projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net: Remove LL_ALLOCATED_SPACE
[GitHub/mt8127/android_kernel_alcatel_ttab.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>
76#include <asm/system.h>
77#include <asm/uaccess.h>
78#include <asm/ioctls.h>
79#include <asm/page.h>
a1f8e7f7 80#include <asm/cacheflush.h>
1da177e4
LT		 81#include <asm/io.h>
82#include <linux/proc_fs.h>
83#include <linux/seq_file.h>
84#include <linux/poll.h>
85#include <linux/module.h>
86#include <linux/init.h>
905db440 87#include <linux/mutex.h>
05423b24 88#include <linux/if_vlan.h>
bfd5f4a3 89#include <linux/virtio_net.h>
ed85b565 90#include <linux/errqueue.h>
614f60fa 91#include <linux/net_tstamp.h>
1da177e4
LT		 92
93#ifdef CONFIG_INET
94#include <net/inet_common.h>
95#endif
96
1da177e4
LT		 97/*
98 Assumptions:
99 - if device has no dev->hard_header routine, it adds and removes ll header
100 inside itself. In this case ll header is invisible outside of device,
101 but higher levels still should reserve dev->hard_header_len.
102 Some devices are enough clever to reallocate skb, when header
103 will not fit to reserved space (tunnel), another ones are silly
104 (PPP).
105 - packet socket receives packets with pulled ll header,
106 so that SOCK_RAW should push it back.
107
108On receive:
109-----------
110
111Incoming, dev->hard_header!=NULL
b0e380b1
ACM		 112 mac_header -> ll header
113 data -> data
1da177e4
LT		 114
115Outgoing, dev->hard_header!=NULL
b0e380b1
ACM		 116 mac_header -> ll header
117 data -> ll header
1da177e4
LT		 118
119Incoming, dev->hard_header==NULL
b0e380b1
ACM		 120 mac_header -> UNKNOWN position. It is very likely, that it points to ll
121 header. PPP makes it, that is wrong, because introduce
db0c58f9 122 assymetry between rx and tx paths.
b0e380b1 123 data -> data
1da177e4
LT		 124
125Outgoing, dev->hard_header==NULL
b0e380b1
ACM		 126 mac_header -> data. ll header is still not built!
127 data -> data
1da177e4
LT		 128
129Resume
130 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
131
132
133On transmit:
134------------
135
136dev->hard_header != NULL
b0e380b1
ACM		 137 mac_header -> ll header
138 data -> ll header
1da177e4
LT		 139
140dev->hard_header == NULL (ll header is added by device, we cannot control it)
b0e380b1
ACM		 141 mac_header -> data
142 data -> data
1da177e4
LT		 143
144 We should set nh.raw on output to correct posistion,
145 packet classifier depends on it.
146 */
147
1da177e4
LT		 148/* Private packet socket structures. */
149
40d4e3df 150struct packet_mclist {
1da177e4
LT		 151 struct packet_mclist *next;
152 int ifindex;
153 int count;
154 unsigned short type;
155 unsigned short alen;
0fb375fb
EB		 156 unsigned char addr[MAX_ADDR_LEN];
157};
158/* identical to struct packet_mreq except it has
159 * a longer address field.
160 */
40d4e3df 161struct packet_mreq_max {
0fb375fb
EB		 162 int mr_ifindex;
163 unsigned short mr_type;
164 unsigned short mr_alen;
165 unsigned char mr_address[MAX_ADDR_LEN];
1da177e4 166};
a2efcfa0 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
172#define V3_ALIGNMENT (8)
173
bc59ba39 174#define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
f6fb8f10 175
176#define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179/* kbdq - kernel block descriptor queue */
bc59ba39 180struct tpacket_kbdq_core {
f6fb8f10 181 struct pgv *pkbdq;
182 unsigned int feature_req_word;
183 unsigned int hdrlen;
184 unsigned char reset_pending_on_curr_blk;
185 unsigned char delete_blk_timer;
186 unsigned short kactive_blk_num;
187 unsigned short blk_sizeof_priv;
188
189 /* last_kactive_blk_num:
190 * trick to see if user-space has caught up
191 * in order to avoid refreshing timer when every single pkt arrives.
192 */
193 unsigned short last_kactive_blk_num;
194
195 char *pkblk_start;
196 char *pkblk_end;
197 int kblk_size;
198 unsigned int knum_blocks;
199 uint64_t knxt_seq_num;
200 char *prev;
201 char *nxt_offset;
202 struct sk_buff *skb;
203
204 atomic_t blk_fill_in_prog;
205
206 /* Default is set to 8ms */
207#define DEFAULT_PRB_RETIRE_TOV (8)
208
209 unsigned short retire_blk_tov;
210 unsigned short version;
211 unsigned long tov_in_jiffies;
212
213 /* timer to retire an outstanding block */
214 struct timer_list retire_blk_timer;
215};
216
217#define PGV_FROM_VMALLOC 1
0e3125c7
NH		 218struct pgv {
219 char *buffer;
0e3125c7
NH		 220};
221
69e3c75f 222struct packet_ring_buffer {
0e3125c7 223 struct pgv *pg_vec;
69e3c75f
JB		 224 unsigned int head;
225 unsigned int frames_per_block;
226 unsigned int frame_size;
227 unsigned int frame_max;
228
229 unsigned int pg_vec_order;
230 unsigned int pg_vec_pages;
231 unsigned int pg_vec_len;
232
bc59ba39 233 struct tpacket_kbdq_core prb_bdqc;
69e3c75f
JB		 234 atomic_t pending;
235};
236
f6fb8f10 237#define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
238#define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
239#define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
240#define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
241#define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
242#define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
243#define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
244
69e3c75f
JB		 245struct packet_sock;
246static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
1da177e4 247
f6fb8f10 248static void *packet_previous_frame(struct packet_sock *po,
249 struct packet_ring_buffer *rb,
250 int status);
251static void packet_increment_head(struct packet_ring_buffer *buff);
bc59ba39 252static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
253 struct tpacket_block_desc *);
254static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
f6fb8f10 255 struct packet_sock *);
bc59ba39 256static void prb_retire_current_block(struct tpacket_kbdq_core *,
f6fb8f10 257 struct packet_sock *, unsigned int status);
bc59ba39 258static int prb_queue_frozen(struct tpacket_kbdq_core *);
259static void prb_open_block(struct tpacket_kbdq_core *,
260 struct tpacket_block_desc *);
f6fb8f10 261static void prb_retire_rx_blk_timer_expired(unsigned long);
bc59ba39 262static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
263static void prb_init_blk_timer(struct packet_sock *,
264 struct tpacket_kbdq_core *,
265 void (*func) (unsigned long));
266static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
267static void prb_clear_rxhash(struct tpacket_kbdq_core *,
268 struct tpacket3_hdr *);
269static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
270 struct tpacket3_hdr *);
1da177e4
LT		 271static void packet_flush_mclist(struct sock *sk);
272
dc99f600 273struct packet_fanout;
1da177e4
LT		 274struct packet_sock {
275 /* struct sock has to be the first member of packet_sock */
276 struct sock sk;
dc99f600 277 struct packet_fanout *fanout;
1da177e4 278 struct tpacket_stats stats;
f6fb8f10 279 union tpacket_stats_u stats_u;
69e3c75f
JB		 280 struct packet_ring_buffer rx_ring;
281 struct packet_ring_buffer tx_ring;
1da177e4 282 int copy_thresh;
1da177e4 283 spinlock_t bind_lock;
905db440 284 struct mutex pg_vec_lock;
8dc41944 285 unsigned int running:1, /* prot_hook is attached*/
80feaacb 286 auxdata:1,
bfd5f4a3
SS		 287 origdev:1,
288 has_vnet_hdr:1;
1da177e4 289 int ifindex; /* bound device */
0e11c91e 290 __be16 num;
1da177e4 291 struct packet_mclist *mclist;
1da177e4 292 atomic_t mapped;
bbd6ef87
PM		 293 enum tpacket_versions tp_version;
294 unsigned int tp_hdrlen;
8913336a 295 unsigned int tp_reserve;
69e3c75f 296 unsigned int tp_loss:1;
614f60fa 297 unsigned int tp_tstamp;
94b05952 298 struct packet_type prot_hook ____cacheline_aligned_in_smp;
1da177e4
LT		 299};
300
dc99f600
DM		 301#define PACKET_FANOUT_MAX 256
302
303struct packet_fanout {
304#ifdef CONFIG_NET_NS
305 struct net *net;
306#endif
307 unsigned int num_members;
308 u16 id;
309 u8 type;
7736d33f 310 u8 defrag;
dc99f600
DM		 311 atomic_t rr_cur;
312 struct list_head list;
313 struct sock *arr[PACKET_FANOUT_MAX];
314 spinlock_t lock;
315 atomic_t sk_ref;
316 struct packet_type prot_hook ____cacheline_aligned_in_smp;
317};
318
ffbc6111
HX		 319struct packet_skb_cb {
320 unsigned int origlen;
321 union {
322 struct sockaddr_pkt pkt;
323 struct sockaddr_ll ll;
324 } sa;
325};
326
327#define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
8dc41944 328
bc59ba39 329#define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
f6fb8f10 330#define GET_PBLOCK_DESC(x, bid) \
bc59ba39 331 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
f6fb8f10 332#define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
bc59ba39 333 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
f6fb8f10 334#define GET_NEXT_PRB_BLK_NUM(x) \
335 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
336 ((x)->kactive_blk_num+1) : 0)
337
eea49cc9 338static struct packet_sock *pkt_sk(struct sock *sk)
ce06b03e
DM		 339{
340 return (struct packet_sock *)sk;
341}
342
dc99f600
DM		 343static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
344static void __fanout_link(struct sock *sk, struct packet_sock *po);
345
ce06b03e
DM		 346/* register_prot_hook must be invoked with the po->bind_lock held,
347 * or from a context in which asynchronous accesses to the packet
348 * socket is not possible (packet_create()).
349 */
350static void register_prot_hook(struct sock *sk)
351{
352 struct packet_sock *po = pkt_sk(sk);
353 if (!po->running) {
dc99f600
DM		 354 if (po->fanout)
355 __fanout_link(sk, po);
356 else
357 dev_add_pack(&po->prot_hook);
ce06b03e
DM		 358 sock_hold(sk);
359 po->running = 1;
360 }
361}
362
363/* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
364 * held. If the sync parameter is true, we will temporarily drop
365 * the po->bind_lock and do a synchronize_net to make sure no
366 * asynchronous packet processing paths still refer to the elements
367 * of po->prot_hook. If the sync parameter is false, it is the
368 * callers responsibility to take care of this.
369 */
370static void __unregister_prot_hook(struct sock *sk, bool sync)
371{
372 struct packet_sock *po = pkt_sk(sk);
373
374 po->running = 0;
dc99f600
DM		 375 if (po->fanout)
376 __fanout_unlink(sk, po);
377 else
378 __dev_remove_pack(&po->prot_hook);
ce06b03e
DM		 379 __sock_put(sk);
380
381 if (sync) {
382 spin_unlock(&po->bind_lock);
383 synchronize_net();
384 spin_lock(&po->bind_lock);
385 }
386}
387
388static void unregister_prot_hook(struct sock *sk, bool sync)
389{
390 struct packet_sock *po = pkt_sk(sk);
391
392 if (po->running)
393 __unregister_prot_hook(sk, sync);
394}
395
f6dafa95 396static inline __pure struct page *pgv_to_page(void *addr)
0af55bb5
CG		 397{
398 if (is_vmalloc_addr(addr))
399 return vmalloc_to_page(addr);
400 return virt_to_page(addr);
401}
402
69e3c75f 403static void __packet_set_status(struct packet_sock *po, void *frame, int status)
1da177e4 404{
bbd6ef87
PM		 405 union {
406 struct tpacket_hdr *h1;
407 struct tpacket2_hdr *h2;
408 void *raw;
409 } h;
1da177e4 410
69e3c75f 411 h.raw = frame;
bbd6ef87
PM		 412 switch (po->tp_version) {
413 case TPACKET_V1:
69e3c75f 414 h.h1->tp_status = status;
0af55bb5 415 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
bbd6ef87
PM		 416 break;
417 case TPACKET_V2:
69e3c75f 418 h.h2->tp_status = status;
0af55bb5 419 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
bbd6ef87 420 break;
f6fb8f10 421 case TPACKET_V3:
69e3c75f 422 default:
f6fb8f10 423 WARN(1, "TPACKET version not supported.\n");
69e3c75f 424 BUG();
bbd6ef87 425 }
69e3c75f
JB		 426
427 smp_wmb();
bbd6ef87
PM		 428}
429
69e3c75f 430static int __packet_get_status(struct packet_sock *po, void *frame)
bbd6ef87
PM		 431{
432 union {
433 struct tpacket_hdr *h1;
434 struct tpacket2_hdr *h2;
435 void *raw;
436 } h;
437
69e3c75f
JB		 438 smp_rmb();
439
bbd6ef87
PM		 440 h.raw = frame;
441 switch (po->tp_version) {
442 case TPACKET_V1:
0af55bb5 443 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
69e3c75f 444 return h.h1->tp_status;
bbd6ef87 445 case TPACKET_V2:
0af55bb5 446 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
69e3c75f 447 return h.h2->tp_status;
f6fb8f10 448 case TPACKET_V3:
69e3c75f 449 default:
f6fb8f10 450 WARN(1, "TPACKET version not supported.\n");
69e3c75f
JB		 451 BUG();
452 return 0;
bbd6ef87 453 }
1da177e4 454}
69e3c75f
JB		 455
456static void *packet_lookup_frame(struct packet_sock *po,
457 struct packet_ring_buffer *rb,
458 unsigned int position,
459 int status)
460{
461 unsigned int pg_vec_pos, frame_offset;
462 union {
463 struct tpacket_hdr *h1;
464 struct tpacket2_hdr *h2;
465 void *raw;
466 } h;
467
468 pg_vec_pos = position / rb->frames_per_block;
469 frame_offset = position % rb->frames_per_block;
470
0e3125c7
NH		 471 h.raw = rb->pg_vec[pg_vec_pos].buffer +
472 (frame_offset * rb->frame_size);
69e3c75f
JB		 473
474 if (status != __packet_get_status(po, h.raw))
475 return NULL;
476
477 return h.raw;
478}
479
eea49cc9 480static void *packet_current_frame(struct packet_sock *po,
69e3c75f
JB		 481 struct packet_ring_buffer *rb,
482 int status)
483{
484 return packet_lookup_frame(po, rb, rb->head, status);
485}
486
bc59ba39 487static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
f6fb8f10 488{
489 del_timer_sync(&pkc->retire_blk_timer);
490}
491
492static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
493 int tx_ring,
494 struct sk_buff_head *rb_queue)
495{
bc59ba39 496 struct tpacket_kbdq_core *pkc;
f6fb8f10 497
498 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
499
500 spin_lock(&rb_queue->lock);
501 pkc->delete_blk_timer = 1;
502 spin_unlock(&rb_queue->lock);
503
504 prb_del_retire_blk_timer(pkc);
505}
506
507static void prb_init_blk_timer(struct packet_sock *po,
bc59ba39 508 struct tpacket_kbdq_core *pkc,
f6fb8f10 509 void (*func) (unsigned long))
510{
511 init_timer(&pkc->retire_blk_timer);
512 pkc->retire_blk_timer.data = (long)po;
513 pkc->retire_blk_timer.function = func;
514 pkc->retire_blk_timer.expires = jiffies;
515}
516
517static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
518{
bc59ba39 519 struct tpacket_kbdq_core *pkc;
f6fb8f10 520
521 if (tx_ring)
522 BUG();
523
524 pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
525 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
526}
527
528static int prb_calc_retire_blk_tmo(struct packet_sock *po,
529 int blk_size_in_bytes)
530{
531 struct net_device *dev;
532 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
4bc71cb9
JP		 533 struct ethtool_cmd ecmd;
534 int err;
f6fb8f10 535
4bc71cb9
JP		 536 rtnl_lock();
537 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
538 if (unlikely(!dev)) {
539 rtnl_unlock();
f6fb8f10 540 return DEFAULT_PRB_RETIRE_TOV;
4bc71cb9
JP		 541 }
542 err = __ethtool_get_settings(dev, &ecmd);
543 rtnl_unlock();
544 if (!err) {
545 switch (ecmd.speed) {
546 case SPEED_10000:
547 msec = 1;
548 div = 10000/1000;
549 break;
550 case SPEED_1000:
551 msec = 1;
552 div = 1000/1000;
553 break;
554 /*
555 * If the link speed is so slow you don't really
556 * need to worry about perf anyways
557 */
558 case SPEED_100:
559 case SPEED_10:
560 default:
561 return DEFAULT_PRB_RETIRE_TOV;
f6fb8f10 562 }
563 }
564
565 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
566
567 if (div)
568 mbits /= div;
569
570 tmo = mbits * msec;
571
572 if (div)
573 return tmo+1;
574 return tmo;
575}
576
bc59ba39 577static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
f6fb8f10 578 union tpacket_req_u *req_u)
579{
580 p1->feature_req_word = req_u->req3.tp_feature_req_word;
581}
582
583static void init_prb_bdqc(struct packet_sock *po,
584 struct packet_ring_buffer *rb,
585 struct pgv *pg_vec,
586 union tpacket_req_u *req_u, int tx_ring)
587{
bc59ba39 588 struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
589 struct tpacket_block_desc *pbd;
f6fb8f10 590
591 memset(p1, 0x0, sizeof(*p1));
592
593 p1->knxt_seq_num = 1;
594 p1->pkbdq = pg_vec;
bc59ba39 595 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
f6fb8f10 596 p1->pkblk_start = (char *)pg_vec[0].buffer;
597 p1->kblk_size = req_u->req3.tp_block_size;
598 p1->knum_blocks = req_u->req3.tp_block_nr;
599 p1->hdrlen = po->tp_hdrlen;
600 p1->version = po->tp_version;
601 p1->last_kactive_blk_num = 0;
602 po->stats_u.stats3.tp_freeze_q_cnt = 0;
603 if (req_u->req3.tp_retire_blk_tov)
604 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
605 else
606 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
607 req_u->req3.tp_block_size);
608 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
609 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
610
611 prb_init_ft_ops(p1, req_u);
612 prb_setup_retire_blk_timer(po, tx_ring);
613 prb_open_block(p1, pbd);
614}
615
616/* Do NOT update the last_blk_num first.
617 * Assumes sk_buff_head lock is held.
618 */
bc59ba39 619static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
f6fb8f10 620{
621 mod_timer(&pkc->retire_blk_timer,
622 jiffies + pkc->tov_in_jiffies);
623 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
624}
625
626/*
627 * Timer logic:
628 * 1) We refresh the timer only when we open a block.
629 * By doing this we don't waste cycles refreshing the timer
630 * on packet-by-packet basis.
631 *
632 * With a 1MB block-size, on a 1Gbps line, it will take
633 * i) ~8 ms to fill a block + ii) memcpy etc.
634 * In this cut we are not accounting for the memcpy time.
635 *
636 * So, if the user sets the 'tmo' to 10ms then the timer
637 * will never fire while the block is still getting filled
638 * (which is what we want). However, the user could choose
639 * to close a block early and that's fine.
640 *
641 * But when the timer does fire, we check whether or not to refresh it.
642 * Since the tmo granularity is in msecs, it is not too expensive
643 * to refresh the timer, lets say every '8' msecs.
644 * Either the user can set the 'tmo' or we can derive it based on
645 * a) line-speed and b) block-size.
646 * prb_calc_retire_blk_tmo() calculates the tmo.
647 *
648 */
649static void prb_retire_rx_blk_timer_expired(unsigned long data)
650{
651 struct packet_sock *po = (struct packet_sock *)data;
bc59ba39 652 struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
f6fb8f10 653 unsigned int frozen;
bc59ba39 654 struct tpacket_block_desc *pbd;
f6fb8f10 655
656 spin_lock(&po->sk.sk_receive_queue.lock);
657
658 frozen = prb_queue_frozen(pkc);
659 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
660
661 if (unlikely(pkc->delete_blk_timer))
662 goto out;
663
664 /* We only need to plug the race when the block is partially filled.
665 * tpacket_rcv:
666 * lock(); increment BLOCK_NUM_PKTS; unlock()
667 * copy_bits() is in progress ...
668 * timer fires on other cpu:
669 * we can't retire the current block because copy_bits
670 * is in progress.
671 *
672 */
673 if (BLOCK_NUM_PKTS(pbd)) {
674 while (atomic_read(&pkc->blk_fill_in_prog)) {
675 /* Waiting for skb_copy_bits to finish... */
676 cpu_relax();
677 }
678 }
679
680 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
681 if (!frozen) {
682 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
683 if (!prb_dispatch_next_block(pkc, po))
684 goto refresh_timer;
685 else
686 goto out;
687 } else {
688 /* Case 1. Queue was frozen because user-space was
689 * lagging behind.
690 */
691 if (prb_curr_blk_in_use(pkc, pbd)) {
692 /*
693 * Ok, user-space is still behind.
694 * So just refresh the timer.
695 */
696 goto refresh_timer;
697 } else {
698 /* Case 2. queue was frozen,user-space caught up,
699 * now the link went idle && the timer fired.
700 * We don't have a block to close.So we open this
701 * block and restart the timer.
702 * opening a block thaws the queue,restarts timer
703 * Thawing/timer-refresh is a side effect.
704 */
705 prb_open_block(pkc, pbd);
706 goto out;
707 }
708 }
709 }
710
711refresh_timer:
712 _prb_refresh_rx_retire_blk_timer(pkc);
713
714out:
715 spin_unlock(&po->sk.sk_receive_queue.lock);
716}
717
eea49cc9 718static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
bc59ba39 719 struct tpacket_block_desc *pbd1, __u32 status)
f6fb8f10 720{
721 /* Flush everything minus the block header */
722
723#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
724 u8 *start, *end;
725
726 start = (u8 *)pbd1;
727
728 /* Skip the block header(we know header WILL fit in 4K) */
729 start += PAGE_SIZE;
730
731 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
732 for (; start < end; start += PAGE_SIZE)
733 flush_dcache_page(pgv_to_page(start));
734
735 smp_wmb();
736#endif
737
738 /* Now update the block status. */
739
740 BLOCK_STATUS(pbd1) = status;
741
742 /* Flush the block header */
743
744#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
745 start = (u8 *)pbd1;
746 flush_dcache_page(pgv_to_page(start));
747
748 smp_wmb();
749#endif
750}
751
752/*
753 * Side effect:
754 *
755 * 1) flush the block
756 * 2) Increment active_blk_num
757 *
758 * Note:We DONT refresh the timer on purpose.
759 * Because almost always the next block will be opened.
760 */
bc59ba39 761static void prb_close_block(struct tpacket_kbdq_core *pkc1,
762 struct tpacket_block_desc *pbd1,
f6fb8f10 763 struct packet_sock *po, unsigned int stat)
764{
765 __u32 status = TP_STATUS_USER | stat;
766
767 struct tpacket3_hdr *last_pkt;
bc59ba39 768 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
f6fb8f10 769
770 if (po->stats.tp_drops)
771 status |= TP_STATUS_LOSING;
772
773 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
774 last_pkt->tp_next_offset = 0;
775
776 /* Get the ts of the last pkt */
777 if (BLOCK_NUM_PKTS(pbd1)) {
778 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
779 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
780 } else {
781 /* Ok, we tmo'd - so get the current time */
782 struct timespec ts;
783 getnstimeofday(&ts);
784 h1->ts_last_pkt.ts_sec = ts.tv_sec;
785 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
786 }
787
788 smp_wmb();
789
790 /* Flush the block */
791 prb_flush_block(pkc1, pbd1, status);
792
793 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
794}
795
eea49cc9 796static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
f6fb8f10 797{
798 pkc->reset_pending_on_curr_blk = 0;
799}
800
801/*
802 * Side effect of opening a block:
803 *
804 * 1) prb_queue is thawed.
805 * 2) retire_blk_timer is refreshed.
806 *
807 */
bc59ba39 808static void prb_open_block(struct tpacket_kbdq_core *pkc1,
809 struct tpacket_block_desc *pbd1)
f6fb8f10 810{
811 struct timespec ts;
bc59ba39 812 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
f6fb8f10 813
814 smp_rmb();
815
816 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
817
818 /* We could have just memset this but we will lose the
819 * flexibility of making the priv area sticky
820 */
821 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
822 BLOCK_NUM_PKTS(pbd1) = 0;
823 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
824 getnstimeofday(&ts);
825 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827 pkc1->pkblk_start = (char *)pbd1;
828 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
829 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
830 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
832 pbd1->version = pkc1->version;
833 pkc1->prev = pkc1->nxt_offset;
834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
835 prb_thaw_queue(pkc1);
836 _prb_refresh_rx_retire_blk_timer(pkc1);
837
838 smp_wmb();
839
840 return;
841 }
842
843 WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
844 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
845 dump_stack();
846 BUG();
847}
848
849/*
850 * Queue freeze logic:
851 * 1) Assume tp_block_nr = 8 blocks.
852 * 2) At time 't0', user opens Rx ring.
853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854 * 4) user-space is either sleeping or processing block '0'.
855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856 * it will close block-7,loop around and try to fill block '0'.
857 * call-flow:
858 * __packet_lookup_frame_in_block
859 * prb_retire_current_block()
860 * prb_dispatch_next_block()
861 * |->(BLOCK_STATUS == USER) evaluates to true
862 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
863 * 6) Now there are two cases:
864 * 6.1) Link goes idle right after the queue is frozen.
865 * But remember, the last open_block() refreshed the timer.
866 * When this timer expires,it will refresh itself so that we can
867 * re-open block-0 in near future.
868 * 6.2) Link is busy and keeps on receiving packets. This is a simple
869 * case and __packet_lookup_frame_in_block will check if block-0
870 * is free and can now be re-used.
871 */
eea49cc9 872static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
f6fb8f10 873 struct packet_sock *po)
874{
875 pkc->reset_pending_on_curr_blk = 1;
876 po->stats_u.stats3.tp_freeze_q_cnt++;
877}
878
879#define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880
881/*
882 * If the next block is free then we will dispatch it
883 * and return a good offset.
884 * Else, we will freeze the queue.
885 * So, caller must check the return value.
886 */
bc59ba39 887static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
f6fb8f10 888 struct packet_sock *po)
889{
bc59ba39 890 struct tpacket_block_desc *pbd;
f6fb8f10 891
892 smp_rmb();
893
894 /* 1. Get current block num */
895 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896
897 /* 2. If this block is currently in_use then freeze the queue */
898 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899 prb_freeze_queue(pkc, po);
900 return NULL;
901 }
902
903 /*
904 * 3.
905 * open this block and return the offset where the first packet
906 * needs to get stored.
907 */
908 prb_open_block(pkc, pbd);
909 return (void *)pkc->nxt_offset;
910}
911
bc59ba39 912static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
f6fb8f10 913 struct packet_sock *po, unsigned int status)
914{
bc59ba39 915 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
f6fb8f10 916
917 /* retire/close the current block */
918 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919 /*
920 * Plug the case where copy_bits() is in progress on
921 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922 * have space to copy the pkt in the current block and
923 * called prb_retire_current_block()
924 *
925 * We don't need to worry about the TMO case because
926 * the timer-handler already handled this case.
927 */
928 if (!(status & TP_STATUS_BLK_TMO)) {
929 while (atomic_read(&pkc->blk_fill_in_prog)) {
930 /* Waiting for skb_copy_bits to finish... */
931 cpu_relax();
932 }
933 }
934 prb_close_block(pkc, pbd, po, status);
935 return;
936 }
937
938 WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
939 dump_stack();
940 BUG();
941}
942
eea49cc9 943static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
bc59ba39 944 struct tpacket_block_desc *pbd)
f6fb8f10 945{
946 return TP_STATUS_USER & BLOCK_STATUS(pbd);
947}
948
eea49cc9 949static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
f6fb8f10 950{
951 return pkc->reset_pending_on_curr_blk;
952}
953
eea49cc9 954static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
f6fb8f10 955{
bc59ba39 956 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
f6fb8f10 957 atomic_dec(&pkc->blk_fill_in_prog);
958}
959
eea49cc9 960static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
f6fb8f10 961 struct tpacket3_hdr *ppd)
962{
963 ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
964}
965
eea49cc9 966static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
f6fb8f10 967 struct tpacket3_hdr *ppd)
968{
969 ppd->hv1.tp_rxhash = 0;
970}
971
eea49cc9 972static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
f6fb8f10 973 struct tpacket3_hdr *ppd)
974{
975 if (vlan_tx_tag_present(pkc->skb)) {
976 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
977 ppd->tp_status = TP_STATUS_VLAN_VALID;
978 } else {
979 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
980 }
981}
982
bc59ba39 983static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
f6fb8f10 984 struct tpacket3_hdr *ppd)
985{
986 prb_fill_vlan_info(pkc, ppd);
987
988 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989 prb_fill_rxhash(pkc, ppd);
990 else
991 prb_clear_rxhash(pkc, ppd);
992}
993
eea49cc9 994static void prb_fill_curr_block(char *curr,
bc59ba39 995 struct tpacket_kbdq_core *pkc,
996 struct tpacket_block_desc *pbd,
f6fb8f10 997 unsigned int len)
998{
999 struct tpacket3_hdr *ppd;
1000
1001 ppd = (struct tpacket3_hdr *)curr;
1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1003 pkc->prev = curr;
1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_NUM_PKTS(pbd) += 1;
1007 atomic_inc(&pkc->blk_fill_in_prog);
1008 prb_run_all_ft_ops(pkc, ppd);
1009}
1010
1011/* Assumes caller has the sk->rx_queue.lock */
1012static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013 struct sk_buff *skb,
1014 int status,
1015 unsigned int len
1016 )
1017{
bc59ba39 1018 struct tpacket_kbdq_core *pkc;
1019 struct tpacket_block_desc *pbd;
f6fb8f10 1020 char *curr, *end;
1021
1022 pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
1025 /* Queue is frozen when user space is lagging behind */
1026 if (prb_queue_frozen(pkc)) {
1027 /*
1028 * Check if that last block which caused the queue to freeze,
1029 * is still in_use by user-space.
1030 */
1031 if (prb_curr_blk_in_use(pkc, pbd)) {
1032 /* Can't record this packet */
1033 return NULL;
1034 } else {
1035 /*
1036 * Ok, the block was released by user-space.
1037 * Now let's open that block.
1038 * opening a block also thaws the queue.
1039 * Thawing is a side effect.
1040 */
1041 prb_open_block(pkc, pbd);
1042 }
1043 }
1044
1045 smp_mb();
1046 curr = pkc->nxt_offset;
1047 pkc->skb = skb;
1048 end = (char *) ((char *)pbd + pkc->kblk_size);
1049
1050 /* first try the current block */
1051 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052 prb_fill_curr_block(curr, pkc, pbd, len);
1053 return (void *)curr;
1054 }
1055
1056 /* Ok, close the current block */
1057 prb_retire_current_block(pkc, po, 0);
1058
1059 /* Now, try to dispatch the next block */
1060 curr = (char *)prb_dispatch_next_block(pkc, po);
1061 if (curr) {
1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 prb_fill_curr_block(curr, pkc, pbd, len);
1064 return (void *)curr;
1065 }
1066
1067 /*
1068 * No free blocks are available.user_space hasn't caught up yet.
1069 * Queue was just frozen and now this packet will get dropped.
1070 */
1071 return NULL;
1072}
1073
eea49cc9 1074static void *packet_current_rx_frame(struct packet_sock *po,
f6fb8f10 1075 struct sk_buff *skb,
1076 int status, unsigned int len)
1077{
1078 char *curr = NULL;
1079 switch (po->tp_version) {
1080 case TPACKET_V1:
1081 case TPACKET_V2:
1082 curr = packet_lookup_frame(po, &po->rx_ring,
1083 po->rx_ring.head, status);
1084 return curr;
1085 case TPACKET_V3:
1086 return __packet_lookup_frame_in_block(po, skb, status, len);
1087 default:
1088 WARN(1, "TPACKET version not supported\n");
1089 BUG();
1090 return 0;
1091 }
1092}
1093
eea49cc9 1094static void *prb_lookup_block(struct packet_sock *po,
f6fb8f10 1095 struct packet_ring_buffer *rb,
1096 unsigned int previous,
1097 int status)
1098{
bc59ba39 1099 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1100 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
f6fb8f10 1101
1102 if (status != BLOCK_STATUS(pbd))
1103 return NULL;
1104 return pbd;
1105}
1106
eea49cc9 1107static int prb_previous_blk_num(struct packet_ring_buffer *rb)
f6fb8f10 1108{
1109 unsigned int prev;
1110 if (rb->prb_bdqc.kactive_blk_num)
1111 prev = rb->prb_bdqc.kactive_blk_num-1;
1112 else
1113 prev = rb->prb_bdqc.knum_blocks-1;
1114 return prev;
1115}
1116
1117/* Assumes caller has held the rx_queue.lock */
eea49cc9 1118static void *__prb_previous_block(struct packet_sock *po,
f6fb8f10 1119 struct packet_ring_buffer *rb,
1120 int status)
1121{
1122 unsigned int previous = prb_previous_blk_num(rb);
1123 return prb_lookup_block(po, rb, previous, status);
1124}
1125
eea49cc9 1126static void *packet_previous_rx_frame(struct packet_sock *po,
f6fb8f10 1127 struct packet_ring_buffer *rb,
1128 int status)
1129{
1130 if (po->tp_version <= TPACKET_V2)
1131 return packet_previous_frame(po, rb, status);
1132
1133 return __prb_previous_block(po, rb, status);
1134}
1135
eea49cc9 1136static void packet_increment_rx_head(struct packet_sock *po,
f6fb8f10 1137 struct packet_ring_buffer *rb)
1138{
1139 switch (po->tp_version) {
1140 case TPACKET_V1:
1141 case TPACKET_V2:
1142 return packet_increment_head(rb);
1143 case TPACKET_V3:
1144 default:
1145 WARN(1, "TPACKET version not supported.\n");
1146 BUG();
1147 return;
1148 }
1149}
1150
eea49cc9 1151static void *packet_previous_frame(struct packet_sock *po,
69e3c75f
JB		 1152 struct packet_ring_buffer *rb,
1153 int status)
1154{
1155 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156 return packet_lookup_frame(po, rb, previous, status);
1157}
1158
eea49cc9 1159static void packet_increment_head(struct packet_ring_buffer *buff)
69e3c75f
JB		 1160{
1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162}
1163
1da177e4
LT		 1164static void packet_sock_destruct(struct sock *sk)
1165{
ed85b565
RC		 1166 skb_queue_purge(&sk->sk_error_queue);
1167
547b792c
IJ		 1168 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1169 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1da177e4
LT		 1170
1171 if (!sock_flag(sk, SOCK_DEAD)) {
40d4e3df 1172 pr_err("Attempt to release alive packet socket: %p\n", sk);
1da177e4
LT		 1173 return;
1174 }
1175
17ab56a2 1176 sk_refcnt_debug_dec(sk);
1da177e4
LT		 1177}
1178
dc99f600
DM		 1179static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1180{
1181 int x = atomic_read(&f->rr_cur) + 1;
1182
1183 if (x >= num)
1184 x = 0;
1185
1186 return x;
1187}
1188
1189static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1190{
1191 u32 idx, hash = skb->rxhash;
1192
1193 idx = ((u64)hash * num) >> 32;
1194
1195 return f->arr[idx];
1196}
1197
1198static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1199{
1200 int cur, old;
1201
1202 cur = atomic_read(&f->rr_cur);
1203 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1204 fanout_rr_next(f, num))) != cur)
1205 cur = old;
1206 return f->arr[cur];
1207}
1208
95ec3eb4
DM		 1209static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1210{
1211 unsigned int cpu = smp_processor_id();
1212
1213 return f->arr[cpu % num];
1214}
1215
95ec3eb4
DM		 1216static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1217 struct packet_type *pt, struct net_device *orig_dev)
dc99f600
DM		 1218{
1219 struct packet_fanout *f = pt->af_packet_priv;
1220 unsigned int num = f->num_members;
1221 struct packet_sock *po;
1222 struct sock *sk;
1223
1224 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1225 !num) {
1226 kfree_skb(skb);
1227 return 0;
1228 }
1229
95ec3eb4
DM		 1230 switch (f->type) {
1231 case PACKET_FANOUT_HASH:
1232 default:
1233 if (f->defrag) {
bc416d97 1234 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
95ec3eb4
DM		 1235 if (!skb)
1236 return 0;
1237 }
1238 skb_get_rxhash(skb);
1239 sk = fanout_demux_hash(f, skb, num);
1240 break;
1241 case PACKET_FANOUT_LB:
1242 sk = fanout_demux_lb(f, skb, num);
1243 break;
1244 case PACKET_FANOUT_CPU:
1245 sk = fanout_demux_cpu(f, skb, num);
1246 break;
dc99f600
DM		 1247 }
1248
dc99f600
DM		 1249 po = pkt_sk(sk);
1250
1251 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1252}
1253
1254static DEFINE_MUTEX(fanout_mutex);
1255static LIST_HEAD(fanout_list);
1256
1257static void __fanout_link(struct sock *sk, struct packet_sock *po)
1258{
1259 struct packet_fanout *f = po->fanout;
1260
1261 spin_lock(&f->lock);
1262 f->arr[f->num_members] = sk;
1263 smp_wmb();
1264 f->num_members++;
1265 spin_unlock(&f->lock);
1266}
1267
1268static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1269{
1270 struct packet_fanout *f = po->fanout;
1271 int i;
1272
1273 spin_lock(&f->lock);
1274 for (i = 0; i < f->num_members; i++) {
1275 if (f->arr[i] == sk)
1276 break;
1277 }
1278 BUG_ON(i >= f->num_members);
1279 f->arr[i] = f->arr[f->num_members - 1];
1280 f->num_members--;
1281 spin_unlock(&f->lock);
1282}
1283
7736d33f 1284static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
dc99f600
DM		 1285{
1286 struct packet_sock *po = pkt_sk(sk);
1287 struct packet_fanout *f, *match;
7736d33f
DM		 1288 u8 type = type_flags & 0xff;
1289 u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
dc99f600
DM		 1290 int err;
1291
1292 switch (type) {
1293 case PACKET_FANOUT_HASH:
1294 case PACKET_FANOUT_LB:
95ec3eb4 1295 case PACKET_FANOUT_CPU:
dc99f600
DM		 1296 break;
1297 default:
1298 return -EINVAL;
1299 }
1300
1301 if (!po->running)
1302 return -EINVAL;
1303
1304 if (po->fanout)
1305 return -EALREADY;
1306
1307 mutex_lock(&fanout_mutex);
1308 match = NULL;
1309 list_for_each_entry(f, &fanout_list, list) {
1310 if (f->id == id &&
1311 read_pnet(&f->net) == sock_net(sk)) {
1312 match = f;
1313 break;
1314 }
1315 }
afe62c68 1316 err = -EINVAL;
7736d33f 1317 if (match && match->defrag != defrag)
afe62c68 1318 goto out;
dc99f600 1319 if (!match) {
afe62c68 1320 err = -ENOMEM;
dc99f600 1321 match = kzalloc(sizeof(*match), GFP_KERNEL);
afe62c68
ED		 1322 if (!match)
1323 goto out;
1324 write_pnet(&match->net, sock_net(sk));
1325 match->id = id;
1326 match->type = type;
1327 match->defrag = defrag;
1328 atomic_set(&match->rr_cur, 0);
1329 INIT_LIST_HEAD(&match->list);
1330 spin_lock_init(&match->lock);
1331 atomic_set(&match->sk_ref, 0);
1332 match->prot_hook.type = po->prot_hook.type;
1333 match->prot_hook.dev = po->prot_hook.dev;
1334 match->prot_hook.func = packet_rcv_fanout;
1335 match->prot_hook.af_packet_priv = match;
1336 dev_add_pack(&match->prot_hook);
1337 list_add(&match->list, &fanout_list);
dc99f600 1338 }
afe62c68
ED		 1339 err = -EINVAL;
1340 if (match->type == type &&
1341 match->prot_hook.type == po->prot_hook.type &&
1342 match->prot_hook.dev == po->prot_hook.dev) {
1343 err = -ENOSPC;
1344 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1345 __dev_remove_pack(&po->prot_hook);
1346 po->fanout = match;
1347 atomic_inc(&match->sk_ref);
1348 __fanout_link(sk, po);
1349 err = 0;
dc99f600
DM		 1350 }
1351 }
afe62c68 1352out:
dc99f600
DM		 1353 mutex_unlock(&fanout_mutex);
1354 return err;
1355}
1356
1357static void fanout_release(struct sock *sk)
1358{
1359 struct packet_sock *po = pkt_sk(sk);
1360 struct packet_fanout *f;
1361
1362 f = po->fanout;
1363 if (!f)
1364 return;
1365
1366 po->fanout = NULL;
1367
1368 mutex_lock(&fanout_mutex);
1369 if (atomic_dec_and_test(&f->sk_ref)) {
1370 list_del(&f->list);
1371 dev_remove_pack(&f->prot_hook);
1372 kfree(f);
1373 }
1374 mutex_unlock(&fanout_mutex);
1375}
1da177e4 1376
90ddc4f0 1377static const struct proto_ops packet_ops;
1da177e4 1378
90ddc4f0 1379static const struct proto_ops packet_ops_spkt;
1da177e4 1380
40d4e3df
ED		 1381static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1382 struct packet_type *pt, struct net_device *orig_dev)
1da177e4
LT		 1383{
1384 struct sock *sk;
1385 struct sockaddr_pkt *spkt;
1386
1387 /*
1388 * When we registered the protocol we saved the socket in the data
1389 * field for just this event.
1390 */
1391
1392 sk = pt->af_packet_priv;
1ce4f28b 1393
1da177e4
LT		 1394 /*
1395 * Yank back the headers [hope the device set this
1396 * right or kerboom...]
1397 *
1398 * Incoming packets have ll header pulled,
1399 * push it back.
1400 *
98e399f8 1401 * For outgoing ones skb->data == skb_mac_header(skb)
1da177e4
LT		 1402 * so that this procedure is noop.
1403 */
1404
1405 if (skb->pkt_type == PACKET_LOOPBACK)
1406 goto out;
1407
09ad9bc7 1408 if (!net_eq(dev_net(dev), sock_net(sk)))
d12d01d6
DL		 1409 goto out;
1410
40d4e3df
ED		 1411 skb = skb_share_check(skb, GFP_ATOMIC);
1412 if (skb == NULL)
1da177e4
LT		 1413 goto oom;
1414
1415 /* drop any routing info */
adf30907 1416 skb_dst_drop(skb);
1da177e4 1417
84531c24
PO		 1418 /* drop conntrack reference */
1419 nf_reset(skb);
1420
ffbc6111 1421 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1da177e4 1422
98e399f8 1423 skb_push(skb, skb->data - skb_mac_header(skb));
1da177e4
LT		 1424
1425 /*
1426 * The SOCK_PACKET socket receives _all_ frames.
1427 */
1428
1429 spkt->spkt_family = dev->type;
1430 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1431 spkt->spkt_protocol = skb->protocol;
1432
1433 /*
1434 * Charge the memory to the socket. This is done specifically
1435 * to prevent sockets using all the memory up.
1436 */
1437
40d4e3df 1438 if (sock_queue_rcv_skb(sk, skb) == 0)
1da177e4
LT		 1439 return 0;
1440
1441out:
1442 kfree_skb(skb);
1443oom:
1444 return 0;
1445}
1446
1447
1448/*
1449 * Output a raw packet to a device layer. This bypasses all the other
1450 * protocol layers and you must therefore supply it with a complete frame
1451 */
1ce4f28b 1452
1da177e4
LT		 1453static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1454 struct msghdr *msg, size_t len)
1455{
1456 struct sock *sk = sock->sk;
40d4e3df 1457 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1a35ca80 1458 struct sk_buff *skb = NULL;
1da177e4 1459 struct net_device *dev;
40d4e3df 1460 __be16 proto = 0;
1da177e4 1461 int err;
1ce4f28b 1462
1da177e4 1463 /*
1ce4f28b 1464 * Get and verify the address.
1da177e4
LT		 1465 */
1466
40d4e3df 1467 if (saddr) {
1da177e4 1468 if (msg->msg_namelen < sizeof(struct sockaddr))
40d4e3df
ED		 1469 return -EINVAL;
1470 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1471 proto = saddr->spkt_protocol;
1472 } else
1473 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1da177e4
LT		 1474
1475 /*
1ce4f28b 1476 * Find the device first to size check it
1da177e4
LT		 1477 */
1478
1479 saddr->spkt_device[13] = 0;
1a35ca80 1480retry:
654d1f8a
ED		 1481 rcu_read_lock();
1482 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1da177e4
LT		 1483 err = -ENODEV;
1484 if (dev == NULL)
1485 goto out_unlock;
1ce4f28b 1486
d5e76b0a
DM		 1487 err = -ENETDOWN;
1488 if (!(dev->flags & IFF_UP))
1489 goto out_unlock;
1490
1da177e4 1491 /*
40d4e3df
ED		 1492 * You may not queue a frame bigger than the mtu. This is the lowest level
1493 * raw protocol and you must do your own fragmentation at this level.
1da177e4 1494 */
1ce4f28b 1495
1da177e4 1496 err = -EMSGSIZE;
57f89bfa 1497 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1da177e4
LT		 1498 goto out_unlock;
1499
1a35ca80
ED		 1500 if (!skb) {
1501 size_t reserved = LL_RESERVED_SPACE(dev);
1502 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1503
1504 rcu_read_unlock();
1505 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1506 if (skb == NULL)
1507 return -ENOBUFS;
1508 /* FIXME: Save some space for broken drivers that write a hard
1509 * header at transmission time by themselves. PPP is the notable
1510 * one here. This should really be fixed at the driver level.
1511 */
1512 skb_reserve(skb, reserved);
1513 skb_reset_network_header(skb);
1514
1515 /* Try to align data part correctly */
1516 if (hhlen) {
1517 skb->data -= hhlen;
1518 skb->tail -= hhlen;
1519 if (len < hhlen)
1520 skb_reset_network_header(skb);
1521 }
1522 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1523 if (err)
1524 goto out_free;
1525 goto retry;
1da177e4
LT		 1526 }
1527
57f89bfa
BG		 1528 if (len > (dev->mtu + dev->hard_header_len)) {
1529 /* Earlier code assumed this would be a VLAN pkt,
1530 * double-check this now that we have the actual
1531 * packet in hand.
1532 */
1533 struct ethhdr *ehdr;
1534 skb_reset_mac_header(skb);
1535 ehdr = eth_hdr(skb);
1536 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1537 err = -EMSGSIZE;
1538 goto out_unlock;
1539 }
1540 }
1a35ca80 1541
1da177e4
LT		 1542 skb->protocol = proto;
1543 skb->dev = dev;
1544 skb->priority = sk->sk_priority;
2d37a186 1545 skb->mark = sk->sk_mark;
2244d07b 1546 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
ed85b565
RC		 1547 if (err < 0)
1548 goto out_unlock;
1da177e4
LT		 1549
1550 dev_queue_xmit(skb);
654d1f8a 1551 rcu_read_unlock();
40d4e3df 1552 return len;
1da177e4 1553
1da177e4 1554out_unlock:
654d1f8a 1555 rcu_read_unlock();
1a35ca80
ED		 1556out_free:
1557 kfree_skb(skb);
1da177e4
LT		 1558 return err;
1559}
1da177e4 1560
eea49cc9 1561static unsigned int run_filter(const struct sk_buff *skb,
62ab0812 1562 const struct sock *sk,
dbcb5855 1563 unsigned int res)
1da177e4
LT		 1564{
1565 struct sk_filter *filter;
fda9ef5d 1566
80f8f102
ED		 1567 rcu_read_lock();
1568 filter = rcu_dereference(sk->sk_filter);
dbcb5855 1569 if (filter != NULL)
0a14842f 1570 res = SK_RUN_FILTER(filter, skb);
80f8f102 1571 rcu_read_unlock();
1da177e4 1572
dbcb5855 1573 return res;
1da177e4
LT		 1574}
1575
1576/*
62ab0812
ED		 1577 * This function makes lazy skb cloning in hope that most of packets
1578 * are discarded by BPF.
1579 *
1580 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1581 * and skb->cb are mangled. It works because (and until) packets
1582 * falling here are owned by current CPU. Output packets are cloned
1583 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1584 * sequencially, so that if we return skb to original state on exit,
1585 * we will not harm anyone.
1da177e4
LT		 1586 */
1587
40d4e3df
ED		 1588static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1589 struct packet_type *pt, struct net_device *orig_dev)
1da177e4
LT		 1590{
1591 struct sock *sk;
1592 struct sockaddr_ll *sll;
1593 struct packet_sock *po;
40d4e3df 1594 u8 *skb_head = skb->data;
1da177e4 1595 int skb_len = skb->len;
dbcb5855 1596 unsigned int snaplen, res;
1da177e4
LT		 1597
1598 if (skb->pkt_type == PACKET_LOOPBACK)
1599 goto drop;
1600
1601 sk = pt->af_packet_priv;
1602 po = pkt_sk(sk);
1603
09ad9bc7 1604 if (!net_eq(dev_net(dev), sock_net(sk)))
d12d01d6
DL		 1605 goto drop;
1606
1da177e4
LT		 1607 skb->dev = dev;
1608
3b04ddde 1609 if (dev->header_ops) {
1da177e4 1610 /* The device has an explicit notion of ll header,
62ab0812
ED		 1611 * exported to higher levels.
1612 *
1613 * Otherwise, the device hides details of its frame
1614 * structure, so that corresponding packet head is
1615 * never delivered to user.
1da177e4
LT		 1616 */
1617 if (sk->sk_type != SOCK_DGRAM)
98e399f8 1618 skb_push(skb, skb->data - skb_mac_header(skb));
1da177e4
LT		 1619 else if (skb->pkt_type == PACKET_OUTGOING) {
1620 /* Special case: outgoing packets have ll header at head */
bbe735e4 1621 skb_pull(skb, skb_network_offset(skb));
1da177e4
LT		 1622 }
1623 }
1624
1625 snaplen = skb->len;
1626
dbcb5855
DM		 1627 res = run_filter(skb, sk, snaplen);
1628 if (!res)
fda9ef5d 1629 goto drop_n_restore;
dbcb5855
DM		 1630 if (snaplen > res)
1631 snaplen = res;
1da177e4
LT		 1632
1633 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1634 (unsigned)sk->sk_rcvbuf)
1635 goto drop_n_acct;
1636
1637 if (skb_shared(skb)) {
1638 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1639 if (nskb == NULL)
1640 goto drop_n_acct;
1641
1642 if (skb_head != skb->data) {
1643 skb->data = skb_head;
1644 skb->len = skb_len;
1645 }
1646 kfree_skb(skb);
1647 skb = nskb;
1648 }
1649
ffbc6111
HX		 1650 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1651 sizeof(skb->cb));
1652
1653 sll = &PACKET_SKB_CB(skb)->sa.ll;
1da177e4
LT		 1654 sll->sll_family = AF_PACKET;
1655 sll->sll_hatype = dev->type;
1656 sll->sll_protocol = skb->protocol;
1657 sll->sll_pkttype = skb->pkt_type;
8032b464 1658 if (unlikely(po->origdev))
80feaacb
PWJ		 1659 sll->sll_ifindex = orig_dev->ifindex;
1660 else
1661 sll->sll_ifindex = dev->ifindex;
1da177e4 1662
b95cce35 1663 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1da177e4 1664
ffbc6111 1665 PACKET_SKB_CB(skb)->origlen = skb->len;
8dc41944 1666
1da177e4
LT		 1667 if (pskb_trim(skb, snaplen))
1668 goto drop_n_acct;
1669
1670 skb_set_owner_r(skb, sk);
1671 skb->dev = NULL;
adf30907 1672 skb_dst_drop(skb);
1da177e4 1673
84531c24
PO		 1674 /* drop conntrack reference */
1675 nf_reset(skb);
1676
1da177e4
LT		 1677 spin_lock(&sk->sk_receive_queue.lock);
1678 po->stats.tp_packets++;
3b885787 1679 skb->dropcount = atomic_read(&sk->sk_drops);
1da177e4
LT		 1680 __skb_queue_tail(&sk->sk_receive_queue, skb);
1681 spin_unlock(&sk->sk_receive_queue.lock);
1682 sk->sk_data_ready(sk, skb->len);
1683 return 0;
1684
1685drop_n_acct:
7091fbd8
WB		 1686 spin_lock(&sk->sk_receive_queue.lock);
1687 po->stats.tp_drops++;
1688 atomic_inc(&sk->sk_drops);
1689 spin_unlock(&sk->sk_receive_queue.lock);
1da177e4
LT		 1690
1691drop_n_restore:
1692 if (skb_head != skb->data && skb_shared(skb)) {
1693 skb->data = skb_head;
1694 skb->len = skb_len;
1695 }
1696drop:
ead2ceb0 1697 consume_skb(skb);
1da177e4
LT		 1698 return 0;
1699}
1700
40d4e3df
ED		 1701static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1702 struct packet_type *pt, struct net_device *orig_dev)
1da177e4
LT		 1703{
1704 struct sock *sk;
1705 struct packet_sock *po;
1706 struct sockaddr_ll *sll;
bbd6ef87
PM		 1707 union {
1708 struct tpacket_hdr *h1;
1709 struct tpacket2_hdr *h2;
f6fb8f10 1710 struct tpacket3_hdr *h3;
bbd6ef87
PM		 1711 void *raw;
1712 } h;
40d4e3df 1713 u8 *skb_head = skb->data;
1da177e4 1714 int skb_len = skb->len;
dbcb5855 1715 unsigned int snaplen, res;
f6fb8f10 1716 unsigned long status = TP_STATUS_USER;
bbd6ef87 1717 unsigned short macoff, netoff, hdrlen;
1da177e4 1718 struct sk_buff *copy_skb = NULL;
b7aa0bf7 1719 struct timeval tv;
bbd6ef87 1720 struct timespec ts;
614f60fa 1721 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1da177e4
LT		 1722
1723 if (skb->pkt_type == PACKET_LOOPBACK)
1724 goto drop;
1725
1726 sk = pt->af_packet_priv;
1727 po = pkt_sk(sk);
1728
09ad9bc7 1729 if (!net_eq(dev_net(dev), sock_net(sk)))
d12d01d6
DL		 1730 goto drop;
1731
3b04ddde 1732 if (dev->header_ops) {
1da177e4 1733 if (sk->sk_type != SOCK_DGRAM)
98e399f8 1734 skb_push(skb, skb->data - skb_mac_header(skb));
1da177e4
LT		 1735 else if (skb->pkt_type == PACKET_OUTGOING) {
1736 /* Special case: outgoing packets have ll header at head */
bbe735e4 1737 skb_pull(skb, skb_network_offset(skb));
1da177e4
LT		 1738 }
1739 }
1740
8dc41944
HX		 1741 if (skb->ip_summed == CHECKSUM_PARTIAL)
1742 status |= TP_STATUS_CSUMNOTREADY;
1743
1da177e4
LT		 1744 snaplen = skb->len;
1745
dbcb5855
DM		 1746 res = run_filter(skb, sk, snaplen);
1747 if (!res)
fda9ef5d 1748 goto drop_n_restore;
dbcb5855
DM		 1749 if (snaplen > res)
1750 snaplen = res;
1da177e4
LT		 1751
1752 if (sk->sk_type == SOCK_DGRAM) {
8913336a
PM		 1753 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1754 po->tp_reserve;
1da177e4 1755 } else {
bbe735e4 1756 unsigned maclen = skb_network_offset(skb);
bbd6ef87 1757 netoff = TPACKET_ALIGN(po->tp_hdrlen +
8913336a
PM		 1758 (maclen < 16 ? 16 : maclen)) +
1759 po->tp_reserve;
1da177e4
LT		 1760 macoff = netoff - maclen;
1761 }
f6fb8f10 1762 if (po->tp_version <= TPACKET_V2) {
1763 if (macoff + snaplen > po->rx_ring.frame_size) {
1764 if (po->copy_thresh &&
1765 atomic_read(&sk->sk_rmem_alloc) + skb->truesize
1766 < (unsigned)sk->sk_rcvbuf) {
1767 if (skb_shared(skb)) {
1768 copy_skb = skb_clone(skb, GFP_ATOMIC);
1769 } else {
1770 copy_skb = skb_get(skb);
1771 skb_head = skb->data;
1772 }
1773 if (copy_skb)
1774 skb_set_owner_r(copy_skb, sk);
1da177e4 1775 }
f6fb8f10 1776 snaplen = po->rx_ring.frame_size - macoff;
1777 if ((int)snaplen < 0)
1778 snaplen = 0;
1da177e4 1779 }
1da177e4 1780 }
1da177e4 1781 spin_lock(&sk->sk_receive_queue.lock);
f6fb8f10 1782 h.raw = packet_current_rx_frame(po, skb,
1783 TP_STATUS_KERNEL, (macoff+snaplen));
bbd6ef87 1784 if (!h.raw)
1da177e4 1785 goto ring_is_full;
f6fb8f10 1786 if (po->tp_version <= TPACKET_V2) {
1787 packet_increment_rx_head(po, &po->rx_ring);
1788 /*
1789 * LOSING will be reported till you read the stats,
1790 * because it's COR - Clear On Read.
1791 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1792 * at packet level.
1793 */
1794 if (po->stats.tp_drops)
1795 status |= TP_STATUS_LOSING;
1796 }
1da177e4
LT		 1797 po->stats.tp_packets++;
1798 if (copy_skb) {
1799 status |= TP_STATUS_COPY;
1800 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1801 }
1da177e4
LT		 1802 spin_unlock(&sk->sk_receive_queue.lock);
1803
bbd6ef87 1804 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1da177e4 1805
bbd6ef87
PM		 1806 switch (po->tp_version) {
1807 case TPACKET_V1:
1808 h.h1->tp_len = skb->len;
1809 h.h1->tp_snaplen = snaplen;
1810 h.h1->tp_mac = macoff;
1811 h.h1->tp_net = netoff;
614f60fa
SM		 1812 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1813 && shhwtstamps->syststamp.tv64)
1814 tv = ktime_to_timeval(shhwtstamps->syststamp);
1815 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1816 && shhwtstamps->hwtstamp.tv64)
1817 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1818 else if (skb->tstamp.tv64)
bbd6ef87
PM		 1819 tv = ktime_to_timeval(skb->tstamp);
1820 else
1821 do_gettimeofday(&tv);
1822 h.h1->tp_sec = tv.tv_sec;
1823 h.h1->tp_usec = tv.tv_usec;
1824 hdrlen = sizeof(*h.h1);
1825 break;
1826 case TPACKET_V2:
1827 h.h2->tp_len = skb->len;
1828 h.h2->tp_snaplen = snaplen;
1829 h.h2->tp_mac = macoff;
1830 h.h2->tp_net = netoff;
614f60fa
SM		 1831 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1832 && shhwtstamps->syststamp.tv64)
1833 ts = ktime_to_timespec(shhwtstamps->syststamp);
1834 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1835 && shhwtstamps->hwtstamp.tv64)
1836 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1837 else if (skb->tstamp.tv64)
bbd6ef87
PM		 1838 ts = ktime_to_timespec(skb->tstamp);
1839 else
1840 getnstimeofday(&ts);
1841 h.h2->tp_sec = ts.tv_sec;
1842 h.h2->tp_nsec = ts.tv_nsec;
a3bcc23e
BG		 1843 if (vlan_tx_tag_present(skb)) {
1844 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1845 status |= TP_STATUS_VLAN_VALID;
1846 } else {
1847 h.h2->tp_vlan_tci = 0;
1848 }
13fcb7bd 1849 h.h2->tp_padding = 0;
bbd6ef87
PM		 1850 hdrlen = sizeof(*h.h2);
1851 break;
f6fb8f10 1852 case TPACKET_V3:
1853 /* tp_nxt_offset,vlan are already populated above.
1854 * So DONT clear those fields here
1855 */
1856 h.h3->tp_status |= status;
1857 h.h3->tp_len = skb->len;
1858 h.h3->tp_snaplen = snaplen;
1859 h.h3->tp_mac = macoff;
1860 h.h3->tp_net = netoff;
1861 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1862 && shhwtstamps->syststamp.tv64)
1863 ts = ktime_to_timespec(shhwtstamps->syststamp);
1864 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1865 && shhwtstamps->hwtstamp.tv64)
1866 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1867 else if (skb->tstamp.tv64)
1868 ts = ktime_to_timespec(skb->tstamp);
1869 else
1870 getnstimeofday(&ts);
1871 h.h3->tp_sec = ts.tv_sec;
1872 h.h3->tp_nsec = ts.tv_nsec;
1873 hdrlen = sizeof(*h.h3);
1874 break;
bbd6ef87
PM		 1875 default:
1876 BUG();
1877 }
1da177e4 1878
bbd6ef87 1879 sll = h.raw + TPACKET_ALIGN(hdrlen);
b95cce35 1880 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1da177e4
LT		 1881 sll->sll_family = AF_PACKET;
1882 sll->sll_hatype = dev->type;
1883 sll->sll_protocol = skb->protocol;
1884 sll->sll_pkttype = skb->pkt_type;
8032b464 1885 if (unlikely(po->origdev))
80feaacb
PWJ		 1886 sll->sll_ifindex = orig_dev->ifindex;
1887 else
1888 sll->sll_ifindex = dev->ifindex;
1da177e4 1889
e16aa207 1890 smp_mb();
f6dafa95 1891#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1da177e4 1892 {
0af55bb5
CG		 1893 u8 *start, *end;
1894
f6fb8f10 1895 if (po->tp_version <= TPACKET_V2) {
1896 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1897 + macoff + snaplen);
1898 for (start = h.raw; start < end; start += PAGE_SIZE)
1899 flush_dcache_page(pgv_to_page(start));
1900 }
cc9f01b2 1901 smp_wmb();
1da177e4 1902 }
f6dafa95 1903#endif
f6fb8f10 1904 if (po->tp_version <= TPACKET_V2)
1905 __packet_set_status(po, h.raw, status);
1906 else
1907 prb_clear_blk_fill_status(&po->rx_ring);
1da177e4
LT		 1908
1909 sk->sk_data_ready(sk, 0);
1910
1911drop_n_restore:
1912 if (skb_head != skb->data && skb_shared(skb)) {
1913 skb->data = skb_head;
1914 skb->len = skb_len;
1915 }
1916drop:
1ce4f28b 1917 kfree_skb(skb);
1da177e4
LT		 1918 return 0;
1919
1920ring_is_full:
1921 po->stats.tp_drops++;
1922 spin_unlock(&sk->sk_receive_queue.lock);
1923
1924 sk->sk_data_ready(sk, 0);
acb5d75b 1925 kfree_skb(copy_skb);
1da177e4
LT		 1926 goto drop_n_restore;
1927}
1928
69e3c75f
JB		 1929static void tpacket_destruct_skb(struct sk_buff *skb)
1930{
1931 struct packet_sock *po = pkt_sk(skb->sk);
40d4e3df 1932 void *ph;
1da177e4 1933
69e3c75f
JB		 1934 if (likely(po->tx_ring.pg_vec)) {
1935 ph = skb_shinfo(skb)->destructor_arg;
1936 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1937 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1938 atomic_dec(&po->tx_ring.pending);
1939 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1940 }
1941
1942 sock_wfree(skb);
1943}
1944
40d4e3df
ED		 1945static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1946 void *frame, struct net_device *dev, int size_max,
ae641949 1947 __be16 proto, unsigned char *addr, int hlen)
69e3c75f
JB		 1948{
1949 union {
1950 struct tpacket_hdr *h1;
1951 struct tpacket2_hdr *h2;
1952 void *raw;
1953 } ph;
1954 int to_write, offset, len, tp_len, nr_frags, len_max;
1955 struct socket *sock = po->sk.sk_socket;
1956 struct page *page;
1957 void *data;
1958 int err;
1959
1960 ph.raw = frame;
1961
1962 skb->protocol = proto;
1963 skb->dev = dev;
1964 skb->priority = po->sk.sk_priority;
2d37a186 1965 skb->mark = po->sk.sk_mark;
69e3c75f
JB		 1966 skb_shinfo(skb)->destructor_arg = ph.raw;
1967
1968 switch (po->tp_version) {
1969 case TPACKET_V2:
1970 tp_len = ph.h2->tp_len;
1971 break;
1972 default:
1973 tp_len = ph.h1->tp_len;
1974 break;
1975 }
1976 if (unlikely(tp_len > size_max)) {
40d4e3df 1977 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
69e3c75f
JB		 1978 return -EMSGSIZE;
1979 }
1980
ae641949 1981 skb_reserve(skb, hlen);
69e3c75f
JB		 1982 skb_reset_network_header(skb);
1983
1984 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1985 to_write = tp_len;
1986
1987 if (sock->type == SOCK_DGRAM) {
1988 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1989 NULL, tp_len);
1990 if (unlikely(err < 0))
1991 return -EINVAL;
40d4e3df 1992 } else if (dev->hard_header_len) {
69e3c75f
JB		 1993 /* net device doesn't like empty head */
1994 if (unlikely(tp_len <= dev->hard_header_len)) {
40d4e3df
ED		 1995 pr_err("packet size is too short (%d < %d)\n",
1996 tp_len, dev->hard_header_len);
69e3c75f
JB		 1997 return -EINVAL;
1998 }
1999
2000 skb_push(skb, dev->hard_header_len);
2001 err = skb_store_bits(skb, 0, data,
2002 dev->hard_header_len);
2003 if (unlikely(err))
2004 return err;
2005
2006 data += dev->hard_header_len;
2007 to_write -= dev->hard_header_len;
2008 }
2009
2010 err = -EFAULT;
69e3c75f
JB		 2011 offset = offset_in_page(data);
2012 len_max = PAGE_SIZE - offset;
2013 len = ((to_write > len_max) ? len_max : to_write);
2014
2015 skb->data_len = to_write;
2016 skb->len += to_write;
2017 skb->truesize += to_write;
2018 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2019
2020 while (likely(to_write)) {
2021 nr_frags = skb_shinfo(skb)->nr_frags;
2022
2023 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
40d4e3df
ED		 2024 pr_err("Packet exceed the number of skb frags(%lu)\n",
2025 MAX_SKB_FRAGS);
69e3c75f
JB		 2026 return -EFAULT;
2027 }
2028
0af55bb5
CG		 2029 page = pgv_to_page(data);
2030 data += len;
69e3c75f
JB		 2031 flush_dcache_page(page);
2032 get_page(page);
0af55bb5 2033 skb_fill_page_desc(skb, nr_frags, page, offset, len);
69e3c75f
JB		 2034 to_write -= len;
2035 offset = 0;
2036 len_max = PAGE_SIZE;
2037 len = ((to_write > len_max) ? len_max : to_write);
2038 }
2039
2040 return tp_len;
2041}
2042
2043static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2044{
69e3c75f
JB		 2045 struct sk_buff *skb;
2046 struct net_device *dev;
2047 __be16 proto;
827d9780
BG		 2048 bool need_rls_dev = false;
2049 int err, reserve = 0;
40d4e3df
ED		 2050 void *ph;
2051 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
69e3c75f
JB		 2052 int tp_len, size_max;
2053 unsigned char *addr;
2054 int len_sum = 0;
2055 int status = 0;
ae641949 2056 int hlen, tlen;
69e3c75f 2057
69e3c75f
JB		 2058 mutex_lock(&po->pg_vec_lock);
2059
2060 err = -EBUSY;
2061 if (saddr == NULL) {
827d9780 2062 dev = po->prot_hook.dev;
69e3c75f
JB		 2063 proto = po->num;
2064 addr = NULL;
2065 } else {
2066 err = -EINVAL;
2067 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2068 goto out;
2069 if (msg->msg_namelen < (saddr->sll_halen
2070 + offsetof(struct sockaddr_ll,
2071 sll_addr)))
2072 goto out;
69e3c75f
JB		 2073 proto = saddr->sll_protocol;
2074 addr = saddr->sll_addr;
827d9780
BG		 2075 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2076 need_rls_dev = true;
69e3c75f
JB		 2077 }
2078
69e3c75f
JB		 2079 err = -ENXIO;
2080 if (unlikely(dev == NULL))
2081 goto out;
2082
2083 reserve = dev->hard_header_len;
2084
2085 err = -ENETDOWN;
2086 if (unlikely(!(dev->flags & IFF_UP)))
2087 goto out_put;
2088
2089 size_max = po->tx_ring.frame_size
b5dd884e 2090 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
69e3c75f
JB		 2091
2092 if (size_max > dev->mtu + reserve)
2093 size_max = dev->mtu + reserve;
2094
2095 do {
2096 ph = packet_current_frame(po, &po->tx_ring,
2097 TP_STATUS_SEND_REQUEST);
2098
2099 if (unlikely(ph == NULL)) {
2100 schedule();
2101 continue;
2102 }
2103
2104 status = TP_STATUS_SEND_REQUEST;
ae641949
HX		 2105 hlen = LL_RESERVED_SPACE(dev);
2106 tlen = dev->needed_tailroom;
69e3c75f 2107 skb = sock_alloc_send_skb(&po->sk,
ae641949 2108 hlen + tlen + sizeof(struct sockaddr_ll),
69e3c75f
JB		 2109 0, &err);
2110
2111 if (unlikely(skb == NULL))
2112 goto out_status;
2113
2114 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
ae641949 2115 addr, hlen);
69e3c75f
JB		 2116
2117 if (unlikely(tp_len < 0)) {
2118 if (po->tp_loss) {
2119 __packet_set_status(po, ph,
2120 TP_STATUS_AVAILABLE);
2121 packet_increment_head(&po->tx_ring);
2122 kfree_skb(skb);
2123 continue;
2124 } else {
2125 status = TP_STATUS_WRONG_FORMAT;
2126 err = tp_len;
2127 goto out_status;
2128 }
2129 }
2130
2131 skb->destructor = tpacket_destruct_skb;
2132 __packet_set_status(po, ph, TP_STATUS_SENDING);
2133 atomic_inc(&po->tx_ring.pending);
2134
2135 status = TP_STATUS_SEND_REQUEST;
2136 err = dev_queue_xmit(skb);
eb70df13
JP		 2137 if (unlikely(err > 0)) {
2138 err = net_xmit_errno(err);
2139 if (err && __packet_get_status(po, ph) ==
2140 TP_STATUS_AVAILABLE) {
2141 /* skb was destructed already */
2142 skb = NULL;
2143 goto out_status;
2144 }
2145 /*
2146 * skb was dropped but not destructed yet;
2147 * let's treat it like congestion or err < 0
2148 */
2149 err = 0;
2150 }
69e3c75f
JB		 2151 packet_increment_head(&po->tx_ring);
2152 len_sum += tp_len;
f64f9e71
JP		 2153 } while (likely((ph != NULL) ||
2154 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2155 (atomic_read(&po->tx_ring.pending))))
2156 );
69e3c75f
JB		 2157
2158 err = len_sum;
2159 goto out_put;
2160
69e3c75f
JB		 2161out_status:
2162 __packet_set_status(po, ph, status);
2163 kfree_skb(skb);
2164out_put:
827d9780
BG		 2165 if (need_rls_dev)
2166 dev_put(dev);
69e3c75f
JB		 2167out:
2168 mutex_unlock(&po->pg_vec_lock);
2169 return err;
2170}
69e3c75f 2171
eea49cc9
OJ		 2172static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2173 size_t reserve, size_t len,
2174 size_t linear, int noblock,
2175 int *err)
bfd5f4a3
SS		 2176{
2177 struct sk_buff *skb;
2178
2179 /* Under a page? Don't bother with paged skb. */
2180 if (prepad + len < PAGE_SIZE || !linear)
2181 linear = len;
2182
2183 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2184 err);
2185 if (!skb)
2186 return NULL;
2187
2188 skb_reserve(skb, reserve);
2189 skb_put(skb, linear);
2190 skb->data_len = len - linear;
2191 skb->len += len - linear;
2192
2193 return skb;
2194}
2195
69e3c75f 2196static int packet_snd(struct socket *sock,
1da177e4
LT		 2197 struct msghdr *msg, size_t len)
2198{
2199 struct sock *sk = sock->sk;
40d4e3df 2200 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1da177e4
LT		 2201 struct sk_buff *skb;
2202 struct net_device *dev;
0e11c91e 2203 __be16 proto;
827d9780 2204 bool need_rls_dev = false;
1da177e4 2205 unsigned char *addr;
827d9780 2206 int err, reserve = 0;
bfd5f4a3
SS		 2207 struct virtio_net_hdr vnet_hdr = { 0 };
2208 int offset = 0;
2209 int vnet_hdr_len;
2210 struct packet_sock *po = pkt_sk(sk);
2211 unsigned short gso_type = 0;
ae641949 2212 int hlen, tlen;
1da177e4
LT		 2213
2214 /*
1ce4f28b 2215 * Get and verify the address.
1da177e4 2216 */
1ce4f28b 2217
1da177e4 2218 if (saddr == NULL) {
827d9780 2219 dev = po->prot_hook.dev;
1da177e4
LT		 2220 proto = po->num;
2221 addr = NULL;
2222 } else {
2223 err = -EINVAL;
2224 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2225 goto out;
0fb375fb
EB		 2226 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2227 goto out;
1da177e4
LT		 2228 proto = saddr->sll_protocol;
2229 addr = saddr->sll_addr;
827d9780
BG		 2230 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2231 need_rls_dev = true;
1da177e4
LT		 2232 }
2233
1da177e4
LT		 2234 err = -ENXIO;
2235 if (dev == NULL)
2236 goto out_unlock;
2237 if (sock->type == SOCK_RAW)
2238 reserve = dev->hard_header_len;
2239
d5e76b0a
DM		 2240 err = -ENETDOWN;
2241 if (!(dev->flags & IFF_UP))
2242 goto out_unlock;
2243
bfd5f4a3
SS		 2244 if (po->has_vnet_hdr) {
2245 vnet_hdr_len = sizeof(vnet_hdr);
2246
2247 err = -EINVAL;
2248 if (len < vnet_hdr_len)
2249 goto out_unlock;
2250
2251 len -= vnet_hdr_len;
2252
2253 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2254 vnet_hdr_len);
2255 if (err < 0)
2256 goto out_unlock;
2257
2258 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2259 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2260 vnet_hdr.hdr_len))
2261 vnet_hdr.hdr_len = vnet_hdr.csum_start +
2262 vnet_hdr.csum_offset + 2;
2263
2264 err = -EINVAL;
2265 if (vnet_hdr.hdr_len > len)
2266 goto out_unlock;
2267
2268 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2269 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2270 case VIRTIO_NET_HDR_GSO_TCPV4:
2271 gso_type = SKB_GSO_TCPV4;
2272 break;
2273 case VIRTIO_NET_HDR_GSO_TCPV6:
2274 gso_type = SKB_GSO_TCPV6;
2275 break;
2276 case VIRTIO_NET_HDR_GSO_UDP:
2277 gso_type = SKB_GSO_UDP;
2278 break;
2279 default:
2280 goto out_unlock;
2281 }
2282
2283 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2284 gso_type |= SKB_GSO_TCP_ECN;
2285
2286 if (vnet_hdr.gso_size == 0)
2287 goto out_unlock;
2288
2289 }
2290 }
2291
1da177e4 2292 err = -EMSGSIZE;
57f89bfa 2293 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1da177e4
LT		 2294 goto out_unlock;
2295
bfd5f4a3 2296 err = -ENOBUFS;
ae641949
HX		 2297 hlen = LL_RESERVED_SPACE(dev);
2298 tlen = dev->needed_tailroom;
2299 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
bfd5f4a3 2300 msg->msg_flags & MSG_DONTWAIT, &err);
40d4e3df 2301 if (skb == NULL)
1da177e4
LT		 2302 goto out_unlock;
2303
bfd5f4a3 2304 skb_set_network_header(skb, reserve);
1da177e4 2305
0c4e8581
SH		 2306 err = -EINVAL;
2307 if (sock->type == SOCK_DGRAM &&
bfd5f4a3 2308 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
0c4e8581 2309 goto out_free;
1da177e4
LT		 2310
2311 /* Returns -EFAULT on error */
bfd5f4a3 2312 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1da177e4
LT		 2313 if (err)
2314 goto out_free;
2244d07b 2315 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
ed85b565
RC		 2316 if (err < 0)
2317 goto out_free;
1da177e4 2318
57f89bfa
BG		 2319 if (!gso_type && (len > dev->mtu + reserve)) {
2320 /* Earlier code assumed this would be a VLAN pkt,
2321 * double-check this now that we have the actual
2322 * packet in hand.
2323 */
2324 struct ethhdr *ehdr;
2325 skb_reset_mac_header(skb);
2326 ehdr = eth_hdr(skb);
2327 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2328 err = -EMSGSIZE;
2329 goto out_free;
2330 }
2331 }
2332
1da177e4
LT		 2333 skb->protocol = proto;
2334 skb->dev = dev;
2335 skb->priority = sk->sk_priority;
2d37a186 2336 skb->mark = sk->sk_mark;
1da177e4 2337
bfd5f4a3
SS		 2338 if (po->has_vnet_hdr) {
2339 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2340 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2341 vnet_hdr.csum_offset)) {
2342 err = -EINVAL;
2343 goto out_free;
2344 }
2345 }
2346
2347 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2348 skb_shinfo(skb)->gso_type = gso_type;
2349
2350 /* Header must be checked, and gso_segs computed. */
2351 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2352 skb_shinfo(skb)->gso_segs = 0;
2353
2354 len += vnet_hdr_len;
2355 }
2356
1da177e4
LT		 2357 /*
2358 * Now send it
2359 */
2360
2361 err = dev_queue_xmit(skb);
2362 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2363 goto out_unlock;
2364
827d9780
BG		 2365 if (need_rls_dev)
2366 dev_put(dev);
1da177e4 2367
40d4e3df 2368 return len;
1da177e4
LT		 2369
2370out_free:
2371 kfree_skb(skb);
2372out_unlock:
827d9780 2373 if (dev && need_rls_dev)
1da177e4
LT		 2374 dev_put(dev);
2375out:
2376 return err;
2377}
2378
69e3c75f
JB		 2379static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2380 struct msghdr *msg, size_t len)
2381{
69e3c75f
JB		 2382 struct sock *sk = sock->sk;
2383 struct packet_sock *po = pkt_sk(sk);
2384 if (po->tx_ring.pg_vec)
2385 return tpacket_snd(po, msg);
2386 else
69e3c75f
JB		 2387 return packet_snd(sock, msg, len);
2388}
2389
1da177e4
LT		 2390/*
2391 * Close a PACKET socket. This is fairly simple. We immediately go
2392 * to 'closed' state and remove our protocol entry in the device list.
2393 */
2394
2395static int packet_release(struct socket *sock)
2396{
2397 struct sock *sk = sock->sk;
2398 struct packet_sock *po;
d12d01d6 2399 struct net *net;
f6fb8f10 2400 union tpacket_req_u req_u;
1da177e4
LT		 2401
2402 if (!sk)
2403 return 0;
2404
3b1e0a65 2405 net = sock_net(sk);
1da177e4
LT		 2406 po = pkt_sk(sk);
2407
808f5114 2408 spin_lock_bh(&net->packet.sklist_lock);
2409 sk_del_node_init_rcu(sk);
920de804 2410 sock_prot_inuse_add(net, sk->sk_prot, -1);
808f5114 2411 spin_unlock_bh(&net->packet.sklist_lock);
1da177e4 2412
808f5114 2413 spin_lock(&po->bind_lock);
ce06b03e 2414 unregister_prot_hook(sk, false);
160ff18a
BG		 2415 if (po->prot_hook.dev) {
2416 dev_put(po->prot_hook.dev);
2417 po->prot_hook.dev = NULL;
2418 }
808f5114 2419 spin_unlock(&po->bind_lock);
1da177e4 2420
1da177e4 2421 packet_flush_mclist(sk);
1da177e4 2422
f6fb8f10 2423 memset(&req_u, 0, sizeof(req_u));
69e3c75f
JB		 2424
2425 if (po->rx_ring.pg_vec)
f6fb8f10 2426 packet_set_ring(sk, &req_u, 1, 0);
69e3c75f
JB		 2427
2428 if (po->tx_ring.pg_vec)
f6fb8f10 2429 packet_set_ring(sk, &req_u, 1, 1);
1da177e4 2430
dc99f600
DM		 2431 fanout_release(sk);
2432
808f5114 2433 synchronize_net();
1da177e4
LT		 2434 /*
2435 * Now the socket is dead. No more input will appear.
2436 */
1da177e4
LT		 2437 sock_orphan(sk);
2438 sock->sk = NULL;
2439
2440 /* Purge queues */
2441
2442 skb_queue_purge(&sk->sk_receive_queue);
17ab56a2 2443 sk_refcnt_debug_release(sk);
1da177e4
LT		 2444
2445 sock_put(sk);
2446 return 0;
2447}
2448
2449/*
2450 * Attach a packet hook.
2451 */
2452
0e11c91e 2453static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1da177e4
LT		 2454{
2455 struct packet_sock *po = pkt_sk(sk);
dc99f600
DM		 2456
2457 if (po->fanout)
2458 return -EINVAL;
1da177e4
LT		 2459
2460 lock_sock(sk);
2461
2462 spin_lock(&po->bind_lock);
ce06b03e 2463 unregister_prot_hook(sk, true);
1da177e4
LT		 2464 po->num = protocol;
2465 po->prot_hook.type = protocol;
160ff18a
BG		 2466 if (po->prot_hook.dev)
2467 dev_put(po->prot_hook.dev);
1da177e4
LT		 2468 po->prot_hook.dev = dev;
2469
2470 po->ifindex = dev ? dev->ifindex : 0;
2471
2472 if (protocol == 0)
2473 goto out_unlock;
2474
be85d4ad 2475 if (!dev || (dev->flags & IFF_UP)) {
ce06b03e 2476 register_prot_hook(sk);
be85d4ad
UT		 2477 } else {
2478 sk->sk_err = ENETDOWN;
2479 if (!sock_flag(sk, SOCK_DEAD))
2480 sk->sk_error_report(sk);
1da177e4
LT		 2481 }
2482
2483out_unlock:
2484 spin_unlock(&po->bind_lock);
2485 release_sock(sk);
2486 return 0;
2487}
2488
2489/*
2490 * Bind a packet socket to a device
2491 */
2492
40d4e3df
ED		 2493static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2494 int addr_len)
1da177e4 2495{
40d4e3df 2496 struct sock *sk = sock->sk;
1da177e4
LT		 2497 char name[15];
2498 struct net_device *dev;
2499 int err = -ENODEV;
1ce4f28b 2500
1da177e4
LT		 2501 /*
2502 * Check legality
2503 */
1ce4f28b 2504
8ae55f04 2505 if (addr_len != sizeof(struct sockaddr))
1da177e4 2506 return -EINVAL;
40d4e3df 2507 strlcpy(name, uaddr->sa_data, sizeof(name));
1da177e4 2508
3b1e0a65 2509 dev = dev_get_by_name(sock_net(sk), name);
160ff18a 2510 if (dev)
1da177e4 2511 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1da177e4
LT		 2512 return err;
2513}
1da177e4
LT		 2514
2515static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2516{
40d4e3df
ED		 2517 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2518 struct sock *sk = sock->sk;
1da177e4
LT		 2519 struct net_device *dev = NULL;
2520 int err;
2521
2522
2523 /*
2524 * Check legality
2525 */
1ce4f28b 2526
1da177e4
LT		 2527 if (addr_len < sizeof(struct sockaddr_ll))
2528 return -EINVAL;
2529 if (sll->sll_family != AF_PACKET)
2530 return -EINVAL;
2531
2532 if (sll->sll_ifindex) {
2533 err = -ENODEV;
3b1e0a65 2534 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1da177e4
LT		 2535 if (dev == NULL)
2536 goto out;
2537 }
2538 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1da177e4
LT		 2539
2540out:
2541 return err;
2542}
2543
2544static struct proto packet_proto = {
2545 .name = "PACKET",
2546 .owner = THIS_MODULE,
2547 .obj_size = sizeof(struct packet_sock),
2548};
2549
2550/*
1ce4f28b 2551 * Create a packet of type SOCK_PACKET.
1da177e4
LT		 2552 */
2553
3f378b68
EP		 2554static int packet_create(struct net *net, struct socket *sock, int protocol,
2555 int kern)
1da177e4
LT		 2556{
2557 struct sock *sk;
2558 struct packet_sock *po;
0e11c91e 2559 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1da177e4
LT		 2560 int err;
2561
2562 if (!capable(CAP_NET_RAW))
2563 return -EPERM;
be02097c
DM		 2564 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2565 sock->type != SOCK_PACKET)
1da177e4
LT		 2566 return -ESOCKTNOSUPPORT;
2567
2568 sock->state = SS_UNCONNECTED;
2569
2570 err = -ENOBUFS;
6257ff21 2571 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1da177e4
LT		 2572 if (sk == NULL)
2573 goto out;
2574
2575 sock->ops = &packet_ops;
1da177e4
LT		 2576 if (sock->type == SOCK_PACKET)
2577 sock->ops = &packet_ops_spkt;
be02097c 2578
1da177e4
LT		 2579 sock_init_data(sock, sk);
2580
2581 po = pkt_sk(sk);
2582 sk->sk_family = PF_PACKET;
0e11c91e 2583 po->num = proto;
1da177e4
LT		 2584
2585 sk->sk_destruct = packet_sock_destruct;
17ab56a2 2586 sk_refcnt_debug_inc(sk);
1da177e4
LT		 2587
2588 /*
2589 * Attach a protocol block
2590 */
2591
2592 spin_lock_init(&po->bind_lock);
905db440 2593 mutex_init(&po->pg_vec_lock);
1da177e4 2594 po->prot_hook.func = packet_rcv;
be02097c 2595
1da177e4
LT		 2596 if (sock->type == SOCK_PACKET)
2597 po->prot_hook.func = packet_rcv_spkt;
be02097c 2598
1da177e4
LT		 2599 po->prot_hook.af_packet_priv = sk;
2600
0e11c91e
AV		 2601 if (proto) {
2602 po->prot_hook.type = proto;
ce06b03e 2603 register_prot_hook(sk);
1da177e4
LT		 2604 }
2605
808f5114 2606 spin_lock_bh(&net->packet.sklist_lock);
2607 sk_add_node_rcu(sk, &net->packet.sklist);
3680453c 2608 sock_prot_inuse_add(net, &packet_proto, 1);
808f5114 2609 spin_unlock_bh(&net->packet.sklist_lock);
2610
40d4e3df 2611 return 0;
1da177e4
LT		 2612out:
2613 return err;
2614}
2615
ed85b565
RC		 2616static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2617{
2618 struct sock_exterr_skb *serr;
2619 struct sk_buff *skb, *skb2;
2620 int copied, err;
2621
2622 err = -EAGAIN;
2623 skb = skb_dequeue(&sk->sk_error_queue);
2624 if (skb == NULL)
2625 goto out;
2626
2627 copied = skb->len;
2628 if (copied > len) {
2629 msg->msg_flags |= MSG_TRUNC;
2630 copied = len;
2631 }
2632 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2633 if (err)
2634 goto out_free_skb;
2635
2636 sock_recv_timestamp(msg, sk, skb);
2637
2638 serr = SKB_EXT_ERR(skb);
2639 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2640 sizeof(serr->ee), &serr->ee);
2641
2642 msg->msg_flags |= MSG_ERRQUEUE;
2643 err = copied;
2644
2645 /* Reset and regenerate socket error */
2646 spin_lock_bh(&sk->sk_error_queue.lock);
2647 sk->sk_err = 0;
2648 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2649 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2650 spin_unlock_bh(&sk->sk_error_queue.lock);
2651 sk->sk_error_report(sk);
2652 } else
2653 spin_unlock_bh(&sk->sk_error_queue.lock);
2654
2655out_free_skb:
2656 kfree_skb(skb);
2657out:
2658 return err;
2659}
2660
1da177e4
LT		 2661/*
2662 * Pull a packet from our receive queue and hand it to the user.
2663 * If necessary we block.
2664 */
2665
2666static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2667 struct msghdr *msg, size_t len, int flags)
2668{
2669 struct sock *sk = sock->sk;
2670 struct sk_buff *skb;
2671 int copied, err;
0fb375fb 2672 struct sockaddr_ll *sll;
bfd5f4a3 2673 int vnet_hdr_len = 0;
1da177e4
LT		 2674
2675 err = -EINVAL;
ed85b565 2676 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1da177e4
LT		 2677 goto out;
2678
2679#if 0
2680 /* What error should we return now? EUNATTACH? */
2681 if (pkt_sk(sk)->ifindex < 0)
2682 return -ENODEV;
2683#endif
2684
ed85b565
RC		 2685 if (flags & MSG_ERRQUEUE) {
2686 err = packet_recv_error(sk, msg, len);
2687 goto out;
2688 }
2689
1da177e4
LT		 2690 /*
2691 * Call the generic datagram receiver. This handles all sorts
2692 * of horrible races and re-entrancy so we can forget about it
2693 * in the protocol layers.
2694 *
2695 * Now it will return ENETDOWN, if device have just gone down,
2696 * but then it will block.
2697 */
2698
40d4e3df 2699 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1da177e4
LT		 2700
2701 /*
1ce4f28b 2702 * An error occurred so return it. Because skb_recv_datagram()
1da177e4
LT		 2703 * handles the blocking we don't see and worry about blocking
2704 * retries.
2705 */
2706
8ae55f04 2707 if (skb == NULL)
1da177e4
LT		 2708 goto out;
2709
bfd5f4a3
SS		 2710 if (pkt_sk(sk)->has_vnet_hdr) {
2711 struct virtio_net_hdr vnet_hdr = { 0 };
2712
2713 err = -EINVAL;
2714 vnet_hdr_len = sizeof(vnet_hdr);
1f18b717 2715 if (len < vnet_hdr_len)
bfd5f4a3
SS		 2716 goto out_free;
2717
1f18b717
MK		 2718 len -= vnet_hdr_len;
2719
bfd5f4a3
SS		 2720 if (skb_is_gso(skb)) {
2721 struct skb_shared_info *sinfo = skb_shinfo(skb);
2722
2723 /* This is a hint as to how much should be linear. */
2724 vnet_hdr.hdr_len = skb_headlen(skb);
2725 vnet_hdr.gso_size = sinfo->gso_size;
2726 if (sinfo->gso_type & SKB_GSO_TCPV4)
2727 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2728 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2729 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2730 else if (sinfo->gso_type & SKB_GSO_UDP)
2731 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2732 else if (sinfo->gso_type & SKB_GSO_FCOE)
2733 goto out_free;
2734 else
2735 BUG();
2736 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2737 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2738 } else
2739 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2740
2741 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2742 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
55508d60 2743 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
bfd5f4a3 2744 vnet_hdr.csum_offset = skb->csum_offset;
10a8d94a
JW		 2745 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2746 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
bfd5f4a3
SS		 2747 } /* else everything is zero */
2748
2749 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2750 vnet_hdr_len);
2751 if (err < 0)
2752 goto out_free;
2753 }
2754
0fb375fb
EB		 2755 /*
2756 * If the address length field is there to be filled in, we fill
2757 * it in now.
2758 */
2759
ffbc6111 2760 sll = &PACKET_SKB_CB(skb)->sa.ll;
0fb375fb
EB		 2761 if (sock->type == SOCK_PACKET)
2762 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2763 else
2764 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2765
1da177e4
LT		 2766 /*
2767 * You lose any data beyond the buffer you gave. If it worries a
2768 * user program they can ask the device for its MTU anyway.
2769 */
2770
2771 copied = skb->len;
40d4e3df
ED		 2772 if (copied > len) {
2773 copied = len;
2774 msg->msg_flags |= MSG_TRUNC;
1da177e4
LT		 2775 }
2776
2777 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2778 if (err)
2779 goto out_free;
2780
3b885787 2781 sock_recv_ts_and_drops(msg, sk, skb);
1da177e4
LT		 2782
2783 if (msg->msg_name)
ffbc6111
HX		 2784 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2785 msg->msg_namelen);
1da177e4 2786
8dc41944 2787 if (pkt_sk(sk)->auxdata) {
ffbc6111
HX		 2788 struct tpacket_auxdata aux;
2789
2790 aux.tp_status = TP_STATUS_USER;
2791 if (skb->ip_summed == CHECKSUM_PARTIAL)
2792 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2793 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2794 aux.tp_snaplen = skb->len;
2795 aux.tp_mac = 0;
bbe735e4 2796 aux.tp_net = skb_network_offset(skb);
a3bcc23e
BG		 2797 if (vlan_tx_tag_present(skb)) {
2798 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2799 aux.tp_status |= TP_STATUS_VLAN_VALID;
2800 } else {
2801 aux.tp_vlan_tci = 0;
2802 }
13fcb7bd 2803 aux.tp_padding = 0;
ffbc6111 2804 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
8dc41944
HX		 2805 }
2806
1da177e4
LT		 2807 /*
2808 * Free or return the buffer as appropriate. Again this
2809 * hides all the races and re-entrancy issues from us.
2810 */
bfd5f4a3 2811 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1da177e4
LT		 2812
2813out_free:
2814 skb_free_datagram(sk, skb);
2815out:
2816 return err;
2817}
2818
1da177e4
LT		 2819static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2820 int *uaddr_len, int peer)
2821{
2822 struct net_device *dev;
2823 struct sock *sk = sock->sk;
2824
2825 if (peer)
2826 return -EOPNOTSUPP;
2827
2828 uaddr->sa_family = AF_PACKET;
654d1f8a
ED		 2829 rcu_read_lock();
2830 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2831 if (dev)
67286640 2832 strncpy(uaddr->sa_data, dev->name, 14);
654d1f8a 2833 else
1da177e4 2834 memset(uaddr->sa_data, 0, 14);
654d1f8a 2835 rcu_read_unlock();
1da177e4
LT		 2836 *uaddr_len = sizeof(*uaddr);
2837
2838 return 0;
2839}
1da177e4
LT		 2840
2841static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2842 int *uaddr_len, int peer)
2843{
2844 struct net_device *dev;
2845 struct sock *sk = sock->sk;
2846 struct packet_sock *po = pkt_sk(sk);
13cfa97b 2847 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1da177e4
LT		 2848
2849 if (peer)
2850 return -EOPNOTSUPP;
2851
2852 sll->sll_family = AF_PACKET;
2853 sll->sll_ifindex = po->ifindex;
2854 sll->sll_protocol = po->num;
67286640 2855 sll->sll_pkttype = 0;
654d1f8a
ED		 2856 rcu_read_lock();
2857 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1da177e4
LT		 2858 if (dev) {
2859 sll->sll_hatype = dev->type;
2860 sll->sll_halen = dev->addr_len;
2861 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1da177e4
LT		 2862 } else {
2863 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
2864 sll->sll_halen = 0;
2865 }
654d1f8a 2866 rcu_read_unlock();
0fb375fb 2867 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1da177e4
LT		 2868
2869 return 0;
2870}
2871
2aeb0b88
WC		 2872static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2873 int what)
1da177e4
LT		 2874{
2875 switch (i->type) {
2876 case PACKET_MR_MULTICAST:
1162563f
JP		 2877 if (i->alen != dev->addr_len)
2878 return -EINVAL;
1da177e4 2879 if (what > 0)
22bedad3 2880 return dev_mc_add(dev, i->addr);
1da177e4 2881 else
22bedad3 2882 return dev_mc_del(dev, i->addr);
1da177e4
LT		 2883 break;
2884 case PACKET_MR_PROMISC:
2aeb0b88 2885 return dev_set_promiscuity(dev, what);
1da177e4
LT		 2886 break;
2887 case PACKET_MR_ALLMULTI:
2aeb0b88 2888 return dev_set_allmulti(dev, what);
1da177e4 2889 break;
d95ed927 2890 case PACKET_MR_UNICAST:
1162563f
JP		 2891 if (i->alen != dev->addr_len)
2892 return -EINVAL;
d95ed927 2893 if (what > 0)
a748ee24 2894 return dev_uc_add(dev, i->addr);
d95ed927 2895 else
a748ee24 2896 return dev_uc_del(dev, i->addr);
d95ed927 2897 break;
40d4e3df
ED		 2898 default:
2899 break;
1da177e4 2900 }
2aeb0b88 2901 return 0;
1da177e4
LT		 2902}
2903
2904static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2905{
40d4e3df 2906 for ( ; i; i = i->next) {
1da177e4
LT		 2907 if (i->ifindex == dev->ifindex)
2908 packet_dev_mc(dev, i, what);
2909 }
2910}
2911
0fb375fb 2912static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1da177e4
LT		 2913{
2914 struct packet_sock *po = pkt_sk(sk);
2915 struct packet_mclist *ml, *i;
2916 struct net_device *dev;
2917 int err;
2918
2919 rtnl_lock();
2920
2921 err = -ENODEV;
3b1e0a65 2922 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1da177e4
LT		 2923 if (!dev)
2924 goto done;
2925
2926 err = -EINVAL;
1162563f 2927 if (mreq->mr_alen > dev->addr_len)
1da177e4
LT		 2928 goto done;
2929
2930 err = -ENOBUFS;
8b3a7005 2931 i = kmalloc(sizeof(*i), GFP_KERNEL);
1da177e4
LT		 2932 if (i == NULL)
2933 goto done;
2934
2935 err = 0;
2936 for (ml = po->mclist; ml; ml = ml->next) {
2937 if (ml->ifindex == mreq->mr_ifindex &&
2938 ml->type == mreq->mr_type &&
2939 ml->alen == mreq->mr_alen &&
2940 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2941 ml->count++;
2942 /* Free the new element ... */
2943 kfree(i);
2944 goto done;
2945 }
2946 }
2947
2948 i->type = mreq->mr_type;
2949 i->ifindex = mreq->mr_ifindex;
2950 i->alen = mreq->mr_alen;
2951 memcpy(i->addr, mreq->mr_address, i->alen);
2952 i->count = 1;
2953 i->next = po->mclist;
2954 po->mclist = i;
2aeb0b88
WC		 2955 err = packet_dev_mc(dev, i, 1);
2956 if (err) {
2957 po->mclist = i->next;
2958 kfree(i);
2959 }
1da177e4
LT		 2960
2961done:
2962 rtnl_unlock();
2963 return err;
2964}
2965
0fb375fb 2966static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1da177e4
LT		 2967{
2968 struct packet_mclist *ml, **mlp;
2969
2970 rtnl_lock();
2971
2972 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2973 if (ml->ifindex == mreq->mr_ifindex &&
2974 ml->type == mreq->mr_type &&
2975 ml->alen == mreq->mr_alen &&
2976 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2977 if (--ml->count == 0) {
2978 struct net_device *dev;
2979 *mlp = ml->next;
ad959e76
ED		 2980 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2981 if (dev)
1da177e4 2982 packet_dev_mc(dev, ml, -1);
1da177e4
LT		 2983 kfree(ml);
2984 }
2985 rtnl_unlock();
2986 return 0;
2987 }
2988 }
2989 rtnl_unlock();
2990 return -EADDRNOTAVAIL;
2991}
2992
2993static void packet_flush_mclist(struct sock *sk)
2994{
2995 struct packet_sock *po = pkt_sk(sk);
2996 struct packet_mclist *ml;
2997
2998 if (!po->mclist)
2999 return;
3000
3001 rtnl_lock();
3002 while ((ml = po->mclist) != NULL) {
3003 struct net_device *dev;
3004
3005 po->mclist = ml->next;
ad959e76
ED		 3006 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3007 if (dev != NULL)
1da177e4 3008 packet_dev_mc(dev, ml, -1);
1da177e4
LT		 3009 kfree(ml);
3010 }
3011 rtnl_unlock();
3012}
1da177e4
LT		 3013
3014static int
b7058842 3015packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1da177e4
LT		 3016{
3017 struct sock *sk = sock->sk;
8dc41944 3018 struct packet_sock *po = pkt_sk(sk);
1da177e4
LT		 3019 int ret;
3020
3021 if (level != SOL_PACKET)
3022 return -ENOPROTOOPT;
3023
69e3c75f 3024 switch (optname) {
1ce4f28b 3025 case PACKET_ADD_MEMBERSHIP:
1da177e4
LT		 3026 case PACKET_DROP_MEMBERSHIP:
3027 {
0fb375fb
EB		 3028 struct packet_mreq_max mreq;
3029 int len = optlen;
3030 memset(&mreq, 0, sizeof(mreq));
3031 if (len < sizeof(struct packet_mreq))
1da177e4 3032 return -EINVAL;
0fb375fb
EB		 3033 if (len > sizeof(mreq))
3034 len = sizeof(mreq);
40d4e3df 3035 if (copy_from_user(&mreq, optval, len))
1da177e4 3036 return -EFAULT;
0fb375fb
EB		 3037 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3038 return -EINVAL;
1da177e4
LT		 3039 if (optname == PACKET_ADD_MEMBERSHIP)
3040 ret = packet_mc_add(sk, &mreq);
3041 else
3042 ret = packet_mc_drop(sk, &mreq);
3043 return ret;
3044 }
a2efcfa0 3045
1da177e4 3046 case PACKET_RX_RING:
69e3c75f 3047 case PACKET_TX_RING:
1da177e4 3048 {
f6fb8f10 3049 union tpacket_req_u req_u;
3050 int len;
1da177e4 3051
f6fb8f10 3052 switch (po->tp_version) {
3053 case TPACKET_V1:
3054 case TPACKET_V2:
3055 len = sizeof(req_u.req);
3056 break;
3057 case TPACKET_V3:
3058 default:
3059 len = sizeof(req_u.req3);
3060 break;
3061 }
3062 if (optlen < len)
1da177e4 3063 return -EINVAL;
bfd5f4a3
SS		 3064 if (pkt_sk(sk)->has_vnet_hdr)
3065 return -EINVAL;
f6fb8f10 3066 if (copy_from_user(&req_u.req, optval, len))
1da177e4 3067 return -EFAULT;
f6fb8f10 3068 return packet_set_ring(sk, &req_u, 0,
3069 optname == PACKET_TX_RING);
1da177e4
LT		 3070 }
3071 case PACKET_COPY_THRESH:
3072 {
3073 int val;
3074
40d4e3df 3075 if (optlen != sizeof(val))
1da177e4 3076 return -EINVAL;
40d4e3df 3077 if (copy_from_user(&val, optval, sizeof(val)))
1da177e4
LT		 3078 return -EFAULT;
3079
3080 pkt_sk(sk)->copy_thresh = val;
3081 return 0;
3082 }
bbd6ef87
PM		 3083 case PACKET_VERSION:
3084 {
3085 int val;
3086
3087 if (optlen != sizeof(val))
3088 return -EINVAL;
69e3c75f 3089 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
bbd6ef87
PM		 3090 return -EBUSY;
3091 if (copy_from_user(&val, optval, sizeof(val)))
3092 return -EFAULT;
3093 switch (val) {
3094 case TPACKET_V1:
3095 case TPACKET_V2:
f6fb8f10 3096 case TPACKET_V3:
bbd6ef87
PM		 3097 po->tp_version = val;
3098 return 0;
3099 default:
3100 return -EINVAL;
3101 }
3102 }
8913336a
PM		 3103 case PACKET_RESERVE:
3104 {
3105 unsigned int val;
3106
3107 if (optlen != sizeof(val))
3108 return -EINVAL;
69e3c75f 3109 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
8913336a
PM		 3110 return -EBUSY;
3111 if (copy_from_user(&val, optval, sizeof(val)))
3112 return -EFAULT;
3113 po->tp_reserve = val;
3114 return 0;
3115 }
69e3c75f
JB		 3116 case PACKET_LOSS:
3117 {
3118 unsigned int val;
3119
3120 if (optlen != sizeof(val))
3121 return -EINVAL;
3122 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3123 return -EBUSY;
3124 if (copy_from_user(&val, optval, sizeof(val)))
3125 return -EFAULT;
3126 po->tp_loss = !!val;
3127 return 0;
3128 }
8dc41944
HX		 3129 case PACKET_AUXDATA:
3130 {
3131 int val;
3132
3133 if (optlen < sizeof(val))
3134 return -EINVAL;
3135 if (copy_from_user(&val, optval, sizeof(val)))
3136 return -EFAULT;
3137
3138 po->auxdata = !!val;
3139 return 0;
3140 }
80feaacb
PWJ		 3141 case PACKET_ORIGDEV:
3142 {
3143 int val;
3144
3145 if (optlen < sizeof(val))
3146 return -EINVAL;
3147 if (copy_from_user(&val, optval, sizeof(val)))
3148 return -EFAULT;
3149
3150 po->origdev = !!val;
3151 return 0;
3152 }
bfd5f4a3
SS		 3153 case PACKET_VNET_HDR:
3154 {
3155 int val;
3156
3157 if (sock->type != SOCK_RAW)
3158 return -EINVAL;
3159 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3160 return -EBUSY;
3161 if (optlen < sizeof(val))
3162 return -EINVAL;
3163 if (copy_from_user(&val, optval, sizeof(val)))
3164 return -EFAULT;
3165
3166 po->has_vnet_hdr = !!val;
3167 return 0;
3168 }
614f60fa
SM		 3169 case PACKET_TIMESTAMP:
3170 {
3171 int val;
3172
3173 if (optlen != sizeof(val))
3174 return -EINVAL;
3175 if (copy_from_user(&val, optval, sizeof(val)))
3176 return -EFAULT;
3177
3178 po->tp_tstamp = val;
3179 return 0;
3180 }
dc99f600
DM		 3181 case PACKET_FANOUT:
3182 {
3183 int val;
3184
3185 if (optlen != sizeof(val))
3186 return -EINVAL;
3187 if (copy_from_user(&val, optval, sizeof(val)))
3188 return -EFAULT;
3189
3190 return fanout_add(sk, val & 0xffff, val >> 16);
3191 }
1da177e4
LT		 3192 default:
3193 return -ENOPROTOOPT;
3194 }
3195}
3196
3197static int packet_getsockopt(struct socket *sock, int level, int optname,
3198 char __user *optval, int __user *optlen)
3199{
3200 int len;
8dc41944 3201 int val;
1da177e4
LT		 3202 struct sock *sk = sock->sk;
3203 struct packet_sock *po = pkt_sk(sk);
8dc41944
HX		 3204 void *data;
3205 struct tpacket_stats st;
f6fb8f10 3206 union tpacket_stats_u st_u;
1da177e4
LT		 3207
3208 if (level != SOL_PACKET)
3209 return -ENOPROTOOPT;
3210
8ae55f04
KK		 3211 if (get_user(len, optlen))
3212 return -EFAULT;
1da177e4
LT		 3213
3214 if (len < 0)
3215 return -EINVAL;
1ce4f28b 3216
69e3c75f 3217 switch (optname) {
1da177e4 3218 case PACKET_STATISTICS:
f6fb8f10 3219 if (po->tp_version == TPACKET_V3) {
3220 len = sizeof(struct tpacket_stats_v3);
3221 } else {
3222 if (len > sizeof(struct tpacket_stats))
3223 len = sizeof(struct tpacket_stats);
3224 }
1da177e4 3225 spin_lock_bh(&sk->sk_receive_queue.lock);
f6fb8f10 3226 if (po->tp_version == TPACKET_V3) {
3227 memcpy(&st_u.stats3, &po->stats,
3228 sizeof(struct tpacket_stats));
3229 st_u.stats3.tp_freeze_q_cnt =
3230 po->stats_u.stats3.tp_freeze_q_cnt;
3231 st_u.stats3.tp_packets += po->stats.tp_drops;
3232 data = &st_u.stats3;
3233 } else {
3234 st = po->stats;
3235 st.tp_packets += st.tp_drops;
3236 data = &st;
3237 }
1da177e4
LT		 3238 memset(&po->stats, 0, sizeof(st));
3239 spin_unlock_bh(&sk->sk_receive_queue.lock);
8dc41944
HX		 3240 break;
3241 case PACKET_AUXDATA:
3242 if (len > sizeof(int))
3243 len = sizeof(int);
3244 val = po->auxdata;
3245
80feaacb
PWJ		 3246 data = &val;
3247 break;
3248 case PACKET_ORIGDEV:
3249 if (len > sizeof(int))
3250 len = sizeof(int);
3251 val = po->origdev;
3252
bfd5f4a3
SS		 3253 data = &val;
3254 break;
3255 case PACKET_VNET_HDR:
3256 if (len > sizeof(int))
3257 len = sizeof(int);
3258 val = po->has_vnet_hdr;
3259
8dc41944 3260 data = &val;
1da177e4 3261 break;
bbd6ef87
PM		 3262 case PACKET_VERSION:
3263 if (len > sizeof(int))
3264 len = sizeof(int);
3265 val = po->tp_version;
3266 data = &val;
3267 break;
3268 case PACKET_HDRLEN:
3269 if (len > sizeof(int))
3270 len = sizeof(int);
3271 if (copy_from_user(&val, optval, len))
3272 return -EFAULT;
3273 switch (val) {
3274 case TPACKET_V1:
3275 val = sizeof(struct tpacket_hdr);
3276 break;
3277 case TPACKET_V2:
3278 val = sizeof(struct tpacket2_hdr);
3279 break;
f6fb8f10 3280 case TPACKET_V3:
3281 val = sizeof(struct tpacket3_hdr);
3282 break;
bbd6ef87
PM		 3283 default:
3284 return -EINVAL;
3285 }
3286 data = &val;
3287 break;
8913336a
PM		 3288 case PACKET_RESERVE:
3289 if (len > sizeof(unsigned int))
3290 len = sizeof(unsigned int);
3291 val = po->tp_reserve;
3292 data = &val;
3293 break;
69e3c75f
JB		 3294 case PACKET_LOSS:
3295 if (len > sizeof(unsigned int))
3296 len = sizeof(unsigned int);
3297 val = po->tp_loss;
3298 data = &val;
3299 break;
614f60fa
SM		 3300 case PACKET_TIMESTAMP:
3301 if (len > sizeof(int))
3302 len = sizeof(int);
3303 val = po->tp_tstamp;
3304 data = &val;
3305 break;
dc99f600
DM		 3306 case PACKET_FANOUT:
3307 if (len > sizeof(int))
3308 len = sizeof(int);
3309 val = (po->fanout ?
3310 ((u32)po->fanout->id |
3311 ((u32)po->fanout->type << 16)) :
3312 0);
3313 data = &val;
3314 break;
1da177e4
LT		 3315 default:
3316 return -ENOPROTOOPT;
3317 }
3318
8ae55f04
KK		 3319 if (put_user(len, optlen))
3320 return -EFAULT;
8dc41944
HX		 3321 if (copy_to_user(optval, data, len))
3322 return -EFAULT;
8ae55f04 3323 return 0;
1da177e4
LT		 3324}
3325
3326
3327static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3328{
3329 struct sock *sk;
3330 struct hlist_node *node;
ad930650 3331 struct net_device *dev = data;
c346dca1 3332 struct net *net = dev_net(dev);
1da177e4 3333
808f5114 3334 rcu_read_lock();
3335 sk_for_each_rcu(sk, node, &net->packet.sklist) {
1da177e4
LT		 3336 struct packet_sock *po = pkt_sk(sk);
3337
3338 switch (msg) {
3339 case NETDEV_UNREGISTER:
1da177e4
LT		 3340 if (po->mclist)
3341 packet_dev_mclist(dev, po->mclist, -1);
a2efcfa0
DM		 3342 /* fallthrough */
3343
1da177e4
LT		 3344 case NETDEV_DOWN:
3345 if (dev->ifindex == po->ifindex) {
3346 spin_lock(&po->bind_lock);
3347 if (po->running) {
ce06b03e 3348 __unregister_prot_hook(sk, false);
1da177e4
LT		 3349 sk->sk_err = ENETDOWN;
3350 if (!sock_flag(sk, SOCK_DEAD))
3351 sk->sk_error_report(sk);
3352 }
3353 if (msg == NETDEV_UNREGISTER) {
3354 po->ifindex = -1;
160ff18a
BG		 3355 if (po->prot_hook.dev)
3356 dev_put(po->prot_hook.dev);
1da177e4
LT		 3357 po->prot_hook.dev = NULL;
3358 }
3359 spin_unlock(&po->bind_lock);
3360 }
3361 break;
3362 case NETDEV_UP:
808f5114 3363 if (dev->ifindex == po->ifindex) {
3364 spin_lock(&po->bind_lock);
ce06b03e
DM		 3365 if (po->num)
3366 register_prot_hook(sk);
808f5114 3367 spin_unlock(&po->bind_lock);
1da177e4 3368 }
1da177e4
LT		 3369 break;
3370 }
3371 }
808f5114 3372 rcu_read_unlock();
1da177e4
LT		 3373 return NOTIFY_DONE;
3374}
3375
3376
3377static int packet_ioctl(struct socket *sock, unsigned int cmd,
3378 unsigned long arg)
3379{
3380 struct sock *sk = sock->sk;
3381
69e3c75f 3382 switch (cmd) {
40d4e3df
ED		 3383 case SIOCOUTQ:
3384 {
3385 int amount = sk_wmem_alloc_get(sk);
31e6d363 3386
40d4e3df
ED		 3387 return put_user(amount, (int __user *)arg);
3388 }
3389 case SIOCINQ:
3390 {
3391 struct sk_buff *skb;
3392 int amount = 0;
3393
3394 spin_lock_bh(&sk->sk_receive_queue.lock);
3395 skb = skb_peek(&sk->sk_receive_queue);
3396 if (skb)
3397 amount = skb->len;
3398 spin_unlock_bh(&sk->sk_receive_queue.lock);
3399 return put_user(amount, (int __user *)arg);
3400 }
3401 case SIOCGSTAMP:
3402 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3403 case SIOCGSTAMPNS:
3404 return sock_get_timestampns(sk, (struct timespec __user *)arg);
1ce4f28b 3405
1da177e4 3406#ifdef CONFIG_INET
40d4e3df
ED		 3407 case SIOCADDRT:
3408 case SIOCDELRT:
3409 case SIOCDARP:
3410 case SIOCGARP:
3411 case SIOCSARP:
3412 case SIOCGIFADDR:
3413 case SIOCSIFADDR:
3414 case SIOCGIFBRDADDR:
3415 case SIOCSIFBRDADDR:
3416 case SIOCGIFNETMASK:
3417 case SIOCSIFNETMASK:
3418 case SIOCGIFDSTADDR:
3419 case SIOCSIFDSTADDR:
3420 case SIOCSIFFLAGS:
40d4e3df 3421 return inet_dgram_ops.ioctl(sock, cmd, arg);
1da177e4
LT		 3422#endif
3423
40d4e3df
ED		 3424 default:
3425 return -ENOIOCTLCMD;
1da177e4
LT		 3426 }
3427 return 0;
3428}
3429
40d4e3df 3430static unsigned int packet_poll(struct file *file, struct socket *sock,
1da177e4
LT		 3431 poll_table *wait)
3432{
3433 struct sock *sk = sock->sk;
3434 struct packet_sock *po = pkt_sk(sk);
3435 unsigned int mask = datagram_poll(file, sock, wait);
3436
3437 spin_lock_bh(&sk->sk_receive_queue.lock);
69e3c75f 3438 if (po->rx_ring.pg_vec) {
f6fb8f10 3439 if (!packet_previous_rx_frame(po, &po->rx_ring,
3440 TP_STATUS_KERNEL))
1da177e4
LT		 3441 mask |= POLLIN | POLLRDNORM;
3442 }
3443 spin_unlock_bh(&sk->sk_receive_queue.lock);
69e3c75f
JB		 3444 spin_lock_bh(&sk->sk_write_queue.lock);
3445 if (po->tx_ring.pg_vec) {
3446 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3447 mask |= POLLOUT | POLLWRNORM;
3448 }
3449 spin_unlock_bh(&sk->sk_write_queue.lock);
1da177e4
LT		 3450 return mask;
3451}
3452
3453
3454/* Dirty? Well, I still did not learn better way to account
3455 * for user mmaps.
3456 */
3457
3458static void packet_mm_open(struct vm_area_struct *vma)
3459{
3460 struct file *file = vma->vm_file;
40d4e3df 3461 struct socket *sock = file->private_data;
1da177e4 3462 struct sock *sk = sock->sk;
1ce4f28b 3463
1da177e4
LT		 3464 if (sk)
3465 atomic_inc(&pkt_sk(sk)->mapped);
3466}
3467
3468static void packet_mm_close(struct vm_area_struct *vma)
3469{
3470 struct file *file = vma->vm_file;
40d4e3df 3471 struct socket *sock = file->private_data;
1da177e4 3472 struct sock *sk = sock->sk;
1ce4f28b 3473
1da177e4
LT		 3474 if (sk)
3475 atomic_dec(&pkt_sk(sk)->mapped);
3476}
3477
f0f37e2f 3478static const struct vm_operations_struct packet_mmap_ops = {
40d4e3df
ED		 3479 .open = packet_mm_open,
3480 .close = packet_mm_close,
1da177e4
LT		 3481};
3482
0e3125c7
NH		 3483static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3484 unsigned int len)
1da177e4
LT		 3485{
3486 int i;
3487
4ebf0ae2 3488 for (i = 0; i < len; i++) {
0e3125c7 3489 if (likely(pg_vec[i].buffer)) {
c56b4d90 3490 if (is_vmalloc_addr(pg_vec[i].buffer))
0e3125c7
NH		 3491 vfree(pg_vec[i].buffer);
3492 else
3493 free_pages((unsigned long)pg_vec[i].buffer,
3494 order);
3495 pg_vec[i].buffer = NULL;
3496 }
1da177e4
LT		 3497 }
3498 kfree(pg_vec);
3499}
3500
eea49cc9 3501static char *alloc_one_pg_vec_page(unsigned long order)
4ebf0ae2 3502{
0e3125c7
NH		 3503 char *buffer = NULL;
3504 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3505 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3506
3507 buffer = (char *) __get_free_pages(gfp_flags, order);
3508
3509 if (buffer)
3510 return buffer;
3511
3512 /*
3513 * __get_free_pages failed, fall back to vmalloc
3514 */
bbce5a59 3515 buffer = vzalloc((1 << order) * PAGE_SIZE);
719bfeaa 3516
0e3125c7
NH		 3517 if (buffer)
3518 return buffer;
3519
3520 /*
3521 * vmalloc failed, lets dig into swap here
3522 */
0e3125c7
NH		 3523 gfp_flags &= ~__GFP_NORETRY;
3524 buffer = (char *)__get_free_pages(gfp_flags, order);
3525 if (buffer)
3526 return buffer;
3527
3528 /*
3529 * complete and utter failure
3530 */
3531 return NULL;
4ebf0ae2
DM		 3532}
3533
0e3125c7 3534static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4ebf0ae2
DM		 3535{
3536 unsigned int block_nr = req->tp_block_nr;
0e3125c7 3537 struct pgv *pg_vec;
4ebf0ae2
DM		 3538 int i;
3539
0e3125c7 3540 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4ebf0ae2
DM		 3541 if (unlikely(!pg_vec))
3542 goto out;
3543
3544 for (i = 0; i < block_nr; i++) {
c56b4d90 3545 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
0e3125c7 3546 if (unlikely(!pg_vec[i].buffer))
4ebf0ae2
DM		 3547 goto out_free_pgvec;
3548 }
3549
3550out:
3551 return pg_vec;
3552
3553out_free_pgvec:
3554 free_pg_vec(pg_vec, order, block_nr);
3555 pg_vec = NULL;
3556 goto out;
3557}
1da177e4 3558
f6fb8f10 3559static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
69e3c75f 3560 int closing, int tx_ring)
1da177e4 3561{
0e3125c7 3562 struct pgv *pg_vec = NULL;
1da177e4 3563 struct packet_sock *po = pkt_sk(sk);
0e11c91e 3564 int was_running, order = 0;
69e3c75f
JB		 3565 struct packet_ring_buffer *rb;
3566 struct sk_buff_head *rb_queue;
0e11c91e 3567 __be16 num;
f6fb8f10 3568 int err = -EINVAL;
3569 /* Added to avoid minimal code churn */
3570 struct tpacket_req *req = &req_u->req;
3571
3572 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3573 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3574 WARN(1, "Tx-ring is not supported.\n");
3575 goto out;
3576 }
1ce4f28b 3577
69e3c75f
JB		 3578 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3579 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
1da177e4 3580
69e3c75f
JB		 3581 err = -EBUSY;
3582 if (!closing) {
3583 if (atomic_read(&po->mapped))
3584 goto out;
3585 if (atomic_read(&rb->pending))
3586 goto out;
3587 }
1da177e4 3588
69e3c75f
JB		 3589 if (req->tp_block_nr) {
3590 /* Sanity tests and some calculations */
3591 err = -EBUSY;
3592 if (unlikely(rb->pg_vec))
3593 goto out;
1da177e4 3594
bbd6ef87
PM		 3595 switch (po->tp_version) {
3596 case TPACKET_V1:
3597 po->tp_hdrlen = TPACKET_HDRLEN;
3598 break;
3599 case TPACKET_V2:
3600 po->tp_hdrlen = TPACKET2_HDRLEN;
3601 break;
f6fb8f10 3602 case TPACKET_V3:
3603 po->tp_hdrlen = TPACKET3_HDRLEN;
3604 break;
bbd6ef87
PM		 3605 }
3606
69e3c75f 3607 err = -EINVAL;
4ebf0ae2 3608 if (unlikely((int)req->tp_block_size <= 0))
69e3c75f 3609 goto out;
4ebf0ae2 3610 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
69e3c75f 3611 goto out;
8913336a 3612 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
69e3c75f
JB		 3613 po->tp_reserve))
3614 goto out;
4ebf0ae2 3615 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
69e3c75f 3616 goto out;
1da177e4 3617
69e3c75f
JB		 3618 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3619 if (unlikely(rb->frames_per_block <= 0))
3620 goto out;
3621 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3622 req->tp_frame_nr))
3623 goto out;
1da177e4
LT		 3624
3625 err = -ENOMEM;
4ebf0ae2
DM		 3626 order = get_order(req->tp_block_size);
3627 pg_vec = alloc_pg_vec(req, order);
3628 if (unlikely(!pg_vec))
1da177e4 3629 goto out;
f6fb8f10 3630 switch (po->tp_version) {
3631 case TPACKET_V3:
3632 /* Transmit path is not supported. We checked
3633 * it above but just being paranoid
3634 */
3635 if (!tx_ring)
3636 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3637 break;
3638 default:
3639 break;
3640 }
69e3c75f
JB		 3641 }
3642 /* Done */
3643 else {
3644 err = -EINVAL;
4ebf0ae2 3645 if (unlikely(req->tp_frame_nr))
69e3c75f 3646 goto out;
1da177e4
LT		 3647 }
3648
3649 lock_sock(sk);
3650
3651 /* Detach socket from network */
3652 spin_lock(&po->bind_lock);
3653 was_running = po->running;
3654 num = po->num;
3655 if (was_running) {
1da177e4 3656 po->num = 0;
ce06b03e 3657 __unregister_prot_hook(sk, false);
1da177e4
LT		 3658 }
3659 spin_unlock(&po->bind_lock);
1ce4f28b 3660
1da177e4
LT		 3661 synchronize_net();
3662
3663 err = -EBUSY;
905db440 3664 mutex_lock(&po->pg_vec_lock);
1da177e4
LT		 3665 if (closing || atomic_read(&po->mapped) == 0) {
3666 err = 0;
69e3c75f 3667 spin_lock_bh(&rb_queue->lock);
c053fd96 3668 swap(rb->pg_vec, pg_vec);
69e3c75f
JB		 3669 rb->frame_max = (req->tp_frame_nr - 1);
3670 rb->head = 0;
3671 rb->frame_size = req->tp_frame_size;
3672 spin_unlock_bh(&rb_queue->lock);
3673
c053fd96
CG		 3674 swap(rb->pg_vec_order, order);
3675 swap(rb->pg_vec_len, req->tp_block_nr);
69e3c75f
JB		 3676
3677 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3678 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3679 tpacket_rcv : packet_rcv;
3680 skb_queue_purge(rb_queue);
1da177e4 3681 if (atomic_read(&po->mapped))
40d4e3df
ED		 3682 pr_err("packet_mmap: vma is busy: %d\n",
3683 atomic_read(&po->mapped));
1da177e4 3684 }
905db440 3685 mutex_unlock(&po->pg_vec_lock);
1da177e4
LT		 3686
3687 spin_lock(&po->bind_lock);
ce06b03e 3688 if (was_running) {
1da177e4 3689 po->num = num;
ce06b03e 3690 register_prot_hook(sk);
1da177e4
LT		 3691 }
3692 spin_unlock(&po->bind_lock);
f6fb8f10 3693 if (closing && (po->tp_version > TPACKET_V2)) {
3694 /* Because we don't support block-based V3 on tx-ring */
3695 if (!tx_ring)
3696 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3697 }
1da177e4
LT		 3698 release_sock(sk);
3699
1da177e4
LT		 3700 if (pg_vec)
3701 free_pg_vec(pg_vec, order, req->tp_block_nr);
3702out:
3703 return err;
3704}
3705
69e3c75f
JB		 3706static int packet_mmap(struct file *file, struct socket *sock,
3707 struct vm_area_struct *vma)
1da177e4
LT		 3708{
3709 struct sock *sk = sock->sk;
3710 struct packet_sock *po = pkt_sk(sk);
69e3c75f
JB		 3711 unsigned long size, expected_size;
3712 struct packet_ring_buffer *rb;
1da177e4
LT		 3713 unsigned long start;
3714 int err = -EINVAL;
3715 int i;
3716
3717 if (vma->vm_pgoff)
3718 return -EINVAL;
3719
905db440 3720 mutex_lock(&po->pg_vec_lock);
69e3c75f
JB		 3721
3722 expected_size = 0;
3723 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3724 if (rb->pg_vec) {
3725 expected_size += rb->pg_vec_len
3726 * rb->pg_vec_pages
3727 * PAGE_SIZE;
3728 }
3729 }
3730
3731 if (expected_size == 0)
1da177e4 3732 goto out;
69e3c75f
JB		 3733
3734 size = vma->vm_end - vma->vm_start;
3735 if (size != expected_size)
1da177e4
LT		 3736 goto out;
3737
1da177e4 3738 start = vma->vm_start;
69e3c75f
JB		 3739 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3740 if (rb->pg_vec == NULL)
3741 continue;
3742
3743 for (i = 0; i < rb->pg_vec_len; i++) {
0e3125c7
NH		 3744 struct page *page;
3745 void *kaddr = rb->pg_vec[i].buffer;
69e3c75f
JB		 3746 int pg_num;
3747
c56b4d90
CG		 3748 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3749 page = pgv_to_page(kaddr);
69e3c75f
JB		 3750 err = vm_insert_page(vma, start, page);
3751 if (unlikely(err))
3752 goto out;
3753 start += PAGE_SIZE;
0e3125c7 3754 kaddr += PAGE_SIZE;
69e3c75f 3755 }
4ebf0ae2 3756 }
1da177e4 3757 }
69e3c75f 3758
4ebf0ae2 3759 atomic_inc(&po->mapped);
1da177e4
LT		 3760 vma->vm_ops = &packet_mmap_ops;
3761 err = 0;
3762
3763out:
905db440 3764 mutex_unlock(&po->pg_vec_lock);
1da177e4
LT		 3765 return err;
3766}
1da177e4 3767
90ddc4f0 3768static const struct proto_ops packet_ops_spkt = {
1da177e4
LT		 3769 .family = PF_PACKET,
3770 .owner = THIS_MODULE,
3771 .release = packet_release,
3772 .bind = packet_bind_spkt,
3773 .connect = sock_no_connect,
3774 .socketpair = sock_no_socketpair,
3775 .accept = sock_no_accept,
3776 .getname = packet_getname_spkt,
3777 .poll = datagram_poll,
3778 .ioctl = packet_ioctl,
3779 .listen = sock_no_listen,
3780 .shutdown = sock_no_shutdown,
3781 .setsockopt = sock_no_setsockopt,
3782 .getsockopt = sock_no_getsockopt,
3783 .sendmsg = packet_sendmsg_spkt,
3784 .recvmsg = packet_recvmsg,
3785 .mmap = sock_no_mmap,
3786 .sendpage = sock_no_sendpage,
3787};
1da177e4 3788
90ddc4f0 3789static const struct proto_ops packet_ops = {
1da177e4
LT		 3790 .family = PF_PACKET,
3791 .owner = THIS_MODULE,
3792 .release = packet_release,
3793 .bind = packet_bind,
3794 .connect = sock_no_connect,
3795 .socketpair = sock_no_socketpair,
3796 .accept = sock_no_accept,
1ce4f28b 3797 .getname = packet_getname,
1da177e4
LT		 3798 .poll = packet_poll,
3799 .ioctl = packet_ioctl,
3800 .listen = sock_no_listen,
3801 .shutdown = sock_no_shutdown,
3802 .setsockopt = packet_setsockopt,
3803 .getsockopt = packet_getsockopt,
3804 .sendmsg = packet_sendmsg,
3805 .recvmsg = packet_recvmsg,
3806 .mmap = packet_mmap,
3807 .sendpage = sock_no_sendpage,
3808};
3809
ec1b4cf7 3810static const struct net_proto_family packet_family_ops = {
1da177e4
LT		 3811 .family = PF_PACKET,
3812 .create = packet_create,
3813 .owner = THIS_MODULE,
3814};
3815
3816static struct notifier_block packet_netdev_notifier = {
40d4e3df 3817 .notifier_call = packet_notifier,
1da177e4
LT		 3818};
3819
3820#ifdef CONFIG_PROC_FS
1da177e4
LT		 3821
3822static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
808f5114 3823 __acquires(RCU)
1da177e4 3824{
e372c414 3825 struct net *net = seq_file_net(seq);
808f5114 3826
3827 rcu_read_lock();
3828 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
1da177e4
LT		 3829}
3830
3831static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3832{
1bf40954 3833 struct net *net = seq_file_net(seq);
808f5114 3834 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
1da177e4
LT		 3835}
3836
3837static void packet_seq_stop(struct seq_file *seq, void *v)
808f5114 3838 __releases(RCU)
1da177e4 3839{
808f5114 3840 rcu_read_unlock();
1da177e4
LT		 3841}
3842
1ce4f28b 3843static int packet_seq_show(struct seq_file *seq, void *v)
1da177e4
LT		 3844{
3845 if (v == SEQ_START_TOKEN)
3846 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
3847 else {
b7ceabd9 3848 struct sock *s = sk_entry(v);
1da177e4
LT		 3849 const struct packet_sock *po = pkt_sk(s);
3850
3851 seq_printf(seq,
71338aa7 3852 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
1da177e4
LT		 3853 s,
3854 atomic_read(&s->sk_refcnt),
3855 s->sk_type,
3856 ntohs(po->num),
3857 po->ifindex,
3858 po->running,
3859 atomic_read(&s->sk_rmem_alloc),
3860 sock_i_uid(s),
40d4e3df 3861 sock_i_ino(s));
1da177e4
LT		 3862 }
3863
3864 return 0;
3865}
3866
56b3d975 3867static const struct seq_operations packet_seq_ops = {
1da177e4
LT		 3868 .start = packet_seq_start,
3869 .next = packet_seq_next,
3870 .stop = packet_seq_stop,
3871 .show = packet_seq_show,
3872};
3873
3874static int packet_seq_open(struct inode *inode, struct file *file)
3875{
e372c414
DL		 3876 return seq_open_net(inode, file, &packet_seq_ops,
3877 sizeof(struct seq_net_private));
1da177e4
LT		 3878}
3879
da7071d7 3880static const struct file_operations packet_seq_fops = {
1da177e4
LT		 3881 .owner = THIS_MODULE,
3882 .open = packet_seq_open,
3883 .read = seq_read,
3884 .llseek = seq_lseek,
e372c414 3885 .release = seq_release_net,
1da177e4
LT		 3886};
3887
3888#endif
3889
2c8c1e72 3890static int __net_init packet_net_init(struct net *net)
d12d01d6 3891{
808f5114 3892 spin_lock_init(&net->packet.sklist_lock);
2aaef4e4 3893 INIT_HLIST_HEAD(&net->packet.sklist);
d12d01d6
DL		 3894
3895 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3896 return -ENOMEM;
3897
3898 return 0;
3899}
3900
2c8c1e72 3901static void __net_exit packet_net_exit(struct net *net)
d12d01d6
DL		 3902{
3903 proc_net_remove(net, "packet");
3904}
3905
3906static struct pernet_operations packet_net_ops = {
3907 .init = packet_net_init,
3908 .exit = packet_net_exit,
3909};
3910
3911
1da177e4
LT		 3912static void __exit packet_exit(void)
3913{
1da177e4 3914 unregister_netdevice_notifier(&packet_netdev_notifier);
d12d01d6 3915 unregister_pernet_subsys(&packet_net_ops);
1da177e4
LT		 3916 sock_unregister(PF_PACKET);
3917 proto_unregister(&packet_proto);
3918}
3919
3920static int __init packet_init(void)
3921{
3922 int rc = proto_register(&packet_proto, 0);
3923
3924 if (rc != 0)
3925 goto out;
3926
3927 sock_register(&packet_family_ops);
d12d01d6 3928 register_pernet_subsys(&packet_net_ops);
1da177e4 3929 register_netdevice_notifier(&packet_netdev_notifier);
1da177e4
LT		 3930out:
3931 return rc;
3932}
3933
3934module_init(packet_init);
3935module_exit(packet_exit);
3936MODULE_LICENSE("GPL");
3937MODULE_ALIAS_NETPROTO(PF_PACKET);
kernel source for telekom puls (alcatel/mediatek)