blob: ee93abc71cb8aaefde9f24bbd6a5e5326fc5aee3 [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>
Randy Dunlap4fc268d2006-01-11 12:17:47 -080056#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <linux/fcntl.h>
58#include <linux/socket.h>
59#include <linux/in.h>
60#include <linux/inet.h>
61#include <linux/netdevice.h>
62#include <linux/if_packet.h>
63#include <linux/wireless.h>
64#include <linux/kmod.h>
65#include <net/ip.h>
66#include <net/protocol.h>
67#include <linux/skbuff.h>
68#include <net/sock.h>
69#include <linux/errno.h>
70#include <linux/timer.h>
71#include <asm/system.h>
72#include <asm/uaccess.h>
73#include <asm/ioctls.h>
74#include <asm/page.h>
75#include <asm/io.h>
76#include <linux/proc_fs.h>
77#include <linux/seq_file.h>
78#include <linux/poll.h>
79#include <linux/module.h>
80#include <linux/init.h>
81
82#ifdef CONFIG_INET
83#include <net/inet_common.h>
84#endif
85
86#define CONFIG_SOCK_PACKET 1
87
88/*
89 Proposed replacement for SIOC{ADD,DEL}MULTI and
90 IFF_PROMISC, IFF_ALLMULTI flags.
91
92 It is more expensive, but I believe,
93 it is really correct solution: reentereble, safe and fault tolerant.
94
95 IFF_PROMISC/IFF_ALLMULTI/SIOC{ADD/DEL}MULTI are faked by keeping
96 reference count and global flag, so that real status is
97 (gflag|(count != 0)), so that we can use obsolete faulty interface
98 not harming clever users.
99 */
100#define CONFIG_PACKET_MULTICAST 1
101
102/*
103 Assumptions:
104 - if device has no dev->hard_header routine, it adds and removes ll header
105 inside itself. In this case ll header is invisible outside of device,
106 but higher levels still should reserve dev->hard_header_len.
107 Some devices are enough clever to reallocate skb, when header
108 will not fit to reserved space (tunnel), another ones are silly
109 (PPP).
110 - packet socket receives packets with pulled ll header,
111 so that SOCK_RAW should push it back.
112
113On receive:
114-----------
115
116Incoming, dev->hard_header!=NULL
117 mac.raw -> ll header
118 data -> data
119
120Outgoing, dev->hard_header!=NULL
121 mac.raw -> ll header
122 data -> ll header
123
124Incoming, dev->hard_header==NULL
125 mac.raw -> UNKNOWN position. It is very likely, that it points to ll header.
126 PPP makes it, that is wrong, because introduce assymetry
127 between rx and tx paths.
128 data -> data
129
130Outgoing, dev->hard_header==NULL
131 mac.raw -> data. ll header is still not built!
132 data -> data
133
134Resume
135 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
136
137
138On transmit:
139------------
140
141dev->hard_header != NULL
142 mac.raw -> ll header
143 data -> ll header
144
145dev->hard_header == NULL (ll header is added by device, we cannot control it)
146 mac.raw -> data
147 data -> data
148
149 We should set nh.raw on output to correct posistion,
150 packet classifier depends on it.
151 */
152
153/* List of all packet sockets. */
154static HLIST_HEAD(packet_sklist);
155static DEFINE_RWLOCK(packet_sklist_lock);
156
157static atomic_t packet_socks_nr;
158
159
160/* Private packet socket structures. */
161
162#ifdef CONFIG_PACKET_MULTICAST
163struct packet_mclist
164{
165 struct packet_mclist *next;
166 int ifindex;
167 int count;
168 unsigned short type;
169 unsigned short alen;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -0700170 unsigned char addr[MAX_ADDR_LEN];
171};
172/* identical to struct packet_mreq except it has
173 * a longer address field.
174 */
175struct packet_mreq_max
176{
177 int mr_ifindex;
178 unsigned short mr_type;
179 unsigned short mr_alen;
180 unsigned char mr_address[MAX_ADDR_LEN];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181};
182#endif
183#ifdef CONFIG_PACKET_MMAP
184static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
185#endif
186
187static void packet_flush_mclist(struct sock *sk);
188
189struct packet_sock {
190 /* struct sock has to be the first member of packet_sock */
191 struct sock sk;
192 struct tpacket_stats stats;
193#ifdef CONFIG_PACKET_MMAP
194 char * *pg_vec;
195 unsigned int head;
196 unsigned int frames_per_block;
197 unsigned int frame_size;
198 unsigned int frame_max;
199 int copy_thresh;
200#endif
201 struct packet_type prot_hook;
202 spinlock_t bind_lock;
203 char running; /* prot_hook is attached*/
204 int ifindex; /* bound device */
205 unsigned short num;
206#ifdef CONFIG_PACKET_MULTICAST
207 struct packet_mclist *mclist;
208#endif
209#ifdef CONFIG_PACKET_MMAP
210 atomic_t mapped;
211 unsigned int pg_vec_order;
212 unsigned int pg_vec_pages;
213 unsigned int pg_vec_len;
214#endif
215};
216
217#ifdef CONFIG_PACKET_MMAP
218
219static inline char *packet_lookup_frame(struct packet_sock *po, unsigned int position)
220{
221 unsigned int pg_vec_pos, frame_offset;
222 char *frame;
223
224 pg_vec_pos = position / po->frames_per_block;
225 frame_offset = position % po->frames_per_block;
226
227 frame = po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size);
228
229 return frame;
230}
231#endif
232
233static inline struct packet_sock *pkt_sk(struct sock *sk)
234{
235 return (struct packet_sock *)sk;
236}
237
238static void packet_sock_destruct(struct sock *sk)
239{
240 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
241 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
242
243 if (!sock_flag(sk, SOCK_DEAD)) {
244 printk("Attempt to release alive packet socket: %p\n", sk);
245 return;
246 }
247
248 atomic_dec(&packet_socks_nr);
249#ifdef PACKET_REFCNT_DEBUG
250 printk(KERN_DEBUG "PACKET socket %p is free, %d are alive\n", sk, atomic_read(&packet_socks_nr));
251#endif
252}
253
254
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800255static const struct proto_ops packet_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257#ifdef CONFIG_SOCK_PACKET
Eric Dumazet90ddc4f2005-12-22 12:49:22 -0800258static const struct proto_ops packet_ops_spkt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
David S. Millerf2ccd8f2005-08-09 19:34:12 -0700260static 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 -0700261{
262 struct sock *sk;
263 struct sockaddr_pkt *spkt;
264
265 /*
266 * When we registered the protocol we saved the socket in the data
267 * field for just this event.
268 */
269
270 sk = pt->af_packet_priv;
271
272 /*
273 * Yank back the headers [hope the device set this
274 * right or kerboom...]
275 *
276 * Incoming packets have ll header pulled,
277 * push it back.
278 *
279 * For outgoing ones skb->data == skb->mac.raw
280 * so that this procedure is noop.
281 */
282
283 if (skb->pkt_type == PACKET_LOOPBACK)
284 goto out;
285
286 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
287 goto oom;
288
289 /* drop any routing info */
290 dst_release(skb->dst);
291 skb->dst = NULL;
292
Phil Oester84531c22005-07-12 11:57:52 -0700293 /* drop conntrack reference */
294 nf_reset(skb);
295
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 spkt = (struct sockaddr_pkt*)skb->cb;
297
298 skb_push(skb, skb->data-skb->mac.raw);
299
300 /*
301 * The SOCK_PACKET socket receives _all_ frames.
302 */
303
304 spkt->spkt_family = dev->type;
305 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
306 spkt->spkt_protocol = skb->protocol;
307
308 /*
309 * Charge the memory to the socket. This is done specifically
310 * to prevent sockets using all the memory up.
311 */
312
313 if (sock_queue_rcv_skb(sk,skb) == 0)
314 return 0;
315
316out:
317 kfree_skb(skb);
318oom:
319 return 0;
320}
321
322
323/*
324 * Output a raw packet to a device layer. This bypasses all the other
325 * protocol layers and you must therefore supply it with a complete frame
326 */
327
328static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
329 struct msghdr *msg, size_t len)
330{
331 struct sock *sk = sock->sk;
332 struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
333 struct sk_buff *skb;
334 struct net_device *dev;
335 unsigned short proto=0;
336 int err;
337
338 /*
339 * Get and verify the address.
340 */
341
342 if (saddr)
343 {
344 if (msg->msg_namelen < sizeof(struct sockaddr))
345 return(-EINVAL);
346 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
347 proto=saddr->spkt_protocol;
348 }
349 else
350 return(-ENOTCONN); /* SOCK_PACKET must be sent giving an address */
351
352 /*
353 * Find the device first to size check it
354 */
355
356 saddr->spkt_device[13] = 0;
357 dev = dev_get_by_name(saddr->spkt_device);
358 err = -ENODEV;
359 if (dev == NULL)
360 goto out_unlock;
361
362 /*
363 * You may not queue a frame bigger than the mtu. This is the lowest level
364 * raw protocol and you must do your own fragmentation at this level.
365 */
366
367 err = -EMSGSIZE;
368 if(len>dev->mtu+dev->hard_header_len)
369 goto out_unlock;
370
371 err = -ENOBUFS;
372 skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
373
374 /*
375 * If the write buffer is full, then tough. At this level the user gets to
376 * deal with the problem - do your own algorithmic backoffs. That's far
377 * more flexible.
378 */
379
380 if (skb == NULL)
381 goto out_unlock;
382
383 /*
384 * Fill it in
385 */
386
387 /* FIXME: Save some space for broken drivers that write a
388 * hard header at transmission time by themselves. PPP is the
389 * notable one here. This should really be fixed at the driver level.
390 */
391 skb_reserve(skb, LL_RESERVED_SPACE(dev));
392 skb->nh.raw = skb->data;
393
394 /* Try to align data part correctly */
395 if (dev->hard_header) {
396 skb->data -= dev->hard_header_len;
397 skb->tail -= dev->hard_header_len;
398 if (len < dev->hard_header_len)
399 skb->nh.raw = skb->data;
400 }
401
402 /* Returns -EFAULT on error */
403 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
404 skb->protocol = proto;
405 skb->dev = dev;
406 skb->priority = sk->sk_priority;
407 if (err)
408 goto out_free;
409
410 err = -ENETDOWN;
411 if (!(dev->flags & IFF_UP))
412 goto out_free;
413
414 /*
415 * Now send it
416 */
417
418 dev_queue_xmit(skb);
419 dev_put(dev);
420 return(len);
421
422out_free:
423 kfree_skb(skb);
424out_unlock:
425 if (dev)
426 dev_put(dev);
427 return err;
428}
429#endif
430
431static inline unsigned run_filter(struct sk_buff *skb, struct sock *sk, unsigned res)
432{
433 struct sk_filter *filter;
434
435 bh_lock_sock(sk);
436 filter = sk->sk_filter;
437 /*
438 * Our caller already checked that filter != NULL but we need to
439 * verify that under bh_lock_sock() to be safe
440 */
441 if (likely(filter != NULL))
442 res = sk_run_filter(skb, filter->insns, filter->len);
443 bh_unlock_sock(sk);
444
445 return res;
446}
447
448/*
449 This function makes lazy skb cloning in hope that most of packets
450 are discarded by BPF.
451
452 Note tricky part: we DO mangle shared skb! skb->data, skb->len
453 and skb->cb are mangled. It works because (and until) packets
454 falling here are owned by current CPU. Output packets are cloned
455 by dev_queue_xmit_nit(), input packets are processed by net_bh
456 sequencially, so that if we return skb to original state on exit,
457 we will not harm anyone.
458 */
459
David S. Millerf2ccd8f2005-08-09 19:34:12 -0700460static 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 -0700461{
462 struct sock *sk;
463 struct sockaddr_ll *sll;
464 struct packet_sock *po;
465 u8 * skb_head = skb->data;
466 int skb_len = skb->len;
467 unsigned snaplen;
468
469 if (skb->pkt_type == PACKET_LOOPBACK)
470 goto drop;
471
472 sk = pt->af_packet_priv;
473 po = pkt_sk(sk);
474
475 skb->dev = dev;
476
477 if (dev->hard_header) {
478 /* The device has an explicit notion of ll header,
479 exported to higher levels.
480
481 Otherwise, the device hides datails of it frame
482 structure, so that corresponding packet head
483 never delivered to user.
484 */
485 if (sk->sk_type != SOCK_DGRAM)
486 skb_push(skb, skb->data - skb->mac.raw);
487 else if (skb->pkt_type == PACKET_OUTGOING) {
488 /* Special case: outgoing packets have ll header at head */
489 skb_pull(skb, skb->nh.raw - skb->data);
490 }
491 }
492
493 snaplen = skb->len;
494
495 if (sk->sk_filter) {
496 unsigned res = run_filter(skb, sk, snaplen);
497 if (res == 0)
498 goto drop_n_restore;
499 if (snaplen > res)
500 snaplen = res;
501 }
502
503 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
504 (unsigned)sk->sk_rcvbuf)
505 goto drop_n_acct;
506
507 if (skb_shared(skb)) {
508 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
509 if (nskb == NULL)
510 goto drop_n_acct;
511
512 if (skb_head != skb->data) {
513 skb->data = skb_head;
514 skb->len = skb_len;
515 }
516 kfree_skb(skb);
517 skb = nskb;
518 }
519
520 sll = (struct sockaddr_ll*)skb->cb;
521 sll->sll_family = AF_PACKET;
522 sll->sll_hatype = dev->type;
523 sll->sll_protocol = skb->protocol;
524 sll->sll_pkttype = skb->pkt_type;
525 sll->sll_ifindex = dev->ifindex;
526 sll->sll_halen = 0;
527
528 if (dev->hard_header_parse)
529 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
530
531 if (pskb_trim(skb, snaplen))
532 goto drop_n_acct;
533
534 skb_set_owner_r(skb, sk);
535 skb->dev = NULL;
536 dst_release(skb->dst);
537 skb->dst = NULL;
538
Phil Oester84531c22005-07-12 11:57:52 -0700539 /* drop conntrack reference */
540 nf_reset(skb);
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 spin_lock(&sk->sk_receive_queue.lock);
543 po->stats.tp_packets++;
544 __skb_queue_tail(&sk->sk_receive_queue, skb);
545 spin_unlock(&sk->sk_receive_queue.lock);
546 sk->sk_data_ready(sk, skb->len);
547 return 0;
548
549drop_n_acct:
550 spin_lock(&sk->sk_receive_queue.lock);
551 po->stats.tp_drops++;
552 spin_unlock(&sk->sk_receive_queue.lock);
553
554drop_n_restore:
555 if (skb_head != skb->data && skb_shared(skb)) {
556 skb->data = skb_head;
557 skb->len = skb_len;
558 }
559drop:
560 kfree_skb(skb);
561 return 0;
562}
563
564#ifdef CONFIG_PACKET_MMAP
David S. Millerf2ccd8f2005-08-09 19:34:12 -0700565static 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 -0700566{
567 struct sock *sk;
568 struct packet_sock *po;
569 struct sockaddr_ll *sll;
570 struct tpacket_hdr *h;
571 u8 * skb_head = skb->data;
572 int skb_len = skb->len;
573 unsigned snaplen;
574 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
575 unsigned short macoff, netoff;
576 struct sk_buff *copy_skb = NULL;
577
578 if (skb->pkt_type == PACKET_LOOPBACK)
579 goto drop;
580
581 sk = pt->af_packet_priv;
582 po = pkt_sk(sk);
583
584 if (dev->hard_header) {
585 if (sk->sk_type != SOCK_DGRAM)
586 skb_push(skb, skb->data - skb->mac.raw);
587 else if (skb->pkt_type == PACKET_OUTGOING) {
588 /* Special case: outgoing packets have ll header at head */
589 skb_pull(skb, skb->nh.raw - skb->data);
590 if (skb->ip_summed == CHECKSUM_HW)
591 status |= TP_STATUS_CSUMNOTREADY;
592 }
593 }
594
595 snaplen = skb->len;
596
597 if (sk->sk_filter) {
598 unsigned res = run_filter(skb, sk, snaplen);
599 if (res == 0)
600 goto drop_n_restore;
601 if (snaplen > res)
602 snaplen = res;
603 }
604
605 if (sk->sk_type == SOCK_DGRAM) {
606 macoff = netoff = TPACKET_ALIGN(TPACKET_HDRLEN) + 16;
607 } else {
608 unsigned maclen = skb->nh.raw - skb->data;
609 netoff = TPACKET_ALIGN(TPACKET_HDRLEN + (maclen < 16 ? 16 : maclen));
610 macoff = netoff - maclen;
611 }
612
613 if (macoff + snaplen > po->frame_size) {
614 if (po->copy_thresh &&
615 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
616 (unsigned)sk->sk_rcvbuf) {
617 if (skb_shared(skb)) {
618 copy_skb = skb_clone(skb, GFP_ATOMIC);
619 } else {
620 copy_skb = skb_get(skb);
621 skb_head = skb->data;
622 }
623 if (copy_skb)
624 skb_set_owner_r(copy_skb, sk);
625 }
626 snaplen = po->frame_size - macoff;
627 if ((int)snaplen < 0)
628 snaplen = 0;
629 }
630 if (snaplen > skb->len-skb->data_len)
631 snaplen = skb->len-skb->data_len;
632
633 spin_lock(&sk->sk_receive_queue.lock);
634 h = (struct tpacket_hdr *)packet_lookup_frame(po, po->head);
635
636 if (h->tp_status)
637 goto ring_is_full;
638 po->head = po->head != po->frame_max ? po->head+1 : 0;
639 po->stats.tp_packets++;
640 if (copy_skb) {
641 status |= TP_STATUS_COPY;
642 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
643 }
644 if (!po->stats.tp_drops)
645 status &= ~TP_STATUS_LOSING;
646 spin_unlock(&sk->sk_receive_queue.lock);
647
648 memcpy((u8*)h + macoff, skb->data, snaplen);
649
650 h->tp_len = skb->len;
651 h->tp_snaplen = snaplen;
652 h->tp_mac = macoff;
653 h->tp_net = netoff;
Patrick McHardya61bbcf2005-08-14 17:24:31 -0700654 if (skb->tstamp.off_sec == 0) {
655 __net_timestamp(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 sock_enable_timestamp(sk);
657 }
Herbert Xu325ed822005-10-03 13:57:23 -0700658 h->tp_sec = skb->tstamp.off_sec;
659 h->tp_usec = skb->tstamp.off_usec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
661 sll = (struct sockaddr_ll*)((u8*)h + TPACKET_ALIGN(sizeof(*h)));
662 sll->sll_halen = 0;
663 if (dev->hard_header_parse)
664 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
665 sll->sll_family = AF_PACKET;
666 sll->sll_hatype = dev->type;
667 sll->sll_protocol = skb->protocol;
668 sll->sll_pkttype = skb->pkt_type;
669 sll->sll_ifindex = dev->ifindex;
670
671 h->tp_status = status;
672 mb();
673
674 {
675 struct page *p_start, *p_end;
676 u8 *h_end = (u8 *)h + macoff + snaplen - 1;
677
678 p_start = virt_to_page(h);
679 p_end = virt_to_page(h_end);
680 while (p_start <= p_end) {
681 flush_dcache_page(p_start);
682 p_start++;
683 }
684 }
685
686 sk->sk_data_ready(sk, 0);
687
688drop_n_restore:
689 if (skb_head != skb->data && skb_shared(skb)) {
690 skb->data = skb_head;
691 skb->len = skb_len;
692 }
693drop:
694 kfree_skb(skb);
695 return 0;
696
697ring_is_full:
698 po->stats.tp_drops++;
699 spin_unlock(&sk->sk_receive_queue.lock);
700
701 sk->sk_data_ready(sk, 0);
702 if (copy_skb)
703 kfree_skb(copy_skb);
704 goto drop_n_restore;
705}
706
707#endif
708
709
710static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
711 struct msghdr *msg, size_t len)
712{
713 struct sock *sk = sock->sk;
714 struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
715 struct sk_buff *skb;
716 struct net_device *dev;
717 unsigned short proto;
718 unsigned char *addr;
719 int ifindex, err, reserve = 0;
720
721 /*
722 * Get and verify the address.
723 */
724
725 if (saddr == NULL) {
726 struct packet_sock *po = pkt_sk(sk);
727
728 ifindex = po->ifindex;
729 proto = po->num;
730 addr = NULL;
731 } else {
732 err = -EINVAL;
733 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
734 goto out;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -0700735 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
736 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 ifindex = saddr->sll_ifindex;
738 proto = saddr->sll_protocol;
739 addr = saddr->sll_addr;
740 }
741
742
743 dev = dev_get_by_index(ifindex);
744 err = -ENXIO;
745 if (dev == NULL)
746 goto out_unlock;
747 if (sock->type == SOCK_RAW)
748 reserve = dev->hard_header_len;
749
750 err = -EMSGSIZE;
751 if (len > dev->mtu+reserve)
752 goto out_unlock;
753
754 skb = sock_alloc_send_skb(sk, len + LL_RESERVED_SPACE(dev),
755 msg->msg_flags & MSG_DONTWAIT, &err);
756 if (skb==NULL)
757 goto out_unlock;
758
759 skb_reserve(skb, LL_RESERVED_SPACE(dev));
760 skb->nh.raw = skb->data;
761
762 if (dev->hard_header) {
763 int res;
764 err = -EINVAL;
765 res = dev->hard_header(skb, dev, ntohs(proto), addr, NULL, len);
766 if (sock->type != SOCK_DGRAM) {
767 skb->tail = skb->data;
768 skb->len = 0;
769 } else if (res < 0)
770 goto out_free;
771 }
772
773 /* Returns -EFAULT on error */
774 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
775 if (err)
776 goto out_free;
777
778 skb->protocol = proto;
779 skb->dev = dev;
780 skb->priority = sk->sk_priority;
781
782 err = -ENETDOWN;
783 if (!(dev->flags & IFF_UP))
784 goto out_free;
785
786 /*
787 * Now send it
788 */
789
790 err = dev_queue_xmit(skb);
791 if (err > 0 && (err = net_xmit_errno(err)) != 0)
792 goto out_unlock;
793
794 dev_put(dev);
795
796 return(len);
797
798out_free:
799 kfree_skb(skb);
800out_unlock:
801 if (dev)
802 dev_put(dev);
803out:
804 return err;
805}
806
807/*
808 * Close a PACKET socket. This is fairly simple. We immediately go
809 * to 'closed' state and remove our protocol entry in the device list.
810 */
811
812static int packet_release(struct socket *sock)
813{
814 struct sock *sk = sock->sk;
815 struct packet_sock *po;
816
817 if (!sk)
818 return 0;
819
820 po = pkt_sk(sk);
821
822 write_lock_bh(&packet_sklist_lock);
823 sk_del_node_init(sk);
824 write_unlock_bh(&packet_sklist_lock);
825
826 /*
827 * Unhook packet receive handler.
828 */
829
830 if (po->running) {
831 /*
832 * Remove the protocol hook
833 */
834 dev_remove_pack(&po->prot_hook);
835 po->running = 0;
836 po->num = 0;
837 __sock_put(sk);
838 }
839
840#ifdef CONFIG_PACKET_MULTICAST
841 packet_flush_mclist(sk);
842#endif
843
844#ifdef CONFIG_PACKET_MMAP
845 if (po->pg_vec) {
846 struct tpacket_req req;
847 memset(&req, 0, sizeof(req));
848 packet_set_ring(sk, &req, 1);
849 }
850#endif
851
852 /*
853 * Now the socket is dead. No more input will appear.
854 */
855
856 sock_orphan(sk);
857 sock->sk = NULL;
858
859 /* Purge queues */
860
861 skb_queue_purge(&sk->sk_receive_queue);
862
863 sock_put(sk);
864 return 0;
865}
866
867/*
868 * Attach a packet hook.
869 */
870
871static int packet_do_bind(struct sock *sk, struct net_device *dev, int protocol)
872{
873 struct packet_sock *po = pkt_sk(sk);
874 /*
875 * Detach an existing hook if present.
876 */
877
878 lock_sock(sk);
879
880 spin_lock(&po->bind_lock);
881 if (po->running) {
882 __sock_put(sk);
883 po->running = 0;
884 po->num = 0;
885 spin_unlock(&po->bind_lock);
886 dev_remove_pack(&po->prot_hook);
887 spin_lock(&po->bind_lock);
888 }
889
890 po->num = protocol;
891 po->prot_hook.type = protocol;
892 po->prot_hook.dev = dev;
893
894 po->ifindex = dev ? dev->ifindex : 0;
895
896 if (protocol == 0)
897 goto out_unlock;
898
899 if (dev) {
900 if (dev->flags&IFF_UP) {
901 dev_add_pack(&po->prot_hook);
902 sock_hold(sk);
903 po->running = 1;
904 } else {
905 sk->sk_err = ENETDOWN;
906 if (!sock_flag(sk, SOCK_DEAD))
907 sk->sk_error_report(sk);
908 }
909 } else {
910 dev_add_pack(&po->prot_hook);
911 sock_hold(sk);
912 po->running = 1;
913 }
914
915out_unlock:
916 spin_unlock(&po->bind_lock);
917 release_sock(sk);
918 return 0;
919}
920
921/*
922 * Bind a packet socket to a device
923 */
924
925#ifdef CONFIG_SOCK_PACKET
926
927static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
928{
929 struct sock *sk=sock->sk;
930 char name[15];
931 struct net_device *dev;
932 int err = -ENODEV;
933
934 /*
935 * Check legality
936 */
937
938 if(addr_len!=sizeof(struct sockaddr))
939 return -EINVAL;
940 strlcpy(name,uaddr->sa_data,sizeof(name));
941
942 dev = dev_get_by_name(name);
943 if (dev) {
944 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
945 dev_put(dev);
946 }
947 return err;
948}
949#endif
950
951static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
952{
953 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
954 struct sock *sk=sock->sk;
955 struct net_device *dev = NULL;
956 int err;
957
958
959 /*
960 * Check legality
961 */
962
963 if (addr_len < sizeof(struct sockaddr_ll))
964 return -EINVAL;
965 if (sll->sll_family != AF_PACKET)
966 return -EINVAL;
967
968 if (sll->sll_ifindex) {
969 err = -ENODEV;
970 dev = dev_get_by_index(sll->sll_ifindex);
971 if (dev == NULL)
972 goto out;
973 }
974 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
975 if (dev)
976 dev_put(dev);
977
978out:
979 return err;
980}
981
982static struct proto packet_proto = {
983 .name = "PACKET",
984 .owner = THIS_MODULE,
985 .obj_size = sizeof(struct packet_sock),
986};
987
988/*
989 * Create a packet of type SOCK_PACKET.
990 */
991
992static int packet_create(struct socket *sock, int protocol)
993{
994 struct sock *sk;
995 struct packet_sock *po;
996 int err;
997
998 if (!capable(CAP_NET_RAW))
999 return -EPERM;
1000 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW
1001#ifdef CONFIG_SOCK_PACKET
1002 && sock->type != SOCK_PACKET
1003#endif
1004 )
1005 return -ESOCKTNOSUPPORT;
1006
1007 sock->state = SS_UNCONNECTED;
1008
1009 err = -ENOBUFS;
1010 sk = sk_alloc(PF_PACKET, GFP_KERNEL, &packet_proto, 1);
1011 if (sk == NULL)
1012 goto out;
1013
1014 sock->ops = &packet_ops;
1015#ifdef CONFIG_SOCK_PACKET
1016 if (sock->type == SOCK_PACKET)
1017 sock->ops = &packet_ops_spkt;
1018#endif
1019 sock_init_data(sock, sk);
1020
1021 po = pkt_sk(sk);
1022 sk->sk_family = PF_PACKET;
1023 po->num = protocol;
1024
1025 sk->sk_destruct = packet_sock_destruct;
1026 atomic_inc(&packet_socks_nr);
1027
1028 /*
1029 * Attach a protocol block
1030 */
1031
1032 spin_lock_init(&po->bind_lock);
1033 po->prot_hook.func = packet_rcv;
1034#ifdef CONFIG_SOCK_PACKET
1035 if (sock->type == SOCK_PACKET)
1036 po->prot_hook.func = packet_rcv_spkt;
1037#endif
1038 po->prot_hook.af_packet_priv = sk;
1039
1040 if (protocol) {
1041 po->prot_hook.type = protocol;
1042 dev_add_pack(&po->prot_hook);
1043 sock_hold(sk);
1044 po->running = 1;
1045 }
1046
1047 write_lock_bh(&packet_sklist_lock);
1048 sk_add_node(sk, &packet_sklist);
1049 write_unlock_bh(&packet_sklist_lock);
1050 return(0);
1051out:
1052 return err;
1053}
1054
1055/*
1056 * Pull a packet from our receive queue and hand it to the user.
1057 * If necessary we block.
1058 */
1059
1060static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1061 struct msghdr *msg, size_t len, int flags)
1062{
1063 struct sock *sk = sock->sk;
1064 struct sk_buff *skb;
1065 int copied, err;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001066 struct sockaddr_ll *sll;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067
1068 err = -EINVAL;
1069 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1070 goto out;
1071
1072#if 0
1073 /* What error should we return now? EUNATTACH? */
1074 if (pkt_sk(sk)->ifindex < 0)
1075 return -ENODEV;
1076#endif
1077
1078 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 * Call the generic datagram receiver. This handles all sorts
1080 * of horrible races and re-entrancy so we can forget about it
1081 * in the protocol layers.
1082 *
1083 * Now it will return ENETDOWN, if device have just gone down,
1084 * but then it will block.
1085 */
1086
1087 skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1088
1089 /*
1090 * An error occurred so return it. Because skb_recv_datagram()
1091 * handles the blocking we don't see and worry about blocking
1092 * retries.
1093 */
1094
1095 if(skb==NULL)
1096 goto out;
1097
1098 /*
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001099 * If the address length field is there to be filled in, we fill
1100 * it in now.
1101 */
1102
1103 sll = (struct sockaddr_ll*)skb->cb;
1104 if (sock->type == SOCK_PACKET)
1105 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1106 else
1107 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1108
1109 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 * You lose any data beyond the buffer you gave. If it worries a
1111 * user program they can ask the device for its MTU anyway.
1112 */
1113
1114 copied = skb->len;
1115 if (copied > len)
1116 {
1117 copied=len;
1118 msg->msg_flags|=MSG_TRUNC;
1119 }
1120
1121 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1122 if (err)
1123 goto out_free;
1124
1125 sock_recv_timestamp(msg, sk, skb);
1126
1127 if (msg->msg_name)
1128 memcpy(msg->msg_name, skb->cb, msg->msg_namelen);
1129
1130 /*
1131 * Free or return the buffer as appropriate. Again this
1132 * hides all the races and re-entrancy issues from us.
1133 */
1134 err = (flags&MSG_TRUNC) ? skb->len : copied;
1135
1136out_free:
1137 skb_free_datagram(sk, skb);
1138out:
1139 return err;
1140}
1141
1142#ifdef CONFIG_SOCK_PACKET
1143static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1144 int *uaddr_len, int peer)
1145{
1146 struct net_device *dev;
1147 struct sock *sk = sock->sk;
1148
1149 if (peer)
1150 return -EOPNOTSUPP;
1151
1152 uaddr->sa_family = AF_PACKET;
1153 dev = dev_get_by_index(pkt_sk(sk)->ifindex);
1154 if (dev) {
1155 strlcpy(uaddr->sa_data, dev->name, 15);
1156 dev_put(dev);
1157 } else
1158 memset(uaddr->sa_data, 0, 14);
1159 *uaddr_len = sizeof(*uaddr);
1160
1161 return 0;
1162}
1163#endif
1164
1165static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1166 int *uaddr_len, int peer)
1167{
1168 struct net_device *dev;
1169 struct sock *sk = sock->sk;
1170 struct packet_sock *po = pkt_sk(sk);
1171 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1172
1173 if (peer)
1174 return -EOPNOTSUPP;
1175
1176 sll->sll_family = AF_PACKET;
1177 sll->sll_ifindex = po->ifindex;
1178 sll->sll_protocol = po->num;
1179 dev = dev_get_by_index(po->ifindex);
1180 if (dev) {
1181 sll->sll_hatype = dev->type;
1182 sll->sll_halen = dev->addr_len;
1183 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1184 dev_put(dev);
1185 } else {
1186 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1187 sll->sll_halen = 0;
1188 }
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001189 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191 return 0;
1192}
1193
1194#ifdef CONFIG_PACKET_MULTICAST
1195static void packet_dev_mc(struct net_device *dev, struct packet_mclist *i, int what)
1196{
1197 switch (i->type) {
1198 case PACKET_MR_MULTICAST:
1199 if (what > 0)
1200 dev_mc_add(dev, i->addr, i->alen, 0);
1201 else
1202 dev_mc_delete(dev, i->addr, i->alen, 0);
1203 break;
1204 case PACKET_MR_PROMISC:
1205 dev_set_promiscuity(dev, what);
1206 break;
1207 case PACKET_MR_ALLMULTI:
1208 dev_set_allmulti(dev, what);
1209 break;
1210 default:;
1211 }
1212}
1213
1214static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1215{
1216 for ( ; i; i=i->next) {
1217 if (i->ifindex == dev->ifindex)
1218 packet_dev_mc(dev, i, what);
1219 }
1220}
1221
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001222static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223{
1224 struct packet_sock *po = pkt_sk(sk);
1225 struct packet_mclist *ml, *i;
1226 struct net_device *dev;
1227 int err;
1228
1229 rtnl_lock();
1230
1231 err = -ENODEV;
1232 dev = __dev_get_by_index(mreq->mr_ifindex);
1233 if (!dev)
1234 goto done;
1235
1236 err = -EINVAL;
1237 if (mreq->mr_alen > dev->addr_len)
1238 goto done;
1239
1240 err = -ENOBUFS;
Kris Katterjohn8b3a7002006-01-11 15:56:43 -08001241 i = kmalloc(sizeof(*i), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (i == NULL)
1243 goto done;
1244
1245 err = 0;
1246 for (ml = po->mclist; ml; ml = ml->next) {
1247 if (ml->ifindex == mreq->mr_ifindex &&
1248 ml->type == mreq->mr_type &&
1249 ml->alen == mreq->mr_alen &&
1250 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1251 ml->count++;
1252 /* Free the new element ... */
1253 kfree(i);
1254 goto done;
1255 }
1256 }
1257
1258 i->type = mreq->mr_type;
1259 i->ifindex = mreq->mr_ifindex;
1260 i->alen = mreq->mr_alen;
1261 memcpy(i->addr, mreq->mr_address, i->alen);
1262 i->count = 1;
1263 i->next = po->mclist;
1264 po->mclist = i;
1265 packet_dev_mc(dev, i, +1);
1266
1267done:
1268 rtnl_unlock();
1269 return err;
1270}
1271
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001272static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273{
1274 struct packet_mclist *ml, **mlp;
1275
1276 rtnl_lock();
1277
1278 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1279 if (ml->ifindex == mreq->mr_ifindex &&
1280 ml->type == mreq->mr_type &&
1281 ml->alen == mreq->mr_alen &&
1282 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1283 if (--ml->count == 0) {
1284 struct net_device *dev;
1285 *mlp = ml->next;
1286 dev = dev_get_by_index(ml->ifindex);
1287 if (dev) {
1288 packet_dev_mc(dev, ml, -1);
1289 dev_put(dev);
1290 }
1291 kfree(ml);
1292 }
1293 rtnl_unlock();
1294 return 0;
1295 }
1296 }
1297 rtnl_unlock();
1298 return -EADDRNOTAVAIL;
1299}
1300
1301static void packet_flush_mclist(struct sock *sk)
1302{
1303 struct packet_sock *po = pkt_sk(sk);
1304 struct packet_mclist *ml;
1305
1306 if (!po->mclist)
1307 return;
1308
1309 rtnl_lock();
1310 while ((ml = po->mclist) != NULL) {
1311 struct net_device *dev;
1312
1313 po->mclist = ml->next;
1314 if ((dev = dev_get_by_index(ml->ifindex)) != NULL) {
1315 packet_dev_mc(dev, ml, -1);
1316 dev_put(dev);
1317 }
1318 kfree(ml);
1319 }
1320 rtnl_unlock();
1321}
1322#endif
1323
1324static int
1325packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1326{
1327 struct sock *sk = sock->sk;
1328 int ret;
1329
1330 if (level != SOL_PACKET)
1331 return -ENOPROTOOPT;
1332
1333 switch(optname) {
1334#ifdef CONFIG_PACKET_MULTICAST
1335 case PACKET_ADD_MEMBERSHIP:
1336 case PACKET_DROP_MEMBERSHIP:
1337 {
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001338 struct packet_mreq_max mreq;
1339 int len = optlen;
1340 memset(&mreq, 0, sizeof(mreq));
1341 if (len < sizeof(struct packet_mreq))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 return -EINVAL;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001343 if (len > sizeof(mreq))
1344 len = sizeof(mreq);
1345 if (copy_from_user(&mreq,optval,len))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 return -EFAULT;
Eric W. Biederman0fb375f2005-09-21 00:11:37 -07001347 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1348 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 if (optname == PACKET_ADD_MEMBERSHIP)
1350 ret = packet_mc_add(sk, &mreq);
1351 else
1352 ret = packet_mc_drop(sk, &mreq);
1353 return ret;
1354 }
1355#endif
1356#ifdef CONFIG_PACKET_MMAP
1357 case PACKET_RX_RING:
1358 {
1359 struct tpacket_req req;
1360
1361 if (optlen<sizeof(req))
1362 return -EINVAL;
1363 if (copy_from_user(&req,optval,sizeof(req)))
1364 return -EFAULT;
1365 return packet_set_ring(sk, &req, 0);
1366 }
1367 case PACKET_COPY_THRESH:
1368 {
1369 int val;
1370
1371 if (optlen!=sizeof(val))
1372 return -EINVAL;
1373 if (copy_from_user(&val,optval,sizeof(val)))
1374 return -EFAULT;
1375
1376 pkt_sk(sk)->copy_thresh = val;
1377 return 0;
1378 }
1379#endif
1380 default:
1381 return -ENOPROTOOPT;
1382 }
1383}
1384
1385static int packet_getsockopt(struct socket *sock, int level, int optname,
1386 char __user *optval, int __user *optlen)
1387{
1388 int len;
1389 struct sock *sk = sock->sk;
1390 struct packet_sock *po = pkt_sk(sk);
1391
1392 if (level != SOL_PACKET)
1393 return -ENOPROTOOPT;
1394
1395 if (get_user(len,optlen))
1396 return -EFAULT;
1397
1398 if (len < 0)
1399 return -EINVAL;
1400
1401 switch(optname) {
1402 case PACKET_STATISTICS:
1403 {
1404 struct tpacket_stats st;
1405
1406 if (len > sizeof(struct tpacket_stats))
1407 len = sizeof(struct tpacket_stats);
1408 spin_lock_bh(&sk->sk_receive_queue.lock);
1409 st = po->stats;
1410 memset(&po->stats, 0, sizeof(st));
1411 spin_unlock_bh(&sk->sk_receive_queue.lock);
1412 st.tp_packets += st.tp_drops;
1413
1414 if (copy_to_user(optval, &st, len))
1415 return -EFAULT;
1416 break;
1417 }
1418 default:
1419 return -ENOPROTOOPT;
1420 }
1421
1422 if (put_user(len, optlen))
1423 return -EFAULT;
1424 return 0;
1425}
1426
1427
1428static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1429{
1430 struct sock *sk;
1431 struct hlist_node *node;
1432 struct net_device *dev = (struct net_device*)data;
1433
1434 read_lock(&packet_sklist_lock);
1435 sk_for_each(sk, node, &packet_sklist) {
1436 struct packet_sock *po = pkt_sk(sk);
1437
1438 switch (msg) {
1439 case NETDEV_UNREGISTER:
1440#ifdef CONFIG_PACKET_MULTICAST
1441 if (po->mclist)
1442 packet_dev_mclist(dev, po->mclist, -1);
1443 // fallthrough
1444#endif
1445 case NETDEV_DOWN:
1446 if (dev->ifindex == po->ifindex) {
1447 spin_lock(&po->bind_lock);
1448 if (po->running) {
1449 __dev_remove_pack(&po->prot_hook);
1450 __sock_put(sk);
1451 po->running = 0;
1452 sk->sk_err = ENETDOWN;
1453 if (!sock_flag(sk, SOCK_DEAD))
1454 sk->sk_error_report(sk);
1455 }
1456 if (msg == NETDEV_UNREGISTER) {
1457 po->ifindex = -1;
1458 po->prot_hook.dev = NULL;
1459 }
1460 spin_unlock(&po->bind_lock);
1461 }
1462 break;
1463 case NETDEV_UP:
1464 spin_lock(&po->bind_lock);
1465 if (dev->ifindex == po->ifindex && po->num &&
1466 !po->running) {
1467 dev_add_pack(&po->prot_hook);
1468 sock_hold(sk);
1469 po->running = 1;
1470 }
1471 spin_unlock(&po->bind_lock);
1472 break;
1473 }
1474 }
1475 read_unlock(&packet_sklist_lock);
1476 return NOTIFY_DONE;
1477}
1478
1479
1480static int packet_ioctl(struct socket *sock, unsigned int cmd,
1481 unsigned long arg)
1482{
1483 struct sock *sk = sock->sk;
1484
1485 switch(cmd) {
1486 case SIOCOUTQ:
1487 {
1488 int amount = atomic_read(&sk->sk_wmem_alloc);
1489 return put_user(amount, (int __user *)arg);
1490 }
1491 case SIOCINQ:
1492 {
1493 struct sk_buff *skb;
1494 int amount = 0;
1495
1496 spin_lock_bh(&sk->sk_receive_queue.lock);
1497 skb = skb_peek(&sk->sk_receive_queue);
1498 if (skb)
1499 amount = skb->len;
1500 spin_unlock_bh(&sk->sk_receive_queue.lock);
1501 return put_user(amount, (int __user *)arg);
1502 }
1503 case SIOCGSTAMP:
1504 return sock_get_timestamp(sk, (struct timeval __user *)arg);
1505
1506#ifdef CONFIG_INET
1507 case SIOCADDRT:
1508 case SIOCDELRT:
1509 case SIOCDARP:
1510 case SIOCGARP:
1511 case SIOCSARP:
1512 case SIOCGIFADDR:
1513 case SIOCSIFADDR:
1514 case SIOCGIFBRDADDR:
1515 case SIOCSIFBRDADDR:
1516 case SIOCGIFNETMASK:
1517 case SIOCSIFNETMASK:
1518 case SIOCGIFDSTADDR:
1519 case SIOCSIFDSTADDR:
1520 case SIOCSIFFLAGS:
1521 return inet_dgram_ops.ioctl(sock, cmd, arg);
1522#endif
1523
1524 default:
Christoph Hellwigb5e5fa52006-01-03 14:18:33 -08001525 return -ENOIOCTLCMD;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 }
1527 return 0;
1528}
1529
1530#ifndef CONFIG_PACKET_MMAP
1531#define packet_mmap sock_no_mmap
1532#define packet_poll datagram_poll
1533#else
1534
1535static unsigned int packet_poll(struct file * file, struct socket *sock,
1536 poll_table *wait)
1537{
1538 struct sock *sk = sock->sk;
1539 struct packet_sock *po = pkt_sk(sk);
1540 unsigned int mask = datagram_poll(file, sock, wait);
1541
1542 spin_lock_bh(&sk->sk_receive_queue.lock);
1543 if (po->pg_vec) {
1544 unsigned last = po->head ? po->head-1 : po->frame_max;
1545 struct tpacket_hdr *h;
1546
1547 h = (struct tpacket_hdr *)packet_lookup_frame(po, last);
1548
1549 if (h->tp_status)
1550 mask |= POLLIN | POLLRDNORM;
1551 }
1552 spin_unlock_bh(&sk->sk_receive_queue.lock);
1553 return mask;
1554}
1555
1556
1557/* Dirty? Well, I still did not learn better way to account
1558 * for user mmaps.
1559 */
1560
1561static void packet_mm_open(struct vm_area_struct *vma)
1562{
1563 struct file *file = vma->vm_file;
Eric Dumazetb69aee02005-09-06 14:42:45 -07001564 struct socket * sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 struct sock *sk = sock->sk;
1566
1567 if (sk)
1568 atomic_inc(&pkt_sk(sk)->mapped);
1569}
1570
1571static void packet_mm_close(struct vm_area_struct *vma)
1572{
1573 struct file *file = vma->vm_file;
Eric Dumazetb69aee02005-09-06 14:42:45 -07001574 struct socket * sock = file->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 struct sock *sk = sock->sk;
1576
1577 if (sk)
1578 atomic_dec(&pkt_sk(sk)->mapped);
1579}
1580
1581static struct vm_operations_struct packet_mmap_ops = {
1582 .open = packet_mm_open,
1583 .close =packet_mm_close,
1584};
1585
1586static inline struct page *pg_vec_endpage(char *one_pg_vec, unsigned int order)
1587{
1588 return virt_to_page(one_pg_vec + (PAGE_SIZE << order) - 1);
1589}
1590
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001591static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592{
1593 int i;
1594
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001595 for (i = 0; i < len; i++) {
1596 if (likely(pg_vec[i]))
1597 free_pages((unsigned long) pg_vec[i], order);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 }
1599 kfree(pg_vec);
1600}
1601
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001602static inline char *alloc_one_pg_vec_page(unsigned long order)
1603{
1604 return (char *) __get_free_pages(GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
1605 order);
1606}
1607
1608static char **alloc_pg_vec(struct tpacket_req *req, int order)
1609{
1610 unsigned int block_nr = req->tp_block_nr;
1611 char **pg_vec;
1612 int i;
1613
1614 pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
1615 if (unlikely(!pg_vec))
1616 goto out;
1617
1618 for (i = 0; i < block_nr; i++) {
1619 pg_vec[i] = alloc_one_pg_vec_page(order);
1620 if (unlikely(!pg_vec[i]))
1621 goto out_free_pgvec;
1622 }
1623
1624out:
1625 return pg_vec;
1626
1627out_free_pgvec:
1628 free_pg_vec(pg_vec, order, block_nr);
1629 pg_vec = NULL;
1630 goto out;
1631}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
1633static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
1634{
1635 char **pg_vec = NULL;
1636 struct packet_sock *po = pkt_sk(sk);
1637 int was_running, num, order = 0;
1638 int err = 0;
1639
1640 if (req->tp_block_nr) {
1641 int i, l;
1642
1643 /* Sanity tests and some calculations */
1644
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001645 if (unlikely(po->pg_vec))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 return -EBUSY;
1647
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001648 if (unlikely((int)req->tp_block_size <= 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 return -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001650 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 return -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001652 if (unlikely(req->tp_frame_size < TPACKET_HDRLEN))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 return -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001654 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 return -EINVAL;
1656
1657 po->frames_per_block = req->tp_block_size/req->tp_frame_size;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001658 if (unlikely(po->frames_per_block <= 0))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 return -EINVAL;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001660 if (unlikely((po->frames_per_block * req->tp_block_nr) !=
1661 req->tp_frame_nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
1664 err = -ENOMEM;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001665 order = get_order(req->tp_block_size);
1666 pg_vec = alloc_pg_vec(req, order);
1667 if (unlikely(!pg_vec))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669
1670 l = 0;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001671 for (i = 0; i < req->tp_block_nr; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 char *ptr = pg_vec[i];
1673 struct tpacket_hdr *header;
1674 int k;
1675
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001676 for (k = 0; k < po->frames_per_block; k++) {
1677 header = (struct tpacket_hdr *) ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 header->tp_status = TP_STATUS_KERNEL;
1679 ptr += req->tp_frame_size;
1680 }
1681 }
1682 /* Done */
1683 } else {
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001684 if (unlikely(req->tp_frame_nr))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 return -EINVAL;
1686 }
1687
1688 lock_sock(sk);
1689
1690 /* Detach socket from network */
1691 spin_lock(&po->bind_lock);
1692 was_running = po->running;
1693 num = po->num;
1694 if (was_running) {
1695 __dev_remove_pack(&po->prot_hook);
1696 po->num = 0;
1697 po->running = 0;
1698 __sock_put(sk);
1699 }
1700 spin_unlock(&po->bind_lock);
1701
1702 synchronize_net();
1703
1704 err = -EBUSY;
1705 if (closing || atomic_read(&po->mapped) == 0) {
1706 err = 0;
1707#define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
1708
1709 spin_lock_bh(&sk->sk_receive_queue.lock);
1710 pg_vec = XC(po->pg_vec, pg_vec);
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001711 po->frame_max = (req->tp_frame_nr - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 po->head = 0;
1713 po->frame_size = req->tp_frame_size;
1714 spin_unlock_bh(&sk->sk_receive_queue.lock);
1715
1716 order = XC(po->pg_vec_order, order);
1717 req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
1718
1719 po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
1720 po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
1721 skb_queue_purge(&sk->sk_receive_queue);
1722#undef XC
1723 if (atomic_read(&po->mapped))
1724 printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
1725 }
1726
1727 spin_lock(&po->bind_lock);
1728 if (was_running && !po->running) {
1729 sock_hold(sk);
1730 po->running = 1;
1731 po->num = num;
1732 dev_add_pack(&po->prot_hook);
1733 }
1734 spin_unlock(&po->bind_lock);
1735
1736 release_sock(sk);
1737
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 if (pg_vec)
1739 free_pg_vec(pg_vec, order, req->tp_block_nr);
1740out:
1741 return err;
1742}
1743
1744static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1745{
1746 struct sock *sk = sock->sk;
1747 struct packet_sock *po = pkt_sk(sk);
1748 unsigned long size;
1749 unsigned long start;
1750 int err = -EINVAL;
1751 int i;
1752
1753 if (vma->vm_pgoff)
1754 return -EINVAL;
1755
1756 size = vma->vm_end - vma->vm_start;
1757
1758 lock_sock(sk);
1759 if (po->pg_vec == NULL)
1760 goto out;
1761 if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
1762 goto out;
1763
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 start = vma->vm_start;
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001765 for (i = 0; i < po->pg_vec_len; i++) {
1766 struct page *page = virt_to_page(po->pg_vec[i]);
1767 int pg_num;
1768
1769 for (pg_num = 0; pg_num < po->pg_vec_pages; pg_num++, page++) {
1770 err = vm_insert_page(vma, start, page);
1771 if (unlikely(err))
1772 goto out;
1773 start += PAGE_SIZE;
1774 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 }
David S. Miller4ebf0ae2005-12-06 16:38:35 -08001776 atomic_inc(&po->mapped);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 vma->vm_ops = &packet_mmap_ops;
1778 err = 0;
1779
1780out:
1781 release_sock(sk);
1782 return err;
1783}
1784#endif
1785
1786
1787#ifdef CONFIG_SOCK_PACKET
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08001788static const struct proto_ops packet_ops_spkt = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 .family = PF_PACKET,
1790 .owner = THIS_MODULE,
1791 .release = packet_release,
1792 .bind = packet_bind_spkt,
1793 .connect = sock_no_connect,
1794 .socketpair = sock_no_socketpair,
1795 .accept = sock_no_accept,
1796 .getname = packet_getname_spkt,
1797 .poll = datagram_poll,
1798 .ioctl = packet_ioctl,
1799 .listen = sock_no_listen,
1800 .shutdown = sock_no_shutdown,
1801 .setsockopt = sock_no_setsockopt,
1802 .getsockopt = sock_no_getsockopt,
1803 .sendmsg = packet_sendmsg_spkt,
1804 .recvmsg = packet_recvmsg,
1805 .mmap = sock_no_mmap,
1806 .sendpage = sock_no_sendpage,
1807};
1808#endif
1809
Eric Dumazet90ddc4f2005-12-22 12:49:22 -08001810static const struct proto_ops packet_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 .family = PF_PACKET,
1812 .owner = THIS_MODULE,
1813 .release = packet_release,
1814 .bind = packet_bind,
1815 .connect = sock_no_connect,
1816 .socketpair = sock_no_socketpair,
1817 .accept = sock_no_accept,
1818 .getname = packet_getname,
1819 .poll = packet_poll,
1820 .ioctl = packet_ioctl,
1821 .listen = sock_no_listen,
1822 .shutdown = sock_no_shutdown,
1823 .setsockopt = packet_setsockopt,
1824 .getsockopt = packet_getsockopt,
1825 .sendmsg = packet_sendmsg,
1826 .recvmsg = packet_recvmsg,
1827 .mmap = packet_mmap,
1828 .sendpage = sock_no_sendpage,
1829};
1830
1831static struct net_proto_family packet_family_ops = {
1832 .family = PF_PACKET,
1833 .create = packet_create,
1834 .owner = THIS_MODULE,
1835};
1836
1837static struct notifier_block packet_netdev_notifier = {
1838 .notifier_call =packet_notifier,
1839};
1840
1841#ifdef CONFIG_PROC_FS
1842static inline struct sock *packet_seq_idx(loff_t off)
1843{
1844 struct sock *s;
1845 struct hlist_node *node;
1846
1847 sk_for_each(s, node, &packet_sklist) {
1848 if (!off--)
1849 return s;
1850 }
1851 return NULL;
1852}
1853
1854static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
1855{
1856 read_lock(&packet_sklist_lock);
1857 return *pos ? packet_seq_idx(*pos - 1) : SEQ_START_TOKEN;
1858}
1859
1860static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1861{
1862 ++*pos;
1863 return (v == SEQ_START_TOKEN)
1864 ? sk_head(&packet_sklist)
1865 : sk_next((struct sock*)v) ;
1866}
1867
1868static void packet_seq_stop(struct seq_file *seq, void *v)
1869{
1870 read_unlock(&packet_sklist_lock);
1871}
1872
1873static int packet_seq_show(struct seq_file *seq, void *v)
1874{
1875 if (v == SEQ_START_TOKEN)
1876 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
1877 else {
1878 struct sock *s = v;
1879 const struct packet_sock *po = pkt_sk(s);
1880
1881 seq_printf(seq,
1882 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
1883 s,
1884 atomic_read(&s->sk_refcnt),
1885 s->sk_type,
1886 ntohs(po->num),
1887 po->ifindex,
1888 po->running,
1889 atomic_read(&s->sk_rmem_alloc),
1890 sock_i_uid(s),
1891 sock_i_ino(s) );
1892 }
1893
1894 return 0;
1895}
1896
1897static struct seq_operations packet_seq_ops = {
1898 .start = packet_seq_start,
1899 .next = packet_seq_next,
1900 .stop = packet_seq_stop,
1901 .show = packet_seq_show,
1902};
1903
1904static int packet_seq_open(struct inode *inode, struct file *file)
1905{
1906 return seq_open(file, &packet_seq_ops);
1907}
1908
1909static struct file_operations packet_seq_fops = {
1910 .owner = THIS_MODULE,
1911 .open = packet_seq_open,
1912 .read = seq_read,
1913 .llseek = seq_lseek,
1914 .release = seq_release,
1915};
1916
1917#endif
1918
1919static void __exit packet_exit(void)
1920{
1921 proc_net_remove("packet");
1922 unregister_netdevice_notifier(&packet_netdev_notifier);
1923 sock_unregister(PF_PACKET);
1924 proto_unregister(&packet_proto);
1925}
1926
1927static int __init packet_init(void)
1928{
1929 int rc = proto_register(&packet_proto, 0);
1930
1931 if (rc != 0)
1932 goto out;
1933
1934 sock_register(&packet_family_ops);
1935 register_netdevice_notifier(&packet_netdev_notifier);
1936 proc_net_fops_create("packet", 0, &packet_seq_fops);
1937out:
1938 return rc;
1939}
1940
1941module_init(packet_init);
1942module_exit(packet_exit);
1943MODULE_LICENSE("GPL");
1944MODULE_ALIAS_NETPROTO(PF_PACKET);