]> 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
kill zerocopy_sg_from_iovec()
[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 }
6ce8e9ce 1679 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1a35ca80
ED		 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
9c707762
WB		 2098static bool ll_header_truncated(const struct net_device *dev, int len)
2099{
2100 /* net device doesn't like empty head */
2101 if (unlikely(len <= dev->hard_header_len)) {
2102 net_warn_ratelimited("%s: packet size is too short (%d < %d)\n",
2103 current->comm, len, dev->hard_header_len);
2104 return true;
2105 }
2106
2107 return false;
2108}
2109
40d4e3df
ED		 2110static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2111 void *frame, struct net_device *dev, int size_max,
ae641949 2112 __be16 proto, unsigned char *addr, int hlen)
69e3c75f 2113{
184f489e 2114 union tpacket_uhdr ph;
09effa67 2115 int to_write, offset, len, tp_len, nr_frags, len_max;
69e3c75f
JB		 2116 struct socket *sock = po->sk.sk_socket;
2117 struct page *page;
2118 void *data;
2119 int err;
2120
2121 ph.raw = frame;
2122
2123 skb->protocol = proto;
2124 skb->dev = dev;
2125 skb->priority = po->sk.sk_priority;
2d37a186 2126 skb->mark = po->sk.sk_mark;
2e31396f 2127 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
69e3c75f
JB		 2128 skb_shinfo(skb)->destructor_arg = ph.raw;
2129
2130 switch (po->tp_version) {
2131 case TPACKET_V2:
2132 tp_len = ph.h2->tp_len;
2133 break;
2134 default:
2135 tp_len = ph.h1->tp_len;
2136 break;
2137 }
09effa67
DM		 2138 if (unlikely(tp_len > size_max)) {
2139 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2140 return -EMSGSIZE;
2141 }
69e3c75f 2142
ae641949 2143 skb_reserve(skb, hlen);
69e3c75f 2144 skb_reset_network_header(skb);
c1aad275 2145
d346a3fa
DB		 2146 if (!packet_use_direct_xmit(po))
2147 skb_probe_transport_header(skb, 0);
2148 if (unlikely(po->tp_tx_has_off)) {
5920cd3a
PC		 2149 int off_min, off_max, off;
2150 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2151 off_max = po->tx_ring.frame_size - tp_len;
2152 if (sock->type == SOCK_DGRAM) {
2153 switch (po->tp_version) {
2154 case TPACKET_V2:
2155 off = ph.h2->tp_net;
2156 break;
2157 default:
2158 off = ph.h1->tp_net;
2159 break;
2160 }
2161 } else {
2162 switch (po->tp_version) {
2163 case TPACKET_V2:
2164 off = ph.h2->tp_mac;
2165 break;
2166 default:
2167 off = ph.h1->tp_mac;
2168 break;
2169 }
2170 }
2171 if (unlikely((off < off_min) || (off_max < off)))
2172 return -EINVAL;
2173 data = ph.raw + off;
2174 } else {
2175 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2176 }
69e3c75f
JB		 2177 to_write = tp_len;
2178
2179 if (sock->type == SOCK_DGRAM) {
2180 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2181 NULL, tp_len);
2182 if (unlikely(err < 0))
2183 return -EINVAL;
40d4e3df 2184 } else if (dev->hard_header_len) {
9c707762 2185 if (ll_header_truncated(dev, tp_len))
69e3c75f 2186 return -EINVAL;
69e3c75f
JB		 2187
2188 skb_push(skb, dev->hard_header_len);
2189 err = skb_store_bits(skb, 0, data,
2190 dev->hard_header_len);
2191 if (unlikely(err))
2192 return err;
2193
2194 data += dev->hard_header_len;
2195 to_write -= dev->hard_header_len;
2196 }
2197
69e3c75f
JB		 2198 offset = offset_in_page(data);
2199 len_max = PAGE_SIZE - offset;
2200 len = ((to_write > len_max) ? len_max : to_write);
2201
2202 skb->data_len = to_write;
2203 skb->len += to_write;
2204 skb->truesize += to_write;
2205 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2206
2207 while (likely(to_write)) {
2208 nr_frags = skb_shinfo(skb)->nr_frags;
2209
2210 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
40d4e3df
ED		 2211 pr_err("Packet exceed the number of skb frags(%lu)\n",
2212 MAX_SKB_FRAGS);
69e3c75f
JB		 2213 return -EFAULT;
2214 }
2215
0af55bb5
CG		 2216 page = pgv_to_page(data);
2217 data += len;
69e3c75f
JB		 2218 flush_dcache_page(page);
2219 get_page(page);
0af55bb5 2220 skb_fill_page_desc(skb, nr_frags, page, offset, len);
69e3c75f
JB		 2221 to_write -= len;
2222 offset = 0;
2223 len_max = PAGE_SIZE;
2224 len = ((to_write > len_max) ? len_max : to_write);
2225 }
2226
2227 return tp_len;
2228}
2229
2230static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2231{
69e3c75f
JB		 2232 struct sk_buff *skb;
2233 struct net_device *dev;
2234 __be16 proto;
09effa67 2235 int err, reserve = 0;
40d4e3df 2236 void *ph;
342dfc30 2237 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
87a2fd28 2238 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
69e3c75f
JB		 2239 int tp_len, size_max;
2240 unsigned char *addr;
2241 int len_sum = 0;
9e67030a 2242 int status = TP_STATUS_AVAILABLE;
ae641949 2243 int hlen, tlen;
69e3c75f 2244
69e3c75f
JB		 2245 mutex_lock(&po->pg_vec_lock);
2246
66e56cd4 2247 if (likely(saddr == NULL)) {
e40526cb 2248 dev = packet_cached_dev_get(po);
69e3c75f
JB		 2249 proto = po->num;
2250 addr = NULL;
2251 } else {
2252 err = -EINVAL;
2253 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2254 goto out;
2255 if (msg->msg_namelen < (saddr->sll_halen
2256 + offsetof(struct sockaddr_ll,
2257 sll_addr)))
2258 goto out;
69e3c75f
JB		 2259 proto = saddr->sll_protocol;
2260 addr = saddr->sll_addr;
827d9780 2261 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
69e3c75f
JB		 2262 }
2263
69e3c75f
JB		 2264 err = -ENXIO;
2265 if (unlikely(dev == NULL))
2266 goto out;
69e3c75f
JB		 2267 err = -ENETDOWN;
2268 if (unlikely(!(dev->flags & IFF_UP)))
2269 goto out_put;
2270
52f1454f 2271 reserve = dev->hard_header_len + VLAN_HLEN;
69e3c75f 2272 size_max = po->tx_ring.frame_size
b5dd884e 2273 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
69e3c75f 2274
09effa67
DM		 2275 if (size_max > dev->mtu + reserve)
2276 size_max = dev->mtu + reserve;
2277
69e3c75f
JB		 2278 do {
2279 ph = packet_current_frame(po, &po->tx_ring,
87a2fd28 2280 TP_STATUS_SEND_REQUEST);
69e3c75f 2281 if (unlikely(ph == NULL)) {
87a2fd28
DB		 2282 if (need_wait && need_resched())
2283 schedule();
69e3c75f
JB		 2284 continue;
2285 }
2286
2287 status = TP_STATUS_SEND_REQUEST;
ae641949
HX		 2288 hlen = LL_RESERVED_SPACE(dev);
2289 tlen = dev->needed_tailroom;
69e3c75f 2290 skb = sock_alloc_send_skb(&po->sk,
ae641949 2291 hlen + tlen + sizeof(struct sockaddr_ll),
69e3c75f
JB		 2292 0, &err);
2293
2294 if (unlikely(skb == NULL))
2295 goto out_status;
2296
2297 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
52f1454f
DB		 2298 addr, hlen);
2299 if (tp_len > dev->mtu + dev->hard_header_len) {
2300 struct ethhdr *ehdr;
2301 /* Earlier code assumed this would be a VLAN pkt,
2302 * double-check this now that we have the actual
2303 * packet in hand.
2304 */
69e3c75f 2305
52f1454f
DB		 2306 skb_reset_mac_header(skb);
2307 ehdr = eth_hdr(skb);
2308 if (ehdr->h_proto != htons(ETH_P_8021Q))
2309 tp_len = -EMSGSIZE;
2310 }
69e3c75f
JB		 2311 if (unlikely(tp_len < 0)) {
2312 if (po->tp_loss) {
2313 __packet_set_status(po, ph,
2314 TP_STATUS_AVAILABLE);
2315 packet_increment_head(&po->tx_ring);
2316 kfree_skb(skb);
2317 continue;
2318 } else {
2319 status = TP_STATUS_WRONG_FORMAT;
2320 err = tp_len;
2321 goto out_status;
2322 }
2323 }
2324
0fd5d57b
DB		 2325 packet_pick_tx_queue(dev, skb);
2326
69e3c75f
JB		 2327 skb->destructor = tpacket_destruct_skb;
2328 __packet_set_status(po, ph, TP_STATUS_SENDING);
b0138408 2329 packet_inc_pending(&po->tx_ring);
69e3c75f
JB		 2330
2331 status = TP_STATUS_SEND_REQUEST;
d346a3fa 2332 err = po->xmit(skb);
eb70df13
JP		 2333 if (unlikely(err > 0)) {
2334 err = net_xmit_errno(err);
2335 if (err && __packet_get_status(po, ph) ==
2336 TP_STATUS_AVAILABLE) {
2337 /* skb was destructed already */
2338 skb = NULL;
2339 goto out_status;
2340 }
2341 /*
2342 * skb was dropped but not destructed yet;
2343 * let's treat it like congestion or err < 0
2344 */
2345 err = 0;
2346 }
69e3c75f
JB		 2347 packet_increment_head(&po->tx_ring);
2348 len_sum += tp_len;
b0138408
DB		 2349 } while (likely((ph != NULL) ||
2350 /* Note: packet_read_pending() might be slow if we have
2351 * to call it as it's per_cpu variable, but in fast-path
2352 * we already short-circuit the loop with the first
2353 * condition, and luckily don't have to go that path
2354 * anyway.
2355 */
2356 (need_wait && packet_read_pending(&po->tx_ring))));
69e3c75f
JB		 2357
2358 err = len_sum;
2359 goto out_put;
2360
69e3c75f
JB		 2361out_status:
2362 __packet_set_status(po, ph, status);
2363 kfree_skb(skb);
2364out_put:
e40526cb 2365 dev_put(dev);
69e3c75f
JB		 2366out:
2367 mutex_unlock(&po->pg_vec_lock);
2368 return err;
2369}
69e3c75f 2370
eea49cc9
OJ		 2371static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2372 size_t reserve, size_t len,
2373 size_t linear, int noblock,
2374 int *err)
bfd5f4a3
SS		 2375{
2376 struct sk_buff *skb;
2377
2378 /* Under a page? Don't bother with paged skb. */
2379 if (prepad + len < PAGE_SIZE || !linear)
2380 linear = len;
2381
2382 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
28d64271 2383 err, 0);
bfd5f4a3
SS		 2384 if (!skb)
2385 return NULL;
2386
2387 skb_reserve(skb, reserve);
2388 skb_put(skb, linear);
2389 skb->data_len = len - linear;
2390 skb->len += len - linear;
2391
2392 return skb;
2393}
2394
d346a3fa 2395static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
1da177e4
LT		 2396{
2397 struct sock *sk = sock->sk;
342dfc30 2398 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
1da177e4
LT		 2399 struct sk_buff *skb;
2400 struct net_device *dev;
0e11c91e 2401 __be16 proto;
1da177e4 2402 unsigned char *addr;
827d9780 2403 int err, reserve = 0;
bfd5f4a3
SS		 2404 struct virtio_net_hdr vnet_hdr = { 0 };
2405 int offset = 0;
2406 int vnet_hdr_len;
2407 struct packet_sock *po = pkt_sk(sk);
2408 unsigned short gso_type = 0;
ae641949 2409 int hlen, tlen;
3bdc0eba 2410 int extra_len = 0;
1da177e4
LT		 2411
2412 /*
1ce4f28b 2413 * Get and verify the address.
1da177e4 2414 */
1ce4f28b 2415
66e56cd4 2416 if (likely(saddr == NULL)) {
e40526cb 2417 dev = packet_cached_dev_get(po);
1da177e4
LT		 2418 proto = po->num;
2419 addr = NULL;
2420 } else {
2421 err = -EINVAL;
2422 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2423 goto out;
0fb375fb
EB		 2424 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2425 goto out;
1da177e4
LT		 2426 proto = saddr->sll_protocol;
2427 addr = saddr->sll_addr;
827d9780 2428 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
1da177e4
LT		 2429 }
2430
1da177e4 2431 err = -ENXIO;
e40526cb 2432 if (unlikely(dev == NULL))
1da177e4 2433 goto out_unlock;
d5e76b0a 2434 err = -ENETDOWN;
e40526cb 2435 if (unlikely(!(dev->flags & IFF_UP)))
d5e76b0a
DM		 2436 goto out_unlock;
2437
e40526cb
DB		 2438 if (sock->type == SOCK_RAW)
2439 reserve = dev->hard_header_len;
bfd5f4a3
SS		 2440 if (po->has_vnet_hdr) {
2441 vnet_hdr_len = sizeof(vnet_hdr);
2442
2443 err = -EINVAL;
2444 if (len < vnet_hdr_len)
2445 goto out_unlock;
2446
2447 len -= vnet_hdr_len;
2448
6ce8e9ce 2449 err = memcpy_from_msg((void *)&vnet_hdr, msg, vnet_hdr_len);
bfd5f4a3
SS		 2450 if (err < 0)
2451 goto out_unlock;
2452
2453 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2454 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2455 vnet_hdr.hdr_len))
2456 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2457 vnet_hdr.csum_offset + 2;
2458
2459 err = -EINVAL;
2460 if (vnet_hdr.hdr_len > len)
2461 goto out_unlock;
2462
2463 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2464 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2465 case VIRTIO_NET_HDR_GSO_TCPV4:
2466 gso_type = SKB_GSO_TCPV4;
2467 break;
2468 case VIRTIO_NET_HDR_GSO_TCPV6:
2469 gso_type = SKB_GSO_TCPV6;
2470 break;
2471 case VIRTIO_NET_HDR_GSO_UDP:
2472 gso_type = SKB_GSO_UDP;
2473 break;
2474 default:
2475 goto out_unlock;
2476 }
2477
2478 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2479 gso_type |= SKB_GSO_TCP_ECN;
2480
2481 if (vnet_hdr.gso_size == 0)
2482 goto out_unlock;
2483
2484 }
2485 }
2486
3bdc0eba
BG		 2487 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2488 if (!netif_supports_nofcs(dev)) {
2489 err = -EPROTONOSUPPORT;
2490 goto out_unlock;
2491 }
2492 extra_len = 4; /* We're doing our own CRC */
2493 }
2494
1da177e4 2495 err = -EMSGSIZE;
3bdc0eba 2496 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
1da177e4
LT		 2497 goto out_unlock;
2498
bfd5f4a3 2499 err = -ENOBUFS;
ae641949
HX		 2500 hlen = LL_RESERVED_SPACE(dev);
2501 tlen = dev->needed_tailroom;
2502 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
bfd5f4a3 2503 msg->msg_flags & MSG_DONTWAIT, &err);
40d4e3df 2504 if (skb == NULL)
1da177e4
LT		 2505 goto out_unlock;
2506
bfd5f4a3 2507 skb_set_network_header(skb, reserve);
1da177e4 2508
0c4e8581 2509 err = -EINVAL;
9c707762
WB		 2510 if (sock->type == SOCK_DGRAM) {
2511 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2512 if (unlikely(offset) < 0)
2513 goto out_free;
2514 } else {
2515 if (ll_header_truncated(dev, len))
2516 goto out_free;
2517 }
1da177e4
LT		 2518
2519 /* Returns -EFAULT on error */
bfd5f4a3 2520 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1da177e4
LT		 2521 if (err)
2522 goto out_free;
bf84a010
DB		 2523
2524 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1da177e4 2525
3bdc0eba 2526 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
09effa67
DM		 2527 /* Earlier code assumed this would be a VLAN pkt,
2528 * double-check this now that we have the actual
2529 * packet in hand.
2530 */
2531 struct ethhdr *ehdr;
2532 skb_reset_mac_header(skb);
2533 ehdr = eth_hdr(skb);
2534 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2535 err = -EMSGSIZE;
2536 goto out_free;
2537 }
57f89bfa
BG		 2538 }
2539
09effa67
DM		 2540 skb->protocol = proto;
2541 skb->dev = dev;
1da177e4 2542 skb->priority = sk->sk_priority;
2d37a186 2543 skb->mark = sk->sk_mark;
0fd5d57b
DB		 2544
2545 packet_pick_tx_queue(dev, skb);
1da177e4 2546
bfd5f4a3
SS		 2547 if (po->has_vnet_hdr) {
2548 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2549 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2550 vnet_hdr.csum_offset)) {
2551 err = -EINVAL;
2552 goto out_free;
2553 }
2554 }
2555
2556 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2557 skb_shinfo(skb)->gso_type = gso_type;
2558
2559 /* Header must be checked, and gso_segs computed. */
2560 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2561 skb_shinfo(skb)->gso_segs = 0;
2562
2563 len += vnet_hdr_len;
2564 }
2565
d346a3fa
DB		 2566 if (!packet_use_direct_xmit(po))
2567 skb_probe_transport_header(skb, reserve);
3bdc0eba
BG		 2568 if (unlikely(extra_len == 4))
2569 skb->no_fcs = 1;
2570
d346a3fa 2571 err = po->xmit(skb);
1da177e4
LT		 2572 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2573 goto out_unlock;
2574
e40526cb 2575 dev_put(dev);
1da177e4 2576
40d4e3df 2577 return len;
1da177e4
LT		 2578
2579out_free:
2580 kfree_skb(skb);
2581out_unlock:
e40526cb 2582 if (dev)
1da177e4
LT		 2583 dev_put(dev);
2584out:
2585 return err;
2586}
2587
69e3c75f
JB		 2588static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2589 struct msghdr *msg, size_t len)
2590{
69e3c75f
JB		 2591 struct sock *sk = sock->sk;
2592 struct packet_sock *po = pkt_sk(sk);
d346a3fa 2593
69e3c75f
JB		 2594 if (po->tx_ring.pg_vec)
2595 return tpacket_snd(po, msg);
2596 else
69e3c75f
JB		 2597 return packet_snd(sock, msg, len);
2598}
2599
1da177e4
LT		 2600/*
2601 * Close a PACKET socket. This is fairly simple. We immediately go
2602 * to 'closed' state and remove our protocol entry in the device list.
2603 */
2604
2605static int packet_release(struct socket *sock)
2606{
2607 struct sock *sk = sock->sk;
2608 struct packet_sock *po;
d12d01d6 2609 struct net *net;
f6fb8f10 2610 union tpacket_req_u req_u;
1da177e4
LT		 2611
2612 if (!sk)
2613 return 0;
2614
3b1e0a65 2615 net = sock_net(sk);
1da177e4
LT		 2616 po = pkt_sk(sk);
2617
0fa7fa98 2618 mutex_lock(&net->packet.sklist_lock);
808f5114 2619 sk_del_node_init_rcu(sk);
0fa7fa98
PE		 2620 mutex_unlock(&net->packet.sklist_lock);
2621
2622 preempt_disable();
920de804 2623 sock_prot_inuse_add(net, sk->sk_prot, -1);
0fa7fa98 2624 preempt_enable();
1da177e4 2625
808f5114 2626 spin_lock(&po->bind_lock);
ce06b03e 2627 unregister_prot_hook(sk, false);
66e56cd4
DB		 2628 packet_cached_dev_reset(po);
2629
160ff18a
BG		 2630 if (po->prot_hook.dev) {
2631 dev_put(po->prot_hook.dev);
2632 po->prot_hook.dev = NULL;
2633 }
808f5114 2634 spin_unlock(&po->bind_lock);
1da177e4 2635
1da177e4 2636 packet_flush_mclist(sk);
1da177e4 2637
9665d5d6
PS		 2638 if (po->rx_ring.pg_vec) {
2639 memset(&req_u, 0, sizeof(req_u));
f6fb8f10 2640 packet_set_ring(sk, &req_u, 1, 0);
9665d5d6 2641 }
69e3c75f 2642
9665d5d6
PS		 2643 if (po->tx_ring.pg_vec) {
2644 memset(&req_u, 0, sizeof(req_u));
f6fb8f10 2645 packet_set_ring(sk, &req_u, 1, 1);
9665d5d6 2646 }
1da177e4 2647
dc99f600
DM		 2648 fanout_release(sk);
2649
808f5114 2650 synchronize_net();
1da177e4
LT		 2651 /*
2652 * Now the socket is dead. No more input will appear.
2653 */
1da177e4
LT		 2654 sock_orphan(sk);
2655 sock->sk = NULL;
2656
2657 /* Purge queues */
2658
2659 skb_queue_purge(&sk->sk_receive_queue);
b0138408 2660 packet_free_pending(po);
17ab56a2 2661 sk_refcnt_debug_release(sk);
1da177e4
LT		 2662
2663 sock_put(sk);
2664 return 0;
2665}
2666
2667/*
2668 * Attach a packet hook.
2669 */
2670
902fefb8 2671static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
1da177e4
LT		 2672{
2673 struct packet_sock *po = pkt_sk(sk);
902fefb8
DB		 2674 const struct net_device *dev_curr;
2675 __be16 proto_curr;
2676 bool need_rehook;
dc99f600 2677
aef950b4
WY		 2678 if (po->fanout) {
2679 if (dev)
2680 dev_put(dev);
2681
dc99f600 2682 return -EINVAL;
aef950b4 2683 }
1da177e4
LT		 2684
2685 lock_sock(sk);
1da177e4 2686 spin_lock(&po->bind_lock);
66e56cd4 2687
902fefb8
DB		 2688 proto_curr = po->prot_hook.type;
2689 dev_curr = po->prot_hook.dev;
2690
2691 need_rehook = proto_curr != proto || dev_curr != dev;
2692
2693 if (need_rehook) {
2694 unregister_prot_hook(sk, true);
1da177e4 2695
902fefb8
DB		 2696 po->num = proto;
2697 po->prot_hook.type = proto;
1da177e4 2698
902fefb8
DB		 2699 if (po->prot_hook.dev)
2700 dev_put(po->prot_hook.dev);
2701
2702 po->prot_hook.dev = dev;
2703
2704 po->ifindex = dev ? dev->ifindex : 0;
2705 packet_cached_dev_assign(po, dev);
2706 }
66e56cd4 2707
902fefb8 2708 if (proto == 0 || !need_rehook)
1da177e4
LT		 2709 goto out_unlock;
2710
be85d4ad 2711 if (!dev || (dev->flags & IFF_UP)) {
ce06b03e 2712 register_prot_hook(sk);
be85d4ad
UT		 2713 } else {
2714 sk->sk_err = ENETDOWN;
2715 if (!sock_flag(sk, SOCK_DEAD))
2716 sk->sk_error_report(sk);
1da177e4
LT		 2717 }
2718
2719out_unlock:
2720 spin_unlock(&po->bind_lock);
2721 release_sock(sk);
2722 return 0;
2723}
2724
2725/*
2726 * Bind a packet socket to a device
2727 */
2728
40d4e3df
ED		 2729static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2730 int addr_len)
1da177e4 2731{
40d4e3df 2732 struct sock *sk = sock->sk;
1da177e4
LT		 2733 char name[15];
2734 struct net_device *dev;
2735 int err = -ENODEV;
1ce4f28b 2736
1da177e4
LT		 2737 /*
2738 * Check legality
2739 */
1ce4f28b 2740
8ae55f04 2741 if (addr_len != sizeof(struct sockaddr))
1da177e4 2742 return -EINVAL;
40d4e3df 2743 strlcpy(name, uaddr->sa_data, sizeof(name));
1da177e4 2744
3b1e0a65 2745 dev = dev_get_by_name(sock_net(sk), name);
160ff18a 2746 if (dev)
1da177e4 2747 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1da177e4
LT		 2748 return err;
2749}
1da177e4
LT		 2750
2751static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2752{
40d4e3df
ED		 2753 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2754 struct sock *sk = sock->sk;
1da177e4
LT		 2755 struct net_device *dev = NULL;
2756 int err;
2757
2758
2759 /*
2760 * Check legality
2761 */
1ce4f28b 2762
1da177e4
LT		 2763 if (addr_len < sizeof(struct sockaddr_ll))
2764 return -EINVAL;
2765 if (sll->sll_family != AF_PACKET)
2766 return -EINVAL;
2767
2768 if (sll->sll_ifindex) {
2769 err = -ENODEV;
3b1e0a65 2770 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1da177e4
LT		 2771 if (dev == NULL)
2772 goto out;
2773 }
2774 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1da177e4
LT		 2775
2776out:
2777 return err;
2778}
2779
2780static struct proto packet_proto = {
2781 .name = "PACKET",
2782 .owner = THIS_MODULE,
2783 .obj_size = sizeof(struct packet_sock),
2784};
2785
2786/*
1ce4f28b 2787 * Create a packet of type SOCK_PACKET.
1da177e4
LT		 2788 */
2789
3f378b68
EP		 2790static int packet_create(struct net *net, struct socket *sock, int protocol,
2791 int kern)
1da177e4
LT		 2792{
2793 struct sock *sk;
2794 struct packet_sock *po;
0e11c91e 2795 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1da177e4
LT		 2796 int err;
2797
df008c91 2798 if (!ns_capable(net->user_ns, CAP_NET_RAW))
1da177e4 2799 return -EPERM;
be02097c
DM		 2800 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2801 sock->type != SOCK_PACKET)
1da177e4
LT		 2802 return -ESOCKTNOSUPPORT;
2803
2804 sock->state = SS_UNCONNECTED;
2805
2806 err = -ENOBUFS;
6257ff21 2807 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1da177e4
LT		 2808 if (sk == NULL)
2809 goto out;
2810
2811 sock->ops = &packet_ops;
1da177e4
LT		 2812 if (sock->type == SOCK_PACKET)
2813 sock->ops = &packet_ops_spkt;
be02097c 2814
1da177e4
LT		 2815 sock_init_data(sock, sk);
2816
2817 po = pkt_sk(sk);
2818 sk->sk_family = PF_PACKET;
0e11c91e 2819 po->num = proto;
d346a3fa 2820 po->xmit = dev_queue_xmit;
66e56cd4 2821
b0138408
DB		 2822 err = packet_alloc_pending(po);
2823 if (err)
2824 goto out2;
2825
66e56cd4 2826 packet_cached_dev_reset(po);
1da177e4
LT		 2827
2828 sk->sk_destruct = packet_sock_destruct;
17ab56a2 2829 sk_refcnt_debug_inc(sk);
1da177e4
LT		 2830
2831 /*
2832 * Attach a protocol block
2833 */
2834
2835 spin_lock_init(&po->bind_lock);
905db440 2836 mutex_init(&po->pg_vec_lock);
1da177e4 2837 po->prot_hook.func = packet_rcv;
be02097c 2838
1da177e4
LT		 2839 if (sock->type == SOCK_PACKET)
2840 po->prot_hook.func = packet_rcv_spkt;
be02097c 2841
1da177e4
LT		 2842 po->prot_hook.af_packet_priv = sk;
2843
0e11c91e
AV		 2844 if (proto) {
2845 po->prot_hook.type = proto;
ce06b03e 2846 register_prot_hook(sk);
1da177e4
LT		 2847 }
2848
0fa7fa98 2849 mutex_lock(&net->packet.sklist_lock);
808f5114 2850 sk_add_node_rcu(sk, &net->packet.sklist);
0fa7fa98
PE		 2851 mutex_unlock(&net->packet.sklist_lock);
2852
2853 preempt_disable();
3680453c 2854 sock_prot_inuse_add(net, &packet_proto, 1);
0fa7fa98 2855 preempt_enable();
808f5114 2856
40d4e3df 2857 return 0;
b0138408
DB		 2858out2:
2859 sk_free(sk);
1da177e4
LT		 2860out:
2861 return err;
2862}
2863
2864/*
2865 * Pull a packet from our receive queue and hand it to the user.
2866 * If necessary we block.
2867 */
2868
2869static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2870 struct msghdr *msg, size_t len, int flags)
2871{
2872 struct sock *sk = sock->sk;
2873 struct sk_buff *skb;
2874 int copied, err;
bfd5f4a3 2875 int vnet_hdr_len = 0;
1da177e4
LT		 2876
2877 err = -EINVAL;
ed85b565 2878 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1da177e4
LT		 2879 goto out;
2880
2881#if 0
2882 /* What error should we return now? EUNATTACH? */
2883 if (pkt_sk(sk)->ifindex < 0)
2884 return -ENODEV;
2885#endif
2886
ed85b565 2887 if (flags & MSG_ERRQUEUE) {
cb820f8e
RC		 2888 err = sock_recv_errqueue(sk, msg, len,
2889 SOL_PACKET, PACKET_TX_TIMESTAMP);
ed85b565
RC		 2890 goto out;
2891 }
2892
1da177e4
LT		 2893 /*
2894 * Call the generic datagram receiver. This handles all sorts
2895 * of horrible races and re-entrancy so we can forget about it
2896 * in the protocol layers.
2897 *
2898 * Now it will return ENETDOWN, if device have just gone down,
2899 * but then it will block.
2900 */
2901
40d4e3df 2902 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1da177e4
LT		 2903
2904 /*
1ce4f28b 2905 * An error occurred so return it. Because skb_recv_datagram()
1da177e4
LT		 2906 * handles the blocking we don't see and worry about blocking
2907 * retries.
2908 */
2909
8ae55f04 2910 if (skb == NULL)
1da177e4
LT		 2911 goto out;
2912
bfd5f4a3
SS		 2913 if (pkt_sk(sk)->has_vnet_hdr) {
2914 struct virtio_net_hdr vnet_hdr = { 0 };
2915
2916 err = -EINVAL;
2917 vnet_hdr_len = sizeof(vnet_hdr);
1f18b717 2918 if (len < vnet_hdr_len)
bfd5f4a3
SS		 2919 goto out_free;
2920
1f18b717
MK		 2921 len -= vnet_hdr_len;
2922
bfd5f4a3
SS		 2923 if (skb_is_gso(skb)) {
2924 struct skb_shared_info *sinfo = skb_shinfo(skb);
2925
2926 /* This is a hint as to how much should be linear. */
2927 vnet_hdr.hdr_len = skb_headlen(skb);
2928 vnet_hdr.gso_size = sinfo->gso_size;
2929 if (sinfo->gso_type & SKB_GSO_TCPV4)
2930 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2931 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2932 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2933 else if (sinfo->gso_type & SKB_GSO_UDP)
2934 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2935 else if (sinfo->gso_type & SKB_GSO_FCOE)
2936 goto out_free;
2937 else
2938 BUG();
2939 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2940 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2941 } else
2942 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2943
2944 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2945 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
55508d60 2946 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
bfd5f4a3 2947 vnet_hdr.csum_offset = skb->csum_offset;
10a8d94a
JW		 2948 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2949 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
bfd5f4a3
SS		 2950 } /* else everything is zero */
2951
7eab8d9e 2952 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
bfd5f4a3
SS		 2953 if (err < 0)
2954 goto out_free;
2955 }
2956
f3d33426
HFS		 2957 /* You lose any data beyond the buffer you gave. If it worries
2958 * a user program they can ask the device for its MTU
2959 * anyway.
1da177e4 2960 */
1da177e4 2961 copied = skb->len;
40d4e3df
ED		 2962 if (copied > len) {
2963 copied = len;
2964 msg->msg_flags |= MSG_TRUNC;
1da177e4
LT		 2965 }
2966
51f3d02b 2967 err = skb_copy_datagram_msg(skb, 0, msg, copied);
1da177e4
LT		 2968 if (err)
2969 goto out_free;
2970
3b885787 2971 sock_recv_ts_and_drops(msg, sk, skb);
1da177e4 2972
f3d33426
HFS		 2973 if (msg->msg_name) {
2974 /* If the address length field is there to be filled
2975 * in, we fill it in now.
2976 */
2977 if (sock->type == SOCK_PACKET) {
342dfc30 2978 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
f3d33426
HFS		 2979 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2980 } else {
2981 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2982 msg->msg_namelen = sll->sll_halen +
2983 offsetof(struct sockaddr_ll, sll_addr);
2984 }
ffbc6111
HX		 2985 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2986 msg->msg_namelen);
f3d33426 2987 }
1da177e4 2988
8dc41944 2989 if (pkt_sk(sk)->auxdata) {
ffbc6111
HX		 2990 struct tpacket_auxdata aux;
2991
2992 aux.tp_status = TP_STATUS_USER;
2993 if (skb->ip_summed == CHECKSUM_PARTIAL)
2994 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2995 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2996 aux.tp_snaplen = skb->len;
2997 aux.tp_mac = 0;
bbe735e4 2998 aux.tp_net = skb_network_offset(skb);
a3bcc23e
BG		 2999 if (vlan_tx_tag_present(skb)) {
3000 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
a0cdfcf3
AW		 3001 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3002 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
a3bcc23e
BG		 3003 } else {
3004 aux.tp_vlan_tci = 0;
a0cdfcf3 3005 aux.tp_vlan_tpid = 0;
a3bcc23e 3006 }
ffbc6111 3007 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
8dc41944
HX		 3008 }
3009
1da177e4
LT		 3010 /*
3011 * Free or return the buffer as appropriate. Again this
3012 * hides all the races and re-entrancy issues from us.
3013 */
bfd5f4a3 3014 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1da177e4
LT		 3015
3016out_free:
3017 skb_free_datagram(sk, skb);
3018out:
3019 return err;
3020}
3021
1da177e4
LT		 3022static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3023 int *uaddr_len, int peer)
3024{
3025 struct net_device *dev;
3026 struct sock *sk = sock->sk;
3027
3028 if (peer)
3029 return -EOPNOTSUPP;
3030
3031 uaddr->sa_family = AF_PACKET;
2dc85bf3 3032 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
654d1f8a
ED		 3033 rcu_read_lock();
3034 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3035 if (dev)
2dc85bf3 3036 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
654d1f8a 3037 rcu_read_unlock();
1da177e4
LT		 3038 *uaddr_len = sizeof(*uaddr);
3039
3040 return 0;
3041}
1da177e4
LT		 3042
3043static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3044 int *uaddr_len, int peer)
3045{
3046 struct net_device *dev;
3047 struct sock *sk = sock->sk;
3048 struct packet_sock *po = pkt_sk(sk);
13cfa97b 3049 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1da177e4
LT		 3050
3051 if (peer)
3052 return -EOPNOTSUPP;
3053
3054 sll->sll_family = AF_PACKET;
3055 sll->sll_ifindex = po->ifindex;
3056 sll->sll_protocol = po->num;
67286640 3057 sll->sll_pkttype = 0;
654d1f8a
ED		 3058 rcu_read_lock();
3059 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1da177e4
LT		 3060 if (dev) {
3061 sll->sll_hatype = dev->type;
3062 sll->sll_halen = dev->addr_len;
3063 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1da177e4
LT		 3064 } else {
3065 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3066 sll->sll_halen = 0;
3067 }
654d1f8a 3068 rcu_read_unlock();
0fb375fb 3069 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1da177e4
LT		 3070
3071 return 0;
3072}
3073
2aeb0b88
WC		 3074static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3075 int what)
1da177e4
LT		 3076{
3077 switch (i->type) {
3078 case PACKET_MR_MULTICAST:
1162563f
JP		 3079 if (i->alen != dev->addr_len)
3080 return -EINVAL;
1da177e4 3081 if (what > 0)
22bedad3 3082 return dev_mc_add(dev, i->addr);
1da177e4 3083 else
22bedad3 3084 return dev_mc_del(dev, i->addr);
1da177e4
LT		 3085 break;
3086 case PACKET_MR_PROMISC:
2aeb0b88 3087 return dev_set_promiscuity(dev, what);
1da177e4 3088 case PACKET_MR_ALLMULTI:
2aeb0b88 3089 return dev_set_allmulti(dev, what);
d95ed927 3090 case PACKET_MR_UNICAST:
1162563f
JP		 3091 if (i->alen != dev->addr_len)
3092 return -EINVAL;
d95ed927 3093 if (what > 0)
a748ee24 3094 return dev_uc_add(dev, i->addr);
d95ed927 3095 else
a748ee24 3096 return dev_uc_del(dev, i->addr);
d95ed927 3097 break;
40d4e3df
ED		 3098 default:
3099 break;
1da177e4 3100 }
2aeb0b88 3101 return 0;
1da177e4
LT		 3102}
3103
3104static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
3105{
40d4e3df 3106 for ( ; i; i = i->next) {
1da177e4
LT		 3107 if (i->ifindex == dev->ifindex)
3108 packet_dev_mc(dev, i, what);
3109 }
3110}
3111
0fb375fb 3112static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1da177e4
LT		 3113{
3114 struct packet_sock *po = pkt_sk(sk);
3115 struct packet_mclist *ml, *i;
3116 struct net_device *dev;
3117 int err;
3118
3119 rtnl_lock();
3120
3121 err = -ENODEV;
3b1e0a65 3122 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1da177e4
LT		 3123 if (!dev)
3124 goto done;
3125
3126 err = -EINVAL;
1162563f 3127 if (mreq->mr_alen > dev->addr_len)
1da177e4
LT		 3128 goto done;
3129
3130 err = -ENOBUFS;
8b3a7005 3131 i = kmalloc(sizeof(*i), GFP_KERNEL);
1da177e4
LT		 3132 if (i == NULL)
3133 goto done;
3134
3135 err = 0;
3136 for (ml = po->mclist; ml; ml = ml->next) {
3137 if (ml->ifindex == mreq->mr_ifindex &&
3138 ml->type == mreq->mr_type &&
3139 ml->alen == mreq->mr_alen &&
3140 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3141 ml->count++;
3142 /* Free the new element ... */
3143 kfree(i);
3144 goto done;
3145 }
3146 }
3147
3148 i->type = mreq->mr_type;
3149 i->ifindex = mreq->mr_ifindex;
3150 i->alen = mreq->mr_alen;
3151 memcpy(i->addr, mreq->mr_address, i->alen);
3152 i->count = 1;
3153 i->next = po->mclist;
3154 po->mclist = i;
2aeb0b88
WC		 3155 err = packet_dev_mc(dev, i, 1);
3156 if (err) {
3157 po->mclist = i->next;
3158 kfree(i);
3159 }
1da177e4
LT		 3160
3161done:
3162 rtnl_unlock();
3163 return err;
3164}
3165
0fb375fb 3166static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1da177e4
LT		 3167{
3168 struct packet_mclist *ml, **mlp;
3169
3170 rtnl_lock();
3171
3172 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3173 if (ml->ifindex == mreq->mr_ifindex &&
3174 ml->type == mreq->mr_type &&
3175 ml->alen == mreq->mr_alen &&
3176 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3177 if (--ml->count == 0) {
3178 struct net_device *dev;
3179 *mlp = ml->next;
ad959e76
ED		 3180 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3181 if (dev)
1da177e4 3182 packet_dev_mc(dev, ml, -1);
1da177e4
LT		 3183 kfree(ml);
3184 }
3185 rtnl_unlock();
3186 return 0;
3187 }
3188 }
3189 rtnl_unlock();
3190 return -EADDRNOTAVAIL;
3191}
3192
3193static void packet_flush_mclist(struct sock *sk)
3194{
3195 struct packet_sock *po = pkt_sk(sk);
3196 struct packet_mclist *ml;
3197
3198 if (!po->mclist)
3199 return;
3200
3201 rtnl_lock();
3202 while ((ml = po->mclist) != NULL) {
3203 struct net_device *dev;
3204
3205 po->mclist = ml->next;
ad959e76
ED		 3206 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3207 if (dev != NULL)
1da177e4 3208 packet_dev_mc(dev, ml, -1);
1da177e4
LT		 3209 kfree(ml);
3210 }
3211 rtnl_unlock();
3212}
1da177e4
LT		 3213
3214static int
b7058842 3215packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1da177e4
LT		 3216{
3217 struct sock *sk = sock->sk;
8dc41944 3218 struct packet_sock *po = pkt_sk(sk);
1da177e4
LT		 3219 int ret;
3220
3221 if (level != SOL_PACKET)
3222 return -ENOPROTOOPT;
3223
69e3c75f 3224 switch (optname) {
1ce4f28b 3225 case PACKET_ADD_MEMBERSHIP:
1da177e4
LT		 3226 case PACKET_DROP_MEMBERSHIP:
3227 {
0fb375fb
EB		 3228 struct packet_mreq_max mreq;
3229 int len = optlen;
3230 memset(&mreq, 0, sizeof(mreq));
3231 if (len < sizeof(struct packet_mreq))
1da177e4 3232 return -EINVAL;
0fb375fb
EB		 3233 if (len > sizeof(mreq))
3234 len = sizeof(mreq);
40d4e3df 3235 if (copy_from_user(&mreq, optval, len))
1da177e4 3236 return -EFAULT;
0fb375fb
EB		 3237 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3238 return -EINVAL;
1da177e4
LT		 3239 if (optname == PACKET_ADD_MEMBERSHIP)
3240 ret = packet_mc_add(sk, &mreq);
3241 else
3242 ret = packet_mc_drop(sk, &mreq);
3243 return ret;
3244 }
a2efcfa0 3245
1da177e4 3246 case PACKET_RX_RING:
69e3c75f 3247 case PACKET_TX_RING:
1da177e4 3248 {
f6fb8f10 3249 union tpacket_req_u req_u;
3250 int len;
1da177e4 3251
f6fb8f10 3252 switch (po->tp_version) {
3253 case TPACKET_V1:
3254 case TPACKET_V2:
3255 len = sizeof(req_u.req);
3256 break;
3257 case TPACKET_V3:
3258 default:
3259 len = sizeof(req_u.req3);
3260 break;
3261 }
3262 if (optlen < len)
1da177e4 3263 return -EINVAL;
bfd5f4a3
SS		 3264 if (pkt_sk(sk)->has_vnet_hdr)
3265 return -EINVAL;
f6fb8f10 3266 if (copy_from_user(&req_u.req, optval, len))
1da177e4 3267 return -EFAULT;
f6fb8f10 3268 return packet_set_ring(sk, &req_u, 0,
3269 optname == PACKET_TX_RING);
1da177e4
LT		 3270 }
3271 case PACKET_COPY_THRESH:
3272 {
3273 int val;
3274
40d4e3df 3275 if (optlen != sizeof(val))
1da177e4 3276 return -EINVAL;
40d4e3df 3277 if (copy_from_user(&val, optval, sizeof(val)))
1da177e4
LT		 3278 return -EFAULT;
3279
3280 pkt_sk(sk)->copy_thresh = val;
3281 return 0;
3282 }
bbd6ef87
PM		 3283 case PACKET_VERSION:
3284 {
3285 int val;
3286
3287 if (optlen != sizeof(val))
3288 return -EINVAL;
69e3c75f 3289 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
bbd6ef87
PM		 3290 return -EBUSY;
3291 if (copy_from_user(&val, optval, sizeof(val)))
3292 return -EFAULT;
3293 switch (val) {
3294 case TPACKET_V1:
3295 case TPACKET_V2:
f6fb8f10 3296 case TPACKET_V3:
bbd6ef87
PM		 3297 po->tp_version = val;
3298 return 0;
3299 default:
3300 return -EINVAL;
3301 }
3302 }
8913336a
PM		 3303 case PACKET_RESERVE:
3304 {
3305 unsigned int val;
3306
3307 if (optlen != sizeof(val))
3308 return -EINVAL;
69e3c75f 3309 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
8913336a
PM		 3310 return -EBUSY;
3311 if (copy_from_user(&val, optval, sizeof(val)))
3312 return -EFAULT;
3313 po->tp_reserve = val;
3314 return 0;
3315 }
69e3c75f
JB		 3316 case PACKET_LOSS:
3317 {
3318 unsigned int val;
3319
3320 if (optlen != sizeof(val))
3321 return -EINVAL;
3322 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3323 return -EBUSY;
3324 if (copy_from_user(&val, optval, sizeof(val)))
3325 return -EFAULT;
3326 po->tp_loss = !!val;
3327 return 0;
3328 }
8dc41944
HX		 3329 case PACKET_AUXDATA:
3330 {
3331 int val;
3332
3333 if (optlen < sizeof(val))
3334 return -EINVAL;
3335 if (copy_from_user(&val, optval, sizeof(val)))
3336 return -EFAULT;
3337
3338 po->auxdata = !!val;
3339 return 0;
3340 }
80feaacb
PWJ		 3341 case PACKET_ORIGDEV:
3342 {
3343 int val;
3344
3345 if (optlen < sizeof(val))
3346 return -EINVAL;
3347 if (copy_from_user(&val, optval, sizeof(val)))
3348 return -EFAULT;
3349
3350 po->origdev = !!val;
3351 return 0;
3352 }
bfd5f4a3
SS		 3353 case PACKET_VNET_HDR:
3354 {
3355 int val;
3356
3357 if (sock->type != SOCK_RAW)
3358 return -EINVAL;
3359 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3360 return -EBUSY;
3361 if (optlen < sizeof(val))
3362 return -EINVAL;
3363 if (copy_from_user(&val, optval, sizeof(val)))
3364 return -EFAULT;
3365
3366 po->has_vnet_hdr = !!val;
3367 return 0;
3368 }
614f60fa
SM		 3369 case PACKET_TIMESTAMP:
3370 {
3371 int val;
3372
3373 if (optlen != sizeof(val))
3374 return -EINVAL;
3375 if (copy_from_user(&val, optval, sizeof(val)))
3376 return -EFAULT;
3377
3378 po->tp_tstamp = val;
3379 return 0;
3380 }
dc99f600
DM		 3381 case PACKET_FANOUT:
3382 {
3383 int val;
3384
3385 if (optlen != sizeof(val))
3386 return -EINVAL;
3387 if (copy_from_user(&val, optval, sizeof(val)))
3388 return -EFAULT;
3389
3390 return fanout_add(sk, val & 0xffff, val >> 16);
3391 }
5920cd3a
PC		 3392 case PACKET_TX_HAS_OFF:
3393 {
3394 unsigned int val;
3395
3396 if (optlen != sizeof(val))
3397 return -EINVAL;
3398 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3399 return -EBUSY;
3400 if (copy_from_user(&val, optval, sizeof(val)))
3401 return -EFAULT;
3402 po->tp_tx_has_off = !!val;
3403 return 0;
3404 }
d346a3fa
DB		 3405 case PACKET_QDISC_BYPASS:
3406 {
3407 int val;
3408
3409 if (optlen != sizeof(val))
3410 return -EINVAL;
3411 if (copy_from_user(&val, optval, sizeof(val)))
3412 return -EFAULT;
3413
3414 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3415 return 0;
3416 }
1da177e4
LT		 3417 default:
3418 return -ENOPROTOOPT;
3419 }
3420}
3421
3422static int packet_getsockopt(struct socket *sock, int level, int optname,
3423 char __user *optval, int __user *optlen)
3424{
3425 int len;
c06fff6e 3426 int val, lv = sizeof(val);
1da177e4
LT		 3427 struct sock *sk = sock->sk;
3428 struct packet_sock *po = pkt_sk(sk);
c06fff6e 3429 void *data = &val;
ee80fbf3 3430 union tpacket_stats_u st;
1da177e4
LT		 3431
3432 if (level != SOL_PACKET)
3433 return -ENOPROTOOPT;
3434
8ae55f04
KK		 3435 if (get_user(len, optlen))
3436 return -EFAULT;
1da177e4
LT		 3437
3438 if (len < 0)
3439 return -EINVAL;
1ce4f28b 3440
69e3c75f 3441 switch (optname) {
1da177e4 3442 case PACKET_STATISTICS:
1da177e4 3443 spin_lock_bh(&sk->sk_receive_queue.lock);
ee80fbf3
DB		 3444 memcpy(&st, &po->stats, sizeof(st));
3445 memset(&po->stats, 0, sizeof(po->stats));
3446 spin_unlock_bh(&sk->sk_receive_queue.lock);
3447
f6fb8f10 3448 if (po->tp_version == TPACKET_V3) {
c06fff6e 3449 lv = sizeof(struct tpacket_stats_v3);
8bcdeaff 3450 st.stats3.tp_packets += st.stats3.tp_drops;
ee80fbf3 3451 data = &st.stats3;
f6fb8f10 3452 } else {
c06fff6e 3453 lv = sizeof(struct tpacket_stats);
8bcdeaff 3454 st.stats1.tp_packets += st.stats1.tp_drops;
ee80fbf3 3455 data = &st.stats1;
f6fb8f10 3456 }
ee80fbf3 3457
8dc41944
HX		 3458 break;
3459 case PACKET_AUXDATA:
8dc41944 3460 val = po->auxdata;
80feaacb
PWJ		 3461 break;
3462 case PACKET_ORIGDEV:
80feaacb 3463 val = po->origdev;
bfd5f4a3
SS		 3464 break;
3465 case PACKET_VNET_HDR:
bfd5f4a3 3466 val = po->has_vnet_hdr;
1da177e4 3467 break;
bbd6ef87 3468 case PACKET_VERSION:
bbd6ef87 3469 val = po->tp_version;
bbd6ef87
PM		 3470 break;
3471 case PACKET_HDRLEN:
3472 if (len > sizeof(int))
3473 len = sizeof(int);
3474 if (copy_from_user(&val, optval, len))
3475 return -EFAULT;
3476 switch (val) {
3477 case TPACKET_V1:
3478 val = sizeof(struct tpacket_hdr);
3479 break;
3480 case TPACKET_V2:
3481 val = sizeof(struct tpacket2_hdr);
3482 break;
f6fb8f10 3483 case TPACKET_V3:
3484 val = sizeof(struct tpacket3_hdr);
3485 break;
bbd6ef87
PM		 3486 default:
3487 return -EINVAL;
3488 }
bbd6ef87 3489 break;
8913336a 3490 case PACKET_RESERVE:
8913336a 3491 val = po->tp_reserve;
8913336a 3492 break;
69e3c75f 3493 case PACKET_LOSS:
69e3c75f 3494 val = po->tp_loss;
69e3c75f 3495 break;
614f60fa 3496 case PACKET_TIMESTAMP:
614f60fa 3497 val = po->tp_tstamp;
614f60fa 3498 break;
dc99f600 3499 case PACKET_FANOUT:
dc99f600
DM		 3500 val = (po->fanout ?
3501 ((u32)po->fanout->id |
77f65ebd
WB		 3502 ((u32)po->fanout->type << 16) |
3503 ((u32)po->fanout->flags << 24)) :
dc99f600 3504 0);
dc99f600 3505 break;
5920cd3a
PC		 3506 case PACKET_TX_HAS_OFF:
3507 val = po->tp_tx_has_off;
3508 break;
d346a3fa
DB		 3509 case PACKET_QDISC_BYPASS:
3510 val = packet_use_direct_xmit(po);
3511 break;
1da177e4
LT		 3512 default:
3513 return -ENOPROTOOPT;
3514 }
3515
c06fff6e
ED		 3516 if (len > lv)
3517 len = lv;
8ae55f04
KK		 3518 if (put_user(len, optlen))
3519 return -EFAULT;
8dc41944
HX		 3520 if (copy_to_user(optval, data, len))
3521 return -EFAULT;
8ae55f04 3522 return 0;
1da177e4
LT		 3523}
3524
3525
351638e7
JP		 3526static int packet_notifier(struct notifier_block *this,
3527 unsigned long msg, void *ptr)
1da177e4
LT		 3528{
3529 struct sock *sk;
351638e7 3530 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
c346dca1 3531 struct net *net = dev_net(dev);
1da177e4 3532
808f5114 3533 rcu_read_lock();
b67bfe0d 3534 sk_for_each_rcu(sk, &net->packet.sklist) {
1da177e4
LT		 3535 struct packet_sock *po = pkt_sk(sk);
3536
3537 switch (msg) {
3538 case NETDEV_UNREGISTER:
1da177e4
LT		 3539 if (po->mclist)
3540 packet_dev_mclist(dev, po->mclist, -1);
a2efcfa0
DM		 3541 /* fallthrough */
3542
1da177e4
LT		 3543 case NETDEV_DOWN:
3544 if (dev->ifindex == po->ifindex) {
3545 spin_lock(&po->bind_lock);
3546 if (po->running) {
ce06b03e 3547 __unregister_prot_hook(sk, false);
1da177e4
LT		 3548 sk->sk_err = ENETDOWN;
3549 if (!sock_flag(sk, SOCK_DEAD))
3550 sk->sk_error_report(sk);
3551 }
3552 if (msg == NETDEV_UNREGISTER) {
66e56cd4 3553 packet_cached_dev_reset(po);
1da177e4 3554 po->ifindex = -1;
160ff18a
BG		 3555 if (po->prot_hook.dev)
3556 dev_put(po->prot_hook.dev);
1da177e4
LT		 3557 po->prot_hook.dev = NULL;
3558 }
3559 spin_unlock(&po->bind_lock);
3560 }
3561 break;
3562 case NETDEV_UP:
808f5114 3563 if (dev->ifindex == po->ifindex) {
3564 spin_lock(&po->bind_lock);
ce06b03e
DM		 3565 if (po->num)
3566 register_prot_hook(sk);
808f5114 3567 spin_unlock(&po->bind_lock);
1da177e4 3568 }
1da177e4
LT		 3569 break;
3570 }
3571 }
808f5114 3572 rcu_read_unlock();
1da177e4
LT		 3573 return NOTIFY_DONE;
3574}
3575
3576
3577static int packet_ioctl(struct socket *sock, unsigned int cmd,
3578 unsigned long arg)
3579{
3580 struct sock *sk = sock->sk;
3581
69e3c75f 3582 switch (cmd) {
40d4e3df
ED		 3583 case SIOCOUTQ:
3584 {
3585 int amount = sk_wmem_alloc_get(sk);
31e6d363 3586
40d4e3df
ED		 3587 return put_user(amount, (int __user *)arg);
3588 }
3589 case SIOCINQ:
3590 {
3591 struct sk_buff *skb;
3592 int amount = 0;
3593
3594 spin_lock_bh(&sk->sk_receive_queue.lock);
3595 skb = skb_peek(&sk->sk_receive_queue);
3596 if (skb)
3597 amount = skb->len;
3598 spin_unlock_bh(&sk->sk_receive_queue.lock);
3599 return put_user(amount, (int __user *)arg);
3600 }
3601 case SIOCGSTAMP:
3602 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3603 case SIOCGSTAMPNS:
3604 return sock_get_timestampns(sk, (struct timespec __user *)arg);
1ce4f28b 3605
1da177e4 3606#ifdef CONFIG_INET
40d4e3df
ED		 3607 case SIOCADDRT:
3608 case SIOCDELRT:
3609 case SIOCDARP:
3610 case SIOCGARP:
3611 case SIOCSARP:
3612 case SIOCGIFADDR:
3613 case SIOCSIFADDR:
3614 case SIOCGIFBRDADDR:
3615 case SIOCSIFBRDADDR:
3616 case SIOCGIFNETMASK:
3617 case SIOCSIFNETMASK:
3618 case SIOCGIFDSTADDR:
3619 case SIOCSIFDSTADDR:
3620 case SIOCSIFFLAGS:
40d4e3df 3621 return inet_dgram_ops.ioctl(sock, cmd, arg);
1da177e4
LT		 3622#endif
3623
40d4e3df
ED		 3624 default:
3625 return -ENOIOCTLCMD;
1da177e4
LT		 3626 }
3627 return 0;
3628}
3629
40d4e3df 3630static unsigned int packet_poll(struct file *file, struct socket *sock,
1da177e4
LT		 3631 poll_table *wait)
3632{
3633 struct sock *sk = sock->sk;
3634 struct packet_sock *po = pkt_sk(sk);
3635 unsigned int mask = datagram_poll(file, sock, wait);
3636
3637 spin_lock_bh(&sk->sk_receive_queue.lock);
69e3c75f 3638 if (po->rx_ring.pg_vec) {
f6fb8f10 3639 if (!packet_previous_rx_frame(po, &po->rx_ring,
3640 TP_STATUS_KERNEL))
1da177e4
LT		 3641 mask |= POLLIN | POLLRDNORM;
3642 }
3643 spin_unlock_bh(&sk->sk_receive_queue.lock);
69e3c75f
JB		 3644 spin_lock_bh(&sk->sk_write_queue.lock);
3645 if (po->tx_ring.pg_vec) {
3646 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3647 mask |= POLLOUT | POLLWRNORM;
3648 }
3649 spin_unlock_bh(&sk->sk_write_queue.lock);
1da177e4
LT		 3650 return mask;
3651}
3652
3653
3654/* Dirty? Well, I still did not learn better way to account
3655 * for user mmaps.
3656 */
3657
3658static void packet_mm_open(struct vm_area_struct *vma)
3659{
3660 struct file *file = vma->vm_file;
40d4e3df 3661 struct socket *sock = file->private_data;
1da177e4 3662 struct sock *sk = sock->sk;
1ce4f28b 3663
1da177e4
LT		 3664 if (sk)
3665 atomic_inc(&pkt_sk(sk)->mapped);
3666}
3667
3668static void packet_mm_close(struct vm_area_struct *vma)
3669{
3670 struct file *file = vma->vm_file;
40d4e3df 3671 struct socket *sock = file->private_data;
1da177e4 3672 struct sock *sk = sock->sk;
1ce4f28b 3673
1da177e4
LT		 3674 if (sk)
3675 atomic_dec(&pkt_sk(sk)->mapped);
3676}
3677
f0f37e2f 3678static const struct vm_operations_struct packet_mmap_ops = {
40d4e3df
ED		 3679 .open = packet_mm_open,
3680 .close = packet_mm_close,
1da177e4
LT		 3681};
3682
0e3125c7
NH		 3683static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3684 unsigned int len)
1da177e4
LT		 3685{
3686 int i;
3687
4ebf0ae2 3688 for (i = 0; i < len; i++) {
0e3125c7 3689 if (likely(pg_vec[i].buffer)) {
c56b4d90 3690 if (is_vmalloc_addr(pg_vec[i].buffer))
0e3125c7
NH		 3691 vfree(pg_vec[i].buffer);
3692 else
3693 free_pages((unsigned long)pg_vec[i].buffer,
3694 order);
3695 pg_vec[i].buffer = NULL;
3696 }
1da177e4
LT		 3697 }
3698 kfree(pg_vec);
3699}
3700
eea49cc9 3701static char *alloc_one_pg_vec_page(unsigned long order)
4ebf0ae2 3702{
f0d4eb29 3703 char *buffer;
0e3125c7
NH		 3704 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3705 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3706
3707 buffer = (char *) __get_free_pages(gfp_flags, order);
0e3125c7
NH		 3708 if (buffer)
3709 return buffer;
3710
f0d4eb29 3711 /* __get_free_pages failed, fall back to vmalloc */
bbce5a59 3712 buffer = vzalloc((1 << order) * PAGE_SIZE);
0e3125c7
NH		 3713 if (buffer)
3714 return buffer;
3715
f0d4eb29 3716 /* vmalloc failed, lets dig into swap here */
0e3125c7 3717 gfp_flags &= ~__GFP_NORETRY;
f0d4eb29 3718 buffer = (char *) __get_free_pages(gfp_flags, order);
0e3125c7
NH		 3719 if (buffer)
3720 return buffer;
3721
f0d4eb29 3722 /* complete and utter failure */
0e3125c7 3723 return NULL;
4ebf0ae2
DM		 3724}
3725
0e3125c7 3726static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4ebf0ae2
DM		 3727{
3728 unsigned int block_nr = req->tp_block_nr;
0e3125c7 3729 struct pgv *pg_vec;
4ebf0ae2
DM		 3730 int i;
3731
0e3125c7 3732 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4ebf0ae2
DM		 3733 if (unlikely(!pg_vec))
3734 goto out;
3735
3736 for (i = 0; i < block_nr; i++) {
c56b4d90 3737 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
0e3125c7 3738 if (unlikely(!pg_vec[i].buffer))
4ebf0ae2
DM		 3739 goto out_free_pgvec;
3740 }
3741
3742out:
3743 return pg_vec;
3744
3745out_free_pgvec:
3746 free_pg_vec(pg_vec, order, block_nr);
3747 pg_vec = NULL;
3748 goto out;
3749}
1da177e4 3750
f6fb8f10 3751static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
69e3c75f 3752 int closing, int tx_ring)
1da177e4 3753{
0e3125c7 3754 struct pgv *pg_vec = NULL;
1da177e4 3755 struct packet_sock *po = pkt_sk(sk);
0e11c91e 3756 int was_running, order = 0;
69e3c75f
JB		 3757 struct packet_ring_buffer *rb;
3758 struct sk_buff_head *rb_queue;
0e11c91e 3759 __be16 num;
f6fb8f10 3760 int err = -EINVAL;
3761 /* Added to avoid minimal code churn */
3762 struct tpacket_req *req = &req_u->req;
3763
3764 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3765 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3766 WARN(1, "Tx-ring is not supported.\n");
3767 goto out;
3768 }
1ce4f28b 3769
69e3c75f
JB		 3770 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3771 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
1da177e4 3772
69e3c75f
JB		 3773 err = -EBUSY;
3774 if (!closing) {
3775 if (atomic_read(&po->mapped))
3776 goto out;
b0138408 3777 if (packet_read_pending(rb))
69e3c75f
JB		 3778 goto out;
3779 }
1da177e4 3780
69e3c75f
JB		 3781 if (req->tp_block_nr) {
3782 /* Sanity tests and some calculations */
3783 err = -EBUSY;
3784 if (unlikely(rb->pg_vec))
3785 goto out;
1da177e4 3786
bbd6ef87
PM		 3787 switch (po->tp_version) {
3788 case TPACKET_V1:
3789 po->tp_hdrlen = TPACKET_HDRLEN;
3790 break;
3791 case TPACKET_V2:
3792 po->tp_hdrlen = TPACKET2_HDRLEN;
3793 break;
f6fb8f10 3794 case TPACKET_V3:
3795 po->tp_hdrlen = TPACKET3_HDRLEN;
3796 break;
bbd6ef87
PM		 3797 }
3798
69e3c75f 3799 err = -EINVAL;
4ebf0ae2 3800 if (unlikely((int)req->tp_block_size <= 0))
69e3c75f 3801 goto out;
4ebf0ae2 3802 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
69e3c75f 3803 goto out;
dc808110
ED		 3804 if (po->tp_version >= TPACKET_V3 &&
3805 (int)(req->tp_block_size -
3806 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3807 goto out;
8913336a 3808 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
69e3c75f
JB		 3809 po->tp_reserve))
3810 goto out;
4ebf0ae2 3811 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
69e3c75f 3812 goto out;
1da177e4 3813
69e3c75f
JB		 3814 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3815 if (unlikely(rb->frames_per_block <= 0))
3816 goto out;
3817 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3818 req->tp_frame_nr))
3819 goto out;
1da177e4
LT		 3820
3821 err = -ENOMEM;
4ebf0ae2
DM		 3822 order = get_order(req->tp_block_size);
3823 pg_vec = alloc_pg_vec(req, order);
3824 if (unlikely(!pg_vec))
1da177e4 3825 goto out;
f6fb8f10 3826 switch (po->tp_version) {
3827 case TPACKET_V3:
3828 /* Transmit path is not supported. We checked
3829 * it above but just being paranoid
3830 */
3831 if (!tx_ring)
3832 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
d7cf0c34 3833 break;
f6fb8f10 3834 default:
3835 break;
3836 }
69e3c75f
JB		 3837 }
3838 /* Done */
3839 else {
3840 err = -EINVAL;
4ebf0ae2 3841 if (unlikely(req->tp_frame_nr))
69e3c75f 3842 goto out;
1da177e4
LT		 3843 }
3844
3845 lock_sock(sk);
3846
3847 /* Detach socket from network */
3848 spin_lock(&po->bind_lock);
3849 was_running = po->running;
3850 num = po->num;
3851 if (was_running) {
1da177e4 3852 po->num = 0;
ce06b03e 3853 __unregister_prot_hook(sk, false);
1da177e4
LT		 3854 }
3855 spin_unlock(&po->bind_lock);
1ce4f28b 3856
1da177e4
LT		 3857 synchronize_net();
3858
3859 err = -EBUSY;
905db440 3860 mutex_lock(&po->pg_vec_lock);
1da177e4
LT		 3861 if (closing || atomic_read(&po->mapped) == 0) {
3862 err = 0;
69e3c75f 3863 spin_lock_bh(&rb_queue->lock);
c053fd96 3864 swap(rb->pg_vec, pg_vec);
69e3c75f
JB		 3865 rb->frame_max = (req->tp_frame_nr - 1);
3866 rb->head = 0;
3867 rb->frame_size = req->tp_frame_size;
3868 spin_unlock_bh(&rb_queue->lock);
3869
c053fd96
CG		 3870 swap(rb->pg_vec_order, order);
3871 swap(rb->pg_vec_len, req->tp_block_nr);
69e3c75f
JB		 3872
3873 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3874 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3875 tpacket_rcv : packet_rcv;
3876 skb_queue_purge(rb_queue);
1da177e4 3877 if (atomic_read(&po->mapped))
40d4e3df
ED		 3878 pr_err("packet_mmap: vma is busy: %d\n",
3879 atomic_read(&po->mapped));
1da177e4 3880 }
905db440 3881 mutex_unlock(&po->pg_vec_lock);
1da177e4
LT		 3882
3883 spin_lock(&po->bind_lock);
ce06b03e 3884 if (was_running) {
1da177e4 3885 po->num = num;
ce06b03e 3886 register_prot_hook(sk);
1da177e4
LT		 3887 }
3888 spin_unlock(&po->bind_lock);
f6fb8f10 3889 if (closing && (po->tp_version > TPACKET_V2)) {
3890 /* Because we don't support block-based V3 on tx-ring */
3891 if (!tx_ring)
3892 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3893 }
1da177e4
LT		 3894 release_sock(sk);
3895
1da177e4
LT		 3896 if (pg_vec)
3897 free_pg_vec(pg_vec, order, req->tp_block_nr);
3898out:
3899 return err;
3900}
3901
69e3c75f
JB		 3902static int packet_mmap(struct file *file, struct socket *sock,
3903 struct vm_area_struct *vma)
1da177e4
LT		 3904{
3905 struct sock *sk = sock->sk;
3906 struct packet_sock *po = pkt_sk(sk);
69e3c75f
JB		 3907 unsigned long size, expected_size;
3908 struct packet_ring_buffer *rb;
1da177e4
LT		 3909 unsigned long start;
3910 int err = -EINVAL;
3911 int i;
3912
3913 if (vma->vm_pgoff)
3914 return -EINVAL;
3915
905db440 3916 mutex_lock(&po->pg_vec_lock);
69e3c75f
JB		 3917
3918 expected_size = 0;
3919 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3920 if (rb->pg_vec) {
3921 expected_size += rb->pg_vec_len
3922 * rb->pg_vec_pages
3923 * PAGE_SIZE;
3924 }
3925 }
3926
3927 if (expected_size == 0)
1da177e4 3928 goto out;
69e3c75f
JB		 3929
3930 size = vma->vm_end - vma->vm_start;
3931 if (size != expected_size)
1da177e4
LT		 3932 goto out;
3933
1da177e4 3934 start = vma->vm_start;
69e3c75f
JB		 3935 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3936 if (rb->pg_vec == NULL)
3937 continue;
3938
3939 for (i = 0; i < rb->pg_vec_len; i++) {
0e3125c7
NH		 3940 struct page *page;
3941 void *kaddr = rb->pg_vec[i].buffer;
69e3c75f
JB		 3942 int pg_num;
3943
c56b4d90
CG		 3944 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3945 page = pgv_to_page(kaddr);
69e3c75f
JB		 3946 err = vm_insert_page(vma, start, page);
3947 if (unlikely(err))
3948 goto out;
3949 start += PAGE_SIZE;
0e3125c7 3950 kaddr += PAGE_SIZE;
69e3c75f 3951 }
4ebf0ae2 3952 }
1da177e4 3953 }
69e3c75f 3954
4ebf0ae2 3955 atomic_inc(&po->mapped);
1da177e4
LT		 3956 vma->vm_ops = &packet_mmap_ops;
3957 err = 0;
3958
3959out:
905db440 3960 mutex_unlock(&po->pg_vec_lock);
1da177e4
LT		 3961 return err;
3962}
1da177e4 3963
90ddc4f0 3964static const struct proto_ops packet_ops_spkt = {
1da177e4
LT		 3965 .family = PF_PACKET,
3966 .owner = THIS_MODULE,
3967 .release = packet_release,
3968 .bind = packet_bind_spkt,
3969 .connect = sock_no_connect,
3970 .socketpair = sock_no_socketpair,
3971 .accept = sock_no_accept,
3972 .getname = packet_getname_spkt,
3973 .poll = datagram_poll,
3974 .ioctl = packet_ioctl,
3975 .listen = sock_no_listen,
3976 .shutdown = sock_no_shutdown,
3977 .setsockopt = sock_no_setsockopt,
3978 .getsockopt = sock_no_getsockopt,
3979 .sendmsg = packet_sendmsg_spkt,
3980 .recvmsg = packet_recvmsg,
3981 .mmap = sock_no_mmap,
3982 .sendpage = sock_no_sendpage,
3983};
1da177e4 3984
90ddc4f0 3985static const struct proto_ops packet_ops = {
1da177e4
LT		 3986 .family = PF_PACKET,
3987 .owner = THIS_MODULE,
3988 .release = packet_release,
3989 .bind = packet_bind,
3990 .connect = sock_no_connect,
3991 .socketpair = sock_no_socketpair,
3992 .accept = sock_no_accept,
1ce4f28b 3993 .getname = packet_getname,
1da177e4
LT		 3994 .poll = packet_poll,
3995 .ioctl = packet_ioctl,
3996 .listen = sock_no_listen,
3997 .shutdown = sock_no_shutdown,
3998 .setsockopt = packet_setsockopt,
3999 .getsockopt = packet_getsockopt,
4000 .sendmsg = packet_sendmsg,
4001 .recvmsg = packet_recvmsg,
4002 .mmap = packet_mmap,
4003 .sendpage = sock_no_sendpage,
4004};
4005
ec1b4cf7 4006static const struct net_proto_family packet_family_ops = {
1da177e4
LT		 4007 .family = PF_PACKET,
4008 .create = packet_create,
4009 .owner = THIS_MODULE,
4010};
4011
4012static struct notifier_block packet_netdev_notifier = {
40d4e3df 4013 .notifier_call = packet_notifier,
1da177e4
LT		 4014};
4015
4016#ifdef CONFIG_PROC_FS
1da177e4
LT		 4017
4018static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
808f5114 4019 __acquires(RCU)
1da177e4 4020{
e372c414 4021 struct net *net = seq_file_net(seq);
808f5114 4022
4023 rcu_read_lock();
4024 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
1da177e4
LT		 4025}
4026
4027static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4028{
1bf40954 4029 struct net *net = seq_file_net(seq);
808f5114 4030 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
1da177e4
LT		 4031}
4032
4033static void packet_seq_stop(struct seq_file *seq, void *v)
808f5114 4034 __releases(RCU)
1da177e4 4035{
808f5114 4036 rcu_read_unlock();
1da177e4
LT		 4037}
4038
1ce4f28b 4039static int packet_seq_show(struct seq_file *seq, void *v)
1da177e4
LT		 4040{
4041 if (v == SEQ_START_TOKEN)
4042 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4043 else {
b7ceabd9 4044 struct sock *s = sk_entry(v);
1da177e4
LT		 4045 const struct packet_sock *po = pkt_sk(s);
4046
4047 seq_printf(seq,
71338aa7 4048 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
1da177e4
LT		 4049 s,
4050 atomic_read(&s->sk_refcnt),
4051 s->sk_type,
4052 ntohs(po->num),
4053 po->ifindex,
4054 po->running,
4055 atomic_read(&s->sk_rmem_alloc),
a7cb5a49 4056 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
40d4e3df 4057 sock_i_ino(s));
1da177e4
LT		 4058 }
4059
4060 return 0;
4061}
4062
56b3d975 4063static const struct seq_operations packet_seq_ops = {
1da177e4
LT		 4064 .start = packet_seq_start,
4065 .next = packet_seq_next,
4066 .stop = packet_seq_stop,
4067 .show = packet_seq_show,
4068};
4069
4070static int packet_seq_open(struct inode *inode, struct file *file)
4071{
e372c414
DL		 4072 return seq_open_net(inode, file, &packet_seq_ops,
4073 sizeof(struct seq_net_private));
1da177e4
LT		 4074}
4075
da7071d7 4076static const struct file_operations packet_seq_fops = {
1da177e4
LT		 4077 .owner = THIS_MODULE,
4078 .open = packet_seq_open,
4079 .read = seq_read,
4080 .llseek = seq_lseek,
e372c414 4081 .release = seq_release_net,
1da177e4
LT		 4082};
4083
4084#endif
4085
2c8c1e72 4086static int __net_init packet_net_init(struct net *net)
d12d01d6 4087{
0fa7fa98 4088 mutex_init(&net->packet.sklist_lock);
2aaef4e4 4089 INIT_HLIST_HEAD(&net->packet.sklist);
d12d01d6 4090
d4beaa66 4091 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
d12d01d6
DL		 4092 return -ENOMEM;
4093
4094 return 0;
4095}
4096
2c8c1e72 4097static void __net_exit packet_net_exit(struct net *net)
d12d01d6 4098{
ece31ffd 4099 remove_proc_entry("packet", net->proc_net);
d12d01d6
DL		 4100}
4101
4102static struct pernet_operations packet_net_ops = {
4103 .init = packet_net_init,
4104 .exit = packet_net_exit,
4105};
4106
4107
1da177e4
LT		 4108static void __exit packet_exit(void)
4109{
1da177e4 4110 unregister_netdevice_notifier(&packet_netdev_notifier);
d12d01d6 4111 unregister_pernet_subsys(&packet_net_ops);
1da177e4
LT		 4112 sock_unregister(PF_PACKET);
4113 proto_unregister(&packet_proto);
4114}
4115
4116static int __init packet_init(void)
4117{
4118 int rc = proto_register(&packet_proto, 0);
4119
4120 if (rc != 0)
4121 goto out;
4122
4123 sock_register(&packet_family_ops);
d12d01d6 4124 register_pernet_subsys(&packet_net_ops);
1da177e4 4125 register_netdevice_notifier(&packet_netdev_notifier);
1da177e4
LT		 4126out:
4127 return rc;
4128}
4129
4130module_init(packet_init);
4131module_exit(packet_exit);
4132MODULE_LICENSE("GPL");
4133MODULE_ALIAS_NETPROTO(PF_PACKET);
ubuntu-zesty-kernel mirror