ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / net / packet / af_packet.c
blobfafb96830e78b9d6fd9469a1a100a562bbb22961
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
12 * Fixes:
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
41 * and packet_mreq.
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>
52 #include <linux/mm.h>
53 #include <linux/capability.h>
54 #include <linux/fcntl.h>
55 #include <linux/socket.h>
56 #include <linux/in.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 <linux/vmalloc.h>
65 #include <net/net_namespace.h>
66 #include <net/ip.h>
67 #include <net/protocol.h>
68 #include <linux/skbuff.h>
69 #include <net/sock.h>
70 #include <linux/errno.h>
71 #include <linux/timer.h>
72 #include <asm/system.h>
73 #include <asm/uaccess.h>
74 #include <asm/ioctls.h>
75 #include <asm/page.h>
76 #include <asm/cacheflush.h>
77 #include <asm/io.h>
78 #include <linux/proc_fs.h>
79 #include <linux/seq_file.h>
80 #include <linux/poll.h>
81 #include <linux/module.h>
82 #include <linux/init.h>
83 #include <linux/mutex.h>
84 #include <linux/if_vlan.h>
85 #include <linux/virtio_net.h>
86 #include <linux/errqueue.h>
87 #include <linux/net_tstamp.h>
89 #ifdef CONFIG_INET
90 #include <net/inet_common.h>
91 #endif
94 Assumptions:
95 - if device has no dev->hard_header routine, it adds and removes ll header
96 inside itself. In this case ll header is invisible outside of device,
97 but higher levels still should reserve dev->hard_header_len.
98 Some devices are enough clever to reallocate skb, when header
99 will not fit to reserved space (tunnel), another ones are silly
100 (PPP).
101 - packet socket receives packets with pulled ll header,
102 so that SOCK_RAW should push it back.
104 On receive:
105 -----------
107 Incoming, dev->hard_header!=NULL
108 mac_header -> ll header
109 data -> data
111 Outgoing, dev->hard_header!=NULL
112 mac_header -> ll header
113 data -> ll header
115 Incoming, dev->hard_header==NULL
116 mac_header -> UNKNOWN position. It is very likely, that it points to ll
117 header. PPP makes it, that is wrong, because introduce
118 assymetry between rx and tx paths.
119 data -> data
121 Outgoing, dev->hard_header==NULL
122 mac_header -> data. ll header is still not built!
123 data -> data
125 Resume
126 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
129 On transmit:
130 ------------
132 dev->hard_header != NULL
133 mac_header -> ll header
134 data -> ll header
136 dev->hard_header == NULL (ll header is added by device, we cannot control it)
137 mac_header -> data
138 data -> data
140 We should set nh.raw on output to correct posistion,
141 packet classifier depends on it.
144 /* Private packet socket structures. */
146 struct packet_mclist {
147 struct packet_mclist *next;
148 int ifindex;
149 int count;
150 unsigned short type;
151 unsigned short alen;
152 unsigned char addr[MAX_ADDR_LEN];
154 /* identical to struct packet_mreq except it has
155 * a longer address field.
157 struct packet_mreq_max {
158 int mr_ifindex;
159 unsigned short mr_type;
160 unsigned short mr_alen;
161 unsigned char mr_address[MAX_ADDR_LEN];
164 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
165 int closing, int tx_ring);
167 struct pgv {
168 char *buffer;
171 struct packet_ring_buffer {
172 struct pgv *pg_vec;
173 unsigned int head;
174 unsigned int frames_per_block;
175 unsigned int frame_size;
176 unsigned int frame_max;
178 unsigned int pg_vec_order;
179 unsigned int pg_vec_pages;
180 unsigned int pg_vec_len;
182 atomic_t pending;
185 struct packet_sock;
186 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
188 static void packet_flush_mclist(struct sock *sk);
190 struct packet_sock {
191 /* struct sock has to be the first member of packet_sock */
192 struct sock sk;
193 struct tpacket_stats stats;
194 struct packet_ring_buffer rx_ring;
195 struct packet_ring_buffer tx_ring;
196 int copy_thresh;
197 spinlock_t bind_lock;
198 struct mutex pg_vec_lock;
199 unsigned int running:1, /* prot_hook is attached*/
200 auxdata:1,
201 origdev:1,
202 has_vnet_hdr:1;
203 int ifindex; /* bound device */
204 __be16 num;
205 struct packet_mclist *mclist;
206 atomic_t mapped;
207 enum tpacket_versions tp_version;
208 unsigned int tp_hdrlen;
209 unsigned int tp_reserve;
210 unsigned int tp_loss:1;
211 unsigned int tp_tstamp;
212 struct packet_type prot_hook ____cacheline_aligned_in_smp;
215 struct packet_skb_cb {
216 unsigned int origlen;
217 union {
218 struct sockaddr_pkt pkt;
219 struct sockaddr_ll ll;
220 } sa;
223 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
225 static inline __pure struct page *pgv_to_page(void *addr)
227 if (is_vmalloc_addr(addr))
228 return vmalloc_to_page(addr);
229 return virt_to_page(addr);
232 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
234 union {
235 struct tpacket_hdr *h1;
236 struct tpacket2_hdr *h2;
237 void *raw;
238 } h;
240 h.raw = frame;
241 switch (po->tp_version) {
242 case TPACKET_V1:
243 h.h1->tp_status = status;
244 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
245 break;
246 case TPACKET_V2:
247 h.h2->tp_status = status;
248 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
249 break;
250 default:
251 pr_err("TPACKET version not supported\n");
252 BUG();
255 smp_wmb();
258 static int __packet_get_status(struct packet_sock *po, void *frame)
260 union {
261 struct tpacket_hdr *h1;
262 struct tpacket2_hdr *h2;
263 void *raw;
264 } h;
266 smp_rmb();
268 h.raw = frame;
269 switch (po->tp_version) {
270 case TPACKET_V1:
271 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
272 return h.h1->tp_status;
273 case TPACKET_V2:
274 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
275 return h.h2->tp_status;
276 default:
277 pr_err("TPACKET version not supported\n");
278 BUG();
279 return 0;
283 static void *packet_lookup_frame(struct packet_sock *po,
284 struct packet_ring_buffer *rb,
285 unsigned int position,
286 int status)
288 unsigned int pg_vec_pos, frame_offset;
289 union {
290 struct tpacket_hdr *h1;
291 struct tpacket2_hdr *h2;
292 void *raw;
293 } h;
295 pg_vec_pos = position / rb->frames_per_block;
296 frame_offset = position % rb->frames_per_block;
298 h.raw = rb->pg_vec[pg_vec_pos].buffer +
299 (frame_offset * rb->frame_size);
301 if (status != __packet_get_status(po, h.raw))
302 return NULL;
304 return h.raw;
307 static inline void *packet_current_frame(struct packet_sock *po,
308 struct packet_ring_buffer *rb,
309 int status)
311 return packet_lookup_frame(po, rb, rb->head, status);
314 static inline void *packet_previous_frame(struct packet_sock *po,
315 struct packet_ring_buffer *rb,
316 int status)
318 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
319 return packet_lookup_frame(po, rb, previous, status);
322 static inline void packet_increment_head(struct packet_ring_buffer *buff)
324 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
327 static inline struct packet_sock *pkt_sk(struct sock *sk)
329 return (struct packet_sock *)sk;
332 static void packet_sock_destruct(struct sock *sk)
334 skb_queue_purge(&sk->sk_error_queue);
336 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
337 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
339 if (!sock_flag(sk, SOCK_DEAD)) {
340 pr_err("Attempt to release alive packet socket: %p\n", sk);
341 return;
344 sk_refcnt_debug_dec(sk);
348 static const struct proto_ops packet_ops;
350 static const struct proto_ops packet_ops_spkt;
352 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
353 struct packet_type *pt, struct net_device *orig_dev)
355 struct sock *sk;
356 struct sockaddr_pkt *spkt;
359 * When we registered the protocol we saved the socket in the data
360 * field for just this event.
363 sk = pt->af_packet_priv;
366 * Yank back the headers [hope the device set this
367 * right or kerboom...]
369 * Incoming packets have ll header pulled,
370 * push it back.
372 * For outgoing ones skb->data == skb_mac_header(skb)
373 * so that this procedure is noop.
376 if (skb->pkt_type == PACKET_LOOPBACK)
377 goto out;
379 if (!net_eq(dev_net(dev), sock_net(sk)))
380 goto out;
382 skb = skb_share_check(skb, GFP_ATOMIC);
383 if (skb == NULL)
384 goto oom;
386 /* drop any routing info */
387 skb_dst_drop(skb);
389 /* drop conntrack reference */
390 nf_reset(skb);
392 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
394 skb_push(skb, skb->data - skb_mac_header(skb));
397 * The SOCK_PACKET socket receives _all_ frames.
400 spkt->spkt_family = dev->type;
401 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
402 spkt->spkt_protocol = skb->protocol;
405 * Charge the memory to the socket. This is done specifically
406 * to prevent sockets using all the memory up.
409 if (sock_queue_rcv_skb(sk, skb) == 0)
410 return 0;
412 out:
413 kfree_skb(skb);
414 oom:
415 return 0;
420 * Output a raw packet to a device layer. This bypasses all the other
421 * protocol layers and you must therefore supply it with a complete frame
424 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
425 struct msghdr *msg, size_t len)
427 struct sock *sk = sock->sk;
428 struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
429 struct sk_buff *skb = NULL;
430 struct net_device *dev;
431 __be16 proto = 0;
432 int err;
435 * Get and verify the address.
438 if (saddr) {
439 if (msg->msg_namelen < sizeof(struct sockaddr))
440 return -EINVAL;
441 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
442 proto = saddr->spkt_protocol;
443 } else
444 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
447 * Find the device first to size check it
450 saddr->spkt_device[13] = 0;
451 retry:
452 rcu_read_lock();
453 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
454 err = -ENODEV;
455 if (dev == NULL)
456 goto out_unlock;
458 err = -ENETDOWN;
459 if (!(dev->flags & IFF_UP))
460 goto out_unlock;
463 * You may not queue a frame bigger than the mtu. This is the lowest level
464 * raw protocol and you must do your own fragmentation at this level.
467 err = -EMSGSIZE;
468 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
469 goto out_unlock;
471 if (!skb) {
472 size_t reserved = LL_RESERVED_SPACE(dev);
473 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
475 rcu_read_unlock();
476 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
477 if (skb == NULL)
478 return -ENOBUFS;
479 /* FIXME: Save some space for broken drivers that write a hard
480 * header at transmission time by themselves. PPP is the notable
481 * one here. This should really be fixed at the driver level.
483 skb_reserve(skb, reserved);
484 skb_reset_network_header(skb);
486 /* Try to align data part correctly */
487 if (hhlen) {
488 skb->data -= hhlen;
489 skb->tail -= hhlen;
490 if (len < hhlen)
491 skb_reset_network_header(skb);
493 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
494 if (err)
495 goto out_free;
496 goto retry;
499 if (len > (dev->mtu + dev->hard_header_len)) {
500 /* Earlier code assumed this would be a VLAN pkt,
501 * double-check this now that we have the actual
502 * packet in hand.
504 struct ethhdr *ehdr;
505 skb_reset_mac_header(skb);
506 ehdr = eth_hdr(skb);
507 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
508 err = -EMSGSIZE;
509 goto out_unlock;
513 skb->protocol = proto;
514 skb->dev = dev;
515 skb->priority = sk->sk_priority;
516 skb->mark = sk->sk_mark;
517 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
518 if (err < 0)
519 goto out_unlock;
521 dev_queue_xmit(skb);
522 rcu_read_unlock();
523 return len;
525 out_unlock:
526 rcu_read_unlock();
527 out_free:
528 kfree_skb(skb);
529 return err;
532 static inline unsigned int run_filter(const struct sk_buff *skb,
533 const struct sock *sk,
534 unsigned int res)
536 struct sk_filter *filter;
538 rcu_read_lock();
539 filter = rcu_dereference(sk->sk_filter);
540 if (filter != NULL)
541 res = SK_RUN_FILTER(filter, skb);
542 rcu_read_unlock();
544 return res;
548 * This function makes lazy skb cloning in hope that most of packets
549 * are discarded by BPF.
551 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
552 * and skb->cb are mangled. It works because (and until) packets
553 * falling here are owned by current CPU. Output packets are cloned
554 * by dev_queue_xmit_nit(), input packets are processed by net_bh
555 * sequencially, so that if we return skb to original state on exit,
556 * we will not harm anyone.
559 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
560 struct packet_type *pt, struct net_device *orig_dev)
562 struct sock *sk;
563 struct sockaddr_ll *sll;
564 struct packet_sock *po;
565 u8 *skb_head = skb->data;
566 int skb_len = skb->len;
567 unsigned int snaplen, res;
569 if (skb->pkt_type == PACKET_LOOPBACK)
570 goto drop;
572 sk = pt->af_packet_priv;
573 po = pkt_sk(sk);
575 if (!net_eq(dev_net(dev), sock_net(sk)))
576 goto drop;
578 skb->dev = dev;
580 if (dev->header_ops) {
581 /* The device has an explicit notion of ll header,
582 * exported to higher levels.
584 * Otherwise, the device hides details of its frame
585 * structure, so that corresponding packet head is
586 * never delivered to user.
588 if (sk->sk_type != SOCK_DGRAM)
589 skb_push(skb, skb->data - skb_mac_header(skb));
590 else if (skb->pkt_type == PACKET_OUTGOING) {
591 /* Special case: outgoing packets have ll header at head */
592 skb_pull(skb, skb_network_offset(skb));
596 snaplen = skb->len;
598 res = run_filter(skb, sk, snaplen);
599 if (!res)
600 goto drop_n_restore;
601 if (snaplen > res)
602 snaplen = res;
604 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
605 (unsigned)sk->sk_rcvbuf)
606 goto drop_n_acct;
608 if (skb_shared(skb)) {
609 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
610 if (nskb == NULL)
611 goto drop_n_acct;
613 if (skb_head != skb->data) {
614 skb->data = skb_head;
615 skb->len = skb_len;
617 kfree_skb(skb);
618 skb = nskb;
621 BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
622 sizeof(skb->cb));
624 sll = &PACKET_SKB_CB(skb)->sa.ll;
625 sll->sll_family = AF_PACKET;
626 sll->sll_hatype = dev->type;
627 sll->sll_protocol = skb->protocol;
628 sll->sll_pkttype = skb->pkt_type;
629 if (unlikely(po->origdev))
630 sll->sll_ifindex = orig_dev->ifindex;
631 else
632 sll->sll_ifindex = dev->ifindex;
634 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
636 PACKET_SKB_CB(skb)->origlen = skb->len;
638 if (pskb_trim(skb, snaplen))
639 goto drop_n_acct;
641 skb_set_owner_r(skb, sk);
642 skb->dev = NULL;
643 skb_dst_drop(skb);
645 /* drop conntrack reference */
646 nf_reset(skb);
648 spin_lock(&sk->sk_receive_queue.lock);
649 po->stats.tp_packets++;
650 skb->dropcount = atomic_read(&sk->sk_drops);
651 __skb_queue_tail(&sk->sk_receive_queue, skb);
652 spin_unlock(&sk->sk_receive_queue.lock);
653 sk->sk_data_ready(sk, skb->len);
654 return 0;
656 drop_n_acct:
657 spin_lock(&sk->sk_receive_queue.lock);
658 po->stats.tp_drops++;
659 atomic_inc(&sk->sk_drops);
660 spin_unlock(&sk->sk_receive_queue.lock);
662 drop_n_restore:
663 if (skb_head != skb->data && skb_shared(skb)) {
664 skb->data = skb_head;
665 skb->len = skb_len;
667 drop:
668 consume_skb(skb);
669 return 0;
672 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
673 struct packet_type *pt, struct net_device *orig_dev)
675 struct sock *sk;
676 struct packet_sock *po;
677 struct sockaddr_ll *sll;
678 union {
679 struct tpacket_hdr *h1;
680 struct tpacket2_hdr *h2;
681 void *raw;
682 } h;
683 u8 *skb_head = skb->data;
684 int skb_len = skb->len;
685 unsigned int snaplen, res;
686 unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
687 unsigned short macoff, netoff, hdrlen;
688 struct sk_buff *copy_skb = NULL;
689 struct timeval tv;
690 struct timespec ts;
691 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
693 if (skb->pkt_type == PACKET_LOOPBACK)
694 goto drop;
696 sk = pt->af_packet_priv;
697 po = pkt_sk(sk);
699 if (!net_eq(dev_net(dev), sock_net(sk)))
700 goto drop;
702 if (dev->header_ops) {
703 if (sk->sk_type != SOCK_DGRAM)
704 skb_push(skb, skb->data - skb_mac_header(skb));
705 else if (skb->pkt_type == PACKET_OUTGOING) {
706 /* Special case: outgoing packets have ll header at head */
707 skb_pull(skb, skb_network_offset(skb));
711 if (skb->ip_summed == CHECKSUM_PARTIAL)
712 status |= TP_STATUS_CSUMNOTREADY;
714 snaplen = skb->len;
716 res = run_filter(skb, sk, snaplen);
717 if (!res)
718 goto drop_n_restore;
719 if (snaplen > res)
720 snaplen = res;
722 if (sk->sk_type == SOCK_DGRAM) {
723 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
724 po->tp_reserve;
725 } else {
726 unsigned maclen = skb_network_offset(skb);
727 netoff = TPACKET_ALIGN(po->tp_hdrlen +
728 (maclen < 16 ? 16 : maclen)) +
729 po->tp_reserve;
730 macoff = netoff - maclen;
733 if (macoff + snaplen > po->rx_ring.frame_size) {
734 if (po->copy_thresh &&
735 atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
736 (unsigned)sk->sk_rcvbuf) {
737 if (skb_shared(skb)) {
738 copy_skb = skb_clone(skb, GFP_ATOMIC);
739 } else {
740 copy_skb = skb_get(skb);
741 skb_head = skb->data;
743 if (copy_skb)
744 skb_set_owner_r(copy_skb, sk);
746 snaplen = po->rx_ring.frame_size - macoff;
747 if ((int)snaplen < 0)
748 snaplen = 0;
751 spin_lock(&sk->sk_receive_queue.lock);
752 h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
753 if (!h.raw)
754 goto ring_is_full;
755 packet_increment_head(&po->rx_ring);
756 po->stats.tp_packets++;
757 if (copy_skb) {
758 status |= TP_STATUS_COPY;
759 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
761 if (!po->stats.tp_drops)
762 status &= ~TP_STATUS_LOSING;
763 spin_unlock(&sk->sk_receive_queue.lock);
765 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
767 switch (po->tp_version) {
768 case TPACKET_V1:
769 h.h1->tp_len = skb->len;
770 h.h1->tp_snaplen = snaplen;
771 h.h1->tp_mac = macoff;
772 h.h1->tp_net = netoff;
773 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
774 && shhwtstamps->syststamp.tv64)
775 tv = ktime_to_timeval(shhwtstamps->syststamp);
776 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
777 && shhwtstamps->hwtstamp.tv64)
778 tv = ktime_to_timeval(shhwtstamps->hwtstamp);
779 else if (skb->tstamp.tv64)
780 tv = ktime_to_timeval(skb->tstamp);
781 else
782 do_gettimeofday(&tv);
783 h.h1->tp_sec = tv.tv_sec;
784 h.h1->tp_usec = tv.tv_usec;
785 hdrlen = sizeof(*h.h1);
786 break;
787 case TPACKET_V2:
788 h.h2->tp_len = skb->len;
789 h.h2->tp_snaplen = snaplen;
790 h.h2->tp_mac = macoff;
791 h.h2->tp_net = netoff;
792 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
793 && shhwtstamps->syststamp.tv64)
794 ts = ktime_to_timespec(shhwtstamps->syststamp);
795 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
796 && shhwtstamps->hwtstamp.tv64)
797 ts = ktime_to_timespec(shhwtstamps->hwtstamp);
798 else if (skb->tstamp.tv64)
799 ts = ktime_to_timespec(skb->tstamp);
800 else
801 getnstimeofday(&ts);
802 h.h2->tp_sec = ts.tv_sec;
803 h.h2->tp_nsec = ts.tv_nsec;
804 if (vlan_tx_tag_present(skb)) {
805 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
806 status |= TP_STATUS_VLAN_VALID;
807 } else {
808 h.h2->tp_vlan_tci = 0;
810 h.h2->tp_padding = 0;
811 hdrlen = sizeof(*h.h2);
812 break;
813 default:
814 BUG();
817 sll = h.raw + TPACKET_ALIGN(hdrlen);
818 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
819 sll->sll_family = AF_PACKET;
820 sll->sll_hatype = dev->type;
821 sll->sll_protocol = skb->protocol;
822 sll->sll_pkttype = skb->pkt_type;
823 if (unlikely(po->origdev))
824 sll->sll_ifindex = orig_dev->ifindex;
825 else
826 sll->sll_ifindex = dev->ifindex;
828 __packet_set_status(po, h.raw, status);
829 smp_mb();
830 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
832 u8 *start, *end;
834 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
835 for (start = h.raw; start < end; start += PAGE_SIZE)
836 flush_dcache_page(pgv_to_page(start));
838 #endif
840 sk->sk_data_ready(sk, 0);
842 drop_n_restore:
843 if (skb_head != skb->data && skb_shared(skb)) {
844 skb->data = skb_head;
845 skb->len = skb_len;
847 drop:
848 kfree_skb(skb);
849 return 0;
851 ring_is_full:
852 po->stats.tp_drops++;
853 spin_unlock(&sk->sk_receive_queue.lock);
855 sk->sk_data_ready(sk, 0);
856 kfree_skb(copy_skb);
857 goto drop_n_restore;
860 static void tpacket_destruct_skb(struct sk_buff *skb)
862 struct packet_sock *po = pkt_sk(skb->sk);
863 void *ph;
865 BUG_ON(skb == NULL);
867 if (likely(po->tx_ring.pg_vec)) {
868 ph = skb_shinfo(skb)->destructor_arg;
869 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
870 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
871 atomic_dec(&po->tx_ring.pending);
872 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
875 sock_wfree(skb);
878 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
879 void *frame, struct net_device *dev, int size_max,
880 __be16 proto, unsigned char *addr)
882 union {
883 struct tpacket_hdr *h1;
884 struct tpacket2_hdr *h2;
885 void *raw;
886 } ph;
887 int to_write, offset, len, tp_len, nr_frags, len_max;
888 struct socket *sock = po->sk.sk_socket;
889 struct page *page;
890 void *data;
891 int err;
893 ph.raw = frame;
895 skb->protocol = proto;
896 skb->dev = dev;
897 skb->priority = po->sk.sk_priority;
898 skb->mark = po->sk.sk_mark;
899 skb_shinfo(skb)->destructor_arg = ph.raw;
901 switch (po->tp_version) {
902 case TPACKET_V2:
903 tp_len = ph.h2->tp_len;
904 break;
905 default:
906 tp_len = ph.h1->tp_len;
907 break;
909 if (unlikely(tp_len > size_max)) {
910 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
911 return -EMSGSIZE;
914 skb_reserve(skb, LL_RESERVED_SPACE(dev));
915 skb_reset_network_header(skb);
917 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
918 to_write = tp_len;
920 if (sock->type == SOCK_DGRAM) {
921 err = dev_hard_header(skb, dev, ntohs(proto), addr,
922 NULL, tp_len);
923 if (unlikely(err < 0))
924 return -EINVAL;
925 } else if (dev->hard_header_len) {
926 /* net device doesn't like empty head */
927 if (unlikely(tp_len <= dev->hard_header_len)) {
928 pr_err("packet size is too short (%d < %d)\n",
929 tp_len, dev->hard_header_len);
930 return -EINVAL;
933 skb_push(skb, dev->hard_header_len);
934 err = skb_store_bits(skb, 0, data,
935 dev->hard_header_len);
936 if (unlikely(err))
937 return err;
939 data += dev->hard_header_len;
940 to_write -= dev->hard_header_len;
943 err = -EFAULT;
944 offset = offset_in_page(data);
945 len_max = PAGE_SIZE - offset;
946 len = ((to_write > len_max) ? len_max : to_write);
948 skb->data_len = to_write;
949 skb->len += to_write;
950 skb->truesize += to_write;
951 atomic_add(to_write, &po->sk.sk_wmem_alloc);
953 while (likely(to_write)) {
954 nr_frags = skb_shinfo(skb)->nr_frags;
956 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
957 pr_err("Packet exceed the number of skb frags(%lu)\n",
958 MAX_SKB_FRAGS);
959 return -EFAULT;
962 page = pgv_to_page(data);
963 data += len;
964 flush_dcache_page(page);
965 get_page(page);
966 skb_fill_page_desc(skb, nr_frags, page, offset, len);
967 to_write -= len;
968 offset = 0;
969 len_max = PAGE_SIZE;
970 len = ((to_write > len_max) ? len_max : to_write);
973 return tp_len;
976 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
978 struct sk_buff *skb;
979 struct net_device *dev;
980 __be16 proto;
981 int ifindex, err, reserve = 0;
982 void *ph;
983 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
984 int tp_len, size_max;
985 unsigned char *addr;
986 int len_sum = 0;
987 int status = 0;
989 mutex_lock(&po->pg_vec_lock);
991 err = -EBUSY;
992 if (saddr == NULL) {
993 ifindex = po->ifindex;
994 proto = po->num;
995 addr = NULL;
996 } else {
997 err = -EINVAL;
998 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
999 goto out;
1000 if (msg->msg_namelen < (saddr->sll_halen
1001 + offsetof(struct sockaddr_ll,
1002 sll_addr)))
1003 goto out;
1004 ifindex = saddr->sll_ifindex;
1005 proto = saddr->sll_protocol;
1006 addr = saddr->sll_addr;
1009 dev = dev_get_by_index(sock_net(&po->sk), ifindex);
1010 err = -ENXIO;
1011 if (unlikely(dev == NULL))
1012 goto out;
1014 reserve = dev->hard_header_len;
1016 err = -ENETDOWN;
1017 if (unlikely(!(dev->flags & IFF_UP)))
1018 goto out_put;
1020 size_max = po->tx_ring.frame_size
1021 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1023 if (size_max > dev->mtu + reserve)
1024 size_max = dev->mtu + reserve;
1026 do {
1027 ph = packet_current_frame(po, &po->tx_ring,
1028 TP_STATUS_SEND_REQUEST);
1030 if (unlikely(ph == NULL)) {
1031 schedule();
1032 continue;
1035 status = TP_STATUS_SEND_REQUEST;
1036 skb = sock_alloc_send_skb(&po->sk,
1037 LL_ALLOCATED_SPACE(dev)
1038 + sizeof(struct sockaddr_ll),
1039 0, &err);
1041 if (unlikely(skb == NULL))
1042 goto out_status;
1044 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1045 addr);
1047 if (unlikely(tp_len < 0)) {
1048 if (po->tp_loss) {
1049 __packet_set_status(po, ph,
1050 TP_STATUS_AVAILABLE);
1051 packet_increment_head(&po->tx_ring);
1052 kfree_skb(skb);
1053 continue;
1054 } else {
1055 status = TP_STATUS_WRONG_FORMAT;
1056 err = tp_len;
1057 goto out_status;
1061 skb->destructor = tpacket_destruct_skb;
1062 __packet_set_status(po, ph, TP_STATUS_SENDING);
1063 atomic_inc(&po->tx_ring.pending);
1065 status = TP_STATUS_SEND_REQUEST;
1066 err = dev_queue_xmit(skb);
1067 if (unlikely(err > 0)) {
1068 err = net_xmit_errno(err);
1069 if (err && __packet_get_status(po, ph) ==
1070 TP_STATUS_AVAILABLE) {
1071 /* skb was destructed already */
1072 skb = NULL;
1073 goto out_status;
1076 * skb was dropped but not destructed yet;
1077 * let's treat it like congestion or err < 0
1079 err = 0;
1081 packet_increment_head(&po->tx_ring);
1082 len_sum += tp_len;
1083 } while (likely((ph != NULL) ||
1084 ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1085 (atomic_read(&po->tx_ring.pending))))
1088 err = len_sum;
1089 goto out_put;
1091 out_status:
1092 __packet_set_status(po, ph, status);
1093 kfree_skb(skb);
1094 out_put:
1095 dev_put(dev);
1096 out:
1097 mutex_unlock(&po->pg_vec_lock);
1098 return err;
1101 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1102 size_t reserve, size_t len,
1103 size_t linear, int noblock,
1104 int *err)
1106 struct sk_buff *skb;
1108 /* Under a page? Don't bother with paged skb. */
1109 if (prepad + len < PAGE_SIZE || !linear)
1110 linear = len;
1112 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1113 err);
1114 if (!skb)
1115 return NULL;
1117 skb_reserve(skb, reserve);
1118 skb_put(skb, linear);
1119 skb->data_len = len - linear;
1120 skb->len += len - linear;
1122 return skb;
1125 static int packet_snd(struct socket *sock,
1126 struct msghdr *msg, size_t len)
1128 struct sock *sk = sock->sk;
1129 struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1130 struct sk_buff *skb;
1131 struct net_device *dev;
1132 __be16 proto;
1133 unsigned char *addr;
1134 int ifindex, err, reserve = 0;
1135 struct virtio_net_hdr vnet_hdr = { 0 };
1136 int offset = 0;
1137 int vnet_hdr_len;
1138 struct packet_sock *po = pkt_sk(sk);
1139 unsigned short gso_type = 0;
1142 * Get and verify the address.
1145 if (saddr == NULL) {
1146 ifindex = po->ifindex;
1147 proto = po->num;
1148 addr = NULL;
1149 } else {
1150 err = -EINVAL;
1151 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1152 goto out;
1153 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1154 goto out;
1155 ifindex = saddr->sll_ifindex;
1156 proto = saddr->sll_protocol;
1157 addr = saddr->sll_addr;
1161 dev = dev_get_by_index(sock_net(sk), ifindex);
1162 err = -ENXIO;
1163 if (dev == NULL)
1164 goto out_unlock;
1165 if (sock->type == SOCK_RAW)
1166 reserve = dev->hard_header_len;
1168 err = -ENETDOWN;
1169 if (!(dev->flags & IFF_UP))
1170 goto out_unlock;
1172 if (po->has_vnet_hdr) {
1173 vnet_hdr_len = sizeof(vnet_hdr);
1175 err = -EINVAL;
1176 if (len < vnet_hdr_len)
1177 goto out_unlock;
1179 len -= vnet_hdr_len;
1181 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1182 vnet_hdr_len);
1183 if (err < 0)
1184 goto out_unlock;
1186 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1187 (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1188 vnet_hdr.hdr_len))
1189 vnet_hdr.hdr_len = vnet_hdr.csum_start +
1190 vnet_hdr.csum_offset + 2;
1192 err = -EINVAL;
1193 if (vnet_hdr.hdr_len > len)
1194 goto out_unlock;
1196 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1197 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1198 case VIRTIO_NET_HDR_GSO_TCPV4:
1199 gso_type = SKB_GSO_TCPV4;
1200 break;
1201 case VIRTIO_NET_HDR_GSO_TCPV6:
1202 gso_type = SKB_GSO_TCPV6;
1203 break;
1204 case VIRTIO_NET_HDR_GSO_UDP:
1205 gso_type = SKB_GSO_UDP;
1206 break;
1207 default:
1208 goto out_unlock;
1211 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1212 gso_type |= SKB_GSO_TCP_ECN;
1214 if (vnet_hdr.gso_size == 0)
1215 goto out_unlock;
1220 err = -EMSGSIZE;
1221 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1222 goto out_unlock;
1224 err = -ENOBUFS;
1225 skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1226 LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1227 msg->msg_flags & MSG_DONTWAIT, &err);
1228 if (skb == NULL)
1229 goto out_unlock;
1231 skb_set_network_header(skb, reserve);
1233 err = -EINVAL;
1234 if (sock->type == SOCK_DGRAM &&
1235 (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1236 goto out_free;
1238 /* Returns -EFAULT on error */
1239 err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1240 if (err)
1241 goto out_free;
1242 err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1243 if (err < 0)
1244 goto out_free;
1246 if (!gso_type && (len > dev->mtu + reserve)) {
1247 /* Earlier code assumed this would be a VLAN pkt,
1248 * double-check this now that we have the actual
1249 * packet in hand.
1251 struct ethhdr *ehdr;
1252 skb_reset_mac_header(skb);
1253 ehdr = eth_hdr(skb);
1254 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1255 err = -EMSGSIZE;
1256 goto out_free;
1260 skb->protocol = proto;
1261 skb->dev = dev;
1262 skb->priority = sk->sk_priority;
1263 skb->mark = sk->sk_mark;
1265 if (po->has_vnet_hdr) {
1266 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1267 if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1268 vnet_hdr.csum_offset)) {
1269 err = -EINVAL;
1270 goto out_free;
1274 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1275 skb_shinfo(skb)->gso_type = gso_type;
1277 /* Header must be checked, and gso_segs computed. */
1278 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1279 skb_shinfo(skb)->gso_segs = 0;
1281 len += vnet_hdr_len;
1285 * Now send it
1288 err = dev_queue_xmit(skb);
1289 if (err > 0 && (err = net_xmit_errno(err)) != 0)
1290 goto out_unlock;
1292 dev_put(dev);
1294 return len;
1296 out_free:
1297 kfree_skb(skb);
1298 out_unlock:
1299 if (dev)
1300 dev_put(dev);
1301 out:
1302 return err;
1305 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1306 struct msghdr *msg, size_t len)
1308 struct sock *sk = sock->sk;
1309 struct packet_sock *po = pkt_sk(sk);
1310 if (po->tx_ring.pg_vec)
1311 return tpacket_snd(po, msg);
1312 else
1313 return packet_snd(sock, msg, len);
1317 * Close a PACKET socket. This is fairly simple. We immediately go
1318 * to 'closed' state and remove our protocol entry in the device list.
1321 static int packet_release(struct socket *sock)
1323 struct sock *sk = sock->sk;
1324 struct packet_sock *po;
1325 struct net *net;
1326 struct tpacket_req req;
1328 if (!sk)
1329 return 0;
1331 net = sock_net(sk);
1332 po = pkt_sk(sk);
1334 spin_lock_bh(&net->packet.sklist_lock);
1335 sk_del_node_init_rcu(sk);
1336 sock_prot_inuse_add(net, sk->sk_prot, -1);
1337 spin_unlock_bh(&net->packet.sklist_lock);
1339 spin_lock(&po->bind_lock);
1340 if (po->running) {
1342 * Remove from protocol table
1344 po->running = 0;
1345 po->num = 0;
1346 __dev_remove_pack(&po->prot_hook);
1347 __sock_put(sk);
1349 spin_unlock(&po->bind_lock);
1351 packet_flush_mclist(sk);
1353 memset(&req, 0, sizeof(req));
1355 if (po->rx_ring.pg_vec)
1356 packet_set_ring(sk, &req, 1, 0);
1358 if (po->tx_ring.pg_vec)
1359 packet_set_ring(sk, &req, 1, 1);
1361 synchronize_net();
1363 * Now the socket is dead. No more input will appear.
1365 sock_orphan(sk);
1366 sock->sk = NULL;
1368 /* Purge queues */
1370 skb_queue_purge(&sk->sk_receive_queue);
1371 sk_refcnt_debug_release(sk);
1373 sock_put(sk);
1374 return 0;
1378 * Attach a packet hook.
1381 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1383 struct packet_sock *po = pkt_sk(sk);
1385 * Detach an existing hook if present.
1388 lock_sock(sk);
1390 spin_lock(&po->bind_lock);
1391 if (po->running) {
1392 __sock_put(sk);
1393 po->running = 0;
1394 po->num = 0;
1395 spin_unlock(&po->bind_lock);
1396 dev_remove_pack(&po->prot_hook);
1397 spin_lock(&po->bind_lock);
1400 po->num = protocol;
1401 po->prot_hook.type = protocol;
1402 po->prot_hook.dev = dev;
1404 po->ifindex = dev ? dev->ifindex : 0;
1406 if (protocol == 0)
1407 goto out_unlock;
1409 if (!dev || (dev->flags & IFF_UP)) {
1410 dev_add_pack(&po->prot_hook);
1411 sock_hold(sk);
1412 po->running = 1;
1413 } else {
1414 sk->sk_err = ENETDOWN;
1415 if (!sock_flag(sk, SOCK_DEAD))
1416 sk->sk_error_report(sk);
1419 out_unlock:
1420 spin_unlock(&po->bind_lock);
1421 release_sock(sk);
1422 return 0;
1426 * Bind a packet socket to a device
1429 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1430 int addr_len)
1432 struct sock *sk = sock->sk;
1433 char name[15];
1434 struct net_device *dev;
1435 int err = -ENODEV;
1438 * Check legality
1441 if (addr_len != sizeof(struct sockaddr))
1442 return -EINVAL;
1443 strlcpy(name, uaddr->sa_data, sizeof(name));
1445 dev = dev_get_by_name(sock_net(sk), name);
1446 if (dev) {
1447 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1448 dev_put(dev);
1450 return err;
1453 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1455 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1456 struct sock *sk = sock->sk;
1457 struct net_device *dev = NULL;
1458 int err;
1462 * Check legality
1465 if (addr_len < sizeof(struct sockaddr_ll))
1466 return -EINVAL;
1467 if (sll->sll_family != AF_PACKET)
1468 return -EINVAL;
1470 if (sll->sll_ifindex) {
1471 err = -ENODEV;
1472 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1473 if (dev == NULL)
1474 goto out;
1476 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1477 if (dev)
1478 dev_put(dev);
1480 out:
1481 return err;
1484 static struct proto packet_proto = {
1485 .name = "PACKET",
1486 .owner = THIS_MODULE,
1487 .obj_size = sizeof(struct packet_sock),
1491 * Create a packet of type SOCK_PACKET.
1494 static int packet_create(struct net *net, struct socket *sock, int protocol,
1495 int kern)
1497 struct sock *sk;
1498 struct packet_sock *po;
1499 __be16 proto = (__force __be16)protocol; /* weird, but documented */
1500 int err;
1502 if (!capable(CAP_NET_RAW))
1503 return -EPERM;
1504 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1505 sock->type != SOCK_PACKET)
1506 return -ESOCKTNOSUPPORT;
1508 sock->state = SS_UNCONNECTED;
1510 err = -ENOBUFS;
1511 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1512 if (sk == NULL)
1513 goto out;
1515 sock->ops = &packet_ops;
1516 if (sock->type == SOCK_PACKET)
1517 sock->ops = &packet_ops_spkt;
1519 sock_init_data(sock, sk);
1521 po = pkt_sk(sk);
1522 sk->sk_family = PF_PACKET;
1523 po->num = proto;
1525 sk->sk_destruct = packet_sock_destruct;
1526 sk_refcnt_debug_inc(sk);
1529 * Attach a protocol block
1532 spin_lock_init(&po->bind_lock);
1533 mutex_init(&po->pg_vec_lock);
1534 po->prot_hook.func = packet_rcv;
1536 if (sock->type == SOCK_PACKET)
1537 po->prot_hook.func = packet_rcv_spkt;
1539 po->prot_hook.af_packet_priv = sk;
1541 if (proto) {
1542 po->prot_hook.type = proto;
1543 dev_add_pack(&po->prot_hook);
1544 sock_hold(sk);
1545 po->running = 1;
1548 spin_lock_bh(&net->packet.sklist_lock);
1549 sk_add_node_rcu(sk, &net->packet.sklist);
1550 sock_prot_inuse_add(net, &packet_proto, 1);
1551 spin_unlock_bh(&net->packet.sklist_lock);
1553 return 0;
1554 out:
1555 return err;
1558 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1560 struct sock_exterr_skb *serr;
1561 struct sk_buff *skb, *skb2;
1562 int copied, err;
1564 err = -EAGAIN;
1565 skb = skb_dequeue(&sk->sk_error_queue);
1566 if (skb == NULL)
1567 goto out;
1569 copied = skb->len;
1570 if (copied > len) {
1571 msg->msg_flags |= MSG_TRUNC;
1572 copied = len;
1574 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1575 if (err)
1576 goto out_free_skb;
1578 sock_recv_timestamp(msg, sk, skb);
1580 serr = SKB_EXT_ERR(skb);
1581 put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1582 sizeof(serr->ee), &serr->ee);
1584 msg->msg_flags |= MSG_ERRQUEUE;
1585 err = copied;
1587 /* Reset and regenerate socket error */
1588 spin_lock_bh(&sk->sk_error_queue.lock);
1589 sk->sk_err = 0;
1590 if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1591 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1592 spin_unlock_bh(&sk->sk_error_queue.lock);
1593 sk->sk_error_report(sk);
1594 } else
1595 spin_unlock_bh(&sk->sk_error_queue.lock);
1597 out_free_skb:
1598 kfree_skb(skb);
1599 out:
1600 return err;
1604 * Pull a packet from our receive queue and hand it to the user.
1605 * If necessary we block.
1608 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1609 struct msghdr *msg, size_t len, int flags)
1611 struct sock *sk = sock->sk;
1612 struct sk_buff *skb;
1613 int copied, err;
1614 struct sockaddr_ll *sll;
1615 int vnet_hdr_len = 0;
1617 err = -EINVAL;
1618 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1619 goto out;
1621 #if 0
1622 /* What error should we return now? EUNATTACH? */
1623 if (pkt_sk(sk)->ifindex < 0)
1624 return -ENODEV;
1625 #endif
1627 if (flags & MSG_ERRQUEUE) {
1628 err = packet_recv_error(sk, msg, len);
1629 goto out;
1633 * Call the generic datagram receiver. This handles all sorts
1634 * of horrible races and re-entrancy so we can forget about it
1635 * in the protocol layers.
1637 * Now it will return ENETDOWN, if device have just gone down,
1638 * but then it will block.
1641 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1644 * An error occurred so return it. Because skb_recv_datagram()
1645 * handles the blocking we don't see and worry about blocking
1646 * retries.
1649 if (skb == NULL)
1650 goto out;
1652 if (pkt_sk(sk)->has_vnet_hdr) {
1653 struct virtio_net_hdr vnet_hdr = { 0 };
1655 err = -EINVAL;
1656 vnet_hdr_len = sizeof(vnet_hdr);
1657 if (len < vnet_hdr_len)
1658 goto out_free;
1660 len -= vnet_hdr_len;
1662 if (skb_is_gso(skb)) {
1663 struct skb_shared_info *sinfo = skb_shinfo(skb);
1665 /* This is a hint as to how much should be linear. */
1666 vnet_hdr.hdr_len = skb_headlen(skb);
1667 vnet_hdr.gso_size = sinfo->gso_size;
1668 if (sinfo->gso_type & SKB_GSO_TCPV4)
1669 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1670 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1671 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1672 else if (sinfo->gso_type & SKB_GSO_UDP)
1673 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1674 else if (sinfo->gso_type & SKB_GSO_FCOE)
1675 goto out_free;
1676 else
1677 BUG();
1678 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1679 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1680 } else
1681 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1683 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1684 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1685 vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1686 vnet_hdr.csum_offset = skb->csum_offset;
1687 } /* else everything is zero */
1689 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1690 vnet_hdr_len);
1691 if (err < 0)
1692 goto out_free;
1696 * If the address length field is there to be filled in, we fill
1697 * it in now.
1700 sll = &PACKET_SKB_CB(skb)->sa.ll;
1701 if (sock->type == SOCK_PACKET)
1702 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1703 else
1704 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1707 * You lose any data beyond the buffer you gave. If it worries a
1708 * user program they can ask the device for its MTU anyway.
1711 copied = skb->len;
1712 if (copied > len) {
1713 copied = len;
1714 msg->msg_flags |= MSG_TRUNC;
1717 err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1718 if (err)
1719 goto out_free;
1721 sock_recv_ts_and_drops(msg, sk, skb);
1723 if (msg->msg_name)
1724 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1725 msg->msg_namelen);
1727 if (pkt_sk(sk)->auxdata) {
1728 struct tpacket_auxdata aux;
1730 aux.tp_status = TP_STATUS_USER;
1731 if (skb->ip_summed == CHECKSUM_PARTIAL)
1732 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1733 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1734 aux.tp_snaplen = skb->len;
1735 aux.tp_mac = 0;
1736 aux.tp_net = skb_network_offset(skb);
1737 if (vlan_tx_tag_present(skb)) {
1738 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1739 aux.tp_status |= TP_STATUS_VLAN_VALID;
1740 } else {
1741 aux.tp_vlan_tci = 0;
1743 aux.tp_padding = 0;
1744 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1748 * Free or return the buffer as appropriate. Again this
1749 * hides all the races and re-entrancy issues from us.
1751 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1753 out_free:
1754 skb_free_datagram(sk, skb);
1755 out:
1756 return err;
1759 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1760 int *uaddr_len, int peer)
1762 struct net_device *dev;
1763 struct sock *sk = sock->sk;
1765 if (peer)
1766 return -EOPNOTSUPP;
1768 uaddr->sa_family = AF_PACKET;
1769 rcu_read_lock();
1770 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1771 if (dev)
1772 strncpy(uaddr->sa_data, dev->name, 14);
1773 else
1774 memset(uaddr->sa_data, 0, 14);
1775 rcu_read_unlock();
1776 *uaddr_len = sizeof(*uaddr);
1778 return 0;
1781 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1782 int *uaddr_len, int peer)
1784 struct net_device *dev;
1785 struct sock *sk = sock->sk;
1786 struct packet_sock *po = pkt_sk(sk);
1787 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1789 if (peer)
1790 return -EOPNOTSUPP;
1792 sll->sll_family = AF_PACKET;
1793 sll->sll_ifindex = po->ifindex;
1794 sll->sll_protocol = po->num;
1795 sll->sll_pkttype = 0;
1796 rcu_read_lock();
1797 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1798 if (dev) {
1799 sll->sll_hatype = dev->type;
1800 sll->sll_halen = dev->addr_len;
1801 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1802 } else {
1803 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
1804 sll->sll_halen = 0;
1806 rcu_read_unlock();
1807 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1809 return 0;
1812 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1813 int what)
1815 switch (i->type) {
1816 case PACKET_MR_MULTICAST:
1817 if (i->alen != dev->addr_len)
1818 return -EINVAL;
1819 if (what > 0)
1820 return dev_mc_add(dev, i->addr);
1821 else
1822 return dev_mc_del(dev, i->addr);
1823 break;
1824 case PACKET_MR_PROMISC:
1825 return dev_set_promiscuity(dev, what);
1826 break;
1827 case PACKET_MR_ALLMULTI:
1828 return dev_set_allmulti(dev, what);
1829 break;
1830 case PACKET_MR_UNICAST:
1831 if (i->alen != dev->addr_len)
1832 return -EINVAL;
1833 if (what > 0)
1834 return dev_uc_add(dev, i->addr);
1835 else
1836 return dev_uc_del(dev, i->addr);
1837 break;
1838 default:
1839 break;
1841 return 0;
1844 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1846 for ( ; i; i = i->next) {
1847 if (i->ifindex == dev->ifindex)
1848 packet_dev_mc(dev, i, what);
1852 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1854 struct packet_sock *po = pkt_sk(sk);
1855 struct packet_mclist *ml, *i;
1856 struct net_device *dev;
1857 int err;
1859 rtnl_lock();
1861 err = -ENODEV;
1862 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1863 if (!dev)
1864 goto done;
1866 err = -EINVAL;
1867 if (mreq->mr_alen > dev->addr_len)
1868 goto done;
1870 err = -ENOBUFS;
1871 i = kmalloc(sizeof(*i), GFP_KERNEL);
1872 if (i == NULL)
1873 goto done;
1875 err = 0;
1876 for (ml = po->mclist; ml; ml = ml->next) {
1877 if (ml->ifindex == mreq->mr_ifindex &&
1878 ml->type == mreq->mr_type &&
1879 ml->alen == mreq->mr_alen &&
1880 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1881 ml->count++;
1882 /* Free the new element ... */
1883 kfree(i);
1884 goto done;
1888 i->type = mreq->mr_type;
1889 i->ifindex = mreq->mr_ifindex;
1890 i->alen = mreq->mr_alen;
1891 memcpy(i->addr, mreq->mr_address, i->alen);
1892 i->count = 1;
1893 i->next = po->mclist;
1894 po->mclist = i;
1895 err = packet_dev_mc(dev, i, 1);
1896 if (err) {
1897 po->mclist = i->next;
1898 kfree(i);
1901 done:
1902 rtnl_unlock();
1903 return err;
1906 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1908 struct packet_mclist *ml, **mlp;
1910 rtnl_lock();
1912 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1913 if (ml->ifindex == mreq->mr_ifindex &&
1914 ml->type == mreq->mr_type &&
1915 ml->alen == mreq->mr_alen &&
1916 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1917 if (--ml->count == 0) {
1918 struct net_device *dev;
1919 *mlp = ml->next;
1920 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1921 if (dev)
1922 packet_dev_mc(dev, ml, -1);
1923 kfree(ml);
1925 rtnl_unlock();
1926 return 0;
1929 rtnl_unlock();
1930 return -EADDRNOTAVAIL;
1933 static void packet_flush_mclist(struct sock *sk)
1935 struct packet_sock *po = pkt_sk(sk);
1936 struct packet_mclist *ml;
1938 if (!po->mclist)
1939 return;
1941 rtnl_lock();
1942 while ((ml = po->mclist) != NULL) {
1943 struct net_device *dev;
1945 po->mclist = ml->next;
1946 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1947 if (dev != NULL)
1948 packet_dev_mc(dev, ml, -1);
1949 kfree(ml);
1951 rtnl_unlock();
1954 static int
1955 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1957 struct sock *sk = sock->sk;
1958 struct packet_sock *po = pkt_sk(sk);
1959 int ret;
1961 if (level != SOL_PACKET)
1962 return -ENOPROTOOPT;
1964 switch (optname) {
1965 case PACKET_ADD_MEMBERSHIP:
1966 case PACKET_DROP_MEMBERSHIP:
1968 struct packet_mreq_max mreq;
1969 int len = optlen;
1970 memset(&mreq, 0, sizeof(mreq));
1971 if (len < sizeof(struct packet_mreq))
1972 return -EINVAL;
1973 if (len > sizeof(mreq))
1974 len = sizeof(mreq);
1975 if (copy_from_user(&mreq, optval, len))
1976 return -EFAULT;
1977 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1978 return -EINVAL;
1979 if (optname == PACKET_ADD_MEMBERSHIP)
1980 ret = packet_mc_add(sk, &mreq);
1981 else
1982 ret = packet_mc_drop(sk, &mreq);
1983 return ret;
1986 case PACKET_RX_RING:
1987 case PACKET_TX_RING:
1989 struct tpacket_req req;
1991 if (optlen < sizeof(req))
1992 return -EINVAL;
1993 if (pkt_sk(sk)->has_vnet_hdr)
1994 return -EINVAL;
1995 if (copy_from_user(&req, optval, sizeof(req)))
1996 return -EFAULT;
1997 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1999 case PACKET_COPY_THRESH:
2001 int val;
2003 if (optlen != sizeof(val))
2004 return -EINVAL;
2005 if (copy_from_user(&val, optval, sizeof(val)))
2006 return -EFAULT;
2008 pkt_sk(sk)->copy_thresh = val;
2009 return 0;
2011 case PACKET_VERSION:
2013 int val;
2015 if (optlen != sizeof(val))
2016 return -EINVAL;
2017 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2018 return -EBUSY;
2019 if (copy_from_user(&val, optval, sizeof(val)))
2020 return -EFAULT;
2021 switch (val) {
2022 case TPACKET_V1:
2023 case TPACKET_V2:
2024 po->tp_version = val;
2025 return 0;
2026 default:
2027 return -EINVAL;
2030 case PACKET_RESERVE:
2032 unsigned int val;
2034 if (optlen != sizeof(val))
2035 return -EINVAL;
2036 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2037 return -EBUSY;
2038 if (copy_from_user(&val, optval, sizeof(val)))
2039 return -EFAULT;
2040 po->tp_reserve = val;
2041 return 0;
2043 case PACKET_LOSS:
2045 unsigned int val;
2047 if (optlen != sizeof(val))
2048 return -EINVAL;
2049 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2050 return -EBUSY;
2051 if (copy_from_user(&val, optval, sizeof(val)))
2052 return -EFAULT;
2053 po->tp_loss = !!val;
2054 return 0;
2056 case PACKET_AUXDATA:
2058 int val;
2060 if (optlen < sizeof(val))
2061 return -EINVAL;
2062 if (copy_from_user(&val, optval, sizeof(val)))
2063 return -EFAULT;
2065 po->auxdata = !!val;
2066 return 0;
2068 case PACKET_ORIGDEV:
2070 int val;
2072 if (optlen < sizeof(val))
2073 return -EINVAL;
2074 if (copy_from_user(&val, optval, sizeof(val)))
2075 return -EFAULT;
2077 po->origdev = !!val;
2078 return 0;
2080 case PACKET_VNET_HDR:
2082 int val;
2084 if (sock->type != SOCK_RAW)
2085 return -EINVAL;
2086 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2087 return -EBUSY;
2088 if (optlen < sizeof(val))
2089 return -EINVAL;
2090 if (copy_from_user(&val, optval, sizeof(val)))
2091 return -EFAULT;
2093 po->has_vnet_hdr = !!val;
2094 return 0;
2096 case PACKET_TIMESTAMP:
2098 int val;
2100 if (optlen != sizeof(val))
2101 return -EINVAL;
2102 if (copy_from_user(&val, optval, sizeof(val)))
2103 return -EFAULT;
2105 po->tp_tstamp = val;
2106 return 0;
2108 default:
2109 return -ENOPROTOOPT;
2113 static int packet_getsockopt(struct socket *sock, int level, int optname,
2114 char __user *optval, int __user *optlen)
2116 int len;
2117 int val;
2118 struct sock *sk = sock->sk;
2119 struct packet_sock *po = pkt_sk(sk);
2120 void *data;
2121 struct tpacket_stats st;
2123 if (level != SOL_PACKET)
2124 return -ENOPROTOOPT;
2126 if (get_user(len, optlen))
2127 return -EFAULT;
2129 if (len < 0)
2130 return -EINVAL;
2132 switch (optname) {
2133 case PACKET_STATISTICS:
2134 if (len > sizeof(struct tpacket_stats))
2135 len = sizeof(struct tpacket_stats);
2136 spin_lock_bh(&sk->sk_receive_queue.lock);
2137 st = po->stats;
2138 memset(&po->stats, 0, sizeof(st));
2139 spin_unlock_bh(&sk->sk_receive_queue.lock);
2140 st.tp_packets += st.tp_drops;
2142 data = &st;
2143 break;
2144 case PACKET_AUXDATA:
2145 if (len > sizeof(int))
2146 len = sizeof(int);
2147 val = po->auxdata;
2149 data = &val;
2150 break;
2151 case PACKET_ORIGDEV:
2152 if (len > sizeof(int))
2153 len = sizeof(int);
2154 val = po->origdev;
2156 data = &val;
2157 break;
2158 case PACKET_VNET_HDR:
2159 if (len > sizeof(int))
2160 len = sizeof(int);
2161 val = po->has_vnet_hdr;
2163 data = &val;
2164 break;
2165 case PACKET_VERSION:
2166 if (len > sizeof(int))
2167 len = sizeof(int);
2168 val = po->tp_version;
2169 data = &val;
2170 break;
2171 case PACKET_HDRLEN:
2172 if (len > sizeof(int))
2173 len = sizeof(int);
2174 if (copy_from_user(&val, optval, len))
2175 return -EFAULT;
2176 switch (val) {
2177 case TPACKET_V1:
2178 val = sizeof(struct tpacket_hdr);
2179 break;
2180 case TPACKET_V2:
2181 val = sizeof(struct tpacket2_hdr);
2182 break;
2183 default:
2184 return -EINVAL;
2186 data = &val;
2187 break;
2188 case PACKET_RESERVE:
2189 if (len > sizeof(unsigned int))
2190 len = sizeof(unsigned int);
2191 val = po->tp_reserve;
2192 data = &val;
2193 break;
2194 case PACKET_LOSS:
2195 if (len > sizeof(unsigned int))
2196 len = sizeof(unsigned int);
2197 val = po->tp_loss;
2198 data = &val;
2199 break;
2200 case PACKET_TIMESTAMP:
2201 if (len > sizeof(int))
2202 len = sizeof(int);
2203 val = po->tp_tstamp;
2204 data = &val;
2205 break;
2206 default:
2207 return -ENOPROTOOPT;
2210 if (put_user(len, optlen))
2211 return -EFAULT;
2212 if (copy_to_user(optval, data, len))
2213 return -EFAULT;
2214 return 0;
2218 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2220 struct sock *sk;
2221 struct hlist_node *node;
2222 struct net_device *dev = data;
2223 struct net *net = dev_net(dev);
2225 rcu_read_lock();
2226 sk_for_each_rcu(sk, node, &net->packet.sklist) {
2227 struct packet_sock *po = pkt_sk(sk);
2229 switch (msg) {
2230 case NETDEV_UNREGISTER:
2231 if (po->mclist)
2232 packet_dev_mclist(dev, po->mclist, -1);
2233 /* fallthrough */
2235 case NETDEV_DOWN:
2236 if (dev->ifindex == po->ifindex) {
2237 spin_lock(&po->bind_lock);
2238 if (po->running) {
2239 __dev_remove_pack(&po->prot_hook);
2240 __sock_put(sk);
2241 po->running = 0;
2242 sk->sk_err = ENETDOWN;
2243 if (!sock_flag(sk, SOCK_DEAD))
2244 sk->sk_error_report(sk);
2246 if (msg == NETDEV_UNREGISTER) {
2247 po->ifindex = -1;
2248 po->prot_hook.dev = NULL;
2250 spin_unlock(&po->bind_lock);
2252 break;
2253 case NETDEV_UP:
2254 if (dev->ifindex == po->ifindex) {
2255 spin_lock(&po->bind_lock);
2256 if (po->num && !po->running) {
2257 dev_add_pack(&po->prot_hook);
2258 sock_hold(sk);
2259 po->running = 1;
2261 spin_unlock(&po->bind_lock);
2263 break;
2266 rcu_read_unlock();
2267 return NOTIFY_DONE;
2271 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2272 unsigned long arg)
2274 struct sock *sk = sock->sk;
2276 switch (cmd) {
2277 case SIOCOUTQ:
2279 int amount = sk_wmem_alloc_get(sk);
2281 return put_user(amount, (int __user *)arg);
2283 case SIOCINQ:
2285 struct sk_buff *skb;
2286 int amount = 0;
2288 spin_lock_bh(&sk->sk_receive_queue.lock);
2289 skb = skb_peek(&sk->sk_receive_queue);
2290 if (skb)
2291 amount = skb->len;
2292 spin_unlock_bh(&sk->sk_receive_queue.lock);
2293 return put_user(amount, (int __user *)arg);
2295 case SIOCGSTAMP:
2296 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2297 case SIOCGSTAMPNS:
2298 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2300 #ifdef CONFIG_INET
2301 case SIOCADDRT:
2302 case SIOCDELRT:
2303 case SIOCDARP:
2304 case SIOCGARP:
2305 case SIOCSARP:
2306 case SIOCGIFADDR:
2307 case SIOCSIFADDR:
2308 case SIOCGIFBRDADDR:
2309 case SIOCSIFBRDADDR:
2310 case SIOCGIFNETMASK:
2311 case SIOCSIFNETMASK:
2312 case SIOCGIFDSTADDR:
2313 case SIOCSIFDSTADDR:
2314 case SIOCSIFFLAGS:
2315 return inet_dgram_ops.ioctl(sock, cmd, arg);
2316 #endif
2318 default:
2319 return -ENOIOCTLCMD;
2321 return 0;
2324 static unsigned int packet_poll(struct file *file, struct socket *sock,
2325 poll_table *wait)
2327 struct sock *sk = sock->sk;
2328 struct packet_sock *po = pkt_sk(sk);
2329 unsigned int mask = datagram_poll(file, sock, wait);
2331 spin_lock_bh(&sk->sk_receive_queue.lock);
2332 if (po->rx_ring.pg_vec) {
2333 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2334 mask |= POLLIN | POLLRDNORM;
2336 spin_unlock_bh(&sk->sk_receive_queue.lock);
2337 spin_lock_bh(&sk->sk_write_queue.lock);
2338 if (po->tx_ring.pg_vec) {
2339 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2340 mask |= POLLOUT | POLLWRNORM;
2342 spin_unlock_bh(&sk->sk_write_queue.lock);
2343 return mask;
2347 /* Dirty? Well, I still did not learn better way to account
2348 * for user mmaps.
2351 static void packet_mm_open(struct vm_area_struct *vma)
2353 struct file *file = vma->vm_file;
2354 struct socket *sock = file->private_data;
2355 struct sock *sk = sock->sk;
2357 if (sk)
2358 atomic_inc(&pkt_sk(sk)->mapped);
2361 static void packet_mm_close(struct vm_area_struct *vma)
2363 struct file *file = vma->vm_file;
2364 struct socket *sock = file->private_data;
2365 struct sock *sk = sock->sk;
2367 if (sk)
2368 atomic_dec(&pkt_sk(sk)->mapped);
2371 static const struct vm_operations_struct packet_mmap_ops = {
2372 .open = packet_mm_open,
2373 .close = packet_mm_close,
2376 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2377 unsigned int len)
2379 int i;
2381 for (i = 0; i < len; i++) {
2382 if (likely(pg_vec[i].buffer)) {
2383 if (is_vmalloc_addr(pg_vec[i].buffer))
2384 vfree(pg_vec[i].buffer);
2385 else
2386 free_pages((unsigned long)pg_vec[i].buffer,
2387 order);
2388 pg_vec[i].buffer = NULL;
2391 kfree(pg_vec);
2394 static inline char *alloc_one_pg_vec_page(unsigned long order)
2396 char *buffer = NULL;
2397 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2398 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2400 buffer = (char *) __get_free_pages(gfp_flags, order);
2402 if (buffer)
2403 return buffer;
2406 * __get_free_pages failed, fall back to vmalloc
2408 buffer = vzalloc((1 << order) * PAGE_SIZE);
2410 if (buffer)
2411 return buffer;
2414 * vmalloc failed, lets dig into swap here
2416 gfp_flags &= ~__GFP_NORETRY;
2417 buffer = (char *)__get_free_pages(gfp_flags, order);
2418 if (buffer)
2419 return buffer;
2422 * complete and utter failure
2424 return NULL;
2427 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2429 unsigned int block_nr = req->tp_block_nr;
2430 struct pgv *pg_vec;
2431 int i;
2433 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2434 if (unlikely(!pg_vec))
2435 goto out;
2437 for (i = 0; i < block_nr; i++) {
2438 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2439 if (unlikely(!pg_vec[i].buffer))
2440 goto out_free_pgvec;
2443 out:
2444 return pg_vec;
2446 out_free_pgvec:
2447 free_pg_vec(pg_vec, order, block_nr);
2448 pg_vec = NULL;
2449 goto out;
2452 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2453 int closing, int tx_ring)
2455 struct pgv *pg_vec = NULL;
2456 struct packet_sock *po = pkt_sk(sk);
2457 int was_running, order = 0;
2458 struct packet_ring_buffer *rb;
2459 struct sk_buff_head *rb_queue;
2460 __be16 num;
2461 int err;
2463 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2464 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2466 err = -EBUSY;
2467 if (!closing) {
2468 if (atomic_read(&po->mapped))
2469 goto out;
2470 if (atomic_read(&rb->pending))
2471 goto out;
2474 if (req->tp_block_nr) {
2475 /* Sanity tests and some calculations */
2476 err = -EBUSY;
2477 if (unlikely(rb->pg_vec))
2478 goto out;
2480 switch (po->tp_version) {
2481 case TPACKET_V1:
2482 po->tp_hdrlen = TPACKET_HDRLEN;
2483 break;
2484 case TPACKET_V2:
2485 po->tp_hdrlen = TPACKET2_HDRLEN;
2486 break;
2489 err = -EINVAL;
2490 if (unlikely((int)req->tp_block_size <= 0))
2491 goto out;
2492 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2493 goto out;
2494 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2495 po->tp_reserve))
2496 goto out;
2497 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2498 goto out;
2500 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2501 if (unlikely(rb->frames_per_block <= 0))
2502 goto out;
2503 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2504 req->tp_frame_nr))
2505 goto out;
2507 err = -ENOMEM;
2508 order = get_order(req->tp_block_size);
2509 pg_vec = alloc_pg_vec(req, order);
2510 if (unlikely(!pg_vec))
2511 goto out;
2513 /* Done */
2514 else {
2515 err = -EINVAL;
2516 if (unlikely(req->tp_frame_nr))
2517 goto out;
2520 lock_sock(sk);
2522 /* Detach socket from network */
2523 spin_lock(&po->bind_lock);
2524 was_running = po->running;
2525 num = po->num;
2526 if (was_running) {
2527 __dev_remove_pack(&po->prot_hook);
2528 po->num = 0;
2529 po->running = 0;
2530 __sock_put(sk);
2532 spin_unlock(&po->bind_lock);
2534 synchronize_net();
2536 err = -EBUSY;
2537 mutex_lock(&po->pg_vec_lock);
2538 if (closing || atomic_read(&po->mapped) == 0) {
2539 err = 0;
2540 spin_lock_bh(&rb_queue->lock);
2541 swap(rb->pg_vec, pg_vec);
2542 rb->frame_max = (req->tp_frame_nr - 1);
2543 rb->head = 0;
2544 rb->frame_size = req->tp_frame_size;
2545 spin_unlock_bh(&rb_queue->lock);
2547 swap(rb->pg_vec_order, order);
2548 swap(rb->pg_vec_len, req->tp_block_nr);
2550 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2551 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2552 tpacket_rcv : packet_rcv;
2553 skb_queue_purge(rb_queue);
2554 if (atomic_read(&po->mapped))
2555 pr_err("packet_mmap: vma is busy: %d\n",
2556 atomic_read(&po->mapped));
2558 mutex_unlock(&po->pg_vec_lock);
2560 spin_lock(&po->bind_lock);
2561 if (was_running && !po->running) {
2562 sock_hold(sk);
2563 po->running = 1;
2564 po->num = num;
2565 dev_add_pack(&po->prot_hook);
2567 spin_unlock(&po->bind_lock);
2569 release_sock(sk);
2571 if (pg_vec)
2572 free_pg_vec(pg_vec, order, req->tp_block_nr);
2573 out:
2574 return err;
2577 static int packet_mmap(struct file *file, struct socket *sock,
2578 struct vm_area_struct *vma)
2580 struct sock *sk = sock->sk;
2581 struct packet_sock *po = pkt_sk(sk);
2582 unsigned long size, expected_size;
2583 struct packet_ring_buffer *rb;
2584 unsigned long start;
2585 int err = -EINVAL;
2586 int i;
2588 if (vma->vm_pgoff)
2589 return -EINVAL;
2591 mutex_lock(&po->pg_vec_lock);
2593 expected_size = 0;
2594 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2595 if (rb->pg_vec) {
2596 expected_size += rb->pg_vec_len
2597 * rb->pg_vec_pages
2598 * PAGE_SIZE;
2602 if (expected_size == 0)
2603 goto out;
2605 size = vma->vm_end - vma->vm_start;
2606 if (size != expected_size)
2607 goto out;
2609 start = vma->vm_start;
2610 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2611 if (rb->pg_vec == NULL)
2612 continue;
2614 for (i = 0; i < rb->pg_vec_len; i++) {
2615 struct page *page;
2616 void *kaddr = rb->pg_vec[i].buffer;
2617 int pg_num;
2619 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2620 page = pgv_to_page(kaddr);
2621 err = vm_insert_page(vma, start, page);
2622 if (unlikely(err))
2623 goto out;
2624 start += PAGE_SIZE;
2625 kaddr += PAGE_SIZE;
2630 atomic_inc(&po->mapped);
2631 vma->vm_ops = &packet_mmap_ops;
2632 err = 0;
2634 out:
2635 mutex_unlock(&po->pg_vec_lock);
2636 return err;
2639 static const struct proto_ops packet_ops_spkt = {
2640 .family = PF_PACKET,
2641 .owner = THIS_MODULE,
2642 .release = packet_release,
2643 .bind = packet_bind_spkt,
2644 .connect = sock_no_connect,
2645 .socketpair = sock_no_socketpair,
2646 .accept = sock_no_accept,
2647 .getname = packet_getname_spkt,
2648 .poll = datagram_poll,
2649 .ioctl = packet_ioctl,
2650 .listen = sock_no_listen,
2651 .shutdown = sock_no_shutdown,
2652 .setsockopt = sock_no_setsockopt,
2653 .getsockopt = sock_no_getsockopt,
2654 .sendmsg = packet_sendmsg_spkt,
2655 .recvmsg = packet_recvmsg,
2656 .mmap = sock_no_mmap,
2657 .sendpage = sock_no_sendpage,
2660 static const struct proto_ops packet_ops = {
2661 .family = PF_PACKET,
2662 .owner = THIS_MODULE,
2663 .release = packet_release,
2664 .bind = packet_bind,
2665 .connect = sock_no_connect,
2666 .socketpair = sock_no_socketpair,
2667 .accept = sock_no_accept,
2668 .getname = packet_getname,
2669 .poll = packet_poll,
2670 .ioctl = packet_ioctl,
2671 .listen = sock_no_listen,
2672 .shutdown = sock_no_shutdown,
2673 .setsockopt = packet_setsockopt,
2674 .getsockopt = packet_getsockopt,
2675 .sendmsg = packet_sendmsg,
2676 .recvmsg = packet_recvmsg,
2677 .mmap = packet_mmap,
2678 .sendpage = sock_no_sendpage,
2681 static const struct net_proto_family packet_family_ops = {
2682 .family = PF_PACKET,
2683 .create = packet_create,
2684 .owner = THIS_MODULE,
2687 static struct notifier_block packet_netdev_notifier = {
2688 .notifier_call = packet_notifier,
2691 #ifdef CONFIG_PROC_FS
2693 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2694 __acquires(RCU)
2696 struct net *net = seq_file_net(seq);
2698 rcu_read_lock();
2699 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2702 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2704 struct net *net = seq_file_net(seq);
2705 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2708 static void packet_seq_stop(struct seq_file *seq, void *v)
2709 __releases(RCU)
2711 rcu_read_unlock();
2714 static int packet_seq_show(struct seq_file *seq, void *v)
2716 if (v == SEQ_START_TOKEN)
2717 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
2718 else {
2719 struct sock *s = sk_entry(v);
2720 const struct packet_sock *po = pkt_sk(s);
2722 seq_printf(seq,
2723 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
2725 atomic_read(&s->sk_refcnt),
2726 s->sk_type,
2727 ntohs(po->num),
2728 po->ifindex,
2729 po->running,
2730 atomic_read(&s->sk_rmem_alloc),
2731 sock_i_uid(s),
2732 sock_i_ino(s));
2735 return 0;
2738 static const struct seq_operations packet_seq_ops = {
2739 .start = packet_seq_start,
2740 .next = packet_seq_next,
2741 .stop = packet_seq_stop,
2742 .show = packet_seq_show,
2745 static int packet_seq_open(struct inode *inode, struct file *file)
2747 return seq_open_net(inode, file, &packet_seq_ops,
2748 sizeof(struct seq_net_private));
2751 static const struct file_operations packet_seq_fops = {
2752 .owner = THIS_MODULE,
2753 .open = packet_seq_open,
2754 .read = seq_read,
2755 .llseek = seq_lseek,
2756 .release = seq_release_net,
2759 #endif
2761 static int __net_init packet_net_init(struct net *net)
2763 spin_lock_init(&net->packet.sklist_lock);
2764 INIT_HLIST_HEAD(&net->packet.sklist);
2766 if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2767 return -ENOMEM;
2769 return 0;
2772 static void __net_exit packet_net_exit(struct net *net)
2774 proc_net_remove(net, "packet");
2777 static struct pernet_operations packet_net_ops = {
2778 .init = packet_net_init,
2779 .exit = packet_net_exit,
2783 static void __exit packet_exit(void)
2785 unregister_netdevice_notifier(&packet_netdev_notifier);
2786 unregister_pernet_subsys(&packet_net_ops);
2787 sock_unregister(PF_PACKET);
2788 proto_unregister(&packet_proto);
2791 static int __init packet_init(void)
2793 int rc = proto_register(&packet_proto, 0);
2795 if (rc != 0)
2796 goto out;
2798 sock_register(&packet_family_ops);
2799 register_pernet_subsys(&packet_net_ops);
2800 register_netdevice_notifier(&packet_netdev_notifier);
2801 out:
2802 return rc;
2805 module_init(packet_init);
2806 module_exit(packet_exit);
2807 MODULE_LICENSE("GPL");
2808 MODULE_ALIAS_NETPROTO(PF_PACKET);