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 struct packet_type prot_hook;
192 spinlock_t bind_lock;
193 struct mutex pg_vec_lock;
194 unsigned int running:1, /* prot_hook is attached*/
197 int ifindex; /* bound device */
199 struct packet_mclist *mclist;
200 #ifdef CONFIG_PACKET_MMAP
202 enum tpacket_versions tp_version;
203 unsigned int tp_hdrlen;
204 unsigned int tp_reserve;
205 unsigned int tp_loss:1;
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_queue_tail(&sk->sk_receive_queue, skb);
631 spin_unlock(&sk->sk_receive_queue.lock);
632 sk->sk_data_ready(sk, skb->len);
636 spin_lock(&sk->sk_receive_queue.lock);
637 po->stats.tp_drops++;
638 spin_unlock(&sk->sk_receive_queue.lock);
641 if (skb_head != skb->data && skb_shared(skb)) {
642 skb->data = skb_head;
650 #ifdef CONFIG_PACKET_MMAP
651 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
652 struct packet_type *pt, struct net_device *orig_dev)
655 struct packet_sock *po;
656 struct sockaddr_ll *sll;
658 struct tpacket_hdr *h1;
659 struct tpacket2_hdr *h2;
662 u8 *skb_head = skb->data;
663 int skb_len = skb->len;
664 unsigned int snaplen, res;
665 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
666 unsigned short macoff, netoff, hdrlen;
667 struct sk_buff *copy_skb = NULL;
671 if (skb->pkt_type == PACKET_LOOPBACK)
674 sk = pt->af_packet_priv;
677 if (dev_net(dev) != sock_net(sk))
680 if (dev->header_ops) {
681 if (sk->sk_type != SOCK_DGRAM)
682 skb_push(skb, skb->data - skb_mac_header(skb));
683 else if (skb->pkt_type == PACKET_OUTGOING) {
684 /* Special case: outgoing packets have ll header at head */
685 skb_pull(skb, skb_network_offset(skb));
689 if (skb->ip_summed == CHECKSUM_PARTIAL)
690 status |= TP_STATUS_CSUMNOTREADY;
694 res = run_filter(skb, sk, snaplen);
700 if (sk->sk_type == SOCK_DGRAM) {
701 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
704 unsigned maclen = skb_network_offset(skb);
705 netoff = TPACKET_ALIGN(po->tp_hdrlen +
706 (maclen < 16 ? 16 : maclen)) +
708 macoff = netoff - maclen;
711 if (macoff + snaplen > po->rx_ring.frame_size) {
712 if (po->copy_thresh &&
713 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
714 (unsigned)sk->sk_rcvbuf) {
715 if (skb_shared(skb)) {
716 copy_skb = skb_clone(skb, GFP_ATOMIC);
718 copy_skb = skb_get(skb);
719 skb_head = skb->data;
722 skb_set_owner_r(copy_skb, sk);
724 snaplen = po->rx_ring.frame_size - macoff;
725 if ((int)snaplen < 0)
729 spin_lock(&sk->sk_receive_queue.lock);
730 h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
733 packet_increment_head(&po->rx_ring);
734 po->stats.tp_packets++;
736 status |= TP_STATUS_COPY;
737 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
739 if (!po->stats.tp_drops)
740 status &= ~TP_STATUS_LOSING;
741 spin_unlock(&sk->sk_receive_queue.lock);
743 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
745 switch (po->tp_version) {
747 h.h1->tp_len = skb->len;
748 h.h1->tp_snaplen = snaplen;
749 h.h1->tp_mac = macoff;
750 h.h1->tp_net = netoff;
751 if (skb->tstamp.tv64)
752 tv = ktime_to_timeval(skb->tstamp);
754 do_gettimeofday(&tv);
755 h.h1->tp_sec = tv.tv_sec;
756 h.h1->tp_usec = tv.tv_usec;
757 hdrlen = sizeof(*h.h1);
760 h.h2->tp_len = skb->len;
761 h.h2->tp_snaplen = snaplen;
762 h.h2->tp_mac = macoff;
763 h.h2->tp_net = netoff;
764 if (skb->tstamp.tv64)
765 ts = ktime_to_timespec(skb->tstamp);
768 h.h2->tp_sec = ts.tv_sec;
769 h.h2->tp_nsec = ts.tv_nsec;
770 h.h2->tp_vlan_tci = skb->vlan_tci;
771 hdrlen = sizeof(*h.h2);
777 sll = h.raw + TPACKET_ALIGN(hdrlen);
778 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
779 sll->sll_family = AF_PACKET;
780 sll->sll_hatype = dev->type;
781 sll->sll_protocol = skb->protocol;
782 sll->sll_pkttype = skb->pkt_type;
783 if (unlikely(po->origdev))
784 sll->sll_ifindex = orig_dev->ifindex;
786 sll->sll_ifindex = dev->ifindex;
788 __packet_set_status(po, h.raw, status);
791 struct page *p_start, *p_end;
792 u8 *h_end = h.raw + macoff + snaplen - 1;
794 p_start = virt_to_page(h.raw);
795 p_end = virt_to_page(h_end);
796 while (p_start <= p_end) {
797 flush_dcache_page(p_start);
802 sk->sk_data_ready(sk, 0);
805 if (skb_head != skb->data && skb_shared(skb)) {
806 skb->data = skb_head;
814 po->stats.tp_drops++;
815 spin_unlock(&sk->sk_receive_queue.lock);
817 sk->sk_data_ready(sk, 0);
822 static void tpacket_destruct_skb(struct sk_buff *skb)
824 struct packet_sock *po = pkt_sk(skb->sk);
829 if (likely(po->tx_ring.pg_vec)) {
830 ph = skb_shinfo(skb)->destructor_arg;
831 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
832 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
833 atomic_dec(&po->tx_ring.pending);
834 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
840 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
841 void *frame, struct net_device *dev, int size_max,
842 __be16 proto, unsigned char *addr)
845 struct tpacket_hdr *h1;
846 struct tpacket2_hdr *h2;
849 int to_write, offset, len, tp_len, nr_frags, len_max;
850 struct socket *sock = po->sk.sk_socket;
857 skb->protocol = proto;
859 skb->priority = po->sk.sk_priority;
860 skb->mark = po->sk.sk_mark;
861 skb_shinfo(skb)->destructor_arg = ph.raw;
863 switch (po->tp_version) {
865 tp_len = ph.h2->tp_len;
868 tp_len = ph.h1->tp_len;
871 if (unlikely(tp_len > size_max)) {
872 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
876 skb_reserve(skb, LL_RESERVED_SPACE(dev));
877 skb_reset_network_header(skb);
879 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
882 if (sock->type == SOCK_DGRAM) {
883 err = dev_hard_header(skb, dev, ntohs(proto), addr,
885 if (unlikely(err < 0))
887 } else if (dev->hard_header_len) {
888 /* net device doesn't like empty head */
889 if (unlikely(tp_len <= dev->hard_header_len)) {
890 pr_err("packet size is too short (%d < %d)\n",
891 tp_len, dev->hard_header_len);
895 skb_push(skb, dev->hard_header_len);
896 err = skb_store_bits(skb, 0, data,
897 dev->hard_header_len);
901 data += dev->hard_header_len;
902 to_write -= dev->hard_header_len;
906 page = virt_to_page(data);
907 offset = offset_in_page(data);
908 len_max = PAGE_SIZE - offset;
909 len = ((to_write > len_max) ? len_max : to_write);
911 skb->data_len = to_write;
912 skb->len += to_write;
913 skb->truesize += to_write;
914 atomic_add(to_write, &po->sk.sk_wmem_alloc);
916 while (likely(to_write)) {
917 nr_frags = skb_shinfo(skb)->nr_frags;
919 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
920 pr_err("Packet exceed the number of skb frags(%lu)\n",
925 flush_dcache_page(page);
927 skb_fill_page_desc(skb,
929 page++, offset, len);
933 len = ((to_write > len_max) ? len_max : to_write);
939 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
943 struct net_device *dev;
945 int ifindex, err, reserve = 0;
947 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
948 int tp_len, size_max;
953 sock = po->sk.sk_socket;
955 mutex_lock(&po->pg_vec_lock);
959 ifindex = po->ifindex;
964 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
966 if (msg->msg_namelen < (saddr->sll_halen
967 + offsetof(struct sockaddr_ll,
970 ifindex = saddr->sll_ifindex;
971 proto = saddr->sll_protocol;
972 addr = saddr->sll_addr;
975 dev = dev_get_by_index(sock_net(&po->sk), ifindex);
977 if (unlikely(dev == NULL))
980 reserve = dev->hard_header_len;
983 if (unlikely(!(dev->flags & IFF_UP)))
986 size_max = po->tx_ring.frame_size
987 - sizeof(struct skb_shared_info)
989 - LL_ALLOCATED_SPACE(dev)
990 - sizeof(struct sockaddr_ll);
992 if (size_max > dev->mtu + reserve)
993 size_max = dev->mtu + reserve;
996 ph = packet_current_frame(po, &po->tx_ring,
997 TP_STATUS_SEND_REQUEST);
999 if (unlikely(ph == NULL)) {
1004 status = TP_STATUS_SEND_REQUEST;
1005 skb = sock_alloc_send_skb(&po->sk,
1006 LL_ALLOCATED_SPACE(dev)
1007 + sizeof(struct sockaddr_ll),
1010 if (unlikely(skb == NULL))
1013 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1016 if (unlikely(tp_len < 0)) {
1018 __packet_set_status(po, ph,
1019 TP_STATUS_AVAILABLE);
1020 packet_increment_head(&po->tx_ring);
1024 status = TP_STATUS_WRONG_FORMAT;
1030 skb->destructor = tpacket_destruct_skb;
1031 __packet_set_status(po, ph, TP_STATUS_SENDING);
1032 atomic_inc(&po->tx_ring.pending);
1034 status = TP_STATUS_SEND_REQUEST;
1035 err = dev_queue_xmit(skb);
1036 if (unlikely(err > 0 && (err = net_xmit_errno(err)) != 0))
1038 packet_increment_head(&po->tx_ring);
1040 } while (likely((ph != NULL) || ((!(msg->msg_flags & MSG_DONTWAIT))
1041 && (atomic_read(&po->tx_ring.pending))))
1048 skb->destructor = sock_wfree;
1049 atomic_dec(&po->tx_ring.pending);
1051 __packet_set_status(po, ph, status);
1056 mutex_unlock(&po->pg_vec_lock);
1061 static int packet_snd(struct socket *sock,
1062 struct msghdr *msg, size_t len)
1064 struct sock *sk = sock->sk;
1065 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1066 struct sk_buff *skb;
1067 struct net_device *dev;
1069 unsigned char *addr;
1070 int ifindex, err, reserve = 0;
1073 * Get and verify the address.
1076 if (saddr == NULL) {
1077 struct packet_sock *po = pkt_sk(sk);
1079 ifindex = po->ifindex;
1084 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1086 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1088 ifindex = saddr->sll_ifindex;
1089 proto = saddr->sll_protocol;
1090 addr = saddr->sll_addr;
1094 dev = dev_get_by_index(sock_net(sk), ifindex);
1098 if (sock->type == SOCK_RAW)
1099 reserve = dev->hard_header_len;
1102 if (!(dev->flags & IFF_UP))
1106 if (len > dev->mtu+reserve)
1109 skb = sock_alloc_send_skb(sk, len + LL_ALLOCATED_SPACE(dev),
1110 msg->msg_flags & MSG_DONTWAIT, &err);
1114 skb_reserve(skb, LL_RESERVED_SPACE(dev));
1115 skb_reset_network_header(skb);
1118 if (sock->type == SOCK_DGRAM &&
1119 dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len) < 0)
1122 /* Returns -EFAULT on error */
1123 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1127 skb->protocol = proto;
1129 skb->priority = sk->sk_priority;
1130 skb->mark = sk->sk_mark;
1136 err = dev_queue_xmit(skb);
1137 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1153 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1154 struct msghdr *msg, size_t len)
1156 #ifdef CONFIG_PACKET_MMAP
1157 struct sock *sk = sock->sk;
1158 struct packet_sock *po = pkt_sk(sk);
1159 if (po->tx_ring.pg_vec)
1160 return tpacket_snd(po, msg);
1163 return packet_snd(sock, msg, len);
1167 * Close a PACKET socket. This is fairly simple. We immediately go
1168 * to 'closed' state and remove our protocol entry in the device list.
1171 static int packet_release(struct socket *sock)
1173 struct sock *sk = sock->sk;
1174 struct packet_sock *po;
1176 #ifdef CONFIG_PACKET_MMAP
1177 struct tpacket_req req;
1186 write_lock_bh(&net->packet.sklist_lock);
1187 sk_del_node_init(sk);
1188 sock_prot_inuse_add(net, sk->sk_prot, -1);
1189 write_unlock_bh(&net->packet.sklist_lock);
1192 * Unhook packet receive handler.
1197 * Remove the protocol hook
1199 dev_remove_pack(&po->prot_hook);
1205 packet_flush_mclist(sk);
1207 #ifdef CONFIG_PACKET_MMAP
1208 memset(&req, 0, sizeof(req));
1210 if (po->rx_ring.pg_vec)
1211 packet_set_ring(sk, &req, 1, 0);
1213 if (po->tx_ring.pg_vec)
1214 packet_set_ring(sk, &req, 1, 1);
1218 * Now the socket is dead. No more input will appear.
1226 skb_queue_purge(&sk->sk_receive_queue);
1227 sk_refcnt_debug_release(sk);
1234 * Attach a packet hook.
1237 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1239 struct packet_sock *po = pkt_sk(sk);
1241 * Detach an existing hook if present.
1246 spin_lock(&po->bind_lock);
1251 spin_unlock(&po->bind_lock);
1252 dev_remove_pack(&po->prot_hook);
1253 spin_lock(&po->bind_lock);
1257 po->prot_hook.type = protocol;
1258 po->prot_hook.dev = dev;
1260 po->ifindex = dev ? dev->ifindex : 0;
1265 if (!dev || (dev->flags & IFF_UP)) {
1266 dev_add_pack(&po->prot_hook);
1270 sk->sk_err = ENETDOWN;
1271 if (!sock_flag(sk, SOCK_DEAD))
1272 sk->sk_error_report(sk);
1276 spin_unlock(&po->bind_lock);
1282 * Bind a packet socket to a device
1285 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1288 struct sock *sk = sock->sk;
1290 struct net_device *dev;
1297 if (addr_len != sizeof(struct sockaddr))
1299 strlcpy(name, uaddr->sa_data, sizeof(name));
1301 dev = dev_get_by_name(sock_net(sk), name);
1303 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1309 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1311 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1312 struct sock *sk = sock->sk;
1313 struct net_device *dev = NULL;
1321 if (addr_len < sizeof(struct sockaddr_ll))
1323 if (sll->sll_family != AF_PACKET)
1326 if (sll->sll_ifindex) {
1328 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1332 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1340 static struct proto packet_proto = {
1342 .owner = THIS_MODULE,
1343 .obj_size = sizeof(struct packet_sock),
1347 * Create a packet of type SOCK_PACKET.
1350 static int packet_create(struct net *net, struct socket *sock, int protocol)
1353 struct packet_sock *po;
1354 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1357 if (!capable(CAP_NET_RAW))
1359 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1360 sock->type != SOCK_PACKET)
1361 return -ESOCKTNOSUPPORT;
1363 sock->state = SS_UNCONNECTED;
1366 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1370 sock->ops = &packet_ops;
1371 if (sock->type == SOCK_PACKET)
1372 sock->ops = &packet_ops_spkt;
1374 sock_init_data(sock, sk);
1377 sk->sk_family = PF_PACKET;
1380 sk->sk_destruct = packet_sock_destruct;
1381 sk_refcnt_debug_inc(sk);
1384 * Attach a protocol block
1387 spin_lock_init(&po->bind_lock);
1388 mutex_init(&po->pg_vec_lock);
1389 po->prot_hook.func = packet_rcv;
1391 if (sock->type == SOCK_PACKET)
1392 po->prot_hook.func = packet_rcv_spkt;
1394 po->prot_hook.af_packet_priv = sk;
1397 po->prot_hook.type = proto;
1398 dev_add_pack(&po->prot_hook);
1403 write_lock_bh(&net->packet.sklist_lock);
1404 sk_add_node(sk, &net->packet.sklist);
1405 sock_prot_inuse_add(net, &packet_proto, 1);
1406 write_unlock_bh(&net->packet.sklist_lock);
1413 * Pull a packet from our receive queue and hand it to the user.
1414 * If necessary we block.
1417 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1418 struct msghdr *msg, size_t len, int flags)
1420 struct sock *sk = sock->sk;
1421 struct sk_buff *skb;
1423 struct sockaddr_ll *sll;
1426 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1430 /* What error should we return now? EUNATTACH? */
1431 if (pkt_sk(sk)->ifindex < 0)
1436 * Call the generic datagram receiver. This handles all sorts
1437 * of horrible races and re-entrancy so we can forget about it
1438 * in the protocol layers.
1440 * Now it will return ENETDOWN, if device have just gone down,
1441 * but then it will block.
1444 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1447 * An error occurred so return it. Because skb_recv_datagram()
1448 * handles the blocking we don't see and worry about blocking
1456 * If the address length field is there to be filled in, we fill
1460 sll = &PACKET_SKB_CB(skb)->sa.ll;
1461 if (sock->type == SOCK_PACKET)
1462 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1464 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1467 * You lose any data beyond the buffer you gave. If it worries a
1468 * user program they can ask the device for its MTU anyway.
1474 msg->msg_flags |= MSG_TRUNC;
1477 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1481 sock_recv_timestamp(msg, sk, skb);
1484 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1487 if (pkt_sk(sk)->auxdata) {
1488 struct tpacket_auxdata aux;
1490 aux.tp_status = TP_STATUS_USER;
1491 if (skb->ip_summed == CHECKSUM_PARTIAL)
1492 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1493 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1494 aux.tp_snaplen = skb->len;
1496 aux.tp_net = skb_network_offset(skb);
1497 aux.tp_vlan_tci = skb->vlan_tci;
1499 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1503 * Free or return the buffer as appropriate. Again this
1504 * hides all the races and re-entrancy issues from us.
1506 err = (flags&MSG_TRUNC) ? skb->len : copied;
1509 skb_free_datagram(sk, skb);
1514 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1515 int *uaddr_len, int peer)
1517 struct net_device *dev;
1518 struct sock *sk = sock->sk;
1523 uaddr->sa_family = AF_PACKET;
1524 dev = dev_get_by_index(sock_net(sk), pkt_sk(sk)->ifindex);
1526 strlcpy(uaddr->sa_data, dev->name, 15);
1529 memset(uaddr->sa_data, 0, 14);
1530 *uaddr_len = sizeof(*uaddr);
1535 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1536 int *uaddr_len, int peer)
1538 struct net_device *dev;
1539 struct sock *sk = sock->sk;
1540 struct packet_sock *po = pkt_sk(sk);
1541 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1546 sll->sll_family = AF_PACKET;
1547 sll->sll_ifindex = po->ifindex;
1548 sll->sll_protocol = po->num;
1549 dev = dev_get_by_index(sock_net(sk), po->ifindex);
1551 sll->sll_hatype = dev->type;
1552 sll->sll_halen = dev->addr_len;
1553 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1556 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1559 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1564 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1568 case PACKET_MR_MULTICAST:
1570 return dev_mc_add(dev, i->addr, i->alen, 0);
1572 return dev_mc_delete(dev, i->addr, i->alen, 0);
1574 case PACKET_MR_PROMISC:
1575 return dev_set_promiscuity(dev, what);
1577 case PACKET_MR_ALLMULTI:
1578 return dev_set_allmulti(dev, what);
1580 case PACKET_MR_UNICAST:
1582 return dev_unicast_add(dev, i->addr);
1584 return dev_unicast_delete(dev, i->addr);
1592 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1594 for ( ; i; i = i->next) {
1595 if (i->ifindex == dev->ifindex)
1596 packet_dev_mc(dev, i, what);
1600 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1602 struct packet_sock *po = pkt_sk(sk);
1603 struct packet_mclist *ml, *i;
1604 struct net_device *dev;
1610 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1615 if (mreq->mr_alen > dev->addr_len)
1619 i = kmalloc(sizeof(*i), GFP_KERNEL);
1624 for (ml = po->mclist; ml; ml = ml->next) {
1625 if (ml->ifindex == mreq->mr_ifindex &&
1626 ml->type == mreq->mr_type &&
1627 ml->alen == mreq->mr_alen &&
1628 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1630 /* Free the new element ... */
1636 i->type = mreq->mr_type;
1637 i->ifindex = mreq->mr_ifindex;
1638 i->alen = mreq->mr_alen;
1639 memcpy(i->addr, mreq->mr_address, i->alen);
1641 i->next = po->mclist;
1643 err = packet_dev_mc(dev, i, 1);
1645 po->mclist = i->next;
1654 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1656 struct packet_mclist *ml, **mlp;
1660 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1661 if (ml->ifindex == mreq->mr_ifindex &&
1662 ml->type == mreq->mr_type &&
1663 ml->alen == mreq->mr_alen &&
1664 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1665 if (--ml->count == 0) {
1666 struct net_device *dev;
1668 dev = dev_get_by_index(sock_net(sk), ml->ifindex);
1670 packet_dev_mc(dev, ml, -1);
1680 return -EADDRNOTAVAIL;
1683 static void packet_flush_mclist(struct sock *sk)
1685 struct packet_sock *po = pkt_sk(sk);
1686 struct packet_mclist *ml;
1692 while ((ml = po->mclist) != NULL) {
1693 struct net_device *dev;
1695 po->mclist = ml->next;
1696 dev = dev_get_by_index(sock_net(sk), ml->ifindex);
1698 packet_dev_mc(dev, ml, -1);
1707 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1709 struct sock *sk = sock->sk;
1710 struct packet_sock *po = pkt_sk(sk);
1713 if (level != SOL_PACKET)
1714 return -ENOPROTOOPT;
1717 case PACKET_ADD_MEMBERSHIP:
1718 case PACKET_DROP_MEMBERSHIP:
1720 struct packet_mreq_max mreq;
1722 memset(&mreq, 0, sizeof(mreq));
1723 if (len < sizeof(struct packet_mreq))
1725 if (len > sizeof(mreq))
1727 if (copy_from_user(&mreq, optval, len))
1729 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1731 if (optname == PACKET_ADD_MEMBERSHIP)
1732 ret = packet_mc_add(sk, &mreq);
1734 ret = packet_mc_drop(sk, &mreq);
1738 #ifdef CONFIG_PACKET_MMAP
1739 case PACKET_RX_RING:
1740 case PACKET_TX_RING:
1742 struct tpacket_req req;
1744 if (optlen < sizeof(req))
1746 if (copy_from_user(&req, optval, sizeof(req)))
1748 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1750 case PACKET_COPY_THRESH:
1754 if (optlen != sizeof(val))
1756 if (copy_from_user(&val, optval, sizeof(val)))
1759 pkt_sk(sk)->copy_thresh = val;
1762 case PACKET_VERSION:
1766 if (optlen != sizeof(val))
1768 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1770 if (copy_from_user(&val, optval, sizeof(val)))
1775 po->tp_version = val;
1781 case PACKET_RESERVE:
1785 if (optlen != sizeof(val))
1787 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1789 if (copy_from_user(&val, optval, sizeof(val)))
1791 po->tp_reserve = val;
1798 if (optlen != sizeof(val))
1800 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1802 if (copy_from_user(&val, optval, sizeof(val)))
1804 po->tp_loss = !!val;
1808 case PACKET_AUXDATA:
1812 if (optlen < sizeof(val))
1814 if (copy_from_user(&val, optval, sizeof(val)))
1817 po->auxdata = !!val;
1820 case PACKET_ORIGDEV:
1824 if (optlen < sizeof(val))
1826 if (copy_from_user(&val, optval, sizeof(val)))
1829 po->origdev = !!val;
1833 return -ENOPROTOOPT;
1837 static int packet_getsockopt(struct socket *sock, int level, int optname,
1838 char __user *optval, int __user *optlen)
1842 struct sock *sk = sock->sk;
1843 struct packet_sock *po = pkt_sk(sk);
1845 struct tpacket_stats st;
1847 if (level != SOL_PACKET)
1848 return -ENOPROTOOPT;
1850 if (get_user(len, optlen))
1857 case PACKET_STATISTICS:
1858 if (len > sizeof(struct tpacket_stats))
1859 len = sizeof(struct tpacket_stats);
1860 spin_lock_bh(&sk->sk_receive_queue.lock);
1862 memset(&po->stats, 0, sizeof(st));
1863 spin_unlock_bh(&sk->sk_receive_queue.lock);
1864 st.tp_packets += st.tp_drops;
1868 case PACKET_AUXDATA:
1869 if (len > sizeof(int))
1875 case PACKET_ORIGDEV:
1876 if (len > sizeof(int))
1882 #ifdef CONFIG_PACKET_MMAP
1883 case PACKET_VERSION:
1884 if (len > sizeof(int))
1886 val = po->tp_version;
1890 if (len > sizeof(int))
1892 if (copy_from_user(&val, optval, len))
1896 val = sizeof(struct tpacket_hdr);
1899 val = sizeof(struct tpacket2_hdr);
1906 case PACKET_RESERVE:
1907 if (len > sizeof(unsigned int))
1908 len = sizeof(unsigned int);
1909 val = po->tp_reserve;
1913 if (len > sizeof(unsigned int))
1914 len = sizeof(unsigned int);
1920 return -ENOPROTOOPT;
1923 if (put_user(len, optlen))
1925 if (copy_to_user(optval, data, len))
1931 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1934 struct hlist_node *node;
1935 struct net_device *dev = data;
1936 struct net *net = dev_net(dev);
1938 read_lock(&net->packet.sklist_lock);
1939 sk_for_each(sk, node, &net->packet.sklist) {
1940 struct packet_sock *po = pkt_sk(sk);
1943 case NETDEV_UNREGISTER:
1945 packet_dev_mclist(dev, po->mclist, -1);
1949 if (dev->ifindex == po->ifindex) {
1950 spin_lock(&po->bind_lock);
1952 __dev_remove_pack(&po->prot_hook);
1955 sk->sk_err = ENETDOWN;
1956 if (!sock_flag(sk, SOCK_DEAD))
1957 sk->sk_error_report(sk);
1959 if (msg == NETDEV_UNREGISTER) {
1961 po->prot_hook.dev = NULL;
1963 spin_unlock(&po->bind_lock);
1967 spin_lock(&po->bind_lock);
1968 if (dev->ifindex == po->ifindex && po->num &&
1970 dev_add_pack(&po->prot_hook);
1974 spin_unlock(&po->bind_lock);
1978 read_unlock(&net->packet.sklist_lock);
1983 static int packet_ioctl(struct socket *sock, unsigned int cmd,
1986 struct sock *sk = sock->sk;
1991 int amount = sk_wmem_alloc_get(sk);
1993 return put_user(amount, (int __user *)arg);
1997 struct sk_buff *skb;
2000 spin_lock_bh(&sk->sk_receive_queue.lock);
2001 skb = skb_peek(&sk->sk_receive_queue);
2004 spin_unlock_bh(&sk->sk_receive_queue.lock);
2005 return put_user(amount, (int __user *)arg);
2008 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2010 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2020 case SIOCGIFBRDADDR:
2021 case SIOCSIFBRDADDR:
2022 case SIOCGIFNETMASK:
2023 case SIOCSIFNETMASK:
2024 case SIOCGIFDSTADDR:
2025 case SIOCSIFDSTADDR:
2027 if (!net_eq(sock_net(sk), &init_net))
2028 return -ENOIOCTLCMD;
2029 return inet_dgram_ops.ioctl(sock, cmd, arg);
2033 return -ENOIOCTLCMD;
2038 #ifndef CONFIG_PACKET_MMAP
2039 #define packet_mmap sock_no_mmap
2040 #define packet_poll datagram_poll
2043 static unsigned int packet_poll(struct file *file, struct socket *sock,
2046 struct sock *sk = sock->sk;
2047 struct packet_sock *po = pkt_sk(sk);
2048 unsigned int mask = datagram_poll(file, sock, wait);
2050 spin_lock_bh(&sk->sk_receive_queue.lock);
2051 if (po->rx_ring.pg_vec) {
2052 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2053 mask |= POLLIN | POLLRDNORM;
2055 spin_unlock_bh(&sk->sk_receive_queue.lock);
2056 spin_lock_bh(&sk->sk_write_queue.lock);
2057 if (po->tx_ring.pg_vec) {
2058 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2059 mask |= POLLOUT | POLLWRNORM;
2061 spin_unlock_bh(&sk->sk_write_queue.lock);
2066 /* Dirty? Well, I still did not learn better way to account
2070 static void packet_mm_open(struct vm_area_struct *vma)
2072 struct file *file = vma->vm_file;
2073 struct socket *sock = file->private_data;
2074 struct sock *sk = sock->sk;
2077 atomic_inc(&pkt_sk(sk)->mapped);
2080 static void packet_mm_close(struct vm_area_struct *vma)
2082 struct file *file = vma->vm_file;
2083 struct socket *sock = file->private_data;
2084 struct sock *sk = sock->sk;
2087 atomic_dec(&pkt_sk(sk)->mapped);
2090 static const struct vm_operations_struct packet_mmap_ops = {
2091 .open = packet_mm_open,
2092 .close = packet_mm_close,
2095 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
2099 for (i = 0; i < len; i++) {
2100 if (likely(pg_vec[i]))
2101 free_pages((unsigned long) pg_vec[i], order);
2106 static inline char *alloc_one_pg_vec_page(unsigned long order)
2108 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO | __GFP_NOWARN;
2110 return (char *) __get_free_pages(gfp_flags, order);
2113 static char **alloc_pg_vec(struct tpacket_req *req, int order)
2115 unsigned int block_nr = req->tp_block_nr;
2119 pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
2120 if (unlikely(!pg_vec))
2123 for (i = 0; i < block_nr; i++) {
2124 pg_vec[i] = alloc_one_pg_vec_page(order);
2125 if (unlikely(!pg_vec[i]))
2126 goto out_free_pgvec;
2133 free_pg_vec(pg_vec, order, block_nr);
2138 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2139 int closing, int tx_ring)
2141 char **pg_vec = NULL;
2142 struct packet_sock *po = pkt_sk(sk);
2143 int was_running, order = 0;
2144 struct packet_ring_buffer *rb;
2145 struct sk_buff_head *rb_queue;
2149 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2150 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2154 if (atomic_read(&po->mapped))
2156 if (atomic_read(&rb->pending))
2160 if (req->tp_block_nr) {
2161 /* Sanity tests and some calculations */
2163 if (unlikely(rb->pg_vec))
2166 switch (po->tp_version) {
2168 po->tp_hdrlen = TPACKET_HDRLEN;
2171 po->tp_hdrlen = TPACKET2_HDRLEN;
2176 if (unlikely((int)req->tp_block_size <= 0))
2178 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2180 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2183 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2186 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2187 if (unlikely(rb->frames_per_block <= 0))
2189 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2194 order = get_order(req->tp_block_size);
2195 pg_vec = alloc_pg_vec(req, order);
2196 if (unlikely(!pg_vec))
2202 if (unlikely(req->tp_frame_nr))
2208 /* Detach socket from network */
2209 spin_lock(&po->bind_lock);
2210 was_running = po->running;
2213 __dev_remove_pack(&po->prot_hook);
2218 spin_unlock(&po->bind_lock);
2223 mutex_lock(&po->pg_vec_lock);
2224 if (closing || atomic_read(&po->mapped) == 0) {
2226 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
2227 spin_lock_bh(&rb_queue->lock);
2228 pg_vec = XC(rb->pg_vec, pg_vec);
2229 rb->frame_max = (req->tp_frame_nr - 1);
2231 rb->frame_size = req->tp_frame_size;
2232 spin_unlock_bh(&rb_queue->lock);
2234 order = XC(rb->pg_vec_order, order);
2235 req->tp_block_nr = XC(rb->pg_vec_len, req->tp_block_nr);
2237 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2238 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2239 tpacket_rcv : packet_rcv;
2240 skb_queue_purge(rb_queue);
2242 if (atomic_read(&po->mapped))
2243 pr_err("packet_mmap: vma is busy: %d\n",
2244 atomic_read(&po->mapped));
2246 mutex_unlock(&po->pg_vec_lock);
2248 spin_lock(&po->bind_lock);
2249 if (was_running && !po->running) {
2253 dev_add_pack(&po->prot_hook);
2255 spin_unlock(&po->bind_lock);
2260 free_pg_vec(pg_vec, order, req->tp_block_nr);
2265 static int packet_mmap(struct file *file, struct socket *sock,
2266 struct vm_area_struct *vma)
2268 struct sock *sk = sock->sk;
2269 struct packet_sock *po = pkt_sk(sk);
2270 unsigned long size, expected_size;
2271 struct packet_ring_buffer *rb;
2272 unsigned long start;
2279 mutex_lock(&po->pg_vec_lock);
2282 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2284 expected_size += rb->pg_vec_len
2290 if (expected_size == 0)
2293 size = vma->vm_end - vma->vm_start;
2294 if (size != expected_size)
2297 start = vma->vm_start;
2298 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2299 if (rb->pg_vec == NULL)
2302 for (i = 0; i < rb->pg_vec_len; i++) {
2303 struct page *page = virt_to_page(rb->pg_vec[i]);
2306 for (pg_num = 0; pg_num < rb->pg_vec_pages;
2308 err = vm_insert_page(vma, start, page);
2316 atomic_inc(&po->mapped);
2317 vma->vm_ops = &packet_mmap_ops;
2321 mutex_unlock(&po->pg_vec_lock);
2327 static const struct proto_ops packet_ops_spkt = {
2328 .family = PF_PACKET,
2329 .owner = THIS_MODULE,
2330 .release = packet_release,
2331 .bind = packet_bind_spkt,
2332 .connect = sock_no_connect,
2333 .socketpair = sock_no_socketpair,
2334 .accept = sock_no_accept,
2335 .getname = packet_getname_spkt,
2336 .poll = datagram_poll,
2337 .ioctl = packet_ioctl,
2338 .listen = sock_no_listen,
2339 .shutdown = sock_no_shutdown,
2340 .setsockopt = sock_no_setsockopt,
2341 .getsockopt = sock_no_getsockopt,
2342 .sendmsg = packet_sendmsg_spkt,
2343 .recvmsg = packet_recvmsg,
2344 .mmap = sock_no_mmap,
2345 .sendpage = sock_no_sendpage,
2348 static const struct proto_ops packet_ops = {
2349 .family = PF_PACKET,
2350 .owner = THIS_MODULE,
2351 .release = packet_release,
2352 .bind = packet_bind,
2353 .connect = sock_no_connect,
2354 .socketpair = sock_no_socketpair,
2355 .accept = sock_no_accept,
2356 .getname = packet_getname,
2357 .poll = packet_poll,
2358 .ioctl = packet_ioctl,
2359 .listen = sock_no_listen,
2360 .shutdown = sock_no_shutdown,
2361 .setsockopt = packet_setsockopt,
2362 .getsockopt = packet_getsockopt,
2363 .sendmsg = packet_sendmsg,
2364 .recvmsg = packet_recvmsg,
2365 .mmap = packet_mmap,
2366 .sendpage = sock_no_sendpage,
2369 static const struct net_proto_family packet_family_ops = {
2370 .family = PF_PACKET,
2371 .create = packet_create,
2372 .owner = THIS_MODULE,
2375 static struct notifier_block packet_netdev_notifier = {
2376 .notifier_call = packet_notifier,
2379 #ifdef CONFIG_PROC_FS
2380 static inline struct sock *packet_seq_idx(struct net *net, loff_t off)
2383 struct hlist_node *node;
2385 sk_for_each(s, node, &net->packet.sklist) {
2392 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2393 __acquires(seq_file_net(seq)->packet.sklist_lock)
2395 struct net *net = seq_file_net(seq);
2396 read_lock(&net->packet.sklist_lock);
2397 return *pos ? packet_seq_idx(net, *pos - 1) : SEQ_START_TOKEN;
2400 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2402 struct net *net = seq_file_net(seq);
2404 return (v == SEQ_START_TOKEN)
2405 ? sk_head(&net->packet.sklist)
2406 : sk_next((struct sock *)v) ;
2409 static void packet_seq_stop(struct seq_file *seq, void *v)
2410 __releases(seq_file_net(seq)->packet.sklist_lock)
2412 struct net *net = seq_file_net(seq);
2413 read_unlock(&net->packet.sklist_lock);
2416 static int packet_seq_show(struct seq_file *seq, void *v)
2418 if (v == SEQ_START_TOKEN)
2419 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2422 const struct packet_sock *po = pkt_sk(s);
2425 "%p %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2427 atomic_read(&s->sk_refcnt),
2432 atomic_read(&s->sk_rmem_alloc),
2440 static const struct seq_operations packet_seq_ops = {
2441 .start = packet_seq_start,
2442 .next = packet_seq_next,
2443 .stop = packet_seq_stop,
2444 .show = packet_seq_show,
2447 static int packet_seq_open(struct inode *inode, struct file *file)
2449 return seq_open_net(inode, file, &packet_seq_ops,
2450 sizeof(struct seq_net_private));
2453 static const struct file_operations packet_seq_fops = {
2454 .owner = THIS_MODULE,
2455 .open = packet_seq_open,
2457 .llseek = seq_lseek,
2458 .release = seq_release_net,
2463 static int packet_net_init(struct net *net)
2465 rwlock_init(&net->packet.sklist_lock);
2466 INIT_HLIST_HEAD(&net->packet.sklist);
2468 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2474 static void packet_net_exit(struct net *net)
2476 proc_net_remove(net, "packet");
2479 static struct pernet_operations packet_net_ops = {
2480 .init = packet_net_init,
2481 .exit = packet_net_exit,
2485 static void __exit packet_exit(void)
2487 unregister_netdevice_notifier(&packet_netdev_notifier);
2488 unregister_pernet_subsys(&packet_net_ops);
2489 sock_unregister(PF_PACKET);
2490 proto_unregister(&packet_proto);
2493 static int __init packet_init(void)
2495 int rc = proto_register(&packet_proto, 0);
2500 sock_register(&packet_family_ops);
2501 register_pernet_subsys(&packet_net_ops);
2502 register_netdevice_notifier(&packet_netdev_notifier);
2507 module_init(packet_init);
2508 module_exit(packet_exit);
2509 MODULE_LICENSE("GPL");
2510 MODULE_ALIAS_NETPROTO(PF_PACKET);