}
4104
4105static int __init packet_init(void)
4106{
4107 int rc = proto_register(&packet_proto, 0);
4108
4109 if (rc != 0)
4110 goto out;
4111
4112 sock_register(&packet_family_ops);
d12d01d6 4113 register_pernet_subsys(&packet_net_ops);
1da177e4 4114 register_netdevice_notifier(&packet_netdev_notifier);
1da177e4
LT		 4115out:
4116 return rc;
4117}
4118
4119module_init(packet_init);
4120module_exit(packet_exit);
4121MODULE_LICENSE("GPL");
4122MODULE_ALIAS_NETPROTO(PF_PACKET);
ubuntu-zesty-kernel mirror