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 $
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.
39 * 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
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.
52 #include <linux/types.h>
54 #include <linux/capability.h>
55 #include <linux/fcntl.h>
56 #include <linux/socket.h>
58 #include <linux/inet.h>
59 #include <linux/netdevice.h>
60 #include <linux/if_packet.h>
61 #include <linux/wireless.h>
62 #include <linux/kernel.h>
63 #include <linux/kmod.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.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>
74 #include <asm/cacheflush.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>
83 #include <net/inet_common.h>
88 - if device has no dev->hard_header routine, it adds and removes ll header
89 inside itself. In this case ll header is invisible outside of device,
90 but higher levels still should reserve dev->hard_header_len.
91 Some devices are enough clever to reallocate skb, when header
92 will not fit to reserved space (tunnel), another ones are silly
94 - packet socket receives packets with pulled ll header,
95 so that SOCK_RAW should push it back.
100 Incoming, dev->hard_header!=NULL
101 mac_header -> ll header
104 Outgoing, dev->hard_header!=NULL
105 mac_header -> ll header
108 Incoming, dev->hard_header==NULL
109 mac_header -> UNKNOWN position. It is very likely, that it points to ll
110 header. PPP makes it, that is wrong, because introduce
111 assymetry between rx and tx paths.
114 Outgoing, dev->hard_header==NULL
115 mac_header -> data. ll header is still not built!
119 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
125 dev->hard_header != NULL
126 mac_header -> ll header
129 dev->hard_header == NULL (ll header is added by device, we cannot control it)
133 We should set nh.raw on output to correct posistion,
134 packet classifier depends on it.
137 /* List of all packet sockets. */
138 static HLIST_HEAD(packet_sklist);
139 static DEFINE_RWLOCK(packet_sklist_lock);
141 static atomic_t packet_socks_nr;
144 /* Private packet socket structures. */
148 struct packet_mclist *next;
153 unsigned char addr[MAX_ADDR_LEN];
155 /* identical to struct packet_mreq except it has
156 * a longer address field.
158 struct packet_mreq_max
161 unsigned short mr_type;
162 unsigned short mr_alen;
163 unsigned char mr_address[MAX_ADDR_LEN];
166 #ifdef CONFIG_PACKET_MMAP
167 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
170 static void packet_flush_mclist(struct sock *sk);
173 /* struct sock has to be the first member of packet_sock */
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 unsigned int running:1, /* prot_hook is attached*/
189 int ifindex; /* bound device */
191 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 struct packet_skb_cb {
201 unsigned int origlen;
203 struct sockaddr_pkt pkt;
204 struct sockaddr_ll ll;
208 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
210 #ifdef CONFIG_PACKET_MMAP
212 static inline struct tpacket_hdr *packet_lookup_frame(struct packet_sock *po, unsigned int position)
214 unsigned int pg_vec_pos, frame_offset;
216 pg_vec_pos = position / po->frames_per_block;
217 frame_offset = position % po->frames_per_block;
219 return (struct tpacket_hdr *)(po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size));
223 static inline struct packet_sock *pkt_sk(struct sock *sk)
225 return (struct packet_sock *)sk;
228 static void packet_sock_destruct(struct sock *sk)
230 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
231 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
233 if (!sock_flag(sk, SOCK_DEAD)) {
234 printk("Attempt to release alive packet socket: %p\n", sk);
238 atomic_dec(&packet_socks_nr);
239 #ifdef PACKET_REFCNT_DEBUG
240 printk(KERN_DEBUG "PACKET socket %p is free, %d are alive\n", sk, atomic_read(&packet_socks_nr));
245 static const struct proto_ops packet_ops;
247 static const struct proto_ops packet_ops_spkt;
249 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
252 struct sockaddr_pkt *spkt;
255 * When we registered the protocol we saved the socket in the data
256 * field for just this event.
259 sk = pt->af_packet_priv;
262 * Yank back the headers [hope the device set this
263 * right or kerboom...]
265 * Incoming packets have ll header pulled,
268 * For outgoing ones skb->data == skb_mac_header(skb)
269 * so that this procedure is noop.
272 if (skb->pkt_type == PACKET_LOOPBACK)
275 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
278 /* drop any routing info */
279 dst_release(skb->dst);
282 /* drop conntrack reference */
285 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
287 skb_push(skb, skb->data - skb_mac_header(skb));
290 * The SOCK_PACKET socket receives _all_ frames.
293 spkt->spkt_family = dev->type;
294 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
295 spkt->spkt_protocol = skb->protocol;
298 * Charge the memory to the socket. This is done specifically
299 * to prevent sockets using all the memory up.
302 if (sock_queue_rcv_skb(sk,skb) == 0)
313 * Output a raw packet to a device layer. This bypasses all the other
314 * protocol layers and you must therefore supply it with a complete frame
317 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
318 struct msghdr *msg, size_t len)
320 struct sock *sk = sock->sk;
321 struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
323 struct net_device *dev;
328 * Get and verify the address.
333 if (msg->msg_namelen < sizeof(struct sockaddr))
335 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
336 proto=saddr->spkt_protocol;
339 return(-ENOTCONN); /* SOCK_PACKET must be sent giving an address */
342 * Find the device first to size check it
345 saddr->spkt_device[13] = 0;
346 dev = dev_get_by_name(saddr->spkt_device);
352 if (!(dev->flags & IFF_UP))
356 * You may not queue a frame bigger than the mtu. This is the lowest level
357 * raw protocol and you must do your own fragmentation at this level.
361 if (len > dev->mtu + dev->hard_header_len)
365 skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
368 * If the write buffer is full, then tough. At this level the user gets to
369 * deal with the problem - do your own algorithmic backoffs. That's far
380 /* FIXME: Save some space for broken drivers that write a
381 * hard header at transmission time by themselves. PPP is the
382 * notable one here. This should really be fixed at the driver level.
384 skb_reserve(skb, LL_RESERVED_SPACE(dev));
385 skb_reset_network_header(skb);
387 /* Try to align data part correctly */
388 if (dev->hard_header) {
389 skb->data -= dev->hard_header_len;
390 skb->tail -= dev->hard_header_len;
391 if (len < dev->hard_header_len)
392 skb_reset_network_header(skb);
395 /* Returns -EFAULT on error */
396 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
397 skb->protocol = proto;
399 skb->priority = sk->sk_priority;
419 static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
422 struct sk_filter *filter;
425 filter = rcu_dereference(sk->sk_filter);
427 res = sk_run_filter(skb, filter->insns, filter->len);
428 rcu_read_unlock_bh();
434 This function makes lazy skb cloning in hope that most of packets
435 are discarded by BPF.
437 Note tricky part: we DO mangle shared skb! skb->data, skb->len
438 and skb->cb are mangled. It works because (and until) packets
439 falling here are owned by current CPU. Output packets are cloned
440 by dev_queue_xmit_nit(), input packets are processed by net_bh
441 sequencially, so that if we return skb to original state on exit,
442 we will not harm anyone.
445 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
448 struct sockaddr_ll *sll;
449 struct packet_sock *po;
450 u8 * skb_head = skb->data;
451 int skb_len = skb->len;
452 unsigned int snaplen, res;
454 if (skb->pkt_type == PACKET_LOOPBACK)
457 sk = pt->af_packet_priv;
462 if (dev->hard_header) {
463 /* The device has an explicit notion of ll header,
464 exported to higher levels.
466 Otherwise, the device hides datails of it frame
467 structure, so that corresponding packet head
468 never delivered to user.
470 if (sk->sk_type != SOCK_DGRAM)
471 skb_push(skb, skb->data - skb_mac_header(skb));
472 else if (skb->pkt_type == PACKET_OUTGOING) {
473 /* Special case: outgoing packets have ll header at head */
474 skb_pull(skb, skb_network_offset(skb));
480 res = run_filter(skb, sk, snaplen);
486 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
487 (unsigned)sk->sk_rcvbuf)
490 if (skb_shared(skb)) {
491 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
495 if (skb_head != skb->data) {
496 skb->data = skb_head;
503 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
506 sll = &PACKET_SKB_CB(skb)->sa.ll;
507 sll->sll_family = AF_PACKET;
508 sll->sll_hatype = dev->type;
509 sll->sll_protocol = skb->protocol;
510 sll->sll_pkttype = skb->pkt_type;
511 if (unlikely(po->origdev) && skb->pkt_type == PACKET_HOST)
512 sll->sll_ifindex = orig_dev->ifindex;
514 sll->sll_ifindex = dev->ifindex;
517 if (dev->hard_header_parse)
518 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
520 PACKET_SKB_CB(skb)->origlen = skb->len;
522 if (pskb_trim(skb, snaplen))
525 skb_set_owner_r(skb, sk);
527 dst_release(skb->dst);
530 /* drop conntrack reference */
533 spin_lock(&sk->sk_receive_queue.lock);
534 po->stats.tp_packets++;
535 __skb_queue_tail(&sk->sk_receive_queue, skb);
536 spin_unlock(&sk->sk_receive_queue.lock);
537 sk->sk_data_ready(sk, skb->len);
541 spin_lock(&sk->sk_receive_queue.lock);
542 po->stats.tp_drops++;
543 spin_unlock(&sk->sk_receive_queue.lock);
546 if (skb_head != skb->data && skb_shared(skb)) {
547 skb->data = skb_head;
555 #ifdef CONFIG_PACKET_MMAP
556 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
559 struct packet_sock *po;
560 struct sockaddr_ll *sll;
561 struct tpacket_hdr *h;
562 u8 * skb_head = skb->data;
563 int skb_len = skb->len;
564 unsigned int snaplen, res;
565 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
566 unsigned short macoff, netoff;
567 struct sk_buff *copy_skb = NULL;
570 if (skb->pkt_type == PACKET_LOOPBACK)
573 sk = pt->af_packet_priv;
576 if (dev->hard_header) {
577 if (sk->sk_type != SOCK_DGRAM)
578 skb_push(skb, skb->data - skb_mac_header(skb));
579 else if (skb->pkt_type == PACKET_OUTGOING) {
580 /* Special case: outgoing packets have ll header at head */
581 skb_pull(skb, skb_network_offset(skb));
585 if (skb->ip_summed == CHECKSUM_PARTIAL)
586 status |= TP_STATUS_CSUMNOTREADY;
590 res = run_filter(skb, sk, snaplen);
596 if (sk->sk_type == SOCK_DGRAM) {
597 macoff = netoff = TPACKET_ALIGN(TPACKET_HDRLEN) + 16;
599 unsigned maclen = skb_network_offset(skb);
600 netoff = TPACKET_ALIGN(TPACKET_HDRLEN + (maclen < 16 ? 16 : maclen));
601 macoff = netoff - maclen;
604 if (macoff + snaplen > po->frame_size) {
605 if (po->copy_thresh &&
606 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
607 (unsigned)sk->sk_rcvbuf) {
608 if (skb_shared(skb)) {
609 copy_skb = skb_clone(skb, GFP_ATOMIC);
611 copy_skb = skb_get(skb);
612 skb_head = skb->data;
615 skb_set_owner_r(copy_skb, sk);
617 snaplen = po->frame_size - macoff;
618 if ((int)snaplen < 0)
622 spin_lock(&sk->sk_receive_queue.lock);
623 h = packet_lookup_frame(po, po->head);
627 po->head = po->head != po->frame_max ? po->head+1 : 0;
628 po->stats.tp_packets++;
630 status |= TP_STATUS_COPY;
631 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
633 if (!po->stats.tp_drops)
634 status &= ~TP_STATUS_LOSING;
635 spin_unlock(&sk->sk_receive_queue.lock);
637 skb_copy_bits(skb, 0, (u8*)h + macoff, snaplen);
639 h->tp_len = skb->len;
640 h->tp_snaplen = snaplen;
643 if (skb->tstamp.tv64)
644 tv = ktime_to_timeval(skb->tstamp);
646 do_gettimeofday(&tv);
647 h->tp_sec = tv.tv_sec;
648 h->tp_usec = tv.tv_usec;
650 sll = (struct sockaddr_ll*)((u8*)h + TPACKET_ALIGN(sizeof(*h)));
652 if (dev->hard_header_parse)
653 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
654 sll->sll_family = AF_PACKET;
655 sll->sll_hatype = dev->type;
656 sll->sll_protocol = skb->protocol;
657 sll->sll_pkttype = skb->pkt_type;
658 if (unlikely(po->origdev) && skb->pkt_type == PACKET_HOST)
659 sll->sll_ifindex = orig_dev->ifindex;
661 sll->sll_ifindex = dev->ifindex;
663 h->tp_status = status;
667 struct page *p_start, *p_end;
668 u8 *h_end = (u8 *)h + macoff + snaplen - 1;
670 p_start = virt_to_page(h);
671 p_end = virt_to_page(h_end);
672 while (p_start <= p_end) {
673 flush_dcache_page(p_start);
678 sk->sk_data_ready(sk, 0);
681 if (skb_head != skb->data && skb_shared(skb)) {
682 skb->data = skb_head;
690 po->stats.tp_drops++;
691 spin_unlock(&sk->sk_receive_queue.lock);
693 sk->sk_data_ready(sk, 0);
702 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
703 struct msghdr *msg, size_t len)
705 struct sock *sk = sock->sk;
706 struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
708 struct net_device *dev;
711 int ifindex, err, reserve = 0;
714 * Get and verify the address.
718 struct packet_sock *po = pkt_sk(sk);
720 ifindex = po->ifindex;
725 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
727 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
729 ifindex = saddr->sll_ifindex;
730 proto = saddr->sll_protocol;
731 addr = saddr->sll_addr;
735 dev = dev_get_by_index(ifindex);
739 if (sock->type == SOCK_RAW)
740 reserve = dev->hard_header_len;
743 if (!(dev->flags & IFF_UP))
747 if (len > dev->mtu+reserve)
750 skb = sock_alloc_send_skb(sk, len + LL_RESERVED_SPACE(dev),
751 msg->msg_flags & MSG_DONTWAIT, &err);
755 skb_reserve(skb, LL_RESERVED_SPACE(dev));
756 skb_reset_network_header(skb);
758 if (dev->hard_header) {
761 res = dev->hard_header(skb, dev, ntohs(proto), addr, NULL, len);
762 if (sock->type != SOCK_DGRAM) {
763 skb_reset_tail_pointer(skb);
769 /* Returns -EFAULT on error */
770 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
774 skb->protocol = proto;
776 skb->priority = sk->sk_priority;
782 err = dev_queue_xmit(skb);
783 if (err > 0 && (err = net_xmit_errno(err)) != 0)
800 * Close a PACKET socket. This is fairly simple. We immediately go
801 * to 'closed' state and remove our protocol entry in the device list.
804 static int packet_release(struct socket *sock)
806 struct sock *sk = sock->sk;
807 struct packet_sock *po;
814 write_lock_bh(&packet_sklist_lock);
815 sk_del_node_init(sk);
816 write_unlock_bh(&packet_sklist_lock);
819 * Unhook packet receive handler.
824 * Remove the protocol hook
826 dev_remove_pack(&po->prot_hook);
832 packet_flush_mclist(sk);
834 #ifdef CONFIG_PACKET_MMAP
836 struct tpacket_req req;
837 memset(&req, 0, sizeof(req));
838 packet_set_ring(sk, &req, 1);
843 * Now the socket is dead. No more input will appear.
851 skb_queue_purge(&sk->sk_receive_queue);
858 * Attach a packet hook.
861 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
863 struct packet_sock *po = pkt_sk(sk);
865 * Detach an existing hook if present.
870 spin_lock(&po->bind_lock);
875 spin_unlock(&po->bind_lock);
876 dev_remove_pack(&po->prot_hook);
877 spin_lock(&po->bind_lock);
881 po->prot_hook.type = protocol;
882 po->prot_hook.dev = dev;
884 po->ifindex = dev ? dev->ifindex : 0;
890 if (dev->flags&IFF_UP) {
891 dev_add_pack(&po->prot_hook);
895 sk->sk_err = ENETDOWN;
896 if (!sock_flag(sk, SOCK_DEAD))
897 sk->sk_error_report(sk);
900 dev_add_pack(&po->prot_hook);
906 spin_unlock(&po->bind_lock);
912 * Bind a packet socket to a device
915 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
917 struct sock *sk=sock->sk;
919 struct net_device *dev;
926 if (addr_len != sizeof(struct sockaddr))
928 strlcpy(name,uaddr->sa_data,sizeof(name));
930 dev = dev_get_by_name(name);
932 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
938 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
940 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
941 struct sock *sk=sock->sk;
942 struct net_device *dev = NULL;
950 if (addr_len < sizeof(struct sockaddr_ll))
952 if (sll->sll_family != AF_PACKET)
955 if (sll->sll_ifindex) {
957 dev = dev_get_by_index(sll->sll_ifindex);
961 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
969 static struct proto packet_proto = {
971 .owner = THIS_MODULE,
972 .obj_size = sizeof(struct packet_sock),
976 * Create a packet of type SOCK_PACKET.
979 static int packet_create(struct socket *sock, int protocol)
982 struct packet_sock *po;
983 __be16 proto = (__force __be16)protocol; /* weird, but documented */
986 if (!capable(CAP_NET_RAW))
988 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
989 sock->type != SOCK_PACKET)
990 return -ESOCKTNOSUPPORT;
992 sock->state = SS_UNCONNECTED;
995 sk = sk_alloc(PF_PACKET, GFP_KERNEL, &packet_proto, 1);
999 sock->ops = &packet_ops;
1000 if (sock->type == SOCK_PACKET)
1001 sock->ops = &packet_ops_spkt;
1003 sock_init_data(sock, sk);
1006 sk->sk_family = PF_PACKET;
1009 sk->sk_destruct = packet_sock_destruct;
1010 atomic_inc(&packet_socks_nr);
1013 * Attach a protocol block
1016 spin_lock_init(&po->bind_lock);
1017 po->prot_hook.func = packet_rcv;
1019 if (sock->type == SOCK_PACKET)
1020 po->prot_hook.func = packet_rcv_spkt;
1022 po->prot_hook.af_packet_priv = sk;
1025 po->prot_hook.type = proto;
1026 dev_add_pack(&po->prot_hook);
1031 write_lock_bh(&packet_sklist_lock);
1032 sk_add_node(sk, &packet_sklist);
1033 write_unlock_bh(&packet_sklist_lock);
1040 * Pull a packet from our receive queue and hand it to the user.
1041 * If necessary we block.
1044 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1045 struct msghdr *msg, size_t len, int flags)
1047 struct sock *sk = sock->sk;
1048 struct sk_buff *skb;
1050 struct sockaddr_ll *sll;
1053 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1057 /* What error should we return now? EUNATTACH? */
1058 if (pkt_sk(sk)->ifindex < 0)
1063 * Call the generic datagram receiver. This handles all sorts
1064 * of horrible races and re-entrancy so we can forget about it
1065 * in the protocol layers.
1067 * Now it will return ENETDOWN, if device have just gone down,
1068 * but then it will block.
1071 skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1074 * An error occurred so return it. Because skb_recv_datagram()
1075 * handles the blocking we don't see and worry about blocking
1083 * If the address length field is there to be filled in, we fill
1087 sll = &PACKET_SKB_CB(skb)->sa.ll;
1088 if (sock->type == SOCK_PACKET)
1089 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1091 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1094 * You lose any data beyond the buffer you gave. If it worries a
1095 * user program they can ask the device for its MTU anyway.
1102 msg->msg_flags|=MSG_TRUNC;
1105 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1109 sock_recv_timestamp(msg, sk, skb);
1112 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1115 if (pkt_sk(sk)->auxdata) {
1116 struct tpacket_auxdata aux;
1118 aux.tp_status = TP_STATUS_USER;
1119 if (skb->ip_summed == CHECKSUM_PARTIAL)
1120 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1121 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1122 aux.tp_snaplen = skb->len;
1124 aux.tp_net = skb_network_offset(skb);
1126 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1130 * Free or return the buffer as appropriate. Again this
1131 * hides all the races and re-entrancy issues from us.
1133 err = (flags&MSG_TRUNC) ? skb->len : copied;
1136 skb_free_datagram(sk, skb);
1141 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1142 int *uaddr_len, int peer)
1144 struct net_device *dev;
1145 struct sock *sk = sock->sk;
1150 uaddr->sa_family = AF_PACKET;
1151 dev = dev_get_by_index(pkt_sk(sk)->ifindex);
1153 strlcpy(uaddr->sa_data, dev->name, 15);
1156 memset(uaddr->sa_data, 0, 14);
1157 *uaddr_len = sizeof(*uaddr);
1162 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1163 int *uaddr_len, int peer)
1165 struct net_device *dev;
1166 struct sock *sk = sock->sk;
1167 struct packet_sock *po = pkt_sk(sk);
1168 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1173 sll->sll_family = AF_PACKET;
1174 sll->sll_ifindex = po->ifindex;
1175 sll->sll_protocol = po->num;
1176 dev = dev_get_by_index(po->ifindex);
1178 sll->sll_hatype = dev->type;
1179 sll->sll_halen = dev->addr_len;
1180 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1183 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1186 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1191 static void packet_dev_mc(struct net_device *dev, struct packet_mclist *i, int what)
1194 case PACKET_MR_MULTICAST:
1196 dev_mc_add(dev, i->addr, i->alen, 0);
1198 dev_mc_delete(dev, i->addr, i->alen, 0);
1200 case PACKET_MR_PROMISC:
1201 dev_set_promiscuity(dev, what);
1203 case PACKET_MR_ALLMULTI:
1204 dev_set_allmulti(dev, what);
1210 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1212 for ( ; i; i=i->next) {
1213 if (i->ifindex == dev->ifindex)
1214 packet_dev_mc(dev, i, what);
1218 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1220 struct packet_sock *po = pkt_sk(sk);
1221 struct packet_mclist *ml, *i;
1222 struct net_device *dev;
1228 dev = __dev_get_by_index(mreq->mr_ifindex);
1233 if (mreq->mr_alen > dev->addr_len)
1237 i = kmalloc(sizeof(*i), GFP_KERNEL);
1242 for (ml = po->mclist; ml; ml = ml->next) {
1243 if (ml->ifindex == mreq->mr_ifindex &&
1244 ml->type == mreq->mr_type &&
1245 ml->alen == mreq->mr_alen &&
1246 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1248 /* Free the new element ... */
1254 i->type = mreq->mr_type;
1255 i->ifindex = mreq->mr_ifindex;
1256 i->alen = mreq->mr_alen;
1257 memcpy(i->addr, mreq->mr_address, i->alen);
1259 i->next = po->mclist;
1261 packet_dev_mc(dev, i, +1);
1268 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1270 struct packet_mclist *ml, **mlp;
1274 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1275 if (ml->ifindex == mreq->mr_ifindex &&
1276 ml->type == mreq->mr_type &&
1277 ml->alen == mreq->mr_alen &&
1278 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1279 if (--ml->count == 0) {
1280 struct net_device *dev;
1282 dev = dev_get_by_index(ml->ifindex);
1284 packet_dev_mc(dev, ml, -1);
1294 return -EADDRNOTAVAIL;
1297 static void packet_flush_mclist(struct sock *sk)
1299 struct packet_sock *po = pkt_sk(sk);
1300 struct packet_mclist *ml;
1306 while ((ml = po->mclist) != NULL) {
1307 struct net_device *dev;
1309 po->mclist = ml->next;
1310 if ((dev = dev_get_by_index(ml->ifindex)) != NULL) {
1311 packet_dev_mc(dev, ml, -1);
1320 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1322 struct sock *sk = sock->sk;
1323 struct packet_sock *po = pkt_sk(sk);
1326 if (level != SOL_PACKET)
1327 return -ENOPROTOOPT;
1330 case PACKET_ADD_MEMBERSHIP:
1331 case PACKET_DROP_MEMBERSHIP:
1333 struct packet_mreq_max mreq;
1335 memset(&mreq, 0, sizeof(mreq));
1336 if (len < sizeof(struct packet_mreq))
1338 if (len > sizeof(mreq))
1340 if (copy_from_user(&mreq,optval,len))
1342 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1344 if (optname == PACKET_ADD_MEMBERSHIP)
1345 ret = packet_mc_add(sk, &mreq);
1347 ret = packet_mc_drop(sk, &mreq);
1351 #ifdef CONFIG_PACKET_MMAP
1352 case PACKET_RX_RING:
1354 struct tpacket_req req;
1356 if (optlen<sizeof(req))
1358 if (copy_from_user(&req,optval,sizeof(req)))
1360 return packet_set_ring(sk, &req, 0);
1362 case PACKET_COPY_THRESH:
1366 if (optlen!=sizeof(val))
1368 if (copy_from_user(&val,optval,sizeof(val)))
1371 pkt_sk(sk)->copy_thresh = val;
1375 case PACKET_AUXDATA:
1379 if (optlen < sizeof(val))
1381 if (copy_from_user(&val, optval, sizeof(val)))
1384 po->auxdata = !!val;
1387 case PACKET_ORIGDEV:
1391 if (optlen < sizeof(val))
1393 if (copy_from_user(&val, optval, sizeof(val)))
1396 po->origdev = !!val;
1400 return -ENOPROTOOPT;
1404 static int packet_getsockopt(struct socket *sock, int level, int optname,
1405 char __user *optval, int __user *optlen)
1409 struct sock *sk = sock->sk;
1410 struct packet_sock *po = pkt_sk(sk);
1412 struct tpacket_stats st;
1414 if (level != SOL_PACKET)
1415 return -ENOPROTOOPT;
1417 if (get_user(len, optlen))
1424 case PACKET_STATISTICS:
1425 if (len > sizeof(struct tpacket_stats))
1426 len = sizeof(struct tpacket_stats);
1427 spin_lock_bh(&sk->sk_receive_queue.lock);
1429 memset(&po->stats, 0, sizeof(st));
1430 spin_unlock_bh(&sk->sk_receive_queue.lock);
1431 st.tp_packets += st.tp_drops;
1435 case PACKET_AUXDATA:
1436 if (len > sizeof(int))
1442 case PACKET_ORIGDEV:
1443 if (len > sizeof(int))
1450 return -ENOPROTOOPT;
1453 if (put_user(len, optlen))
1455 if (copy_to_user(optval, data, len))
1461 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1464 struct hlist_node *node;
1465 struct net_device *dev = data;
1467 read_lock(&packet_sklist_lock);
1468 sk_for_each(sk, node, &packet_sklist) {
1469 struct packet_sock *po = pkt_sk(sk);
1472 case NETDEV_UNREGISTER:
1474 packet_dev_mclist(dev, po->mclist, -1);
1478 if (dev->ifindex == po->ifindex) {
1479 spin_lock(&po->bind_lock);
1481 __dev_remove_pack(&po->prot_hook);
1484 sk->sk_err = ENETDOWN;
1485 if (!sock_flag(sk, SOCK_DEAD))
1486 sk->sk_error_report(sk);
1488 if (msg == NETDEV_UNREGISTER) {
1490 po->prot_hook.dev = NULL;
1492 spin_unlock(&po->bind_lock);
1496 spin_lock(&po->bind_lock);
1497 if (dev->ifindex == po->ifindex && po->num &&
1499 dev_add_pack(&po->prot_hook);
1503 spin_unlock(&po->bind_lock);
1507 read_unlock(&packet_sklist_lock);
1512 static int packet_ioctl(struct socket *sock, unsigned int cmd,
1515 struct sock *sk = sock->sk;
1520 int amount = atomic_read(&sk->sk_wmem_alloc);
1521 return put_user(amount, (int __user *)arg);
1525 struct sk_buff *skb;
1528 spin_lock_bh(&sk->sk_receive_queue.lock);
1529 skb = skb_peek(&sk->sk_receive_queue);
1532 spin_unlock_bh(&sk->sk_receive_queue.lock);
1533 return put_user(amount, (int __user *)arg);
1536 return sock_get_timestamp(sk, (struct timeval __user *)arg);
1538 return sock_get_timestampns(sk, (struct timespec __user *)arg);
1548 case SIOCGIFBRDADDR:
1549 case SIOCSIFBRDADDR:
1550 case SIOCGIFNETMASK:
1551 case SIOCSIFNETMASK:
1552 case SIOCGIFDSTADDR:
1553 case SIOCSIFDSTADDR:
1555 return inet_dgram_ops.ioctl(sock, cmd, arg);
1559 return -ENOIOCTLCMD;
1564 #ifndef CONFIG_PACKET_MMAP
1565 #define packet_mmap sock_no_mmap
1566 #define packet_poll datagram_poll
1569 static unsigned int packet_poll(struct file * file, struct socket *sock,
1572 struct sock *sk = sock->sk;
1573 struct packet_sock *po = pkt_sk(sk);
1574 unsigned int mask = datagram_poll(file, sock, wait);
1576 spin_lock_bh(&sk->sk_receive_queue.lock);
1578 unsigned last = po->head ? po->head-1 : po->frame_max;
1579 struct tpacket_hdr *h;
1581 h = packet_lookup_frame(po, last);
1584 mask |= POLLIN | POLLRDNORM;
1586 spin_unlock_bh(&sk->sk_receive_queue.lock);
1591 /* Dirty? Well, I still did not learn better way to account
1595 static void packet_mm_open(struct vm_area_struct *vma)
1597 struct file *file = vma->vm_file;
1598 struct socket * sock = file->private_data;
1599 struct sock *sk = sock->sk;
1602 atomic_inc(&pkt_sk(sk)->mapped);
1605 static void packet_mm_close(struct vm_area_struct *vma)
1607 struct file *file = vma->vm_file;
1608 struct socket * sock = file->private_data;
1609 struct sock *sk = sock->sk;
1612 atomic_dec(&pkt_sk(sk)->mapped);
1615 static struct vm_operations_struct packet_mmap_ops = {
1616 .open = packet_mm_open,
1617 .close =packet_mm_close,
1620 static inline struct page *pg_vec_endpage(char *one_pg_vec, unsigned int order)
1622 return virt_to_page(one_pg_vec + (PAGE_SIZE << order) - 1);
1625 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
1629 for (i = 0; i < len; i++) {
1630 if (likely(pg_vec[i]))
1631 free_pages((unsigned long) pg_vec[i], order);
1636 static inline char *alloc_one_pg_vec_page(unsigned long order)
1638 return (char *) __get_free_pages(GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
1642 static char **alloc_pg_vec(struct tpacket_req *req, int order)
1644 unsigned int block_nr = req->tp_block_nr;
1648 pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
1649 if (unlikely(!pg_vec))
1652 for (i = 0; i < block_nr; i++) {
1653 pg_vec[i] = alloc_one_pg_vec_page(order);
1654 if (unlikely(!pg_vec[i]))
1655 goto out_free_pgvec;
1662 free_pg_vec(pg_vec, order, block_nr);
1667 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
1669 char **pg_vec = NULL;
1670 struct packet_sock *po = pkt_sk(sk);
1671 int was_running, order = 0;
1675 if (req->tp_block_nr) {
1678 /* Sanity tests and some calculations */
1680 if (unlikely(po->pg_vec))
1683 if (unlikely((int)req->tp_block_size <= 0))
1685 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
1687 if (unlikely(req->tp_frame_size < TPACKET_HDRLEN))
1689 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
1692 po->frames_per_block = req->tp_block_size/req->tp_frame_size;
1693 if (unlikely(po->frames_per_block <= 0))
1695 if (unlikely((po->frames_per_block * req->tp_block_nr) !=
1700 order = get_order(req->tp_block_size);
1701 pg_vec = alloc_pg_vec(req, order);
1702 if (unlikely(!pg_vec))
1706 for (i = 0; i < req->tp_block_nr; i++) {
1707 char *ptr = pg_vec[i];
1708 struct tpacket_hdr *header;
1711 for (k = 0; k < po->frames_per_block; k++) {
1712 header = (struct tpacket_hdr *) ptr;
1713 header->tp_status = TP_STATUS_KERNEL;
1714 ptr += req->tp_frame_size;
1719 if (unlikely(req->tp_frame_nr))
1725 /* Detach socket from network */
1726 spin_lock(&po->bind_lock);
1727 was_running = po->running;
1730 __dev_remove_pack(&po->prot_hook);
1735 spin_unlock(&po->bind_lock);
1740 if (closing || atomic_read(&po->mapped) == 0) {
1742 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
1744 spin_lock_bh(&sk->sk_receive_queue.lock);
1745 pg_vec = XC(po->pg_vec, pg_vec);
1746 po->frame_max = (req->tp_frame_nr - 1);
1748 po->frame_size = req->tp_frame_size;
1749 spin_unlock_bh(&sk->sk_receive_queue.lock);
1751 order = XC(po->pg_vec_order, order);
1752 req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
1754 po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
1755 po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
1756 skb_queue_purge(&sk->sk_receive_queue);
1758 if (atomic_read(&po->mapped))
1759 printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
1762 spin_lock(&po->bind_lock);
1763 if (was_running && !po->running) {
1767 dev_add_pack(&po->prot_hook);
1769 spin_unlock(&po->bind_lock);
1774 free_pg_vec(pg_vec, order, req->tp_block_nr);
1779 static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1781 struct sock *sk = sock->sk;
1782 struct packet_sock *po = pkt_sk(sk);
1784 unsigned long start;
1791 size = vma->vm_end - vma->vm_start;
1794 if (po->pg_vec == NULL)
1796 if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
1799 start = vma->vm_start;
1800 for (i = 0; i < po->pg_vec_len; i++) {
1801 struct page *page = virt_to_page(po->pg_vec[i]);
1804 for (pg_num = 0; pg_num < po->pg_vec_pages; pg_num++, page++) {
1805 err = vm_insert_page(vma, start, page);
1811 atomic_inc(&po->mapped);
1812 vma->vm_ops = &packet_mmap_ops;
1822 static const struct proto_ops packet_ops_spkt = {
1823 .family = PF_PACKET,
1824 .owner = THIS_MODULE,
1825 .release = packet_release,
1826 .bind = packet_bind_spkt,
1827 .connect = sock_no_connect,
1828 .socketpair = sock_no_socketpair,
1829 .accept = sock_no_accept,
1830 .getname = packet_getname_spkt,
1831 .poll = datagram_poll,
1832 .ioctl = packet_ioctl,
1833 .listen = sock_no_listen,
1834 .shutdown = sock_no_shutdown,
1835 .setsockopt = sock_no_setsockopt,
1836 .getsockopt = sock_no_getsockopt,
1837 .sendmsg = packet_sendmsg_spkt,
1838 .recvmsg = packet_recvmsg,
1839 .mmap = sock_no_mmap,
1840 .sendpage = sock_no_sendpage,
1843 static const struct proto_ops packet_ops = {
1844 .family = PF_PACKET,
1845 .owner = THIS_MODULE,
1846 .release = packet_release,
1847 .bind = packet_bind,
1848 .connect = sock_no_connect,
1849 .socketpair = sock_no_socketpair,
1850 .accept = sock_no_accept,
1851 .getname = packet_getname,
1852 .poll = packet_poll,
1853 .ioctl = packet_ioctl,
1854 .listen = sock_no_listen,
1855 .shutdown = sock_no_shutdown,
1856 .setsockopt = packet_setsockopt,
1857 .getsockopt = packet_getsockopt,
1858 .sendmsg = packet_sendmsg,
1859 .recvmsg = packet_recvmsg,
1860 .mmap = packet_mmap,
1861 .sendpage = sock_no_sendpage,
1864 static struct net_proto_family packet_family_ops = {
1865 .family = PF_PACKET,
1866 .create = packet_create,
1867 .owner = THIS_MODULE,
1870 static struct notifier_block packet_netdev_notifier = {
1871 .notifier_call =packet_notifier,
1874 #ifdef CONFIG_PROC_FS
1875 static inline struct sock *packet_seq_idx(loff_t off)
1878 struct hlist_node *node;
1880 sk_for_each(s, node, &packet_sklist) {
1887 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
1889 read_lock(&packet_sklist_lock);
1890 return *pos ? packet_seq_idx(*pos - 1) : SEQ_START_TOKEN;
1893 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1896 return (v == SEQ_START_TOKEN)
1897 ? sk_head(&packet_sklist)
1898 : sk_next((struct sock*)v) ;
1901 static void packet_seq_stop(struct seq_file *seq, void *v)
1903 read_unlock(&packet_sklist_lock);
1906 static int packet_seq_show(struct seq_file *seq, void *v)
1908 if (v == SEQ_START_TOKEN)
1909 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
1912 const struct packet_sock *po = pkt_sk(s);
1915 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
1917 atomic_read(&s->sk_refcnt),
1922 atomic_read(&s->sk_rmem_alloc),
1930 static const struct seq_operations packet_seq_ops = {
1931 .start = packet_seq_start,
1932 .next = packet_seq_next,
1933 .stop = packet_seq_stop,
1934 .show = packet_seq_show,
1937 static int packet_seq_open(struct inode *inode, struct file *file)
1939 return seq_open(file, &packet_seq_ops);
1942 static const struct file_operations packet_seq_fops = {
1943 .owner = THIS_MODULE,
1944 .open = packet_seq_open,
1946 .llseek = seq_lseek,
1947 .release = seq_release,
1952 static void __exit packet_exit(void)
1954 proc_net_remove("packet");
1955 unregister_netdevice_notifier(&packet_netdev_notifier);
1956 sock_unregister(PF_PACKET);
1957 proto_unregister(&packet_proto);
1960 static int __init packet_init(void)
1962 int rc = proto_register(&packet_proto, 0);
1967 sock_register(&packet_family_ops);
1968 register_netdevice_notifier(&packet_netdev_notifier);
1969 proc_net_fops_create("packet", 0, &packet_seq_fops);
1974 module_init(packet_init);
1975 module_exit(packet_exit);
1976 MODULE_LICENSE("GPL");
1977 MODULE_ALIAS_NETPROTO(PF_PACKET);