ba248d93399a95213717b8e170cfd635717b42d9
[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 <linux/slab.h>
64 #include <linux/vmalloc.h>
65 #include <net/net_namespace.h>
66 #include <net/ip.h>
67 #include <net/protocol.h>
68 #include <linux/skbuff.h>
69 #include <net/sock.h>
70 #include <linux/errno.h>
71 #include <linux/timer.h>
72 #include <asm/system.h>
73 #include <asm/uaccess.h>
74 #include <asm/ioctls.h>
75 #include <asm/page.h>
76 #include <asm/cacheflush.h>
77 #include <asm/io.h>
78 #include <linux/proc_fs.h>
79 #include <linux/seq_file.h>
80 #include <linux/poll.h>
81 #include <linux/module.h>
82 #include <linux/init.h>
83 #include <linux/mutex.h>
84 #include <linux/if_vlan.h>
85 #include <linux/virtio_net.h>
86 #include <linux/errqueue.h>
87 #include <linux/net_tstamp.h>
88
89 #ifdef CONFIG_INET
90 #include <net/inet_common.h>
91 #endif
92
93 /*
94    Assumptions:
95    - if device has no dev->hard_header routine, it adds and removes ll header
96      inside itself. In this case ll header is invisible outside of device,
97      but higher levels still should reserve dev->hard_header_len.
98      Some devices are enough clever to reallocate skb, when header
99      will not fit to reserved space (tunnel), another ones are silly
100      (PPP).
101    - packet socket receives packets with pulled ll header,
102      so that SOCK_RAW should push it back.
103
104 On receive:
105 -----------
106
107 Incoming, dev->hard_header!=NULL
108    mac_header -> ll header
109    data       -> data
110
111 Outgoing, dev->hard_header!=NULL
112    mac_header -> ll header
113    data       -> ll header
114
115 Incoming, dev->hard_header==NULL
116    mac_header -> UNKNOWN position. It is very likely, that it points to ll
117                  header.  PPP makes it, that is wrong, because introduce
118                  assymetry between rx and tx paths.
119    data       -> data
120
121 Outgoing, dev->hard_header==NULL
122    mac_header -> data. ll header is still not built!
123    data       -> data
124
125 Resume
126   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
127
128
129 On transmit:
130 ------------
131
132 dev->hard_header != NULL
133    mac_header -> ll header
134    data       -> ll header
135
136 dev->hard_header == NULL (ll header is added by device, we cannot control it)
137    mac_header -> data
138    data       -> data
139
140    We should set nh.raw on output to correct posistion,
141    packet classifier depends on it.
142  */
143
144 /* Private packet socket structures. */
145
146 struct packet_mclist {
147         struct packet_mclist    *next;
148         int                     ifindex;
149         int                     count;
150         unsigned short          type;
151         unsigned short          alen;
152         unsigned char           addr[MAX_ADDR_LEN];
153 };
154 /* identical to struct packet_mreq except it has
155  * a longer address field.
156  */
157 struct packet_mreq_max {
158         int             mr_ifindex;
159         unsigned short  mr_type;
160         unsigned short  mr_alen;
161         unsigned char   mr_address[MAX_ADDR_LEN];
162 };
163
164 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
165                 int closing, int tx_ring);
166
167 struct pgv {
168         char *buffer;
169 };
170
171 struct packet_ring_buffer {
172         struct pgv              *pg_vec;
173         unsigned int            head;
174         unsigned int            frames_per_block;
175         unsigned int            frame_size;
176         unsigned int            frame_max;
177
178         unsigned int            pg_vec_order;
179         unsigned int            pg_vec_pages;
180         unsigned int            pg_vec_len;
181
182         atomic_t                pending;
183 };
184
185 struct packet_sock;
186 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
187
188 static void packet_flush_mclist(struct sock *sk);
189
190 struct packet_sock {
191         /* struct sock has to be the first member of packet_sock */
192         struct sock             sk;
193         struct tpacket_stats    stats;
194         struct packet_ring_buffer       rx_ring;
195         struct packet_ring_buffer       tx_ring;
196         int                     copy_thresh;
197         spinlock_t              bind_lock;
198         struct mutex            pg_vec_lock;
199         unsigned int            running:1,      /* prot_hook is attached*/
200                                 auxdata:1,
201                                 origdev:1,
202                                 has_vnet_hdr:1;
203         int                     ifindex;        /* bound device         */
204         __be16                  num;
205         struct packet_mclist    *mclist;
206         atomic_t                mapped;
207         enum tpacket_versions   tp_version;
208         unsigned int            tp_hdrlen;
209         unsigned int            tp_reserve;
210         unsigned int            tp_loss:1;
211         unsigned int            tp_tstamp;
212         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
213 };
214
215 struct packet_skb_cb {
216         unsigned int origlen;
217         union {
218                 struct sockaddr_pkt pkt;
219                 struct sockaddr_ll ll;
220         } sa;
221 };
222
223 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
224
225 static inline __pure struct page *pgv_to_page(void *addr)
226 {
227         if (is_vmalloc_addr(addr))
228                 return vmalloc_to_page(addr);
229         return virt_to_page(addr);
230 }
231
232 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
233 {
234         union {
235                 struct tpacket_hdr *h1;
236                 struct tpacket2_hdr *h2;
237                 void *raw;
238         } h;
239
240         h.raw = frame;
241         switch (po->tp_version) {
242         case TPACKET_V1:
243                 h.h1->tp_status = status;
244                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
245                 break;
246         case TPACKET_V2:
247                 h.h2->tp_status = status;
248                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
249                 break;
250         default:
251                 pr_err("TPACKET version not supported\n");
252                 BUG();
253         }
254
255         smp_wmb();
256 }
257
258 static int __packet_get_status(struct packet_sock *po, void *frame)
259 {
260         union {
261                 struct tpacket_hdr *h1;
262                 struct tpacket2_hdr *h2;
263                 void *raw;
264         } h;
265
266         smp_rmb();
267
268         h.raw = frame;
269         switch (po->tp_version) {
270         case TPACKET_V1:
271                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
272                 return h.h1->tp_status;
273         case TPACKET_V2:
274                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
275                 return h.h2->tp_status;
276         default:
277                 pr_err("TPACKET version not supported\n");
278                 BUG();
279                 return 0;
280         }
281 }
282
283 static void *packet_lookup_frame(struct packet_sock *po,
284                 struct packet_ring_buffer *rb,
285                 unsigned int position,
286                 int status)
287 {
288         unsigned int pg_vec_pos, frame_offset;
289         union {
290                 struct tpacket_hdr *h1;
291                 struct tpacket2_hdr *h2;
292                 void *raw;
293         } h;
294
295         pg_vec_pos = position / rb->frames_per_block;
296         frame_offset = position % rb->frames_per_block;
297
298         h.raw = rb->pg_vec[pg_vec_pos].buffer +
299                 (frame_offset * rb->frame_size);
300
301         if (status != __packet_get_status(po, h.raw))
302                 return NULL;
303
304         return h.raw;
305 }
306
307 static inline void *packet_current_frame(struct packet_sock *po,
308                 struct packet_ring_buffer *rb,
309                 int status)
310 {
311         return packet_lookup_frame(po, rb, rb->head, status);
312 }
313
314 static inline void *packet_previous_frame(struct packet_sock *po,
315                 struct packet_ring_buffer *rb,
316                 int status)
317 {
318         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
319         return packet_lookup_frame(po, rb, previous, status);
320 }
321
322 static inline void packet_increment_head(struct packet_ring_buffer *buff)
323 {
324         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
325 }
326
327 static inline struct packet_sock *pkt_sk(struct sock *sk)
328 {
329         return (struct packet_sock *)sk;
330 }
331
332 static void packet_sock_destruct(struct sock *sk)
333 {
334         skb_queue_purge(&sk->sk_error_queue);
335
336         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
337         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
338
339         if (!sock_flag(sk, SOCK_DEAD)) {
340                 pr_err("Attempt to release alive packet socket: %p\n", sk);
341                 return;
342         }
343
344         sk_refcnt_debug_dec(sk);
345 }
346
347
348 static const struct proto_ops packet_ops;
349
350 static const struct proto_ops packet_ops_spkt;
351
352 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
353                            struct packet_type *pt, struct net_device *orig_dev)
354 {
355         struct sock *sk;
356         struct sockaddr_pkt *spkt;
357
358         /*
359          *      When we registered the protocol we saved the socket in the data
360          *      field for just this event.
361          */
362
363         sk = pt->af_packet_priv;
364
365         /*
366          *      Yank back the headers [hope the device set this
367          *      right or kerboom...]
368          *
369          *      Incoming packets have ll header pulled,
370          *      push it back.
371          *
372          *      For outgoing ones skb->data == skb_mac_header(skb)
373          *      so that this procedure is noop.
374          */
375
376         if (skb->pkt_type == PACKET_LOOPBACK)
377                 goto out;
378
379         if (!net_eq(dev_net(dev), sock_net(sk)))
380                 goto out;
381
382         skb = skb_share_check(skb, GFP_ATOMIC);
383         if (skb == NULL)
384                 goto oom;
385
386         /* drop any routing info */
387         skb_dst_drop(skb);
388
389         /* drop conntrack reference */
390         nf_reset(skb);
391
392         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
393
394         skb_push(skb, skb->data - skb_mac_header(skb));
395
396         /*
397          *      The SOCK_PACKET socket receives _all_ frames.
398          */
399
400         spkt->spkt_family = dev->type;
401         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
402         spkt->spkt_protocol = skb->protocol;
403
404         /*
405          *      Charge the memory to the socket. This is done specifically
406          *      to prevent sockets using all the memory up.
407          */
408
409         if (sock_queue_rcv_skb(sk, skb) == 0)
410                 return 0;
411
412 out:
413         kfree_skb(skb);
414 oom:
415         return 0;
416 }
417
418
419 /*
420  *      Output a raw packet to a device layer. This bypasses all the other
421  *      protocol layers and you must therefore supply it with a complete frame
422  */
423
424 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
425                                struct msghdr *msg, size_t len)
426 {
427         struct sock *sk = sock->sk;
428         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
429         struct sk_buff *skb = NULL;
430         struct net_device *dev;
431         __be16 proto = 0;
432         int err;
433
434         /*
435          *      Get and verify the address.
436          */
437
438         if (saddr) {
439                 if (msg->msg_namelen < sizeof(struct sockaddr))
440                         return -EINVAL;
441                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
442                         proto = saddr->spkt_protocol;
443         } else
444                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
445
446         /*
447          *      Find the device first to size check it
448          */
449
450         saddr->spkt_device[13] = 0;
451 retry:
452         rcu_read_lock();
453         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
454         err = -ENODEV;
455         if (dev == NULL)
456                 goto out_unlock;
457
458         err = -ENETDOWN;
459         if (!(dev->flags & IFF_UP))
460                 goto out_unlock;
461
462         /*
463          * You may not queue a frame bigger than the mtu. This is the lowest level
464          * raw protocol and you must do your own fragmentation at this level.
465          */
466
467         err = -EMSGSIZE;
468         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
469                 goto out_unlock;
470
471         if (!skb) {
472                 size_t reserved = LL_RESERVED_SPACE(dev);
473                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
474
475                 rcu_read_unlock();
476                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
477                 if (skb == NULL)
478                         return -ENOBUFS;
479                 /* FIXME: Save some space for broken drivers that write a hard
480                  * header at transmission time by themselves. PPP is the notable
481                  * one here. This should really be fixed at the driver level.
482                  */
483                 skb_reserve(skb, reserved);
484                 skb_reset_network_header(skb);
485
486                 /* Try to align data part correctly */
487                 if (hhlen) {
488                         skb->data -= hhlen;
489                         skb->tail -= hhlen;
490                         if (len < hhlen)
491                                 skb_reset_network_header(skb);
492                 }
493                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
494                 if (err)
495                         goto out_free;
496                 goto retry;
497         }
498
499         if (len > (dev->mtu + dev->hard_header_len)) {
500                 /* Earlier code assumed this would be a VLAN pkt,
501                  * double-check this now that we have the actual
502                  * packet in hand.
503                  */
504                 struct ethhdr *ehdr;
505                 skb_reset_mac_header(skb);
506                 ehdr = eth_hdr(skb);
507                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
508                         err = -EMSGSIZE;
509                         goto out_unlock;
510                 }
511         }
512
513         skb->protocol = proto;
514         skb->dev = dev;
515         skb->priority = sk->sk_priority;
516         skb->mark = sk->sk_mark;
517         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
518         if (err < 0)
519                 goto out_unlock;
520
521         dev_queue_xmit(skb);
522         rcu_read_unlock();
523         return len;
524
525 out_unlock:
526         rcu_read_unlock();
527 out_free:
528         kfree_skb(skb);
529         return err;
530 }
531
532 static inline unsigned int run_filter(const struct sk_buff *skb,
533                                       const struct sock *sk,
534                                       unsigned int res)
535 {
536         struct sk_filter *filter;
537
538         rcu_read_lock();
539         filter = rcu_dereference(sk->sk_filter);
540         if (filter != NULL)
541                 res = SK_RUN_FILTER(filter, skb);
542         rcu_read_unlock();
543
544         return res;
545 }
546
547 /*
548  * This function makes lazy skb cloning in hope that most of packets
549  * are discarded by BPF.
550  *
551  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
552  * and skb->cb are mangled. It works because (and until) packets
553  * falling here are owned by current CPU. Output packets are cloned
554  * by dev_queue_xmit_nit(), input packets are processed by net_bh
555  * sequencially, so that if we return skb to original state on exit,
556  * we will not harm anyone.
557  */
558
559 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
560                       struct packet_type *pt, struct net_device *orig_dev)
561 {
562         struct sock *sk;
563         struct sockaddr_ll *sll;
564         struct packet_sock *po;
565         u8 *skb_head = skb->data;
566         int skb_len = skb->len;
567         unsigned int snaplen, res;
568
569         if (skb->pkt_type == PACKET_LOOPBACK)
570                 goto drop;
571
572         sk = pt->af_packet_priv;
573         po = pkt_sk(sk);
574
575         if (!net_eq(dev_net(dev), sock_net(sk)))
576                 goto drop;
577
578         skb->dev = dev;
579
580         if (dev->header_ops) {
581                 /* The device has an explicit notion of ll header,
582                  * exported to higher levels.
583                  *
584                  * Otherwise, the device hides details of its frame
585                  * structure, so that corresponding packet head is
586                  * never delivered to user.
587                  */
588                 if (sk->sk_type != SOCK_DGRAM)
589                         skb_push(skb, skb->data - skb_mac_header(skb));
590                 else if (skb->pkt_type == PACKET_OUTGOING) {
591                         /* Special case: outgoing packets have ll header at head */
592                         skb_pull(skb, skb_network_offset(skb));
593                 }
594         }
595
596         snaplen = skb->len;
597
598         res = run_filter(skb, sk, snaplen);
599         if (!res)
600                 goto drop_n_restore;
601         if (snaplen > res)
602                 snaplen = res;
603
604         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
605             (unsigned)sk->sk_rcvbuf)
606                 goto drop_n_acct;
607
608         if (skb_shared(skb)) {
609                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
610                 if (nskb == NULL)
611                         goto drop_n_acct;
612
613                 if (skb_head != skb->data) {
614                         skb->data = skb_head;
615                         skb->len = skb_len;
616                 }
617                 kfree_skb(skb);
618                 skb = nskb;
619         }
620
621         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
622                      sizeof(skb->cb));
623
624         sll = &PACKET_SKB_CB(skb)->sa.ll;
625         sll->sll_family = AF_PACKET;
626         sll->sll_hatype = dev->type;
627         sll->sll_protocol = skb->protocol;
628         sll->sll_pkttype = skb->pkt_type;
629         if (unlikely(po->origdev))
630                 sll->sll_ifindex = orig_dev->ifindex;
631         else
632                 sll->sll_ifindex = dev->ifindex;
633
634         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
635
636         PACKET_SKB_CB(skb)->origlen = skb->len;
637
638         if (pskb_trim(skb, snaplen))
639                 goto drop_n_acct;
640
641         skb_set_owner_r(skb, sk);
642         skb->dev = NULL;
643         skb_dst_drop(skb);
644
645         /* drop conntrack reference */
646         nf_reset(skb);
647
648         spin_lock(&sk->sk_receive_queue.lock);
649         po->stats.tp_packets++;
650         skb->dropcount = atomic_read(&sk->sk_drops);
651         __skb_queue_tail(&sk->sk_receive_queue, skb);
652         spin_unlock(&sk->sk_receive_queue.lock);
653         sk->sk_data_ready(sk, skb->len);
654         return 0;
655
656 drop_n_acct:
657         po->stats.tp_drops = atomic_inc_return(&sk->sk_drops);
658
659 drop_n_restore:
660         if (skb_head != skb->data && skb_shared(skb)) {
661                 skb->data = skb_head;
662                 skb->len = skb_len;
663         }
664 drop:
665         consume_skb(skb);
666         return 0;
667 }
668
669 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
670                        struct packet_type *pt, struct net_device *orig_dev)
671 {
672         struct sock *sk;
673         struct packet_sock *po;
674         struct sockaddr_ll *sll;
675         union {
676                 struct tpacket_hdr *h1;
677                 struct tpacket2_hdr *h2;
678                 void *raw;
679         } h;
680         u8 *skb_head = skb->data;
681         int skb_len = skb->len;
682         unsigned int snaplen, res;
683         unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
684         unsigned short macoff, netoff, hdrlen;
685         struct sk_buff *copy_skb = NULL;
686         struct timeval tv;
687         struct timespec ts;
688         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
689
690         if (skb->pkt_type == PACKET_LOOPBACK)
691                 goto drop;
692
693         sk = pt->af_packet_priv;
694         po = pkt_sk(sk);
695
696         if (!net_eq(dev_net(dev), sock_net(sk)))
697                 goto drop;
698
699         if (dev->header_ops) {
700                 if (sk->sk_type != SOCK_DGRAM)
701                         skb_push(skb, skb->data - skb_mac_header(skb));
702                 else if (skb->pkt_type == PACKET_OUTGOING) {
703                         /* Special case: outgoing packets have ll header at head */
704                         skb_pull(skb, skb_network_offset(skb));
705                 }
706         }
707
708         if (skb->ip_summed == CHECKSUM_PARTIAL)
709                 status |= TP_STATUS_CSUMNOTREADY;
710
711         snaplen = skb->len;
712
713         res = run_filter(skb, sk, snaplen);
714         if (!res)
715                 goto drop_n_restore;
716         if (snaplen > res)
717                 snaplen = res;
718
719         if (sk->sk_type == SOCK_DGRAM) {
720                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
721                                   po->tp_reserve;
722         } else {
723                 unsigned maclen = skb_network_offset(skb);
724                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
725                                        (maclen < 16 ? 16 : maclen)) +
726                         po->tp_reserve;
727                 macoff = netoff - maclen;
728         }
729
730         if (macoff + snaplen > po->rx_ring.frame_size) {
731                 if (po->copy_thresh &&
732                     atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
733                     (unsigned)sk->sk_rcvbuf) {
734                         if (skb_shared(skb)) {
735                                 copy_skb = skb_clone(skb, GFP_ATOMIC);
736                         } else {
737                                 copy_skb = skb_get(skb);
738                                 skb_head = skb->data;
739                         }
740                         if (copy_skb)
741                                 skb_set_owner_r(copy_skb, sk);
742                 }
743                 snaplen = po->rx_ring.frame_size - macoff;
744                 if ((int)snaplen < 0)
745                         snaplen = 0;
746         }
747
748         spin_lock(&sk->sk_receive_queue.lock);
749         h.raw = packet_current_frame(po, &po->rx_ring, TP_STATUS_KERNEL);
750         if (!h.raw)
751                 goto ring_is_full;
752         packet_increment_head(&po->rx_ring);
753         po->stats.tp_packets++;
754         if (copy_skb) {
755                 status |= TP_STATUS_COPY;
756                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
757         }
758         if (!po->stats.tp_drops)
759                 status &= ~TP_STATUS_LOSING;
760         spin_unlock(&sk->sk_receive_queue.lock);
761
762         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
763
764         switch (po->tp_version) {
765         case TPACKET_V1:
766                 h.h1->tp_len = skb->len;
767                 h.h1->tp_snaplen = snaplen;
768                 h.h1->tp_mac = macoff;
769                 h.h1->tp_net = netoff;
770                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
771                                 && shhwtstamps->syststamp.tv64)
772                         tv = ktime_to_timeval(shhwtstamps->syststamp);
773                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
774                                 && shhwtstamps->hwtstamp.tv64)
775                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
776                 else if (skb->tstamp.tv64)
777                         tv = ktime_to_timeval(skb->tstamp);
778                 else
779                         do_gettimeofday(&tv);
780                 h.h1->tp_sec = tv.tv_sec;
781                 h.h1->tp_usec = tv.tv_usec;
782                 hdrlen = sizeof(*h.h1);
783                 break;
784         case TPACKET_V2:
785                 h.h2->tp_len = skb->len;
786                 h.h2->tp_snaplen = snaplen;
787                 h.h2->tp_mac = macoff;
788                 h.h2->tp_net = netoff;
789                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
790                                 && shhwtstamps->syststamp.tv64)
791                         ts = ktime_to_timespec(shhwtstamps->syststamp);
792                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
793                                 && shhwtstamps->hwtstamp.tv64)
794                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
795                 else if (skb->tstamp.tv64)
796                         ts = ktime_to_timespec(skb->tstamp);
797                 else
798                         getnstimeofday(&ts);
799                 h.h2->tp_sec = ts.tv_sec;
800                 h.h2->tp_nsec = ts.tv_nsec;
801                 if (vlan_tx_tag_present(skb)) {
802                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
803                         status |= TP_STATUS_VLAN_VALID;
804                 } else {
805                         h.h2->tp_vlan_tci = 0;
806                 }
807                 hdrlen = sizeof(*h.h2);
808                 break;
809         default:
810                 BUG();
811         }
812
813         sll = h.raw + TPACKET_ALIGN(hdrlen);
814         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
815         sll->sll_family = AF_PACKET;
816         sll->sll_hatype = dev->type;
817         sll->sll_protocol = skb->protocol;
818         sll->sll_pkttype = skb->pkt_type;
819         if (unlikely(po->origdev))
820                 sll->sll_ifindex = orig_dev->ifindex;
821         else
822                 sll->sll_ifindex = dev->ifindex;
823
824         __packet_set_status(po, h.raw, status);
825         smp_mb();
826 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
827         {
828                 u8 *start, *end;
829
830                 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
831                 for (start = h.raw; start < end; start += PAGE_SIZE)
832                         flush_dcache_page(pgv_to_page(start));
833         }
834 #endif
835
836         sk->sk_data_ready(sk, 0);
837
838 drop_n_restore:
839         if (skb_head != skb->data && skb_shared(skb)) {
840                 skb->data = skb_head;
841                 skb->len = skb_len;
842         }
843 drop:
844         kfree_skb(skb);
845         return 0;
846
847 ring_is_full:
848         po->stats.tp_drops++;
849         spin_unlock(&sk->sk_receive_queue.lock);
850
851         sk->sk_data_ready(sk, 0);
852         kfree_skb(copy_skb);
853         goto drop_n_restore;
854 }
855
856 static void tpacket_destruct_skb(struct sk_buff *skb)
857 {
858         struct packet_sock *po = pkt_sk(skb->sk);
859         void *ph;
860
861         BUG_ON(skb == NULL);
862
863         if (likely(po->tx_ring.pg_vec)) {
864                 ph = skb_shinfo(skb)->destructor_arg;
865                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
866                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
867                 atomic_dec(&po->tx_ring.pending);
868                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
869         }
870
871         sock_wfree(skb);
872 }
873
874 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
875                 void *frame, struct net_device *dev, int size_max,
876                 __be16 proto, unsigned char *addr)
877 {
878         union {
879                 struct tpacket_hdr *h1;
880                 struct tpacket2_hdr *h2;
881                 void *raw;
882         } ph;
883         int to_write, offset, len, tp_len, nr_frags, len_max;
884         struct socket *sock = po->sk.sk_socket;
885         struct page *page;
886         void *data;
887         int err;
888
889         ph.raw = frame;
890
891         skb->protocol = proto;
892         skb->dev = dev;
893         skb->priority = po->sk.sk_priority;
894         skb->mark = po->sk.sk_mark;
895         skb_shinfo(skb)->destructor_arg = ph.raw;
896
897         switch (po->tp_version) {
898         case TPACKET_V2:
899                 tp_len = ph.h2->tp_len;
900                 break;
901         default:
902                 tp_len = ph.h1->tp_len;
903                 break;
904         }
905         if (unlikely(tp_len > size_max)) {
906                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
907                 return -EMSGSIZE;
908         }
909
910         skb_reserve(skb, LL_RESERVED_SPACE(dev));
911         skb_reset_network_header(skb);
912
913         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
914         to_write = tp_len;
915
916         if (sock->type == SOCK_DGRAM) {
917                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
918                                 NULL, tp_len);
919                 if (unlikely(err < 0))
920                         return -EINVAL;
921         } else if (dev->hard_header_len) {
922                 /* net device doesn't like empty head */
923                 if (unlikely(tp_len <= dev->hard_header_len)) {
924                         pr_err("packet size is too short (%d < %d)\n",
925                                tp_len, dev->hard_header_len);
926                         return -EINVAL;
927                 }
928
929                 skb_push(skb, dev->hard_header_len);
930                 err = skb_store_bits(skb, 0, data,
931                                 dev->hard_header_len);
932                 if (unlikely(err))
933                         return err;
934
935                 data += dev->hard_header_len;
936                 to_write -= dev->hard_header_len;
937         }
938
939         err = -EFAULT;
940         offset = offset_in_page(data);
941         len_max = PAGE_SIZE - offset;
942         len = ((to_write > len_max) ? len_max : to_write);
943
944         skb->data_len = to_write;
945         skb->len += to_write;
946         skb->truesize += to_write;
947         atomic_add(to_write, &po->sk.sk_wmem_alloc);
948
949         while (likely(to_write)) {
950                 nr_frags = skb_shinfo(skb)->nr_frags;
951
952                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
953                         pr_err("Packet exceed the number of skb frags(%lu)\n",
954                                MAX_SKB_FRAGS);
955                         return -EFAULT;
956                 }
957
958                 page = pgv_to_page(data);
959                 data += len;
960                 flush_dcache_page(page);
961                 get_page(page);
962                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
963                 to_write -= len;
964                 offset = 0;
965                 len_max = PAGE_SIZE;
966                 len = ((to_write > len_max) ? len_max : to_write);
967         }
968
969         return tp_len;
970 }
971
972 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
973 {
974         struct sk_buff *skb;
975         struct net_device *dev;
976         __be16 proto;
977         int ifindex, err, reserve = 0;
978         void *ph;
979         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
980         int tp_len, size_max;
981         unsigned char *addr;
982         int len_sum = 0;
983         int status = 0;
984
985         mutex_lock(&po->pg_vec_lock);
986
987         err = -EBUSY;
988         if (saddr == NULL) {
989                 ifindex = po->ifindex;
990                 proto   = po->num;
991                 addr    = NULL;
992         } else {
993                 err = -EINVAL;
994                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
995                         goto out;
996                 if (msg->msg_namelen < (saddr->sll_halen
997                                         + offsetof(struct sockaddr_ll,
998                                                 sll_addr)))
999                         goto out;
1000                 ifindex = saddr->sll_ifindex;
1001                 proto   = saddr->sll_protocol;
1002                 addr    = saddr->sll_addr;
1003         }
1004
1005         dev = dev_get_by_index(sock_net(&po->sk), ifindex);
1006         err = -ENXIO;
1007         if (unlikely(dev == NULL))
1008                 goto out;
1009
1010         reserve = dev->hard_header_len;
1011
1012         err = -ENETDOWN;
1013         if (unlikely(!(dev->flags & IFF_UP)))
1014                 goto out_put;
1015
1016         size_max = po->tx_ring.frame_size
1017                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1018
1019         if (size_max > dev->mtu + reserve)
1020                 size_max = dev->mtu + reserve;
1021
1022         do {
1023                 ph = packet_current_frame(po, &po->tx_ring,
1024                                 TP_STATUS_SEND_REQUEST);
1025
1026                 if (unlikely(ph == NULL)) {
1027                         schedule();
1028                         continue;
1029                 }
1030
1031                 status = TP_STATUS_SEND_REQUEST;
1032                 skb = sock_alloc_send_skb(&po->sk,
1033                                 LL_ALLOCATED_SPACE(dev)
1034                                 + sizeof(struct sockaddr_ll),
1035                                 0, &err);
1036
1037                 if (unlikely(skb == NULL))
1038                         goto out_status;
1039
1040                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1041                                 addr);
1042
1043                 if (unlikely(tp_len < 0)) {
1044                         if (po->tp_loss) {
1045                                 __packet_set_status(po, ph,
1046                                                 TP_STATUS_AVAILABLE);
1047                                 packet_increment_head(&po->tx_ring);
1048                                 kfree_skb(skb);
1049                                 continue;
1050                         } else {
1051                                 status = TP_STATUS_WRONG_FORMAT;
1052                                 err = tp_len;
1053                                 goto out_status;
1054                         }
1055                 }
1056
1057                 skb->destructor = tpacket_destruct_skb;
1058                 __packet_set_status(po, ph, TP_STATUS_SENDING);
1059                 atomic_inc(&po->tx_ring.pending);
1060
1061                 status = TP_STATUS_SEND_REQUEST;
1062                 err = dev_queue_xmit(skb);
1063                 if (unlikely(err > 0)) {
1064                         err = net_xmit_errno(err);
1065                         if (err && __packet_get_status(po, ph) ==
1066                                    TP_STATUS_AVAILABLE) {
1067                                 /* skb was destructed already */
1068                                 skb = NULL;
1069                                 goto out_status;
1070                         }
1071                         /*
1072                          * skb was dropped but not destructed yet;
1073                          * let's treat it like congestion or err < 0
1074                          */
1075                         err = 0;
1076                 }
1077                 packet_increment_head(&po->tx_ring);
1078                 len_sum += tp_len;
1079         } while (likely((ph != NULL) ||
1080                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1081                          (atomic_read(&po->tx_ring.pending))))
1082                 );
1083
1084         err = len_sum;
1085         goto out_put;
1086
1087 out_status:
1088         __packet_set_status(po, ph, status);
1089         kfree_skb(skb);
1090 out_put:
1091         dev_put(dev);
1092 out:
1093         mutex_unlock(&po->pg_vec_lock);
1094         return err;
1095 }
1096
1097 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1098                                                size_t reserve, size_t len,
1099                                                size_t linear, int noblock,
1100                                                int *err)
1101 {
1102         struct sk_buff *skb;
1103
1104         /* Under a page?  Don't bother with paged skb. */
1105         if (prepad + len < PAGE_SIZE || !linear)
1106                 linear = len;
1107
1108         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1109                                    err);
1110         if (!skb)
1111                 return NULL;
1112
1113         skb_reserve(skb, reserve);
1114         skb_put(skb, linear);
1115         skb->data_len = len - linear;
1116         skb->len += len - linear;
1117
1118         return skb;
1119 }
1120
1121 static int packet_snd(struct socket *sock,
1122                           struct msghdr *msg, size_t len)
1123 {
1124         struct sock *sk = sock->sk;
1125         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1126         struct sk_buff *skb;
1127         struct net_device *dev;
1128         __be16 proto;
1129         unsigned char *addr;
1130         int ifindex, err, reserve = 0;
1131         struct virtio_net_hdr vnet_hdr = { 0 };
1132         int offset = 0;
1133         int vnet_hdr_len;
1134         struct packet_sock *po = pkt_sk(sk);
1135         unsigned short gso_type = 0;
1136
1137         /*
1138          *      Get and verify the address.
1139          */
1140
1141         if (saddr == NULL) {
1142                 ifindex = po->ifindex;
1143                 proto   = po->num;
1144                 addr    = NULL;
1145         } else {
1146                 err = -EINVAL;
1147                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1148                         goto out;
1149                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1150                         goto out;
1151                 ifindex = saddr->sll_ifindex;
1152                 proto   = saddr->sll_protocol;
1153                 addr    = saddr->sll_addr;
1154         }
1155
1156
1157         dev = dev_get_by_index(sock_net(sk), ifindex);
1158         err = -ENXIO;
1159         if (dev == NULL)
1160                 goto out_unlock;
1161         if (sock->type == SOCK_RAW)
1162                 reserve = dev->hard_header_len;
1163
1164         err = -ENETDOWN;
1165         if (!(dev->flags & IFF_UP))
1166                 goto out_unlock;
1167
1168         if (po->has_vnet_hdr) {
1169                 vnet_hdr_len = sizeof(vnet_hdr);
1170
1171                 err = -EINVAL;
1172                 if (len < vnet_hdr_len)
1173                         goto out_unlock;
1174
1175                 len -= vnet_hdr_len;
1176
1177                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1178                                        vnet_hdr_len);
1179                 if (err < 0)
1180                         goto out_unlock;
1181
1182                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1183                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1184                       vnet_hdr.hdr_len))
1185                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
1186                                                  vnet_hdr.csum_offset + 2;
1187
1188                 err = -EINVAL;
1189                 if (vnet_hdr.hdr_len > len)
1190                         goto out_unlock;
1191
1192                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1193                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1194                         case VIRTIO_NET_HDR_GSO_TCPV4:
1195                                 gso_type = SKB_GSO_TCPV4;
1196                                 break;
1197                         case VIRTIO_NET_HDR_GSO_TCPV6:
1198                                 gso_type = SKB_GSO_TCPV6;
1199                                 break;
1200                         case VIRTIO_NET_HDR_GSO_UDP:
1201                                 gso_type = SKB_GSO_UDP;
1202                                 break;
1203                         default:
1204                                 goto out_unlock;
1205                         }
1206
1207                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1208                                 gso_type |= SKB_GSO_TCP_ECN;
1209
1210                         if (vnet_hdr.gso_size == 0)
1211                                 goto out_unlock;
1212
1213                 }
1214         }
1215
1216         err = -EMSGSIZE;
1217         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1218                 goto out_unlock;
1219
1220         err = -ENOBUFS;
1221         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1222                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1223                                msg->msg_flags & MSG_DONTWAIT, &err);
1224         if (skb == NULL)
1225                 goto out_unlock;
1226
1227         skb_set_network_header(skb, reserve);
1228
1229         err = -EINVAL;
1230         if (sock->type == SOCK_DGRAM &&
1231             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1232                 goto out_free;
1233
1234         /* Returns -EFAULT on error */
1235         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1236         if (err)
1237                 goto out_free;
1238         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1239         if (err < 0)
1240                 goto out_free;
1241
1242         if (!gso_type && (len > dev->mtu + reserve)) {
1243                 /* Earlier code assumed this would be a VLAN pkt,
1244                  * double-check this now that we have the actual
1245                  * packet in hand.
1246                  */
1247                 struct ethhdr *ehdr;
1248                 skb_reset_mac_header(skb);
1249                 ehdr = eth_hdr(skb);
1250                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1251                         err = -EMSGSIZE;
1252                         goto out_free;
1253                 }
1254         }
1255
1256         skb->protocol = proto;
1257         skb->dev = dev;
1258         skb->priority = sk->sk_priority;
1259         skb->mark = sk->sk_mark;
1260
1261         if (po->has_vnet_hdr) {
1262                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1263                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1264                                                   vnet_hdr.csum_offset)) {
1265                                 err = -EINVAL;
1266                                 goto out_free;
1267                         }
1268                 }
1269
1270                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1271                 skb_shinfo(skb)->gso_type = gso_type;
1272
1273                 /* Header must be checked, and gso_segs computed. */
1274                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1275                 skb_shinfo(skb)->gso_segs = 0;
1276
1277                 len += vnet_hdr_len;
1278         }
1279
1280         /*
1281          *      Now send it
1282          */
1283
1284         err = dev_queue_xmit(skb);
1285         if (err > 0 && (err = net_xmit_errno(err)) != 0)
1286                 goto out_unlock;
1287
1288         dev_put(dev);
1289
1290         return len;
1291
1292 out_free:
1293         kfree_skb(skb);
1294 out_unlock:
1295         if (dev)
1296                 dev_put(dev);
1297 out:
1298         return err;
1299 }
1300
1301 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1302                 struct msghdr *msg, size_t len)
1303 {
1304         struct sock *sk = sock->sk;
1305         struct packet_sock *po = pkt_sk(sk);
1306         if (po->tx_ring.pg_vec)
1307                 return tpacket_snd(po, msg);
1308         else
1309                 return packet_snd(sock, msg, len);
1310 }
1311
1312 /*
1313  *      Close a PACKET socket. This is fairly simple. We immediately go
1314  *      to 'closed' state and remove our protocol entry in the device list.
1315  */
1316
1317 static int packet_release(struct socket *sock)
1318 {
1319         struct sock *sk = sock->sk;
1320         struct packet_sock *po;
1321         struct net *net;
1322         struct tpacket_req req;
1323
1324         if (!sk)
1325                 return 0;
1326
1327         net = sock_net(sk);
1328         po = pkt_sk(sk);
1329
1330         spin_lock_bh(&net->packet.sklist_lock);
1331         sk_del_node_init_rcu(sk);
1332         sock_prot_inuse_add(net, sk->sk_prot, -1);
1333         spin_unlock_bh(&net->packet.sklist_lock);
1334
1335         spin_lock(&po->bind_lock);
1336         if (po->running) {
1337                 /*
1338                  * Remove from protocol table
1339                  */
1340                 po->running = 0;
1341                 po->num = 0;
1342                 __dev_remove_pack(&po->prot_hook);
1343                 __sock_put(sk);
1344         }
1345         spin_unlock(&po->bind_lock);
1346
1347         packet_flush_mclist(sk);
1348
1349         memset(&req, 0, sizeof(req));
1350
1351         if (po->rx_ring.pg_vec)
1352                 packet_set_ring(sk, &req, 1, 0);
1353
1354         if (po->tx_ring.pg_vec)
1355                 packet_set_ring(sk, &req, 1, 1);
1356
1357         synchronize_net();
1358         /*
1359          *      Now the socket is dead. No more input will appear.
1360          */
1361         sock_orphan(sk);
1362         sock->sk = NULL;
1363
1364         /* Purge queues */
1365
1366         skb_queue_purge(&sk->sk_receive_queue);
1367         sk_refcnt_debug_release(sk);
1368
1369         sock_put(sk);
1370         return 0;
1371 }
1372
1373 /*
1374  *      Attach a packet hook.
1375  */
1376
1377 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1378 {
1379         struct packet_sock *po = pkt_sk(sk);
1380         /*
1381          *      Detach an existing hook if present.
1382          */
1383
1384         lock_sock(sk);
1385
1386         spin_lock(&po->bind_lock);
1387         if (po->running) {
1388                 __sock_put(sk);
1389                 po->running = 0;
1390                 po->num = 0;
1391                 spin_unlock(&po->bind_lock);
1392                 dev_remove_pack(&po->prot_hook);
1393                 spin_lock(&po->bind_lock);
1394         }
1395
1396         po->num = protocol;
1397         po->prot_hook.type = protocol;
1398         po->prot_hook.dev = dev;
1399
1400         po->ifindex = dev ? dev->ifindex : 0;
1401
1402         if (protocol == 0)
1403                 goto out_unlock;
1404
1405         if (!dev || (dev->flags & IFF_UP)) {
1406                 dev_add_pack(&po->prot_hook);
1407                 sock_hold(sk);
1408                 po->running = 1;
1409         } else {
1410                 sk->sk_err = ENETDOWN;
1411                 if (!sock_flag(sk, SOCK_DEAD))
1412                         sk->sk_error_report(sk);
1413         }
1414
1415 out_unlock:
1416         spin_unlock(&po->bind_lock);
1417         release_sock(sk);
1418         return 0;
1419 }
1420
1421 /*
1422  *      Bind a packet socket to a device
1423  */
1424
1425 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1426                             int addr_len)
1427 {
1428         struct sock *sk = sock->sk;
1429         char name[15];
1430         struct net_device *dev;
1431         int err = -ENODEV;
1432
1433         /*
1434          *      Check legality
1435          */
1436
1437         if (addr_len != sizeof(struct sockaddr))
1438                 return -EINVAL;
1439         strlcpy(name, uaddr->sa_data, sizeof(name));
1440
1441         dev = dev_get_by_name(sock_net(sk), name);
1442         if (dev) {
1443                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1444                 dev_put(dev);
1445         }
1446         return err;
1447 }
1448
1449 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1450 {
1451         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1452         struct sock *sk = sock->sk;
1453         struct net_device *dev = NULL;
1454         int err;
1455
1456
1457         /*
1458          *      Check legality
1459          */
1460
1461         if (addr_len < sizeof(struct sockaddr_ll))
1462                 return -EINVAL;
1463         if (sll->sll_family != AF_PACKET)
1464                 return -EINVAL;
1465
1466         if (sll->sll_ifindex) {
1467                 err = -ENODEV;
1468                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1469                 if (dev == NULL)
1470                         goto out;
1471         }
1472         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1473         if (dev)
1474                 dev_put(dev);
1475
1476 out:
1477         return err;
1478 }
1479
1480 static struct proto packet_proto = {
1481         .name     = "PACKET",
1482         .owner    = THIS_MODULE,
1483         .obj_size = sizeof(struct packet_sock),
1484 };
1485
1486 /*
1487  *      Create a packet of type SOCK_PACKET.
1488  */
1489
1490 static int packet_create(struct net *net, struct socket *sock, int protocol,
1491                          int kern)
1492 {
1493         struct sock *sk;
1494         struct packet_sock *po;
1495         __be16 proto = (__force __be16)protocol; /* weird, but documented */
1496         int err;
1497
1498         if (!capable(CAP_NET_RAW))
1499                 return -EPERM;
1500         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1501             sock->type != SOCK_PACKET)
1502                 return -ESOCKTNOSUPPORT;
1503
1504         sock->state = SS_UNCONNECTED;
1505
1506         err = -ENOBUFS;
1507         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1508         if (sk == NULL)
1509                 goto out;
1510
1511         sock->ops = &packet_ops;
1512         if (sock->type == SOCK_PACKET)
1513                 sock->ops = &packet_ops_spkt;
1514
1515         sock_init_data(sock, sk);
1516
1517         po = pkt_sk(sk);
1518         sk->sk_family = PF_PACKET;
1519         po->num = proto;
1520
1521         sk->sk_destruct = packet_sock_destruct;
1522         sk_refcnt_debug_inc(sk);
1523
1524         /*
1525          *      Attach a protocol block
1526          */
1527
1528         spin_lock_init(&po->bind_lock);
1529         mutex_init(&po->pg_vec_lock);
1530         po->prot_hook.func = packet_rcv;
1531
1532         if (sock->type == SOCK_PACKET)
1533                 po->prot_hook.func = packet_rcv_spkt;
1534
1535         po->prot_hook.af_packet_priv = sk;
1536
1537         if (proto) {
1538                 po->prot_hook.type = proto;
1539                 dev_add_pack(&po->prot_hook);
1540                 sock_hold(sk);
1541                 po->running = 1;
1542         }
1543
1544         spin_lock_bh(&net->packet.sklist_lock);
1545         sk_add_node_rcu(sk, &net->packet.sklist);
1546         sock_prot_inuse_add(net, &packet_proto, 1);
1547         spin_unlock_bh(&net->packet.sklist_lock);
1548
1549         return 0;
1550 out:
1551         return err;
1552 }
1553
1554 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1555 {
1556         struct sock_exterr_skb *serr;
1557         struct sk_buff *skb, *skb2;
1558         int copied, err;
1559
1560         err = -EAGAIN;
1561         skb = skb_dequeue(&sk->sk_error_queue);
1562         if (skb == NULL)
1563                 goto out;
1564
1565         copied = skb->len;
1566         if (copied > len) {
1567                 msg->msg_flags |= MSG_TRUNC;
1568                 copied = len;
1569         }
1570         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1571         if (err)
1572                 goto out_free_skb;
1573
1574         sock_recv_timestamp(msg, sk, skb);
1575
1576         serr = SKB_EXT_ERR(skb);
1577         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1578                  sizeof(serr->ee), &serr->ee);
1579
1580         msg->msg_flags |= MSG_ERRQUEUE;
1581         err = copied;
1582
1583         /* Reset and regenerate socket error */
1584         spin_lock_bh(&sk->sk_error_queue.lock);
1585         sk->sk_err = 0;
1586         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1587                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1588                 spin_unlock_bh(&sk->sk_error_queue.lock);
1589                 sk->sk_error_report(sk);
1590         } else
1591                 spin_unlock_bh(&sk->sk_error_queue.lock);
1592
1593 out_free_skb:
1594         kfree_skb(skb);
1595 out:
1596         return err;
1597 }
1598
1599 /*
1600  *      Pull a packet from our receive queue and hand it to the user.
1601  *      If necessary we block.
1602  */
1603
1604 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1605                           struct msghdr *msg, size_t len, int flags)
1606 {
1607         struct sock *sk = sock->sk;
1608         struct sk_buff *skb;
1609         int copied, err;
1610         struct sockaddr_ll *sll;
1611         int vnet_hdr_len = 0;
1612
1613         err = -EINVAL;
1614         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1615                 goto out;
1616
1617 #if 0
1618         /* What error should we return now? EUNATTACH? */
1619         if (pkt_sk(sk)->ifindex < 0)
1620                 return -ENODEV;
1621 #endif
1622
1623         if (flags & MSG_ERRQUEUE) {
1624                 err = packet_recv_error(sk, msg, len);
1625                 goto out;
1626         }
1627
1628         /*
1629          *      Call the generic datagram receiver. This handles all sorts
1630          *      of horrible races and re-entrancy so we can forget about it
1631          *      in the protocol layers.
1632          *
1633          *      Now it will return ENETDOWN, if device have just gone down,
1634          *      but then it will block.
1635          */
1636
1637         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1638
1639         /*
1640          *      An error occurred so return it. Because skb_recv_datagram()
1641          *      handles the blocking we don't see and worry about blocking
1642          *      retries.
1643          */
1644
1645         if (skb == NULL)
1646                 goto out;
1647
1648         if (pkt_sk(sk)->has_vnet_hdr) {
1649                 struct virtio_net_hdr vnet_hdr = { 0 };
1650
1651                 err = -EINVAL;
1652                 vnet_hdr_len = sizeof(vnet_hdr);
1653                 if (len < vnet_hdr_len)
1654                         goto out_free;
1655
1656                 len -= vnet_hdr_len;
1657
1658                 if (skb_is_gso(skb)) {
1659                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1660
1661                         /* This is a hint as to how much should be linear. */
1662                         vnet_hdr.hdr_len = skb_headlen(skb);
1663                         vnet_hdr.gso_size = sinfo->gso_size;
1664                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1665                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1666                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1667                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1668                         else if (sinfo->gso_type & SKB_GSO_UDP)
1669                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1670                         else if (sinfo->gso_type & SKB_GSO_FCOE)
1671                                 goto out_free;
1672                         else
1673                                 BUG();
1674                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1675                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1676                 } else
1677                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1678
1679                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1680                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1681                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1682                         vnet_hdr.csum_offset = skb->csum_offset;
1683                 } /* else everything is zero */
1684
1685                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1686                                      vnet_hdr_len);
1687                 if (err < 0)
1688                         goto out_free;
1689         }
1690
1691         /*
1692          *      If the address length field is there to be filled in, we fill
1693          *      it in now.
1694          */
1695
1696         sll = &PACKET_SKB_CB(skb)->sa.ll;
1697         if (sock->type == SOCK_PACKET)
1698                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1699         else
1700                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1701
1702         /*
1703          *      You lose any data beyond the buffer you gave. If it worries a
1704          *      user program they can ask the device for its MTU anyway.
1705          */
1706
1707         copied = skb->len;
1708         if (copied > len) {
1709                 copied = len;
1710                 msg->msg_flags |= MSG_TRUNC;
1711         }
1712
1713         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1714         if (err)
1715                 goto out_free;
1716
1717         sock_recv_ts_and_drops(msg, sk, skb);
1718
1719         if (msg->msg_name)
1720                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1721                        msg->msg_namelen);
1722
1723         if (pkt_sk(sk)->auxdata) {
1724                 struct tpacket_auxdata aux;
1725
1726                 aux.tp_status = TP_STATUS_USER;
1727                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1728                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1729                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1730                 aux.tp_snaplen = skb->len;
1731                 aux.tp_mac = 0;
1732                 aux.tp_net = skb_network_offset(skb);
1733                 if (vlan_tx_tag_present(skb)) {
1734                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1735                         aux.tp_status |= TP_STATUS_VLAN_VALID;
1736                 } else {
1737                         aux.tp_vlan_tci = 0;
1738                 }
1739                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1740         }
1741
1742         /*
1743          *      Free or return the buffer as appropriate. Again this
1744          *      hides all the races and re-entrancy issues from us.
1745          */
1746         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1747
1748 out_free:
1749         skb_free_datagram(sk, skb);
1750 out:
1751         return err;
1752 }
1753
1754 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1755                                int *uaddr_len, int peer)
1756 {
1757         struct net_device *dev;
1758         struct sock *sk = sock->sk;
1759
1760         if (peer)
1761                 return -EOPNOTSUPP;
1762
1763         uaddr->sa_family = AF_PACKET;
1764         rcu_read_lock();
1765         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1766         if (dev)
1767                 strncpy(uaddr->sa_data, dev->name, 14);
1768         else
1769                 memset(uaddr->sa_data, 0, 14);
1770         rcu_read_unlock();
1771         *uaddr_len = sizeof(*uaddr);
1772
1773         return 0;
1774 }
1775
1776 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1777                           int *uaddr_len, int peer)
1778 {
1779         struct net_device *dev;
1780         struct sock *sk = sock->sk;
1781         struct packet_sock *po = pkt_sk(sk);
1782         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1783
1784         if (peer)
1785                 return -EOPNOTSUPP;
1786
1787         sll->sll_family = AF_PACKET;
1788         sll->sll_ifindex = po->ifindex;
1789         sll->sll_protocol = po->num;
1790         sll->sll_pkttype = 0;
1791         rcu_read_lock();
1792         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1793         if (dev) {
1794                 sll->sll_hatype = dev->type;
1795                 sll->sll_halen = dev->addr_len;
1796                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1797         } else {
1798                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
1799                 sll->sll_halen = 0;
1800         }
1801         rcu_read_unlock();
1802         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1803
1804         return 0;
1805 }
1806
1807 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1808                          int what)
1809 {
1810         switch (i->type) {
1811         case PACKET_MR_MULTICAST:
1812                 if (i->alen != dev->addr_len)
1813                         return -EINVAL;
1814                 if (what > 0)
1815                         return dev_mc_add(dev, i->addr);
1816                 else
1817                         return dev_mc_del(dev, i->addr);
1818                 break;
1819         case PACKET_MR_PROMISC:
1820                 return dev_set_promiscuity(dev, what);
1821                 break;
1822         case PACKET_MR_ALLMULTI:
1823                 return dev_set_allmulti(dev, what);
1824                 break;
1825         case PACKET_MR_UNICAST:
1826                 if (i->alen != dev->addr_len)
1827                         return -EINVAL;
1828                 if (what > 0)
1829                         return dev_uc_add(dev, i->addr);
1830                 else
1831                         return dev_uc_del(dev, i->addr);
1832                 break;
1833         default:
1834                 break;
1835         }
1836         return 0;
1837 }
1838
1839 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1840 {
1841         for ( ; i; i = i->next) {
1842                 if (i->ifindex == dev->ifindex)
1843                         packet_dev_mc(dev, i, what);
1844         }
1845 }
1846
1847 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1848 {
1849         struct packet_sock *po = pkt_sk(sk);
1850         struct packet_mclist *ml, *i;
1851         struct net_device *dev;
1852         int err;
1853
1854         rtnl_lock();
1855
1856         err = -ENODEV;
1857         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1858         if (!dev)
1859                 goto done;
1860
1861         err = -EINVAL;
1862         if (mreq->mr_alen > dev->addr_len)
1863                 goto done;
1864
1865         err = -ENOBUFS;
1866         i = kmalloc(sizeof(*i), GFP_KERNEL);
1867         if (i == NULL)
1868                 goto done;
1869
1870         err = 0;
1871         for (ml = po->mclist; ml; ml = ml->next) {
1872                 if (ml->ifindex == mreq->mr_ifindex &&
1873                     ml->type == mreq->mr_type &&
1874                     ml->alen == mreq->mr_alen &&
1875                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1876                         ml->count++;
1877                         /* Free the new element ... */
1878                         kfree(i);
1879                         goto done;
1880                 }
1881         }
1882
1883         i->type = mreq->mr_type;
1884         i->ifindex = mreq->mr_ifindex;
1885         i->alen = mreq->mr_alen;
1886         memcpy(i->addr, mreq->mr_address, i->alen);
1887         i->count = 1;
1888         i->next = po->mclist;
1889         po->mclist = i;
1890         err = packet_dev_mc(dev, i, 1);
1891         if (err) {
1892                 po->mclist = i->next;
1893                 kfree(i);
1894         }
1895
1896 done:
1897         rtnl_unlock();
1898         return err;
1899 }
1900
1901 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1902 {
1903         struct packet_mclist *ml, **mlp;
1904
1905         rtnl_lock();
1906
1907         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1908                 if (ml->ifindex == mreq->mr_ifindex &&
1909                     ml->type == mreq->mr_type &&
1910                     ml->alen == mreq->mr_alen &&
1911                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1912                         if (--ml->count == 0) {
1913                                 struct net_device *dev;
1914                                 *mlp = ml->next;
1915                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1916                                 if (dev)
1917                                         packet_dev_mc(dev, ml, -1);
1918                                 kfree(ml);
1919                         }
1920                         rtnl_unlock();
1921                         return 0;
1922                 }
1923         }
1924         rtnl_unlock();
1925         return -EADDRNOTAVAIL;
1926 }
1927
1928 static void packet_flush_mclist(struct sock *sk)
1929 {
1930         struct packet_sock *po = pkt_sk(sk);
1931         struct packet_mclist *ml;
1932
1933         if (!po->mclist)
1934                 return;
1935
1936         rtnl_lock();
1937         while ((ml = po->mclist) != NULL) {
1938                 struct net_device *dev;
1939
1940                 po->mclist = ml->next;
1941                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1942                 if (dev != NULL)
1943                         packet_dev_mc(dev, ml, -1);
1944                 kfree(ml);
1945         }
1946         rtnl_unlock();
1947 }
1948
1949 static int
1950 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1951 {
1952         struct sock *sk = sock->sk;
1953         struct packet_sock *po = pkt_sk(sk);
1954         int ret;
1955
1956         if (level != SOL_PACKET)
1957                 return -ENOPROTOOPT;
1958
1959         switch (optname) {
1960         case PACKET_ADD_MEMBERSHIP:
1961         case PACKET_DROP_MEMBERSHIP:
1962         {
1963                 struct packet_mreq_max mreq;
1964                 int len = optlen;
1965                 memset(&mreq, 0, sizeof(mreq));
1966                 if (len < sizeof(struct packet_mreq))
1967                         return -EINVAL;
1968                 if (len > sizeof(mreq))
1969                         len = sizeof(mreq);
1970                 if (copy_from_user(&mreq, optval, len))
1971                         return -EFAULT;
1972                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1973                         return -EINVAL;
1974                 if (optname == PACKET_ADD_MEMBERSHIP)
1975                         ret = packet_mc_add(sk, &mreq);
1976                 else
1977                         ret = packet_mc_drop(sk, &mreq);
1978                 return ret;
1979         }
1980
1981         case PACKET_RX_RING:
1982         case PACKET_TX_RING:
1983         {
1984                 struct tpacket_req req;
1985
1986                 if (optlen < sizeof(req))
1987                         return -EINVAL;
1988                 if (pkt_sk(sk)->has_vnet_hdr)
1989                         return -EINVAL;
1990                 if (copy_from_user(&req, optval, sizeof(req)))
1991                         return -EFAULT;
1992                 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1993         }
1994         case PACKET_COPY_THRESH:
1995         {
1996                 int val;
1997
1998                 if (optlen != sizeof(val))
1999                         return -EINVAL;
2000                 if (copy_from_user(&val, optval, sizeof(val)))
2001                         return -EFAULT;
2002
2003                 pkt_sk(sk)->copy_thresh = val;
2004                 return 0;
2005         }
2006         case PACKET_VERSION:
2007         {
2008                 int val;
2009
2010                 if (optlen != sizeof(val))
2011                         return -EINVAL;
2012                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2013                         return -EBUSY;
2014                 if (copy_from_user(&val, optval, sizeof(val)))
2015                         return -EFAULT;
2016                 switch (val) {
2017                 case TPACKET_V1:
2018                 case TPACKET_V2:
2019                         po->tp_version = val;
2020                         return 0;
2021                 default:
2022                         return -EINVAL;
2023                 }
2024         }
2025         case PACKET_RESERVE:
2026         {
2027                 unsigned int val;
2028
2029                 if (optlen != sizeof(val))
2030                         return -EINVAL;
2031                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2032                         return -EBUSY;
2033                 if (copy_from_user(&val, optval, sizeof(val)))
2034                         return -EFAULT;
2035                 po->tp_reserve = val;
2036                 return 0;
2037         }
2038         case PACKET_LOSS:
2039         {
2040                 unsigned int val;
2041
2042                 if (optlen != sizeof(val))
2043                         return -EINVAL;
2044                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2045                         return -EBUSY;
2046                 if (copy_from_user(&val, optval, sizeof(val)))
2047                         return -EFAULT;
2048                 po->tp_loss = !!val;
2049                 return 0;
2050         }
2051         case PACKET_AUXDATA:
2052         {
2053                 int val;
2054
2055                 if (optlen < sizeof(val))
2056                         return -EINVAL;
2057                 if (copy_from_user(&val, optval, sizeof(val)))
2058                         return -EFAULT;
2059
2060                 po->auxdata = !!val;
2061                 return 0;
2062         }
2063         case PACKET_ORIGDEV:
2064         {
2065                 int val;
2066
2067                 if (optlen < sizeof(val))
2068                         return -EINVAL;
2069                 if (copy_from_user(&val, optval, sizeof(val)))
2070                         return -EFAULT;
2071
2072                 po->origdev = !!val;
2073                 return 0;
2074         }
2075         case PACKET_VNET_HDR:
2076         {
2077                 int val;
2078
2079                 if (sock->type != SOCK_RAW)
2080                         return -EINVAL;
2081                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2082                         return -EBUSY;
2083                 if (optlen < sizeof(val))
2084                         return -EINVAL;
2085                 if (copy_from_user(&val, optval, sizeof(val)))
2086                         return -EFAULT;
2087
2088                 po->has_vnet_hdr = !!val;
2089                 return 0;
2090         }
2091         case PACKET_TIMESTAMP:
2092         {
2093                 int val;
2094
2095                 if (optlen != sizeof(val))
2096                         return -EINVAL;
2097                 if (copy_from_user(&val, optval, sizeof(val)))
2098                         return -EFAULT;
2099
2100                 po->tp_tstamp = val;
2101                 return 0;
2102         }
2103         default:
2104                 return -ENOPROTOOPT;
2105         }
2106 }
2107
2108 static int packet_getsockopt(struct socket *sock, int level, int optname,
2109                              char __user *optval, int __user *optlen)
2110 {
2111         int len;
2112         int val;
2113         struct sock *sk = sock->sk;
2114         struct packet_sock *po = pkt_sk(sk);
2115         void *data;
2116         struct tpacket_stats st;
2117
2118         if (level != SOL_PACKET)
2119                 return -ENOPROTOOPT;
2120
2121         if (get_user(len, optlen))
2122                 return -EFAULT;
2123
2124         if (len < 0)
2125                 return -EINVAL;
2126
2127         switch (optname) {
2128         case PACKET_STATISTICS:
2129                 if (len > sizeof(struct tpacket_stats))
2130                         len = sizeof(struct tpacket_stats);
2131                 spin_lock_bh(&sk->sk_receive_queue.lock);
2132                 st = po->stats;
2133                 memset(&po->stats, 0, sizeof(st));
2134                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2135                 st.tp_packets += st.tp_drops;
2136
2137                 data = &st;
2138                 break;
2139         case PACKET_AUXDATA:
2140                 if (len > sizeof(int))
2141                         len = sizeof(int);
2142                 val = po->auxdata;
2143
2144                 data = &val;
2145                 break;
2146         case PACKET_ORIGDEV:
2147                 if (len > sizeof(int))
2148                         len = sizeof(int);
2149                 val = po->origdev;
2150
2151                 data = &val;
2152                 break;
2153         case PACKET_VNET_HDR:
2154                 if (len > sizeof(int))
2155                         len = sizeof(int);
2156                 val = po->has_vnet_hdr;
2157
2158                 data = &val;
2159                 break;
2160         case PACKET_VERSION:
2161                 if (len > sizeof(int))
2162                         len = sizeof(int);
2163                 val = po->tp_version;
2164                 data = &val;
2165                 break;
2166         case PACKET_HDRLEN:
2167                 if (len > sizeof(int))
2168                         len = sizeof(int);
2169                 if (copy_from_user(&val, optval, len))
2170                         return -EFAULT;
2171                 switch (val) {
2172                 case TPACKET_V1:
2173                         val = sizeof(struct tpacket_hdr);
2174                         break;
2175                 case TPACKET_V2:
2176                         val = sizeof(struct tpacket2_hdr);
2177                         break;
2178                 default:
2179                         return -EINVAL;
2180                 }
2181                 data = &val;
2182                 break;
2183         case PACKET_RESERVE:
2184                 if (len > sizeof(unsigned int))
2185                         len = sizeof(unsigned int);
2186                 val = po->tp_reserve;
2187                 data = &val;
2188                 break;
2189         case PACKET_LOSS:
2190                 if (len > sizeof(unsigned int))
2191                         len = sizeof(unsigned int);
2192                 val = po->tp_loss;
2193                 data = &val;
2194                 break;
2195         case PACKET_TIMESTAMP:
2196                 if (len > sizeof(int))
2197                         len = sizeof(int);
2198                 val = po->tp_tstamp;
2199                 data = &val;
2200                 break;
2201         default:
2202                 return -ENOPROTOOPT;
2203         }
2204
2205         if (put_user(len, optlen))
2206                 return -EFAULT;
2207         if (copy_to_user(optval, data, len))
2208                 return -EFAULT;
2209         return 0;
2210 }
2211
2212
2213 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2214 {
2215         struct sock *sk;
2216         struct hlist_node *node;
2217         struct net_device *dev = data;
2218         struct net *net = dev_net(dev);
2219
2220         rcu_read_lock();
2221         sk_for_each_rcu(sk, node, &net->packet.sklist) {
2222                 struct packet_sock *po = pkt_sk(sk);
2223
2224                 switch (msg) {
2225                 case NETDEV_UNREGISTER:
2226                         if (po->mclist)
2227                                 packet_dev_mclist(dev, po->mclist, -1);
2228                         /* fallthrough */
2229
2230                 case NETDEV_DOWN:
2231                         if (dev->ifindex == po->ifindex) {
2232                                 spin_lock(&po->bind_lock);
2233                                 if (po->running) {
2234                                         __dev_remove_pack(&po->prot_hook);
2235                                         __sock_put(sk);
2236                                         po->running = 0;
2237                                         sk->sk_err = ENETDOWN;
2238                                         if (!sock_flag(sk, SOCK_DEAD))
2239                                                 sk->sk_error_report(sk);
2240                                 }
2241                                 if (msg == NETDEV_UNREGISTER) {
2242                                         po->ifindex = -1;
2243                                         po->prot_hook.dev = NULL;
2244                                 }
2245                                 spin_unlock(&po->bind_lock);
2246                         }
2247                         break;
2248                 case NETDEV_UP:
2249                         if (dev->ifindex == po->ifindex) {
2250                                 spin_lock(&po->bind_lock);
2251                                 if (po->num && !po->running) {
2252                                         dev_add_pack(&po->prot_hook);
2253                                         sock_hold(sk);
2254                                         po->running = 1;
2255                                 }
2256                                 spin_unlock(&po->bind_lock);
2257                         }
2258                         break;
2259                 }
2260         }
2261         rcu_read_unlock();
2262         return NOTIFY_DONE;
2263 }
2264
2265
2266 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2267                         unsigned long arg)
2268 {
2269         struct sock *sk = sock->sk;
2270
2271         switch (cmd) {
2272         case SIOCOUTQ:
2273         {
2274                 int amount = sk_wmem_alloc_get(sk);
2275
2276                 return put_user(amount, (int __user *)arg);
2277         }
2278         case SIOCINQ:
2279         {
2280                 struct sk_buff *skb;
2281                 int amount = 0;
2282
2283                 spin_lock_bh(&sk->sk_receive_queue.lock);
2284                 skb = skb_peek(&sk->sk_receive_queue);
2285                 if (skb)
2286                         amount = skb->len;
2287                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2288                 return put_user(amount, (int __user *)arg);
2289         }
2290         case SIOCGSTAMP:
2291                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2292         case SIOCGSTAMPNS:
2293                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2294
2295 #ifdef CONFIG_INET
2296         case SIOCADDRT:
2297         case SIOCDELRT:
2298         case SIOCDARP:
2299         case SIOCGARP:
2300         case SIOCSARP:
2301         case SIOCGIFADDR:
2302         case SIOCSIFADDR:
2303         case SIOCGIFBRDADDR:
2304         case SIOCSIFBRDADDR:
2305         case SIOCGIFNETMASK:
2306         case SIOCSIFNETMASK:
2307         case SIOCGIFDSTADDR:
2308         case SIOCSIFDSTADDR:
2309         case SIOCSIFFLAGS:
2310                 return inet_dgram_ops.ioctl(sock, cmd, arg);
2311 #endif
2312
2313         default:
2314                 return -ENOIOCTLCMD;
2315         }
2316         return 0;
2317 }
2318
2319 static unsigned int packet_poll(struct file *file, struct socket *sock,
2320                                 poll_table *wait)
2321 {
2322         struct sock *sk = sock->sk;
2323         struct packet_sock *po = pkt_sk(sk);
2324         unsigned int mask = datagram_poll(file, sock, wait);
2325
2326         spin_lock_bh(&sk->sk_receive_queue.lock);
2327         if (po->rx_ring.pg_vec) {
2328                 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2329                         mask |= POLLIN | POLLRDNORM;
2330         }
2331         spin_unlock_bh(&sk->sk_receive_queue.lock);
2332         spin_lock_bh(&sk->sk_write_queue.lock);
2333         if (po->tx_ring.pg_vec) {
2334                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2335                         mask |= POLLOUT | POLLWRNORM;
2336         }
2337         spin_unlock_bh(&sk->sk_write_queue.lock);
2338         return mask;
2339 }
2340
2341
2342 /* Dirty? Well, I still did not learn better way to account
2343  * for user mmaps.
2344  */
2345
2346 static void packet_mm_open(struct vm_area_struct *vma)
2347 {
2348         struct file *file = vma->vm_file;
2349         struct socket *sock = file->private_data;
2350         struct sock *sk = sock->sk;
2351
2352         if (sk)
2353                 atomic_inc(&pkt_sk(sk)->mapped);
2354 }
2355
2356 static void packet_mm_close(struct vm_area_struct *vma)
2357 {
2358         struct file *file = vma->vm_file;
2359         struct socket *sock = file->private_data;
2360         struct sock *sk = sock->sk;
2361
2362         if (sk)
2363                 atomic_dec(&pkt_sk(sk)->mapped);
2364 }
2365
2366 static const struct vm_operations_struct packet_mmap_ops = {
2367         .open   =       packet_mm_open,
2368         .close  =       packet_mm_close,
2369 };
2370
2371 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2372                         unsigned int len)
2373 {
2374         int i;
2375
2376         for (i = 0; i < len; i++) {
2377                 if (likely(pg_vec[i].buffer)) {
2378                         if (is_vmalloc_addr(pg_vec[i].buffer))
2379                                 vfree(pg_vec[i].buffer);
2380                         else
2381                                 free_pages((unsigned long)pg_vec[i].buffer,
2382                                            order);
2383                         pg_vec[i].buffer = NULL;
2384                 }
2385         }
2386         kfree(pg_vec);
2387 }
2388
2389 static inline char *alloc_one_pg_vec_page(unsigned long order)
2390 {
2391         char *buffer = NULL;
2392         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2393                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2394
2395         buffer = (char *) __get_free_pages(gfp_flags, order);
2396
2397         if (buffer)
2398                 return buffer;
2399
2400         /*
2401          * __get_free_pages failed, fall back to vmalloc
2402          */
2403         buffer = vzalloc((1 << order) * PAGE_SIZE);
2404
2405         if (buffer)
2406                 return buffer;
2407
2408         /*
2409          * vmalloc failed, lets dig into swap here
2410          */
2411         gfp_flags &= ~__GFP_NORETRY;
2412         buffer = (char *)__get_free_pages(gfp_flags, order);
2413         if (buffer)
2414                 return buffer;
2415
2416         /*
2417          * complete and utter failure
2418          */
2419         return NULL;
2420 }
2421
2422 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2423 {
2424         unsigned int block_nr = req->tp_block_nr;
2425         struct pgv *pg_vec;
2426         int i;
2427
2428         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2429         if (unlikely(!pg_vec))
2430                 goto out;
2431
2432         for (i = 0; i < block_nr; i++) {
2433                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2434                 if (unlikely(!pg_vec[i].buffer))
2435                         goto out_free_pgvec;
2436         }
2437
2438 out:
2439         return pg_vec;
2440
2441 out_free_pgvec:
2442         free_pg_vec(pg_vec, order, block_nr);
2443         pg_vec = NULL;
2444         goto out;
2445 }
2446
2447 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2448                 int closing, int tx_ring)
2449 {
2450         struct pgv *pg_vec = NULL;
2451         struct packet_sock *po = pkt_sk(sk);
2452         int was_running, order = 0;
2453         struct packet_ring_buffer *rb;
2454         struct sk_buff_head *rb_queue;
2455         __be16 num;
2456         int err;
2457
2458         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2459         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2460
2461         err = -EBUSY;
2462         if (!closing) {
2463                 if (atomic_read(&po->mapped))
2464                         goto out;
2465                 if (atomic_read(&rb->pending))
2466                         goto out;
2467         }
2468
2469         if (req->tp_block_nr) {
2470                 /* Sanity tests and some calculations */
2471                 err = -EBUSY;
2472                 if (unlikely(rb->pg_vec))
2473                         goto out;
2474
2475                 switch (po->tp_version) {
2476                 case TPACKET_V1:
2477                         po->tp_hdrlen = TPACKET_HDRLEN;
2478                         break;
2479                 case TPACKET_V2:
2480                         po->tp_hdrlen = TPACKET2_HDRLEN;
2481                         break;
2482                 }
2483
2484                 err = -EINVAL;
2485                 if (unlikely((int)req->tp_block_size <= 0))
2486                         goto out;
2487                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2488                         goto out;
2489                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2490                                         po->tp_reserve))
2491                         goto out;
2492                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2493                         goto out;
2494
2495                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2496                 if (unlikely(rb->frames_per_block <= 0))
2497                         goto out;
2498                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2499                                         req->tp_frame_nr))
2500                         goto out;
2501
2502                 err = -ENOMEM;
2503                 order = get_order(req->tp_block_size);
2504                 pg_vec = alloc_pg_vec(req, order);
2505                 if (unlikely(!pg_vec))
2506                         goto out;
2507         }
2508         /* Done */
2509         else {
2510                 err = -EINVAL;
2511                 if (unlikely(req->tp_frame_nr))
2512                         goto out;
2513         }
2514
2515         lock_sock(sk);
2516
2517         /* Detach socket from network */
2518         spin_lock(&po->bind_lock);
2519         was_running = po->running;
2520         num = po->num;
2521         if (was_running) {
2522                 __dev_remove_pack(&po->prot_hook);
2523                 po->num = 0;
2524                 po->running = 0;
2525                 __sock_put(sk);
2526         }
2527         spin_unlock(&po->bind_lock);
2528
2529         synchronize_net();
2530
2531         err = -EBUSY;
2532         mutex_lock(&po->pg_vec_lock);
2533         if (closing || atomic_read(&po->mapped) == 0) {
2534                 err = 0;
2535                 spin_lock_bh(&rb_queue->lock);
2536                 swap(rb->pg_vec, pg_vec);
2537                 rb->frame_max = (req->tp_frame_nr - 1);
2538                 rb->head = 0;
2539                 rb->frame_size = req->tp_frame_size;
2540                 spin_unlock_bh(&rb_queue->lock);
2541
2542                 swap(rb->pg_vec_order, order);
2543                 swap(rb->pg_vec_len, req->tp_block_nr);
2544
2545                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2546                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2547                                                 tpacket_rcv : packet_rcv;
2548                 skb_queue_purge(rb_queue);
2549                 if (atomic_read(&po->mapped))
2550                         pr_err("packet_mmap: vma is busy: %d\n",
2551                                atomic_read(&po->mapped));
2552         }
2553         mutex_unlock(&po->pg_vec_lock);
2554
2555         spin_lock(&po->bind_lock);
2556         if (was_running && !po->running) {
2557                 sock_hold(sk);
2558                 po->running = 1;
2559                 po->num = num;
2560                 dev_add_pack(&po->prot_hook);
2561         }
2562         spin_unlock(&po->bind_lock);
2563
2564         release_sock(sk);
2565
2566         if (pg_vec)
2567                 free_pg_vec(pg_vec, order, req->tp_block_nr);
2568 out:
2569         return err;
2570 }
2571
2572 static int packet_mmap(struct file *file, struct socket *sock,
2573                 struct vm_area_struct *vma)
2574 {
2575         struct sock *sk = sock->sk;
2576         struct packet_sock *po = pkt_sk(sk);
2577         unsigned long size, expected_size;
2578         struct packet_ring_buffer *rb;
2579         unsigned long start;
2580         int err = -EINVAL;
2581         int i;
2582
2583         if (vma->vm_pgoff)
2584                 return -EINVAL;
2585
2586         mutex_lock(&po->pg_vec_lock);
2587
2588         expected_size = 0;
2589         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2590                 if (rb->pg_vec) {
2591                         expected_size += rb->pg_vec_len
2592                                                 * rb->pg_vec_pages
2593                                                 * PAGE_SIZE;
2594                 }
2595         }
2596
2597         if (expected_size == 0)
2598                 goto out;
2599
2600         size = vma->vm_end - vma->vm_start;
2601         if (size != expected_size)
2602                 goto out;
2603
2604         start = vma->vm_start;
2605         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2606                 if (rb->pg_vec == NULL)
2607                         continue;
2608
2609                 for (i = 0; i < rb->pg_vec_len; i++) {
2610                         struct page *page;
2611                         void *kaddr = rb->pg_vec[i].buffer;
2612                         int pg_num;
2613
2614                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2615                                 page = pgv_to_page(kaddr);
2616                                 err = vm_insert_page(vma, start, page);
2617                                 if (unlikely(err))
2618                                         goto out;
2619                                 start += PAGE_SIZE;
2620                                 kaddr += PAGE_SIZE;
2621                         }
2622                 }
2623         }
2624
2625         atomic_inc(&po->mapped);
2626         vma->vm_ops = &packet_mmap_ops;
2627         err = 0;
2628
2629 out:
2630         mutex_unlock(&po->pg_vec_lock);
2631         return err;
2632 }
2633
2634 static const struct proto_ops packet_ops_spkt = {
2635         .family =       PF_PACKET,
2636         .owner =        THIS_MODULE,
2637         .release =      packet_release,
2638         .bind =         packet_bind_spkt,
2639         .connect =      sock_no_connect,
2640         .socketpair =   sock_no_socketpair,
2641         .accept =       sock_no_accept,
2642         .getname =      packet_getname_spkt,
2643         .poll =         datagram_poll,
2644         .ioctl =        packet_ioctl,
2645         .listen =       sock_no_listen,
2646         .shutdown =     sock_no_shutdown,
2647         .setsockopt =   sock_no_setsockopt,
2648         .getsockopt =   sock_no_getsockopt,
2649         .sendmsg =      packet_sendmsg_spkt,
2650         .recvmsg =      packet_recvmsg,
2651         .mmap =         sock_no_mmap,
2652         .sendpage =     sock_no_sendpage,
2653 };
2654
2655 static const struct proto_ops packet_ops = {
2656         .family =       PF_PACKET,
2657         .owner =        THIS_MODULE,
2658         .release =      packet_release,
2659         .bind =         packet_bind,
2660         .connect =      sock_no_connect,
2661         .socketpair =   sock_no_socketpair,
2662         .accept =       sock_no_accept,
2663         .getname =      packet_getname,
2664         .poll =         packet_poll,
2665         .ioctl =        packet_ioctl,
2666         .listen =       sock_no_listen,
2667         .shutdown =     sock_no_shutdown,
2668         .setsockopt =   packet_setsockopt,
2669         .getsockopt =   packet_getsockopt,
2670         .sendmsg =      packet_sendmsg,
2671         .recvmsg =      packet_recvmsg,
2672         .mmap =         packet_mmap,
2673         .sendpage =     sock_no_sendpage,
2674 };
2675
2676 static const struct net_proto_family packet_family_ops = {
2677         .family =       PF_PACKET,
2678         .create =       packet_create,
2679         .owner  =       THIS_MODULE,
2680 };
2681
2682 static struct notifier_block packet_netdev_notifier = {
2683         .notifier_call =        packet_notifier,
2684 };
2685
2686 #ifdef CONFIG_PROC_FS
2687
2688 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2689         __acquires(RCU)
2690 {
2691         struct net *net = seq_file_net(seq);
2692
2693         rcu_read_lock();
2694         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2695 }
2696
2697 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2698 {
2699         struct net *net = seq_file_net(seq);
2700         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2701 }
2702
2703 static void packet_seq_stop(struct seq_file *seq, void *v)
2704         __releases(RCU)
2705 {
2706         rcu_read_unlock();
2707 }
2708
2709 static int packet_seq_show(struct seq_file *seq, void *v)
2710 {
2711         if (v == SEQ_START_TOKEN)
2712                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
2713         else {
2714                 struct sock *s = sk_entry(v);
2715                 const struct packet_sock *po = pkt_sk(s);
2716
2717                 seq_printf(seq,
2718                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
2719                            s,
2720                            atomic_read(&s->sk_refcnt),
2721                            s->sk_type,
2722                            ntohs(po->num),
2723                            po->ifindex,
2724                            po->running,
2725                            atomic_read(&s->sk_rmem_alloc),
2726                            sock_i_uid(s),
2727                            sock_i_ino(s));
2728         }
2729
2730         return 0;
2731 }
2732
2733 static const struct seq_operations packet_seq_ops = {
2734         .start  = packet_seq_start,
2735         .next   = packet_seq_next,
2736         .stop   = packet_seq_stop,
2737         .show   = packet_seq_show,
2738 };
2739
2740 static int packet_seq_open(struct inode *inode, struct file *file)
2741 {
2742         return seq_open_net(inode, file, &packet_seq_ops,
2743                             sizeof(struct seq_net_private));
2744 }
2745
2746 static const struct file_operations packet_seq_fops = {
2747         .owner          = THIS_MODULE,
2748         .open           = packet_seq_open,
2749         .read           = seq_read,
2750         .llseek         = seq_lseek,
2751         .release        = seq_release_net,
2752 };
2753
2754 #endif
2755
2756 static int __net_init packet_net_init(struct net *net)
2757 {
2758         spin_lock_init(&net->packet.sklist_lock);
2759         INIT_HLIST_HEAD(&net->packet.sklist);
2760
2761         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2762                 return -ENOMEM;
2763
2764         return 0;
2765 }
2766
2767 static void __net_exit packet_net_exit(struct net *net)
2768 {
2769         proc_net_remove(net, "packet");
2770 }
2771
2772 static struct pernet_operations packet_net_ops = {
2773         .init = packet_net_init,
2774         .exit = packet_net_exit,
2775 };
2776
2777
2778 static void __exit packet_exit(void)
2779 {
2780         unregister_netdevice_notifier(&packet_netdev_notifier);
2781         unregister_pernet_subsys(&packet_net_ops);
2782         sock_unregister(PF_PACKET);
2783         proto_unregister(&packet_proto);
2784 }
2785
2786 static int __init packet_init(void)
2787 {
2788         int rc = proto_register(&packet_proto, 0);
2789
2790         if (rc != 0)
2791                 goto out;
2792
2793         sock_register(&packet_family_ops);
2794         register_pernet_subsys(&packet_net_ops);
2795         register_netdevice_notifier(&packet_netdev_notifier);
2796 out:
2797         return rc;
2798 }
2799
2800 module_init(packet_init);
2801 module_exit(packet_exit);
2802 MODULE_LICENSE("GPL");
2803 MODULE_ALIAS_NETPROTO(PF_PACKET);