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>
53 #include <linux/sched.h>
55 #include <linux/capability.h>
56 #include <linux/fcntl.h>
57 #include <linux/socket.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/if_packet.h>
62 #include <linux/wireless.h>
63 #include <linux/kmod.h>
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>
86 #define CONFIG_SOCK_PACKET 1
89 Proposed replacement for SIOC{ADD,DEL}MULTI and
90 IFF_PROMISC, IFF_ALLMULTI flags.
92 It is more expensive, but I believe,
93 it is really correct solution: reentereble, safe and fault tolerant.
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.
100 #define CONFIG_PACKET_MULTICAST 1
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
110 - packet socket receives packets with pulled ll header,
111 so that SOCK_RAW should push it back.
116 Incoming, dev->hard_header!=NULL
120 Outgoing, dev->hard_header!=NULL
124 Incoming, 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.
130 Outgoing, dev->hard_header==NULL
131 mac.raw -> data. ll header is still not built!
135 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
141 dev->hard_header != NULL
145 dev->hard_header == NULL (ll header is added by device, we cannot control it)
149 We should set nh.raw on output to correct posistion,
150 packet classifier depends on it.
153 /* List of all packet sockets. */
154 static HLIST_HEAD(packet_sklist);
155 static DEFINE_RWLOCK(packet_sklist_lock);
157 static atomic_t packet_socks_nr;
160 /* Private packet socket structures. */
162 #ifdef CONFIG_PACKET_MULTICAST
165 struct packet_mclist *next;
170 unsigned char addr[MAX_ADDR_LEN];
172 /* identical to struct packet_mreq except it has
173 * a longer address field.
175 struct packet_mreq_max
178 unsigned short mr_type;
179 unsigned short mr_alen;
180 unsigned char mr_address[MAX_ADDR_LEN];
183 #ifdef CONFIG_PACKET_MMAP
184 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
187 static void packet_flush_mclist(struct sock *sk);
190 /* struct sock has to be the first member of packet_sock */
192 struct tpacket_stats stats;
193 #ifdef CONFIG_PACKET_MMAP
196 unsigned int frames_per_block;
197 unsigned int frame_size;
198 unsigned int frame_max;
201 struct packet_type prot_hook;
202 spinlock_t bind_lock;
203 unsigned int running:1, /* prot_hook is attached*/
205 int ifindex; /* bound device */
207 #ifdef CONFIG_PACKET_MULTICAST
208 struct packet_mclist *mclist;
210 #ifdef CONFIG_PACKET_MMAP
212 unsigned int pg_vec_order;
213 unsigned int pg_vec_pages;
214 unsigned int pg_vec_len;
218 #define PACKET_SKB_CB(__skb) ((struct tpacket_auxdata *)((__skb)->cb))
220 #ifdef CONFIG_PACKET_MMAP
222 static inline char *packet_lookup_frame(struct packet_sock *po, unsigned int position)
224 unsigned int pg_vec_pos, frame_offset;
227 pg_vec_pos = position / po->frames_per_block;
228 frame_offset = position % po->frames_per_block;
230 frame = po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size);
236 static inline struct packet_sock *pkt_sk(struct sock *sk)
238 return (struct packet_sock *)sk;
241 static void packet_sock_destruct(struct sock *sk)
243 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
244 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
246 if (!sock_flag(sk, SOCK_DEAD)) {
247 printk("Attempt to release alive packet socket: %p\n", sk);
251 atomic_dec(&packet_socks_nr);
252 #ifdef PACKET_REFCNT_DEBUG
253 printk(KERN_DEBUG "PACKET socket %p is free, %d are alive\n", sk, atomic_read(&packet_socks_nr));
258 static const struct proto_ops packet_ops;
260 #ifdef CONFIG_SOCK_PACKET
261 static const struct proto_ops packet_ops_spkt;
263 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
266 struct sockaddr_pkt *spkt;
269 * When we registered the protocol we saved the socket in the data
270 * field for just this event.
273 sk = pt->af_packet_priv;
276 * Yank back the headers [hope the device set this
277 * right or kerboom...]
279 * Incoming packets have ll header pulled,
282 * For outgoing ones skb->data == skb->mac.raw
283 * so that this procedure is noop.
286 if (skb->pkt_type == PACKET_LOOPBACK)
289 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
292 /* drop any routing info */
293 dst_release(skb->dst);
296 /* drop conntrack reference */
299 spkt = (struct sockaddr_pkt*)skb->cb;
301 skb_push(skb, skb->data-skb->mac.raw);
304 * The SOCK_PACKET socket receives _all_ frames.
307 spkt->spkt_family = dev->type;
308 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
309 spkt->spkt_protocol = skb->protocol;
312 * Charge the memory to the socket. This is done specifically
313 * to prevent sockets using all the memory up.
316 if (sock_queue_rcv_skb(sk,skb) == 0)
327 * Output a raw packet to a device layer. This bypasses all the other
328 * protocol layers and you must therefore supply it with a complete frame
331 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
332 struct msghdr *msg, size_t len)
334 struct sock *sk = sock->sk;
335 struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
337 struct net_device *dev;
342 * Get and verify the address.
347 if (msg->msg_namelen < sizeof(struct sockaddr))
349 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
350 proto=saddr->spkt_protocol;
353 return(-ENOTCONN); /* SOCK_PACKET must be sent giving an address */
356 * Find the device first to size check it
359 saddr->spkt_device[13] = 0;
360 dev = dev_get_by_name(saddr->spkt_device);
366 if (!(dev->flags & IFF_UP))
370 * You may not queue a frame bigger than the mtu. This is the lowest level
371 * raw protocol and you must do your own fragmentation at this level.
375 if (len > dev->mtu + dev->hard_header_len)
379 skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
382 * If the write buffer is full, then tough. At this level the user gets to
383 * deal with the problem - do your own algorithmic backoffs. That's far
394 /* FIXME: Save some space for broken drivers that write a
395 * hard header at transmission time by themselves. PPP is the
396 * notable one here. This should really be fixed at the driver level.
398 skb_reserve(skb, LL_RESERVED_SPACE(dev));
399 skb->nh.raw = skb->data;
401 /* Try to align data part correctly */
402 if (dev->hard_header) {
403 skb->data -= dev->hard_header_len;
404 skb->tail -= dev->hard_header_len;
405 if (len < dev->hard_header_len)
406 skb->nh.raw = skb->data;
409 /* Returns -EFAULT on error */
410 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
411 skb->protocol = proto;
413 skb->priority = sk->sk_priority;
434 static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
437 struct sk_filter *filter;
440 filter = rcu_dereference(sk->sk_filter);
442 res = sk_run_filter(skb, filter->insns, filter->len);
443 rcu_read_unlock_bh();
449 This function makes lazy skb cloning in hope that most of packets
450 are discarded by BPF.
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.
460 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
463 struct sockaddr_ll *sll;
464 struct packet_sock *po;
465 u8 * skb_head = skb->data;
466 int skb_len = skb->len;
467 unsigned int snaplen, res;
468 struct tpacket_auxdata *aux;
470 if (skb->pkt_type == PACKET_LOOPBACK)
473 sk = pt->af_packet_priv;
478 if (dev->hard_header) {
479 /* The device has an explicit notion of ll header,
480 exported to higher levels.
482 Otherwise, the device hides datails of it frame
483 structure, so that corresponding packet head
484 never delivered to user.
486 if (sk->sk_type != SOCK_DGRAM)
487 skb_push(skb, skb->data - skb->mac.raw);
488 else if (skb->pkt_type == PACKET_OUTGOING) {
489 /* Special case: outgoing packets have ll header at head */
490 skb_pull(skb, skb->nh.raw - skb->data);
496 res = run_filter(skb, sk, snaplen);
502 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
503 (unsigned)sk->sk_rcvbuf)
506 if (skb_shared(skb)) {
507 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
511 if (skb_head != skb->data) {
512 skb->data = skb_head;
519 sll = (struct sockaddr_ll*)skb->cb;
520 sll->sll_family = AF_PACKET;
521 sll->sll_hatype = dev->type;
522 sll->sll_protocol = skb->protocol;
523 sll->sll_pkttype = skb->pkt_type;
524 sll->sll_ifindex = dev->ifindex;
527 if (dev->hard_header_parse)
528 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
530 aux = PACKET_SKB_CB(skb);
531 aux->tp_status = TP_STATUS_USER;
532 if (skb->ip_summed == CHECKSUM_PARTIAL)
533 aux->tp_status |= TP_STATUS_CSUMNOTREADY;
534 aux->tp_len = skb->len;
535 aux->tp_snaplen = snaplen;
537 aux->tp_net = skb->nh.raw - skb->data;
539 if (pskb_trim(skb, snaplen))
542 skb_set_owner_r(skb, sk);
544 dst_release(skb->dst);
547 /* drop conntrack reference */
550 spin_lock(&sk->sk_receive_queue.lock);
551 po->stats.tp_packets++;
552 __skb_queue_tail(&sk->sk_receive_queue, skb);
553 spin_unlock(&sk->sk_receive_queue.lock);
554 sk->sk_data_ready(sk, skb->len);
558 spin_lock(&sk->sk_receive_queue.lock);
559 po->stats.tp_drops++;
560 spin_unlock(&sk->sk_receive_queue.lock);
563 if (skb_head != skb->data && skb_shared(skb)) {
564 skb->data = skb_head;
572 #ifdef CONFIG_PACKET_MMAP
573 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
576 struct packet_sock *po;
577 struct sockaddr_ll *sll;
578 struct tpacket_hdr *h;
579 u8 * skb_head = skb->data;
580 int skb_len = skb->len;
581 unsigned int snaplen, res;
582 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
583 unsigned short macoff, netoff;
584 struct sk_buff *copy_skb = NULL;
586 if (skb->pkt_type == PACKET_LOOPBACK)
589 sk = pt->af_packet_priv;
592 if (dev->hard_header) {
593 if (sk->sk_type != SOCK_DGRAM)
594 skb_push(skb, skb->data - skb->mac.raw);
595 else if (skb->pkt_type == PACKET_OUTGOING) {
596 /* Special case: outgoing packets have ll header at head */
597 skb_pull(skb, skb->nh.raw - skb->data);
601 if (skb->ip_summed == CHECKSUM_PARTIAL)
602 status |= TP_STATUS_CSUMNOTREADY;
606 res = run_filter(skb, sk, snaplen);
612 if (sk->sk_type == SOCK_DGRAM) {
613 macoff = netoff = TPACKET_ALIGN(TPACKET_HDRLEN) + 16;
615 unsigned maclen = skb->nh.raw - skb->data;
616 netoff = TPACKET_ALIGN(TPACKET_HDRLEN + (maclen < 16 ? 16 : maclen));
617 macoff = netoff - maclen;
620 if (macoff + snaplen > po->frame_size) {
621 if (po->copy_thresh &&
622 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
623 (unsigned)sk->sk_rcvbuf) {
624 if (skb_shared(skb)) {
625 copy_skb = skb_clone(skb, GFP_ATOMIC);
627 copy_skb = skb_get(skb);
628 skb_head = skb->data;
631 skb_set_owner_r(copy_skb, sk);
633 snaplen = po->frame_size - macoff;
634 if ((int)snaplen < 0)
638 spin_lock(&sk->sk_receive_queue.lock);
639 h = (struct tpacket_hdr *)packet_lookup_frame(po, po->head);
643 po->head = po->head != po->frame_max ? po->head+1 : 0;
644 po->stats.tp_packets++;
646 status |= TP_STATUS_COPY;
647 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
649 if (!po->stats.tp_drops)
650 status &= ~TP_STATUS_LOSING;
651 spin_unlock(&sk->sk_receive_queue.lock);
653 skb_copy_bits(skb, 0, (u8*)h + macoff, snaplen);
655 h->tp_len = skb->len;
656 h->tp_snaplen = snaplen;
659 if (skb->tstamp.off_sec == 0) {
660 __net_timestamp(skb);
661 sock_enable_timestamp(sk);
663 h->tp_sec = skb->tstamp.off_sec;
664 h->tp_usec = skb->tstamp.off_usec;
666 sll = (struct sockaddr_ll*)((u8*)h + TPACKET_ALIGN(sizeof(*h)));
668 if (dev->hard_header_parse)
669 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
670 sll->sll_family = AF_PACKET;
671 sll->sll_hatype = dev->type;
672 sll->sll_protocol = skb->protocol;
673 sll->sll_pkttype = skb->pkt_type;
674 sll->sll_ifindex = dev->ifindex;
676 h->tp_status = status;
680 struct page *p_start, *p_end;
681 u8 *h_end = (u8 *)h + macoff + snaplen - 1;
683 p_start = virt_to_page(h);
684 p_end = virt_to_page(h_end);
685 while (p_start <= p_end) {
686 flush_dcache_page(p_start);
691 sk->sk_data_ready(sk, 0);
694 if (skb_head != skb->data && skb_shared(skb)) {
695 skb->data = skb_head;
703 po->stats.tp_drops++;
704 spin_unlock(&sk->sk_receive_queue.lock);
706 sk->sk_data_ready(sk, 0);
715 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
716 struct msghdr *msg, size_t len)
718 struct sock *sk = sock->sk;
719 struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
721 struct net_device *dev;
724 int ifindex, err, reserve = 0;
727 * Get and verify the address.
731 struct packet_sock *po = pkt_sk(sk);
733 ifindex = po->ifindex;
738 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
740 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
742 ifindex = saddr->sll_ifindex;
743 proto = saddr->sll_protocol;
744 addr = saddr->sll_addr;
748 dev = dev_get_by_index(ifindex);
752 if (sock->type == SOCK_RAW)
753 reserve = dev->hard_header_len;
756 if (!(dev->flags & IFF_UP))
760 if (len > dev->mtu+reserve)
763 skb = sock_alloc_send_skb(sk, len + LL_RESERVED_SPACE(dev),
764 msg->msg_flags & MSG_DONTWAIT, &err);
768 skb_reserve(skb, LL_RESERVED_SPACE(dev));
769 skb->nh.raw = skb->data;
771 if (dev->hard_header) {
774 res = dev->hard_header(skb, dev, ntohs(proto), addr, NULL, len);
775 if (sock->type != SOCK_DGRAM) {
776 skb->tail = skb->data;
782 /* Returns -EFAULT on error */
783 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
787 skb->protocol = proto;
789 skb->priority = sk->sk_priority;
795 err = dev_queue_xmit(skb);
796 if (err > 0 && (err = net_xmit_errno(err)) != 0)
813 * Close a PACKET socket. This is fairly simple. We immediately go
814 * to 'closed' state and remove our protocol entry in the device list.
817 static int packet_release(struct socket *sock)
819 struct sock *sk = sock->sk;
820 struct packet_sock *po;
827 write_lock_bh(&packet_sklist_lock);
828 sk_del_node_init(sk);
829 write_unlock_bh(&packet_sklist_lock);
832 * Unhook packet receive handler.
837 * Remove the protocol hook
839 dev_remove_pack(&po->prot_hook);
845 #ifdef CONFIG_PACKET_MULTICAST
846 packet_flush_mclist(sk);
849 #ifdef CONFIG_PACKET_MMAP
851 struct tpacket_req req;
852 memset(&req, 0, sizeof(req));
853 packet_set_ring(sk, &req, 1);
858 * Now the socket is dead. No more input will appear.
866 skb_queue_purge(&sk->sk_receive_queue);
873 * Attach a packet hook.
876 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
878 struct packet_sock *po = pkt_sk(sk);
880 * Detach an existing hook if present.
885 spin_lock(&po->bind_lock);
890 spin_unlock(&po->bind_lock);
891 dev_remove_pack(&po->prot_hook);
892 spin_lock(&po->bind_lock);
896 po->prot_hook.type = protocol;
897 po->prot_hook.dev = dev;
899 po->ifindex = dev ? dev->ifindex : 0;
905 if (dev->flags&IFF_UP) {
906 dev_add_pack(&po->prot_hook);
910 sk->sk_err = ENETDOWN;
911 if (!sock_flag(sk, SOCK_DEAD))
912 sk->sk_error_report(sk);
915 dev_add_pack(&po->prot_hook);
921 spin_unlock(&po->bind_lock);
927 * Bind a packet socket to a device
930 #ifdef CONFIG_SOCK_PACKET
932 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
934 struct sock *sk=sock->sk;
936 struct net_device *dev;
943 if (addr_len != sizeof(struct sockaddr))
945 strlcpy(name,uaddr->sa_data,sizeof(name));
947 dev = dev_get_by_name(name);
949 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
956 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
958 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
959 struct sock *sk=sock->sk;
960 struct net_device *dev = NULL;
968 if (addr_len < sizeof(struct sockaddr_ll))
970 if (sll->sll_family != AF_PACKET)
973 if (sll->sll_ifindex) {
975 dev = dev_get_by_index(sll->sll_ifindex);
979 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
987 static struct proto packet_proto = {
989 .owner = THIS_MODULE,
990 .obj_size = sizeof(struct packet_sock),
994 * Create a packet of type SOCK_PACKET.
997 static int packet_create(struct socket *sock, int protocol)
1000 struct packet_sock *po;
1001 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1004 if (!capable(CAP_NET_RAW))
1006 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW
1007 #ifdef CONFIG_SOCK_PACKET
1008 && sock->type != SOCK_PACKET
1011 return -ESOCKTNOSUPPORT;
1013 sock->state = SS_UNCONNECTED;
1016 sk = sk_alloc(PF_PACKET, GFP_KERNEL, &packet_proto, 1);
1020 sock->ops = &packet_ops;
1021 #ifdef CONFIG_SOCK_PACKET
1022 if (sock->type == SOCK_PACKET)
1023 sock->ops = &packet_ops_spkt;
1025 sock_init_data(sock, sk);
1028 sk->sk_family = PF_PACKET;
1031 sk->sk_destruct = packet_sock_destruct;
1032 atomic_inc(&packet_socks_nr);
1035 * Attach a protocol block
1038 spin_lock_init(&po->bind_lock);
1039 po->prot_hook.func = packet_rcv;
1040 #ifdef CONFIG_SOCK_PACKET
1041 if (sock->type == SOCK_PACKET)
1042 po->prot_hook.func = packet_rcv_spkt;
1044 po->prot_hook.af_packet_priv = sk;
1047 po->prot_hook.type = proto;
1048 dev_add_pack(&po->prot_hook);
1053 write_lock_bh(&packet_sklist_lock);
1054 sk_add_node(sk, &packet_sklist);
1055 write_unlock_bh(&packet_sklist_lock);
1062 * Pull a packet from our receive queue and hand it to the user.
1063 * If necessary we block.
1066 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1067 struct msghdr *msg, size_t len, int flags)
1069 struct sock *sk = sock->sk;
1070 struct sk_buff *skb;
1072 struct sockaddr_ll *sll;
1075 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1079 /* What error should we return now? EUNATTACH? */
1080 if (pkt_sk(sk)->ifindex < 0)
1085 * Call the generic datagram receiver. This handles all sorts
1086 * of horrible races and re-entrancy so we can forget about it
1087 * in the protocol layers.
1089 * Now it will return ENETDOWN, if device have just gone down,
1090 * but then it will block.
1093 skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1096 * An error occurred so return it. Because skb_recv_datagram()
1097 * handles the blocking we don't see and worry about blocking
1105 * If the address length field is there to be filled in, we fill
1109 sll = (struct sockaddr_ll*)skb->cb;
1110 if (sock->type == SOCK_PACKET)
1111 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1113 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1116 * You lose any data beyond the buffer you gave. If it worries a
1117 * user program they can ask the device for its MTU anyway.
1124 msg->msg_flags|=MSG_TRUNC;
1127 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1131 sock_recv_timestamp(msg, sk, skb);
1134 memcpy(msg->msg_name, skb->cb, msg->msg_namelen);
1136 if (pkt_sk(sk)->auxdata) {
1137 struct tpacket_auxdata *aux = PACKET_SKB_CB(skb);
1138 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(*aux), aux);
1142 * Free or return the buffer as appropriate. Again this
1143 * hides all the races and re-entrancy issues from us.
1145 err = (flags&MSG_TRUNC) ? skb->len : copied;
1148 skb_free_datagram(sk, skb);
1153 #ifdef CONFIG_SOCK_PACKET
1154 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1155 int *uaddr_len, int peer)
1157 struct net_device *dev;
1158 struct sock *sk = sock->sk;
1163 uaddr->sa_family = AF_PACKET;
1164 dev = dev_get_by_index(pkt_sk(sk)->ifindex);
1166 strlcpy(uaddr->sa_data, dev->name, 15);
1169 memset(uaddr->sa_data, 0, 14);
1170 *uaddr_len = sizeof(*uaddr);
1176 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1177 int *uaddr_len, int peer)
1179 struct net_device *dev;
1180 struct sock *sk = sock->sk;
1181 struct packet_sock *po = pkt_sk(sk);
1182 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1187 sll->sll_family = AF_PACKET;
1188 sll->sll_ifindex = po->ifindex;
1189 sll->sll_protocol = po->num;
1190 dev = dev_get_by_index(po->ifindex);
1192 sll->sll_hatype = dev->type;
1193 sll->sll_halen = dev->addr_len;
1194 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1197 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1200 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1205 #ifdef CONFIG_PACKET_MULTICAST
1206 static void packet_dev_mc(struct net_device *dev, struct packet_mclist *i, int what)
1209 case PACKET_MR_MULTICAST:
1211 dev_mc_add(dev, i->addr, i->alen, 0);
1213 dev_mc_delete(dev, i->addr, i->alen, 0);
1215 case PACKET_MR_PROMISC:
1216 dev_set_promiscuity(dev, what);
1218 case PACKET_MR_ALLMULTI:
1219 dev_set_allmulti(dev, what);
1225 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1227 for ( ; i; i=i->next) {
1228 if (i->ifindex == dev->ifindex)
1229 packet_dev_mc(dev, i, what);
1233 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1235 struct packet_sock *po = pkt_sk(sk);
1236 struct packet_mclist *ml, *i;
1237 struct net_device *dev;
1243 dev = __dev_get_by_index(mreq->mr_ifindex);
1248 if (mreq->mr_alen > dev->addr_len)
1252 i = kmalloc(sizeof(*i), GFP_KERNEL);
1257 for (ml = po->mclist; ml; ml = ml->next) {
1258 if (ml->ifindex == mreq->mr_ifindex &&
1259 ml->type == mreq->mr_type &&
1260 ml->alen == mreq->mr_alen &&
1261 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1263 /* Free the new element ... */
1269 i->type = mreq->mr_type;
1270 i->ifindex = mreq->mr_ifindex;
1271 i->alen = mreq->mr_alen;
1272 memcpy(i->addr, mreq->mr_address, i->alen);
1274 i->next = po->mclist;
1276 packet_dev_mc(dev, i, +1);
1283 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1285 struct packet_mclist *ml, **mlp;
1289 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1290 if (ml->ifindex == mreq->mr_ifindex &&
1291 ml->type == mreq->mr_type &&
1292 ml->alen == mreq->mr_alen &&
1293 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1294 if (--ml->count == 0) {
1295 struct net_device *dev;
1297 dev = dev_get_by_index(ml->ifindex);
1299 packet_dev_mc(dev, ml, -1);
1309 return -EADDRNOTAVAIL;
1312 static void packet_flush_mclist(struct sock *sk)
1314 struct packet_sock *po = pkt_sk(sk);
1315 struct packet_mclist *ml;
1321 while ((ml = po->mclist) != NULL) {
1322 struct net_device *dev;
1324 po->mclist = ml->next;
1325 if ((dev = dev_get_by_index(ml->ifindex)) != NULL) {
1326 packet_dev_mc(dev, ml, -1);
1336 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1338 struct sock *sk = sock->sk;
1339 struct packet_sock *po = pkt_sk(sk);
1342 if (level != SOL_PACKET)
1343 return -ENOPROTOOPT;
1346 #ifdef CONFIG_PACKET_MULTICAST
1347 case PACKET_ADD_MEMBERSHIP:
1348 case PACKET_DROP_MEMBERSHIP:
1350 struct packet_mreq_max mreq;
1352 memset(&mreq, 0, sizeof(mreq));
1353 if (len < sizeof(struct packet_mreq))
1355 if (len > sizeof(mreq))
1357 if (copy_from_user(&mreq,optval,len))
1359 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1361 if (optname == PACKET_ADD_MEMBERSHIP)
1362 ret = packet_mc_add(sk, &mreq);
1364 ret = packet_mc_drop(sk, &mreq);
1368 #ifdef CONFIG_PACKET_MMAP
1369 case PACKET_RX_RING:
1371 struct tpacket_req req;
1373 if (optlen<sizeof(req))
1375 if (copy_from_user(&req,optval,sizeof(req)))
1377 return packet_set_ring(sk, &req, 0);
1379 case PACKET_COPY_THRESH:
1383 if (optlen!=sizeof(val))
1385 if (copy_from_user(&val,optval,sizeof(val)))
1388 pkt_sk(sk)->copy_thresh = val;
1392 case PACKET_AUXDATA:
1396 if (optlen < sizeof(val))
1398 if (copy_from_user(&val, optval, sizeof(val)))
1401 po->auxdata = !!val;
1405 return -ENOPROTOOPT;
1409 static int packet_getsockopt(struct socket *sock, int level, int optname,
1410 char __user *optval, int __user *optlen)
1414 struct sock *sk = sock->sk;
1415 struct packet_sock *po = pkt_sk(sk);
1417 struct tpacket_stats st;
1419 if (level != SOL_PACKET)
1420 return -ENOPROTOOPT;
1422 if (get_user(len, optlen))
1429 case PACKET_STATISTICS:
1430 if (len > sizeof(struct tpacket_stats))
1431 len = sizeof(struct tpacket_stats);
1432 spin_lock_bh(&sk->sk_receive_queue.lock);
1434 memset(&po->stats, 0, sizeof(st));
1435 spin_unlock_bh(&sk->sk_receive_queue.lock);
1436 st.tp_packets += st.tp_drops;
1440 case PACKET_AUXDATA:
1441 if (len > sizeof(int))
1448 return -ENOPROTOOPT;
1451 if (put_user(len, optlen))
1453 if (copy_to_user(optval, data, len))
1459 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1462 struct hlist_node *node;
1463 struct net_device *dev = (struct net_device*)data;
1465 read_lock(&packet_sklist_lock);
1466 sk_for_each(sk, node, &packet_sklist) {
1467 struct packet_sock *po = pkt_sk(sk);
1470 case NETDEV_UNREGISTER:
1471 #ifdef CONFIG_PACKET_MULTICAST
1473 packet_dev_mclist(dev, po->mclist, -1);
1477 if (dev->ifindex == po->ifindex) {
1478 spin_lock(&po->bind_lock);
1480 __dev_remove_pack(&po->prot_hook);
1483 sk->sk_err = ENETDOWN;
1484 if (!sock_flag(sk, SOCK_DEAD))
1485 sk->sk_error_report(sk);
1487 if (msg == NETDEV_UNREGISTER) {
1489 po->prot_hook.dev = NULL;
1491 spin_unlock(&po->bind_lock);
1495 spin_lock(&po->bind_lock);
1496 if (dev->ifindex == po->ifindex && po->num &&
1498 dev_add_pack(&po->prot_hook);
1502 spin_unlock(&po->bind_lock);
1506 read_unlock(&packet_sklist_lock);
1511 static int packet_ioctl(struct socket *sock, unsigned int cmd,
1514 struct sock *sk = sock->sk;
1519 int amount = atomic_read(&sk->sk_wmem_alloc);
1520 return put_user(amount, (int __user *)arg);
1524 struct sk_buff *skb;
1527 spin_lock_bh(&sk->sk_receive_queue.lock);
1528 skb = skb_peek(&sk->sk_receive_queue);
1531 spin_unlock_bh(&sk->sk_receive_queue.lock);
1532 return put_user(amount, (int __user *)arg);
1535 return sock_get_timestamp(sk, (struct timeval __user *)arg);
1545 case SIOCGIFBRDADDR:
1546 case SIOCSIFBRDADDR:
1547 case SIOCGIFNETMASK:
1548 case SIOCSIFNETMASK:
1549 case SIOCGIFDSTADDR:
1550 case SIOCSIFDSTADDR:
1552 return inet_dgram_ops.ioctl(sock, cmd, arg);
1556 return -ENOIOCTLCMD;
1561 #ifndef CONFIG_PACKET_MMAP
1562 #define packet_mmap sock_no_mmap
1563 #define packet_poll datagram_poll
1566 static unsigned int packet_poll(struct file * file, struct socket *sock,
1569 struct sock *sk = sock->sk;
1570 struct packet_sock *po = pkt_sk(sk);
1571 unsigned int mask = datagram_poll(file, sock, wait);
1573 spin_lock_bh(&sk->sk_receive_queue.lock);
1575 unsigned last = po->head ? po->head-1 : po->frame_max;
1576 struct tpacket_hdr *h;
1578 h = (struct tpacket_hdr *)packet_lookup_frame(po, last);
1581 mask |= POLLIN | POLLRDNORM;
1583 spin_unlock_bh(&sk->sk_receive_queue.lock);
1588 /* Dirty? Well, I still did not learn better way to account
1592 static void packet_mm_open(struct vm_area_struct *vma)
1594 struct file *file = vma->vm_file;
1595 struct socket * sock = file->private_data;
1596 struct sock *sk = sock->sk;
1599 atomic_inc(&pkt_sk(sk)->mapped);
1602 static void packet_mm_close(struct vm_area_struct *vma)
1604 struct file *file = vma->vm_file;
1605 struct socket * sock = file->private_data;
1606 struct sock *sk = sock->sk;
1609 atomic_dec(&pkt_sk(sk)->mapped);
1612 static struct vm_operations_struct packet_mmap_ops = {
1613 .open = packet_mm_open,
1614 .close =packet_mm_close,
1617 static inline struct page *pg_vec_endpage(char *one_pg_vec, unsigned int order)
1619 return virt_to_page(one_pg_vec + (PAGE_SIZE << order) - 1);
1622 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
1626 for (i = 0; i < len; i++) {
1627 if (likely(pg_vec[i]))
1628 free_pages((unsigned long) pg_vec[i], order);
1633 static inline char *alloc_one_pg_vec_page(unsigned long order)
1635 return (char *) __get_free_pages(GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
1639 static char **alloc_pg_vec(struct tpacket_req *req, int order)
1641 unsigned int block_nr = req->tp_block_nr;
1645 pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
1646 if (unlikely(!pg_vec))
1649 for (i = 0; i < block_nr; i++) {
1650 pg_vec[i] = alloc_one_pg_vec_page(order);
1651 if (unlikely(!pg_vec[i]))
1652 goto out_free_pgvec;
1659 free_pg_vec(pg_vec, order, block_nr);
1664 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
1666 char **pg_vec = NULL;
1667 struct packet_sock *po = pkt_sk(sk);
1668 int was_running, order = 0;
1672 if (req->tp_block_nr) {
1675 /* Sanity tests and some calculations */
1677 if (unlikely(po->pg_vec))
1680 if (unlikely((int)req->tp_block_size <= 0))
1682 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
1684 if (unlikely(req->tp_frame_size < TPACKET_HDRLEN))
1686 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
1689 po->frames_per_block = req->tp_block_size/req->tp_frame_size;
1690 if (unlikely(po->frames_per_block <= 0))
1692 if (unlikely((po->frames_per_block * req->tp_block_nr) !=
1697 order = get_order(req->tp_block_size);
1698 pg_vec = alloc_pg_vec(req, order);
1699 if (unlikely(!pg_vec))
1703 for (i = 0; i < req->tp_block_nr; i++) {
1704 char *ptr = pg_vec[i];
1705 struct tpacket_hdr *header;
1708 for (k = 0; k < po->frames_per_block; k++) {
1709 header = (struct tpacket_hdr *) ptr;
1710 header->tp_status = TP_STATUS_KERNEL;
1711 ptr += req->tp_frame_size;
1716 if (unlikely(req->tp_frame_nr))
1722 /* Detach socket from network */
1723 spin_lock(&po->bind_lock);
1724 was_running = po->running;
1727 __dev_remove_pack(&po->prot_hook);
1732 spin_unlock(&po->bind_lock);
1737 if (closing || atomic_read(&po->mapped) == 0) {
1739 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
1741 spin_lock_bh(&sk->sk_receive_queue.lock);
1742 pg_vec = XC(po->pg_vec, pg_vec);
1743 po->frame_max = (req->tp_frame_nr - 1);
1745 po->frame_size = req->tp_frame_size;
1746 spin_unlock_bh(&sk->sk_receive_queue.lock);
1748 order = XC(po->pg_vec_order, order);
1749 req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
1751 po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
1752 po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
1753 skb_queue_purge(&sk->sk_receive_queue);
1755 if (atomic_read(&po->mapped))
1756 printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
1759 spin_lock(&po->bind_lock);
1760 if (was_running && !po->running) {
1764 dev_add_pack(&po->prot_hook);
1766 spin_unlock(&po->bind_lock);
1771 free_pg_vec(pg_vec, order, req->tp_block_nr);
1776 static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1778 struct sock *sk = sock->sk;
1779 struct packet_sock *po = pkt_sk(sk);
1781 unsigned long start;
1788 size = vma->vm_end - vma->vm_start;
1791 if (po->pg_vec == NULL)
1793 if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
1796 start = vma->vm_start;
1797 for (i = 0; i < po->pg_vec_len; i++) {
1798 struct page *page = virt_to_page(po->pg_vec[i]);
1801 for (pg_num = 0; pg_num < po->pg_vec_pages; pg_num++, page++) {
1802 err = vm_insert_page(vma, start, page);
1808 atomic_inc(&po->mapped);
1809 vma->vm_ops = &packet_mmap_ops;
1819 #ifdef CONFIG_SOCK_PACKET
1820 static const struct proto_ops packet_ops_spkt = {
1821 .family = PF_PACKET,
1822 .owner = THIS_MODULE,
1823 .release = packet_release,
1824 .bind = packet_bind_spkt,
1825 .connect = sock_no_connect,
1826 .socketpair = sock_no_socketpair,
1827 .accept = sock_no_accept,
1828 .getname = packet_getname_spkt,
1829 .poll = datagram_poll,
1830 .ioctl = packet_ioctl,
1831 .listen = sock_no_listen,
1832 .shutdown = sock_no_shutdown,
1833 .setsockopt = sock_no_setsockopt,
1834 .getsockopt = sock_no_getsockopt,
1835 .sendmsg = packet_sendmsg_spkt,
1836 .recvmsg = packet_recvmsg,
1837 .mmap = sock_no_mmap,
1838 .sendpage = sock_no_sendpage,
1842 static const struct proto_ops packet_ops = {
1843 .family = PF_PACKET,
1844 .owner = THIS_MODULE,
1845 .release = packet_release,
1846 .bind = packet_bind,
1847 .connect = sock_no_connect,
1848 .socketpair = sock_no_socketpair,
1849 .accept = sock_no_accept,
1850 .getname = packet_getname,
1851 .poll = packet_poll,
1852 .ioctl = packet_ioctl,
1853 .listen = sock_no_listen,
1854 .shutdown = sock_no_shutdown,
1855 .setsockopt = packet_setsockopt,
1856 .getsockopt = packet_getsockopt,
1857 .sendmsg = packet_sendmsg,
1858 .recvmsg = packet_recvmsg,
1859 .mmap = packet_mmap,
1860 .sendpage = sock_no_sendpage,
1863 static struct net_proto_family packet_family_ops = {
1864 .family = PF_PACKET,
1865 .create = packet_create,
1866 .owner = THIS_MODULE,
1869 static struct notifier_block packet_netdev_notifier = {
1870 .notifier_call =packet_notifier,
1873 #ifdef CONFIG_PROC_FS
1874 static inline struct sock *packet_seq_idx(loff_t off)
1877 struct hlist_node *node;
1879 sk_for_each(s, node, &packet_sklist) {
1886 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
1888 read_lock(&packet_sklist_lock);
1889 return *pos ? packet_seq_idx(*pos - 1) : SEQ_START_TOKEN;
1892 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1895 return (v == SEQ_START_TOKEN)
1896 ? sk_head(&packet_sklist)
1897 : sk_next((struct sock*)v) ;
1900 static void packet_seq_stop(struct seq_file *seq, void *v)
1902 read_unlock(&packet_sklist_lock);
1905 static int packet_seq_show(struct seq_file *seq, void *v)
1907 if (v == SEQ_START_TOKEN)
1908 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
1911 const struct packet_sock *po = pkt_sk(s);
1914 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
1916 atomic_read(&s->sk_refcnt),
1921 atomic_read(&s->sk_rmem_alloc),
1929 static struct seq_operations packet_seq_ops = {
1930 .start = packet_seq_start,
1931 .next = packet_seq_next,
1932 .stop = packet_seq_stop,
1933 .show = packet_seq_show,
1936 static int packet_seq_open(struct inode *inode, struct file *file)
1938 return seq_open(file, &packet_seq_ops);
1941 static struct file_operations packet_seq_fops = {
1942 .owner = THIS_MODULE,
1943 .open = packet_seq_open,
1945 .llseek = seq_lseek,
1946 .release = seq_release,
1951 static void __exit packet_exit(void)
1953 proc_net_remove("packet");
1954 unregister_netdevice_notifier(&packet_netdev_notifier);
1955 sock_unregister(PF_PACKET);
1956 proto_unregister(&packet_proto);
1959 static int __init packet_init(void)
1961 int rc = proto_register(&packet_proto, 0);
1966 sock_register(&packet_family_ops);
1967 register_netdevice_notifier(&packet_netdev_notifier);
1968 proc_net_fops_create("packet", 0, &packet_seq_fops);
1973 module_init(packet_init);
1974 module_exit(packet_exit);
1975 MODULE_LICENSE("GPL");
1976 MODULE_ALIAS_NETPROTO(PF_PACKET);