packet: convert socket list to RCU (v3)
[linux-2.6.git] / net / packet / af_packet.c
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.
5  *
6  *              PACKET - implements raw packet sockets.
7  *
8  * Authors:     Ross Biro
9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
11  *
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.
43  *
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.
48  *
49  */
50
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 <net/net_namespace.h>
64 #include <net/ip.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.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>
73 #include <asm/page.h>
74 #include <asm/cacheflush.h>
75 #include <asm/io.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>
82 #include <linux/if_vlan.h>
83 #include <linux/virtio_net.h>
84
85 #ifdef CONFIG_INET
86 #include <net/inet_common.h>
87 #endif
88
89 /*
90    Assumptions:
91    - if device has no dev->hard_header routine, it adds and removes ll header
92      inside itself. In this case ll header is invisible outside of device,
93      but higher levels still should reserve dev->hard_header_len.
94      Some devices are enough clever to reallocate skb, when header
95      will not fit to reserved space (tunnel), another ones are silly
96      (PPP).
97    - packet socket receives packets with pulled ll header,
98      so that SOCK_RAW should push it back.
99
100 On receive:
101 -----------
102
103 Incoming, dev->hard_header!=NULL
104    mac_header -> ll header
105    data       -> data
106
107 Outgoing, dev->hard_header!=NULL
108    mac_header -> ll header
109    data       -> ll header
110
111 Incoming, dev->hard_header==NULL
112    mac_header -> UNKNOWN position. It is very likely, that it points to ll
113                  header.  PPP makes it, that is wrong, because introduce
114                  assymetry between rx and tx paths.
115    data       -> data
116
117 Outgoing, dev->hard_header==NULL
118    mac_header -> data. ll header is still not built!
119    data       -> data
120
121 Resume
122   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
123
124
125 On transmit:
126 ------------
127
128 dev->hard_header != NULL
129    mac_header -> ll header
130    data       -> ll header
131
132 dev->hard_header == NULL (ll header is added by device, we cannot control it)
133    mac_header -> data
134    data       -> data
135
136    We should set nh.raw on output to correct posistion,
137    packet classifier depends on it.
138  */
139
140 /* Private packet socket structures. */
141
142 struct packet_mclist {
143         struct packet_mclist    *next;
144         int                     ifindex;
145         int                     count;
146         unsigned short          type;
147         unsigned short          alen;
148         unsigned char           addr[MAX_ADDR_LEN];
149 };
150 /* identical to struct packet_mreq except it has
151  * a longer address field.
152  */
153 struct packet_mreq_max {
154         int             mr_ifindex;
155         unsigned short  mr_type;
156         unsigned short  mr_alen;
157         unsigned char   mr_address[MAX_ADDR_LEN];
158 };
159
160 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
161                 int closing, int tx_ring);
162
163 struct packet_ring_buffer {
164         char                    **pg_vec;
165         unsigned int            head;
166         unsigned int            frames_per_block;
167         unsigned int            frame_size;
168         unsigned int            frame_max;
169
170         unsigned int            pg_vec_order;
171         unsigned int            pg_vec_pages;
172         unsigned int            pg_vec_len;
173
174         atomic_t                pending;
175 };
176
177 struct packet_sock;
178 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
179
180 static void packet_flush_mclist(struct sock *sk);
181
182 struct packet_sock {
183         /* struct sock has to be the first member of packet_sock */
184         struct sock             sk;
185         struct tpacket_stats    stats;
186         struct packet_ring_buffer       rx_ring;
187         struct packet_ring_buffer       tx_ring;
188         int                     copy_thresh;
189         spinlock_t              bind_lock;
190         struct mutex            pg_vec_lock;
191         unsigned int            running:1,      /* prot_hook is attached*/
192                                 auxdata:1,
193                                 origdev:1,
194                                 has_vnet_hdr:1;
195         int                     ifindex;        /* bound device         */
196         __be16                  num;
197         struct packet_mclist    *mclist;
198         atomic_t                mapped;
199         enum tpacket_versions   tp_version;
200         unsigned int            tp_hdrlen;
201         unsigned int            tp_reserve;
202         unsigned int            tp_loss:1;
203         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
204 };
205
206 struct packet_skb_cb {
207         unsigned int origlen;
208         union {
209                 struct sockaddr_pkt pkt;
210                 struct sockaddr_ll ll;
211         } sa;
212 };
213
214 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
215
216 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
217 {
218         union {
219                 struct tpacket_hdr *h1;
220                 struct tpacket2_hdr *h2;
221                 void *raw;
222         } h;
223
224         h.raw = frame;
225         switch (po->tp_version) {
226         case TPACKET_V1:
227                 h.h1->tp_status = status;
228                 flush_dcache_page(virt_to_page(&h.h1->tp_status));
229                 break;
230         case TPACKET_V2:
231                 h.h2->tp_status = status;
232                 flush_dcache_page(virt_to_page(&h.h2->tp_status));
233                 break;
234         default:
235                 pr_err("TPACKET version not supported\n");
236                 BUG();
237         }
238
239         smp_wmb();
240 }
241
242 static int __packet_get_status(struct packet_sock *po, void *frame)
243 {
244         union {
245                 struct tpacket_hdr *h1;
246                 struct tpacket2_hdr *h2;
247                 void *raw;
248         } h;
249
250         smp_rmb();
251
252         h.raw = frame;
253         switch (po->tp_version) {
254         case TPACKET_V1:
255                 flush_dcache_page(virt_to_page(&h.h1->tp_status));
256                 return h.h1->tp_status;
257         case TPACKET_V2:
258                 flush_dcache_page(virt_to_page(&h.h2->tp_status));
259                 return h.h2->tp_status;
260         default:
261                 pr_err("TPACKET version not supported\n");
262                 BUG();
263                 return 0;
264         }
265 }
266
267 static void *packet_lookup_frame(struct packet_sock *po,
268                 struct packet_ring_buffer *rb,
269                 unsigned int position,
270                 int status)
271 {
272         unsigned int pg_vec_pos, frame_offset;
273         union {
274                 struct tpacket_hdr *h1;
275                 struct tpacket2_hdr *h2;
276                 void *raw;
277         } h;
278
279         pg_vec_pos = position / rb->frames_per_block;
280         frame_offset = position % rb->frames_per_block;
281
282         h.raw = rb->pg_vec[pg_vec_pos] + (frame_offset * rb->frame_size);
283
284         if (status != __packet_get_status(po, h.raw))
285                 return NULL;
286
287         return h.raw;
288 }
289
290 static inline void *packet_current_frame(struct packet_sock *po,
291                 struct packet_ring_buffer *rb,
292                 int status)
293 {
294         return packet_lookup_frame(po, rb, rb->head, status);
295 }
296
297 static inline void *packet_previous_frame(struct packet_sock *po,
298                 struct packet_ring_buffer *rb,
299                 int status)
300 {
301         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
302         return packet_lookup_frame(po, rb, previous, status);
303 }
304
305 static inline void packet_increment_head(struct packet_ring_buffer *buff)
306 {
307         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
308 }
309
310 static inline struct packet_sock *pkt_sk(struct sock *sk)
311 {
312         return (struct packet_sock *)sk;
313 }
314
315 static void packet_sock_destruct(struct sock *sk)
316 {
317         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
318         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
319
320         if (!sock_flag(sk, SOCK_DEAD)) {
321                 pr_err("Attempt to release alive packet socket: %p\n", sk);
322                 return;
323         }
324
325         sk_refcnt_debug_dec(sk);
326 }
327
328
329 static const struct proto_ops packet_ops;
330
331 static const struct proto_ops packet_ops_spkt;
332
333 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
334                            struct packet_type *pt, struct net_device *orig_dev)
335 {
336         struct sock *sk;
337         struct sockaddr_pkt *spkt;
338
339         /*
340          *      When we registered the protocol we saved the socket in the data
341          *      field for just this event.
342          */
343
344         sk = pt->af_packet_priv;
345
346         /*
347          *      Yank back the headers [hope the device set this
348          *      right or kerboom...]
349          *
350          *      Incoming packets have ll header pulled,
351          *      push it back.
352          *
353          *      For outgoing ones skb->data == skb_mac_header(skb)
354          *      so that this procedure is noop.
355          */
356
357         if (skb->pkt_type == PACKET_LOOPBACK)
358                 goto out;
359
360         if (!net_eq(dev_net(dev), sock_net(sk)))
361                 goto out;
362
363         skb = skb_share_check(skb, GFP_ATOMIC);
364         if (skb == NULL)
365                 goto oom;
366
367         /* drop any routing info */
368         skb_dst_drop(skb);
369
370         /* drop conntrack reference */
371         nf_reset(skb);
372
373         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
374
375         skb_push(skb, skb->data - skb_mac_header(skb));
376
377         /*
378          *      The SOCK_PACKET socket receives _all_ frames.
379          */
380
381         spkt->spkt_family = dev->type;
382         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
383         spkt->spkt_protocol = skb->protocol;
384
385         /*
386          *      Charge the memory to the socket. This is done specifically
387          *      to prevent sockets using all the memory up.
388          */
389
390         if (sock_queue_rcv_skb(sk, skb) == 0)
391                 return 0;
392
393 out:
394         kfree_skb(skb);
395 oom:
396         return 0;
397 }
398
399
400 /*
401  *      Output a raw packet to a device layer. This bypasses all the other
402  *      protocol layers and you must therefore supply it with a complete frame
403  */
404
405 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
406                                struct msghdr *msg, size_t len)
407 {
408         struct sock *sk = sock->sk;
409         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
410         struct sk_buff *skb = NULL;
411         struct net_device *dev;
412         __be16 proto = 0;
413         int err;
414
415         /*
416          *      Get and verify the address.
417          */
418
419         if (saddr) {
420                 if (msg->msg_namelen < sizeof(struct sockaddr))
421                         return -EINVAL;
422                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
423                         proto = saddr->spkt_protocol;
424         } else
425                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
426
427         /*
428          *      Find the device first to size check it
429          */
430
431         saddr->spkt_device[13] = 0;
432 retry:
433         rcu_read_lock();
434         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
435         err = -ENODEV;
436         if (dev == NULL)
437                 goto out_unlock;
438
439         err = -ENETDOWN;
440         if (!(dev->flags & IFF_UP))
441                 goto out_unlock;
442
443         /*
444          * You may not queue a frame bigger than the mtu. This is the lowest level
445          * raw protocol and you must do your own fragmentation at this level.
446          */
447
448         err = -EMSGSIZE;
449         if (len > dev->mtu + dev->hard_header_len)
450                 goto out_unlock;
451
452         if (!skb) {
453                 size_t reserved = LL_RESERVED_SPACE(dev);
454                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
455
456                 rcu_read_unlock();
457                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
458                 if (skb == NULL)
459                         return -ENOBUFS;
460                 /* FIXME: Save some space for broken drivers that write a hard
461                  * header at transmission time by themselves. PPP is the notable
462                  * one here. This should really be fixed at the driver level.
463                  */
464                 skb_reserve(skb, reserved);
465                 skb_reset_network_header(skb);
466
467                 /* Try to align data part correctly */
468                 if (hhlen) {
469                         skb->data -= hhlen;
470                         skb->tail -= hhlen;
471                         if (len < hhlen)
472                                 skb_reset_network_header(skb);
473                 }
474                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
475                 if (err)
476                         goto out_free;
477                 goto retry;
478         }
479
480
481         skb->protocol = proto;
482         skb->dev = dev;
483         skb->priority = sk->sk_priority;
484         skb->mark = sk->sk_mark;
485
486         dev_queue_xmit(skb);
487         rcu_read_unlock();
488         return len;
489
490 out_unlock:
491         rcu_read_unlock();
492 out_free:
493         kfree_skb(skb);
494         return err;
495 }
496
497 static inline unsigned int run_filter(struct sk_buff *skb, struct sock *sk,
498                                       unsigned int res)
499 {
500         struct sk_filter *filter;
501
502         rcu_read_lock_bh();
503         filter = rcu_dereference(sk->sk_filter);
504         if (filter != NULL)
505                 res = sk_run_filter(skb, filter->insns, filter->len);
506         rcu_read_unlock_bh();
507
508         return res;
509 }
510
511 /*
512    This function makes lazy skb cloning in hope that most of packets
513    are discarded by BPF.
514
515    Note tricky part: we DO mangle shared skb! skb->data, skb->len
516    and skb->cb are mangled. It works because (and until) packets
517    falling here are owned by current CPU. Output packets are cloned
518    by dev_queue_xmit_nit(), input packets are processed by net_bh
519    sequencially, so that if we return skb to original state on exit,
520    we will not harm anyone.
521  */
522
523 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
524                       struct packet_type *pt, struct net_device *orig_dev)
525 {
526         struct sock *sk;
527         struct sockaddr_ll *sll;
528         struct packet_sock *po;
529         u8 *skb_head = skb->data;
530         int skb_len = skb->len;
531         unsigned int snaplen, res;
532
533         if (skb->pkt_type == PACKET_LOOPBACK)
534                 goto drop;
535
536         sk = pt->af_packet_priv;
537         po = pkt_sk(sk);
538
539         if (!net_eq(dev_net(dev), sock_net(sk)))
540                 goto drop;
541
542         skb->dev = dev;
543
544         if (dev->header_ops) {
545                 /* The device has an explicit notion of ll header,
546                    exported to higher levels.
547
548                    Otherwise, the device hides datails of it frame
549                    structure, so that corresponding packet head
550                    never delivered to user.
551                  */
552                 if (sk->sk_type != SOCK_DGRAM)
553                         skb_push(skb, skb->data - skb_mac_header(skb));
554                 else if (skb->pkt_type == PACKET_OUTGOING) {
555                         /* Special case: outgoing packets have ll header at head */
556                         skb_pull(skb, skb_network_offset(skb));
557                 }
558         }
559
560         snaplen = skb->len;
561
562         res = run_filter(skb, sk, snaplen);
563         if (!res)
564                 goto drop_n_restore;
565         if (snaplen > res)
566                 snaplen = res;
567
568         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
569             (unsigned)sk->sk_rcvbuf)
570                 goto drop_n_acct;
571
572         if (skb_shared(skb)) {
573                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
574                 if (nskb == NULL)
575                         goto drop_n_acct;
576
577                 if (skb_head != skb->data) {
578                         skb->data = skb_head;
579                         skb->len = skb_len;
580                 }
581                 kfree_skb(skb);
582                 skb = nskb;
583         }
584
585         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
586                      sizeof(skb->cb));
587
588         sll = &PACKET_SKB_CB(skb)->sa.ll;
589         sll->sll_family = AF_PACKET;
590         sll->sll_hatype = dev->type;
591         sll->sll_protocol = skb->protocol;
592         sll->sll_pkttype = skb->pkt_type;
593         if (unlikely(po->origdev))
594                 sll->sll_ifindex = orig_dev->ifindex;
595         else
596                 sll->sll_ifindex = dev->ifindex;
597
598         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
599
600         PACKET_SKB_CB(skb)->origlen = skb->len;
601
602         if (pskb_trim(skb, snaplen))
603                 goto drop_n_acct;
604
605         skb_set_owner_r(skb, sk);
606         skb->dev = NULL;
607         skb_dst_drop(skb);
608
609         /* drop conntrack reference */
610         nf_reset(skb);
611
612         spin_lock(&sk->sk_receive_queue.lock);
613         po->stats.tp_packets++;
614         skb->dropcount = atomic_read(&sk->sk_drops);
615         __skb_queue_tail(&sk->sk_receive_queue, skb);
616         spin_unlock(&sk->sk_receive_queue.lock);
617         sk->sk_data_ready(sk, skb->len);
618         return 0;
619
620 drop_n_acct:
621         po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
622
623 drop_n_restore:
624         if (skb_head != skb->data && skb_shared(skb)) {
625                 skb->data = skb_head;
626                 skb->len = skb_len;
627         }
628 drop:
629         consume_skb(skb);
630         return 0;
631 }
632
633 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
634                        struct packet_type *pt, struct net_device *orig_dev)
635 {
636         struct sock *sk;
637         struct packet_sock *po;
638         struct sockaddr_ll *sll;
639         union {
640                 struct tpacket_hdr *h1;
641                 struct tpacket2_hdr *h2;
642                 void *raw;
643         } h;
644         u8 *skb_head = skb->data;
645         int skb_len = skb->len;
646         unsigned int snaplen, res;
647         unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
648         unsigned short macoff, netoff, hdrlen;
649         struct sk_buff *copy_skb = NULL;
650         struct timeval tv;
651         struct timespec ts;
652
653         if (skb->pkt_type == PACKET_LOOPBACK)
654                 goto drop;
655
656         sk = pt->af_packet_priv;
657         po = pkt_sk(sk);
658
659         if (!net_eq(dev_net(dev), sock_net(sk)))
660                 goto drop;
661
662         if (dev->header_ops) {
663                 if (sk->sk_type != SOCK_DGRAM)
664                         skb_push(skb, skb->data - skb_mac_header(skb));
665                 else if (skb->pkt_type == PACKET_OUTGOING) {
666                         /* Special case: outgoing packets have ll header at head */
667                         skb_pull(skb, skb_network_offset(skb));
668                 }
669         }
670
671         if (skb->ip_summed == CHECKSUM_PARTIAL)
672                 status |= TP_STATUS_CSUMNOTREADY;
673
674         snaplen = skb->len;
675
676         res = run_filter(skb, sk, snaplen);
677         if (!res)
678                 goto drop_n_restore;
679         if (snaplen > res)
680                 snaplen = res;
681
682         if (sk->sk_type == SOCK_DGRAM) {
683                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
684                                   po->tp_reserve;
685         } else {
686                 unsigned maclen = skb_network_offset(skb);
687                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
688                                        (maclen < 16 ? 16 : maclen)) +
689                         po->tp_reserve;
690                 macoff = netoff - maclen;
691         }
692
693         if (macoff + snaplen > po->rx_ring.frame_size) {
694                 if (po->copy_thresh &&
695                     atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
696                     (unsigned)sk->sk_rcvbuf) {
697                         if (skb_shared(skb)) {
698                                 copy_skb = skb_clone(skb, GFP_ATOMIC);
699                         } else {
700                                 copy_skb = skb_get(skb);
701                                 skb_head = skb->data;
702                         }
703                         if (copy_skb)
704                                 skb_set_owner_r(copy_skb, sk);
705                 }
706                 snaplen = po->rx_ring.frame_size - macoff;
707                 if ((int)snaplen < 0)
708                         snaplen = 0;
709         }
710
711         spin_lock(&sk->sk_receive_queue.lock);
712         h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
713         if (!h.raw)
714                 goto ring_is_full;
715         packet_increment_head(&po->rx_ring);
716         po->stats.tp_packets++;
717         if (copy_skb) {
718                 status |= TP_STATUS_COPY;
719                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
720         }
721         if (!po->stats.tp_drops)
722                 status &= ~TP_STATUS_LOSING;
723         spin_unlock(&sk->sk_receive_queue.lock);
724
725         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
726
727         switch (po->tp_version) {
728         case TPACKET_V1:
729                 h.h1->tp_len = skb->len;
730                 h.h1->tp_snaplen = snaplen;
731                 h.h1->tp_mac = macoff;
732                 h.h1->tp_net = netoff;
733                 if (skb->tstamp.tv64)
734                         tv = ktime_to_timeval(skb->tstamp);
735                 else
736                         do_gettimeofday(&tv);
737                 h.h1->tp_sec = tv.tv_sec;
738                 h.h1->tp_usec = tv.tv_usec;
739                 hdrlen = sizeof(*h.h1);
740                 break;
741         case TPACKET_V2:
742                 h.h2->tp_len = skb->len;
743                 h.h2->tp_snaplen = snaplen;
744                 h.h2->tp_mac = macoff;
745                 h.h2->tp_net = netoff;
746                 if (skb->tstamp.tv64)
747                         ts = ktime_to_timespec(skb->tstamp);
748                 else
749                         getnstimeofday(&ts);
750                 h.h2->tp_sec = ts.tv_sec;
751                 h.h2->tp_nsec = ts.tv_nsec;
752                 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
753                 hdrlen = sizeof(*h.h2);
754                 break;
755         default:
756                 BUG();
757         }
758
759         sll = h.raw + TPACKET_ALIGN(hdrlen);
760         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
761         sll->sll_family = AF_PACKET;
762         sll->sll_hatype = dev->type;
763         sll->sll_protocol = skb->protocol;
764         sll->sll_pkttype = skb->pkt_type;
765         if (unlikely(po->origdev))
766                 sll->sll_ifindex = orig_dev->ifindex;
767         else
768                 sll->sll_ifindex = dev->ifindex;
769
770         __packet_set_status(po, h.raw, status);
771         smp_mb();
772         {
773                 struct page *p_start, *p_end;
774                 u8 *h_end = h.raw + macoff + snaplen - 1;
775
776                 p_start = virt_to_page(h.raw);
777                 p_end = virt_to_page(h_end);
778                 while (p_start <= p_end) {
779                         flush_dcache_page(p_start);
780                         p_start++;
781                 }
782         }
783
784         sk->sk_data_ready(sk, 0);
785
786 drop_n_restore:
787         if (skb_head != skb->data && skb_shared(skb)) {
788                 skb->data = skb_head;
789                 skb->len = skb_len;
790         }
791 drop:
792         kfree_skb(skb);
793         return 0;
794
795 ring_is_full:
796         po->stats.tp_drops++;
797         spin_unlock(&sk->sk_receive_queue.lock);
798
799         sk->sk_data_ready(sk, 0);
800         kfree_skb(copy_skb);
801         goto drop_n_restore;
802 }
803
804 static void tpacket_destruct_skb(struct sk_buff *skb)
805 {
806         struct packet_sock *po = pkt_sk(skb->sk);
807         void *ph;
808
809         BUG_ON(skb == NULL);
810
811         if (likely(po->tx_ring.pg_vec)) {
812                 ph = skb_shinfo(skb)->destructor_arg;
813                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
814                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
815                 atomic_dec(&po->tx_ring.pending);
816                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
817         }
818
819         sock_wfree(skb);
820 }
821
822 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
823                 void *frame, struct net_device *dev, int size_max,
824                 __be16 proto, unsigned char *addr)
825 {
826         union {
827                 struct tpacket_hdr *h1;
828                 struct tpacket2_hdr *h2;
829                 void *raw;
830         } ph;
831         int to_write, offset, len, tp_len, nr_frags, len_max;
832         struct socket *sock = po->sk.sk_socket;
833         struct page *page;
834         void *data;
835         int err;
836
837         ph.raw = frame;
838
839         skb->protocol = proto;
840         skb->dev = dev;
841         skb->priority = po->sk.sk_priority;
842         skb->mark = po->sk.sk_mark;
843         skb_shinfo(skb)->destructor_arg = ph.raw;
844
845         switch (po->tp_version) {
846         case TPACKET_V2:
847                 tp_len = ph.h2->tp_len;
848                 break;
849         default:
850                 tp_len = ph.h1->tp_len;
851                 break;
852         }
853         if (unlikely(tp_len > size_max)) {
854                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
855                 return -EMSGSIZE;
856         }
857
858         skb_reserve(skb, LL_RESERVED_SPACE(dev));
859         skb_reset_network_header(skb);
860
861         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
862         to_write = tp_len;
863
864         if (sock->type == SOCK_DGRAM) {
865                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
866                                 NULL, tp_len);
867                 if (unlikely(err < 0))
868                         return -EINVAL;
869         } else if (dev->hard_header_len) {
870                 /* net device doesn't like empty head */
871                 if (unlikely(tp_len <= dev->hard_header_len)) {
872                         pr_err("packet size is too short (%d < %d)\n",
873                                tp_len, dev->hard_header_len);
874                         return -EINVAL;
875                 }
876
877                 skb_push(skb, dev->hard_header_len);
878                 err = skb_store_bits(skb, 0, data,
879                                 dev->hard_header_len);
880                 if (unlikely(err))
881                         return err;
882
883                 data += dev->hard_header_len;
884                 to_write -= dev->hard_header_len;
885         }
886
887         err = -EFAULT;
888         page = virt_to_page(data);
889         offset = offset_in_page(data);
890         len_max = PAGE_SIZE - offset;
891         len = ((to_write > len_max) ? len_max : to_write);
892
893         skb->data_len = to_write;
894         skb->len += to_write;
895         skb->truesize += to_write;
896         atomic_add(to_write, &po->sk.sk_wmem_alloc);
897
898         while (likely(to_write)) {
899                 nr_frags = skb_shinfo(skb)->nr_frags;
900
901                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
902                         pr_err("Packet exceed the number of skb frags(%lu)\n",
903                                MAX_SKB_FRAGS);
904                         return -EFAULT;
905                 }
906
907                 flush_dcache_page(page);
908                 get_page(page);
909                 skb_fill_page_desc(skb,
910                                 nr_frags,
911                                 page++, offset, len);
912                 to_write -= len;
913                 offset = 0;
914                 len_max = PAGE_SIZE;
915                 len = ((to_write > len_max) ? len_max : to_write);
916         }
917
918         return tp_len;
919 }
920
921 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
922 {
923         struct socket *sock;
924         struct sk_buff *skb;
925         struct net_device *dev;
926         __be16 proto;
927         int ifindex, err, reserve = 0;
928         void *ph;
929         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
930         int tp_len, size_max;
931         unsigned char *addr;
932         int len_sum = 0;
933         int status = 0;
934
935         sock = po->sk.sk_socket;
936
937         mutex_lock(&po->pg_vec_lock);
938
939         err = -EBUSY;
940         if (saddr == NULL) {
941                 ifindex = po->ifindex;
942                 proto   = po->num;
943                 addr    = NULL;
944         } else {
945                 err = -EINVAL;
946                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
947                         goto out;
948                 if (msg->msg_namelen < (saddr->sll_halen
949                                         + offsetof(struct sockaddr_ll,
950                                                 sll_addr)))
951                         goto out;
952                 ifindex = saddr->sll_ifindex;
953                 proto   = saddr->sll_protocol;
954                 addr    = saddr->sll_addr;
955         }
956
957         dev = dev_get_by_index(sock_net(&po->sk), ifindex);
958         err = -ENXIO;
959         if (unlikely(dev == NULL))
960                 goto out;
961
962         reserve = dev->hard_header_len;
963
964         err = -ENETDOWN;
965         if (unlikely(!(dev->flags & IFF_UP)))
966                 goto out_put;
967
968         size_max = po->tx_ring.frame_size
969                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
970
971         if (size_max > dev->mtu + reserve)
972                 size_max = dev->mtu + reserve;
973
974         do {
975                 ph = packet_current_frame(po, &po->tx_ring,
976                                 TP_STATUS_SEND_REQUEST);
977
978                 if (unlikely(ph == NULL)) {
979                         schedule();
980                         continue;
981                 }
982
983                 status = TP_STATUS_SEND_REQUEST;
984                 skb = sock_alloc_send_skb(&po->sk,
985                                 LL_ALLOCATED_SPACE(dev)
986                                 + sizeof(struct sockaddr_ll),
987                                 0, &err);
988
989                 if (unlikely(skb == NULL))
990                         goto out_status;
991
992                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
993                                 addr);
994
995                 if (unlikely(tp_len < 0)) {
996                         if (po->tp_loss) {
997                                 __packet_set_status(po, ph,
998                                                 TP_STATUS_AVAILABLE);
999                                 packet_increment_head(&po->tx_ring);
1000                                 kfree_skb(skb);
1001                                 continue;
1002                         } else {
1003                                 status = TP_STATUS_WRONG_FORMAT;
1004                                 err = tp_len;
1005                                 goto out_status;
1006                         }
1007                 }
1008
1009                 skb->destructor = tpacket_destruct_skb;
1010                 __packet_set_status(po, ph, TP_STATUS_SENDING);
1011                 atomic_inc(&po->tx_ring.pending);
1012
1013                 status = TP_STATUS_SEND_REQUEST;
1014                 err = dev_queue_xmit(skb);
1015                 if (unlikely(err > 0)) {
1016                         err = net_xmit_errno(err);
1017                         if (err && __packet_get_status(po, ph) ==
1018                                    TP_STATUS_AVAILABLE) {
1019                                 /* skb was destructed already */
1020                                 skb = NULL;
1021                                 goto out_status;
1022                         }
1023                         /*
1024                          * skb was dropped but not destructed yet;
1025                          * let's treat it like congestion or err < 0
1026                          */
1027                         err = 0;
1028                 }
1029                 packet_increment_head(&po->tx_ring);
1030                 len_sum += tp_len;
1031         } while (likely((ph != NULL) ||
1032                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1033                          (atomic_read(&po->tx_ring.pending))))
1034                 );
1035
1036         err = len_sum;
1037         goto out_put;
1038
1039 out_status:
1040         __packet_set_status(po, ph, status);
1041         kfree_skb(skb);
1042 out_put:
1043         dev_put(dev);
1044 out:
1045         mutex_unlock(&po->pg_vec_lock);
1046         return err;
1047 }
1048
1049 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1050                                                size_t reserve, size_t len,
1051                                                size_t linear, int noblock,
1052                                                int *err)
1053 {
1054         struct sk_buff *skb;
1055
1056         /* Under a page?  Don't bother with paged skb. */
1057         if (prepad + len < PAGE_SIZE || !linear)
1058                 linear = len;
1059
1060         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1061                                    err);
1062         if (!skb)
1063                 return NULL;
1064
1065         skb_reserve(skb, reserve);
1066         skb_put(skb, linear);
1067         skb->data_len = len - linear;
1068         skb->len += len - linear;
1069
1070         return skb;
1071 }
1072
1073 static int packet_snd(struct socket *sock,
1074                           struct msghdr *msg, size_t len)
1075 {
1076         struct sock *sk = sock->sk;
1077         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1078         struct sk_buff *skb;
1079         struct net_device *dev;
1080         __be16 proto;
1081         unsigned char *addr;
1082         int ifindex, err, reserve = 0;
1083         struct virtio_net_hdr vnet_hdr = { 0 };
1084         int offset = 0;
1085         int vnet_hdr_len;
1086         struct packet_sock *po = pkt_sk(sk);
1087         unsigned short gso_type = 0;
1088
1089         /*
1090          *      Get and verify the address.
1091          */
1092
1093         if (saddr == NULL) {
1094                 ifindex = po->ifindex;
1095                 proto   = po->num;
1096                 addr    = NULL;
1097         } else {
1098                 err = -EINVAL;
1099                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1100                         goto out;
1101                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1102                         goto out;
1103                 ifindex = saddr->sll_ifindex;
1104                 proto   = saddr->sll_protocol;
1105                 addr    = saddr->sll_addr;
1106         }
1107
1108
1109         dev = dev_get_by_index(sock_net(sk), ifindex);
1110         err = -ENXIO;
1111         if (dev == NULL)
1112                 goto out_unlock;
1113         if (sock->type == SOCK_RAW)
1114                 reserve = dev->hard_header_len;
1115
1116         err = -ENETDOWN;
1117         if (!(dev->flags & IFF_UP))
1118                 goto out_unlock;
1119
1120         if (po->has_vnet_hdr) {
1121                 vnet_hdr_len = sizeof(vnet_hdr);
1122
1123                 err = -EINVAL;
1124                 if (len < vnet_hdr_len)
1125                         goto out_unlock;
1126
1127                 len -= vnet_hdr_len;
1128
1129                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1130                                        vnet_hdr_len);
1131                 if (err < 0)
1132                         goto out_unlock;
1133
1134                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1135                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1136                       vnet_hdr.hdr_len))
1137                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
1138                                                  vnet_hdr.csum_offset + 2;
1139
1140                 err = -EINVAL;
1141                 if (vnet_hdr.hdr_len > len)
1142                         goto out_unlock;
1143
1144                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1145                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1146                         case VIRTIO_NET_HDR_GSO_TCPV4:
1147                                 gso_type = SKB_GSO_TCPV4;
1148                                 break;
1149                         case VIRTIO_NET_HDR_GSO_TCPV6:
1150                                 gso_type = SKB_GSO_TCPV6;
1151                                 break;
1152                         case VIRTIO_NET_HDR_GSO_UDP:
1153                                 gso_type = SKB_GSO_UDP;
1154                                 break;
1155                         default:
1156                                 goto out_unlock;
1157                         }
1158
1159                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1160                                 gso_type |= SKB_GSO_TCP_ECN;
1161
1162                         if (vnet_hdr.gso_size == 0)
1163                                 goto out_unlock;
1164
1165                 }
1166         }
1167
1168         err = -EMSGSIZE;
1169         if (!gso_type && (len > dev->mtu+reserve))
1170                 goto out_unlock;
1171
1172         err = -ENOBUFS;
1173         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1174                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1175                                msg->msg_flags & MSG_DONTWAIT, &err);
1176         if (skb == NULL)
1177                 goto out_unlock;
1178
1179         skb_set_network_header(skb, reserve);
1180
1181         err = -EINVAL;
1182         if (sock->type == SOCK_DGRAM &&
1183             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1184                 goto out_free;
1185
1186         /* Returns -EFAULT on error */
1187         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1188         if (err)
1189                 goto out_free;
1190
1191         skb->protocol = proto;
1192         skb->dev = dev;
1193         skb->priority = sk->sk_priority;
1194         skb->mark = sk->sk_mark;
1195
1196         if (po->has_vnet_hdr) {
1197                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1198                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1199                                                   vnet_hdr.csum_offset)) {
1200                                 err = -EINVAL;
1201                                 goto out_free;
1202                         }
1203                 }
1204
1205                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1206                 skb_shinfo(skb)->gso_type = gso_type;
1207
1208                 /* Header must be checked, and gso_segs computed. */
1209                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1210                 skb_shinfo(skb)->gso_segs = 0;
1211
1212                 len += vnet_hdr_len;
1213         }
1214
1215         /*
1216          *      Now send it
1217          */
1218
1219         err = dev_queue_xmit(skb);
1220         if (err > 0 && (err = net_xmit_errno(err)) != 0)
1221                 goto out_unlock;
1222
1223         dev_put(dev);
1224
1225         return len;
1226
1227 out_free:
1228         kfree_skb(skb);
1229 out_unlock:
1230         if (dev)
1231                 dev_put(dev);
1232 out:
1233         return err;
1234 }
1235
1236 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1237                 struct msghdr *msg, size_t len)
1238 {
1239         struct sock *sk = sock->sk;
1240         struct packet_sock *po = pkt_sk(sk);
1241         if (po->tx_ring.pg_vec)
1242                 return tpacket_snd(po, msg);
1243         else
1244                 return packet_snd(sock, msg, len);
1245 }
1246
1247 /*
1248  *      Close a PACKET socket. This is fairly simple. We immediately go
1249  *      to 'closed' state and remove our protocol entry in the device list.
1250  */
1251
1252 static int packet_release(struct socket *sock)
1253 {
1254         struct sock *sk = sock->sk;
1255         struct packet_sock *po;
1256         struct net *net;
1257         struct tpacket_req req;
1258
1259         if (!sk)
1260                 return 0;
1261
1262         net = sock_net(sk);
1263         po = pkt_sk(sk);
1264
1265         spin_lock_bh(&net->packet.sklist_lock);
1266         sk_del_node_init_rcu(sk);
1267         sock_prot_inuse_add(net, sk->sk_prot, -1);
1268         spin_unlock_bh(&net->packet.sklist_lock);
1269
1270         spin_lock(&po->bind_lock);
1271         if (po->running) {
1272                 /*
1273                  * Remove from protocol table
1274                  */
1275                 po->running = 0;
1276                 po->num = 0;
1277                 __dev_remove_pack(&po->prot_hook);
1278                 __sock_put(sk);
1279         }
1280         spin_unlock(&po->bind_lock);
1281
1282         packet_flush_mclist(sk);
1283
1284         memset(&req, 0, sizeof(req));
1285
1286         if (po->rx_ring.pg_vec)
1287                 packet_set_ring(sk, &req, 1, 0);
1288
1289         if (po->tx_ring.pg_vec)
1290                 packet_set_ring(sk, &req, 1, 1);
1291
1292         synchronize_net();
1293         /*
1294          *      Now the socket is dead. No more input will appear.
1295          */
1296         sock_orphan(sk);
1297         sock->sk = NULL;
1298
1299         /* Purge queues */
1300
1301         skb_queue_purge(&sk->sk_receive_queue);
1302         sk_refcnt_debug_release(sk);
1303
1304         sock_put(sk);
1305         return 0;
1306 }
1307
1308 /*
1309  *      Attach a packet hook.
1310  */
1311
1312 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1313 {
1314         struct packet_sock *po = pkt_sk(sk);
1315         /*
1316          *      Detach an existing hook if present.
1317          */
1318
1319         lock_sock(sk);
1320
1321         spin_lock(&po->bind_lock);
1322         if (po->running) {
1323                 __sock_put(sk);
1324                 po->running = 0;
1325                 po->num = 0;
1326                 spin_unlock(&po->bind_lock);
1327                 dev_remove_pack(&po->prot_hook);
1328                 spin_lock(&po->bind_lock);
1329         }
1330
1331         po->num = protocol;
1332         po->prot_hook.type = protocol;
1333         po->prot_hook.dev = dev;
1334
1335         po->ifindex = dev ? dev->ifindex : 0;
1336
1337         if (protocol == 0)
1338                 goto out_unlock;
1339
1340         if (!dev || (dev->flags & IFF_UP)) {
1341                 dev_add_pack(&po->prot_hook);
1342                 sock_hold(sk);
1343                 po->running = 1;
1344         } else {
1345                 sk->sk_err = ENETDOWN;
1346                 if (!sock_flag(sk, SOCK_DEAD))
1347                         sk->sk_error_report(sk);
1348         }
1349
1350 out_unlock:
1351         spin_unlock(&po->bind_lock);
1352         release_sock(sk);
1353         return 0;
1354 }
1355
1356 /*
1357  *      Bind a packet socket to a device
1358  */
1359
1360 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1361                             int addr_len)
1362 {
1363         struct sock *sk = sock->sk;
1364         char name[15];
1365         struct net_device *dev;
1366         int err = -ENODEV;
1367
1368         /*
1369          *      Check legality
1370          */
1371
1372         if (addr_len != sizeof(struct sockaddr))
1373                 return -EINVAL;
1374         strlcpy(name, uaddr->sa_data, sizeof(name));
1375
1376         dev = dev_get_by_name(sock_net(sk), name);
1377         if (dev) {
1378                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1379                 dev_put(dev);
1380         }
1381         return err;
1382 }
1383
1384 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1385 {
1386         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1387         struct sock *sk = sock->sk;
1388         struct net_device *dev = NULL;
1389         int err;
1390
1391
1392         /*
1393          *      Check legality
1394          */
1395
1396         if (addr_len < sizeof(struct sockaddr_ll))
1397                 return -EINVAL;
1398         if (sll->sll_family != AF_PACKET)
1399                 return -EINVAL;
1400
1401         if (sll->sll_ifindex) {
1402                 err = -ENODEV;
1403                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1404                 if (dev == NULL)
1405                         goto out;
1406         }
1407         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1408         if (dev)
1409                 dev_put(dev);
1410
1411 out:
1412         return err;
1413 }
1414
1415 static struct proto packet_proto = {
1416         .name     = "PACKET",
1417         .owner    = THIS_MODULE,
1418         .obj_size = sizeof(struct packet_sock),
1419 };
1420
1421 /*
1422  *      Create a packet of type SOCK_PACKET.
1423  */
1424
1425 static int packet_create(struct net *net, struct socket *sock, int protocol,
1426                          int kern)
1427 {
1428         struct sock *sk;
1429         struct packet_sock *po;
1430         __be16 proto = (__force __be16)protocol; /* weird, but documented */
1431         int err;
1432
1433         if (!capable(CAP_NET_RAW))
1434                 return -EPERM;
1435         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1436             sock->type != SOCK_PACKET)
1437                 return -ESOCKTNOSUPPORT;
1438
1439         sock->state = SS_UNCONNECTED;
1440
1441         err = -ENOBUFS;
1442         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1443         if (sk == NULL)
1444                 goto out;
1445
1446         sock->ops = &packet_ops;
1447         if (sock->type == SOCK_PACKET)
1448                 sock->ops = &packet_ops_spkt;
1449
1450         sock_init_data(sock, sk);
1451
1452         po = pkt_sk(sk);
1453         sk->sk_family = PF_PACKET;
1454         po->num = proto;
1455
1456         sk->sk_destruct = packet_sock_destruct;
1457         sk_refcnt_debug_inc(sk);
1458
1459         /*
1460          *      Attach a protocol block
1461          */
1462
1463         spin_lock_init(&po->bind_lock);
1464         mutex_init(&po->pg_vec_lock);
1465         po->prot_hook.func = packet_rcv;
1466
1467         if (sock->type == SOCK_PACKET)
1468                 po->prot_hook.func = packet_rcv_spkt;
1469
1470         po->prot_hook.af_packet_priv = sk;
1471
1472         if (proto) {
1473                 po->prot_hook.type = proto;
1474                 dev_add_pack(&po->prot_hook);
1475                 sock_hold(sk);
1476                 po->running = 1;
1477         }
1478
1479         spin_lock_bh(&net->packet.sklist_lock);
1480         sk_add_node_rcu(sk, &net->packet.sklist);
1481         sock_prot_inuse_add(net, &packet_proto, 1);
1482         spin_unlock_bh(&net->packet.sklist_lock);
1483
1484         return 0;
1485 out:
1486         return err;
1487 }
1488
1489 /*
1490  *      Pull a packet from our receive queue and hand it to the user.
1491  *      If necessary we block.
1492  */
1493
1494 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1495                           struct msghdr *msg, size_t len, int flags)
1496 {
1497         struct sock *sk = sock->sk;
1498         struct sk_buff *skb;
1499         int copied, err;
1500         struct sockaddr_ll *sll;
1501         int vnet_hdr_len = 0;
1502
1503         err = -EINVAL;
1504         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1505                 goto out;
1506
1507 #if 0
1508         /* What error should we return now? EUNATTACH? */
1509         if (pkt_sk(sk)->ifindex < 0)
1510                 return -ENODEV;
1511 #endif
1512
1513         /*
1514          *      Call the generic datagram receiver. This handles all sorts
1515          *      of horrible races and re-entrancy so we can forget about it
1516          *      in the protocol layers.
1517          *
1518          *      Now it will return ENETDOWN, if device have just gone down,
1519          *      but then it will block.
1520          */
1521
1522         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1523
1524         /*
1525          *      An error occurred so return it. Because skb_recv_datagram()
1526          *      handles the blocking we don't see and worry about blocking
1527          *      retries.
1528          */
1529
1530         if (skb == NULL)
1531                 goto out;
1532
1533         if (pkt_sk(sk)->has_vnet_hdr) {
1534                 struct virtio_net_hdr vnet_hdr = { 0 };
1535
1536                 err = -EINVAL;
1537                 vnet_hdr_len = sizeof(vnet_hdr);
1538                 if ((len -= vnet_hdr_len) < 0)
1539                         goto out_free;
1540
1541                 if (skb_is_gso(skb)) {
1542                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1543
1544                         /* This is a hint as to how much should be linear. */
1545                         vnet_hdr.hdr_len = skb_headlen(skb);
1546                         vnet_hdr.gso_size = sinfo->gso_size;
1547                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1548                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1549                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1550                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1551                         else if (sinfo->gso_type & SKB_GSO_UDP)
1552                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1553                         else if (sinfo->gso_type & SKB_GSO_FCOE)
1554                                 goto out_free;
1555                         else
1556                                 BUG();
1557                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1558                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1559                 } else
1560                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1561
1562                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1563                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1564                         vnet_hdr.csum_start = skb->csum_start -
1565                                                         skb_headroom(skb);
1566                         vnet_hdr.csum_offset = skb->csum_offset;
1567                 } /* else everything is zero */
1568
1569                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1570                                      vnet_hdr_len);
1571                 if (err < 0)
1572                         goto out_free;
1573         }
1574
1575         /*
1576          *      If the address length field is there to be filled in, we fill
1577          *      it in now.
1578          */
1579
1580         sll = &PACKET_SKB_CB(skb)->sa.ll;
1581         if (sock->type == SOCK_PACKET)
1582                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1583         else
1584                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1585
1586         /*
1587          *      You lose any data beyond the buffer you gave. If it worries a
1588          *      user program they can ask the device for its MTU anyway.
1589          */
1590
1591         copied = skb->len;
1592         if (copied > len) {
1593                 copied = len;
1594                 msg->msg_flags |= MSG_TRUNC;
1595         }
1596
1597         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1598         if (err)
1599                 goto out_free;
1600
1601         sock_recv_ts_and_drops(msg, sk, skb);
1602
1603         if (msg->msg_name)
1604                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1605                        msg->msg_namelen);
1606
1607         if (pkt_sk(sk)->auxdata) {
1608                 struct tpacket_auxdata aux;
1609
1610                 aux.tp_status = TP_STATUS_USER;
1611                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1612                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1613                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1614                 aux.tp_snaplen = skb->len;
1615                 aux.tp_mac = 0;
1616                 aux.tp_net = skb_network_offset(skb);
1617                 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1618
1619                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1620         }
1621
1622         /*
1623          *      Free or return the buffer as appropriate. Again this
1624          *      hides all the races and re-entrancy issues from us.
1625          */
1626         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1627
1628 out_free:
1629         skb_free_datagram(sk, skb);
1630 out:
1631         return err;
1632 }
1633
1634 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1635                                int *uaddr_len, int peer)
1636 {
1637         struct net_device *dev;
1638         struct sock *sk = sock->sk;
1639
1640         if (peer)
1641                 return -EOPNOTSUPP;
1642
1643         uaddr->sa_family = AF_PACKET;
1644         rcu_read_lock();
1645         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1646         if (dev)
1647                 strlcpy(uaddr->sa_data, dev->name, 15);
1648         else
1649                 memset(uaddr->sa_data, 0, 14);
1650         rcu_read_unlock();
1651         *uaddr_len = sizeof(*uaddr);
1652
1653         return 0;
1654 }
1655
1656 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1657                           int *uaddr_len, int peer)
1658 {
1659         struct net_device *dev;
1660         struct sock *sk = sock->sk;
1661         struct packet_sock *po = pkt_sk(sk);
1662         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1663
1664         if (peer)
1665                 return -EOPNOTSUPP;
1666
1667         sll->sll_family = AF_PACKET;
1668         sll->sll_ifindex = po->ifindex;
1669         sll->sll_protocol = po->num;
1670         rcu_read_lock();
1671         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1672         if (dev) {
1673                 sll->sll_hatype = dev->type;
1674                 sll->sll_halen = dev->addr_len;
1675                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1676         } else {
1677                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
1678                 sll->sll_halen = 0;
1679         }
1680         rcu_read_unlock();
1681         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1682
1683         return 0;
1684 }
1685
1686 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1687                          int what)
1688 {
1689         switch (i->type) {
1690         case PACKET_MR_MULTICAST:
1691                 if (what > 0)
1692                         return dev_mc_add(dev, i->addr, i->alen, 0);
1693                 else
1694                         return dev_mc_delete(dev, i->addr, i->alen, 0);
1695                 break;
1696         case PACKET_MR_PROMISC:
1697                 return dev_set_promiscuity(dev, what);
1698                 break;
1699         case PACKET_MR_ALLMULTI:
1700                 return dev_set_allmulti(dev, what);
1701                 break;
1702         case PACKET_MR_UNICAST:
1703                 if (what > 0)
1704                         return dev_unicast_add(dev, i->addr);
1705                 else
1706                         return dev_unicast_delete(dev, i->addr);
1707                 break;
1708         default:
1709                 break;
1710         }
1711         return 0;
1712 }
1713
1714 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1715 {
1716         for ( ; i; i = i->next) {
1717                 if (i->ifindex == dev->ifindex)
1718                         packet_dev_mc(dev, i, what);
1719         }
1720 }
1721
1722 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1723 {
1724         struct packet_sock *po = pkt_sk(sk);
1725         struct packet_mclist *ml, *i;
1726         struct net_device *dev;
1727         int err;
1728
1729         rtnl_lock();
1730
1731         err = -ENODEV;
1732         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1733         if (!dev)
1734                 goto done;
1735
1736         err = -EINVAL;
1737         if (mreq->mr_alen > dev->addr_len)
1738                 goto done;
1739
1740         err = -ENOBUFS;
1741         i = kmalloc(sizeof(*i), GFP_KERNEL);
1742         if (i == NULL)
1743                 goto done;
1744
1745         err = 0;
1746         for (ml = po->mclist; ml; ml = ml->next) {
1747                 if (ml->ifindex == mreq->mr_ifindex &&
1748                     ml->type == mreq->mr_type &&
1749                     ml->alen == mreq->mr_alen &&
1750                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1751                         ml->count++;
1752                         /* Free the new element ... */
1753                         kfree(i);
1754                         goto done;
1755                 }
1756         }
1757
1758         i->type = mreq->mr_type;
1759         i->ifindex = mreq->mr_ifindex;
1760         i->alen = mreq->mr_alen;
1761         memcpy(i->addr, mreq->mr_address, i->alen);
1762         i->count = 1;
1763         i->next = po->mclist;
1764         po->mclist = i;
1765         err = packet_dev_mc(dev, i, 1);
1766         if (err) {
1767                 po->mclist = i->next;
1768                 kfree(i);
1769         }
1770
1771 done:
1772         rtnl_unlock();
1773         return err;
1774 }
1775
1776 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1777 {
1778         struct packet_mclist *ml, **mlp;
1779
1780         rtnl_lock();
1781
1782         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1783                 if (ml->ifindex == mreq->mr_ifindex &&
1784                     ml->type == mreq->mr_type &&
1785                     ml->alen == mreq->mr_alen &&
1786                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1787                         if (--ml->count == 0) {
1788                                 struct net_device *dev;
1789                                 *mlp = ml->next;
1790                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1791                                 if (dev)
1792                                         packet_dev_mc(dev, ml, -1);
1793                                 kfree(ml);
1794                         }
1795                         rtnl_unlock();
1796                         return 0;
1797                 }
1798         }
1799         rtnl_unlock();
1800         return -EADDRNOTAVAIL;
1801 }
1802
1803 static void packet_flush_mclist(struct sock *sk)
1804 {
1805         struct packet_sock *po = pkt_sk(sk);
1806         struct packet_mclist *ml;
1807
1808         if (!po->mclist)
1809                 return;
1810
1811         rtnl_lock();
1812         while ((ml = po->mclist) != NULL) {
1813                 struct net_device *dev;
1814
1815                 po->mclist = ml->next;
1816                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1817                 if (dev != NULL)
1818                         packet_dev_mc(dev, ml, -1);
1819                 kfree(ml);
1820         }
1821         rtnl_unlock();
1822 }
1823
1824 static int
1825 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1826 {
1827         struct sock *sk = sock->sk;
1828         struct packet_sock *po = pkt_sk(sk);
1829         int ret;
1830
1831         if (level != SOL_PACKET)
1832                 return -ENOPROTOOPT;
1833
1834         switch (optname) {
1835         case PACKET_ADD_MEMBERSHIP:
1836         case PACKET_DROP_MEMBERSHIP:
1837         {
1838                 struct packet_mreq_max mreq;
1839                 int len = optlen;
1840                 memset(&mreq, 0, sizeof(mreq));
1841                 if (len < sizeof(struct packet_mreq))
1842                         return -EINVAL;
1843                 if (len > sizeof(mreq))
1844                         len = sizeof(mreq);
1845                 if (copy_from_user(&mreq, optval, len))
1846                         return -EFAULT;
1847                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1848                         return -EINVAL;
1849                 if (optname == PACKET_ADD_MEMBERSHIP)
1850                         ret = packet_mc_add(sk, &mreq);
1851                 else
1852                         ret = packet_mc_drop(sk, &mreq);
1853                 return ret;
1854         }
1855
1856         case PACKET_RX_RING:
1857         case PACKET_TX_RING:
1858         {
1859                 struct tpacket_req req;
1860
1861                 if (optlen < sizeof(req))
1862                         return -EINVAL;
1863                 if (pkt_sk(sk)->has_vnet_hdr)
1864                         return -EINVAL;
1865                 if (copy_from_user(&req, optval, sizeof(req)))
1866                         return -EFAULT;
1867                 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1868         }
1869         case PACKET_COPY_THRESH:
1870         {
1871                 int val;
1872
1873                 if (optlen != sizeof(val))
1874                         return -EINVAL;
1875                 if (copy_from_user(&val, optval, sizeof(val)))
1876                         return -EFAULT;
1877
1878                 pkt_sk(sk)->copy_thresh = val;
1879                 return 0;
1880         }
1881         case PACKET_VERSION:
1882         {
1883                 int val;
1884
1885                 if (optlen != sizeof(val))
1886                         return -EINVAL;
1887                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1888                         return -EBUSY;
1889                 if (copy_from_user(&val, optval, sizeof(val)))
1890                         return -EFAULT;
1891                 switch (val) {
1892                 case TPACKET_V1:
1893                 case TPACKET_V2:
1894                         po->tp_version = val;
1895                         return 0;
1896                 default:
1897                         return -EINVAL;
1898                 }
1899         }
1900         case PACKET_RESERVE:
1901         {
1902                 unsigned int val;
1903
1904                 if (optlen != sizeof(val))
1905                         return -EINVAL;
1906                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1907                         return -EBUSY;
1908                 if (copy_from_user(&val, optval, sizeof(val)))
1909                         return -EFAULT;
1910                 po->tp_reserve = val;
1911                 return 0;
1912         }
1913         case PACKET_LOSS:
1914         {
1915                 unsigned int val;
1916
1917                 if (optlen != sizeof(val))
1918                         return -EINVAL;
1919                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1920                         return -EBUSY;
1921                 if (copy_from_user(&val, optval, sizeof(val)))
1922                         return -EFAULT;
1923                 po->tp_loss = !!val;
1924                 return 0;
1925         }
1926         case PACKET_AUXDATA:
1927         {
1928                 int val;
1929
1930                 if (optlen < sizeof(val))
1931                         return -EINVAL;
1932                 if (copy_from_user(&val, optval, sizeof(val)))
1933                         return -EFAULT;
1934
1935                 po->auxdata = !!val;
1936                 return 0;
1937         }
1938         case PACKET_ORIGDEV:
1939         {
1940                 int val;
1941
1942                 if (optlen < sizeof(val))
1943                         return -EINVAL;
1944                 if (copy_from_user(&val, optval, sizeof(val)))
1945                         return -EFAULT;
1946
1947                 po->origdev = !!val;
1948                 return 0;
1949         }
1950         case PACKET_VNET_HDR:
1951         {
1952                 int val;
1953
1954                 if (sock->type != SOCK_RAW)
1955                         return -EINVAL;
1956                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
1957                         return -EBUSY;
1958                 if (optlen < sizeof(val))
1959                         return -EINVAL;
1960                 if (copy_from_user(&val, optval, sizeof(val)))
1961                         return -EFAULT;
1962
1963                 po->has_vnet_hdr = !!val;
1964                 return 0;
1965         }
1966         default:
1967                 return -ENOPROTOOPT;
1968         }
1969 }
1970
1971 static int packet_getsockopt(struct socket *sock, int level, int optname,
1972                              char __user *optval, int __user *optlen)
1973 {
1974         int len;
1975         int val;
1976         struct sock *sk = sock->sk;
1977         struct packet_sock *po = pkt_sk(sk);
1978         void *data;
1979         struct tpacket_stats st;
1980
1981         if (level != SOL_PACKET)
1982                 return -ENOPROTOOPT;
1983
1984         if (get_user(len, optlen))
1985                 return -EFAULT;
1986
1987         if (len < 0)
1988                 return -EINVAL;
1989
1990         switch (optname) {
1991         case PACKET_STATISTICS:
1992                 if (len > sizeof(struct tpacket_stats))
1993                         len = sizeof(struct tpacket_stats);
1994                 spin_lock_bh(&sk->sk_receive_queue.lock);
1995                 st = po->stats;
1996                 memset(&po->stats, 0, sizeof(st));
1997                 spin_unlock_bh(&sk->sk_receive_queue.lock);
1998                 st.tp_packets += st.tp_drops;
1999
2000                 data = &st;
2001                 break;
2002         case PACKET_AUXDATA:
2003                 if (len > sizeof(int))
2004                         len = sizeof(int);
2005                 val = po->auxdata;
2006
2007                 data = &val;
2008                 break;
2009         case PACKET_ORIGDEV:
2010                 if (len > sizeof(int))
2011                         len = sizeof(int);
2012                 val = po->origdev;
2013
2014                 data = &val;
2015                 break;
2016         case PACKET_VNET_HDR:
2017                 if (len > sizeof(int))
2018                         len = sizeof(int);
2019                 val = po->has_vnet_hdr;
2020
2021                 data = &val;
2022                 break;
2023         case PACKET_VERSION:
2024                 if (len > sizeof(int))
2025                         len = sizeof(int);
2026                 val = po->tp_version;
2027                 data = &val;
2028                 break;
2029         case PACKET_HDRLEN:
2030                 if (len > sizeof(int))
2031                         len = sizeof(int);
2032                 if (copy_from_user(&val, optval, len))
2033                         return -EFAULT;
2034                 switch (val) {
2035                 case TPACKET_V1:
2036                         val = sizeof(struct tpacket_hdr);
2037                         break;
2038                 case TPACKET_V2:
2039                         val = sizeof(struct tpacket2_hdr);
2040                         break;
2041                 default:
2042                         return -EINVAL;
2043                 }
2044                 data = &val;
2045                 break;
2046         case PACKET_RESERVE:
2047                 if (len > sizeof(unsigned int))
2048                         len = sizeof(unsigned int);
2049                 val = po->tp_reserve;
2050                 data = &val;
2051                 break;
2052         case PACKET_LOSS:
2053                 if (len > sizeof(unsigned int))
2054                         len = sizeof(unsigned int);
2055                 val = po->tp_loss;
2056                 data = &val;
2057                 break;
2058         default:
2059                 return -ENOPROTOOPT;
2060         }
2061
2062         if (put_user(len, optlen))
2063                 return -EFAULT;
2064         if (copy_to_user(optval, data, len))
2065                 return -EFAULT;
2066         return 0;
2067 }
2068
2069
2070 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2071 {
2072         struct sock *sk;
2073         struct hlist_node *node;
2074         struct net_device *dev = data;
2075         struct net *net = dev_net(dev);
2076
2077         rcu_read_lock();
2078         sk_for_each_rcu(sk, node, &net->packet.sklist) {
2079                 struct packet_sock *po = pkt_sk(sk);
2080
2081                 switch (msg) {
2082                 case NETDEV_UNREGISTER:
2083                         if (po->mclist)
2084                                 packet_dev_mclist(dev, po->mclist, -1);
2085                         /* fallthrough */
2086
2087                 case NETDEV_DOWN:
2088                         if (dev->ifindex == po->ifindex) {
2089                                 spin_lock(&po->bind_lock);
2090                                 if (po->running) {
2091                                         __dev_remove_pack(&po->prot_hook);
2092                                         __sock_put(sk);
2093                                         po->running = 0;
2094                                         sk->sk_err = ENETDOWN;
2095                                         if (!sock_flag(sk, SOCK_DEAD))
2096                                                 sk->sk_error_report(sk);
2097                                 }
2098                                 if (msg == NETDEV_UNREGISTER) {
2099                                         po->ifindex = -1;
2100                                         po->prot_hook.dev = NULL;
2101                                 }
2102                                 spin_unlock(&po->bind_lock);
2103                         }
2104                         break;
2105                 case NETDEV_UP:
2106                         if (dev->ifindex == po->ifindex) {
2107                                 spin_lock(&po->bind_lock);
2108                                 if (po->num && !po->running) {
2109                                         dev_add_pack(&po->prot_hook);
2110                                         sock_hold(sk);
2111                                         po->running = 1;
2112                                 }
2113                                 spin_unlock(&po->bind_lock);
2114                         }
2115                         break;
2116                 }
2117         }
2118         rcu_read_unlock();
2119         return NOTIFY_DONE;
2120 }
2121
2122
2123 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2124                         unsigned long arg)
2125 {
2126         struct sock *sk = sock->sk;
2127
2128         switch (cmd) {
2129         case SIOCOUTQ:
2130         {
2131                 int amount = sk_wmem_alloc_get(sk);
2132
2133                 return put_user(amount, (int __user *)arg);
2134         }
2135         case SIOCINQ:
2136         {
2137                 struct sk_buff *skb;
2138                 int amount = 0;
2139
2140                 spin_lock_bh(&sk->sk_receive_queue.lock);
2141                 skb = skb_peek(&sk->sk_receive_queue);
2142                 if (skb)
2143                         amount = skb->len;
2144                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2145                 return put_user(amount, (int __user *)arg);
2146         }
2147         case SIOCGSTAMP:
2148                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2149         case SIOCGSTAMPNS:
2150                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2151
2152 #ifdef CONFIG_INET
2153         case SIOCADDRT:
2154         case SIOCDELRT:
2155         case SIOCDARP:
2156         case SIOCGARP:
2157         case SIOCSARP:
2158         case SIOCGIFADDR:
2159         case SIOCSIFADDR:
2160         case SIOCGIFBRDADDR:
2161         case SIOCSIFBRDADDR:
2162         case SIOCGIFNETMASK:
2163         case SIOCSIFNETMASK:
2164         case SIOCGIFDSTADDR:
2165         case SIOCSIFDSTADDR:
2166         case SIOCSIFFLAGS:
2167                 if (!net_eq(sock_net(sk), &init_net))
2168                         return -ENOIOCTLCMD;
2169                 return inet_dgram_ops.ioctl(sock, cmd, arg);
2170 #endif
2171
2172         default:
2173                 return -ENOIOCTLCMD;
2174         }
2175         return 0;
2176 }
2177
2178 static unsigned int packet_poll(struct file *file, struct socket *sock,
2179                                 poll_table *wait)
2180 {
2181         struct sock *sk = sock->sk;
2182         struct packet_sock *po = pkt_sk(sk);
2183         unsigned int mask = datagram_poll(file, sock, wait);
2184
2185         spin_lock_bh(&sk->sk_receive_queue.lock);
2186         if (po->rx_ring.pg_vec) {
2187                 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2188                         mask |= POLLIN | POLLRDNORM;
2189         }
2190         spin_unlock_bh(&sk->sk_receive_queue.lock);
2191         spin_lock_bh(&sk->sk_write_queue.lock);
2192         if (po->tx_ring.pg_vec) {
2193                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2194                         mask |= POLLOUT | POLLWRNORM;
2195         }
2196         spin_unlock_bh(&sk->sk_write_queue.lock);
2197         return mask;
2198 }
2199
2200
2201 /* Dirty? Well, I still did not learn better way to account
2202  * for user mmaps.
2203  */
2204
2205 static void packet_mm_open(struct vm_area_struct *vma)
2206 {
2207         struct file *file = vma->vm_file;
2208         struct socket *sock = file->private_data;
2209         struct sock *sk = sock->sk;
2210
2211         if (sk)
2212                 atomic_inc(&pkt_sk(sk)->mapped);
2213 }
2214
2215 static void packet_mm_close(struct vm_area_struct *vma)
2216 {
2217         struct file *file = vma->vm_file;
2218         struct socket *sock = file->private_data;
2219         struct sock *sk = sock->sk;
2220
2221         if (sk)
2222                 atomic_dec(&pkt_sk(sk)->mapped);
2223 }
2224
2225 static const struct vm_operations_struct packet_mmap_ops = {
2226         .open   =       packet_mm_open,
2227         .close  =       packet_mm_close,
2228 };
2229
2230 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
2231 {
2232         int i;
2233
2234         for (i = 0; i < len; i++) {
2235                 if (likely(pg_vec[i]))
2236                         free_pages((unsigned long) pg_vec[i], order);
2237         }
2238         kfree(pg_vec);
2239 }
2240
2241 static inline char *alloc_one_pg_vec_page(unsigned long order)
2242 {
2243         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO | __GFP_NOWARN;
2244
2245         return (char *) __get_free_pages(gfp_flags, order);
2246 }
2247
2248 static char **alloc_pg_vec(struct tpacket_req *req, int order)
2249 {
2250         unsigned int block_nr = req->tp_block_nr;
2251         char **pg_vec;
2252         int i;
2253
2254         pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
2255         if (unlikely(!pg_vec))
2256                 goto out;
2257
2258         for (i = 0; i < block_nr; i++) {
2259                 pg_vec[i] = alloc_one_pg_vec_page(order);
2260                 if (unlikely(!pg_vec[i]))
2261                         goto out_free_pgvec;
2262         }
2263
2264 out:
2265         return pg_vec;
2266
2267 out_free_pgvec:
2268         free_pg_vec(pg_vec, order, block_nr);
2269         pg_vec = NULL;
2270         goto out;
2271 }
2272
2273 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2274                 int closing, int tx_ring)
2275 {
2276         char **pg_vec = NULL;
2277         struct packet_sock *po = pkt_sk(sk);
2278         int was_running, order = 0;
2279         struct packet_ring_buffer *rb;
2280         struct sk_buff_head *rb_queue;
2281         __be16 num;
2282         int err;
2283
2284         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2285         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2286
2287         err = -EBUSY;
2288         if (!closing) {
2289                 if (atomic_read(&po->mapped))
2290                         goto out;
2291                 if (atomic_read(&rb->pending))
2292                         goto out;
2293         }
2294
2295         if (req->tp_block_nr) {
2296                 /* Sanity tests and some calculations */
2297                 err = -EBUSY;
2298                 if (unlikely(rb->pg_vec))
2299                         goto out;
2300
2301                 switch (po->tp_version) {
2302                 case TPACKET_V1:
2303                         po->tp_hdrlen = TPACKET_HDRLEN;
2304                         break;
2305                 case TPACKET_V2:
2306                         po->tp_hdrlen = TPACKET2_HDRLEN;
2307                         break;
2308                 }
2309
2310                 err = -EINVAL;
2311                 if (unlikely((int)req->tp_block_size <= 0))
2312                         goto out;
2313                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2314                         goto out;
2315                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2316                                         po->tp_reserve))
2317                         goto out;
2318                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2319                         goto out;
2320
2321                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2322                 if (unlikely(rb->frames_per_block <= 0))
2323                         goto out;
2324                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2325                                         req->tp_frame_nr))
2326                         goto out;
2327
2328                 err = -ENOMEM;
2329                 order = get_order(req->tp_block_size);
2330                 pg_vec = alloc_pg_vec(req, order);
2331                 if (unlikely(!pg_vec))
2332                         goto out;
2333         }
2334         /* Done */
2335         else {
2336                 err = -EINVAL;
2337                 if (unlikely(req->tp_frame_nr))
2338                         goto out;
2339         }
2340
2341         lock_sock(sk);
2342
2343         /* Detach socket from network */
2344         spin_lock(&po->bind_lock);
2345         was_running = po->running;
2346         num = po->num;
2347         if (was_running) {
2348                 __dev_remove_pack(&po->prot_hook);
2349                 po->num = 0;
2350                 po->running = 0;
2351                 __sock_put(sk);
2352         }
2353         spin_unlock(&po->bind_lock);
2354
2355         synchronize_net();
2356
2357         err = -EBUSY;
2358         mutex_lock(&po->pg_vec_lock);
2359         if (closing || atomic_read(&po->mapped) == 0) {
2360                 err = 0;
2361 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
2362                 spin_lock_bh(&rb_queue->lock);
2363                 pg_vec = XC(rb->pg_vec, pg_vec);
2364                 rb->frame_max = (req->tp_frame_nr - 1);
2365                 rb->head = 0;
2366                 rb->frame_size = req->tp_frame_size;
2367                 spin_unlock_bh(&rb_queue->lock);
2368
2369                 order = XC(rb->pg_vec_order, order);
2370                 req->tp_block_nr = XC(rb->pg_vec_len, req->tp_block_nr);
2371
2372                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2373                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2374                                                 tpacket_rcv : packet_rcv;
2375                 skb_queue_purge(rb_queue);
2376 #undef XC
2377                 if (atomic_read(&po->mapped))
2378                         pr_err("packet_mmap: vma is busy: %d\n",
2379                                atomic_read(&po->mapped));
2380         }
2381         mutex_unlock(&po->pg_vec_lock);
2382
2383         spin_lock(&po->bind_lock);
2384         if (was_running && !po->running) {
2385                 sock_hold(sk);
2386                 po->running = 1;
2387                 po->num = num;
2388                 dev_add_pack(&po->prot_hook);
2389         }
2390         spin_unlock(&po->bind_lock);
2391
2392         release_sock(sk);
2393
2394         if (pg_vec)
2395                 free_pg_vec(pg_vec, order, req->tp_block_nr);
2396 out:
2397         return err;
2398 }
2399
2400 static int packet_mmap(struct file *file, struct socket *sock,
2401                 struct vm_area_struct *vma)
2402 {
2403         struct sock *sk = sock->sk;
2404         struct packet_sock *po = pkt_sk(sk);
2405         unsigned long size, expected_size;
2406         struct packet_ring_buffer *rb;
2407         unsigned long start;
2408         int err = -EINVAL;
2409         int i;
2410
2411         if (vma->vm_pgoff)
2412                 return -EINVAL;
2413
2414         mutex_lock(&po->pg_vec_lock);
2415
2416         expected_size = 0;
2417         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2418                 if (rb->pg_vec) {
2419                         expected_size += rb->pg_vec_len
2420                                                 * rb->pg_vec_pages
2421                                                 * PAGE_SIZE;
2422                 }
2423         }
2424
2425         if (expected_size == 0)
2426                 goto out;
2427
2428         size = vma->vm_end - vma->vm_start;
2429         if (size != expected_size)
2430                 goto out;
2431
2432         start = vma->vm_start;
2433         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2434                 if (rb->pg_vec == NULL)
2435                         continue;
2436
2437                 for (i = 0; i < rb->pg_vec_len; i++) {
2438                         struct page *page = virt_to_page(rb->pg_vec[i]);
2439                         int pg_num;
2440
2441                         for (pg_num = 0; pg_num < rb->pg_vec_pages;
2442                                         pg_num++, page++) {
2443                                 err = vm_insert_page(vma, start, page);
2444                                 if (unlikely(err))
2445                                         goto out;
2446                                 start += PAGE_SIZE;
2447                         }
2448                 }
2449         }
2450
2451         atomic_inc(&po->mapped);
2452         vma->vm_ops = &packet_mmap_ops;
2453         err = 0;
2454
2455 out:
2456         mutex_unlock(&po->pg_vec_lock);
2457         return err;
2458 }
2459
2460 static const struct proto_ops packet_ops_spkt = {
2461         .family =       PF_PACKET,
2462         .owner =        THIS_MODULE,
2463         .release =      packet_release,
2464         .bind =         packet_bind_spkt,
2465         .connect =      sock_no_connect,
2466         .socketpair =   sock_no_socketpair,
2467         .accept =       sock_no_accept,
2468         .getname =      packet_getname_spkt,
2469         .poll =         datagram_poll,
2470         .ioctl =        packet_ioctl,
2471         .listen =       sock_no_listen,
2472         .shutdown =     sock_no_shutdown,
2473         .setsockopt =   sock_no_setsockopt,
2474         .getsockopt =   sock_no_getsockopt,
2475         .sendmsg =      packet_sendmsg_spkt,
2476         .recvmsg =      packet_recvmsg,
2477         .mmap =         sock_no_mmap,
2478         .sendpage =     sock_no_sendpage,
2479 };
2480
2481 static const struct proto_ops packet_ops = {
2482         .family =       PF_PACKET,
2483         .owner =        THIS_MODULE,
2484         .release =      packet_release,
2485         .bind =         packet_bind,
2486         .connect =      sock_no_connect,
2487         .socketpair =   sock_no_socketpair,
2488         .accept =       sock_no_accept,
2489         .getname =      packet_getname,
2490         .poll =         packet_poll,
2491         .ioctl =        packet_ioctl,
2492         .listen =       sock_no_listen,
2493         .shutdown =     sock_no_shutdown,
2494         .setsockopt =   packet_setsockopt,
2495         .getsockopt =   packet_getsockopt,
2496         .sendmsg =      packet_sendmsg,
2497         .recvmsg =      packet_recvmsg,
2498         .mmap =         packet_mmap,
2499         .sendpage =     sock_no_sendpage,
2500 };
2501
2502 static const struct net_proto_family packet_family_ops = {
2503         .family =       PF_PACKET,
2504         .create =       packet_create,
2505         .owner  =       THIS_MODULE,
2506 };
2507
2508 static struct notifier_block packet_netdev_notifier = {
2509         .notifier_call =        packet_notifier,
2510 };
2511
2512 #ifdef CONFIG_PROC_FS
2513
2514 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2515         __acquires(RCU)
2516 {
2517         struct net *net = seq_file_net(seq);
2518
2519         rcu_read_lock();
2520         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2521 }
2522
2523 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2524 {
2525         struct net *net = seq_file_net(seq);
2526         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2527 }
2528
2529 static void packet_seq_stop(struct seq_file *seq, void *v)
2530         __releases(RCU)
2531 {
2532         rcu_read_unlock();
2533 }
2534
2535 static int packet_seq_show(struct seq_file *seq, void *v)
2536 {
2537         if (v == SEQ_START_TOKEN)
2538                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
2539         else {
2540                 struct sock *s = sk_entry(v);
2541                 const struct packet_sock *po = pkt_sk(s);
2542
2543                 seq_printf(seq,
2544                            "%p %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
2545                            s,
2546                            atomic_read(&s->sk_refcnt),
2547                            s->sk_type,
2548                            ntohs(po->num),
2549                            po->ifindex,
2550                            po->running,
2551                            atomic_read(&s->sk_rmem_alloc),
2552                            sock_i_uid(s),
2553                            sock_i_ino(s));
2554         }
2555
2556         return 0;
2557 }
2558
2559 static const struct seq_operations packet_seq_ops = {
2560         .start  = packet_seq_start,
2561         .next   = packet_seq_next,
2562         .stop   = packet_seq_stop,
2563         .show   = packet_seq_show,
2564 };
2565
2566 static int packet_seq_open(struct inode *inode, struct file *file)
2567 {
2568         return seq_open_net(inode, file, &packet_seq_ops,
2569                             sizeof(struct seq_net_private));
2570 }
2571
2572 static const struct file_operations packet_seq_fops = {
2573         .owner          = THIS_MODULE,
2574         .open           = packet_seq_open,
2575         .read           = seq_read,
2576         .llseek         = seq_lseek,
2577         .release        = seq_release_net,
2578 };
2579
2580 #endif
2581
2582 static int __net_init packet_net_init(struct net *net)
2583 {
2584         spin_lock_init(&net->packet.sklist_lock);
2585         INIT_HLIST_HEAD(&net->packet.sklist);
2586
2587         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2588                 return -ENOMEM;
2589
2590         return 0;
2591 }
2592
2593 static void __net_exit packet_net_exit(struct net *net)
2594 {
2595         proc_net_remove(net, "packet");
2596 }
2597
2598 static struct pernet_operations packet_net_ops = {
2599         .init = packet_net_init,
2600         .exit = packet_net_exit,
2601 };
2602
2603
2604 static void __exit packet_exit(void)
2605 {
2606         unregister_netdevice_notifier(&packet_netdev_notifier);
2607         unregister_pernet_subsys(&packet_net_ops);
2608         sock_unregister(PF_PACKET);
2609         proto_unregister(&packet_proto);
2610 }
2611
2612 static int __init packet_init(void)
2613 {
2614         int rc = proto_register(&packet_proto, 0);
2615
2616         if (rc != 0)
2617                 goto out;
2618
2619         sock_register(&packet_family_ops);
2620         register_pernet_subsys(&packet_net_ops);
2621         register_netdevice_notifier(&packet_netdev_notifier);
2622 out:
2623         return rc;
2624 }
2625
2626 module_init(packet_init);
2627 module_exit(packet_exit);
2628 MODULE_LICENSE("GPL");
2629 MODULE_ALIAS_NETPROTO(PF_PACKET);