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.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
44 * This program is free software; you can redistribute it and/or
45 * modify it under the terms of the GNU General Public License
46 * as published by the Free Software Foundation; either version
47 * 2 of the License, or (at your option) any later version.
51 #include <linux/types.h>
53 #include <linux/capability.h>
54 #include <linux/fcntl.h>
55 #include <linux/socket.h>
57 #include <linux/inet.h>
58 #include <linux/netdevice.h>
59 #include <linux/if_packet.h>
60 #include <linux/wireless.h>
61 #include <linux/kernel.h>
62 #include <linux/kmod.h>
63 #include <net/net_namespace.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>
81 #include <linux/mutex.h>
84 #include <net/inet_common.h>
89 - if device has no dev->hard_header routine, it adds and removes ll header
90 inside itself. In this case ll header is invisible outside of device,
91 but higher levels still should reserve dev->hard_header_len.
92 Some devices are enough clever to reallocate skb, when header
93 will not fit to reserved space (tunnel), another ones are silly
95 - packet socket receives packets with pulled ll header,
96 so that SOCK_RAW should push it back.
101 Incoming, dev->hard_header!=NULL
102 mac_header -> ll header
105 Outgoing, dev->hard_header!=NULL
106 mac_header -> ll header
109 Incoming, dev->hard_header==NULL
110 mac_header -> UNKNOWN position. It is very likely, that it points to ll
111 header. PPP makes it, that is wrong, because introduce
112 assymetry between rx and tx paths.
115 Outgoing, dev->hard_header==NULL
116 mac_header -> data. ll header is still not built!
120 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
126 dev->hard_header != NULL
127 mac_header -> ll header
130 dev->hard_header == NULL (ll header is added by device, we cannot control it)
134 We should set nh.raw on output to correct posistion,
135 packet classifier depends on it.
138 /* Private packet socket structures. */
142 struct packet_mclist *next;
147 unsigned char addr[MAX_ADDR_LEN];
149 /* identical to struct packet_mreq except it has
150 * a longer address field.
152 struct packet_mreq_max
155 unsigned short mr_type;
156 unsigned short mr_alen;
157 unsigned char mr_address[MAX_ADDR_LEN];
160 #ifdef CONFIG_PACKET_MMAP
161 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
162 int closing, int tx_ring);
164 struct packet_ring_buffer {
167 unsigned int frames_per_block;
168 unsigned int frame_size;
169 unsigned int frame_max;
171 unsigned int pg_vec_order;
172 unsigned int pg_vec_pages;
173 unsigned int pg_vec_len;
179 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
182 static void packet_flush_mclist(struct sock *sk);
185 /* struct sock has to be the first member of packet_sock */
187 struct tpacket_stats stats;
188 #ifdef CONFIG_PACKET_MMAP
189 struct packet_ring_buffer rx_ring;
190 struct packet_ring_buffer tx_ring;
193 struct packet_type prot_hook;
194 spinlock_t bind_lock;
195 struct mutex pg_vec_lock;
196 unsigned int running:1, /* prot_hook is attached*/
199 int ifindex; /* bound device */
201 struct packet_mclist *mclist;
202 #ifdef CONFIG_PACKET_MMAP
204 enum tpacket_versions tp_version;
205 unsigned int tp_hdrlen;
206 unsigned int tp_reserve;
207 unsigned int tp_loss:1;
211 struct packet_skb_cb {
212 unsigned int origlen;
214 struct sockaddr_pkt pkt;
215 struct sockaddr_ll ll;
219 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
221 #ifdef CONFIG_PACKET_MMAP
223 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
226 struct tpacket_hdr *h1;
227 struct tpacket2_hdr *h2;
232 switch (po->tp_version) {
234 h.h1->tp_status = status;
235 flush_dcache_page(virt_to_page(&h.h1->tp_status));
238 h.h2->tp_status = status;
239 flush_dcache_page(virt_to_page(&h.h2->tp_status));
242 printk(KERN_ERR "TPACKET version not supported\n");
249 static int __packet_get_status(struct packet_sock *po, void *frame)
252 struct tpacket_hdr *h1;
253 struct tpacket2_hdr *h2;
260 switch (po->tp_version) {
262 flush_dcache_page(virt_to_page(&h.h1->tp_status));
263 return h.h1->tp_status;
265 flush_dcache_page(virt_to_page(&h.h2->tp_status));
266 return h.h2->tp_status;
268 printk(KERN_ERR "TPACKET version not supported\n");
274 static void *packet_lookup_frame(struct packet_sock *po,
275 struct packet_ring_buffer *rb,
276 unsigned int position,
279 unsigned int pg_vec_pos, frame_offset;
281 struct tpacket_hdr *h1;
282 struct tpacket2_hdr *h2;
286 pg_vec_pos = position / rb->frames_per_block;
287 frame_offset = position % rb->frames_per_block;
289 h.raw = rb->pg_vec[pg_vec_pos] + (frame_offset * rb->frame_size);
291 if (status != __packet_get_status(po, h.raw))
297 static inline void *packet_current_frame(struct packet_sock *po,
298 struct packet_ring_buffer *rb,
301 return packet_lookup_frame(po, rb, rb->head, status);
304 static inline void *packet_previous_frame(struct packet_sock *po,
305 struct packet_ring_buffer *rb,
308 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
309 return packet_lookup_frame(po, rb, previous, status);
312 static inline void packet_increment_head(struct packet_ring_buffer *buff)
314 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
319 static inline struct packet_sock *pkt_sk(struct sock *sk)
321 return (struct packet_sock *)sk;
324 static void packet_sock_destruct(struct sock *sk)
326 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
327 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
329 if (!sock_flag(sk, SOCK_DEAD)) {
330 printk("Attempt to release alive packet socket: %p\n", sk);
334 sk_refcnt_debug_dec(sk);
338 static const struct proto_ops packet_ops;
340 static const struct proto_ops packet_ops_spkt;
342 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
345 struct sockaddr_pkt *spkt;
348 * When we registered the protocol we saved the socket in the data
349 * field for just this event.
352 sk = pt->af_packet_priv;
355 * Yank back the headers [hope the device set this
356 * right or kerboom...]
358 * Incoming packets have ll header pulled,
361 * For outgoing ones skb->data == skb_mac_header(skb)
362 * so that this procedure is noop.
365 if (skb->pkt_type == PACKET_LOOPBACK)
368 if (dev_net(dev) != sock_net(sk))
371 if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
374 /* drop any routing info */
375 dst_release(skb->dst);
378 /* drop conntrack reference */
381 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
383 skb_push(skb, skb->data - skb_mac_header(skb));
386 * The SOCK_PACKET socket receives _all_ frames.
389 spkt->spkt_family = dev->type;
390 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
391 spkt->spkt_protocol = skb->protocol;
394 * Charge the memory to the socket. This is done specifically
395 * to prevent sockets using all the memory up.
398 if (sock_queue_rcv_skb(sk,skb) == 0)
409 * Output a raw packet to a device layer. This bypasses all the other
410 * protocol layers and you must therefore supply it with a complete frame
413 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
414 struct msghdr *msg, size_t len)
416 struct sock *sk = sock->sk;
417 struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
419 struct net_device *dev;
424 * Get and verify the address.
429 if (msg->msg_namelen < sizeof(struct sockaddr))
431 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
432 proto=saddr->spkt_protocol;
435 return(-ENOTCONN); /* SOCK_PACKET must be sent giving an address */
438 * Find the device first to size check it
441 saddr->spkt_device[13] = 0;
442 dev = dev_get_by_name(sock_net(sk), saddr->spkt_device);
448 if (!(dev->flags & IFF_UP))
452 * You may not queue a frame bigger than the mtu. This is the lowest level
453 * raw protocol and you must do your own fragmentation at this level.
457 if (len > dev->mtu + dev->hard_header_len)
461 skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
464 * If the write buffer is full, then tough. At this level the user gets to
465 * deal with the problem - do your own algorithmic backoffs. That's far
476 /* FIXME: Save some space for broken drivers that write a
477 * hard header at transmission time by themselves. PPP is the
478 * notable one here. This should really be fixed at the driver level.
480 skb_reserve(skb, LL_RESERVED_SPACE(dev));
481 skb_reset_network_header(skb);
483 /* Try to align data part correctly */
484 if (dev->header_ops) {
485 skb->data -= dev->hard_header_len;
486 skb->tail -= dev->hard_header_len;
487 if (len < dev->hard_header_len)
488 skb_reset_network_header(skb);
491 /* Returns -EFAULT on error */
492 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
493 skb->protocol = proto;
495 skb->priority = sk->sk_priority;
515 static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
518 struct sk_filter *filter;
521 filter = rcu_dereference(sk->sk_filter);
523 res = sk_run_filter(skb, filter->insns, filter->len);
524 rcu_read_unlock_bh();
530 This function makes lazy skb cloning in hope that most of packets
531 are discarded by BPF.
533 Note tricky part: we DO mangle shared skb! skb->data, skb->len
534 and skb->cb are mangled. It works because (and until) packets
535 falling here are owned by current CPU. Output packets are cloned
536 by dev_queue_xmit_nit(), input packets are processed by net_bh
537 sequencially, so that if we return skb to original state on exit,
538 we will not harm anyone.
541 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
544 struct sockaddr_ll *sll;
545 struct packet_sock *po;
546 u8 * skb_head = skb->data;
547 int skb_len = skb->len;
548 unsigned int snaplen, res;
550 if (skb->pkt_type == PACKET_LOOPBACK)
553 sk = pt->af_packet_priv;
556 if (dev_net(dev) != sock_net(sk))
561 if (dev->header_ops) {
562 /* The device has an explicit notion of ll header,
563 exported to higher levels.
565 Otherwise, the device hides datails of it frame
566 structure, so that corresponding packet head
567 never delivered to user.
569 if (sk->sk_type != SOCK_DGRAM)
570 skb_push(skb, skb->data - skb_mac_header(skb));
571 else if (skb->pkt_type == PACKET_OUTGOING) {
572 /* Special case: outgoing packets have ll header at head */
573 skb_pull(skb, skb_network_offset(skb));
579 res = run_filter(skb, sk, snaplen);
585 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
586 (unsigned)sk->sk_rcvbuf)
589 if (skb_shared(skb)) {
590 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
594 if (skb_head != skb->data) {
595 skb->data = skb_head;
602 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
605 sll = &PACKET_SKB_CB(skb)->sa.ll;
606 sll->sll_family = AF_PACKET;
607 sll->sll_hatype = dev->type;
608 sll->sll_protocol = skb->protocol;
609 sll->sll_pkttype = skb->pkt_type;
610 if (unlikely(po->origdev))
611 sll->sll_ifindex = orig_dev->ifindex;
613 sll->sll_ifindex = dev->ifindex;
615 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
617 PACKET_SKB_CB(skb)->origlen = skb->len;
619 if (pskb_trim(skb, snaplen))
622 skb_set_owner_r(skb, sk);
624 dst_release(skb->dst);
627 /* drop conntrack reference */
630 spin_lock(&sk->sk_receive_queue.lock);
631 po->stats.tp_packets++;
632 __skb_queue_tail(&sk->sk_receive_queue, skb);
633 spin_unlock(&sk->sk_receive_queue.lock);
634 sk->sk_data_ready(sk, skb->len);
638 spin_lock(&sk->sk_receive_queue.lock);
639 po->stats.tp_drops++;
640 spin_unlock(&sk->sk_receive_queue.lock);
643 if (skb_head != skb->data && skb_shared(skb)) {
644 skb->data = skb_head;
652 #ifdef CONFIG_PACKET_MMAP
653 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
656 struct packet_sock *po;
657 struct sockaddr_ll *sll;
659 struct tpacket_hdr *h1;
660 struct tpacket2_hdr *h2;
663 u8 * skb_head = skb->data;
664 int skb_len = skb->len;
665 unsigned int snaplen, res;
666 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
667 unsigned short macoff, netoff, hdrlen;
668 struct sk_buff *copy_skb = NULL;
672 if (skb->pkt_type == PACKET_LOOPBACK)
675 sk = pt->af_packet_priv;
678 if (dev_net(dev) != sock_net(sk))
681 if (dev->header_ops) {
682 if (sk->sk_type != SOCK_DGRAM)
683 skb_push(skb, skb->data - skb_mac_header(skb));
684 else if (skb->pkt_type == PACKET_OUTGOING) {
685 /* Special case: outgoing packets have ll header at head */
686 skb_pull(skb, skb_network_offset(skb));
690 if (skb->ip_summed == CHECKSUM_PARTIAL)
691 status |= TP_STATUS_CSUMNOTREADY;
695 res = run_filter(skb, sk, snaplen);
701 if (sk->sk_type == SOCK_DGRAM) {
702 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
705 unsigned maclen = skb_network_offset(skb);
706 netoff = TPACKET_ALIGN(po->tp_hdrlen +
707 (maclen < 16 ? 16 : maclen)) +
709 macoff = netoff - maclen;
712 if (macoff + snaplen > po->rx_ring.frame_size) {
713 if (po->copy_thresh &&
714 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
715 (unsigned)sk->sk_rcvbuf) {
716 if (skb_shared(skb)) {
717 copy_skb = skb_clone(skb, GFP_ATOMIC);
719 copy_skb = skb_get(skb);
720 skb_head = skb->data;
723 skb_set_owner_r(copy_skb, sk);
725 snaplen = po->rx_ring.frame_size - macoff;
726 if ((int)snaplen < 0)
730 spin_lock(&sk->sk_receive_queue.lock);
731 h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
734 packet_increment_head(&po->rx_ring);
735 po->stats.tp_packets++;
737 status |= TP_STATUS_COPY;
738 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
740 if (!po->stats.tp_drops)
741 status &= ~TP_STATUS_LOSING;
742 spin_unlock(&sk->sk_receive_queue.lock);
744 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
746 switch (po->tp_version) {
748 h.h1->tp_len = skb->len;
749 h.h1->tp_snaplen = snaplen;
750 h.h1->tp_mac = macoff;
751 h.h1->tp_net = netoff;
752 if (skb->tstamp.tv64)
753 tv = ktime_to_timeval(skb->tstamp);
755 do_gettimeofday(&tv);
756 h.h1->tp_sec = tv.tv_sec;
757 h.h1->tp_usec = tv.tv_usec;
758 hdrlen = sizeof(*h.h1);
761 h.h2->tp_len = skb->len;
762 h.h2->tp_snaplen = snaplen;
763 h.h2->tp_mac = macoff;
764 h.h2->tp_net = netoff;
765 if (skb->tstamp.tv64)
766 ts = ktime_to_timespec(skb->tstamp);
769 h.h2->tp_sec = ts.tv_sec;
770 h.h2->tp_nsec = ts.tv_nsec;
771 h.h2->tp_vlan_tci = skb->vlan_tci;
772 hdrlen = sizeof(*h.h2);
778 sll = h.raw + TPACKET_ALIGN(hdrlen);
779 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
780 sll->sll_family = AF_PACKET;
781 sll->sll_hatype = dev->type;
782 sll->sll_protocol = skb->protocol;
783 sll->sll_pkttype = skb->pkt_type;
784 if (unlikely(po->origdev))
785 sll->sll_ifindex = orig_dev->ifindex;
787 sll->sll_ifindex = dev->ifindex;
789 __packet_set_status(po, h.raw, status);
792 struct page *p_start, *p_end;
793 u8 *h_end = h.raw + macoff + snaplen - 1;
795 p_start = virt_to_page(h.raw);
796 p_end = virt_to_page(h_end);
797 while (p_start <= p_end) {
798 flush_dcache_page(p_start);
803 sk->sk_data_ready(sk, 0);
806 if (skb_head != skb->data && skb_shared(skb)) {
807 skb->data = skb_head;
815 po->stats.tp_drops++;
816 spin_unlock(&sk->sk_receive_queue.lock);
818 sk->sk_data_ready(sk, 0);
823 static void tpacket_destruct_skb(struct sk_buff *skb)
825 struct packet_sock *po = pkt_sk(skb->sk);
830 if (likely(po->tx_ring.pg_vec)) {
831 ph = skb_shinfo(skb)->destructor_arg;
832 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
833 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
834 atomic_dec(&po->tx_ring.pending);
835 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
841 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff * skb,
842 void * frame, struct net_device *dev, int size_max,
843 __be16 proto, unsigned char * addr)
846 struct tpacket_hdr *h1;
847 struct tpacket2_hdr *h2;
850 int to_write, offset, len, tp_len, nr_frags, len_max;
851 struct socket *sock = po->sk.sk_socket;
858 skb->protocol = proto;
860 skb->priority = po->sk.sk_priority;
861 skb_shinfo(skb)->destructor_arg = ph.raw;
863 switch (po->tp_version) {
865 tp_len = ph.h2->tp_len;
868 tp_len = ph.h1->tp_len;
871 if (unlikely(tp_len > size_max)) {
872 printk(KERN_ERR "packet size is too long (%d > %d)\n",
877 skb_reserve(skb, LL_RESERVED_SPACE(dev));
878 skb_reset_network_header(skb);
880 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
883 if (sock->type == SOCK_DGRAM) {
884 err = dev_hard_header(skb, dev, ntohs(proto), addr,
886 if (unlikely(err < 0))
888 } else if (dev->hard_header_len ) {
889 /* net device doesn't like empty head */
890 if (unlikely(tp_len <= dev->hard_header_len)) {
891 printk(KERN_ERR "packet size is too short "
892 "(%d < %d)\n", tp_len,
893 dev->hard_header_len);
897 skb_push(skb, dev->hard_header_len);
898 err = skb_store_bits(skb, 0, data,
899 dev->hard_header_len);
903 data += dev->hard_header_len;
904 to_write -= dev->hard_header_len;
908 page = virt_to_page(data);
909 offset = offset_in_page(data);
910 len_max = PAGE_SIZE - offset;
911 len = ((to_write > len_max) ? len_max : to_write);
913 skb->data_len = to_write;
914 skb->len += to_write;
915 skb->truesize += to_write;
916 atomic_add(to_write, &po->sk.sk_wmem_alloc);
918 while (likely(to_write)) {
919 nr_frags = skb_shinfo(skb)->nr_frags;
921 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
922 printk(KERN_ERR "Packet exceed the number "
923 "of skb frags(%lu)\n",
928 flush_dcache_page(page);
930 skb_fill_page_desc(skb,
932 page++, offset, len);
936 len = ((to_write > len_max) ? len_max : to_write);
942 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
946 struct net_device *dev;
948 int ifindex, err, reserve = 0;
950 struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
951 int tp_len, size_max;
956 sock = po->sk.sk_socket;
958 mutex_lock(&po->pg_vec_lock);
962 ifindex = po->ifindex;
967 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
969 if (msg->msg_namelen < (saddr->sll_halen
970 + offsetof(struct sockaddr_ll,
973 ifindex = saddr->sll_ifindex;
974 proto = saddr->sll_protocol;
975 addr = saddr->sll_addr;
978 dev = dev_get_by_index(sock_net(&po->sk), ifindex);
980 if (unlikely(dev == NULL))
983 reserve = dev->hard_header_len;
986 if (unlikely(!(dev->flags & IFF_UP)))
989 size_max = po->tx_ring.frame_size
990 - sizeof(struct skb_shared_info)
992 - LL_ALLOCATED_SPACE(dev)
993 - sizeof(struct sockaddr_ll);
995 if (size_max > dev->mtu + reserve)
996 size_max = dev->mtu + reserve;
999 ph = packet_current_frame(po, &po->tx_ring,
1000 TP_STATUS_SEND_REQUEST);
1002 if (unlikely(ph == NULL)) {
1007 status = TP_STATUS_SEND_REQUEST;
1008 skb = sock_alloc_send_skb(&po->sk,
1009 LL_ALLOCATED_SPACE(dev)
1010 + sizeof(struct sockaddr_ll),
1013 if (unlikely(skb == NULL))
1016 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1019 if (unlikely(tp_len < 0)) {
1021 __packet_set_status(po, ph,
1022 TP_STATUS_AVAILABLE);
1023 packet_increment_head(&po->tx_ring);
1027 status = TP_STATUS_WRONG_FORMAT;
1033 skb->destructor = tpacket_destruct_skb;
1034 __packet_set_status(po, ph, TP_STATUS_SENDING);
1035 atomic_inc(&po->tx_ring.pending);
1037 status = TP_STATUS_SEND_REQUEST;
1038 err = dev_queue_xmit(skb);
1039 if (unlikely(err > 0 && (err = net_xmit_errno(err)) != 0))
1041 packet_increment_head(&po->tx_ring);
1044 while (likely((ph != NULL) || ((!(msg->msg_flags & MSG_DONTWAIT))
1045 && (atomic_read(&po->tx_ring.pending))))
1052 skb->destructor = sock_wfree;
1053 atomic_dec(&po->tx_ring.pending);
1055 __packet_set_status(po, ph, status);
1060 mutex_unlock(&po->pg_vec_lock);
1065 static int packet_snd(struct socket *sock,
1066 struct msghdr *msg, size_t len)
1068 struct sock *sk = sock->sk;
1069 struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
1070 struct sk_buff *skb;
1071 struct net_device *dev;
1073 unsigned char *addr;
1074 int ifindex, err, reserve = 0;
1077 * Get and verify the address.
1080 if (saddr == NULL) {
1081 struct packet_sock *po = pkt_sk(sk);
1083 ifindex = po->ifindex;
1088 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1090 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1092 ifindex = saddr->sll_ifindex;
1093 proto = saddr->sll_protocol;
1094 addr = saddr->sll_addr;
1098 dev = dev_get_by_index(sock_net(sk), ifindex);
1102 if (sock->type == SOCK_RAW)
1103 reserve = dev->hard_header_len;
1106 if (!(dev->flags & IFF_UP))
1110 if (len > dev->mtu+reserve)
1113 skb = sock_alloc_send_skb(sk, len + LL_ALLOCATED_SPACE(dev),
1114 msg->msg_flags & MSG_DONTWAIT, &err);
1118 skb_reserve(skb, LL_RESERVED_SPACE(dev));
1119 skb_reset_network_header(skb);
1122 if (sock->type == SOCK_DGRAM &&
1123 dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len) < 0)
1126 /* Returns -EFAULT on error */
1127 err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
1131 skb->protocol = proto;
1133 skb->priority = sk->sk_priority;
1139 err = dev_queue_xmit(skb);
1140 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1156 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1157 struct msghdr *msg, size_t len)
1159 #ifdef CONFIG_PACKET_MMAP
1160 struct sock *sk = sock->sk;
1161 struct packet_sock *po = pkt_sk(sk);
1162 if (po->tx_ring.pg_vec)
1163 return tpacket_snd(po, msg);
1166 return packet_snd(sock, msg, len);
1170 * Close a PACKET socket. This is fairly simple. We immediately go
1171 * to 'closed' state and remove our protocol entry in the device list.
1174 static int packet_release(struct socket *sock)
1176 struct sock *sk = sock->sk;
1177 struct packet_sock *po;
1179 #ifdef CONFIG_PACKET_MMAP
1180 struct tpacket_req req;
1189 write_lock_bh(&net->packet.sklist_lock);
1190 sk_del_node_init(sk);
1191 sock_prot_inuse_add(net, sk->sk_prot, -1);
1192 write_unlock_bh(&net->packet.sklist_lock);
1195 * Unhook packet receive handler.
1200 * Remove the protocol hook
1202 dev_remove_pack(&po->prot_hook);
1208 packet_flush_mclist(sk);
1210 #ifdef CONFIG_PACKET_MMAP
1211 memset(&req, 0, sizeof(req));
1213 if (po->rx_ring.pg_vec)
1214 packet_set_ring(sk, &req, 1, 0);
1216 if (po->tx_ring.pg_vec)
1217 packet_set_ring(sk, &req, 1, 1);
1221 * Now the socket is dead. No more input will appear.
1229 skb_queue_purge(&sk->sk_receive_queue);
1230 sk_refcnt_debug_release(sk);
1237 * Attach a packet hook.
1240 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1242 struct packet_sock *po = pkt_sk(sk);
1244 * Detach an existing hook if present.
1249 spin_lock(&po->bind_lock);
1254 spin_unlock(&po->bind_lock);
1255 dev_remove_pack(&po->prot_hook);
1256 spin_lock(&po->bind_lock);
1260 po->prot_hook.type = protocol;
1261 po->prot_hook.dev = dev;
1263 po->ifindex = dev ? dev->ifindex : 0;
1268 if (!dev || (dev->flags & IFF_UP)) {
1269 dev_add_pack(&po->prot_hook);
1273 sk->sk_err = ENETDOWN;
1274 if (!sock_flag(sk, SOCK_DEAD))
1275 sk->sk_error_report(sk);
1279 spin_unlock(&po->bind_lock);
1285 * Bind a packet socket to a device
1288 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1290 struct sock *sk=sock->sk;
1292 struct net_device *dev;
1299 if (addr_len != sizeof(struct sockaddr))
1301 strlcpy(name,uaddr->sa_data,sizeof(name));
1303 dev = dev_get_by_name(sock_net(sk), name);
1305 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1311 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1313 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1314 struct sock *sk=sock->sk;
1315 struct net_device *dev = NULL;
1323 if (addr_len < sizeof(struct sockaddr_ll))
1325 if (sll->sll_family != AF_PACKET)
1328 if (sll->sll_ifindex) {
1330 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1334 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1342 static struct proto packet_proto = {
1344 .owner = THIS_MODULE,
1345 .obj_size = sizeof(struct packet_sock),
1349 * Create a packet of type SOCK_PACKET.
1352 static int packet_create(struct net *net, struct socket *sock, int protocol)
1355 struct packet_sock *po;
1356 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1359 if (!capable(CAP_NET_RAW))
1361 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1362 sock->type != SOCK_PACKET)
1363 return -ESOCKTNOSUPPORT;
1365 sock->state = SS_UNCONNECTED;
1368 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1372 sock->ops = &packet_ops;
1373 if (sock->type == SOCK_PACKET)
1374 sock->ops = &packet_ops_spkt;
1376 sock_init_data(sock, sk);
1379 sk->sk_family = PF_PACKET;
1382 sk->sk_destruct = packet_sock_destruct;
1383 sk_refcnt_debug_inc(sk);
1386 * Attach a protocol block
1389 spin_lock_init(&po->bind_lock);
1390 mutex_init(&po->pg_vec_lock);
1391 po->prot_hook.func = packet_rcv;
1393 if (sock->type == SOCK_PACKET)
1394 po->prot_hook.func = packet_rcv_spkt;
1396 po->prot_hook.af_packet_priv = sk;
1399 po->prot_hook.type = proto;
1400 dev_add_pack(&po->prot_hook);
1405 write_lock_bh(&net->packet.sklist_lock);
1406 sk_add_node(sk, &net->packet.sklist);
1407 sock_prot_inuse_add(net, &packet_proto, 1);
1408 write_unlock_bh(&net->packet.sklist_lock);
1415 * Pull a packet from our receive queue and hand it to the user.
1416 * If necessary we block.
1419 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1420 struct msghdr *msg, size_t len, int flags)
1422 struct sock *sk = sock->sk;
1423 struct sk_buff *skb;
1425 struct sockaddr_ll *sll;
1428 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1432 /* What error should we return now? EUNATTACH? */
1433 if (pkt_sk(sk)->ifindex < 0)
1438 * Call the generic datagram receiver. This handles all sorts
1439 * of horrible races and re-entrancy so we can forget about it
1440 * in the protocol layers.
1442 * Now it will return ENETDOWN, if device have just gone down,
1443 * but then it will block.
1446 skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1449 * An error occurred so return it. Because skb_recv_datagram()
1450 * handles the blocking we don't see and worry about blocking
1458 * If the address length field is there to be filled in, we fill
1462 sll = &PACKET_SKB_CB(skb)->sa.ll;
1463 if (sock->type == SOCK_PACKET)
1464 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1466 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1469 * You lose any data beyond the buffer you gave. If it worries a
1470 * user program they can ask the device for its MTU anyway.
1477 msg->msg_flags|=MSG_TRUNC;
1480 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1484 sock_recv_timestamp(msg, sk, skb);
1487 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1490 if (pkt_sk(sk)->auxdata) {
1491 struct tpacket_auxdata aux;
1493 aux.tp_status = TP_STATUS_USER;
1494 if (skb->ip_summed == CHECKSUM_PARTIAL)
1495 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1496 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1497 aux.tp_snaplen = skb->len;
1499 aux.tp_net = skb_network_offset(skb);
1500 aux.tp_vlan_tci = skb->vlan_tci;
1502 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1506 * Free or return the buffer as appropriate. Again this
1507 * hides all the races and re-entrancy issues from us.
1509 err = (flags&MSG_TRUNC) ? skb->len : copied;
1512 skb_free_datagram(sk, skb);
1517 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1518 int *uaddr_len, int peer)
1520 struct net_device *dev;
1521 struct sock *sk = sock->sk;
1526 uaddr->sa_family = AF_PACKET;
1527 dev = dev_get_by_index(sock_net(sk), pkt_sk(sk)->ifindex);
1529 strlcpy(uaddr->sa_data, dev->name, 15);
1532 memset(uaddr->sa_data, 0, 14);
1533 *uaddr_len = sizeof(*uaddr);
1538 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1539 int *uaddr_len, int peer)
1541 struct net_device *dev;
1542 struct sock *sk = sock->sk;
1543 struct packet_sock *po = pkt_sk(sk);
1544 struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1549 sll->sll_family = AF_PACKET;
1550 sll->sll_ifindex = po->ifindex;
1551 sll->sll_protocol = po->num;
1552 dev = dev_get_by_index(sock_net(sk), po->ifindex);
1554 sll->sll_hatype = dev->type;
1555 sll->sll_halen = dev->addr_len;
1556 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1559 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1562 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1567 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1571 case PACKET_MR_MULTICAST:
1573 dev_mc_add(dev, i->addr, i->alen, 0);
1575 dev_mc_delete(dev, i->addr, i->alen, 0);
1577 case PACKET_MR_PROMISC:
1578 return dev_set_promiscuity(dev, what);
1580 case PACKET_MR_ALLMULTI:
1581 return dev_set_allmulti(dev, what);
1588 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1590 for ( ; i; i=i->next) {
1591 if (i->ifindex == dev->ifindex)
1592 packet_dev_mc(dev, i, what);
1596 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1598 struct packet_sock *po = pkt_sk(sk);
1599 struct packet_mclist *ml, *i;
1600 struct net_device *dev;
1606 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1611 if (mreq->mr_alen > dev->addr_len)
1615 i = kmalloc(sizeof(*i), GFP_KERNEL);
1620 for (ml = po->mclist; ml; ml = ml->next) {
1621 if (ml->ifindex == mreq->mr_ifindex &&
1622 ml->type == mreq->mr_type &&
1623 ml->alen == mreq->mr_alen &&
1624 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1626 /* Free the new element ... */
1632 i->type = mreq->mr_type;
1633 i->ifindex = mreq->mr_ifindex;
1634 i->alen = mreq->mr_alen;
1635 memcpy(i->addr, mreq->mr_address, i->alen);
1637 i->next = po->mclist;
1639 err = packet_dev_mc(dev, i, 1);
1641 po->mclist = i->next;
1650 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1652 struct packet_mclist *ml, **mlp;
1656 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1657 if (ml->ifindex == mreq->mr_ifindex &&
1658 ml->type == mreq->mr_type &&
1659 ml->alen == mreq->mr_alen &&
1660 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1661 if (--ml->count == 0) {
1662 struct net_device *dev;
1664 dev = dev_get_by_index(sock_net(sk), ml->ifindex);
1666 packet_dev_mc(dev, ml, -1);
1676 return -EADDRNOTAVAIL;
1679 static void packet_flush_mclist(struct sock *sk)
1681 struct packet_sock *po = pkt_sk(sk);
1682 struct packet_mclist *ml;
1688 while ((ml = po->mclist) != NULL) {
1689 struct net_device *dev;
1691 po->mclist = ml->next;
1692 if ((dev = dev_get_by_index(sock_net(sk), ml->ifindex)) != NULL) {
1693 packet_dev_mc(dev, ml, -1);
1702 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1704 struct sock *sk = sock->sk;
1705 struct packet_sock *po = pkt_sk(sk);
1708 if (level != SOL_PACKET)
1709 return -ENOPROTOOPT;
1712 case PACKET_ADD_MEMBERSHIP:
1713 case PACKET_DROP_MEMBERSHIP:
1715 struct packet_mreq_max mreq;
1717 memset(&mreq, 0, sizeof(mreq));
1718 if (len < sizeof(struct packet_mreq))
1720 if (len > sizeof(mreq))
1722 if (copy_from_user(&mreq,optval,len))
1724 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1726 if (optname == PACKET_ADD_MEMBERSHIP)
1727 ret = packet_mc_add(sk, &mreq);
1729 ret = packet_mc_drop(sk, &mreq);
1733 #ifdef CONFIG_PACKET_MMAP
1734 case PACKET_RX_RING:
1735 case PACKET_TX_RING:
1737 struct tpacket_req req;
1739 if (optlen<sizeof(req))
1741 if (copy_from_user(&req,optval,sizeof(req)))
1743 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1745 case PACKET_COPY_THRESH:
1749 if (optlen!=sizeof(val))
1751 if (copy_from_user(&val,optval,sizeof(val)))
1754 pkt_sk(sk)->copy_thresh = val;
1757 case PACKET_VERSION:
1761 if (optlen != sizeof(val))
1763 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1765 if (copy_from_user(&val, optval, sizeof(val)))
1770 po->tp_version = val;
1776 case PACKET_RESERVE:
1780 if (optlen != sizeof(val))
1782 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1784 if (copy_from_user(&val, optval, sizeof(val)))
1786 po->tp_reserve = val;
1793 if (optlen != sizeof(val))
1795 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1797 if (copy_from_user(&val, optval, sizeof(val)))
1799 po->tp_loss = !!val;
1803 case PACKET_AUXDATA:
1807 if (optlen < sizeof(val))
1809 if (copy_from_user(&val, optval, sizeof(val)))
1812 po->auxdata = !!val;
1815 case PACKET_ORIGDEV:
1819 if (optlen < sizeof(val))
1821 if (copy_from_user(&val, optval, sizeof(val)))
1824 po->origdev = !!val;
1828 return -ENOPROTOOPT;
1832 static int packet_getsockopt(struct socket *sock, int level, int optname,
1833 char __user *optval, int __user *optlen)
1837 struct sock *sk = sock->sk;
1838 struct packet_sock *po = pkt_sk(sk);
1840 struct tpacket_stats st;
1842 if (level != SOL_PACKET)
1843 return -ENOPROTOOPT;
1845 if (get_user(len, optlen))
1852 case PACKET_STATISTICS:
1853 if (len > sizeof(struct tpacket_stats))
1854 len = sizeof(struct tpacket_stats);
1855 spin_lock_bh(&sk->sk_receive_queue.lock);
1857 memset(&po->stats, 0, sizeof(st));
1858 spin_unlock_bh(&sk->sk_receive_queue.lock);
1859 st.tp_packets += st.tp_drops;
1863 case PACKET_AUXDATA:
1864 if (len > sizeof(int))
1870 case PACKET_ORIGDEV:
1871 if (len > sizeof(int))
1877 #ifdef CONFIG_PACKET_MMAP
1878 case PACKET_VERSION:
1879 if (len > sizeof(int))
1881 val = po->tp_version;
1885 if (len > sizeof(int))
1887 if (copy_from_user(&val, optval, len))
1891 val = sizeof(struct tpacket_hdr);
1894 val = sizeof(struct tpacket2_hdr);
1901 case PACKET_RESERVE:
1902 if (len > sizeof(unsigned int))
1903 len = sizeof(unsigned int);
1904 val = po->tp_reserve;
1908 if (len > sizeof(unsigned int))
1909 len = sizeof(unsigned int);
1915 return -ENOPROTOOPT;
1918 if (put_user(len, optlen))
1920 if (copy_to_user(optval, data, len))
1926 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1929 struct hlist_node *node;
1930 struct net_device *dev = data;
1931 struct net *net = dev_net(dev);
1933 read_lock(&net->packet.sklist_lock);
1934 sk_for_each(sk, node, &net->packet.sklist) {
1935 struct packet_sock *po = pkt_sk(sk);
1938 case NETDEV_UNREGISTER:
1940 packet_dev_mclist(dev, po->mclist, -1);
1944 if (dev->ifindex == po->ifindex) {
1945 spin_lock(&po->bind_lock);
1947 __dev_remove_pack(&po->prot_hook);
1950 sk->sk_err = ENETDOWN;
1951 if (!sock_flag(sk, SOCK_DEAD))
1952 sk->sk_error_report(sk);
1954 if (msg == NETDEV_UNREGISTER) {
1956 po->prot_hook.dev = NULL;
1958 spin_unlock(&po->bind_lock);
1962 spin_lock(&po->bind_lock);
1963 if (dev->ifindex == po->ifindex && po->num &&
1965 dev_add_pack(&po->prot_hook);
1969 spin_unlock(&po->bind_lock);
1973 read_unlock(&net->packet.sklist_lock);
1978 static int packet_ioctl(struct socket *sock, unsigned int cmd,
1981 struct sock *sk = sock->sk;
1986 int amount = atomic_read(&sk->sk_wmem_alloc);
1987 return put_user(amount, (int __user *)arg);
1991 struct sk_buff *skb;
1994 spin_lock_bh(&sk->sk_receive_queue.lock);
1995 skb = skb_peek(&sk->sk_receive_queue);
1998 spin_unlock_bh(&sk->sk_receive_queue.lock);
1999 return put_user(amount, (int __user *)arg);
2002 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2004 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2014 case SIOCGIFBRDADDR:
2015 case SIOCSIFBRDADDR:
2016 case SIOCGIFNETMASK:
2017 case SIOCSIFNETMASK:
2018 case SIOCGIFDSTADDR:
2019 case SIOCSIFDSTADDR:
2021 if (!net_eq(sock_net(sk), &init_net))
2022 return -ENOIOCTLCMD;
2023 return inet_dgram_ops.ioctl(sock, cmd, arg);
2027 return -ENOIOCTLCMD;
2032 #ifndef CONFIG_PACKET_MMAP
2033 #define packet_mmap sock_no_mmap
2034 #define packet_poll datagram_poll
2037 static unsigned int packet_poll(struct file * file, struct socket *sock,
2040 struct sock *sk = sock->sk;
2041 struct packet_sock *po = pkt_sk(sk);
2042 unsigned int mask = datagram_poll(file, sock, wait);
2044 spin_lock_bh(&sk->sk_receive_queue.lock);
2045 if (po->rx_ring.pg_vec) {
2046 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2047 mask |= POLLIN | POLLRDNORM;
2049 spin_unlock_bh(&sk->sk_receive_queue.lock);
2050 spin_lock_bh(&sk->sk_write_queue.lock);
2051 if (po->tx_ring.pg_vec) {
2052 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2053 mask |= POLLOUT | POLLWRNORM;
2055 spin_unlock_bh(&sk->sk_write_queue.lock);
2060 /* Dirty? Well, I still did not learn better way to account
2064 static void packet_mm_open(struct vm_area_struct *vma)
2066 struct file *file = vma->vm_file;
2067 struct socket * sock = file->private_data;
2068 struct sock *sk = sock->sk;
2071 atomic_inc(&pkt_sk(sk)->mapped);
2074 static void packet_mm_close(struct vm_area_struct *vma)
2076 struct file *file = vma->vm_file;
2077 struct socket * sock = file->private_data;
2078 struct sock *sk = sock->sk;
2081 atomic_dec(&pkt_sk(sk)->mapped);
2084 static struct vm_operations_struct packet_mmap_ops = {
2085 .open = packet_mm_open,
2086 .close =packet_mm_close,
2089 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
2093 for (i = 0; i < len; i++) {
2094 if (likely(pg_vec[i]))
2095 free_pages((unsigned long) pg_vec[i], order);
2100 static inline char *alloc_one_pg_vec_page(unsigned long order)
2102 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO | __GFP_NOWARN;
2104 return (char *) __get_free_pages(gfp_flags, order);
2107 static char **alloc_pg_vec(struct tpacket_req *req, int order)
2109 unsigned int block_nr = req->tp_block_nr;
2113 pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
2114 if (unlikely(!pg_vec))
2117 for (i = 0; i < block_nr; i++) {
2118 pg_vec[i] = alloc_one_pg_vec_page(order);
2119 if (unlikely(!pg_vec[i]))
2120 goto out_free_pgvec;
2127 free_pg_vec(pg_vec, order, block_nr);
2132 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2133 int closing, int tx_ring)
2135 char **pg_vec = NULL;
2136 struct packet_sock *po = pkt_sk(sk);
2137 int was_running, order = 0;
2138 struct packet_ring_buffer *rb;
2139 struct sk_buff_head *rb_queue;
2143 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2144 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2148 if (atomic_read(&po->mapped))
2150 if (atomic_read(&rb->pending))
2154 if (req->tp_block_nr) {
2155 /* Sanity tests and some calculations */
2157 if (unlikely(rb->pg_vec))
2160 switch (po->tp_version) {
2162 po->tp_hdrlen = TPACKET_HDRLEN;
2165 po->tp_hdrlen = TPACKET2_HDRLEN;
2170 if (unlikely((int)req->tp_block_size <= 0))
2172 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2174 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2177 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2180 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2181 if (unlikely(rb->frames_per_block <= 0))
2183 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2188 order = get_order(req->tp_block_size);
2189 pg_vec = alloc_pg_vec(req, order);
2190 if (unlikely(!pg_vec))
2196 if (unlikely(req->tp_frame_nr))
2202 /* Detach socket from network */
2203 spin_lock(&po->bind_lock);
2204 was_running = po->running;
2207 __dev_remove_pack(&po->prot_hook);
2212 spin_unlock(&po->bind_lock);
2217 mutex_lock(&po->pg_vec_lock);
2218 if (closing || atomic_read(&po->mapped) == 0) {
2220 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
2221 spin_lock_bh(&rb_queue->lock);
2222 pg_vec = XC(rb->pg_vec, pg_vec);
2223 rb->frame_max = (req->tp_frame_nr - 1);
2225 rb->frame_size = req->tp_frame_size;
2226 spin_unlock_bh(&rb_queue->lock);
2228 order = XC(rb->pg_vec_order, order);
2229 req->tp_block_nr = XC(rb->pg_vec_len, req->tp_block_nr);
2231 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2232 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2233 tpacket_rcv : packet_rcv;
2234 skb_queue_purge(rb_queue);
2236 if (atomic_read(&po->mapped))
2237 printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n",
2238 atomic_read(&po->mapped));
2240 mutex_unlock(&po->pg_vec_lock);
2242 spin_lock(&po->bind_lock);
2243 if (was_running && !po->running) {
2247 dev_add_pack(&po->prot_hook);
2249 spin_unlock(&po->bind_lock);
2254 free_pg_vec(pg_vec, order, req->tp_block_nr);
2259 static int packet_mmap(struct file *file, struct socket *sock,
2260 struct vm_area_struct *vma)
2262 struct sock *sk = sock->sk;
2263 struct packet_sock *po = pkt_sk(sk);
2264 unsigned long size, expected_size;
2265 struct packet_ring_buffer *rb;
2266 unsigned long start;
2273 mutex_lock(&po->pg_vec_lock);
2276 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2278 expected_size += rb->pg_vec_len
2284 if (expected_size == 0)
2287 size = vma->vm_end - vma->vm_start;
2288 if (size != expected_size)
2291 start = vma->vm_start;
2292 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2293 if (rb->pg_vec == NULL)
2296 for (i = 0; i < rb->pg_vec_len; i++) {
2297 struct page *page = virt_to_page(rb->pg_vec[i]);
2300 for (pg_num = 0; pg_num < rb->pg_vec_pages;
2302 err = vm_insert_page(vma, start, page);
2310 atomic_inc(&po->mapped);
2311 vma->vm_ops = &packet_mmap_ops;
2315 mutex_unlock(&po->pg_vec_lock);
2321 static const struct proto_ops packet_ops_spkt = {
2322 .family = PF_PACKET,
2323 .owner = THIS_MODULE,
2324 .release = packet_release,
2325 .bind = packet_bind_spkt,
2326 .connect = sock_no_connect,
2327 .socketpair = sock_no_socketpair,
2328 .accept = sock_no_accept,
2329 .getname = packet_getname_spkt,
2330 .poll = datagram_poll,
2331 .ioctl = packet_ioctl,
2332 .listen = sock_no_listen,
2333 .shutdown = sock_no_shutdown,
2334 .setsockopt = sock_no_setsockopt,
2335 .getsockopt = sock_no_getsockopt,
2336 .sendmsg = packet_sendmsg_spkt,
2337 .recvmsg = packet_recvmsg,
2338 .mmap = sock_no_mmap,
2339 .sendpage = sock_no_sendpage,
2342 static const struct proto_ops packet_ops = {
2343 .family = PF_PACKET,
2344 .owner = THIS_MODULE,
2345 .release = packet_release,
2346 .bind = packet_bind,
2347 .connect = sock_no_connect,
2348 .socketpair = sock_no_socketpair,
2349 .accept = sock_no_accept,
2350 .getname = packet_getname,
2351 .poll = packet_poll,
2352 .ioctl = packet_ioctl,
2353 .listen = sock_no_listen,
2354 .shutdown = sock_no_shutdown,
2355 .setsockopt = packet_setsockopt,
2356 .getsockopt = packet_getsockopt,
2357 .sendmsg = packet_sendmsg,
2358 .recvmsg = packet_recvmsg,
2359 .mmap = packet_mmap,
2360 .sendpage = sock_no_sendpage,
2363 static struct net_proto_family packet_family_ops = {
2364 .family = PF_PACKET,
2365 .create = packet_create,
2366 .owner = THIS_MODULE,
2369 static struct notifier_block packet_netdev_notifier = {
2370 .notifier_call =packet_notifier,
2373 #ifdef CONFIG_PROC_FS
2374 static inline struct sock *packet_seq_idx(struct net *net, loff_t off)
2377 struct hlist_node *node;
2379 sk_for_each(s, node, &net->packet.sklist) {
2386 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2387 __acquires(seq_file_net(seq)->packet.sklist_lock)
2389 struct net *net = seq_file_net(seq);
2390 read_lock(&net->packet.sklist_lock);
2391 return *pos ? packet_seq_idx(net, *pos - 1) : SEQ_START_TOKEN;
2394 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2396 struct net *net = seq_file_net(seq);
2398 return (v == SEQ_START_TOKEN)
2399 ? sk_head(&net->packet.sklist)
2400 : sk_next((struct sock*)v) ;
2403 static void packet_seq_stop(struct seq_file *seq, void *v)
2404 __releases(seq_file_net(seq)->packet.sklist_lock)
2406 struct net *net = seq_file_net(seq);
2407 read_unlock(&net->packet.sklist_lock);
2410 static int packet_seq_show(struct seq_file *seq, void *v)
2412 if (v == SEQ_START_TOKEN)
2413 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2416 const struct packet_sock *po = pkt_sk(s);
2419 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2421 atomic_read(&s->sk_refcnt),
2426 atomic_read(&s->sk_rmem_alloc),
2434 static const struct seq_operations packet_seq_ops = {
2435 .start = packet_seq_start,
2436 .next = packet_seq_next,
2437 .stop = packet_seq_stop,
2438 .show = packet_seq_show,
2441 static int packet_seq_open(struct inode *inode, struct file *file)
2443 return seq_open_net(inode, file, &packet_seq_ops,
2444 sizeof(struct seq_net_private));
2447 static const struct file_operations packet_seq_fops = {
2448 .owner = THIS_MODULE,
2449 .open = packet_seq_open,
2451 .llseek = seq_lseek,
2452 .release = seq_release_net,
2457 static int packet_net_init(struct net *net)
2459 rwlock_init(&net->packet.sklist_lock);
2460 INIT_HLIST_HEAD(&net->packet.sklist);
2462 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2468 static void packet_net_exit(struct net *net)
2470 proc_net_remove(net, "packet");
2473 static struct pernet_operations packet_net_ops = {
2474 .init = packet_net_init,
2475 .exit = packet_net_exit,
2479 static void __exit packet_exit(void)
2481 unregister_netdevice_notifier(&packet_netdev_notifier);
2482 unregister_pernet_subsys(&packet_net_ops);
2483 sock_unregister(PF_PACKET);
2484 proto_unregister(&packet_proto);
2487 static int __init packet_init(void)
2489 int rc = proto_register(&packet_proto, 0);
2494 sock_register(&packet_family_ops);
2495 register_pernet_subsys(&packet_net_ops);
2496 register_netdevice_notifier(&packet_netdev_notifier);
2501 module_init(packet_init);
2502 module_exit(packet_exit);
2503 MODULE_LICENSE("GPL");
2504 MODULE_ALIAS_NETPROTO(PF_PACKET);