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 <linux/slab.h>
64 #include <net/net_namespace.h>
66 #include <net/protocol.h>
67 #include <linux/skbuff.h>
69 #include <linux/errno.h>
70 #include <linux/timer.h>
71 #include <asm/system.h>
72 #include <asm/uaccess.h>
73 #include <asm/ioctls.h>
75 #include <asm/cacheflush.h>
77 #include <linux/proc_fs.h>
78 #include <linux/seq_file.h>
79 #include <linux/poll.h>
80 #include <linux/module.h>
81 #include <linux/init.h>
82 #include <linux/mutex.h>
83 #include <linux/if_vlan.h>
84 #include <linux/virtio_net.h>
87 #include <net/inet_common.h>
92 - if device has no dev->hard_header routine, it adds and removes ll header
93 inside itself. In this case ll header is invisible outside of device,
94 but higher levels still should reserve dev->hard_header_len.
95 Some devices are enough clever to reallocate skb, when header
96 will not fit to reserved space (tunnel), another ones are silly
98 - packet socket receives packets with pulled ll header,
99 so that SOCK_RAW should push it back.
104 Incoming, dev->hard_header!=NULL
105 mac_header -> ll header
108 Outgoing, dev->hard_header!=NULL
109 mac_header -> ll header
112 Incoming, dev->hard_header==NULL
113 mac_header -> UNKNOWN position. It is very likely, that it points to ll
114 header. PPP makes it, that is wrong, because introduce
115 assymetry between rx and tx paths.
118 Outgoing, dev->hard_header==NULL
119 mac_header -> data. ll header is still not built!
123 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
129 dev->hard_header != NULL
130 mac_header -> ll header
133 dev->hard_header == NULL (ll header is added by device, we cannot control it)
137 We should set nh.raw on output to correct posistion,
138 packet classifier depends on it.
141 /* Private packet socket structures. */
143 struct packet_mclist {
144 struct packet_mclist *next;
149 unsigned char addr[MAX_ADDR_LEN];
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
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);
181 static void packet_flush_mclist(struct sock *sk);
184 /* struct sock has to be the first member of packet_sock */
186 struct tpacket_stats stats;
187 struct packet_ring_buffer rx_ring;
188 struct packet_ring_buffer tx_ring;
190 spinlock_t bind_lock;
191 struct mutex pg_vec_lock;
192 unsigned int running:1, /* prot_hook is attached*/
196 int ifindex; /* bound device */
198 struct packet_mclist *mclist;
200 enum tpacket_versions tp_version;
201 unsigned int tp_hdrlen;
202 unsigned int tp_reserve;
203 unsigned int tp_loss:1;
204 struct packet_type prot_hook ____cacheline_aligned_in_smp;
207 struct packet_skb_cb {
208 unsigned int origlen;
210 struct sockaddr_pkt pkt;
211 struct sockaddr_ll ll;
215 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
217 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
220 struct tpacket_hdr *h1;
221 struct tpacket2_hdr *h2;
226 switch (po->tp_version) {
228 h.h1->tp_status = status;
229 flush_dcache_page(virt_to_page(&h.h1->tp_status));
232 h.h2->tp_status = status;
233 flush_dcache_page(virt_to_page(&h.h2->tp_status));
236 pr_err("TPACKET version not supported\n");
243 static int __packet_get_status(struct packet_sock *po, void *frame)
246 struct tpacket_hdr *h1;
247 struct tpacket2_hdr *h2;
254 switch (po->tp_version) {
256 flush_dcache_page(virt_to_page(&h.h1->tp_status));
257 return h.h1->tp_status;
259 flush_dcache_page(virt_to_page(&h.h2->tp_status));
260 return h.h2->tp_status;
262 pr_err("TPACKET version not supported\n");
268 static void *packet_lookup_frame(struct packet_sock *po,
269 struct packet_ring_buffer *rb,
270 unsigned int position,
273 unsigned int pg_vec_pos, frame_offset;
275 struct tpacket_hdr *h1;
276 struct tpacket2_hdr *h2;
280 pg_vec_pos = position / rb->frames_per_block;
281 frame_offset = position % rb->frames_per_block;
283 h.raw = rb->pg_vec[pg_vec_pos] + (frame_offset * rb->frame_size);
285 if (status != __packet_get_status(po, h.raw))
291 static inline void *packet_current_frame(struct packet_sock *po,
292 struct packet_ring_buffer *rb,
295 return packet_lookup_frame(po, rb, rb->head, status);
298 static inline void *packet_previous_frame(struct packet_sock *po,
299 struct packet_ring_buffer *rb,
302 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
303 return packet_lookup_frame(po, rb, previous, status);
306 static inline void packet_increment_head(struct packet_ring_buffer *buff)
308 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
311 static inline struct packet_sock *pkt_sk(struct sock *sk)
313 return (struct packet_sock *)sk;
316 static void packet_sock_destruct(struct sock *sk)
318 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
319 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
321 if (!sock_flag(sk, SOCK_DEAD)) {
322 pr_err("Attempt to release alive packet socket: %p\n", sk);
326 sk_refcnt_debug_dec(sk);
330 static const struct proto_ops packet_ops;
332 static const struct proto_ops packet_ops_spkt;
334 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
335 struct packet_type *pt, struct net_device *orig_dev)
338 struct sockaddr_pkt *spkt;
341 * When we registered the protocol we saved the socket in the data
342 * field for just this event.
345 sk = pt->af_packet_priv;
348 * Yank back the headers [hope the device set this
349 * right or kerboom...]
351 * Incoming packets have ll header pulled,
354 * For outgoing ones skb->data == skb_mac_header(skb)
355 * so that this procedure is noop.
358 if (skb->pkt_type == PACKET_LOOPBACK)
361 if (!net_eq(dev_net(dev), sock_net(sk)))
364 skb = skb_share_check(skb, GFP_ATOMIC);
368 /* drop any routing info */
371 /* drop conntrack reference */
374 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
376 skb_push(skb, skb->data - skb_mac_header(skb));
379 * The SOCK_PACKET socket receives _all_ frames.
382 spkt->spkt_family = dev->type;
383 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
384 spkt->spkt_protocol = skb->protocol;
387 * Charge the memory to the socket. This is done specifically
388 * to prevent sockets using all the memory up.
391 if (sock_queue_rcv_skb(sk, skb) == 0)
402 * Output a raw packet to a device layer. This bypasses all the other
403 * protocol layers and you must therefore supply it with a complete frame
406 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
407 struct msghdr *msg, size_t len)
409 struct sock *sk = sock->sk;
410 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
411 struct sk_buff *skb = NULL;
412 struct net_device *dev;
417 * Get and verify the address.
421 if (msg->msg_namelen < sizeof(struct sockaddr))
423 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
424 proto = saddr->spkt_protocol;
426 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
429 * Find the device first to size check it
432 saddr->spkt_device[13] = 0;
435 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
441 if (!(dev->flags & IFF_UP))
445 * You may not queue a frame bigger than the mtu. This is the lowest level
446 * raw protocol and you must do your own fragmentation at this level.
450 if (len > dev->mtu + dev->hard_header_len)
454 size_t reserved = LL_RESERVED_SPACE(dev);
455 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
458 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
461 /* FIXME: Save some space for broken drivers that write a hard
462 * header at transmission time by themselves. PPP is the notable
463 * one here. This should really be fixed at the driver level.
465 skb_reserve(skb, reserved);
466 skb_reset_network_header(skb);
468 /* Try to align data part correctly */
473 skb_reset_network_header(skb);
475 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
482 skb->protocol = proto;
484 skb->priority = sk->sk_priority;
485 skb->mark = sk->sk_mark;
498 static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
501 struct sk_filter *filter;
504 filter = rcu_dereference_bh(sk->sk_filter);
506 res = sk_run_filter(skb, filter->insns, filter->len);
507 rcu_read_unlock_bh();
513 This function makes lazy skb cloning in hope that most of packets
514 are discarded by BPF.
516 Note tricky part: we DO mangle shared skb! skb->data, skb->len
517 and skb->cb are mangled. It works because (and until) packets
518 falling here are owned by current CPU. Output packets are cloned
519 by dev_queue_xmit_nit(), input packets are processed by net_bh
520 sequencially, so that if we return skb to original state on exit,
521 we will not harm anyone.
524 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
525 struct packet_type *pt, struct net_device *orig_dev)
528 struct sockaddr_ll *sll;
529 struct packet_sock *po;
530 u8 *skb_head = skb->data;
531 int skb_len = skb->len;
532 unsigned int snaplen, res;
534 if (skb->pkt_type == PACKET_LOOPBACK)
537 sk = pt->af_packet_priv;
540 if (!net_eq(dev_net(dev), sock_net(sk)))
545 if (dev->header_ops) {
546 /* The device has an explicit notion of ll header,
547 exported to higher levels.
549 Otherwise, the device hides datails of it frame
550 structure, so that corresponding packet head
551 never delivered to user.
553 if (sk->sk_type != SOCK_DGRAM)
554 skb_push(skb, skb->data - skb_mac_header(skb));
555 else if (skb->pkt_type == PACKET_OUTGOING) {
556 /* Special case: outgoing packets have ll header at head */
557 skb_pull(skb, skb_network_offset(skb));
563 res = run_filter(skb, sk, snaplen);
569 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
570 (unsigned)sk->sk_rcvbuf)
573 if (skb_shared(skb)) {
574 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
578 if (skb_head != skb->data) {
579 skb->data = skb_head;
586 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
589 sll = &PACKET_SKB_CB(skb)->sa.ll;
590 sll->sll_family = AF_PACKET;
591 sll->sll_hatype = dev->type;
592 sll->sll_protocol = skb->protocol;
593 sll->sll_pkttype = skb->pkt_type;
594 if (unlikely(po->origdev))
595 sll->sll_ifindex = orig_dev->ifindex;
597 sll->sll_ifindex = dev->ifindex;
599 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
601 PACKET_SKB_CB(skb)->origlen = skb->len;
603 if (pskb_trim(skb, snaplen))
606 skb_set_owner_r(skb, sk);
610 /* drop conntrack reference */
613 spin_lock(&sk->sk_receive_queue.lock);
614 po->stats.tp_packets++;
615 skb->dropcount = atomic_read(&sk->sk_drops);
616 __skb_queue_tail(&sk->sk_receive_queue, skb);
617 spin_unlock(&sk->sk_receive_queue.lock);
618 sk->sk_data_ready(sk, skb->len);
622 po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
625 if (skb_head != skb->data && skb_shared(skb)) {
626 skb->data = skb_head;
634 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
635 struct packet_type *pt, struct net_device *orig_dev)
638 struct packet_sock *po;
639 struct sockaddr_ll *sll;
641 struct tpacket_hdr *h1;
642 struct tpacket2_hdr *h2;
645 u8 *skb_head = skb->data;
646 int skb_len = skb->len;
647 unsigned int snaplen, res;
648 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
649 unsigned short macoff, netoff, hdrlen;
650 struct sk_buff *copy_skb = NULL;
654 if (skb->pkt_type == PACKET_LOOPBACK)
657 sk = pt->af_packet_priv;
660 if (!net_eq(dev_net(dev), sock_net(sk)))
663 if (dev->header_ops) {
664 if (sk->sk_type != SOCK_DGRAM)
665 skb_push(skb, skb->data - skb_mac_header(skb));
666 else if (skb->pkt_type == PACKET_OUTGOING) {
667 /* Special case: outgoing packets have ll header at head */
668 skb_pull(skb, skb_network_offset(skb));
672 if (skb->ip_summed == CHECKSUM_PARTIAL)
673 status |= TP_STATUS_CSUMNOTREADY;
677 res = run_filter(skb, sk, snaplen);
683 if (sk->sk_type == SOCK_DGRAM) {
684 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
687 unsigned maclen = skb_network_offset(skb);
688 netoff = TPACKET_ALIGN(po->tp_hdrlen +
689 (maclen < 16 ? 16 : maclen)) +
691 macoff = netoff - maclen;
694 if (macoff + snaplen > po->rx_ring.frame_size) {
695 if (po->copy_thresh &&
696 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
697 (unsigned)sk->sk_rcvbuf) {
698 if (skb_shared(skb)) {
699 copy_skb = skb_clone(skb, GFP_ATOMIC);
701 copy_skb = skb_get(skb);
702 skb_head = skb->data;
705 skb_set_owner_r(copy_skb, sk);
707 snaplen = po->rx_ring.frame_size - macoff;
708 if ((int)snaplen < 0)
712 spin_lock(&sk->sk_receive_queue.lock);
713 h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
716 packet_increment_head(&po->rx_ring);
717 po->stats.tp_packets++;
719 status |= TP_STATUS_COPY;
720 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
722 if (!po->stats.tp_drops)
723 status &= ~TP_STATUS_LOSING;
724 spin_unlock(&sk->sk_receive_queue.lock);
726 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
728 switch (po->tp_version) {
730 h.h1->tp_len = skb->len;
731 h.h1->tp_snaplen = snaplen;
732 h.h1->tp_mac = macoff;
733 h.h1->tp_net = netoff;
734 if (skb->tstamp.tv64)
735 tv = ktime_to_timeval(skb->tstamp);
737 do_gettimeofday(&tv);
738 h.h1->tp_sec = tv.tv_sec;
739 h.h1->tp_usec = tv.tv_usec;
740 hdrlen = sizeof(*h.h1);
743 h.h2->tp_len = skb->len;
744 h.h2->tp_snaplen = snaplen;
745 h.h2->tp_mac = macoff;
746 h.h2->tp_net = netoff;
747 if (skb->tstamp.tv64)
748 ts = ktime_to_timespec(skb->tstamp);
751 h.h2->tp_sec = ts.tv_sec;
752 h.h2->tp_nsec = ts.tv_nsec;
753 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
754 hdrlen = sizeof(*h.h2);
760 sll = h.raw + TPACKET_ALIGN(hdrlen);
761 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
762 sll->sll_family = AF_PACKET;
763 sll->sll_hatype = dev->type;
764 sll->sll_protocol = skb->protocol;
765 sll->sll_pkttype = skb->pkt_type;
766 if (unlikely(po->origdev))
767 sll->sll_ifindex = orig_dev->ifindex;
769 sll->sll_ifindex = dev->ifindex;
771 __packet_set_status(po, h.raw, status);
774 struct page *p_start, *p_end;
775 u8 *h_end = h.raw + macoff + snaplen - 1;
777 p_start = virt_to_page(h.raw);
778 p_end = virt_to_page(h_end);
779 while (p_start <= p_end) {
780 flush_dcache_page(p_start);
785 sk->sk_data_ready(sk, 0);
788 if (skb_head != skb->data && skb_shared(skb)) {
789 skb->data = skb_head;
797 po->stats.tp_drops++;
798 spin_unlock(&sk->sk_receive_queue.lock);
800 sk->sk_data_ready(sk, 0);
805 static void tpacket_destruct_skb(struct sk_buff *skb)
807 struct packet_sock *po = pkt_sk(skb->sk);
812 if (likely(po->tx_ring.pg_vec)) {
813 ph = skb_shinfo(skb)->destructor_arg;
814 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
815 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
816 atomic_dec(&po->tx_ring.pending);
817 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
823 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
824 void *frame, struct net_device *dev, int size_max,
825 __be16 proto, unsigned char *addr)
828 struct tpacket_hdr *h1;
829 struct tpacket2_hdr *h2;
832 int to_write, offset, len, tp_len, nr_frags, len_max;
833 struct socket *sock = po->sk.sk_socket;
840 skb->protocol = proto;
842 skb->priority = po->sk.sk_priority;
843 skb->mark = po->sk.sk_mark;
844 skb_shinfo(skb)->destructor_arg = ph.raw;
846 switch (po->tp_version) {
848 tp_len = ph.h2->tp_len;
851 tp_len = ph.h1->tp_len;
854 if (unlikely(tp_len > size_max)) {
855 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
859 skb_reserve(skb, LL_RESERVED_SPACE(dev));
860 skb_reset_network_header(skb);
862 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
865 if (sock->type == SOCK_DGRAM) {
866 err = dev_hard_header(skb, dev, ntohs(proto), addr,
868 if (unlikely(err < 0))
870 } else if (dev->hard_header_len) {
871 /* net device doesn't like empty head */
872 if (unlikely(tp_len <= dev->hard_header_len)) {
873 pr_err("packet size is too short (%d < %d)\n",
874 tp_len, dev->hard_header_len);
878 skb_push(skb, dev->hard_header_len);
879 err = skb_store_bits(skb, 0, data,
880 dev->hard_header_len);
884 data += dev->hard_header_len;
885 to_write -= dev->hard_header_len;
889 page = virt_to_page(data);
890 offset = offset_in_page(data);
891 len_max = PAGE_SIZE - offset;
892 len = ((to_write > len_max) ? len_max : to_write);
894 skb->data_len = to_write;
895 skb->len += to_write;
896 skb->truesize += to_write;
897 atomic_add(to_write, &po->sk.sk_wmem_alloc);
899 while (likely(to_write)) {
900 nr_frags = skb_shinfo(skb)->nr_frags;
902 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
903 pr_err("Packet exceed the number of skb frags(%lu)\n",
908 flush_dcache_page(page);
910 skb_fill_page_desc(skb,
912 page++, offset, len);
916 len = ((to_write > len_max) ? len_max : to_write);
922 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
926 struct net_device *dev;
928 int ifindex, err, reserve = 0;
930 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
931 int tp_len, size_max;
936 sock = po->sk.sk_socket;
938 mutex_lock(&po->pg_vec_lock);
942 ifindex = po->ifindex;
947 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
949 if (msg->msg_namelen < (saddr->sll_halen
950 + offsetof(struct sockaddr_ll,
953 ifindex = saddr->sll_ifindex;
954 proto = saddr->sll_protocol;
955 addr = saddr->sll_addr;
958 dev = dev_get_by_index(sock_net(&po->sk), ifindex);
960 if (unlikely(dev == NULL))
963 reserve = dev->hard_header_len;
966 if (unlikely(!(dev->flags & IFF_UP)))
969 size_max = po->tx_ring.frame_size
970 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
972 if (size_max > dev->mtu + reserve)
973 size_max = dev->mtu + reserve;
976 ph = packet_current_frame(po, &po->tx_ring,
977 TP_STATUS_SEND_REQUEST);
979 if (unlikely(ph == NULL)) {
984 status = TP_STATUS_SEND_REQUEST;
985 skb = sock_alloc_send_skb(&po->sk,
986 LL_ALLOCATED_SPACE(dev)
987 + sizeof(struct sockaddr_ll),
990 if (unlikely(skb == NULL))
993 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
996 if (unlikely(tp_len < 0)) {
998 __packet_set_status(po, ph,
999 TP_STATUS_AVAILABLE);
1000 packet_increment_head(&po->tx_ring);
1004 status = TP_STATUS_WRONG_FORMAT;
1010 skb->destructor = tpacket_destruct_skb;
1011 __packet_set_status(po, ph, TP_STATUS_SENDING);
1012 atomic_inc(&po->tx_ring.pending);
1014 status = TP_STATUS_SEND_REQUEST;
1015 err = dev_queue_xmit(skb);
1016 if (unlikely(err > 0)) {
1017 err = net_xmit_errno(err);
1018 if (err && __packet_get_status(po, ph) ==
1019 TP_STATUS_AVAILABLE) {
1020 /* skb was destructed already */
1025 * skb was dropped but not destructed yet;
1026 * let's treat it like congestion or err < 0
1030 packet_increment_head(&po->tx_ring);
1032 } while (likely((ph != NULL) ||
1033 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1034 (atomic_read(&po->tx_ring.pending))))
1041 __packet_set_status(po, ph, status);
1046 mutex_unlock(&po->pg_vec_lock);
1050 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1051 size_t reserve, size_t len,
1052 size_t linear, int noblock,
1055 struct sk_buff *skb;
1057 /* Under a page? Don't bother with paged skb. */
1058 if (prepad + len < PAGE_SIZE || !linear)
1061 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1066 skb_reserve(skb, reserve);
1067 skb_put(skb, linear);
1068 skb->data_len = len - linear;
1069 skb->len += len - linear;
1074 static int packet_snd(struct socket *sock,
1075 struct msghdr *msg, size_t len)
1077 struct sock *sk = sock->sk;
1078 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1079 struct sk_buff *skb;
1080 struct net_device *dev;
1082 unsigned char *addr;
1083 int ifindex, err, reserve = 0;
1084 struct virtio_net_hdr vnet_hdr = { 0 };
1087 struct packet_sock *po = pkt_sk(sk);
1088 unsigned short gso_type = 0;
1091 * Get and verify the address.
1094 if (saddr == NULL) {
1095 ifindex = po->ifindex;
1100 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1102 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1104 ifindex = saddr->sll_ifindex;
1105 proto = saddr->sll_protocol;
1106 addr = saddr->sll_addr;
1110 dev = dev_get_by_index(sock_net(sk), ifindex);
1114 if (sock->type == SOCK_RAW)
1115 reserve = dev->hard_header_len;
1118 if (!(dev->flags & IFF_UP))
1121 if (po->has_vnet_hdr) {
1122 vnet_hdr_len = sizeof(vnet_hdr);
1125 if (len < vnet_hdr_len)
1128 len -= vnet_hdr_len;
1130 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1135 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1136 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1138 vnet_hdr.hdr_len = vnet_hdr.csum_start +
1139 vnet_hdr.csum_offset + 2;
1142 if (vnet_hdr.hdr_len > len)
1145 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1146 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1147 case VIRTIO_NET_HDR_GSO_TCPV4:
1148 gso_type = SKB_GSO_TCPV4;
1150 case VIRTIO_NET_HDR_GSO_TCPV6:
1151 gso_type = SKB_GSO_TCPV6;
1153 case VIRTIO_NET_HDR_GSO_UDP:
1154 gso_type = SKB_GSO_UDP;
1160 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1161 gso_type |= SKB_GSO_TCP_ECN;
1163 if (vnet_hdr.gso_size == 0)
1170 if (!gso_type && (len > dev->mtu+reserve))
1174 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1175 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1176 msg->msg_flags & MSG_DONTWAIT, &err);
1180 skb_set_network_header(skb, reserve);
1183 if (sock->type == SOCK_DGRAM &&
1184 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1187 /* Returns -EFAULT on error */
1188 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1192 skb->protocol = proto;
1194 skb->priority = sk->sk_priority;
1195 skb->mark = sk->sk_mark;
1197 if (po->has_vnet_hdr) {
1198 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1199 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1200 vnet_hdr.csum_offset)) {
1206 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1207 skb_shinfo(skb)->gso_type = gso_type;
1209 /* Header must be checked, and gso_segs computed. */
1210 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1211 skb_shinfo(skb)->gso_segs = 0;
1213 len += vnet_hdr_len;
1220 err = dev_queue_xmit(skb);
1221 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1237 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1238 struct msghdr *msg, size_t len)
1240 struct sock *sk = sock->sk;
1241 struct packet_sock *po = pkt_sk(sk);
1242 if (po->tx_ring.pg_vec)
1243 return tpacket_snd(po, msg);
1245 return packet_snd(sock, msg, len);
1249 * Close a PACKET socket. This is fairly simple. We immediately go
1250 * to 'closed' state and remove our protocol entry in the device list.
1253 static int packet_release(struct socket *sock)
1255 struct sock *sk = sock->sk;
1256 struct packet_sock *po;
1258 struct tpacket_req req;
1266 spin_lock_bh(&net->packet.sklist_lock);
1267 sk_del_node_init_rcu(sk);
1268 sock_prot_inuse_add(net, sk->sk_prot, -1);
1269 spin_unlock_bh(&net->packet.sklist_lock);
1271 spin_lock(&po->bind_lock);
1274 * Remove from protocol table
1278 __dev_remove_pack(&po->prot_hook);
1281 spin_unlock(&po->bind_lock);
1283 packet_flush_mclist(sk);
1285 memset(&req, 0, sizeof(req));
1287 if (po->rx_ring.pg_vec)
1288 packet_set_ring(sk, &req, 1, 0);
1290 if (po->tx_ring.pg_vec)
1291 packet_set_ring(sk, &req, 1, 1);
1295 * Now the socket is dead. No more input will appear.
1302 skb_queue_purge(&sk->sk_receive_queue);
1303 sk_refcnt_debug_release(sk);
1310 * Attach a packet hook.
1313 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1315 struct packet_sock *po = pkt_sk(sk);
1317 * Detach an existing hook if present.
1322 spin_lock(&po->bind_lock);
1327 spin_unlock(&po->bind_lock);
1328 dev_remove_pack(&po->prot_hook);
1329 spin_lock(&po->bind_lock);
1333 po->prot_hook.type = protocol;
1334 po->prot_hook.dev = dev;
1336 po->ifindex = dev ? dev->ifindex : 0;
1341 if (!dev || (dev->flags & IFF_UP)) {
1342 dev_add_pack(&po->prot_hook);
1346 sk->sk_err = ENETDOWN;
1347 if (!sock_flag(sk, SOCK_DEAD))
1348 sk->sk_error_report(sk);
1352 spin_unlock(&po->bind_lock);
1358 * Bind a packet socket to a device
1361 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1364 struct sock *sk = sock->sk;
1366 struct net_device *dev;
1373 if (addr_len != sizeof(struct sockaddr))
1375 strlcpy(name, uaddr->sa_data, sizeof(name));
1377 dev = dev_get_by_name(sock_net(sk), name);
1379 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1385 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1387 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1388 struct sock *sk = sock->sk;
1389 struct net_device *dev = NULL;
1397 if (addr_len < sizeof(struct sockaddr_ll))
1399 if (sll->sll_family != AF_PACKET)
1402 if (sll->sll_ifindex) {
1404 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1408 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1416 static struct proto packet_proto = {
1418 .owner = THIS_MODULE,
1419 .obj_size = sizeof(struct packet_sock),
1423 * Create a packet of type SOCK_PACKET.
1426 static int packet_create(struct net *net, struct socket *sock, int protocol,
1430 struct packet_sock *po;
1431 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1434 if (!capable(CAP_NET_RAW))
1436 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1437 sock->type != SOCK_PACKET)
1438 return -ESOCKTNOSUPPORT;
1440 sock->state = SS_UNCONNECTED;
1443 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1447 sock->ops = &packet_ops;
1448 if (sock->type == SOCK_PACKET)
1449 sock->ops = &packet_ops_spkt;
1451 sock_init_data(sock, sk);
1454 sk->sk_family = PF_PACKET;
1457 sk->sk_destruct = packet_sock_destruct;
1458 sk_refcnt_debug_inc(sk);
1461 * Attach a protocol block
1464 spin_lock_init(&po->bind_lock);
1465 mutex_init(&po->pg_vec_lock);
1466 po->prot_hook.func = packet_rcv;
1468 if (sock->type == SOCK_PACKET)
1469 po->prot_hook.func = packet_rcv_spkt;
1471 po->prot_hook.af_packet_priv = sk;
1474 po->prot_hook.type = proto;
1475 dev_add_pack(&po->prot_hook);
1480 spin_lock_bh(&net->packet.sklist_lock);
1481 sk_add_node_rcu(sk, &net->packet.sklist);
1482 sock_prot_inuse_add(net, &packet_proto, 1);
1483 spin_unlock_bh(&net->packet.sklist_lock);
1491 * Pull a packet from our receive queue and hand it to the user.
1492 * If necessary we block.
1495 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1496 struct msghdr *msg, size_t len, int flags)
1498 struct sock *sk = sock->sk;
1499 struct sk_buff *skb;
1501 struct sockaddr_ll *sll;
1502 int vnet_hdr_len = 0;
1505 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1509 /* What error should we return now? EUNATTACH? */
1510 if (pkt_sk(sk)->ifindex < 0)
1515 * Call the generic datagram receiver. This handles all sorts
1516 * of horrible races and re-entrancy so we can forget about it
1517 * in the protocol layers.
1519 * Now it will return ENETDOWN, if device have just gone down,
1520 * but then it will block.
1523 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1526 * An error occurred so return it. Because skb_recv_datagram()
1527 * handles the blocking we don't see and worry about blocking
1534 if (pkt_sk(sk)->has_vnet_hdr) {
1535 struct virtio_net_hdr vnet_hdr = { 0 };
1538 vnet_hdr_len = sizeof(vnet_hdr);
1539 if ((len -= vnet_hdr_len) < 0)
1542 if (skb_is_gso(skb)) {
1543 struct skb_shared_info *sinfo = skb_shinfo(skb);
1545 /* This is a hint as to how much should be linear. */
1546 vnet_hdr.hdr_len = skb_headlen(skb);
1547 vnet_hdr.gso_size = sinfo->gso_size;
1548 if (sinfo->gso_type & SKB_GSO_TCPV4)
1549 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1550 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1551 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1552 else if (sinfo->gso_type & SKB_GSO_UDP)
1553 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1554 else if (sinfo->gso_type & SKB_GSO_FCOE)
1558 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1559 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1561 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1563 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1564 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1565 vnet_hdr.csum_start = skb->csum_start -
1567 vnet_hdr.csum_offset = skb->csum_offset;
1568 } /* else everything is zero */
1570 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1577 * If the address length field is there to be filled in, we fill
1581 sll = &PACKET_SKB_CB(skb)->sa.ll;
1582 if (sock->type == SOCK_PACKET)
1583 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1585 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1588 * You lose any data beyond the buffer you gave. If it worries a
1589 * user program they can ask the device for its MTU anyway.
1595 msg->msg_flags |= MSG_TRUNC;
1598 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1602 sock_recv_ts_and_drops(msg, sk, skb);
1605 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1608 if (pkt_sk(sk)->auxdata) {
1609 struct tpacket_auxdata aux;
1611 aux.tp_status = TP_STATUS_USER;
1612 if (skb->ip_summed == CHECKSUM_PARTIAL)
1613 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1614 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1615 aux.tp_snaplen = skb->len;
1617 aux.tp_net = skb_network_offset(skb);
1618 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1620 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1624 * Free or return the buffer as appropriate. Again this
1625 * hides all the races and re-entrancy issues from us.
1627 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1630 skb_free_datagram(sk, skb);
1635 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1636 int *uaddr_len, int peer)
1638 struct net_device *dev;
1639 struct sock *sk = sock->sk;
1644 uaddr->sa_family = AF_PACKET;
1646 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1648 strlcpy(uaddr->sa_data, dev->name, 15);
1650 memset(uaddr->sa_data, 0, 14);
1652 *uaddr_len = sizeof(*uaddr);
1657 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1658 int *uaddr_len, int peer)
1660 struct net_device *dev;
1661 struct sock *sk = sock->sk;
1662 struct packet_sock *po = pkt_sk(sk);
1663 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1668 sll->sll_family = AF_PACKET;
1669 sll->sll_ifindex = po->ifindex;
1670 sll->sll_protocol = po->num;
1672 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1674 sll->sll_hatype = dev->type;
1675 sll->sll_halen = dev->addr_len;
1676 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1678 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1682 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1687 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1691 case PACKET_MR_MULTICAST:
1692 if (i->alen != dev->addr_len)
1695 return dev_mc_add(dev, i->addr, i->alen, 0);
1697 return dev_mc_delete(dev, i->addr, i->alen, 0);
1699 case PACKET_MR_PROMISC:
1700 return dev_set_promiscuity(dev, what);
1702 case PACKET_MR_ALLMULTI:
1703 return dev_set_allmulti(dev, what);
1705 case PACKET_MR_UNICAST:
1706 if (i->alen != dev->addr_len)
1709 return dev_unicast_add(dev, i->addr);
1711 return dev_unicast_delete(dev, i->addr);
1719 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1721 for ( ; i; i = i->next) {
1722 if (i->ifindex == dev->ifindex)
1723 packet_dev_mc(dev, i, what);
1727 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1729 struct packet_sock *po = pkt_sk(sk);
1730 struct packet_mclist *ml, *i;
1731 struct net_device *dev;
1737 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1742 if (mreq->mr_alen > dev->addr_len)
1746 i = kmalloc(sizeof(*i), GFP_KERNEL);
1751 for (ml = po->mclist; ml; ml = ml->next) {
1752 if (ml->ifindex == mreq->mr_ifindex &&
1753 ml->type == mreq->mr_type &&
1754 ml->alen == mreq->mr_alen &&
1755 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1757 /* Free the new element ... */
1763 i->type = mreq->mr_type;
1764 i->ifindex = mreq->mr_ifindex;
1765 i->alen = mreq->mr_alen;
1766 memcpy(i->addr, mreq->mr_address, i->alen);
1768 i->next = po->mclist;
1770 err = packet_dev_mc(dev, i, 1);
1772 po->mclist = i->next;
1781 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1783 struct packet_mclist *ml, **mlp;
1787 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1788 if (ml->ifindex == mreq->mr_ifindex &&
1789 ml->type == mreq->mr_type &&
1790 ml->alen == mreq->mr_alen &&
1791 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1792 if (--ml->count == 0) {
1793 struct net_device *dev;
1795 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1797 packet_dev_mc(dev, ml, -1);
1805 return -EADDRNOTAVAIL;
1808 static void packet_flush_mclist(struct sock *sk)
1810 struct packet_sock *po = pkt_sk(sk);
1811 struct packet_mclist *ml;
1817 while ((ml = po->mclist) != NULL) {
1818 struct net_device *dev;
1820 po->mclist = ml->next;
1821 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1823 packet_dev_mc(dev, ml, -1);
1830 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1832 struct sock *sk = sock->sk;
1833 struct packet_sock *po = pkt_sk(sk);
1836 if (level != SOL_PACKET)
1837 return -ENOPROTOOPT;
1840 case PACKET_ADD_MEMBERSHIP:
1841 case PACKET_DROP_MEMBERSHIP:
1843 struct packet_mreq_max mreq;
1845 memset(&mreq, 0, sizeof(mreq));
1846 if (len < sizeof(struct packet_mreq))
1848 if (len > sizeof(mreq))
1850 if (copy_from_user(&mreq, optval, len))
1852 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1854 if (optname == PACKET_ADD_MEMBERSHIP)
1855 ret = packet_mc_add(sk, &mreq);
1857 ret = packet_mc_drop(sk, &mreq);
1861 case PACKET_RX_RING:
1862 case PACKET_TX_RING:
1864 struct tpacket_req req;
1866 if (optlen < sizeof(req))
1868 if (pkt_sk(sk)->has_vnet_hdr)
1870 if (copy_from_user(&req, optval, sizeof(req)))
1872 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1874 case PACKET_COPY_THRESH:
1878 if (optlen != sizeof(val))
1880 if (copy_from_user(&val, optval, sizeof(val)))
1883 pkt_sk(sk)->copy_thresh = val;
1886 case PACKET_VERSION:
1890 if (optlen != sizeof(val))
1892 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1894 if (copy_from_user(&val, optval, sizeof(val)))
1899 po->tp_version = val;
1905 case PACKET_RESERVE:
1909 if (optlen != sizeof(val))
1911 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1913 if (copy_from_user(&val, optval, sizeof(val)))
1915 po->tp_reserve = val;
1922 if (optlen != sizeof(val))
1924 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1926 if (copy_from_user(&val, optval, sizeof(val)))
1928 po->tp_loss = !!val;
1931 case PACKET_AUXDATA:
1935 if (optlen < sizeof(val))
1937 if (copy_from_user(&val, optval, sizeof(val)))
1940 po->auxdata = !!val;
1943 case PACKET_ORIGDEV:
1947 if (optlen < sizeof(val))
1949 if (copy_from_user(&val, optval, sizeof(val)))
1952 po->origdev = !!val;
1955 case PACKET_VNET_HDR:
1959 if (sock->type != SOCK_RAW)
1961 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1963 if (optlen < sizeof(val))
1965 if (copy_from_user(&val, optval, sizeof(val)))
1968 po->has_vnet_hdr = !!val;
1972 return -ENOPROTOOPT;
1976 static int packet_getsockopt(struct socket *sock, int level, int optname,
1977 char __user *optval, int __user *optlen)
1981 struct sock *sk = sock->sk;
1982 struct packet_sock *po = pkt_sk(sk);
1984 struct tpacket_stats st;
1986 if (level != SOL_PACKET)
1987 return -ENOPROTOOPT;
1989 if (get_user(len, optlen))
1996 case PACKET_STATISTICS:
1997 if (len > sizeof(struct tpacket_stats))
1998 len = sizeof(struct tpacket_stats);
1999 spin_lock_bh(&sk->sk_receive_queue.lock);
2001 memset(&po->stats, 0, sizeof(st));
2002 spin_unlock_bh(&sk->sk_receive_queue.lock);
2003 st.tp_packets += st.tp_drops;
2007 case PACKET_AUXDATA:
2008 if (len > sizeof(int))
2014 case PACKET_ORIGDEV:
2015 if (len > sizeof(int))
2021 case PACKET_VNET_HDR:
2022 if (len > sizeof(int))
2024 val = po->has_vnet_hdr;
2028 case PACKET_VERSION:
2029 if (len > sizeof(int))
2031 val = po->tp_version;
2035 if (len > sizeof(int))
2037 if (copy_from_user(&val, optval, len))
2041 val = sizeof(struct tpacket_hdr);
2044 val = sizeof(struct tpacket2_hdr);
2051 case PACKET_RESERVE:
2052 if (len > sizeof(unsigned int))
2053 len = sizeof(unsigned int);
2054 val = po->tp_reserve;
2058 if (len > sizeof(unsigned int))
2059 len = sizeof(unsigned int);
2064 return -ENOPROTOOPT;
2067 if (put_user(len, optlen))
2069 if (copy_to_user(optval, data, len))
2075 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2078 struct hlist_node *node;
2079 struct net_device *dev = data;
2080 struct net *net = dev_net(dev);
2083 sk_for_each_rcu(sk, node, &net->packet.sklist) {
2084 struct packet_sock *po = pkt_sk(sk);
2087 case NETDEV_UNREGISTER:
2089 packet_dev_mclist(dev, po->mclist, -1);
2093 if (dev->ifindex == po->ifindex) {
2094 spin_lock(&po->bind_lock);
2096 __dev_remove_pack(&po->prot_hook);
2099 sk->sk_err = ENETDOWN;
2100 if (!sock_flag(sk, SOCK_DEAD))
2101 sk->sk_error_report(sk);
2103 if (msg == NETDEV_UNREGISTER) {
2105 po->prot_hook.dev = NULL;
2107 spin_unlock(&po->bind_lock);
2111 if (dev->ifindex == po->ifindex) {
2112 spin_lock(&po->bind_lock);
2113 if (po->num && !po->running) {
2114 dev_add_pack(&po->prot_hook);
2118 spin_unlock(&po->bind_lock);
2128 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2131 struct sock *sk = sock->sk;
2136 int amount = sk_wmem_alloc_get(sk);
2138 return put_user(amount, (int __user *)arg);
2142 struct sk_buff *skb;
2145 spin_lock_bh(&sk->sk_receive_queue.lock);
2146 skb = skb_peek(&sk->sk_receive_queue);
2149 spin_unlock_bh(&sk->sk_receive_queue.lock);
2150 return put_user(amount, (int __user *)arg);
2153 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2155 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2165 case SIOCGIFBRDADDR:
2166 case SIOCSIFBRDADDR:
2167 case SIOCGIFNETMASK:
2168 case SIOCSIFNETMASK:
2169 case SIOCGIFDSTADDR:
2170 case SIOCSIFDSTADDR:
2172 return inet_dgram_ops.ioctl(sock, cmd, arg);
2176 return -ENOIOCTLCMD;
2181 static unsigned int packet_poll(struct file *file, struct socket *sock,
2184 struct sock *sk = sock->sk;
2185 struct packet_sock *po = pkt_sk(sk);
2186 unsigned int mask = datagram_poll(file, sock, wait);
2188 spin_lock_bh(&sk->sk_receive_queue.lock);
2189 if (po->rx_ring.pg_vec) {
2190 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2191 mask |= POLLIN | POLLRDNORM;
2193 spin_unlock_bh(&sk->sk_receive_queue.lock);
2194 spin_lock_bh(&sk->sk_write_queue.lock);
2195 if (po->tx_ring.pg_vec) {
2196 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2197 mask |= POLLOUT | POLLWRNORM;
2199 spin_unlock_bh(&sk->sk_write_queue.lock);
2204 /* Dirty? Well, I still did not learn better way to account
2208 static void packet_mm_open(struct vm_area_struct *vma)
2210 struct file *file = vma->vm_file;
2211 struct socket *sock = file->private_data;
2212 struct sock *sk = sock->sk;
2215 atomic_inc(&pkt_sk(sk)->mapped);
2218 static void packet_mm_close(struct vm_area_struct *vma)
2220 struct file *file = vma->vm_file;
2221 struct socket *sock = file->private_data;
2222 struct sock *sk = sock->sk;
2225 atomic_dec(&pkt_sk(sk)->mapped);
2228 static const struct vm_operations_struct packet_mmap_ops = {
2229 .open = packet_mm_open,
2230 .close = packet_mm_close,
2233 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
2237 for (i = 0; i < len; i++) {
2238 if (likely(pg_vec[i]))
2239 free_pages((unsigned long) pg_vec[i], order);
2244 static inline char *alloc_one_pg_vec_page(unsigned long order)
2246 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO | __GFP_NOWARN;
2248 return (char *) __get_free_pages(gfp_flags, order);
2251 static char **alloc_pg_vec(struct tpacket_req *req, int order)
2253 unsigned int block_nr = req->tp_block_nr;
2257 pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
2258 if (unlikely(!pg_vec))
2261 for (i = 0; i < block_nr; i++) {
2262 pg_vec[i] = alloc_one_pg_vec_page(order);
2263 if (unlikely(!pg_vec[i]))
2264 goto out_free_pgvec;
2271 free_pg_vec(pg_vec, order, block_nr);
2276 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2277 int closing, int tx_ring)
2279 char **pg_vec = NULL;
2280 struct packet_sock *po = pkt_sk(sk);
2281 int was_running, order = 0;
2282 struct packet_ring_buffer *rb;
2283 struct sk_buff_head *rb_queue;
2287 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2288 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2292 if (atomic_read(&po->mapped))
2294 if (atomic_read(&rb->pending))
2298 if (req->tp_block_nr) {
2299 /* Sanity tests and some calculations */
2301 if (unlikely(rb->pg_vec))
2304 switch (po->tp_version) {
2306 po->tp_hdrlen = TPACKET_HDRLEN;
2309 po->tp_hdrlen = TPACKET2_HDRLEN;
2314 if (unlikely((int)req->tp_block_size <= 0))
2316 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2318 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2321 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2324 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2325 if (unlikely(rb->frames_per_block <= 0))
2327 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2332 order = get_order(req->tp_block_size);
2333 pg_vec = alloc_pg_vec(req, order);
2334 if (unlikely(!pg_vec))
2340 if (unlikely(req->tp_frame_nr))
2346 /* Detach socket from network */
2347 spin_lock(&po->bind_lock);
2348 was_running = po->running;
2351 __dev_remove_pack(&po->prot_hook);
2356 spin_unlock(&po->bind_lock);
2361 mutex_lock(&po->pg_vec_lock);
2362 if (closing || atomic_read(&po->mapped) == 0) {
2364 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
2365 spin_lock_bh(&rb_queue->lock);
2366 pg_vec = XC(rb->pg_vec, pg_vec);
2367 rb->frame_max = (req->tp_frame_nr - 1);
2369 rb->frame_size = req->tp_frame_size;
2370 spin_unlock_bh(&rb_queue->lock);
2372 order = XC(rb->pg_vec_order, order);
2373 req->tp_block_nr = XC(rb->pg_vec_len, req->tp_block_nr);
2375 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2376 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2377 tpacket_rcv : packet_rcv;
2378 skb_queue_purge(rb_queue);
2380 if (atomic_read(&po->mapped))
2381 pr_err("packet_mmap: vma is busy: %d\n",
2382 atomic_read(&po->mapped));
2384 mutex_unlock(&po->pg_vec_lock);
2386 spin_lock(&po->bind_lock);
2387 if (was_running && !po->running) {
2391 dev_add_pack(&po->prot_hook);
2393 spin_unlock(&po->bind_lock);
2398 free_pg_vec(pg_vec, order, req->tp_block_nr);
2403 static int packet_mmap(struct file *file, struct socket *sock,
2404 struct vm_area_struct *vma)
2406 struct sock *sk = sock->sk;
2407 struct packet_sock *po = pkt_sk(sk);
2408 unsigned long size, expected_size;
2409 struct packet_ring_buffer *rb;
2410 unsigned long start;
2417 mutex_lock(&po->pg_vec_lock);
2420 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2422 expected_size += rb->pg_vec_len
2428 if (expected_size == 0)
2431 size = vma->vm_end - vma->vm_start;
2432 if (size != expected_size)
2435 start = vma->vm_start;
2436 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2437 if (rb->pg_vec == NULL)
2440 for (i = 0; i < rb->pg_vec_len; i++) {
2441 struct page *page = virt_to_page(rb->pg_vec[i]);
2444 for (pg_num = 0; pg_num < rb->pg_vec_pages;
2446 err = vm_insert_page(vma, start, page);
2454 atomic_inc(&po->mapped);
2455 vma->vm_ops = &packet_mmap_ops;
2459 mutex_unlock(&po->pg_vec_lock);
2463 static const struct proto_ops packet_ops_spkt = {
2464 .family = PF_PACKET,
2465 .owner = THIS_MODULE,
2466 .release = packet_release,
2467 .bind = packet_bind_spkt,
2468 .connect = sock_no_connect,
2469 .socketpair = sock_no_socketpair,
2470 .accept = sock_no_accept,
2471 .getname = packet_getname_spkt,
2472 .poll = datagram_poll,
2473 .ioctl = packet_ioctl,
2474 .listen = sock_no_listen,
2475 .shutdown = sock_no_shutdown,
2476 .setsockopt = sock_no_setsockopt,
2477 .getsockopt = sock_no_getsockopt,
2478 .sendmsg = packet_sendmsg_spkt,
2479 .recvmsg = packet_recvmsg,
2480 .mmap = sock_no_mmap,
2481 .sendpage = sock_no_sendpage,
2484 static const struct proto_ops packet_ops = {
2485 .family = PF_PACKET,
2486 .owner = THIS_MODULE,
2487 .release = packet_release,
2488 .bind = packet_bind,
2489 .connect = sock_no_connect,
2490 .socketpair = sock_no_socketpair,
2491 .accept = sock_no_accept,
2492 .getname = packet_getname,
2493 .poll = packet_poll,
2494 .ioctl = packet_ioctl,
2495 .listen = sock_no_listen,
2496 .shutdown = sock_no_shutdown,
2497 .setsockopt = packet_setsockopt,
2498 .getsockopt = packet_getsockopt,
2499 .sendmsg = packet_sendmsg,
2500 .recvmsg = packet_recvmsg,
2501 .mmap = packet_mmap,
2502 .sendpage = sock_no_sendpage,
2505 static const struct net_proto_family packet_family_ops = {
2506 .family = PF_PACKET,
2507 .create = packet_create,
2508 .owner = THIS_MODULE,
2511 static struct notifier_block packet_netdev_notifier = {
2512 .notifier_call = packet_notifier,
2515 #ifdef CONFIG_PROC_FS
2517 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2520 struct net *net = seq_file_net(seq);
2523 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2526 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2528 struct net *net = seq_file_net(seq);
2529 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2532 static void packet_seq_stop(struct seq_file *seq, void *v)
2538 static int packet_seq_show(struct seq_file *seq, void *v)
2540 if (v == SEQ_START_TOKEN)
2541 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2543 struct sock *s = sk_entry(v);
2544 const struct packet_sock *po = pkt_sk(s);
2547 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2549 atomic_read(&s->sk_refcnt),
2554 atomic_read(&s->sk_rmem_alloc),
2562 static const struct seq_operations packet_seq_ops = {
2563 .start = packet_seq_start,
2564 .next = packet_seq_next,
2565 .stop = packet_seq_stop,
2566 .show = packet_seq_show,
2569 static int packet_seq_open(struct inode *inode, struct file *file)
2571 return seq_open_net(inode, file, &packet_seq_ops,
2572 sizeof(struct seq_net_private));
2575 static const struct file_operations packet_seq_fops = {
2576 .owner = THIS_MODULE,
2577 .open = packet_seq_open,
2579 .llseek = seq_lseek,
2580 .release = seq_release_net,
2585 static int __net_init packet_net_init(struct net *net)
2587 spin_lock_init(&net->packet.sklist_lock);
2588 INIT_HLIST_HEAD(&net->packet.sklist);
2590 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2596 static void __net_exit packet_net_exit(struct net *net)
2598 proc_net_remove(net, "packet");
2601 static struct pernet_operations packet_net_ops = {
2602 .init = packet_net_init,
2603 .exit = packet_net_exit,
2607 static void __exit packet_exit(void)
2609 unregister_netdevice_notifier(&packet_netdev_notifier);
2610 unregister_pernet_subsys(&packet_net_ops);
2611 sock_unregister(PF_PACKET);
2612 proto_unregister(&packet_proto);
2615 static int __init packet_init(void)
2617 int rc = proto_register(&packet_proto, 0);
2622 sock_register(&packet_family_ops);
2623 register_pernet_subsys(&packet_net_ops);
2624 register_netdevice_notifier(&packet_netdev_notifier);
2629 module_init(packet_init);
2630 module_exit(packet_exit);
2631 MODULE_LICENSE("GPL");
2632 MODULE_ALIAS_NETPROTO(PF_PACKET);