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.
6 * PACKET - implements raw packet sockets.
8 * Version: $Id: af_packet.c,v 1.61 2002/02/08 03:57:19 davem Exp $
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Alan Cox, <gw4pts@gw4pts.ampr.org>
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.
40 * This program is free software; you can redistribute it and/or
41 * modify it under the terms of the GNU General Public License
42 * as published by the Free Software Foundation; either version
43 * 2 of the License, or (at your option) any later version.
47 #include <linux/config.h>
48 #include <linux/types.h>
49 #include <linux/sched.h>
51 #include <linux/fcntl.h>
52 #include <linux/socket.h>
54 #include <linux/inet.h>
55 #include <linux/netdevice.h>
56 #include <linux/if_packet.h>
57 #include <linux/wireless.h>
58 #include <linux/kmod.h>
60 #include <net/protocol.h>
61 #include <linux/skbuff.h>
63 #include <linux/errno.h>
64 #include <linux/timer.h>
65 #include <asm/system.h>
66 #include <asm/uaccess.h>
67 #include <asm/ioctls.h>
70 #include <linux/proc_fs.h>
71 #include <linux/seq_file.h>
72 #include <linux/poll.h>
73 #include <linux/module.h>
74 #include <linux/init.h>
77 #include <net/inet_common.h>
80 #define CONFIG_SOCK_PACKET 1
83 Proposed replacement for SIOC{ADD,DEL}MULTI and
84 IFF_PROMISC, IFF_ALLMULTI flags.
86 It is more expensive, but I believe,
87 it is really correct solution: reentereble, safe and fault tolerant.
89 IFF_PROMISC/IFF_ALLMULTI/SIOC{ADD/DEL}MULTI are faked by keeping
90 reference count and global flag, so that real status is
91 (gflag|(count != 0)), so that we can use obsolete faulty interface
92 not harming clever users.
94 #define CONFIG_PACKET_MULTICAST 1
98 - if device has no dev->hard_header routine, it adds and removes ll header
99 inside itself. In this case ll header is invisible outside of device,
100 but higher levels still should reserve dev->hard_header_len.
101 Some devices are enough clever to reallocate skb, when header
102 will not fit to reserved space (tunnel), another ones are silly
104 - packet socket receives packets with pulled ll header,
105 so that SOCK_RAW should push it back.
110 Incoming, dev->hard_header!=NULL
114 Outgoing, dev->hard_header!=NULL
118 Incoming, dev->hard_header==NULL
119 mac.raw -> UNKNOWN position. It is very likely, that it points to ll header.
120 PPP makes it, that is wrong, because introduce assymetry
121 between rx and tx paths.
124 Outgoing, dev->hard_header==NULL
125 mac.raw -> data. ll header is still not built!
129 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
135 dev->hard_header != NULL
139 dev->hard_header == NULL (ll header is added by device, we cannot control it)
143 We should set nh.raw on output to correct posistion,
144 packet classifier depends on it.
147 /* List of all packet sockets. */
148 HLIST_HEAD(packet_sklist);
149 static rwlock_t packet_sklist_lock = RW_LOCK_UNLOCKED;
151 atomic_t packet_socks_nr;
154 /* Private packet socket structures. */
156 #ifdef CONFIG_PACKET_MULTICAST
159 struct packet_mclist *next;
164 unsigned char addr[8];
167 #ifdef CONFIG_PACKET_MMAP
168 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
171 static void packet_flush_mclist(struct sock *sk);
175 struct tpacket_stats stats;
176 #ifdef CONFIG_PACKET_MMAP
179 unsigned int frames_per_block;
180 unsigned int frame_size;
181 unsigned int frame_max;
184 struct packet_type prot_hook;
185 spinlock_t bind_lock;
186 char running; /* prot_hook is attached*/
187 int ifindex; /* bound device */
189 #ifdef CONFIG_PACKET_MULTICAST
190 struct packet_mclist *mclist;
192 #ifdef CONFIG_PACKET_MMAP
194 unsigned int pg_vec_order;
195 unsigned int pg_vec_pages;
196 unsigned int pg_vec_len;
200 #ifdef CONFIG_PACKET_MMAP
202 static inline char *packet_lookup_frame(struct packet_opt *po, unsigned int position)
204 unsigned int pg_vec_pos, frame_offset;
207 pg_vec_pos = position / po->frames_per_block;
208 frame_offset = position % po->frames_per_block;
210 frame = po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size);
216 #define pkt_sk(__sk) ((struct packet_opt *)(__sk)->sk_protinfo)
218 void packet_sock_destruct(struct sock *sk)
220 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
221 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
222 BUG_ON(sk->sk_nx_info);
223 BUG_ON(sk->sk_vx_info);
225 if (!sock_flag(sk, SOCK_DEAD)) {
226 printk("Attempt to release alive packet socket: %p\n", sk);
232 atomic_dec(&packet_socks_nr);
233 #ifdef PACKET_REFCNT_DEBUG
234 printk(KERN_DEBUG "PACKET socket %p is free, %d are alive\n", sk, atomic_read(&packet_socks_nr));
239 extern struct proto_ops packet_ops;
241 #ifdef CONFIG_SOCK_PACKET
242 extern struct proto_ops packet_ops_spkt;
244 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt)
247 struct sockaddr_pkt *spkt;
250 * When we registered the protocol we saved the socket in the data
251 * field for just this event.
254 sk = pt->af_packet_priv;
257 * Yank back the headers [hope the device set this
258 * right or kerboom...]
260 * Incoming packets have ll header pulled,
263 * For outgoing ones skb->data == skb->mac.raw
264 * so that this procedure is noop.
267 if (skb->pkt_type == PACKET_LOOPBACK)
270 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
273 /* drop any routing info */
274 dst_release(skb->dst);
277 spkt = (struct sockaddr_pkt*)skb->cb;
279 skb_push(skb, skb->data-skb->mac.raw);
282 * The SOCK_PACKET socket receives _all_ frames.
285 spkt->spkt_family = dev->type;
286 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
287 spkt->spkt_protocol = skb->protocol;
290 * Charge the memory to the socket. This is done specifically
291 * to prevent sockets using all the memory up.
294 if (sock_queue_rcv_skb(sk,skb) == 0)
305 * Output a raw packet to a device layer. This bypasses all the other
306 * protocol layers and you must therefore supply it with a complete frame
309 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
310 struct msghdr *msg, size_t len)
312 struct sock *sk = sock->sk;
313 struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
315 struct net_device *dev;
316 unsigned short proto=0;
320 * Get and verify the address.
325 if (msg->msg_namelen < sizeof(struct sockaddr))
327 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
328 proto=saddr->spkt_protocol;
331 return(-ENOTCONN); /* SOCK_PACKET must be sent giving an address */
334 * Find the device first to size check it
337 saddr->spkt_device[13] = 0;
338 dev = dev_get_by_name(saddr->spkt_device);
344 * You may not queue a frame bigger than the mtu. This is the lowest level
345 * raw protocol and you must do your own fragmentation at this level.
349 if(len>dev->mtu+dev->hard_header_len)
353 skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
356 * If the write buffer is full, then tough. At this level the user gets to
357 * deal with the problem - do your own algorithmic backoffs. That's far
368 /* FIXME: Save some space for broken drivers that write a
369 * hard header at transmission time by themselves. PPP is the
370 * notable one here. This should really be fixed at the driver level.
372 skb_reserve(skb, LL_RESERVED_SPACE(dev));
373 skb->nh.raw = skb->data;
375 /* Try to align data part correctly */
376 if (dev->hard_header) {
377 skb->data -= dev->hard_header_len;
378 skb->tail -= dev->hard_header_len;
379 if (len < dev->hard_header_len)
380 skb->nh.raw = skb->data;
383 /* Returns -EFAULT on error */
384 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
385 skb->protocol = proto;
387 skb->priority = sk->sk_priority;
392 if (!(dev->flags & IFF_UP))
412 static inline unsigned run_filter(struct sk_buff *skb, struct sock *sk, unsigned res)
414 struct sk_filter *filter;
417 filter = sk->sk_filter;
419 * Our caller already checked that filter != NULL but we need to
420 * verify that under bh_lock_sock() to be safe
422 if (likely(filter != NULL))
423 res = sk_run_filter(skb, filter->insns, filter->len);
430 This function makes lazy skb cloning in hope that most of packets
431 are discarded by BPF.
433 Note tricky part: we DO mangle shared skb! skb->data, skb->len
434 and skb->cb are mangled. It works because (and until) packets
435 falling here are owned by current CPU. Output packets are cloned
436 by dev_queue_xmit_nit(), input packets are processed by net_bh
437 sequencially, so that if we return skb to original state on exit,
438 we will not harm anyone.
441 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt)
444 struct sockaddr_ll *sll;
445 struct packet_opt *po;
446 u8 * skb_head = skb->data;
447 int skb_len = skb->len;
450 if (skb->pkt_type == PACKET_LOOPBACK)
453 sk = pt->af_packet_priv;
456 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
458 (int) sk->sk_xid > 0 && sk->sk_xid != skb->xid)
464 if (dev->hard_header) {
465 /* The device has an explicit notion of ll header,
466 exported to higher levels.
468 Otherwise, the device hides datails of it frame
469 structure, so that corresponding packet head
470 never delivered to user.
472 if (sk->sk_type != SOCK_DGRAM)
473 skb_push(skb, skb->data - skb->mac.raw);
474 else if (skb->pkt_type == PACKET_OUTGOING) {
475 /* Special case: outgoing packets have ll header at head */
476 skb_pull(skb, skb->nh.raw - skb->data);
483 unsigned res = run_filter(skb, sk, snaplen);
490 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
491 (unsigned)sk->sk_rcvbuf)
494 if (skb_shared(skb)) {
495 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
499 if (skb_head != skb->data) {
500 skb->data = skb_head;
507 sll = (struct sockaddr_ll*)skb->cb;
508 sll->sll_family = AF_PACKET;
509 sll->sll_hatype = dev->type;
510 sll->sll_protocol = skb->protocol;
511 sll->sll_pkttype = skb->pkt_type;
512 sll->sll_ifindex = dev->ifindex;
515 if (dev->hard_header_parse)
516 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
518 if (pskb_trim(skb, snaplen))
521 skb_set_owner_r(skb, sk);
523 dst_release(skb->dst);
526 spin_lock(&sk->sk_receive_queue.lock);
527 po->stats.tp_packets++;
528 __skb_queue_tail(&sk->sk_receive_queue, skb);
529 spin_unlock(&sk->sk_receive_queue.lock);
530 sk->sk_data_ready(sk, skb->len);
534 spin_lock(&sk->sk_receive_queue.lock);
535 po->stats.tp_drops++;
536 spin_unlock(&sk->sk_receive_queue.lock);
539 if (skb_head != skb->data && skb_shared(skb)) {
540 skb->data = skb_head;
548 #ifdef CONFIG_PACKET_MMAP
549 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt)
552 struct packet_opt *po;
553 struct sockaddr_ll *sll;
554 struct tpacket_hdr *h;
555 u8 * skb_head = skb->data;
556 int skb_len = skb->len;
558 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
559 unsigned short macoff, netoff;
560 struct sk_buff *copy_skb = NULL;
562 if (skb->pkt_type == PACKET_LOOPBACK)
565 sk = pt->af_packet_priv;
568 if (dev->hard_header) {
569 if (sk->sk_type != SOCK_DGRAM)
570 skb_push(skb, skb->data - skb->mac.raw);
571 else if (skb->pkt_type == PACKET_OUTGOING) {
572 /* Special case: outgoing packets have ll header at head */
573 skb_pull(skb, skb->nh.raw - skb->data);
574 if (skb->ip_summed == CHECKSUM_HW)
575 status |= TP_STATUS_CSUMNOTREADY;
582 unsigned res = run_filter(skb, sk, snaplen);
589 if (sk->sk_type == SOCK_DGRAM) {
590 macoff = netoff = TPACKET_ALIGN(TPACKET_HDRLEN) + 16;
592 unsigned maclen = skb->nh.raw - skb->data;
593 netoff = TPACKET_ALIGN(TPACKET_HDRLEN + (maclen < 16 ? 16 : maclen));
594 macoff = netoff - maclen;
597 if (macoff + snaplen > po->frame_size) {
598 if (po->copy_thresh &&
599 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
600 (unsigned)sk->sk_rcvbuf) {
601 if (skb_shared(skb)) {
602 copy_skb = skb_clone(skb, GFP_ATOMIC);
604 copy_skb = skb_get(skb);
605 skb_head = skb->data;
608 skb_set_owner_r(copy_skb, sk);
610 snaplen = po->frame_size - macoff;
611 if ((int)snaplen < 0)
614 if (snaplen > skb->len-skb->data_len)
615 snaplen = skb->len-skb->data_len;
617 spin_lock(&sk->sk_receive_queue.lock);
618 h = (struct tpacket_hdr *)packet_lookup_frame(po, po->head);
622 po->head = po->head != po->frame_max ? po->head+1 : 0;
623 po->stats.tp_packets++;
625 status |= TP_STATUS_COPY;
626 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
628 if (!po->stats.tp_drops)
629 status &= ~TP_STATUS_LOSING;
630 spin_unlock(&sk->sk_receive_queue.lock);
632 memcpy((u8*)h + macoff, skb->data, snaplen);
634 h->tp_len = skb->len;
635 h->tp_snaplen = snaplen;
638 if (skb->stamp.tv_sec == 0) {
639 do_gettimeofday(&skb->stamp);
640 sock_enable_timestamp(sk);
642 h->tp_sec = skb->stamp.tv_sec;
643 h->tp_usec = skb->stamp.tv_usec;
645 sll = (struct sockaddr_ll*)((u8*)h + TPACKET_ALIGN(sizeof(*h)));
647 if (dev->hard_header_parse)
648 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
649 sll->sll_family = AF_PACKET;
650 sll->sll_hatype = dev->type;
651 sll->sll_protocol = skb->protocol;
652 sll->sll_pkttype = skb->pkt_type;
653 sll->sll_ifindex = dev->ifindex;
655 h->tp_status = status;
659 struct page *p_start, *p_end;
660 u8 *h_end = (u8 *)h + macoff + snaplen - 1;
662 p_start = virt_to_page(h);
663 p_end = virt_to_page(h_end);
664 while (p_start <= p_end) {
665 flush_dcache_page(p_start);
670 sk->sk_data_ready(sk, 0);
673 if (skb_head != skb->data && skb_shared(skb)) {
674 skb->data = skb_head;
682 po->stats.tp_drops++;
683 spin_unlock(&sk->sk_receive_queue.lock);
685 sk->sk_data_ready(sk, 0);
694 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
695 struct msghdr *msg, size_t len)
697 struct sock *sk = sock->sk;
698 struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
700 struct net_device *dev;
701 unsigned short proto;
703 int ifindex, err, reserve = 0;
706 * Get and verify the address.
710 struct packet_opt *po = pkt_sk(sk);
712 ifindex = po->ifindex;
717 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
719 ifindex = saddr->sll_ifindex;
720 proto = saddr->sll_protocol;
721 addr = saddr->sll_addr;
725 dev = dev_get_by_index(ifindex);
729 if (sock->type == SOCK_RAW)
730 reserve = dev->hard_header_len;
733 if (len > dev->mtu+reserve)
736 skb = sock_alloc_send_skb(sk, len + LL_RESERVED_SPACE(dev),
737 msg->msg_flags & MSG_DONTWAIT, &err);
741 skb_reserve(skb, LL_RESERVED_SPACE(dev));
742 skb->nh.raw = skb->data;
744 if (dev->hard_header) {
747 res = dev->hard_header(skb, dev, ntohs(proto), addr, NULL, len);
748 if (sock->type != SOCK_DGRAM) {
749 skb->tail = skb->data;
755 /* Returns -EFAULT on error */
756 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
760 skb->protocol = proto;
762 skb->priority = sk->sk_priority;
765 if (!(dev->flags & IFF_UP))
772 err = dev_queue_xmit(skb);
773 if (err > 0 && (err = net_xmit_errno(err)) != 0)
790 * Close a PACKET socket. This is fairly simple. We immediately go
791 * to 'closed' state and remove our protocol entry in the device list.
794 static int packet_release(struct socket *sock)
796 struct sock *sk = sock->sk;
797 struct packet_opt *po;
804 write_lock_bh(&packet_sklist_lock);
805 sk_del_node_init(sk);
806 write_unlock_bh(&packet_sklist_lock);
809 * Unhook packet receive handler.
814 * Remove the protocol hook
816 dev_remove_pack(&po->prot_hook);
822 #ifdef CONFIG_PACKET_MULTICAST
823 packet_flush_mclist(sk);
826 #ifdef CONFIG_PACKET_MMAP
828 struct tpacket_req req;
829 memset(&req, 0, sizeof(req));
830 packet_set_ring(sk, &req, 1);
834 clr_vx_info(&sk->sk_vx_info);
835 clr_nx_info(&sk->sk_nx_info);
838 * Now the socket is dead. No more input will appear.
846 skb_queue_purge(&sk->sk_receive_queue);
853 * Attach a packet hook.
856 static int packet_do_bind(struct sock *sk, struct net_device *dev, int protocol)
858 struct packet_opt *po = pkt_sk(sk);
860 * Detach an existing hook if present.
865 spin_lock(&po->bind_lock);
870 spin_unlock(&po->bind_lock);
871 dev_remove_pack(&po->prot_hook);
872 spin_lock(&po->bind_lock);
876 po->prot_hook.type = protocol;
877 po->prot_hook.dev = dev;
879 po->ifindex = dev ? dev->ifindex : 0;
885 if (dev->flags&IFF_UP) {
886 dev_add_pack(&po->prot_hook);
890 sk->sk_err = ENETDOWN;
891 if (!sock_flag(sk, SOCK_DEAD))
892 sk->sk_error_report(sk);
895 dev_add_pack(&po->prot_hook);
901 spin_unlock(&po->bind_lock);
907 * Bind a packet socket to a device
910 #ifdef CONFIG_SOCK_PACKET
912 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
914 struct sock *sk=sock->sk;
916 struct net_device *dev;
923 if(addr_len!=sizeof(struct sockaddr))
925 strlcpy(name,uaddr->sa_data,sizeof(name));
927 dev = dev_get_by_name(name);
929 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
936 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
938 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
939 struct sock *sk=sock->sk;
940 struct net_device *dev = NULL;
948 if (addr_len < sizeof(struct sockaddr_ll))
950 if (sll->sll_family != AF_PACKET)
953 if (sll->sll_ifindex) {
955 dev = dev_get_by_index(sll->sll_ifindex);
959 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
969 * Create a packet of type SOCK_PACKET.
972 static int packet_create(struct socket *sock, int protocol)
975 struct packet_opt *po;
978 if (!capable(CAP_NET_RAW))
980 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW
981 #ifdef CONFIG_SOCK_PACKET
982 && sock->type != SOCK_PACKET
985 return -ESOCKTNOSUPPORT;
987 sock->state = SS_UNCONNECTED;
990 sk = sk_alloc(PF_PACKET, GFP_KERNEL, 1, NULL);
994 sock->ops = &packet_ops;
995 #ifdef CONFIG_SOCK_PACKET
996 if (sock->type == SOCK_PACKET)
997 sock->ops = &packet_ops_spkt;
999 sock_init_data(sock,sk);
1000 sk_set_owner(sk, THIS_MODULE);
1002 po = sk->sk_protinfo = kmalloc(sizeof(*po), GFP_KERNEL);
1005 memset(po, 0, sizeof(*po));
1006 sk->sk_family = PF_PACKET;
1009 sk->sk_destruct = packet_sock_destruct;
1010 atomic_inc(&packet_socks_nr);
1012 set_vx_info(&sk->sk_vx_info, current->vx_info);
1013 sk->sk_xid = vx_current_xid();
1014 set_nx_info(&sk->sk_nx_info, current->nx_info);
1015 sk->sk_nid = nx_current_nid();
1018 * Attach a protocol block
1021 spin_lock_init(&po->bind_lock);
1022 po->prot_hook.func = packet_rcv;
1023 #ifdef CONFIG_SOCK_PACKET
1024 if (sock->type == SOCK_PACKET)
1025 po->prot_hook.func = packet_rcv_spkt;
1027 po->prot_hook.af_packet_priv = sk;
1030 po->prot_hook.type = protocol;
1031 dev_add_pack(&po->prot_hook);
1036 write_lock_bh(&packet_sklist_lock);
1037 sk_add_node(sk, &packet_sklist);
1038 write_unlock_bh(&packet_sklist_lock);
1048 * Pull a packet from our receive queue and hand it to the user.
1049 * If necessary we block.
1052 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1053 struct msghdr *msg, size_t len, int flags)
1055 struct sock *sk = sock->sk;
1056 struct sk_buff *skb;
1060 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1064 /* What error should we return now? EUNATTACH? */
1065 if (pkt_sk(sk)->ifindex < 0)
1070 * If the address length field is there to be filled in, we fill
1074 if (sock->type == SOCK_PACKET)
1075 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1077 msg->msg_namelen = sizeof(struct sockaddr_ll);
1080 * Call the generic datagram receiver. This handles all sorts
1081 * of horrible races and re-entrancy so we can forget about it
1082 * in the protocol layers.
1084 * Now it will return ENETDOWN, if device have just gone down,
1085 * but then it will block.
1088 skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1091 * An error occurred so return it. Because skb_recv_datagram()
1092 * handles the blocking we don't see and worry about blocking
1100 * You lose any data beyond the buffer you gave. If it worries a
1101 * user program they can ask the device for its MTU anyway.
1108 msg->msg_flags|=MSG_TRUNC;
1111 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1115 sock_recv_timestamp(msg, sk, skb);
1118 memcpy(msg->msg_name, skb->cb, msg->msg_namelen);
1121 * Free or return the buffer as appropriate. Again this
1122 * hides all the races and re-entrancy issues from us.
1124 err = (flags&MSG_TRUNC) ? skb->len : copied;
1127 skb_free_datagram(sk, skb);
1132 #ifdef CONFIG_SOCK_PACKET
1133 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1134 int *uaddr_len, int peer)
1136 struct net_device *dev;
1137 struct sock *sk = sock->sk;
1142 uaddr->sa_family = AF_PACKET;
1143 dev = dev_get_by_index(pkt_sk(sk)->ifindex);
1145 strlcpy(uaddr->sa_data, dev->name, 15);
1148 memset(uaddr->sa_data, 0, 14);
1149 *uaddr_len = sizeof(*uaddr);
1155 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1156 int *uaddr_len, int peer)
1158 struct net_device *dev;
1159 struct sock *sk = sock->sk;
1160 struct packet_opt *po = pkt_sk(sk);
1161 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1166 sll->sll_family = AF_PACKET;
1167 sll->sll_ifindex = po->ifindex;
1168 sll->sll_protocol = po->num;
1169 dev = dev_get_by_index(po->ifindex);
1171 sll->sll_hatype = dev->type;
1172 sll->sll_halen = dev->addr_len;
1173 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1176 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1179 *uaddr_len = sizeof(*sll);
1184 #ifdef CONFIG_PACKET_MULTICAST
1185 static void packet_dev_mc(struct net_device *dev, struct packet_mclist *i, int what)
1188 case PACKET_MR_MULTICAST:
1190 dev_mc_add(dev, i->addr, i->alen, 0);
1192 dev_mc_delete(dev, i->addr, i->alen, 0);
1194 case PACKET_MR_PROMISC:
1195 dev_set_promiscuity(dev, what);
1197 case PACKET_MR_ALLMULTI:
1198 dev_set_allmulti(dev, what);
1204 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1206 for ( ; i; i=i->next) {
1207 if (i->ifindex == dev->ifindex)
1208 packet_dev_mc(dev, i, what);
1212 static int packet_mc_add(struct sock *sk, struct packet_mreq *mreq)
1214 struct packet_opt *po = pkt_sk(sk);
1215 struct packet_mclist *ml, *i;
1216 struct net_device *dev;
1222 dev = __dev_get_by_index(mreq->mr_ifindex);
1227 if (mreq->mr_alen > dev->addr_len)
1231 i = (struct packet_mclist *)kmalloc(sizeof(*i), GFP_KERNEL);
1236 for (ml = po->mclist; ml; ml = ml->next) {
1237 if (ml->ifindex == mreq->mr_ifindex &&
1238 ml->type == mreq->mr_type &&
1239 ml->alen == mreq->mr_alen &&
1240 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1242 /* Free the new element ... */
1248 i->type = mreq->mr_type;
1249 i->ifindex = mreq->mr_ifindex;
1250 i->alen = mreq->mr_alen;
1251 memcpy(i->addr, mreq->mr_address, i->alen);
1253 i->next = po->mclist;
1255 packet_dev_mc(dev, i, +1);
1262 static int packet_mc_drop(struct sock *sk, struct packet_mreq *mreq)
1264 struct packet_mclist *ml, **mlp;
1268 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1269 if (ml->ifindex == mreq->mr_ifindex &&
1270 ml->type == mreq->mr_type &&
1271 ml->alen == mreq->mr_alen &&
1272 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1273 if (--ml->count == 0) {
1274 struct net_device *dev;
1276 dev = dev_get_by_index(ml->ifindex);
1278 packet_dev_mc(dev, ml, -1);
1288 return -EADDRNOTAVAIL;
1291 static void packet_flush_mclist(struct sock *sk)
1293 struct packet_opt *po = pkt_sk(sk);
1294 struct packet_mclist *ml;
1300 while ((ml = po->mclist) != NULL) {
1301 struct net_device *dev;
1303 po->mclist = ml->next;
1304 if ((dev = dev_get_by_index(ml->ifindex)) != NULL) {
1305 packet_dev_mc(dev, ml, -1);
1315 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1317 struct sock *sk = sock->sk;
1320 if (level != SOL_PACKET)
1321 return -ENOPROTOOPT;
1324 #ifdef CONFIG_PACKET_MULTICAST
1325 case PACKET_ADD_MEMBERSHIP:
1326 case PACKET_DROP_MEMBERSHIP:
1328 struct packet_mreq mreq;
1329 if (optlen<sizeof(mreq))
1331 if (copy_from_user(&mreq,optval,sizeof(mreq)))
1333 if (optname == PACKET_ADD_MEMBERSHIP)
1334 ret = packet_mc_add(sk, &mreq);
1336 ret = packet_mc_drop(sk, &mreq);
1340 #ifdef CONFIG_PACKET_MMAP
1341 case PACKET_RX_RING:
1343 struct tpacket_req req;
1345 if (optlen<sizeof(req))
1347 if (copy_from_user(&req,optval,sizeof(req)))
1349 return packet_set_ring(sk, &req, 0);
1351 case PACKET_COPY_THRESH:
1355 if (optlen!=sizeof(val))
1357 if (copy_from_user(&val,optval,sizeof(val)))
1360 pkt_sk(sk)->copy_thresh = val;
1365 return -ENOPROTOOPT;
1369 int packet_getsockopt(struct socket *sock, int level, int optname,
1370 char __user *optval, int __user *optlen)
1373 struct sock *sk = sock->sk;
1374 struct packet_opt *po = pkt_sk(sk);
1376 if (level != SOL_PACKET)
1377 return -ENOPROTOOPT;
1379 if (get_user(len,optlen))
1386 case PACKET_STATISTICS:
1388 struct tpacket_stats st;
1390 if (len > sizeof(struct tpacket_stats))
1391 len = sizeof(struct tpacket_stats);
1392 spin_lock_bh(&sk->sk_receive_queue.lock);
1394 memset(&po->stats, 0, sizeof(st));
1395 spin_unlock_bh(&sk->sk_receive_queue.lock);
1396 st.tp_packets += st.tp_drops;
1398 if (copy_to_user(optval, &st, len))
1403 return -ENOPROTOOPT;
1406 if (put_user(len, optlen))
1412 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1415 struct hlist_node *node;
1416 struct net_device *dev = (struct net_device*)data;
1418 read_lock(&packet_sklist_lock);
1419 sk_for_each(sk, node, &packet_sklist) {
1420 struct packet_opt *po = pkt_sk(sk);
1423 case NETDEV_UNREGISTER:
1424 #ifdef CONFIG_PACKET_MULTICAST
1426 packet_dev_mclist(dev, po->mclist, -1);
1430 if (dev->ifindex == po->ifindex) {
1431 spin_lock(&po->bind_lock);
1433 __dev_remove_pack(&po->prot_hook);
1436 sk->sk_err = ENETDOWN;
1437 if (!sock_flag(sk, SOCK_DEAD))
1438 sk->sk_error_report(sk);
1440 if (msg == NETDEV_UNREGISTER) {
1442 po->prot_hook.dev = NULL;
1444 spin_unlock(&po->bind_lock);
1448 spin_lock(&po->bind_lock);
1449 if (dev->ifindex == po->ifindex && po->num &&
1451 dev_add_pack(&po->prot_hook);
1455 spin_unlock(&po->bind_lock);
1459 read_unlock(&packet_sklist_lock);
1464 static int packet_ioctl(struct socket *sock, unsigned int cmd,
1467 struct sock *sk = sock->sk;
1472 int amount = atomic_read(&sk->sk_wmem_alloc);
1473 return put_user(amount, (int __user *)arg);
1477 struct sk_buff *skb;
1480 spin_lock_bh(&sk->sk_receive_queue.lock);
1481 skb = skb_peek(&sk->sk_receive_queue);
1484 spin_unlock_bh(&sk->sk_receive_queue.lock);
1485 return put_user(amount, (int __user *)arg);
1488 return sock_get_timestamp(sk, (struct timeval __user *)arg);
1498 case SIOCGIFBRDADDR:
1499 case SIOCSIFBRDADDR:
1500 case SIOCGIFNETMASK:
1501 case SIOCSIFNETMASK:
1502 case SIOCGIFDSTADDR:
1503 case SIOCSIFDSTADDR:
1505 return inet_dgram_ops.ioctl(sock, cmd, arg);
1509 return dev_ioctl(cmd, (void __user *)arg);
1514 #ifndef CONFIG_PACKET_MMAP
1515 #define packet_mmap sock_no_mmap
1516 #define packet_poll datagram_poll
1519 unsigned int packet_poll(struct file * file, struct socket *sock, poll_table *wait)
1521 struct sock *sk = sock->sk;
1522 struct packet_opt *po = pkt_sk(sk);
1523 unsigned int mask = datagram_poll(file, sock, wait);
1525 spin_lock_bh(&sk->sk_receive_queue.lock);
1527 unsigned last = po->head ? po->head-1 : po->frame_max;
1528 struct tpacket_hdr *h;
1530 h = (struct tpacket_hdr *)packet_lookup_frame(po, last);
1533 mask |= POLLIN | POLLRDNORM;
1535 spin_unlock_bh(&sk->sk_receive_queue.lock);
1540 /* Dirty? Well, I still did not learn better way to account
1544 static void packet_mm_open(struct vm_area_struct *vma)
1546 struct file *file = vma->vm_file;
1547 struct inode *inode = file->f_dentry->d_inode;
1548 struct socket * sock = SOCKET_I(inode);
1549 struct sock *sk = sock->sk;
1552 atomic_inc(&pkt_sk(sk)->mapped);
1555 static void packet_mm_close(struct vm_area_struct *vma)
1557 struct file *file = vma->vm_file;
1558 struct inode *inode = file->f_dentry->d_inode;
1559 struct socket * sock = SOCKET_I(inode);
1560 struct sock *sk = sock->sk;
1563 atomic_dec(&pkt_sk(sk)->mapped);
1566 static struct vm_operations_struct packet_mmap_ops = {
1567 .open = packet_mm_open,
1568 .close =packet_mm_close,
1571 static inline struct page *pg_vec_endpage(char *one_pg_vec, unsigned int order)
1573 return virt_to_page(one_pg_vec + (PAGE_SIZE << order) - 1);
1576 static void free_pg_vec(char **pg_vec, unsigned order, unsigned len)
1580 for (i=0; i<len; i++) {
1582 struct page *page, *pend;
1584 pend = pg_vec_endpage(pg_vec[i], order);
1585 for (page = virt_to_page(pg_vec[i]); page <= pend; page++)
1586 ClearPageReserved(page);
1587 free_pages((unsigned long)pg_vec[i], order);
1594 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
1596 char **pg_vec = NULL;
1597 struct packet_opt *po = pkt_sk(sk);
1598 int was_running, num, order = 0;
1601 if (req->tp_block_nr) {
1604 /* Sanity tests and some calculations */
1609 if ((int)req->tp_block_size <= 0)
1611 if (req->tp_block_size&(PAGE_SIZE-1))
1613 if (req->tp_frame_size < TPACKET_HDRLEN)
1615 if (req->tp_frame_size&(TPACKET_ALIGNMENT-1))
1618 po->frames_per_block = req->tp_block_size/req->tp_frame_size;
1619 if (po->frames_per_block <= 0)
1621 if (po->frames_per_block*req->tp_block_nr != req->tp_frame_nr)
1625 /* Allocate page vector */
1626 while ((PAGE_SIZE<<order) < req->tp_block_size)
1631 pg_vec = kmalloc(req->tp_block_nr*sizeof(char *), GFP_KERNEL);
1634 memset(pg_vec, 0, req->tp_block_nr*sizeof(char **));
1636 for (i=0; i<req->tp_block_nr; i++) {
1637 struct page *page, *pend;
1638 pg_vec[i] = (char *)__get_free_pages(GFP_KERNEL, order);
1640 goto out_free_pgvec;
1642 pend = pg_vec_endpage(pg_vec[i], order);
1643 for (page = virt_to_page(pg_vec[i]); page <= pend; page++)
1644 SetPageReserved(page);
1646 /* Page vector is allocated */
1649 for (i=0; i<req->tp_block_nr; i++) {
1650 char *ptr = pg_vec[i];
1651 struct tpacket_hdr *header;
1654 for (k=0; k<po->frames_per_block; k++) {
1656 header = (struct tpacket_hdr*)ptr;
1657 header->tp_status = TP_STATUS_KERNEL;
1658 ptr += req->tp_frame_size;
1663 if (req->tp_frame_nr)
1669 /* Detach socket from network */
1670 spin_lock(&po->bind_lock);
1671 was_running = po->running;
1674 __dev_remove_pack(&po->prot_hook);
1679 spin_unlock(&po->bind_lock);
1684 if (closing || atomic_read(&po->mapped) == 0) {
1686 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
1688 spin_lock_bh(&sk->sk_receive_queue.lock);
1689 pg_vec = XC(po->pg_vec, pg_vec);
1690 po->frame_max = req->tp_frame_nr-1;
1692 po->frame_size = req->tp_frame_size;
1693 spin_unlock_bh(&sk->sk_receive_queue.lock);
1695 order = XC(po->pg_vec_order, order);
1696 req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
1698 po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
1699 po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
1700 skb_queue_purge(&sk->sk_receive_queue);
1702 if (atomic_read(&po->mapped))
1703 printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
1706 spin_lock(&po->bind_lock);
1707 if (was_running && !po->running) {
1711 dev_add_pack(&po->prot_hook);
1713 spin_unlock(&po->bind_lock);
1719 free_pg_vec(pg_vec, order, req->tp_block_nr);
1724 static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1726 struct sock *sk = sock->sk;
1727 struct packet_opt *po = pkt_sk(sk);
1729 unsigned long start;
1736 size = vma->vm_end - vma->vm_start;
1739 if (po->pg_vec == NULL)
1741 if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
1744 atomic_inc(&po->mapped);
1745 start = vma->vm_start;
1747 for (i=0; i<po->pg_vec_len; i++) {
1748 if (remap_page_range(vma, start, __pa(po->pg_vec[i]),
1749 po->pg_vec_pages*PAGE_SIZE,
1752 start += po->pg_vec_pages*PAGE_SIZE;
1754 vma->vm_ops = &packet_mmap_ops;
1764 #ifdef CONFIG_SOCK_PACKET
1765 struct proto_ops packet_ops_spkt = {
1766 .family = PF_PACKET,
1767 .owner = THIS_MODULE,
1768 .release = packet_release,
1769 .bind = packet_bind_spkt,
1770 .connect = sock_no_connect,
1771 .socketpair = sock_no_socketpair,
1772 .accept = sock_no_accept,
1773 .getname = packet_getname_spkt,
1774 .poll = datagram_poll,
1775 .ioctl = packet_ioctl,
1776 .listen = sock_no_listen,
1777 .shutdown = sock_no_shutdown,
1778 .setsockopt = sock_no_setsockopt,
1779 .getsockopt = sock_no_getsockopt,
1780 .sendmsg = packet_sendmsg_spkt,
1781 .recvmsg = packet_recvmsg,
1782 .mmap = sock_no_mmap,
1783 .sendpage = sock_no_sendpage,
1787 struct proto_ops packet_ops = {
1788 .family = PF_PACKET,
1789 .owner = THIS_MODULE,
1790 .release = packet_release,
1791 .bind = packet_bind,
1792 .connect = sock_no_connect,
1793 .socketpair = sock_no_socketpair,
1794 .accept = sock_no_accept,
1795 .getname = packet_getname,
1796 .poll = packet_poll,
1797 .ioctl = packet_ioctl,
1798 .listen = sock_no_listen,
1799 .shutdown = sock_no_shutdown,
1800 .setsockopt = packet_setsockopt,
1801 .getsockopt = packet_getsockopt,
1802 .sendmsg = packet_sendmsg,
1803 .recvmsg = packet_recvmsg,
1804 .mmap = packet_mmap,
1805 .sendpage = sock_no_sendpage,
1808 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
1809 EXPORT_SYMBOL(packet_ops);
1810 struct net_proto_family packet_family_ops;
1811 EXPORT_SYMBOL(packet_family_ops);
1815 struct net_proto_family packet_family_ops = {
1816 .family = PF_PACKET,
1817 .create = packet_create,
1818 .owner = THIS_MODULE,
1821 static struct notifier_block packet_netdev_notifier = {
1822 .notifier_call =packet_notifier,
1825 #ifdef CONFIG_PROC_FS
1826 static inline struct sock *packet_seq_idx(loff_t off)
1829 struct hlist_node *node;
1831 sk_for_each(s, node, &packet_sklist) {
1838 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
1840 read_lock(&packet_sklist_lock);
1841 return *pos ? packet_seq_idx(*pos - 1) : SEQ_START_TOKEN;
1844 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1847 return (v == SEQ_START_TOKEN)
1848 ? sk_head(&packet_sklist)
1849 : sk_next((struct sock*)v) ;
1852 static void packet_seq_stop(struct seq_file *seq, void *v)
1854 read_unlock(&packet_sklist_lock);
1857 static int packet_seq_show(struct seq_file *seq, void *v)
1859 if (v == SEQ_START_TOKEN)
1860 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
1863 const struct packet_opt *po = pkt_sk(s);
1866 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
1868 atomic_read(&s->sk_refcnt),
1873 atomic_read(&s->sk_rmem_alloc),
1881 static struct seq_operations packet_seq_ops = {
1882 .start = packet_seq_start,
1883 .next = packet_seq_next,
1884 .stop = packet_seq_stop,
1885 .show = packet_seq_show,
1888 static int packet_seq_open(struct inode *inode, struct file *file)
1890 return seq_open(file, &packet_seq_ops);
1893 static struct file_operations packet_seq_fops = {
1894 .owner = THIS_MODULE,
1895 .open = packet_seq_open,
1897 .llseek = seq_lseek,
1898 .release = seq_release,
1903 static void __exit packet_exit(void)
1905 proc_net_remove("packet");
1906 unregister_netdevice_notifier(&packet_netdev_notifier);
1907 sock_unregister(PF_PACKET);
1911 static int __init packet_init(void)
1913 sock_register(&packet_family_ops);
1914 register_netdevice_notifier(&packet_netdev_notifier);
1915 proc_net_fops_create("packet", 0, &packet_seq_fops);
1920 module_init(packet_init);
1921 module_exit(packet_exit);
1922 MODULE_LICENSE("GPL");
1923 MODULE_ALIAS_NETPROTO(PF_PACKET);