blob: 3e2462760413ee5871cbd3061f57be83df6c1aa1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PACKET - implements raw packet sockets.
7 *
8 * Version: $Id: af_packet.c,v 1.61 2002/02/08 03:57:19 davem Exp $
9 *
Jesper Juhl02c30a82005-05-05 16:16:16 -070010 * Authors: Ross Biro
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 *
14 * Fixes:
15 * Alan Cox : verify_area() now used correctly
16 * Alan Cox : new skbuff lists, look ma no backlogs!
17 * Alan Cox : tidied skbuff lists.
18 * Alan Cox : Now uses generic datagram routines I
19 * added. Also fixed the peek/read crash
20 * from all old Linux datagram code.
21 * Alan Cox : Uses the improved datagram code.
22 * Alan Cox : Added NULL's for socket options.
23 * Alan Cox : Re-commented the code.
24 * Alan Cox : Use new kernel side addressing
25 * Rob Janssen : Correct MTU usage.
26 * Dave Platt : Counter leaks caused by incorrect
27 * interrupt locking and some slightly
28 * dubious gcc output. Can you read
29 * compiler: it said _VOLATILE_
30 * Richard Kooijman : Timestamp fixes.
31 * Alan Cox : New buffers. Use sk->mac.raw.
32 * Alan Cox : sendmsg/recvmsg support.
33 * Alan Cox : Protocol setting support
34 * Alexey Kuznetsov : Untied from IPv4 stack.
35 * Cyrus Durgin : Fixed kerneld for kmod.
36 * Michal Ostrowski : Module initialization cleanup.
37 * Ulises Alonso : Frame number limit removal and
38 * packet_set_ring memory leak.
Eric W. Biederman0fb375f2005-09-21 00:11:37 -070039 * Eric Biederman : Allow for > 8 byte hardware addresses.
40 * The convention is that longer addresses
41 * will simply extend the hardware address
42 * byte arrays at the end of sockaddr_ll
43 * and packet_mreq.
Linus Torvalds1da177e2005-04-16 15:20:36 -070044 *
45 * This program is free software; you can redistribute it and/or
46 * modify it under the terms of the GNU General Public License
47 * as published by the Free Software Foundation; either version
48 * 2 of the License, or (at your option) any later version.
49 *
50 */
51
52#include <linux/config.h>
53#include <linux/types.h>
54#include <linux/sched.h>
55#include <linux/mm.h>
56#include <linux/fcntl.h>
57#include <linux/socket.h>
58#include <linux/in.h>
59#include <linux/inet.h>
60#include <linux/netdevice.h>
61#include <linux/if_packet.h>
62#include <linux/wireless.h>
63#include <linux/kmod.h>
64#include <net/ip.h>
65#include <net/protocol.h>
66#include <linux/skbuff.h>
67#include <net/sock.h>
68#include <linux/errno.h>
69#include <linux/timer.h>
70#include <asm/system.h>
71#include <asm/uaccess.h>
72#include <asm/ioctls.h>
73#include <asm/page.h>
74#include <asm/io.h>
75#include <linux/proc_fs.h>
76#include <linux/seq_file.h>
77#include <linux/poll.h>
78#include <linux/module.h>
79#include <linux/init.h>
80
81#ifdef CONFIG_INET
82#include <net/inet_common.h>
83#endif
84
85#define CONFIG_SOCK_PACKET 1
86
87/*
88 Proposed replacement for SIOC{ADD,DEL}MULTI and
89 IFF_PROMISC, IFF_ALLMULTI flags.
90
91 It is more expensive, but I believe,
92 it is really correct solution: reentereble, safe and fault tolerant.
93
94 IFF_PROMISC/IFF_ALLMULTI/SIOC{ADD/DEL}MULTI are faked by keeping
95 reference count and global flag, so that real status is
96 (gflag|(count != 0)), so that we can use obsolete faulty interface
97 not harming clever users.
98 */
99#define CONFIG_PACKET_MULTICAST 1
100
101/*
102 Assumptions:
103 - if device has no dev->hard_header routine, it adds and removes ll header
104 inside itself. In this case ll header is invisible outside of device,
105 but higher levels still should reserve dev->hard_header_len.
106 Some devices are enough clever to reallocate skb, when header
107 will not fit to reserved space (tunnel), another ones are silly
108 (PPP).
109 - packet socket receives packets with pulled ll header,
110 so that SOCK_RAW should push it back.
111
112On receive:
113-----------
114
115Incoming, dev->hard_header!=NULL
116 mac.raw -> ll header
117 data -> data
118
119Outgoing, dev->hard_header!=NULL
120 mac.raw -> ll header
121 data -> ll header
122
123Incoming, dev->hard_header==NULL
124 mac.raw -> UNKNOWN position. It is very likely, that it points to ll header.
125 PPP makes it, that is wrong, because introduce assymetry
126 between rx and tx paths.
127 data -> data
128
129Outgoing, dev->hard_header==NULL
130 mac.raw -> data. ll header is still not built!
131 data -> data
132
133Resume
134 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
135
136
137On transmit:
138------------
139
140dev->hard_header != NULL
141 mac.raw -> ll header
142 data -> ll header
143
144dev->hard_header == NULL (ll header is added by device, we cannot control it)
145 mac.raw -> data
146 data -> data
147
148 We should set nh.raw on output to correct posistion,
149 packet classifier depends on it.
150 */
151
152/* List of all packet sockets. */
153static HLIST_HEAD(packet_sklist);
154static DEFINE_RWLOCK(packet_sklist_lock);
155
156static atomic_t packet_socks_nr;
157
158
159/* Private packet socket structures. */
160
161#ifdef CONFIG_PACKET_MULTICAST
162struct packet_mclist
163{
164 struct packet_mclist *next;
165 int ifindex;
166 int count;
167 unsigned short type;
168 unsigned short alen;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -0700169 unsigned char addr[MAX_ADDR_LEN];
170};
171/* identical to struct packet_mreq except it has
172 * a longer address field.
173 */
174struct packet_mreq_max
175{
176 int mr_ifindex;
177 unsigned short mr_type;
178 unsigned short mr_alen;
179 unsigned char mr_address[MAX_ADDR_LEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180};
181#endif
182#ifdef CONFIG_PACKET_MMAP
183static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
184#endif
185
186static void packet_flush_mclist(struct sock *sk);
187
188struct packet_sock {
189 /* struct sock has to be the first member of packet_sock */
190 struct sock sk;
191 struct tpacket_stats stats;
192#ifdef CONFIG_PACKET_MMAP
193 char * *pg_vec;
194 unsigned int head;
195 unsigned int frames_per_block;
196 unsigned int frame_size;
197 unsigned int frame_max;
198 int copy_thresh;
199#endif
200 struct packet_type prot_hook;
201 spinlock_t bind_lock;
202 char running; /* prot_hook is attached*/
203 int ifindex; /* bound device */
204 unsigned short num;
205#ifdef CONFIG_PACKET_MULTICAST
206 struct packet_mclist *mclist;
207#endif
208#ifdef CONFIG_PACKET_MMAP
209 atomic_t mapped;
210 unsigned int pg_vec_order;
211 unsigned int pg_vec_pages;
212 unsigned int pg_vec_len;
213#endif
214};
215
216#ifdef CONFIG_PACKET_MMAP
217
218static inline char *packet_lookup_frame(struct packet_sock *po, unsigned int position)
219{
220 unsigned int pg_vec_pos, frame_offset;
221 char *frame;
222
223 pg_vec_pos = position / po->frames_per_block;
224 frame_offset = position % po->frames_per_block;
225
226 frame = po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size);
227
228 return frame;
229}
230#endif
231
232static inline struct packet_sock *pkt_sk(struct sock *sk)
233{
234 return (struct packet_sock *)sk;
235}
236
237static void packet_sock_destruct(struct sock *sk)
238{
239 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
240 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
241
242 if (!sock_flag(sk, SOCK_DEAD)) {
243 printk("Attempt to release alive packet socket: %p\n", sk);
244 return;
245 }
246
247 atomic_dec(&packet_socks_nr);
248#ifdef PACKET_REFCNT_DEBUG
249 printk(KERN_DEBUG "PACKET socket %p is free, %d are alive\n", sk, atomic_read(&packet_socks_nr));
250#endif
251}
252
253
254static struct proto_ops packet_ops;
255
256#ifdef CONFIG_SOCK_PACKET
257static struct proto_ops packet_ops_spkt;
258
David S. Millerf2ccd8f2005-08-09 19:34:12 -0700259static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
261 struct sock *sk;
262 struct sockaddr_pkt *spkt;
263
264 /*
265 * When we registered the protocol we saved the socket in the data
266 * field for just this event.
267 */
268
269 sk = pt->af_packet_priv;
270
271 /*
272 * Yank back the headers [hope the device set this
273 * right or kerboom...]
274 *
275 * Incoming packets have ll header pulled,
276 * push it back.
277 *
278 * For outgoing ones skb->data == skb->mac.raw
279 * so that this procedure is noop.
280 */
281
282 if (skb->pkt_type == PACKET_LOOPBACK)
283 goto out;
284
285 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
286 goto oom;
287
288 /* drop any routing info */
289 dst_release(skb->dst);
290 skb->dst = NULL;
291
Phil Oester84531c22005-07-12 11:57:52 -0700292 /* drop conntrack reference */
293 nf_reset(skb);
294
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 spkt = (struct sockaddr_pkt*)skb->cb;
296
297 skb_push(skb, skb->data-skb->mac.raw);
298
299 /*
300 * The SOCK_PACKET socket receives _all_ frames.
301 */
302
303 spkt->spkt_family = dev->type;
304 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
305 spkt->spkt_protocol = skb->protocol;
306
307 /*
308 * Charge the memory to the socket. This is done specifically
309 * to prevent sockets using all the memory up.
310 */
311
312 if (sock_queue_rcv_skb(sk,skb) == 0)
313 return 0;
314
315out:
316 kfree_skb(skb);
317oom:
318 return 0;
319}
320
321
322/*
323 * Output a raw packet to a device layer. This bypasses all the other
324 * protocol layers and you must therefore supply it with a complete frame
325 */
326
327static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
328 struct msghdr *msg, size_t len)
329{
330 struct sock *sk = sock->sk;
331 struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
332 struct sk_buff *skb;
333 struct net_device *dev;
334 unsigned short proto=0;
335 int err;
336
337 /*
338 * Get and verify the address.
339 */
340
341 if (saddr)
342 {
343 if (msg->msg_namelen < sizeof(struct sockaddr))
344 return(-EINVAL);
345 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
346 proto=saddr->spkt_protocol;
347 }
348 else
349 return(-ENOTCONN); /* SOCK_PACKET must be sent giving an address */
350
351 /*
352 * Find the device first to size check it
353 */
354
355 saddr->spkt_device[13] = 0;
356 dev = dev_get_by_name(saddr->spkt_device);
357 err = -ENODEV;
358 if (dev == NULL)
359 goto out_unlock;
360
361 /*
362 * You may not queue a frame bigger than the mtu. This is the lowest level
363 * raw protocol and you must do your own fragmentation at this level.
364 */
365
366 err = -EMSGSIZE;
367 if(len>dev->mtu+dev->hard_header_len)
368 goto out_unlock;
369
370 err = -ENOBUFS;
371 skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
372
373 /*
374 * If the write buffer is full, then tough. At this level the user gets to
375 * deal with the problem - do your own algorithmic backoffs. That's far
376 * more flexible.
377 */
378
379 if (skb == NULL)
380 goto out_unlock;
381
382 /*
383 * Fill it in
384 */
385
386 /* FIXME: Save some space for broken drivers that write a
387 * hard header at transmission time by themselves. PPP is the
388 * notable one here. This should really be fixed at the driver level.
389 */
390 skb_reserve(skb, LL_RESERVED_SPACE(dev));
391 skb->nh.raw = skb->data;
392
393 /* Try to align data part correctly */
394 if (dev->hard_header) {
395 skb->data -= dev->hard_header_len;
396 skb->tail -= dev->hard_header_len;
397 if (len < dev->hard_header_len)
398 skb->nh.raw = skb->data;
399 }
400
401 /* Returns -EFAULT on error */
402 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
403 skb->protocol = proto;
404 skb->dev = dev;
405 skb->priority = sk->sk_priority;
406 if (err)
407 goto out_free;
408
409 err = -ENETDOWN;
410 if (!(dev->flags & IFF_UP))
411 goto out_free;
412
413 /*
414 * Now send it
415 */
416
417 dev_queue_xmit(skb);
418 dev_put(dev);
419 return(len);
420
421out_free:
422 kfree_skb(skb);
423out_unlock:
424 if (dev)
425 dev_put(dev);
426 return err;
427}
428#endif
429
430static inline unsigned run_filter(struct sk_buff *skb, struct sock *sk, unsigned res)
431{
432 struct sk_filter *filter;
433
434 bh_lock_sock(sk);
435 filter = sk->sk_filter;
436 /*
437 * Our caller already checked that filter != NULL but we need to
438 * verify that under bh_lock_sock() to be safe
439 */
440 if (likely(filter != NULL))
441 res = sk_run_filter(skb, filter->insns, filter->len);
442 bh_unlock_sock(sk);
443
444 return res;
445}
446
447/*
448 This function makes lazy skb cloning in hope that most of packets
449 are discarded by BPF.
450
451 Note tricky part: we DO mangle shared skb! skb->data, skb->len
452 and skb->cb are mangled. It works because (and until) packets
453 falling here are owned by current CPU. Output packets are cloned
454 by dev_queue_xmit_nit(), input packets are processed by net_bh
455 sequencially, so that if we return skb to original state on exit,
456 we will not harm anyone.
457 */
458
David S. Millerf2ccd8f2005-08-09 19:34:12 -0700459static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460{
461 struct sock *sk;
462 struct sockaddr_ll *sll;
463 struct packet_sock *po;
464 u8 * skb_head = skb->data;
465 int skb_len = skb->len;
466 unsigned snaplen;
467
468 if (skb->pkt_type == PACKET_LOOPBACK)
469 goto drop;
470
471 sk = pt->af_packet_priv;
472 po = pkt_sk(sk);
473
474 skb->dev = dev;
475
476 if (dev->hard_header) {
477 /* The device has an explicit notion of ll header,
478 exported to higher levels.
479
480 Otherwise, the device hides datails of it frame
481 structure, so that corresponding packet head
482 never delivered to user.
483 */
484 if (sk->sk_type != SOCK_DGRAM)
485 skb_push(skb, skb->data - skb->mac.raw);
486 else if (skb->pkt_type == PACKET_OUTGOING) {
487 /* Special case: outgoing packets have ll header at head */
488 skb_pull(skb, skb->nh.raw - skb->data);
489 }
490 }
491
492 snaplen = skb->len;
493
494 if (sk->sk_filter) {
495 unsigned res = run_filter(skb, sk, snaplen);
496 if (res == 0)
497 goto drop_n_restore;
498 if (snaplen > res)
499 snaplen = res;
500 }
501
502 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
503 (unsigned)sk->sk_rcvbuf)
504 goto drop_n_acct;
505
506 if (skb_shared(skb)) {
507 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
508 if (nskb == NULL)
509 goto drop_n_acct;
510
511 if (skb_head != skb->data) {
512 skb->data = skb_head;
513 skb->len = skb_len;
514 }
515 kfree_skb(skb);
516 skb = nskb;
517 }
518
519 sll = (struct sockaddr_ll*)skb->cb;
520 sll->sll_family = AF_PACKET;
521 sll->sll_hatype = dev->type;
522 sll->sll_protocol = skb->protocol;
523 sll->sll_pkttype = skb->pkt_type;
524 sll->sll_ifindex = dev->ifindex;
525 sll->sll_halen = 0;
526
527 if (dev->hard_header_parse)
528 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
529
530 if (pskb_trim(skb, snaplen))
531 goto drop_n_acct;
532
533 skb_set_owner_r(skb, sk);
534 skb->dev = NULL;
535 dst_release(skb->dst);
536 skb->dst = NULL;
537
Phil Oester84531c22005-07-12 11:57:52 -0700538 /* drop conntrack reference */
539 nf_reset(skb);
540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 spin_lock(&sk->sk_receive_queue.lock);
542 po->stats.tp_packets++;
543 __skb_queue_tail(&sk->sk_receive_queue, skb);
544 spin_unlock(&sk->sk_receive_queue.lock);
545 sk->sk_data_ready(sk, skb->len);
546 return 0;
547
548drop_n_acct:
549 spin_lock(&sk->sk_receive_queue.lock);
550 po->stats.tp_drops++;
551 spin_unlock(&sk->sk_receive_queue.lock);
552
553drop_n_restore:
554 if (skb_head != skb->data && skb_shared(skb)) {
555 skb->data = skb_head;
556 skb->len = skb_len;
557 }
558drop:
559 kfree_skb(skb);
560 return 0;
561}
562
563#ifdef CONFIG_PACKET_MMAP
David S. Millerf2ccd8f2005-08-09 19:34:12 -0700564static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565{
566 struct sock *sk;
567 struct packet_sock *po;
568 struct sockaddr_ll *sll;
569 struct tpacket_hdr *h;
570 u8 * skb_head = skb->data;
571 int skb_len = skb->len;
572 unsigned snaplen;
573 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
574 unsigned short macoff, netoff;
575 struct sk_buff *copy_skb = NULL;
576
577 if (skb->pkt_type == PACKET_LOOPBACK)
578 goto drop;
579
580 sk = pt->af_packet_priv;
581 po = pkt_sk(sk);
582
583 if (dev->hard_header) {
584 if (sk->sk_type != SOCK_DGRAM)
585 skb_push(skb, skb->data - skb->mac.raw);
586 else if (skb->pkt_type == PACKET_OUTGOING) {
587 /* Special case: outgoing packets have ll header at head */
588 skb_pull(skb, skb->nh.raw - skb->data);
589 if (skb->ip_summed == CHECKSUM_HW)
590 status |= TP_STATUS_CSUMNOTREADY;
591 }
592 }
593
594 snaplen = skb->len;
595
596 if (sk->sk_filter) {
597 unsigned res = run_filter(skb, sk, snaplen);
598 if (res == 0)
599 goto drop_n_restore;
600 if (snaplen > res)
601 snaplen = res;
602 }
603
604 if (sk->sk_type == SOCK_DGRAM) {
605 macoff = netoff = TPACKET_ALIGN(TPACKET_HDRLEN) + 16;
606 } else {
607 unsigned maclen = skb->nh.raw - skb->data;
608 netoff = TPACKET_ALIGN(TPACKET_HDRLEN + (maclen < 16 ? 16 : maclen));
609 macoff = netoff - maclen;
610 }
611
612 if (macoff + snaplen > po->frame_size) {
613 if (po->copy_thresh &&
614 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
615 (unsigned)sk->sk_rcvbuf) {
616 if (skb_shared(skb)) {
617 copy_skb = skb_clone(skb, GFP_ATOMIC);
618 } else {
619 copy_skb = skb_get(skb);
620 skb_head = skb->data;
621 }
622 if (copy_skb)
623 skb_set_owner_r(copy_skb, sk);
624 }
625 snaplen = po->frame_size - macoff;
626 if ((int)snaplen < 0)
627 snaplen = 0;
628 }
629 if (snaplen > skb->len-skb->data_len)
630 snaplen = skb->len-skb->data_len;
631
632 spin_lock(&sk->sk_receive_queue.lock);
633 h = (struct tpacket_hdr *)packet_lookup_frame(po, po->head);
634
635 if (h->tp_status)
636 goto ring_is_full;
637 po->head = po->head != po->frame_max ? po->head+1 : 0;
638 po->stats.tp_packets++;
639 if (copy_skb) {
640 status |= TP_STATUS_COPY;
641 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
642 }
643 if (!po->stats.tp_drops)
644 status &= ~TP_STATUS_LOSING;
645 spin_unlock(&sk->sk_receive_queue.lock);
646
647 memcpy((u8*)h + macoff, skb->data, snaplen);
648
649 h->tp_len = skb->len;
650 h->tp_snaplen = snaplen;
651 h->tp_mac = macoff;
652 h->tp_net = netoff;
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700653 if (skb->tstamp.off_sec == 0) {
654 __net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 sock_enable_timestamp(sk);
656 }
Herbert Xu325ed822005-10-03 13:57:23 -0700657 h->tp_sec = skb->tstamp.off_sec;
658 h->tp_usec = skb->tstamp.off_usec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 sll = (struct sockaddr_ll*)((u8*)h + TPACKET_ALIGN(sizeof(*h)));
661 sll->sll_halen = 0;
662 if (dev->hard_header_parse)
663 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
664 sll->sll_family = AF_PACKET;
665 sll->sll_hatype = dev->type;
666 sll->sll_protocol = skb->protocol;
667 sll->sll_pkttype = skb->pkt_type;
668 sll->sll_ifindex = dev->ifindex;
669
670 h->tp_status = status;
671 mb();
672
673 {
674 struct page *p_start, *p_end;
675 u8 *h_end = (u8 *)h + macoff + snaplen - 1;
676
677 p_start = virt_to_page(h);
678 p_end = virt_to_page(h_end);
679 while (p_start <= p_end) {
680 flush_dcache_page(p_start);
681 p_start++;
682 }
683 }
684
685 sk->sk_data_ready(sk, 0);
686
687drop_n_restore:
688 if (skb_head != skb->data && skb_shared(skb)) {
689 skb->data = skb_head;
690 skb->len = skb_len;
691 }
692drop:
693 kfree_skb(skb);
694 return 0;
695
696ring_is_full:
697 po->stats.tp_drops++;
698 spin_unlock(&sk->sk_receive_queue.lock);
699
700 sk->sk_data_ready(sk, 0);
701 if (copy_skb)
702 kfree_skb(copy_skb);
703 goto drop_n_restore;
704}
705
706#endif
707
708
709static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
710 struct msghdr *msg, size_t len)
711{
712 struct sock *sk = sock->sk;
713 struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
714 struct sk_buff *skb;
715 struct net_device *dev;
716 unsigned short proto;
717 unsigned char *addr;
718 int ifindex, err, reserve = 0;
719
720 /*
721 * Get and verify the address.
722 */
723
724 if (saddr == NULL) {
725 struct packet_sock *po = pkt_sk(sk);
726
727 ifindex = po->ifindex;
728 proto = po->num;
729 addr = NULL;
730 } else {
731 err = -EINVAL;
732 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
733 goto out;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -0700734 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
735 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 ifindex = saddr->sll_ifindex;
737 proto = saddr->sll_protocol;
738 addr = saddr->sll_addr;
739 }
740
741
742 dev = dev_get_by_index(ifindex);
743 err = -ENXIO;
744 if (dev == NULL)
745 goto out_unlock;
746 if (sock->type == SOCK_RAW)
747 reserve = dev->hard_header_len;
748
749 err = -EMSGSIZE;
750 if (len > dev->mtu+reserve)
751 goto out_unlock;
752
753 skb = sock_alloc_send_skb(sk, len + LL_RESERVED_SPACE(dev),
754 msg->msg_flags & MSG_DONTWAIT, &err);
755 if (skb==NULL)
756 goto out_unlock;
757
758 skb_reserve(skb, LL_RESERVED_SPACE(dev));
759 skb->nh.raw = skb->data;
760
761 if (dev->hard_header) {
762 int res;
763 err = -EINVAL;
764 res = dev->hard_header(skb, dev, ntohs(proto), addr, NULL, len);
765 if (sock->type != SOCK_DGRAM) {
766 skb->tail = skb->data;
767 skb->len = 0;
768 } else if (res < 0)
769 goto out_free;
770 }
771
772 /* Returns -EFAULT on error */
773 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
774 if (err)
775 goto out_free;
776
777 skb->protocol = proto;
778 skb->dev = dev;
779 skb->priority = sk->sk_priority;
780
781 err = -ENETDOWN;
782 if (!(dev->flags & IFF_UP))
783 goto out_free;
784
785 /*
786 * Now send it
787 */
788
789 err = dev_queue_xmit(skb);
790 if (err > 0 && (err = net_xmit_errno(err)) != 0)
791 goto out_unlock;
792
793 dev_put(dev);
794
795 return(len);
796
797out_free:
798 kfree_skb(skb);
799out_unlock:
800 if (dev)
801 dev_put(dev);
802out:
803 return err;
804}
805
806/*
807 * Close a PACKET socket. This is fairly simple. We immediately go
808 * to 'closed' state and remove our protocol entry in the device list.
809 */
810
811static int packet_release(struct socket *sock)
812{
813 struct sock *sk = sock->sk;
814 struct packet_sock *po;
815
816 if (!sk)
817 return 0;
818
819 po = pkt_sk(sk);
820
821 write_lock_bh(&packet_sklist_lock);
822 sk_del_node_init(sk);
823 write_unlock_bh(&packet_sklist_lock);
824
825 /*
826 * Unhook packet receive handler.
827 */
828
829 if (po->running) {
830 /*
831 * Remove the protocol hook
832 */
833 dev_remove_pack(&po->prot_hook);
834 po->running = 0;
835 po->num = 0;
836 __sock_put(sk);
837 }
838
839#ifdef CONFIG_PACKET_MULTICAST
840 packet_flush_mclist(sk);
841#endif
842
843#ifdef CONFIG_PACKET_MMAP
844 if (po->pg_vec) {
845 struct tpacket_req req;
846 memset(&req, 0, sizeof(req));
847 packet_set_ring(sk, &req, 1);
848 }
849#endif
850
851 /*
852 * Now the socket is dead. No more input will appear.
853 */
854
855 sock_orphan(sk);
856 sock->sk = NULL;
857
858 /* Purge queues */
859
860 skb_queue_purge(&sk->sk_receive_queue);
861
862 sock_put(sk);
863 return 0;
864}
865
866/*
867 * Attach a packet hook.
868 */
869
870static int packet_do_bind(struct sock *sk, struct net_device *dev, int protocol)
871{
872 struct packet_sock *po = pkt_sk(sk);
873 /*
874 * Detach an existing hook if present.
875 */
876
877 lock_sock(sk);
878
879 spin_lock(&po->bind_lock);
880 if (po->running) {
881 __sock_put(sk);
882 po->running = 0;
883 po->num = 0;
884 spin_unlock(&po->bind_lock);
885 dev_remove_pack(&po->prot_hook);
886 spin_lock(&po->bind_lock);
887 }
888
889 po->num = protocol;
890 po->prot_hook.type = protocol;
891 po->prot_hook.dev = dev;
892
893 po->ifindex = dev ? dev->ifindex : 0;
894
895 if (protocol == 0)
896 goto out_unlock;
897
898 if (dev) {
899 if (dev->flags&IFF_UP) {
900 dev_add_pack(&po->prot_hook);
901 sock_hold(sk);
902 po->running = 1;
903 } else {
904 sk->sk_err = ENETDOWN;
905 if (!sock_flag(sk, SOCK_DEAD))
906 sk->sk_error_report(sk);
907 }
908 } else {
909 dev_add_pack(&po->prot_hook);
910 sock_hold(sk);
911 po->running = 1;
912 }
913
914out_unlock:
915 spin_unlock(&po->bind_lock);
916 release_sock(sk);
917 return 0;
918}
919
920/*
921 * Bind a packet socket to a device
922 */
923
924#ifdef CONFIG_SOCK_PACKET
925
926static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
927{
928 struct sock *sk=sock->sk;
929 char name[15];
930 struct net_device *dev;
931 int err = -ENODEV;
932
933 /*
934 * Check legality
935 */
936
937 if(addr_len!=sizeof(struct sockaddr))
938 return -EINVAL;
939 strlcpy(name,uaddr->sa_data,sizeof(name));
940
941 dev = dev_get_by_name(name);
942 if (dev) {
943 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
944 dev_put(dev);
945 }
946 return err;
947}
948#endif
949
950static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
951{
952 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
953 struct sock *sk=sock->sk;
954 struct net_device *dev = NULL;
955 int err;
956
957
958 /*
959 * Check legality
960 */
961
962 if (addr_len < sizeof(struct sockaddr_ll))
963 return -EINVAL;
964 if (sll->sll_family != AF_PACKET)
965 return -EINVAL;
966
967 if (sll->sll_ifindex) {
968 err = -ENODEV;
969 dev = dev_get_by_index(sll->sll_ifindex);
970 if (dev == NULL)
971 goto out;
972 }
973 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
974 if (dev)
975 dev_put(dev);
976
977out:
978 return err;
979}
980
981static struct proto packet_proto = {
982 .name = "PACKET",
983 .owner = THIS_MODULE,
984 .obj_size = sizeof(struct packet_sock),
985};
986
987/*
988 * Create a packet of type SOCK_PACKET.
989 */
990
991static int packet_create(struct socket *sock, int protocol)
992{
993 struct sock *sk;
994 struct packet_sock *po;
995 int err;
996
997 if (!capable(CAP_NET_RAW))
998 return -EPERM;
999 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW
1000#ifdef CONFIG_SOCK_PACKET
1001 && sock->type != SOCK_PACKET
1002#endif
1003 )
1004 return -ESOCKTNOSUPPORT;
1005
1006 sock->state = SS_UNCONNECTED;
1007
1008 err = -ENOBUFS;
1009 sk = sk_alloc(PF_PACKET, GFP_KERNEL, &packet_proto, 1);
1010 if (sk == NULL)
1011 goto out;
1012
1013 sock->ops = &packet_ops;
1014#ifdef CONFIG_SOCK_PACKET
1015 if (sock->type == SOCK_PACKET)
1016 sock->ops = &packet_ops_spkt;
1017#endif
1018 sock_init_data(sock, sk);
1019
1020 po = pkt_sk(sk);
1021 sk->sk_family = PF_PACKET;
1022 po->num = protocol;
1023
1024 sk->sk_destruct = packet_sock_destruct;
1025 atomic_inc(&packet_socks_nr);
1026
1027 /*
1028 * Attach a protocol block
1029 */
1030
1031 spin_lock_init(&po->bind_lock);
1032 po->prot_hook.func = packet_rcv;
1033#ifdef CONFIG_SOCK_PACKET
1034 if (sock->type == SOCK_PACKET)
1035 po->prot_hook.func = packet_rcv_spkt;
1036#endif
1037 po->prot_hook.af_packet_priv = sk;
1038
1039 if (protocol) {
1040 po->prot_hook.type = protocol;
1041 dev_add_pack(&po->prot_hook);
1042 sock_hold(sk);
1043 po->running = 1;
1044 }
1045
1046 write_lock_bh(&packet_sklist_lock);
1047 sk_add_node(sk, &packet_sklist);
1048 write_unlock_bh(&packet_sklist_lock);
1049 return(0);
1050out:
1051 return err;
1052}
1053
1054/*
1055 * Pull a packet from our receive queue and hand it to the user.
1056 * If necessary we block.
1057 */
1058
1059static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1060 struct msghdr *msg, size_t len, int flags)
1061{
1062 struct sock *sk = sock->sk;
1063 struct sk_buff *skb;
1064 int copied, err;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001065 struct sockaddr_ll *sll;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
1067 err = -EINVAL;
1068 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1069 goto out;
1070
1071#if 0
1072 /* What error should we return now? EUNATTACH? */
1073 if (pkt_sk(sk)->ifindex < 0)
1074 return -ENODEV;
1075#endif
1076
1077 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 * Call the generic datagram receiver. This handles all sorts
1079 * of horrible races and re-entrancy so we can forget about it
1080 * in the protocol layers.
1081 *
1082 * Now it will return ENETDOWN, if device have just gone down,
1083 * but then it will block.
1084 */
1085
1086 skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1087
1088 /*
1089 * An error occurred so return it. Because skb_recv_datagram()
1090 * handles the blocking we don't see and worry about blocking
1091 * retries.
1092 */
1093
1094 if(skb==NULL)
1095 goto out;
1096
1097 /*
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001098 * If the address length field is there to be filled in, we fill
1099 * it in now.
1100 */
1101
1102 sll = (struct sockaddr_ll*)skb->cb;
1103 if (sock->type == SOCK_PACKET)
1104 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1105 else
1106 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1107
1108 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 * You lose any data beyond the buffer you gave. If it worries a
1110 * user program they can ask the device for its MTU anyway.
1111 */
1112
1113 copied = skb->len;
1114 if (copied > len)
1115 {
1116 copied=len;
1117 msg->msg_flags|=MSG_TRUNC;
1118 }
1119
1120 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1121 if (err)
1122 goto out_free;
1123
1124 sock_recv_timestamp(msg, sk, skb);
1125
1126 if (msg->msg_name)
1127 memcpy(msg->msg_name, skb->cb, msg->msg_namelen);
1128
1129 /*
1130 * Free or return the buffer as appropriate. Again this
1131 * hides all the races and re-entrancy issues from us.
1132 */
1133 err = (flags&MSG_TRUNC) ? skb->len : copied;
1134
1135out_free:
1136 skb_free_datagram(sk, skb);
1137out:
1138 return err;
1139}
1140
1141#ifdef CONFIG_SOCK_PACKET
1142static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1143 int *uaddr_len, int peer)
1144{
1145 struct net_device *dev;
1146 struct sock *sk = sock->sk;
1147
1148 if (peer)
1149 return -EOPNOTSUPP;
1150
1151 uaddr->sa_family = AF_PACKET;
1152 dev = dev_get_by_index(pkt_sk(sk)->ifindex);
1153 if (dev) {
1154 strlcpy(uaddr->sa_data, dev->name, 15);
1155 dev_put(dev);
1156 } else
1157 memset(uaddr->sa_data, 0, 14);
1158 *uaddr_len = sizeof(*uaddr);
1159
1160 return 0;
1161}
1162#endif
1163
1164static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1165 int *uaddr_len, int peer)
1166{
1167 struct net_device *dev;
1168 struct sock *sk = sock->sk;
1169 struct packet_sock *po = pkt_sk(sk);
1170 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1171
1172 if (peer)
1173 return -EOPNOTSUPP;
1174
1175 sll->sll_family = AF_PACKET;
1176 sll->sll_ifindex = po->ifindex;
1177 sll->sll_protocol = po->num;
1178 dev = dev_get_by_index(po->ifindex);
1179 if (dev) {
1180 sll->sll_hatype = dev->type;
1181 sll->sll_halen = dev->addr_len;
1182 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1183 dev_put(dev);
1184 } else {
1185 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1186 sll->sll_halen = 0;
1187 }
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001188 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189
1190 return 0;
1191}
1192
1193#ifdef CONFIG_PACKET_MULTICAST
1194static void packet_dev_mc(struct net_device *dev, struct packet_mclist *i, int what)
1195{
1196 switch (i->type) {
1197 case PACKET_MR_MULTICAST:
1198 if (what > 0)
1199 dev_mc_add(dev, i->addr, i->alen, 0);
1200 else
1201 dev_mc_delete(dev, i->addr, i->alen, 0);
1202 break;
1203 case PACKET_MR_PROMISC:
1204 dev_set_promiscuity(dev, what);
1205 break;
1206 case PACKET_MR_ALLMULTI:
1207 dev_set_allmulti(dev, what);
1208 break;
1209 default:;
1210 }
1211}
1212
1213static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1214{
1215 for ( ; i; i=i->next) {
1216 if (i->ifindex == dev->ifindex)
1217 packet_dev_mc(dev, i, what);
1218 }
1219}
1220
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001221static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222{
1223 struct packet_sock *po = pkt_sk(sk);
1224 struct packet_mclist *ml, *i;
1225 struct net_device *dev;
1226 int err;
1227
1228 rtnl_lock();
1229
1230 err = -ENODEV;
1231 dev = __dev_get_by_index(mreq->mr_ifindex);
1232 if (!dev)
1233 goto done;
1234
1235 err = -EINVAL;
1236 if (mreq->mr_alen > dev->addr_len)
1237 goto done;
1238
1239 err = -ENOBUFS;
1240 i = (struct packet_mclist *)kmalloc(sizeof(*i), GFP_KERNEL);
1241 if (i == NULL)
1242 goto done;
1243
1244 err = 0;
1245 for (ml = po->mclist; ml; ml = ml->next) {
1246 if (ml->ifindex == mreq->mr_ifindex &&
1247 ml->type == mreq->mr_type &&
1248 ml->alen == mreq->mr_alen &&
1249 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1250 ml->count++;
1251 /* Free the new element ... */
1252 kfree(i);
1253 goto done;
1254 }
1255 }
1256
1257 i->type = mreq->mr_type;
1258 i->ifindex = mreq->mr_ifindex;
1259 i->alen = mreq->mr_alen;
1260 memcpy(i->addr, mreq->mr_address, i->alen);
1261 i->count = 1;
1262 i->next = po->mclist;
1263 po->mclist = i;
1264 packet_dev_mc(dev, i, +1);
1265
1266done:
1267 rtnl_unlock();
1268 return err;
1269}
1270
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001271static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272{
1273 struct packet_mclist *ml, **mlp;
1274
1275 rtnl_lock();
1276
1277 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1278 if (ml->ifindex == mreq->mr_ifindex &&
1279 ml->type == mreq->mr_type &&
1280 ml->alen == mreq->mr_alen &&
1281 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1282 if (--ml->count == 0) {
1283 struct net_device *dev;
1284 *mlp = ml->next;
1285 dev = dev_get_by_index(ml->ifindex);
1286 if (dev) {
1287 packet_dev_mc(dev, ml, -1);
1288 dev_put(dev);
1289 }
1290 kfree(ml);
1291 }
1292 rtnl_unlock();
1293 return 0;
1294 }
1295 }
1296 rtnl_unlock();
1297 return -EADDRNOTAVAIL;
1298}
1299
1300static void packet_flush_mclist(struct sock *sk)
1301{
1302 struct packet_sock *po = pkt_sk(sk);
1303 struct packet_mclist *ml;
1304
1305 if (!po->mclist)
1306 return;
1307
1308 rtnl_lock();
1309 while ((ml = po->mclist) != NULL) {
1310 struct net_device *dev;
1311
1312 po->mclist = ml->next;
1313 if ((dev = dev_get_by_index(ml->ifindex)) != NULL) {
1314 packet_dev_mc(dev, ml, -1);
1315 dev_put(dev);
1316 }
1317 kfree(ml);
1318 }
1319 rtnl_unlock();
1320}
1321#endif
1322
1323static int
1324packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1325{
1326 struct sock *sk = sock->sk;
1327 int ret;
1328
1329 if (level != SOL_PACKET)
1330 return -ENOPROTOOPT;
1331
1332 switch(optname) {
1333#ifdef CONFIG_PACKET_MULTICAST
1334 case PACKET_ADD_MEMBERSHIP:
1335 case PACKET_DROP_MEMBERSHIP:
1336 {
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001337 struct packet_mreq_max mreq;
1338 int len = optlen;
1339 memset(&mreq, 0, sizeof(mreq));
1340 if (len < sizeof(struct packet_mreq))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 return -EINVAL;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001342 if (len > sizeof(mreq))
1343 len = sizeof(mreq);
1344 if (copy_from_user(&mreq,optval,len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 return -EFAULT;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001346 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1347 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 if (optname == PACKET_ADD_MEMBERSHIP)
1349 ret = packet_mc_add(sk, &mreq);
1350 else
1351 ret = packet_mc_drop(sk, &mreq);
1352 return ret;
1353 }
1354#endif
1355#ifdef CONFIG_PACKET_MMAP
1356 case PACKET_RX_RING:
1357 {
1358 struct tpacket_req req;
1359
1360 if (optlen<sizeof(req))
1361 return -EINVAL;
1362 if (copy_from_user(&req,optval,sizeof(req)))
1363 return -EFAULT;
1364 return packet_set_ring(sk, &req, 0);
1365 }
1366 case PACKET_COPY_THRESH:
1367 {
1368 int val;
1369
1370 if (optlen!=sizeof(val))
1371 return -EINVAL;
1372 if (copy_from_user(&val,optval,sizeof(val)))
1373 return -EFAULT;
1374
1375 pkt_sk(sk)->copy_thresh = val;
1376 return 0;
1377 }
1378#endif
1379 default:
1380 return -ENOPROTOOPT;
1381 }
1382}
1383
1384static int packet_getsockopt(struct socket *sock, int level, int optname,
1385 char __user *optval, int __user *optlen)
1386{
1387 int len;
1388 struct sock *sk = sock->sk;
1389 struct packet_sock *po = pkt_sk(sk);
1390
1391 if (level != SOL_PACKET)
1392 return -ENOPROTOOPT;
1393
1394 if (get_user(len,optlen))
1395 return -EFAULT;
1396
1397 if (len < 0)
1398 return -EINVAL;
1399
1400 switch(optname) {
1401 case PACKET_STATISTICS:
1402 {
1403 struct tpacket_stats st;
1404
1405 if (len > sizeof(struct tpacket_stats))
1406 len = sizeof(struct tpacket_stats);
1407 spin_lock_bh(&sk->sk_receive_queue.lock);
1408 st = po->stats;
1409 memset(&po->stats, 0, sizeof(st));
1410 spin_unlock_bh(&sk->sk_receive_queue.lock);
1411 st.tp_packets += st.tp_drops;
1412
1413 if (copy_to_user(optval, &st, len))
1414 return -EFAULT;
1415 break;
1416 }
1417 default:
1418 return -ENOPROTOOPT;
1419 }
1420
1421 if (put_user(len, optlen))
1422 return -EFAULT;
1423 return 0;
1424}
1425
1426
1427static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1428{
1429 struct sock *sk;
1430 struct hlist_node *node;
1431 struct net_device *dev = (struct net_device*)data;
1432
1433 read_lock(&packet_sklist_lock);
1434 sk_for_each(sk, node, &packet_sklist) {
1435 struct packet_sock *po = pkt_sk(sk);
1436
1437 switch (msg) {
1438 case NETDEV_UNREGISTER:
1439#ifdef CONFIG_PACKET_MULTICAST
1440 if (po->mclist)
1441 packet_dev_mclist(dev, po->mclist, -1);
1442 // fallthrough
1443#endif
1444 case NETDEV_DOWN:
1445 if (dev->ifindex == po->ifindex) {
1446 spin_lock(&po->bind_lock);
1447 if (po->running) {
1448 __dev_remove_pack(&po->prot_hook);
1449 __sock_put(sk);
1450 po->running = 0;
1451 sk->sk_err = ENETDOWN;
1452 if (!sock_flag(sk, SOCK_DEAD))
1453 sk->sk_error_report(sk);
1454 }
1455 if (msg == NETDEV_UNREGISTER) {
1456 po->ifindex = -1;
1457 po->prot_hook.dev = NULL;
1458 }
1459 spin_unlock(&po->bind_lock);
1460 }
1461 break;
1462 case NETDEV_UP:
1463 spin_lock(&po->bind_lock);
1464 if (dev->ifindex == po->ifindex && po->num &&
1465 !po->running) {
1466 dev_add_pack(&po->prot_hook);
1467 sock_hold(sk);
1468 po->running = 1;
1469 }
1470 spin_unlock(&po->bind_lock);
1471 break;
1472 }
1473 }
1474 read_unlock(&packet_sklist_lock);
1475 return NOTIFY_DONE;
1476}
1477
1478
1479static int packet_ioctl(struct socket *sock, unsigned int cmd,
1480 unsigned long arg)
1481{
1482 struct sock *sk = sock->sk;
1483
1484 switch(cmd) {
1485 case SIOCOUTQ:
1486 {
1487 int amount = atomic_read(&sk->sk_wmem_alloc);
1488 return put_user(amount, (int __user *)arg);
1489 }
1490 case SIOCINQ:
1491 {
1492 struct sk_buff *skb;
1493 int amount = 0;
1494
1495 spin_lock_bh(&sk->sk_receive_queue.lock);
1496 skb = skb_peek(&sk->sk_receive_queue);
1497 if (skb)
1498 amount = skb->len;
1499 spin_unlock_bh(&sk->sk_receive_queue.lock);
1500 return put_user(amount, (int __user *)arg);
1501 }
1502 case SIOCGSTAMP:
1503 return sock_get_timestamp(sk, (struct timeval __user *)arg);
1504
1505#ifdef CONFIG_INET
1506 case SIOCADDRT:
1507 case SIOCDELRT:
1508 case SIOCDARP:
1509 case SIOCGARP:
1510 case SIOCSARP:
1511 case SIOCGIFADDR:
1512 case SIOCSIFADDR:
1513 case SIOCGIFBRDADDR:
1514 case SIOCSIFBRDADDR:
1515 case SIOCGIFNETMASK:
1516 case SIOCSIFNETMASK:
1517 case SIOCGIFDSTADDR:
1518 case SIOCSIFDSTADDR:
1519 case SIOCSIFFLAGS:
1520 return inet_dgram_ops.ioctl(sock, cmd, arg);
1521#endif
1522
1523 default:
1524 return dev_ioctl(cmd, (void __user *)arg);
1525 }
1526 return 0;
1527}
1528
1529#ifndef CONFIG_PACKET_MMAP
1530#define packet_mmap sock_no_mmap
1531#define packet_poll datagram_poll
1532#else
1533
1534static unsigned int packet_poll(struct file * file, struct socket *sock,
1535 poll_table *wait)
1536{
1537 struct sock *sk = sock->sk;
1538 struct packet_sock *po = pkt_sk(sk);
1539 unsigned int mask = datagram_poll(file, sock, wait);
1540
1541 spin_lock_bh(&sk->sk_receive_queue.lock);
1542 if (po->pg_vec) {
1543 unsigned last = po->head ? po->head-1 : po->frame_max;
1544 struct tpacket_hdr *h;
1545
1546 h = (struct tpacket_hdr *)packet_lookup_frame(po, last);
1547
1548 if (h->tp_status)
1549 mask |= POLLIN | POLLRDNORM;
1550 }
1551 spin_unlock_bh(&sk->sk_receive_queue.lock);
1552 return mask;
1553}
1554
1555
1556/* Dirty? Well, I still did not learn better way to account
1557 * for user mmaps.
1558 */
1559
1560static void packet_mm_open(struct vm_area_struct *vma)
1561{
1562 struct file *file = vma->vm_file;
Eric Dumazetb69aee02005-09-06 14:42:45 -07001563 struct socket * sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 struct sock *sk = sock->sk;
1565
1566 if (sk)
1567 atomic_inc(&pkt_sk(sk)->mapped);
1568}
1569
1570static void packet_mm_close(struct vm_area_struct *vma)
1571{
1572 struct file *file = vma->vm_file;
Eric Dumazetb69aee02005-09-06 14:42:45 -07001573 struct socket * sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 struct sock *sk = sock->sk;
1575
1576 if (sk)
1577 atomic_dec(&pkt_sk(sk)->mapped);
1578}
1579
1580static struct vm_operations_struct packet_mmap_ops = {
1581 .open = packet_mm_open,
1582 .close =packet_mm_close,
1583};
1584
1585static inline struct page *pg_vec_endpage(char *one_pg_vec, unsigned int order)
1586{
1587 return virt_to_page(one_pg_vec + (PAGE_SIZE << order) - 1);
1588}
1589
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001590static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591{
1592 int i;
1593
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001594 for (i = 0; i < len; i++) {
1595 if (likely(pg_vec[i]))
1596 free_pages((unsigned long) pg_vec[i], order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 }
1598 kfree(pg_vec);
1599}
1600
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001601static inline char *alloc_one_pg_vec_page(unsigned long order)
1602{
1603 return (char *) __get_free_pages(GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
1604 order);
1605}
1606
1607static char **alloc_pg_vec(struct tpacket_req *req, int order)
1608{
1609 unsigned int block_nr = req->tp_block_nr;
1610 char **pg_vec;
1611 int i;
1612
1613 pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
1614 if (unlikely(!pg_vec))
1615 goto out;
1616
1617 for (i = 0; i < block_nr; i++) {
1618 pg_vec[i] = alloc_one_pg_vec_page(order);
1619 if (unlikely(!pg_vec[i]))
1620 goto out_free_pgvec;
1621 }
1622
1623out:
1624 return pg_vec;
1625
1626out_free_pgvec:
1627 free_pg_vec(pg_vec, order, block_nr);
1628 pg_vec = NULL;
1629 goto out;
1630}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631
1632static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
1633{
1634 char **pg_vec = NULL;
1635 struct packet_sock *po = pkt_sk(sk);
1636 int was_running, num, order = 0;
1637 int err = 0;
1638
1639 if (req->tp_block_nr) {
1640 int i, l;
1641
1642 /* Sanity tests and some calculations */
1643
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001644 if (unlikely(po->pg_vec))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 return -EBUSY;
1646
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001647 if (unlikely((int)req->tp_block_size <= 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 return -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001649 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 return -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001651 if (unlikely(req->tp_frame_size < TPACKET_HDRLEN))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 return -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001653 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 return -EINVAL;
1655
1656 po->frames_per_block = req->tp_block_size/req->tp_frame_size;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001657 if (unlikely(po->frames_per_block <= 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 return -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001659 if (unlikely((po->frames_per_block * req->tp_block_nr) !=
1660 req->tp_frame_nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662
1663 err = -ENOMEM;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001664 order = get_order(req->tp_block_size);
1665 pg_vec = alloc_pg_vec(req, order);
1666 if (unlikely(!pg_vec))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669 l = 0;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001670 for (i = 0; i < req->tp_block_nr; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671 char *ptr = pg_vec[i];
1672 struct tpacket_hdr *header;
1673 int k;
1674
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001675 for (k = 0; k < po->frames_per_block; k++) {
1676 header = (struct tpacket_hdr *) ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 header->tp_status = TP_STATUS_KERNEL;
1678 ptr += req->tp_frame_size;
1679 }
1680 }
1681 /* Done */
1682 } else {
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001683 if (unlikely(req->tp_frame_nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 return -EINVAL;
1685 }
1686
1687 lock_sock(sk);
1688
1689 /* Detach socket from network */
1690 spin_lock(&po->bind_lock);
1691 was_running = po->running;
1692 num = po->num;
1693 if (was_running) {
1694 __dev_remove_pack(&po->prot_hook);
1695 po->num = 0;
1696 po->running = 0;
1697 __sock_put(sk);
1698 }
1699 spin_unlock(&po->bind_lock);
1700
1701 synchronize_net();
1702
1703 err = -EBUSY;
1704 if (closing || atomic_read(&po->mapped) == 0) {
1705 err = 0;
1706#define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
1707
1708 spin_lock_bh(&sk->sk_receive_queue.lock);
1709 pg_vec = XC(po->pg_vec, pg_vec);
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001710 po->frame_max = (req->tp_frame_nr - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 po->head = 0;
1712 po->frame_size = req->tp_frame_size;
1713 spin_unlock_bh(&sk->sk_receive_queue.lock);
1714
1715 order = XC(po->pg_vec_order, order);
1716 req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
1717
1718 po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
1719 po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
1720 skb_queue_purge(&sk->sk_receive_queue);
1721#undef XC
1722 if (atomic_read(&po->mapped))
1723 printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
1724 }
1725
1726 spin_lock(&po->bind_lock);
1727 if (was_running && !po->running) {
1728 sock_hold(sk);
1729 po->running = 1;
1730 po->num = num;
1731 dev_add_pack(&po->prot_hook);
1732 }
1733 spin_unlock(&po->bind_lock);
1734
1735 release_sock(sk);
1736
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 if (pg_vec)
1738 free_pg_vec(pg_vec, order, req->tp_block_nr);
1739out:
1740 return err;
1741}
1742
1743static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1744{
1745 struct sock *sk = sock->sk;
1746 struct packet_sock *po = pkt_sk(sk);
1747 unsigned long size;
1748 unsigned long start;
1749 int err = -EINVAL;
1750 int i;
1751
1752 if (vma->vm_pgoff)
1753 return -EINVAL;
1754
1755 size = vma->vm_end - vma->vm_start;
1756
1757 lock_sock(sk);
1758 if (po->pg_vec == NULL)
1759 goto out;
1760 if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
1761 goto out;
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 start = vma->vm_start;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001764 for (i = 0; i < po->pg_vec_len; i++) {
1765 struct page *page = virt_to_page(po->pg_vec[i]);
1766 int pg_num;
1767
1768 for (pg_num = 0; pg_num < po->pg_vec_pages; pg_num++, page++) {
1769 err = vm_insert_page(vma, start, page);
1770 if (unlikely(err))
1771 goto out;
1772 start += PAGE_SIZE;
1773 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 }
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001775 atomic_inc(&po->mapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 vma->vm_ops = &packet_mmap_ops;
1777 err = 0;
1778
1779out:
1780 release_sock(sk);
1781 return err;
1782}
1783#endif
1784
1785
1786#ifdef CONFIG_SOCK_PACKET
1787static struct proto_ops packet_ops_spkt = {
1788 .family = PF_PACKET,
1789 .owner = THIS_MODULE,
1790 .release = packet_release,
1791 .bind = packet_bind_spkt,
1792 .connect = sock_no_connect,
1793 .socketpair = sock_no_socketpair,
1794 .accept = sock_no_accept,
1795 .getname = packet_getname_spkt,
1796 .poll = datagram_poll,
1797 .ioctl = packet_ioctl,
1798 .listen = sock_no_listen,
1799 .shutdown = sock_no_shutdown,
1800 .setsockopt = sock_no_setsockopt,
1801 .getsockopt = sock_no_getsockopt,
1802 .sendmsg = packet_sendmsg_spkt,
1803 .recvmsg = packet_recvmsg,
1804 .mmap = sock_no_mmap,
1805 .sendpage = sock_no_sendpage,
1806};
1807#endif
1808
1809static struct proto_ops packet_ops = {
1810 .family = PF_PACKET,
1811 .owner = THIS_MODULE,
1812 .release = packet_release,
1813 .bind = packet_bind,
1814 .connect = sock_no_connect,
1815 .socketpair = sock_no_socketpair,
1816 .accept = sock_no_accept,
1817 .getname = packet_getname,
1818 .poll = packet_poll,
1819 .ioctl = packet_ioctl,
1820 .listen = sock_no_listen,
1821 .shutdown = sock_no_shutdown,
1822 .setsockopt = packet_setsockopt,
1823 .getsockopt = packet_getsockopt,
1824 .sendmsg = packet_sendmsg,
1825 .recvmsg = packet_recvmsg,
1826 .mmap = packet_mmap,
1827 .sendpage = sock_no_sendpage,
1828};
1829
1830static struct net_proto_family packet_family_ops = {
1831 .family = PF_PACKET,
1832 .create = packet_create,
1833 .owner = THIS_MODULE,
1834};
1835
1836static struct notifier_block packet_netdev_notifier = {
1837 .notifier_call =packet_notifier,
1838};
1839
1840#ifdef CONFIG_PROC_FS
1841static inline struct sock *packet_seq_idx(loff_t off)
1842{
1843 struct sock *s;
1844 struct hlist_node *node;
1845
1846 sk_for_each(s, node, &packet_sklist) {
1847 if (!off--)
1848 return s;
1849 }
1850 return NULL;
1851}
1852
1853static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
1854{
1855 read_lock(&packet_sklist_lock);
1856 return *pos ? packet_seq_idx(*pos - 1) : SEQ_START_TOKEN;
1857}
1858
1859static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1860{
1861 ++*pos;
1862 return (v == SEQ_START_TOKEN)
1863 ? sk_head(&packet_sklist)
1864 : sk_next((struct sock*)v) ;
1865}
1866
1867static void packet_seq_stop(struct seq_file *seq, void *v)
1868{
1869 read_unlock(&packet_sklist_lock);
1870}
1871
1872static int packet_seq_show(struct seq_file *seq, void *v)
1873{
1874 if (v == SEQ_START_TOKEN)
1875 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
1876 else {
1877 struct sock *s = v;
1878 const struct packet_sock *po = pkt_sk(s);
1879
1880 seq_printf(seq,
1881 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
1882 s,
1883 atomic_read(&s->sk_refcnt),
1884 s->sk_type,
1885 ntohs(po->num),
1886 po->ifindex,
1887 po->running,
1888 atomic_read(&s->sk_rmem_alloc),
1889 sock_i_uid(s),
1890 sock_i_ino(s) );
1891 }
1892
1893 return 0;
1894}
1895
1896static struct seq_operations packet_seq_ops = {
1897 .start = packet_seq_start,
1898 .next = packet_seq_next,
1899 .stop = packet_seq_stop,
1900 .show = packet_seq_show,
1901};
1902
1903static int packet_seq_open(struct inode *inode, struct file *file)
1904{
1905 return seq_open(file, &packet_seq_ops);
1906}
1907
1908static struct file_operations packet_seq_fops = {
1909 .owner = THIS_MODULE,
1910 .open = packet_seq_open,
1911 .read = seq_read,
1912 .llseek = seq_lseek,
1913 .release = seq_release,
1914};
1915
1916#endif
1917
1918static void __exit packet_exit(void)
1919{
1920 proc_net_remove("packet");
1921 unregister_netdevice_notifier(&packet_netdev_notifier);
1922 sock_unregister(PF_PACKET);
1923 proto_unregister(&packet_proto);
1924}
1925
1926static int __init packet_init(void)
1927{
1928 int rc = proto_register(&packet_proto, 0);
1929
1930 if (rc != 0)
1931 goto out;
1932
1933 sock_register(&packet_family_ops);
1934 register_netdevice_notifier(&packet_netdev_notifier);
1935 proc_net_fops_create("packet", 0, &packet_seq_fops);
1936out:
1937 return rc;
1938}
1939
1940module_init(packet_init);
1941module_exit(packet_exit);
1942MODULE_LICENSE("GPL");
1943MODULE_ALIAS_NETPROTO(PF_PACKET);