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. */
140 struct packet_mclist {
141 struct packet_mclist *next;
146 unsigned char addr[MAX_ADDR_LEN];
148 /* identical to struct packet_mreq except it has
149 * a longer address field.
151 struct packet_mreq_max {
153 unsigned short mr_type;
154 unsigned short mr_alen;
155 unsigned char mr_address[MAX_ADDR_LEN];
158 #ifdef CONFIG_PACKET_MMAP
159 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
160 int closing, int tx_ring);
162 struct packet_ring_buffer {
165 unsigned int frames_per_block;
166 unsigned int frame_size;
167 unsigned int frame_max;
169 unsigned int pg_vec_order;
170 unsigned int pg_vec_pages;
171 unsigned int pg_vec_len;
177 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
180 static void packet_flush_mclist(struct sock *sk);
183 /* struct sock has to be the first member of packet_sock */
185 struct tpacket_stats stats;
186 #ifdef CONFIG_PACKET_MMAP
187 struct packet_ring_buffer rx_ring;
188 struct packet_ring_buffer tx_ring;
191 spinlock_t bind_lock;
192 struct mutex pg_vec_lock;
193 unsigned int running:1, /* prot_hook is attached*/
196 int ifindex; /* bound device */
198 struct packet_mclist *mclist;
199 #ifdef CONFIG_PACKET_MMAP
201 enum tpacket_versions tp_version;
202 unsigned int tp_hdrlen;
203 unsigned int tp_reserve;
204 unsigned int tp_loss:1;
206 struct packet_type prot_hook ____cacheline_aligned_in_smp;
209 struct packet_skb_cb {
210 unsigned int origlen;
212 struct sockaddr_pkt pkt;
213 struct sockaddr_ll ll;
217 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
219 #ifdef CONFIG_PACKET_MMAP
221 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
224 struct tpacket_hdr *h1;
225 struct tpacket2_hdr *h2;
230 switch (po->tp_version) {
232 h.h1->tp_status = status;
233 flush_dcache_page(virt_to_page(&h.h1->tp_status));
236 h.h2->tp_status = status;
237 flush_dcache_page(virt_to_page(&h.h2->tp_status));
240 pr_err("TPACKET version not supported\n");
247 static int __packet_get_status(struct packet_sock *po, void *frame)
250 struct tpacket_hdr *h1;
251 struct tpacket2_hdr *h2;
258 switch (po->tp_version) {
260 flush_dcache_page(virt_to_page(&h.h1->tp_status));
261 return h.h1->tp_status;
263 flush_dcache_page(virt_to_page(&h.h2->tp_status));
264 return h.h2->tp_status;
266 pr_err("TPACKET version not supported\n");
272 static void *packet_lookup_frame(struct packet_sock *po,
273 struct packet_ring_buffer *rb,
274 unsigned int position,
277 unsigned int pg_vec_pos, frame_offset;
279 struct tpacket_hdr *h1;
280 struct tpacket2_hdr *h2;
284 pg_vec_pos = position / rb->frames_per_block;
285 frame_offset = position % rb->frames_per_block;
287 h.raw = rb->pg_vec[pg_vec_pos] + (frame_offset * rb->frame_size);
289 if (status != __packet_get_status(po, h.raw))
295 static inline void *packet_current_frame(struct packet_sock *po,
296 struct packet_ring_buffer *rb,
299 return packet_lookup_frame(po, rb, rb->head, status);
302 static inline void *packet_previous_frame(struct packet_sock *po,
303 struct packet_ring_buffer *rb,
306 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
307 return packet_lookup_frame(po, rb, previous, status);
310 static inline void packet_increment_head(struct packet_ring_buffer *buff)
312 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
317 static inline struct packet_sock *pkt_sk(struct sock *sk)
319 return (struct packet_sock *)sk;
322 static void packet_sock_destruct(struct sock *sk)
324 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
325 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
327 if (!sock_flag(sk, SOCK_DEAD)) {
328 pr_err("Attempt to release alive packet socket: %p\n", sk);
332 sk_refcnt_debug_dec(sk);
336 static const struct proto_ops packet_ops;
338 static const struct proto_ops packet_ops_spkt;
340 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
341 struct packet_type *pt, struct net_device *orig_dev)
344 struct sockaddr_pkt *spkt;
347 * When we registered the protocol we saved the socket in the data
348 * field for just this event.
351 sk = pt->af_packet_priv;
354 * Yank back the headers [hope the device set this
355 * right or kerboom...]
357 * Incoming packets have ll header pulled,
360 * For outgoing ones skb->data == skb_mac_header(skb)
361 * so that this procedure is noop.
364 if (skb->pkt_type == PACKET_LOOPBACK)
367 if (dev_net(dev) != sock_net(sk))
370 skb = skb_share_check(skb, GFP_ATOMIC);
374 /* drop any routing info */
377 /* drop conntrack reference */
380 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
382 skb_push(skb, skb->data - skb_mac_header(skb));
385 * The SOCK_PACKET socket receives _all_ frames.
388 spkt->spkt_family = dev->type;
389 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
390 spkt->spkt_protocol = skb->protocol;
393 * Charge the memory to the socket. This is done specifically
394 * to prevent sockets using all the memory up.
397 if (sock_queue_rcv_skb(sk, skb) == 0)
408 * Output a raw packet to a device layer. This bypasses all the other
409 * protocol layers and you must therefore supply it with a complete frame
412 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
413 struct msghdr *msg, size_t len)
415 struct sock *sk = sock->sk;
416 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
418 struct net_device *dev;
423 * Get and verify the address.
427 if (msg->msg_namelen < sizeof(struct sockaddr))
429 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
430 proto = saddr->spkt_protocol;
432 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
435 * Find the device first to size check it
438 saddr->spkt_device[13] = 0;
439 dev = dev_get_by_name(sock_net(sk), saddr->spkt_device);
445 if (!(dev->flags & IFF_UP))
449 * You may not queue a frame bigger than the mtu. This is the lowest level
450 * raw protocol and you must do your own fragmentation at this level.
454 if (len > dev->mtu + dev->hard_header_len)
458 skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
461 * If the write buffer is full, then tough. At this level the user
462 * gets to deal with the problem - do your own algorithmic backoffs.
463 * That's far more flexible.
473 /* FIXME: Save some space for broken drivers that write a
474 * hard header at transmission time by themselves. PPP is the
475 * notable one here. This should really be fixed at the driver level.
477 skb_reserve(skb, LL_RESERVED_SPACE(dev));
478 skb_reset_network_header(skb);
480 /* Try to align data part correctly */
481 if (dev->header_ops) {
482 skb->data -= dev->hard_header_len;
483 skb->tail -= dev->hard_header_len;
484 if (len < dev->hard_header_len)
485 skb_reset_network_header(skb);
488 /* Returns -EFAULT on error */
489 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
490 skb->protocol = proto;
492 skb->priority = sk->sk_priority;
493 skb->mark = sk->sk_mark;
513 static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
516 struct sk_filter *filter;
519 filter = rcu_dereference(sk->sk_filter);
521 res = sk_run_filter(skb, filter->insns, filter->len);
522 rcu_read_unlock_bh();
528 This function makes lazy skb cloning in hope that most of packets
529 are discarded by BPF.
531 Note tricky part: we DO mangle shared skb! skb->data, skb->len
532 and skb->cb are mangled. It works because (and until) packets
533 falling here are owned by current CPU. Output packets are cloned
534 by dev_queue_xmit_nit(), input packets are processed by net_bh
535 sequencially, so that if we return skb to original state on exit,
536 we will not harm anyone.
539 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
540 struct packet_type *pt, struct net_device *orig_dev)
543 struct sockaddr_ll *sll;
544 struct packet_sock *po;
545 u8 *skb_head = skb->data;
546 int skb_len = skb->len;
547 unsigned int snaplen, res;
549 if (skb->pkt_type == PACKET_LOOPBACK)
552 sk = pt->af_packet_priv;
555 if (dev_net(dev) != sock_net(sk))
560 if (dev->header_ops) {
561 /* The device has an explicit notion of ll header,
562 exported to higher levels.
564 Otherwise, the device hides datails of it frame
565 structure, so that corresponding packet head
566 never delivered to user.
568 if (sk->sk_type != SOCK_DGRAM)
569 skb_push(skb, skb->data - skb_mac_header(skb));
570 else if (skb->pkt_type == PACKET_OUTGOING) {
571 /* Special case: outgoing packets have ll header at head */
572 skb_pull(skb, skb_network_offset(skb));
578 res = run_filter(skb, sk, snaplen);
584 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
585 (unsigned)sk->sk_rcvbuf)
588 if (skb_shared(skb)) {
589 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
593 if (skb_head != skb->data) {
594 skb->data = skb_head;
601 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
604 sll = &PACKET_SKB_CB(skb)->sa.ll;
605 sll->sll_family = AF_PACKET;
606 sll->sll_hatype = dev->type;
607 sll->sll_protocol = skb->protocol;
608 sll->sll_pkttype = skb->pkt_type;
609 if (unlikely(po->origdev))
610 sll->sll_ifindex = orig_dev->ifindex;
612 sll->sll_ifindex = dev->ifindex;
614 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
616 PACKET_SKB_CB(skb)->origlen = skb->len;
618 if (pskb_trim(skb, snaplen))
621 skb_set_owner_r(skb, sk);
625 /* drop conntrack reference */
628 spin_lock(&sk->sk_receive_queue.lock);
629 po->stats.tp_packets++;
630 skb->dropcount = atomic_read(&sk->sk_drops);
631 __skb_queue_tail(&sk->sk_receive_queue, skb);
632 spin_unlock(&sk->sk_receive_queue.lock);
633 sk->sk_data_ready(sk, skb->len);
637 po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
640 if (skb_head != skb->data && skb_shared(skb)) {
641 skb->data = skb_head;
649 #ifdef CONFIG_PACKET_MMAP
650 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
651 struct packet_type *pt, struct net_device *orig_dev)
654 struct packet_sock *po;
655 struct sockaddr_ll *sll;
657 struct tpacket_hdr *h1;
658 struct tpacket2_hdr *h2;
661 u8 *skb_head = skb->data;
662 int skb_len = skb->len;
663 unsigned int snaplen, res;
664 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
665 unsigned short macoff, netoff, hdrlen;
666 struct sk_buff *copy_skb = NULL;
670 if (skb->pkt_type == PACKET_LOOPBACK)
673 sk = pt->af_packet_priv;
676 if (dev_net(dev) != sock_net(sk))
679 if (dev->header_ops) {
680 if (sk->sk_type != SOCK_DGRAM)
681 skb_push(skb, skb->data - skb_mac_header(skb));
682 else if (skb->pkt_type == PACKET_OUTGOING) {
683 /* Special case: outgoing packets have ll header at head */
684 skb_pull(skb, skb_network_offset(skb));
688 if (skb->ip_summed == CHECKSUM_PARTIAL)
689 status |= TP_STATUS_CSUMNOTREADY;
693 res = run_filter(skb, sk, snaplen);
699 if (sk->sk_type == SOCK_DGRAM) {
700 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
703 unsigned maclen = skb_network_offset(skb);
704 netoff = TPACKET_ALIGN(po->tp_hdrlen +
705 (maclen < 16 ? 16 : maclen)) +
707 macoff = netoff - maclen;
710 if (macoff + snaplen > po->rx_ring.frame_size) {
711 if (po->copy_thresh &&
712 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
713 (unsigned)sk->sk_rcvbuf) {
714 if (skb_shared(skb)) {
715 copy_skb = skb_clone(skb, GFP_ATOMIC);
717 copy_skb = skb_get(skb);
718 skb_head = skb->data;
721 skb_set_owner_r(copy_skb, sk);
723 snaplen = po->rx_ring.frame_size - macoff;
724 if ((int)snaplen < 0)
728 spin_lock(&sk->sk_receive_queue.lock);
729 h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
732 packet_increment_head(&po->rx_ring);
733 po->stats.tp_packets++;
735 status |= TP_STATUS_COPY;
736 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
738 if (!po->stats.tp_drops)
739 status &= ~TP_STATUS_LOSING;
740 spin_unlock(&sk->sk_receive_queue.lock);
742 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
744 switch (po->tp_version) {
746 h.h1->tp_len = skb->len;
747 h.h1->tp_snaplen = snaplen;
748 h.h1->tp_mac = macoff;
749 h.h1->tp_net = netoff;
750 if (skb->tstamp.tv64)
751 tv = ktime_to_timeval(skb->tstamp);
753 do_gettimeofday(&tv);
754 h.h1->tp_sec = tv.tv_sec;
755 h.h1->tp_usec = tv.tv_usec;
756 hdrlen = sizeof(*h.h1);
759 h.h2->tp_len = skb->len;
760 h.h2->tp_snaplen = snaplen;
761 h.h2->tp_mac = macoff;
762 h.h2->tp_net = netoff;
763 if (skb->tstamp.tv64)
764 ts = ktime_to_timespec(skb->tstamp);
767 h.h2->tp_sec = ts.tv_sec;
768 h.h2->tp_nsec = ts.tv_nsec;
769 h.h2->tp_vlan_tci = skb->vlan_tci;
770 hdrlen = sizeof(*h.h2);
776 sll = h.raw + TPACKET_ALIGN(hdrlen);
777 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
778 sll->sll_family = AF_PACKET;
779 sll->sll_hatype = dev->type;
780 sll->sll_protocol = skb->protocol;
781 sll->sll_pkttype = skb->pkt_type;
782 if (unlikely(po->origdev))
783 sll->sll_ifindex = orig_dev->ifindex;
785 sll->sll_ifindex = dev->ifindex;
787 __packet_set_status(po, h.raw, status);
790 struct page *p_start, *p_end;
791 u8 *h_end = h.raw + macoff + snaplen - 1;
793 p_start = virt_to_page(h.raw);
794 p_end = virt_to_page(h_end);
795 while (p_start <= p_end) {
796 flush_dcache_page(p_start);
801 sk->sk_data_ready(sk, 0);
804 if (skb_head != skb->data && skb_shared(skb)) {
805 skb->data = skb_head;
813 po->stats.tp_drops++;
814 spin_unlock(&sk->sk_receive_queue.lock);
816 sk->sk_data_ready(sk, 0);
821 static void tpacket_destruct_skb(struct sk_buff *skb)
823 struct packet_sock *po = pkt_sk(skb->sk);
828 if (likely(po->tx_ring.pg_vec)) {
829 ph = skb_shinfo(skb)->destructor_arg;
830 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
831 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
832 atomic_dec(&po->tx_ring.pending);
833 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
839 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
840 void *frame, struct net_device *dev, int size_max,
841 __be16 proto, unsigned char *addr)
844 struct tpacket_hdr *h1;
845 struct tpacket2_hdr *h2;
848 int to_write, offset, len, tp_len, nr_frags, len_max;
849 struct socket *sock = po->sk.sk_socket;
856 skb->protocol = proto;
858 skb->priority = po->sk.sk_priority;
859 skb->mark = po->sk.sk_mark;
860 skb_shinfo(skb)->destructor_arg = ph.raw;
862 switch (po->tp_version) {
864 tp_len = ph.h2->tp_len;
867 tp_len = ph.h1->tp_len;
870 if (unlikely(tp_len > size_max)) {
871 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
875 skb_reserve(skb, LL_RESERVED_SPACE(dev));
876 skb_reset_network_header(skb);
878 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
881 if (sock->type == SOCK_DGRAM) {
882 err = dev_hard_header(skb, dev, ntohs(proto), addr,
884 if (unlikely(err < 0))
886 } else if (dev->hard_header_len) {
887 /* net device doesn't like empty head */
888 if (unlikely(tp_len <= dev->hard_header_len)) {
889 pr_err("packet size is too short (%d < %d)\n",
890 tp_len, dev->hard_header_len);
894 skb_push(skb, dev->hard_header_len);
895 err = skb_store_bits(skb, 0, data,
896 dev->hard_header_len);
900 data += dev->hard_header_len;
901 to_write -= dev->hard_header_len;
905 page = virt_to_page(data);
906 offset = offset_in_page(data);
907 len_max = PAGE_SIZE - offset;
908 len = ((to_write > len_max) ? len_max : to_write);
910 skb->data_len = to_write;
911 skb->len += to_write;
912 skb->truesize += to_write;
913 atomic_add(to_write, &po->sk.sk_wmem_alloc);
915 while (likely(to_write)) {
916 nr_frags = skb_shinfo(skb)->nr_frags;
918 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
919 pr_err("Packet exceed the number of skb frags(%lu)\n",
924 flush_dcache_page(page);
926 skb_fill_page_desc(skb,
928 page++, offset, len);
932 len = ((to_write > len_max) ? len_max : to_write);
938 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
942 struct net_device *dev;
944 int ifindex, err, reserve = 0;
946 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
947 int tp_len, size_max;
952 sock = po->sk.sk_socket;
954 mutex_lock(&po->pg_vec_lock);
958 ifindex = po->ifindex;
963 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
965 if (msg->msg_namelen < (saddr->sll_halen
966 + offsetof(struct sockaddr_ll,
969 ifindex = saddr->sll_ifindex;
970 proto = saddr->sll_protocol;
971 addr = saddr->sll_addr;
974 dev = dev_get_by_index(sock_net(&po->sk), ifindex);
976 if (unlikely(dev == NULL))
979 reserve = dev->hard_header_len;
982 if (unlikely(!(dev->flags & IFF_UP)))
985 size_max = po->tx_ring.frame_size
986 - sizeof(struct skb_shared_info)
988 - LL_ALLOCATED_SPACE(dev)
989 - sizeof(struct sockaddr_ll);
991 if (size_max > dev->mtu + reserve)
992 size_max = dev->mtu + reserve;
995 ph = packet_current_frame(po, &po->tx_ring,
996 TP_STATUS_SEND_REQUEST);
998 if (unlikely(ph == NULL)) {
1003 status = TP_STATUS_SEND_REQUEST;
1004 skb = sock_alloc_send_skb(&po->sk,
1005 LL_ALLOCATED_SPACE(dev)
1006 + sizeof(struct sockaddr_ll),
1009 if (unlikely(skb == NULL))
1012 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1015 if (unlikely(tp_len < 0)) {
1017 __packet_set_status(po, ph,
1018 TP_STATUS_AVAILABLE);
1019 packet_increment_head(&po->tx_ring);
1023 status = TP_STATUS_WRONG_FORMAT;
1029 skb->destructor = tpacket_destruct_skb;
1030 __packet_set_status(po, ph, TP_STATUS_SENDING);
1031 atomic_inc(&po->tx_ring.pending);
1033 status = TP_STATUS_SEND_REQUEST;
1034 err = dev_queue_xmit(skb);
1035 if (unlikely(err > 0 && (err = net_xmit_errno(err)) != 0))
1037 packet_increment_head(&po->tx_ring);
1039 } while (likely((ph != NULL) || ((!(msg->msg_flags & MSG_DONTWAIT))
1040 && (atomic_read(&po->tx_ring.pending))))
1047 skb->destructor = sock_wfree;
1048 atomic_dec(&po->tx_ring.pending);
1050 __packet_set_status(po, ph, status);
1055 mutex_unlock(&po->pg_vec_lock);
1060 static int packet_snd(struct socket *sock,
1061 struct msghdr *msg, size_t len)
1063 struct sock *sk = sock->sk;
1064 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1065 struct sk_buff *skb;
1066 struct net_device *dev;
1068 unsigned char *addr;
1069 int ifindex, err, reserve = 0;
1072 * Get and verify the address.
1075 if (saddr == NULL) {
1076 struct packet_sock *po = pkt_sk(sk);
1078 ifindex = po->ifindex;
1083 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1085 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1087 ifindex = saddr->sll_ifindex;
1088 proto = saddr->sll_protocol;
1089 addr = saddr->sll_addr;
1093 dev = dev_get_by_index(sock_net(sk), ifindex);
1097 if (sock->type == SOCK_RAW)
1098 reserve = dev->hard_header_len;
1101 if (!(dev->flags & IFF_UP))
1105 if (len > dev->mtu+reserve)
1108 skb = sock_alloc_send_skb(sk, len + LL_ALLOCATED_SPACE(dev),
1109 msg->msg_flags & MSG_DONTWAIT, &err);
1113 skb_reserve(skb, LL_RESERVED_SPACE(dev));
1114 skb_reset_network_header(skb);
1117 if (sock->type == SOCK_DGRAM &&
1118 dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len) < 0)
1121 /* Returns -EFAULT on error */
1122 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1126 skb->protocol = proto;
1128 skb->priority = sk->sk_priority;
1129 skb->mark = sk->sk_mark;
1135 err = dev_queue_xmit(skb);
1136 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1152 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1153 struct msghdr *msg, size_t len)
1155 #ifdef CONFIG_PACKET_MMAP
1156 struct sock *sk = sock->sk;
1157 struct packet_sock *po = pkt_sk(sk);
1158 if (po->tx_ring.pg_vec)
1159 return tpacket_snd(po, msg);
1162 return packet_snd(sock, msg, len);
1166 * Close a PACKET socket. This is fairly simple. We immediately go
1167 * to 'closed' state and remove our protocol entry in the device list.
1170 static int packet_release(struct socket *sock)
1172 struct sock *sk = sock->sk;
1173 struct packet_sock *po;
1175 #ifdef CONFIG_PACKET_MMAP
1176 struct tpacket_req req;
1185 write_lock_bh(&net->packet.sklist_lock);
1186 sk_del_node_init(sk);
1187 sock_prot_inuse_add(net, sk->sk_prot, -1);
1188 write_unlock_bh(&net->packet.sklist_lock);
1191 * Unhook packet receive handler.
1196 * Remove the protocol hook
1198 dev_remove_pack(&po->prot_hook);
1204 packet_flush_mclist(sk);
1206 #ifdef CONFIG_PACKET_MMAP
1207 memset(&req, 0, sizeof(req));
1209 if (po->rx_ring.pg_vec)
1210 packet_set_ring(sk, &req, 1, 0);
1212 if (po->tx_ring.pg_vec)
1213 packet_set_ring(sk, &req, 1, 1);
1217 * Now the socket is dead. No more input will appear.
1225 skb_queue_purge(&sk->sk_receive_queue);
1226 sk_refcnt_debug_release(sk);
1233 * Attach a packet hook.
1236 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1238 struct packet_sock *po = pkt_sk(sk);
1240 * Detach an existing hook if present.
1245 spin_lock(&po->bind_lock);
1250 spin_unlock(&po->bind_lock);
1251 dev_remove_pack(&po->prot_hook);
1252 spin_lock(&po->bind_lock);
1256 po->prot_hook.type = protocol;
1257 po->prot_hook.dev = dev;
1259 po->ifindex = dev ? dev->ifindex : 0;
1264 if (!dev || (dev->flags & IFF_UP)) {
1265 dev_add_pack(&po->prot_hook);
1269 sk->sk_err = ENETDOWN;
1270 if (!sock_flag(sk, SOCK_DEAD))
1271 sk->sk_error_report(sk);
1275 spin_unlock(&po->bind_lock);
1281 * Bind a packet socket to a device
1284 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1287 struct sock *sk = sock->sk;
1289 struct net_device *dev;
1296 if (addr_len != sizeof(struct sockaddr))
1298 strlcpy(name, uaddr->sa_data, sizeof(name));
1300 dev = dev_get_by_name(sock_net(sk), name);
1302 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1308 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1310 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1311 struct sock *sk = sock->sk;
1312 struct net_device *dev = NULL;
1320 if (addr_len < sizeof(struct sockaddr_ll))
1322 if (sll->sll_family != AF_PACKET)
1325 if (sll->sll_ifindex) {
1327 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1331 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1339 static struct proto packet_proto = {
1341 .owner = THIS_MODULE,
1342 .obj_size = sizeof(struct packet_sock),
1346 * Create a packet of type SOCK_PACKET.
1349 static int packet_create(struct net *net, struct socket *sock, int protocol)
1352 struct packet_sock *po;
1353 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1356 if (!capable(CAP_NET_RAW))
1358 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1359 sock->type != SOCK_PACKET)
1360 return -ESOCKTNOSUPPORT;
1362 sock->state = SS_UNCONNECTED;
1365 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1369 sock->ops = &packet_ops;
1370 if (sock->type == SOCK_PACKET)
1371 sock->ops = &packet_ops_spkt;
1373 sock_init_data(sock, sk);
1376 sk->sk_family = PF_PACKET;
1379 sk->sk_destruct = packet_sock_destruct;
1380 sk_refcnt_debug_inc(sk);
1383 * Attach a protocol block
1386 spin_lock_init(&po->bind_lock);
1387 mutex_init(&po->pg_vec_lock);
1388 po->prot_hook.func = packet_rcv;
1390 if (sock->type == SOCK_PACKET)
1391 po->prot_hook.func = packet_rcv_spkt;
1393 po->prot_hook.af_packet_priv = sk;
1396 po->prot_hook.type = proto;
1397 dev_add_pack(&po->prot_hook);
1402 write_lock_bh(&net->packet.sklist_lock);
1403 sk_add_node(sk, &net->packet.sklist);
1404 sock_prot_inuse_add(net, &packet_proto, 1);
1405 write_unlock_bh(&net->packet.sklist_lock);
1412 * Pull a packet from our receive queue and hand it to the user.
1413 * If necessary we block.
1416 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1417 struct msghdr *msg, size_t len, int flags)
1419 struct sock *sk = sock->sk;
1420 struct sk_buff *skb;
1422 struct sockaddr_ll *sll;
1425 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1429 /* What error should we return now? EUNATTACH? */
1430 if (pkt_sk(sk)->ifindex < 0)
1435 * Call the generic datagram receiver. This handles all sorts
1436 * of horrible races and re-entrancy so we can forget about it
1437 * in the protocol layers.
1439 * Now it will return ENETDOWN, if device have just gone down,
1440 * but then it will block.
1443 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1446 * An error occurred so return it. Because skb_recv_datagram()
1447 * handles the blocking we don't see and worry about blocking
1455 * If the address length field is there to be filled in, we fill
1459 sll = &PACKET_SKB_CB(skb)->sa.ll;
1460 if (sock->type == SOCK_PACKET)
1461 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1463 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1466 * You lose any data beyond the buffer you gave. If it worries a
1467 * user program they can ask the device for its MTU anyway.
1473 msg->msg_flags |= MSG_TRUNC;
1476 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1480 sock_recv_ts_and_drops(msg, sk, skb);
1483 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1486 if (pkt_sk(sk)->auxdata) {
1487 struct tpacket_auxdata aux;
1489 aux.tp_status = TP_STATUS_USER;
1490 if (skb->ip_summed == CHECKSUM_PARTIAL)
1491 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1492 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1493 aux.tp_snaplen = skb->len;
1495 aux.tp_net = skb_network_offset(skb);
1496 aux.tp_vlan_tci = skb->vlan_tci;
1498 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1502 * Free or return the buffer as appropriate. Again this
1503 * hides all the races and re-entrancy issues from us.
1505 err = (flags&MSG_TRUNC) ? skb->len : copied;
1508 skb_free_datagram(sk, skb);
1513 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1514 int *uaddr_len, int peer)
1516 struct net_device *dev;
1517 struct sock *sk = sock->sk;
1522 uaddr->sa_family = AF_PACKET;
1523 dev = dev_get_by_index(sock_net(sk), pkt_sk(sk)->ifindex);
1525 strlcpy(uaddr->sa_data, dev->name, 15);
1528 memset(uaddr->sa_data, 0, 14);
1529 *uaddr_len = sizeof(*uaddr);
1534 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1535 int *uaddr_len, int peer)
1537 struct net_device *dev;
1538 struct sock *sk = sock->sk;
1539 struct packet_sock *po = pkt_sk(sk);
1540 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1545 sll->sll_family = AF_PACKET;
1546 sll->sll_ifindex = po->ifindex;
1547 sll->sll_protocol = po->num;
1548 dev = dev_get_by_index(sock_net(sk), po->ifindex);
1550 sll->sll_hatype = dev->type;
1551 sll->sll_halen = dev->addr_len;
1552 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1555 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1558 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1563 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1567 case PACKET_MR_MULTICAST:
1569 return dev_mc_add(dev, i->addr, i->alen, 0);
1571 return dev_mc_delete(dev, i->addr, i->alen, 0);
1573 case PACKET_MR_PROMISC:
1574 return dev_set_promiscuity(dev, what);
1576 case PACKET_MR_ALLMULTI:
1577 return dev_set_allmulti(dev, what);
1579 case PACKET_MR_UNICAST:
1581 return dev_unicast_add(dev, i->addr);
1583 return dev_unicast_delete(dev, i->addr);
1591 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1593 for ( ; i; i = i->next) {
1594 if (i->ifindex == dev->ifindex)
1595 packet_dev_mc(dev, i, what);
1599 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1601 struct packet_sock *po = pkt_sk(sk);
1602 struct packet_mclist *ml, *i;
1603 struct net_device *dev;
1609 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1614 if (mreq->mr_alen > dev->addr_len)
1618 i = kmalloc(sizeof(*i), GFP_KERNEL);
1623 for (ml = po->mclist; ml; ml = ml->next) {
1624 if (ml->ifindex == mreq->mr_ifindex &&
1625 ml->type == mreq->mr_type &&
1626 ml->alen == mreq->mr_alen &&
1627 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1629 /* Free the new element ... */
1635 i->type = mreq->mr_type;
1636 i->ifindex = mreq->mr_ifindex;
1637 i->alen = mreq->mr_alen;
1638 memcpy(i->addr, mreq->mr_address, i->alen);
1640 i->next = po->mclist;
1642 err = packet_dev_mc(dev, i, 1);
1644 po->mclist = i->next;
1653 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1655 struct packet_mclist *ml, **mlp;
1659 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1660 if (ml->ifindex == mreq->mr_ifindex &&
1661 ml->type == mreq->mr_type &&
1662 ml->alen == mreq->mr_alen &&
1663 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1664 if (--ml->count == 0) {
1665 struct net_device *dev;
1667 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1669 packet_dev_mc(dev, ml, -1);
1677 return -EADDRNOTAVAIL;
1680 static void packet_flush_mclist(struct sock *sk)
1682 struct packet_sock *po = pkt_sk(sk);
1683 struct packet_mclist *ml;
1689 while ((ml = po->mclist) != NULL) {
1690 struct net_device *dev;
1692 po->mclist = ml->next;
1693 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1695 packet_dev_mc(dev, ml, -1);
1702 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned 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 = sk_wmem_alloc_get(sk);
1988 return put_user(amount, (int __user *)arg);
1992 struct sk_buff *skb;
1995 spin_lock_bh(&sk->sk_receive_queue.lock);
1996 skb = skb_peek(&sk->sk_receive_queue);
1999 spin_unlock_bh(&sk->sk_receive_queue.lock);
2000 return put_user(amount, (int __user *)arg);
2003 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2005 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2015 case SIOCGIFBRDADDR:
2016 case SIOCSIFBRDADDR:
2017 case SIOCGIFNETMASK:
2018 case SIOCSIFNETMASK:
2019 case SIOCGIFDSTADDR:
2020 case SIOCSIFDSTADDR:
2022 if (!net_eq(sock_net(sk), &init_net))
2023 return -ENOIOCTLCMD;
2024 return inet_dgram_ops.ioctl(sock, cmd, arg);
2028 return -ENOIOCTLCMD;
2033 #ifndef CONFIG_PACKET_MMAP
2034 #define packet_mmap sock_no_mmap
2035 #define packet_poll datagram_poll
2038 static unsigned int packet_poll(struct file *file, struct socket *sock,
2041 struct sock *sk = sock->sk;
2042 struct packet_sock *po = pkt_sk(sk);
2043 unsigned int mask = datagram_poll(file, sock, wait);
2045 spin_lock_bh(&sk->sk_receive_queue.lock);
2046 if (po->rx_ring.pg_vec) {
2047 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2048 mask |= POLLIN | POLLRDNORM;
2050 spin_unlock_bh(&sk->sk_receive_queue.lock);
2051 spin_lock_bh(&sk->sk_write_queue.lock);
2052 if (po->tx_ring.pg_vec) {
2053 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2054 mask |= POLLOUT | POLLWRNORM;
2056 spin_unlock_bh(&sk->sk_write_queue.lock);
2061 /* Dirty? Well, I still did not learn better way to account
2065 static void packet_mm_open(struct vm_area_struct *vma)
2067 struct file *file = vma->vm_file;
2068 struct socket *sock = file->private_data;
2069 struct sock *sk = sock->sk;
2072 atomic_inc(&pkt_sk(sk)->mapped);
2075 static void packet_mm_close(struct vm_area_struct *vma)
2077 struct file *file = vma->vm_file;
2078 struct socket *sock = file->private_data;
2079 struct sock *sk = sock->sk;
2082 atomic_dec(&pkt_sk(sk)->mapped);
2085 static const struct vm_operations_struct packet_mmap_ops = {
2086 .open = packet_mm_open,
2087 .close = packet_mm_close,
2090 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
2094 for (i = 0; i < len; i++) {
2095 if (likely(pg_vec[i]))
2096 free_pages((unsigned long) pg_vec[i], order);
2101 static inline char *alloc_one_pg_vec_page(unsigned long order)
2103 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO | __GFP_NOWARN;
2105 return (char *) __get_free_pages(gfp_flags, order);
2108 static char **alloc_pg_vec(struct tpacket_req *req, int order)
2110 unsigned int block_nr = req->tp_block_nr;
2114 pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
2115 if (unlikely(!pg_vec))
2118 for (i = 0; i < block_nr; i++) {
2119 pg_vec[i] = alloc_one_pg_vec_page(order);
2120 if (unlikely(!pg_vec[i]))
2121 goto out_free_pgvec;
2128 free_pg_vec(pg_vec, order, block_nr);
2133 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2134 int closing, int tx_ring)
2136 char **pg_vec = NULL;
2137 struct packet_sock *po = pkt_sk(sk);
2138 int was_running, order = 0;
2139 struct packet_ring_buffer *rb;
2140 struct sk_buff_head *rb_queue;
2144 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2145 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2149 if (atomic_read(&po->mapped))
2151 if (atomic_read(&rb->pending))
2155 if (req->tp_block_nr) {
2156 /* Sanity tests and some calculations */
2158 if (unlikely(rb->pg_vec))
2161 switch (po->tp_version) {
2163 po->tp_hdrlen = TPACKET_HDRLEN;
2166 po->tp_hdrlen = TPACKET2_HDRLEN;
2171 if (unlikely((int)req->tp_block_size <= 0))
2173 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2175 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2178 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2181 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2182 if (unlikely(rb->frames_per_block <= 0))
2184 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2189 order = get_order(req->tp_block_size);
2190 pg_vec = alloc_pg_vec(req, order);
2191 if (unlikely(!pg_vec))
2197 if (unlikely(req->tp_frame_nr))
2203 /* Detach socket from network */
2204 spin_lock(&po->bind_lock);
2205 was_running = po->running;
2208 __dev_remove_pack(&po->prot_hook);
2213 spin_unlock(&po->bind_lock);
2218 mutex_lock(&po->pg_vec_lock);
2219 if (closing || atomic_read(&po->mapped) == 0) {
2221 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
2222 spin_lock_bh(&rb_queue->lock);
2223 pg_vec = XC(rb->pg_vec, pg_vec);
2224 rb->frame_max = (req->tp_frame_nr - 1);
2226 rb->frame_size = req->tp_frame_size;
2227 spin_unlock_bh(&rb_queue->lock);
2229 order = XC(rb->pg_vec_order, order);
2230 req->tp_block_nr = XC(rb->pg_vec_len, req->tp_block_nr);
2232 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2233 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2234 tpacket_rcv : packet_rcv;
2235 skb_queue_purge(rb_queue);
2237 if (atomic_read(&po->mapped))
2238 pr_err("packet_mmap: vma is busy: %d\n",
2239 atomic_read(&po->mapped));
2241 mutex_unlock(&po->pg_vec_lock);
2243 spin_lock(&po->bind_lock);
2244 if (was_running && !po->running) {
2248 dev_add_pack(&po->prot_hook);
2250 spin_unlock(&po->bind_lock);
2255 free_pg_vec(pg_vec, order, req->tp_block_nr);
2260 static int packet_mmap(struct file *file, struct socket *sock,
2261 struct vm_area_struct *vma)
2263 struct sock *sk = sock->sk;
2264 struct packet_sock *po = pkt_sk(sk);
2265 unsigned long size, expected_size;
2266 struct packet_ring_buffer *rb;
2267 unsigned long start;
2274 mutex_lock(&po->pg_vec_lock);
2277 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2279 expected_size += rb->pg_vec_len
2285 if (expected_size == 0)
2288 size = vma->vm_end - vma->vm_start;
2289 if (size != expected_size)
2292 start = vma->vm_start;
2293 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2294 if (rb->pg_vec == NULL)
2297 for (i = 0; i < rb->pg_vec_len; i++) {
2298 struct page *page = virt_to_page(rb->pg_vec[i]);
2301 for (pg_num = 0; pg_num < rb->pg_vec_pages;
2303 err = vm_insert_page(vma, start, page);
2311 atomic_inc(&po->mapped);
2312 vma->vm_ops = &packet_mmap_ops;
2316 mutex_unlock(&po->pg_vec_lock);
2322 static const struct proto_ops packet_ops_spkt = {
2323 .family = PF_PACKET,
2324 .owner = THIS_MODULE,
2325 .release = packet_release,
2326 .bind = packet_bind_spkt,
2327 .connect = sock_no_connect,
2328 .socketpair = sock_no_socketpair,
2329 .accept = sock_no_accept,
2330 .getname = packet_getname_spkt,
2331 .poll = datagram_poll,
2332 .ioctl = packet_ioctl,
2333 .listen = sock_no_listen,
2334 .shutdown = sock_no_shutdown,
2335 .setsockopt = sock_no_setsockopt,
2336 .getsockopt = sock_no_getsockopt,
2337 .sendmsg = packet_sendmsg_spkt,
2338 .recvmsg = packet_recvmsg,
2339 .mmap = sock_no_mmap,
2340 .sendpage = sock_no_sendpage,
2343 static const struct proto_ops packet_ops = {
2344 .family = PF_PACKET,
2345 .owner = THIS_MODULE,
2346 .release = packet_release,
2347 .bind = packet_bind,
2348 .connect = sock_no_connect,
2349 .socketpair = sock_no_socketpair,
2350 .accept = sock_no_accept,
2351 .getname = packet_getname,
2352 .poll = packet_poll,
2353 .ioctl = packet_ioctl,
2354 .listen = sock_no_listen,
2355 .shutdown = sock_no_shutdown,
2356 .setsockopt = packet_setsockopt,
2357 .getsockopt = packet_getsockopt,
2358 .sendmsg = packet_sendmsg,
2359 .recvmsg = packet_recvmsg,
2360 .mmap = packet_mmap,
2361 .sendpage = sock_no_sendpage,
2364 static const struct net_proto_family packet_family_ops = {
2365 .family = PF_PACKET,
2366 .create = packet_create,
2367 .owner = THIS_MODULE,
2370 static struct notifier_block packet_netdev_notifier = {
2371 .notifier_call = packet_notifier,
2374 #ifdef CONFIG_PROC_FS
2375 static inline struct sock *packet_seq_idx(struct net *net, loff_t off)
2378 struct hlist_node *node;
2380 sk_for_each(s, node, &net->packet.sklist) {
2387 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2388 __acquires(seq_file_net(seq)->packet.sklist_lock)
2390 struct net *net = seq_file_net(seq);
2391 read_lock(&net->packet.sklist_lock);
2392 return *pos ? packet_seq_idx(net, *pos - 1) : SEQ_START_TOKEN;
2395 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2397 struct net *net = seq_file_net(seq);
2399 return (v == SEQ_START_TOKEN)
2400 ? sk_head(&net->packet.sklist)
2401 : sk_next((struct sock *)v) ;
2404 static void packet_seq_stop(struct seq_file *seq, void *v)
2405 __releases(seq_file_net(seq)->packet.sklist_lock)
2407 struct net *net = seq_file_net(seq);
2408 read_unlock(&net->packet.sklist_lock);
2411 static int packet_seq_show(struct seq_file *seq, void *v)
2413 if (v == SEQ_START_TOKEN)
2414 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2417 const struct packet_sock *po = pkt_sk(s);
2420 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2422 atomic_read(&s->sk_refcnt),
2427 atomic_read(&s->sk_rmem_alloc),
2435 static const struct seq_operations packet_seq_ops = {
2436 .start = packet_seq_start,
2437 .next = packet_seq_next,
2438 .stop = packet_seq_stop,
2439 .show = packet_seq_show,
2442 static int packet_seq_open(struct inode *inode, struct file *file)
2444 return seq_open_net(inode, file, &packet_seq_ops,
2445 sizeof(struct seq_net_private));
2448 static const struct file_operations packet_seq_fops = {
2449 .owner = THIS_MODULE,
2450 .open = packet_seq_open,
2452 .llseek = seq_lseek,
2453 .release = seq_release_net,
2458 static int packet_net_init(struct net *net)
2460 rwlock_init(&net->packet.sklist_lock);
2461 INIT_HLIST_HEAD(&net->packet.sklist);
2463 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2469 static void packet_net_exit(struct net *net)
2471 proc_net_remove(net, "packet");
2474 static struct pernet_operations packet_net_ops = {
2475 .init = packet_net_init,
2476 .exit = packet_net_exit,
2480 static void __exit packet_exit(void)
2482 unregister_netdevice_notifier(&packet_netdev_notifier);
2483 unregister_pernet_subsys(&packet_net_ops);
2484 sock_unregister(PF_PACKET);
2485 proto_unregister(&packet_proto);
2488 static int __init packet_init(void)
2490 int rc = proto_register(&packet_proto, 0);
2495 sock_register(&packet_family_ops);
2496 register_pernet_subsys(&packet_net_ops);
2497 register_netdevice_notifier(&packet_netdev_notifier);
2502 module_init(packet_init);
2503 module_exit(packet_exit);
2504 MODULE_LICENSE("GPL");
2505 MODULE_ALIAS_NETPROTO(PF_PACKET);