Merge branch 'for-linus' of git://git.kernel.dk/linux-block
[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                 h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
802                 hdrlen = sizeof(*h.h2);
803                 break;
804         default:
805                 BUG();
806         }
807
808         sll = h.raw + TPACKET_ALIGN(hdrlen);
809         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
810         sll->sll_family = AF_PACKET;
811         sll->sll_hatype = dev->type;
812         sll->sll_protocol = skb->protocol;
813         sll->sll_pkttype = skb->pkt_type;
814         if (unlikely(po->origdev))
815                 sll->sll_ifindex = orig_dev->ifindex;
816         else
817                 sll->sll_ifindex = dev->ifindex;
818
819         __packet_set_status(po, h.raw, status);
820         smp_mb();
821 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
822         {
823                 u8 *start, *end;
824
825                 end = (u8 *)PAGE_ALIGN((unsigned long)h.raw + macoff + snaplen);
826                 for (start = h.raw; start < end; start += PAGE_SIZE)
827                         flush_dcache_page(pgv_to_page(start));
828         }
829 #endif
830
831         sk->sk_data_ready(sk, 0);
832
833 drop_n_restore:
834         if (skb_head != skb->data && skb_shared(skb)) {
835                 skb->data = skb_head;
836                 skb->len = skb_len;
837         }
838 drop:
839         kfree_skb(skb);
840         return 0;
841
842 ring_is_full:
843         po->stats.tp_drops++;
844         spin_unlock(&sk->sk_receive_queue.lock);
845
846         sk->sk_data_ready(sk, 0);
847         kfree_skb(copy_skb);
848         goto drop_n_restore;
849 }
850
851 static void tpacket_destruct_skb(struct sk_buff *skb)
852 {
853         struct packet_sock *po = pkt_sk(skb->sk);
854         void *ph;
855
856         BUG_ON(skb == NULL);
857
858         if (likely(po->tx_ring.pg_vec)) {
859                 ph = skb_shinfo(skb)->destructor_arg;
860                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
861                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
862                 atomic_dec(&po->tx_ring.pending);
863                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
864         }
865
866         sock_wfree(skb);
867 }
868
869 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
870                 void *frame, struct net_device *dev, int size_max,
871                 __be16 proto, unsigned char *addr)
872 {
873         union {
874                 struct tpacket_hdr *h1;
875                 struct tpacket2_hdr *h2;
876                 void *raw;
877         } ph;
878         int to_write, offset, len, tp_len, nr_frags, len_max;
879         struct socket *sock = po->sk.sk_socket;
880         struct page *page;
881         void *data;
882         int err;
883
884         ph.raw = frame;
885
886         skb->protocol = proto;
887         skb->dev = dev;
888         skb->priority = po->sk.sk_priority;
889         skb->mark = po->sk.sk_mark;
890         skb_shinfo(skb)->destructor_arg = ph.raw;
891
892         switch (po->tp_version) {
893         case TPACKET_V2:
894                 tp_len = ph.h2->tp_len;
895                 break;
896         default:
897                 tp_len = ph.h1->tp_len;
898                 break;
899         }
900         if (unlikely(tp_len > size_max)) {
901                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
902                 return -EMSGSIZE;
903         }
904
905         skb_reserve(skb, LL_RESERVED_SPACE(dev));
906         skb_reset_network_header(skb);
907
908         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
909         to_write = tp_len;
910
911         if (sock->type == SOCK_DGRAM) {
912                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
913                                 NULL, tp_len);
914                 if (unlikely(err < 0))
915                         return -EINVAL;
916         } else if (dev->hard_header_len) {
917                 /* net device doesn't like empty head */
918                 if (unlikely(tp_len <= dev->hard_header_len)) {
919                         pr_err("packet size is too short (%d < %d)\n",
920                                tp_len, dev->hard_header_len);
921                         return -EINVAL;
922                 }
923
924                 skb_push(skb, dev->hard_header_len);
925                 err = skb_store_bits(skb, 0, data,
926                                 dev->hard_header_len);
927                 if (unlikely(err))
928                         return err;
929
930                 data += dev->hard_header_len;
931                 to_write -= dev->hard_header_len;
932         }
933
934         err = -EFAULT;
935         offset = offset_in_page(data);
936         len_max = PAGE_SIZE - offset;
937         len = ((to_write > len_max) ? len_max : to_write);
938
939         skb->data_len = to_write;
940         skb->len += to_write;
941         skb->truesize += to_write;
942         atomic_add(to_write, &po->sk.sk_wmem_alloc);
943
944         while (likely(to_write)) {
945                 nr_frags = skb_shinfo(skb)->nr_frags;
946
947                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
948                         pr_err("Packet exceed the number of skb frags(%lu)\n",
949                                MAX_SKB_FRAGS);
950                         return -EFAULT;
951                 }
952
953                 page = pgv_to_page(data);
954                 data += len;
955                 flush_dcache_page(page);
956                 get_page(page);
957                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
958                 to_write -= len;
959                 offset = 0;
960                 len_max = PAGE_SIZE;
961                 len = ((to_write > len_max) ? len_max : to_write);
962         }
963
964         return tp_len;
965 }
966
967 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
968 {
969         struct sk_buff *skb;
970         struct net_device *dev;
971         __be16 proto;
972         int ifindex, err, reserve = 0;
973         void *ph;
974         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
975         int tp_len, size_max;
976         unsigned char *addr;
977         int len_sum = 0;
978         int status = 0;
979
980         mutex_lock(&po->pg_vec_lock);
981
982         err = -EBUSY;
983         if (saddr == NULL) {
984                 ifindex = po->ifindex;
985                 proto   = po->num;
986                 addr    = NULL;
987         } else {
988                 err = -EINVAL;
989                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
990                         goto out;
991                 if (msg->msg_namelen < (saddr->sll_halen
992                                         + offsetof(struct sockaddr_ll,
993                                                 sll_addr)))
994                         goto out;
995                 ifindex = saddr->sll_ifindex;
996                 proto   = saddr->sll_protocol;
997                 addr    = saddr->sll_addr;
998         }
999
1000         dev = dev_get_by_index(sock_net(&po->sk), ifindex);
1001         err = -ENXIO;
1002         if (unlikely(dev == NULL))
1003                 goto out;
1004
1005         reserve = dev->hard_header_len;
1006
1007         err = -ENETDOWN;
1008         if (unlikely(!(dev->flags & IFF_UP)))
1009                 goto out_put;
1010
1011         size_max = po->tx_ring.frame_size
1012                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
1013
1014         if (size_max > dev->mtu + reserve)
1015                 size_max = dev->mtu + reserve;
1016
1017         do {
1018                 ph = packet_current_frame(po, &po->tx_ring,
1019                                 TP_STATUS_SEND_REQUEST);
1020
1021                 if (unlikely(ph == NULL)) {
1022                         schedule();
1023                         continue;
1024                 }
1025
1026                 status = TP_STATUS_SEND_REQUEST;
1027                 skb = sock_alloc_send_skb(&po->sk,
1028                                 LL_ALLOCATED_SPACE(dev)
1029                                 + sizeof(struct sockaddr_ll),
1030                                 0, &err);
1031
1032                 if (unlikely(skb == NULL))
1033                         goto out_status;
1034
1035                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
1036                                 addr);
1037
1038                 if (unlikely(tp_len < 0)) {
1039                         if (po->tp_loss) {
1040                                 __packet_set_status(po, ph,
1041                                                 TP_STATUS_AVAILABLE);
1042                                 packet_increment_head(&po->tx_ring);
1043                                 kfree_skb(skb);
1044                                 continue;
1045                         } else {
1046                                 status = TP_STATUS_WRONG_FORMAT;
1047                                 err = tp_len;
1048                                 goto out_status;
1049                         }
1050                 }
1051
1052                 skb->destructor = tpacket_destruct_skb;
1053                 __packet_set_status(po, ph, TP_STATUS_SENDING);
1054                 atomic_inc(&po->tx_ring.pending);
1055
1056                 status = TP_STATUS_SEND_REQUEST;
1057                 err = dev_queue_xmit(skb);
1058                 if (unlikely(err > 0)) {
1059                         err = net_xmit_errno(err);
1060                         if (err && __packet_get_status(po, ph) ==
1061                                    TP_STATUS_AVAILABLE) {
1062                                 /* skb was destructed already */
1063                                 skb = NULL;
1064                                 goto out_status;
1065                         }
1066                         /*
1067                          * skb was dropped but not destructed yet;
1068                          * let's treat it like congestion or err < 0
1069                          */
1070                         err = 0;
1071                 }
1072                 packet_increment_head(&po->tx_ring);
1073                 len_sum += tp_len;
1074         } while (likely((ph != NULL) ||
1075                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
1076                          (atomic_read(&po->tx_ring.pending))))
1077                 );
1078
1079         err = len_sum;
1080         goto out_put;
1081
1082 out_status:
1083         __packet_set_status(po, ph, status);
1084         kfree_skb(skb);
1085 out_put:
1086         dev_put(dev);
1087 out:
1088         mutex_unlock(&po->pg_vec_lock);
1089         return err;
1090 }
1091
1092 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
1093                                                size_t reserve, size_t len,
1094                                                size_t linear, int noblock,
1095                                                int *err)
1096 {
1097         struct sk_buff *skb;
1098
1099         /* Under a page?  Don't bother with paged skb. */
1100         if (prepad + len < PAGE_SIZE || !linear)
1101                 linear = len;
1102
1103         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
1104                                    err);
1105         if (!skb)
1106                 return NULL;
1107
1108         skb_reserve(skb, reserve);
1109         skb_put(skb, linear);
1110         skb->data_len = len - linear;
1111         skb->len += len - linear;
1112
1113         return skb;
1114 }
1115
1116 static int packet_snd(struct socket *sock,
1117                           struct msghdr *msg, size_t len)
1118 {
1119         struct sock *sk = sock->sk;
1120         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
1121         struct sk_buff *skb;
1122         struct net_device *dev;
1123         __be16 proto;
1124         unsigned char *addr;
1125         int ifindex, err, reserve = 0;
1126         struct virtio_net_hdr vnet_hdr = { 0 };
1127         int offset = 0;
1128         int vnet_hdr_len;
1129         struct packet_sock *po = pkt_sk(sk);
1130         unsigned short gso_type = 0;
1131
1132         /*
1133          *      Get and verify the address.
1134          */
1135
1136         if (saddr == NULL) {
1137                 ifindex = po->ifindex;
1138                 proto   = po->num;
1139                 addr    = NULL;
1140         } else {
1141                 err = -EINVAL;
1142                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
1143                         goto out;
1144                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
1145                         goto out;
1146                 ifindex = saddr->sll_ifindex;
1147                 proto   = saddr->sll_protocol;
1148                 addr    = saddr->sll_addr;
1149         }
1150
1151
1152         dev = dev_get_by_index(sock_net(sk), ifindex);
1153         err = -ENXIO;
1154         if (dev == NULL)
1155                 goto out_unlock;
1156         if (sock->type == SOCK_RAW)
1157                 reserve = dev->hard_header_len;
1158
1159         err = -ENETDOWN;
1160         if (!(dev->flags & IFF_UP))
1161                 goto out_unlock;
1162
1163         if (po->has_vnet_hdr) {
1164                 vnet_hdr_len = sizeof(vnet_hdr);
1165
1166                 err = -EINVAL;
1167                 if (len < vnet_hdr_len)
1168                         goto out_unlock;
1169
1170                 len -= vnet_hdr_len;
1171
1172                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
1173                                        vnet_hdr_len);
1174                 if (err < 0)
1175                         goto out_unlock;
1176
1177                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
1178                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
1179                       vnet_hdr.hdr_len))
1180                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
1181                                                  vnet_hdr.csum_offset + 2;
1182
1183                 err = -EINVAL;
1184                 if (vnet_hdr.hdr_len > len)
1185                         goto out_unlock;
1186
1187                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
1188                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
1189                         case VIRTIO_NET_HDR_GSO_TCPV4:
1190                                 gso_type = SKB_GSO_TCPV4;
1191                                 break;
1192                         case VIRTIO_NET_HDR_GSO_TCPV6:
1193                                 gso_type = SKB_GSO_TCPV6;
1194                                 break;
1195                         case VIRTIO_NET_HDR_GSO_UDP:
1196                                 gso_type = SKB_GSO_UDP;
1197                                 break;
1198                         default:
1199                                 goto out_unlock;
1200                         }
1201
1202                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
1203                                 gso_type |= SKB_GSO_TCP_ECN;
1204
1205                         if (vnet_hdr.gso_size == 0)
1206                                 goto out_unlock;
1207
1208                 }
1209         }
1210
1211         err = -EMSGSIZE;
1212         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
1213                 goto out_unlock;
1214
1215         err = -ENOBUFS;
1216         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
1217                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
1218                                msg->msg_flags & MSG_DONTWAIT, &err);
1219         if (skb == NULL)
1220                 goto out_unlock;
1221
1222         skb_set_network_header(skb, reserve);
1223
1224         err = -EINVAL;
1225         if (sock->type == SOCK_DGRAM &&
1226             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
1227                 goto out_free;
1228
1229         /* Returns -EFAULT on error */
1230         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
1231         if (err)
1232                 goto out_free;
1233         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1234         if (err < 0)
1235                 goto out_free;
1236
1237         if (!gso_type && (len > dev->mtu + reserve)) {
1238                 /* Earlier code assumed this would be a VLAN pkt,
1239                  * double-check this now that we have the actual
1240                  * packet in hand.
1241                  */
1242                 struct ethhdr *ehdr;
1243                 skb_reset_mac_header(skb);
1244                 ehdr = eth_hdr(skb);
1245                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1246                         err = -EMSGSIZE;
1247                         goto out_free;
1248                 }
1249         }
1250
1251         skb->protocol = proto;
1252         skb->dev = dev;
1253         skb->priority = sk->sk_priority;
1254         skb->mark = sk->sk_mark;
1255
1256         if (po->has_vnet_hdr) {
1257                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1258                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
1259                                                   vnet_hdr.csum_offset)) {
1260                                 err = -EINVAL;
1261                                 goto out_free;
1262                         }
1263                 }
1264
1265                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
1266                 skb_shinfo(skb)->gso_type = gso_type;
1267
1268                 /* Header must be checked, and gso_segs computed. */
1269                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1270                 skb_shinfo(skb)->gso_segs = 0;
1271
1272                 len += vnet_hdr_len;
1273         }
1274
1275         /*
1276          *      Now send it
1277          */
1278
1279         err = dev_queue_xmit(skb);
1280         if (err > 0 && (err = net_xmit_errno(err)) != 0)
1281                 goto out_unlock;
1282
1283         dev_put(dev);
1284
1285         return len;
1286
1287 out_free:
1288         kfree_skb(skb);
1289 out_unlock:
1290         if (dev)
1291                 dev_put(dev);
1292 out:
1293         return err;
1294 }
1295
1296 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
1297                 struct msghdr *msg, size_t len)
1298 {
1299         struct sock *sk = sock->sk;
1300         struct packet_sock *po = pkt_sk(sk);
1301         if (po->tx_ring.pg_vec)
1302                 return tpacket_snd(po, msg);
1303         else
1304                 return packet_snd(sock, msg, len);
1305 }
1306
1307 /*
1308  *      Close a PACKET socket. This is fairly simple. We immediately go
1309  *      to 'closed' state and remove our protocol entry in the device list.
1310  */
1311
1312 static int packet_release(struct socket *sock)
1313 {
1314         struct sock *sk = sock->sk;
1315         struct packet_sock *po;
1316         struct net *net;
1317         struct tpacket_req req;
1318
1319         if (!sk)
1320                 return 0;
1321
1322         net = sock_net(sk);
1323         po = pkt_sk(sk);
1324
1325         spin_lock_bh(&net->packet.sklist_lock);
1326         sk_del_node_init_rcu(sk);
1327         sock_prot_inuse_add(net, sk->sk_prot, -1);
1328         spin_unlock_bh(&net->packet.sklist_lock);
1329
1330         spin_lock(&po->bind_lock);
1331         if (po->running) {
1332                 /*
1333                  * Remove from protocol table
1334                  */
1335                 po->running = 0;
1336                 po->num = 0;
1337                 __dev_remove_pack(&po->prot_hook);
1338                 __sock_put(sk);
1339         }
1340         spin_unlock(&po->bind_lock);
1341
1342         packet_flush_mclist(sk);
1343
1344         memset(&req, 0, sizeof(req));
1345
1346         if (po->rx_ring.pg_vec)
1347                 packet_set_ring(sk, &req, 1, 0);
1348
1349         if (po->tx_ring.pg_vec)
1350                 packet_set_ring(sk, &req, 1, 1);
1351
1352         synchronize_net();
1353         /*
1354          *      Now the socket is dead. No more input will appear.
1355          */
1356         sock_orphan(sk);
1357         sock->sk = NULL;
1358
1359         /* Purge queues */
1360
1361         skb_queue_purge(&sk->sk_receive_queue);
1362         sk_refcnt_debug_release(sk);
1363
1364         sock_put(sk);
1365         return 0;
1366 }
1367
1368 /*
1369  *      Attach a packet hook.
1370  */
1371
1372 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
1373 {
1374         struct packet_sock *po = pkt_sk(sk);
1375         /*
1376          *      Detach an existing hook if present.
1377          */
1378
1379         lock_sock(sk);
1380
1381         spin_lock(&po->bind_lock);
1382         if (po->running) {
1383                 __sock_put(sk);
1384                 po->running = 0;
1385                 po->num = 0;
1386                 spin_unlock(&po->bind_lock);
1387                 dev_remove_pack(&po->prot_hook);
1388                 spin_lock(&po->bind_lock);
1389         }
1390
1391         po->num = protocol;
1392         po->prot_hook.type = protocol;
1393         po->prot_hook.dev = dev;
1394
1395         po->ifindex = dev ? dev->ifindex : 0;
1396
1397         if (protocol == 0)
1398                 goto out_unlock;
1399
1400         if (!dev || (dev->flags & IFF_UP)) {
1401                 dev_add_pack(&po->prot_hook);
1402                 sock_hold(sk);
1403                 po->running = 1;
1404         } else {
1405                 sk->sk_err = ENETDOWN;
1406                 if (!sock_flag(sk, SOCK_DEAD))
1407                         sk->sk_error_report(sk);
1408         }
1409
1410 out_unlock:
1411         spin_unlock(&po->bind_lock);
1412         release_sock(sk);
1413         return 0;
1414 }
1415
1416 /*
1417  *      Bind a packet socket to a device
1418  */
1419
1420 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
1421                             int addr_len)
1422 {
1423         struct sock *sk = sock->sk;
1424         char name[15];
1425         struct net_device *dev;
1426         int err = -ENODEV;
1427
1428         /*
1429          *      Check legality
1430          */
1431
1432         if (addr_len != sizeof(struct sockaddr))
1433                 return -EINVAL;
1434         strlcpy(name, uaddr->sa_data, sizeof(name));
1435
1436         dev = dev_get_by_name(sock_net(sk), name);
1437         if (dev) {
1438                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
1439                 dev_put(dev);
1440         }
1441         return err;
1442 }
1443
1444 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
1445 {
1446         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
1447         struct sock *sk = sock->sk;
1448         struct net_device *dev = NULL;
1449         int err;
1450
1451
1452         /*
1453          *      Check legality
1454          */
1455
1456         if (addr_len < sizeof(struct sockaddr_ll))
1457                 return -EINVAL;
1458         if (sll->sll_family != AF_PACKET)
1459                 return -EINVAL;
1460
1461         if (sll->sll_ifindex) {
1462                 err = -ENODEV;
1463                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
1464                 if (dev == NULL)
1465                         goto out;
1466         }
1467         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
1468         if (dev)
1469                 dev_put(dev);
1470
1471 out:
1472         return err;
1473 }
1474
1475 static struct proto packet_proto = {
1476         .name     = "PACKET",
1477         .owner    = THIS_MODULE,
1478         .obj_size = sizeof(struct packet_sock),
1479 };
1480
1481 /*
1482  *      Create a packet of type SOCK_PACKET.
1483  */
1484
1485 static int packet_create(struct net *net, struct socket *sock, int protocol,
1486                          int kern)
1487 {
1488         struct sock *sk;
1489         struct packet_sock *po;
1490         __be16 proto = (__force __be16)protocol; /* weird, but documented */
1491         int err;
1492
1493         if (!capable(CAP_NET_RAW))
1494                 return -EPERM;
1495         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
1496             sock->type != SOCK_PACKET)
1497                 return -ESOCKTNOSUPPORT;
1498
1499         sock->state = SS_UNCONNECTED;
1500
1501         err = -ENOBUFS;
1502         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
1503         if (sk == NULL)
1504                 goto out;
1505
1506         sock->ops = &packet_ops;
1507         if (sock->type == SOCK_PACKET)
1508                 sock->ops = &packet_ops_spkt;
1509
1510         sock_init_data(sock, sk);
1511
1512         po = pkt_sk(sk);
1513         sk->sk_family = PF_PACKET;
1514         po->num = proto;
1515
1516         sk->sk_destruct = packet_sock_destruct;
1517         sk_refcnt_debug_inc(sk);
1518
1519         /*
1520          *      Attach a protocol block
1521          */
1522
1523         spin_lock_init(&po->bind_lock);
1524         mutex_init(&po->pg_vec_lock);
1525         po->prot_hook.func = packet_rcv;
1526
1527         if (sock->type == SOCK_PACKET)
1528                 po->prot_hook.func = packet_rcv_spkt;
1529
1530         po->prot_hook.af_packet_priv = sk;
1531
1532         if (proto) {
1533                 po->prot_hook.type = proto;
1534                 dev_add_pack(&po->prot_hook);
1535                 sock_hold(sk);
1536                 po->running = 1;
1537         }
1538
1539         spin_lock_bh(&net->packet.sklist_lock);
1540         sk_add_node_rcu(sk, &net->packet.sklist);
1541         sock_prot_inuse_add(net, &packet_proto, 1);
1542         spin_unlock_bh(&net->packet.sklist_lock);
1543
1544         return 0;
1545 out:
1546         return err;
1547 }
1548
1549 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
1550 {
1551         struct sock_exterr_skb *serr;
1552         struct sk_buff *skb, *skb2;
1553         int copied, err;
1554
1555         err = -EAGAIN;
1556         skb = skb_dequeue(&sk->sk_error_queue);
1557         if (skb == NULL)
1558                 goto out;
1559
1560         copied = skb->len;
1561         if (copied > len) {
1562                 msg->msg_flags |= MSG_TRUNC;
1563                 copied = len;
1564         }
1565         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1566         if (err)
1567                 goto out_free_skb;
1568
1569         sock_recv_timestamp(msg, sk, skb);
1570
1571         serr = SKB_EXT_ERR(skb);
1572         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
1573                  sizeof(serr->ee), &serr->ee);
1574
1575         msg->msg_flags |= MSG_ERRQUEUE;
1576         err = copied;
1577
1578         /* Reset and regenerate socket error */
1579         spin_lock_bh(&sk->sk_error_queue.lock);
1580         sk->sk_err = 0;
1581         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
1582                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
1583                 spin_unlock_bh(&sk->sk_error_queue.lock);
1584                 sk->sk_error_report(sk);
1585         } else
1586                 spin_unlock_bh(&sk->sk_error_queue.lock);
1587
1588 out_free_skb:
1589         kfree_skb(skb);
1590 out:
1591         return err;
1592 }
1593
1594 /*
1595  *      Pull a packet from our receive queue and hand it to the user.
1596  *      If necessary we block.
1597  */
1598
1599 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1600                           struct msghdr *msg, size_t len, int flags)
1601 {
1602         struct sock *sk = sock->sk;
1603         struct sk_buff *skb;
1604         int copied, err;
1605         struct sockaddr_ll *sll;
1606         int vnet_hdr_len = 0;
1607
1608         err = -EINVAL;
1609         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
1610                 goto out;
1611
1612 #if 0
1613         /* What error should we return now? EUNATTACH? */
1614         if (pkt_sk(sk)->ifindex < 0)
1615                 return -ENODEV;
1616 #endif
1617
1618         if (flags & MSG_ERRQUEUE) {
1619                 err = packet_recv_error(sk, msg, len);
1620                 goto out;
1621         }
1622
1623         /*
1624          *      Call the generic datagram receiver. This handles all sorts
1625          *      of horrible races and re-entrancy so we can forget about it
1626          *      in the protocol layers.
1627          *
1628          *      Now it will return ENETDOWN, if device have just gone down,
1629          *      but then it will block.
1630          */
1631
1632         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
1633
1634         /*
1635          *      An error occurred so return it. Because skb_recv_datagram()
1636          *      handles the blocking we don't see and worry about blocking
1637          *      retries.
1638          */
1639
1640         if (skb == NULL)
1641                 goto out;
1642
1643         if (pkt_sk(sk)->has_vnet_hdr) {
1644                 struct virtio_net_hdr vnet_hdr = { 0 };
1645
1646                 err = -EINVAL;
1647                 vnet_hdr_len = sizeof(vnet_hdr);
1648                 if (len < vnet_hdr_len)
1649                         goto out_free;
1650
1651                 len -= vnet_hdr_len;
1652
1653                 if (skb_is_gso(skb)) {
1654                         struct skb_shared_info *sinfo = skb_shinfo(skb);
1655
1656                         /* This is a hint as to how much should be linear. */
1657                         vnet_hdr.hdr_len = skb_headlen(skb);
1658                         vnet_hdr.gso_size = sinfo->gso_size;
1659                         if (sinfo->gso_type & SKB_GSO_TCPV4)
1660                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1661                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
1662                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1663                         else if (sinfo->gso_type & SKB_GSO_UDP)
1664                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
1665                         else if (sinfo->gso_type & SKB_GSO_FCOE)
1666                                 goto out_free;
1667                         else
1668                                 BUG();
1669                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1670                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1671                 } else
1672                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
1673
1674                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1675                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1676                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
1677                         vnet_hdr.csum_offset = skb->csum_offset;
1678                 } /* else everything is zero */
1679
1680                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
1681                                      vnet_hdr_len);
1682                 if (err < 0)
1683                         goto out_free;
1684         }
1685
1686         /*
1687          *      If the address length field is there to be filled in, we fill
1688          *      it in now.
1689          */
1690
1691         sll = &PACKET_SKB_CB(skb)->sa.ll;
1692         if (sock->type == SOCK_PACKET)
1693                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1694         else
1695                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1696
1697         /*
1698          *      You lose any data beyond the buffer you gave. If it worries a
1699          *      user program they can ask the device for its MTU anyway.
1700          */
1701
1702         copied = skb->len;
1703         if (copied > len) {
1704                 copied = len;
1705                 msg->msg_flags |= MSG_TRUNC;
1706         }
1707
1708         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1709         if (err)
1710                 goto out_free;
1711
1712         sock_recv_ts_and_drops(msg, sk, skb);
1713
1714         if (msg->msg_name)
1715                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
1716                        msg->msg_namelen);
1717
1718         if (pkt_sk(sk)->auxdata) {
1719                 struct tpacket_auxdata aux;
1720
1721                 aux.tp_status = TP_STATUS_USER;
1722                 if (skb->ip_summed == CHECKSUM_PARTIAL)
1723                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
1724                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
1725                 aux.tp_snaplen = skb->len;
1726                 aux.tp_mac = 0;
1727                 aux.tp_net = skb_network_offset(skb);
1728                 aux.tp_vlan_tci = vlan_tx_tag_get(skb);
1729
1730                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
1731         }
1732
1733         /*
1734          *      Free or return the buffer as appropriate. Again this
1735          *      hides all the races and re-entrancy issues from us.
1736          */
1737         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
1738
1739 out_free:
1740         skb_free_datagram(sk, skb);
1741 out:
1742         return err;
1743 }
1744
1745 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1746                                int *uaddr_len, int peer)
1747 {
1748         struct net_device *dev;
1749         struct sock *sk = sock->sk;
1750
1751         if (peer)
1752                 return -EOPNOTSUPP;
1753
1754         uaddr->sa_family = AF_PACKET;
1755         rcu_read_lock();
1756         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
1757         if (dev)
1758                 strncpy(uaddr->sa_data, dev->name, 14);
1759         else
1760                 memset(uaddr->sa_data, 0, 14);
1761         rcu_read_unlock();
1762         *uaddr_len = sizeof(*uaddr);
1763
1764         return 0;
1765 }
1766
1767 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1768                           int *uaddr_len, int peer)
1769 {
1770         struct net_device *dev;
1771         struct sock *sk = sock->sk;
1772         struct packet_sock *po = pkt_sk(sk);
1773         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
1774
1775         if (peer)
1776                 return -EOPNOTSUPP;
1777
1778         sll->sll_family = AF_PACKET;
1779         sll->sll_ifindex = po->ifindex;
1780         sll->sll_protocol = po->num;
1781         sll->sll_pkttype = 0;
1782         rcu_read_lock();
1783         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
1784         if (dev) {
1785                 sll->sll_hatype = dev->type;
1786                 sll->sll_halen = dev->addr_len;
1787                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1788         } else {
1789                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
1790                 sll->sll_halen = 0;
1791         }
1792         rcu_read_unlock();
1793         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1794
1795         return 0;
1796 }
1797
1798 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
1799                          int what)
1800 {
1801         switch (i->type) {
1802         case PACKET_MR_MULTICAST:
1803                 if (i->alen != dev->addr_len)
1804                         return -EINVAL;
1805                 if (what > 0)
1806                         return dev_mc_add(dev, i->addr);
1807                 else
1808                         return dev_mc_del(dev, i->addr);
1809                 break;
1810         case PACKET_MR_PROMISC:
1811                 return dev_set_promiscuity(dev, what);
1812                 break;
1813         case PACKET_MR_ALLMULTI:
1814                 return dev_set_allmulti(dev, what);
1815                 break;
1816         case PACKET_MR_UNICAST:
1817                 if (i->alen != dev->addr_len)
1818                         return -EINVAL;
1819                 if (what > 0)
1820                         return dev_uc_add(dev, i->addr);
1821                 else
1822                         return dev_uc_del(dev, i->addr);
1823                 break;
1824         default:
1825                 break;
1826         }
1827         return 0;
1828 }
1829
1830 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1831 {
1832         for ( ; i; i = i->next) {
1833                 if (i->ifindex == dev->ifindex)
1834                         packet_dev_mc(dev, i, what);
1835         }
1836 }
1837
1838 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1839 {
1840         struct packet_sock *po = pkt_sk(sk);
1841         struct packet_mclist *ml, *i;
1842         struct net_device *dev;
1843         int err;
1844
1845         rtnl_lock();
1846
1847         err = -ENODEV;
1848         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
1849         if (!dev)
1850                 goto done;
1851
1852         err = -EINVAL;
1853         if (mreq->mr_alen > dev->addr_len)
1854                 goto done;
1855
1856         err = -ENOBUFS;
1857         i = kmalloc(sizeof(*i), GFP_KERNEL);
1858         if (i == NULL)
1859                 goto done;
1860
1861         err = 0;
1862         for (ml = po->mclist; ml; ml = ml->next) {
1863                 if (ml->ifindex == mreq->mr_ifindex &&
1864                     ml->type == mreq->mr_type &&
1865                     ml->alen == mreq->mr_alen &&
1866                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1867                         ml->count++;
1868                         /* Free the new element ... */
1869                         kfree(i);
1870                         goto done;
1871                 }
1872         }
1873
1874         i->type = mreq->mr_type;
1875         i->ifindex = mreq->mr_ifindex;
1876         i->alen = mreq->mr_alen;
1877         memcpy(i->addr, mreq->mr_address, i->alen);
1878         i->count = 1;
1879         i->next = po->mclist;
1880         po->mclist = i;
1881         err = packet_dev_mc(dev, i, 1);
1882         if (err) {
1883                 po->mclist = i->next;
1884                 kfree(i);
1885         }
1886
1887 done:
1888         rtnl_unlock();
1889         return err;
1890 }
1891
1892 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1893 {
1894         struct packet_mclist *ml, **mlp;
1895
1896         rtnl_lock();
1897
1898         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1899                 if (ml->ifindex == mreq->mr_ifindex &&
1900                     ml->type == mreq->mr_type &&
1901                     ml->alen == mreq->mr_alen &&
1902                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1903                         if (--ml->count == 0) {
1904                                 struct net_device *dev;
1905                                 *mlp = ml->next;
1906                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1907                                 if (dev)
1908                                         packet_dev_mc(dev, ml, -1);
1909                                 kfree(ml);
1910                         }
1911                         rtnl_unlock();
1912                         return 0;
1913                 }
1914         }
1915         rtnl_unlock();
1916         return -EADDRNOTAVAIL;
1917 }
1918
1919 static void packet_flush_mclist(struct sock *sk)
1920 {
1921         struct packet_sock *po = pkt_sk(sk);
1922         struct packet_mclist *ml;
1923
1924         if (!po->mclist)
1925                 return;
1926
1927         rtnl_lock();
1928         while ((ml = po->mclist) != NULL) {
1929                 struct net_device *dev;
1930
1931                 po->mclist = ml->next;
1932                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
1933                 if (dev != NULL)
1934                         packet_dev_mc(dev, ml, -1);
1935                 kfree(ml);
1936         }
1937         rtnl_unlock();
1938 }
1939
1940 static int
1941 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
1942 {
1943         struct sock *sk = sock->sk;
1944         struct packet_sock *po = pkt_sk(sk);
1945         int ret;
1946
1947         if (level != SOL_PACKET)
1948                 return -ENOPROTOOPT;
1949
1950         switch (optname) {
1951         case PACKET_ADD_MEMBERSHIP:
1952         case PACKET_DROP_MEMBERSHIP:
1953         {
1954                 struct packet_mreq_max mreq;
1955                 int len = optlen;
1956                 memset(&mreq, 0, sizeof(mreq));
1957                 if (len < sizeof(struct packet_mreq))
1958                         return -EINVAL;
1959                 if (len > sizeof(mreq))
1960                         len = sizeof(mreq);
1961                 if (copy_from_user(&mreq, optval, len))
1962                         return -EFAULT;
1963                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1964                         return -EINVAL;
1965                 if (optname == PACKET_ADD_MEMBERSHIP)
1966                         ret = packet_mc_add(sk, &mreq);
1967                 else
1968                         ret = packet_mc_drop(sk, &mreq);
1969                 return ret;
1970         }
1971
1972         case PACKET_RX_RING:
1973         case PACKET_TX_RING:
1974         {
1975                 struct tpacket_req req;
1976
1977                 if (optlen < sizeof(req))
1978                         return -EINVAL;
1979                 if (pkt_sk(sk)->has_vnet_hdr)
1980                         return -EINVAL;
1981                 if (copy_from_user(&req, optval, sizeof(req)))
1982                         return -EFAULT;
1983                 return packet_set_ring(sk, &req, 0, optname == PACKET_TX_RING);
1984         }
1985         case PACKET_COPY_THRESH:
1986         {
1987                 int val;
1988
1989                 if (optlen != sizeof(val))
1990                         return -EINVAL;
1991                 if (copy_from_user(&val, optval, sizeof(val)))
1992                         return -EFAULT;
1993
1994                 pkt_sk(sk)->copy_thresh = val;
1995                 return 0;
1996         }
1997         case PACKET_VERSION:
1998         {
1999                 int val;
2000
2001                 if (optlen != sizeof(val))
2002                         return -EINVAL;
2003                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2004                         return -EBUSY;
2005                 if (copy_from_user(&val, optval, sizeof(val)))
2006                         return -EFAULT;
2007                 switch (val) {
2008                 case TPACKET_V1:
2009                 case TPACKET_V2:
2010                         po->tp_version = val;
2011                         return 0;
2012                 default:
2013                         return -EINVAL;
2014                 }
2015         }
2016         case PACKET_RESERVE:
2017         {
2018                 unsigned int val;
2019
2020                 if (optlen != sizeof(val))
2021                         return -EINVAL;
2022                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2023                         return -EBUSY;
2024                 if (copy_from_user(&val, optval, sizeof(val)))
2025                         return -EFAULT;
2026                 po->tp_reserve = val;
2027                 return 0;
2028         }
2029         case PACKET_LOSS:
2030         {
2031                 unsigned int val;
2032
2033                 if (optlen != sizeof(val))
2034                         return -EINVAL;
2035                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2036                         return -EBUSY;
2037                 if (copy_from_user(&val, optval, sizeof(val)))
2038                         return -EFAULT;
2039                 po->tp_loss = !!val;
2040                 return 0;
2041         }
2042         case PACKET_AUXDATA:
2043         {
2044                 int val;
2045
2046                 if (optlen < sizeof(val))
2047                         return -EINVAL;
2048                 if (copy_from_user(&val, optval, sizeof(val)))
2049                         return -EFAULT;
2050
2051                 po->auxdata = !!val;
2052                 return 0;
2053         }
2054         case PACKET_ORIGDEV:
2055         {
2056                 int val;
2057
2058                 if (optlen < sizeof(val))
2059                         return -EINVAL;
2060                 if (copy_from_user(&val, optval, sizeof(val)))
2061                         return -EFAULT;
2062
2063                 po->origdev = !!val;
2064                 return 0;
2065         }
2066         case PACKET_VNET_HDR:
2067         {
2068                 int val;
2069
2070                 if (sock->type != SOCK_RAW)
2071                         return -EINVAL;
2072                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
2073                         return -EBUSY;
2074                 if (optlen < sizeof(val))
2075                         return -EINVAL;
2076                 if (copy_from_user(&val, optval, sizeof(val)))
2077                         return -EFAULT;
2078
2079                 po->has_vnet_hdr = !!val;
2080                 return 0;
2081         }
2082         case PACKET_TIMESTAMP:
2083         {
2084                 int val;
2085
2086                 if (optlen != sizeof(val))
2087                         return -EINVAL;
2088                 if (copy_from_user(&val, optval, sizeof(val)))
2089                         return -EFAULT;
2090
2091                 po->tp_tstamp = val;
2092                 return 0;
2093         }
2094         default:
2095                 return -ENOPROTOOPT;
2096         }
2097 }
2098
2099 static int packet_getsockopt(struct socket *sock, int level, int optname,
2100                              char __user *optval, int __user *optlen)
2101 {
2102         int len;
2103         int val;
2104         struct sock *sk = sock->sk;
2105         struct packet_sock *po = pkt_sk(sk);
2106         void *data;
2107         struct tpacket_stats st;
2108
2109         if (level != SOL_PACKET)
2110                 return -ENOPROTOOPT;
2111
2112         if (get_user(len, optlen))
2113                 return -EFAULT;
2114
2115         if (len < 0)
2116                 return -EINVAL;
2117
2118         switch (optname) {
2119         case PACKET_STATISTICS:
2120                 if (len > sizeof(struct tpacket_stats))
2121                         len = sizeof(struct tpacket_stats);
2122                 spin_lock_bh(&sk->sk_receive_queue.lock);
2123                 st = po->stats;
2124                 memset(&po->stats, 0, sizeof(st));
2125                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2126                 st.tp_packets += st.tp_drops;
2127
2128                 data = &st;
2129                 break;
2130         case PACKET_AUXDATA:
2131                 if (len > sizeof(int))
2132                         len = sizeof(int);
2133                 val = po->auxdata;
2134
2135                 data = &val;
2136                 break;
2137         case PACKET_ORIGDEV:
2138                 if (len > sizeof(int))
2139                         len = sizeof(int);
2140                 val = po->origdev;
2141
2142                 data = &val;
2143                 break;
2144         case PACKET_VNET_HDR:
2145                 if (len > sizeof(int))
2146                         len = sizeof(int);
2147                 val = po->has_vnet_hdr;
2148
2149                 data = &val;
2150                 break;
2151         case PACKET_VERSION:
2152                 if (len > sizeof(int))
2153                         len = sizeof(int);
2154                 val = po->tp_version;
2155                 data = &val;
2156                 break;
2157         case PACKET_HDRLEN:
2158                 if (len > sizeof(int))
2159                         len = sizeof(int);
2160                 if (copy_from_user(&val, optval, len))
2161                         return -EFAULT;
2162                 switch (val) {
2163                 case TPACKET_V1:
2164                         val = sizeof(struct tpacket_hdr);
2165                         break;
2166                 case TPACKET_V2:
2167                         val = sizeof(struct tpacket2_hdr);
2168                         break;
2169                 default:
2170                         return -EINVAL;
2171                 }
2172                 data = &val;
2173                 break;
2174         case PACKET_RESERVE:
2175                 if (len > sizeof(unsigned int))
2176                         len = sizeof(unsigned int);
2177                 val = po->tp_reserve;
2178                 data = &val;
2179                 break;
2180         case PACKET_LOSS:
2181                 if (len > sizeof(unsigned int))
2182                         len = sizeof(unsigned int);
2183                 val = po->tp_loss;
2184                 data = &val;
2185                 break;
2186         case PACKET_TIMESTAMP:
2187                 if (len > sizeof(int))
2188                         len = sizeof(int);
2189                 val = po->tp_tstamp;
2190                 data = &val;
2191                 break;
2192         default:
2193                 return -ENOPROTOOPT;
2194         }
2195
2196         if (put_user(len, optlen))
2197                 return -EFAULT;
2198         if (copy_to_user(optval, data, len))
2199                 return -EFAULT;
2200         return 0;
2201 }
2202
2203
2204 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
2205 {
2206         struct sock *sk;
2207         struct hlist_node *node;
2208         struct net_device *dev = data;
2209         struct net *net = dev_net(dev);
2210
2211         rcu_read_lock();
2212         sk_for_each_rcu(sk, node, &net->packet.sklist) {
2213                 struct packet_sock *po = pkt_sk(sk);
2214
2215                 switch (msg) {
2216                 case NETDEV_UNREGISTER:
2217                         if (po->mclist)
2218                                 packet_dev_mclist(dev, po->mclist, -1);
2219                         /* fallthrough */
2220
2221                 case NETDEV_DOWN:
2222                         if (dev->ifindex == po->ifindex) {
2223                                 spin_lock(&po->bind_lock);
2224                                 if (po->running) {
2225                                         __dev_remove_pack(&po->prot_hook);
2226                                         __sock_put(sk);
2227                                         po->running = 0;
2228                                         sk->sk_err = ENETDOWN;
2229                                         if (!sock_flag(sk, SOCK_DEAD))
2230                                                 sk->sk_error_report(sk);
2231                                 }
2232                                 if (msg == NETDEV_UNREGISTER) {
2233                                         po->ifindex = -1;
2234                                         po->prot_hook.dev = NULL;
2235                                 }
2236                                 spin_unlock(&po->bind_lock);
2237                         }
2238                         break;
2239                 case NETDEV_UP:
2240                         if (dev->ifindex == po->ifindex) {
2241                                 spin_lock(&po->bind_lock);
2242                                 if (po->num && !po->running) {
2243                                         dev_add_pack(&po->prot_hook);
2244                                         sock_hold(sk);
2245                                         po->running = 1;
2246                                 }
2247                                 spin_unlock(&po->bind_lock);
2248                         }
2249                         break;
2250                 }
2251         }
2252         rcu_read_unlock();
2253         return NOTIFY_DONE;
2254 }
2255
2256
2257 static int packet_ioctl(struct socket *sock, unsigned int cmd,
2258                         unsigned long arg)
2259 {
2260         struct sock *sk = sock->sk;
2261
2262         switch (cmd) {
2263         case SIOCOUTQ:
2264         {
2265                 int amount = sk_wmem_alloc_get(sk);
2266
2267                 return put_user(amount, (int __user *)arg);
2268         }
2269         case SIOCINQ:
2270         {
2271                 struct sk_buff *skb;
2272                 int amount = 0;
2273
2274                 spin_lock_bh(&sk->sk_receive_queue.lock);
2275                 skb = skb_peek(&sk->sk_receive_queue);
2276                 if (skb)
2277                         amount = skb->len;
2278                 spin_unlock_bh(&sk->sk_receive_queue.lock);
2279                 return put_user(amount, (int __user *)arg);
2280         }
2281         case SIOCGSTAMP:
2282                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
2283         case SIOCGSTAMPNS:
2284                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
2285
2286 #ifdef CONFIG_INET
2287         case SIOCADDRT:
2288         case SIOCDELRT:
2289         case SIOCDARP:
2290         case SIOCGARP:
2291         case SIOCSARP:
2292         case SIOCGIFADDR:
2293         case SIOCSIFADDR:
2294         case SIOCGIFBRDADDR:
2295         case SIOCSIFBRDADDR:
2296         case SIOCGIFNETMASK:
2297         case SIOCSIFNETMASK:
2298         case SIOCGIFDSTADDR:
2299         case SIOCSIFDSTADDR:
2300         case SIOCSIFFLAGS:
2301                 return inet_dgram_ops.ioctl(sock, cmd, arg);
2302 #endif
2303
2304         default:
2305                 return -ENOIOCTLCMD;
2306         }
2307         return 0;
2308 }
2309
2310 static unsigned int packet_poll(struct file *file, struct socket *sock,
2311                                 poll_table *wait)
2312 {
2313         struct sock *sk = sock->sk;
2314         struct packet_sock *po = pkt_sk(sk);
2315         unsigned int mask = datagram_poll(file, sock, wait);
2316
2317         spin_lock_bh(&sk->sk_receive_queue.lock);
2318         if (po->rx_ring.pg_vec) {
2319                 if (!packet_previous_frame(po, &po->rx_ring, TP_STATUS_KERNEL))
2320                         mask |= POLLIN | POLLRDNORM;
2321         }
2322         spin_unlock_bh(&sk->sk_receive_queue.lock);
2323         spin_lock_bh(&sk->sk_write_queue.lock);
2324         if (po->tx_ring.pg_vec) {
2325                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
2326                         mask |= POLLOUT | POLLWRNORM;
2327         }
2328         spin_unlock_bh(&sk->sk_write_queue.lock);
2329         return mask;
2330 }
2331
2332
2333 /* Dirty? Well, I still did not learn better way to account
2334  * for user mmaps.
2335  */
2336
2337 static void packet_mm_open(struct vm_area_struct *vma)
2338 {
2339         struct file *file = vma->vm_file;
2340         struct socket *sock = file->private_data;
2341         struct sock *sk = sock->sk;
2342
2343         if (sk)
2344                 atomic_inc(&pkt_sk(sk)->mapped);
2345 }
2346
2347 static void packet_mm_close(struct vm_area_struct *vma)
2348 {
2349         struct file *file = vma->vm_file;
2350         struct socket *sock = file->private_data;
2351         struct sock *sk = sock->sk;
2352
2353         if (sk)
2354                 atomic_dec(&pkt_sk(sk)->mapped);
2355 }
2356
2357 static const struct vm_operations_struct packet_mmap_ops = {
2358         .open   =       packet_mm_open,
2359         .close  =       packet_mm_close,
2360 };
2361
2362 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
2363                         unsigned int len)
2364 {
2365         int i;
2366
2367         for (i = 0; i < len; i++) {
2368                 if (likely(pg_vec[i].buffer)) {
2369                         if (is_vmalloc_addr(pg_vec[i].buffer))
2370                                 vfree(pg_vec[i].buffer);
2371                         else
2372                                 free_pages((unsigned long)pg_vec[i].buffer,
2373                                            order);
2374                         pg_vec[i].buffer = NULL;
2375                 }
2376         }
2377         kfree(pg_vec);
2378 }
2379
2380 static inline char *alloc_one_pg_vec_page(unsigned long order)
2381 {
2382         char *buffer = NULL;
2383         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
2384                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
2385
2386         buffer = (char *) __get_free_pages(gfp_flags, order);
2387
2388         if (buffer)
2389                 return buffer;
2390
2391         /*
2392          * __get_free_pages failed, fall back to vmalloc
2393          */
2394         buffer = vzalloc((1 << order) * PAGE_SIZE);
2395
2396         if (buffer)
2397                 return buffer;
2398
2399         /*
2400          * vmalloc failed, lets dig into swap here
2401          */
2402         gfp_flags &= ~__GFP_NORETRY;
2403         buffer = (char *)__get_free_pages(gfp_flags, order);
2404         if (buffer)
2405                 return buffer;
2406
2407         /*
2408          * complete and utter failure
2409          */
2410         return NULL;
2411 }
2412
2413 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
2414 {
2415         unsigned int block_nr = req->tp_block_nr;
2416         struct pgv *pg_vec;
2417         int i;
2418
2419         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
2420         if (unlikely(!pg_vec))
2421                 goto out;
2422
2423         for (i = 0; i < block_nr; i++) {
2424                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
2425                 if (unlikely(!pg_vec[i].buffer))
2426                         goto out_free_pgvec;
2427         }
2428
2429 out:
2430         return pg_vec;
2431
2432 out_free_pgvec:
2433         free_pg_vec(pg_vec, order, block_nr);
2434         pg_vec = NULL;
2435         goto out;
2436 }
2437
2438 static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
2439                 int closing, int tx_ring)
2440 {
2441         struct pgv *pg_vec = NULL;
2442         struct packet_sock *po = pkt_sk(sk);
2443         int was_running, order = 0;
2444         struct packet_ring_buffer *rb;
2445         struct sk_buff_head *rb_queue;
2446         __be16 num;
2447         int err;
2448
2449         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
2450         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
2451
2452         err = -EBUSY;
2453         if (!closing) {
2454                 if (atomic_read(&po->mapped))
2455                         goto out;
2456                 if (atomic_read(&rb->pending))
2457                         goto out;
2458         }
2459
2460         if (req->tp_block_nr) {
2461                 /* Sanity tests and some calculations */
2462                 err = -EBUSY;
2463                 if (unlikely(rb->pg_vec))
2464                         goto out;
2465
2466                 switch (po->tp_version) {
2467                 case TPACKET_V1:
2468                         po->tp_hdrlen = TPACKET_HDRLEN;
2469                         break;
2470                 case TPACKET_V2:
2471                         po->tp_hdrlen = TPACKET2_HDRLEN;
2472                         break;
2473                 }
2474
2475                 err = -EINVAL;
2476                 if (unlikely((int)req->tp_block_size <= 0))
2477                         goto out;
2478                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
2479                         goto out;
2480                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
2481                                         po->tp_reserve))
2482                         goto out;
2483                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
2484                         goto out;
2485
2486                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
2487                 if (unlikely(rb->frames_per_block <= 0))
2488                         goto out;
2489                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
2490                                         req->tp_frame_nr))
2491                         goto out;
2492
2493                 err = -ENOMEM;
2494                 order = get_order(req->tp_block_size);
2495                 pg_vec = alloc_pg_vec(req, order);
2496                 if (unlikely(!pg_vec))
2497                         goto out;
2498         }
2499         /* Done */
2500         else {
2501                 err = -EINVAL;
2502                 if (unlikely(req->tp_frame_nr))
2503                         goto out;
2504         }
2505
2506         lock_sock(sk);
2507
2508         /* Detach socket from network */
2509         spin_lock(&po->bind_lock);
2510         was_running = po->running;
2511         num = po->num;
2512         if (was_running) {
2513                 __dev_remove_pack(&po->prot_hook);
2514                 po->num = 0;
2515                 po->running = 0;
2516                 __sock_put(sk);
2517         }
2518         spin_unlock(&po->bind_lock);
2519
2520         synchronize_net();
2521
2522         err = -EBUSY;
2523         mutex_lock(&po->pg_vec_lock);
2524         if (closing || atomic_read(&po->mapped) == 0) {
2525                 err = 0;
2526                 spin_lock_bh(&rb_queue->lock);
2527                 swap(rb->pg_vec, pg_vec);
2528                 rb->frame_max = (req->tp_frame_nr - 1);
2529                 rb->head = 0;
2530                 rb->frame_size = req->tp_frame_size;
2531                 spin_unlock_bh(&rb_queue->lock);
2532
2533                 swap(rb->pg_vec_order, order);
2534                 swap(rb->pg_vec_len, req->tp_block_nr);
2535
2536                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
2537                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
2538                                                 tpacket_rcv : packet_rcv;
2539                 skb_queue_purge(rb_queue);
2540                 if (atomic_read(&po->mapped))
2541                         pr_err("packet_mmap: vma is busy: %d\n",
2542                                atomic_read(&po->mapped));
2543         }
2544         mutex_unlock(&po->pg_vec_lock);
2545
2546         spin_lock(&po->bind_lock);
2547         if (was_running && !po->running) {
2548                 sock_hold(sk);
2549                 po->running = 1;
2550                 po->num = num;
2551                 dev_add_pack(&po->prot_hook);
2552         }
2553         spin_unlock(&po->bind_lock);
2554
2555         release_sock(sk);
2556
2557         if (pg_vec)
2558                 free_pg_vec(pg_vec, order, req->tp_block_nr);
2559 out:
2560         return err;
2561 }
2562
2563 static int packet_mmap(struct file *file, struct socket *sock,
2564                 struct vm_area_struct *vma)
2565 {
2566         struct sock *sk = sock->sk;
2567         struct packet_sock *po = pkt_sk(sk);
2568         unsigned long size, expected_size;
2569         struct packet_ring_buffer *rb;
2570         unsigned long start;
2571         int err = -EINVAL;
2572         int i;
2573
2574         if (vma->vm_pgoff)
2575                 return -EINVAL;
2576
2577         mutex_lock(&po->pg_vec_lock);
2578
2579         expected_size = 0;
2580         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2581                 if (rb->pg_vec) {
2582                         expected_size += rb->pg_vec_len
2583                                                 * rb->pg_vec_pages
2584                                                 * PAGE_SIZE;
2585                 }
2586         }
2587
2588         if (expected_size == 0)
2589                 goto out;
2590
2591         size = vma->vm_end - vma->vm_start;
2592         if (size != expected_size)
2593                 goto out;
2594
2595         start = vma->vm_start;
2596         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
2597                 if (rb->pg_vec == NULL)
2598                         continue;
2599
2600                 for (i = 0; i < rb->pg_vec_len; i++) {
2601                         struct page *page;
2602                         void *kaddr = rb->pg_vec[i].buffer;
2603                         int pg_num;
2604
2605                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
2606                                 page = pgv_to_page(kaddr);
2607                                 err = vm_insert_page(vma, start, page);
2608                                 if (unlikely(err))
2609                                         goto out;
2610                                 start += PAGE_SIZE;
2611                                 kaddr += PAGE_SIZE;
2612                         }
2613                 }
2614         }
2615
2616         atomic_inc(&po->mapped);
2617         vma->vm_ops = &packet_mmap_ops;
2618         err = 0;
2619
2620 out:
2621         mutex_unlock(&po->pg_vec_lock);
2622         return err;
2623 }
2624
2625 static const struct proto_ops packet_ops_spkt = {
2626         .family =       PF_PACKET,
2627         .owner =        THIS_MODULE,
2628         .release =      packet_release,
2629         .bind =         packet_bind_spkt,
2630         .connect =      sock_no_connect,
2631         .socketpair =   sock_no_socketpair,
2632         .accept =       sock_no_accept,
2633         .getname =      packet_getname_spkt,
2634         .poll =         datagram_poll,
2635         .ioctl =        packet_ioctl,
2636         .listen =       sock_no_listen,
2637         .shutdown =     sock_no_shutdown,
2638         .setsockopt =   sock_no_setsockopt,
2639         .getsockopt =   sock_no_getsockopt,
2640         .sendmsg =      packet_sendmsg_spkt,
2641         .recvmsg =      packet_recvmsg,
2642         .mmap =         sock_no_mmap,
2643         .sendpage =     sock_no_sendpage,
2644 };
2645
2646 static const struct proto_ops packet_ops = {
2647         .family =       PF_PACKET,
2648         .owner =        THIS_MODULE,
2649         .release =      packet_release,
2650         .bind =         packet_bind,
2651         .connect =      sock_no_connect,
2652         .socketpair =   sock_no_socketpair,
2653         .accept =       sock_no_accept,
2654         .getname =      packet_getname,
2655         .poll =         packet_poll,
2656         .ioctl =        packet_ioctl,
2657         .listen =       sock_no_listen,
2658         .shutdown =     sock_no_shutdown,
2659         .setsockopt =   packet_setsockopt,
2660         .getsockopt =   packet_getsockopt,
2661         .sendmsg =      packet_sendmsg,
2662         .recvmsg =      packet_recvmsg,
2663         .mmap =         packet_mmap,
2664         .sendpage =     sock_no_sendpage,
2665 };
2666
2667 static const struct net_proto_family packet_family_ops = {
2668         .family =       PF_PACKET,
2669         .create =       packet_create,
2670         .owner  =       THIS_MODULE,
2671 };
2672
2673 static struct notifier_block packet_netdev_notifier = {
2674         .notifier_call =        packet_notifier,
2675 };
2676
2677 #ifdef CONFIG_PROC_FS
2678
2679 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
2680         __acquires(RCU)
2681 {
2682         struct net *net = seq_file_net(seq);
2683
2684         rcu_read_lock();
2685         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
2686 }
2687
2688 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2689 {
2690         struct net *net = seq_file_net(seq);
2691         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
2692 }
2693
2694 static void packet_seq_stop(struct seq_file *seq, void *v)
2695         __releases(RCU)
2696 {
2697         rcu_read_unlock();
2698 }
2699
2700 static int packet_seq_show(struct seq_file *seq, void *v)
2701 {
2702         if (v == SEQ_START_TOKEN)
2703                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
2704         else {
2705                 struct sock *s = sk_entry(v);
2706                 const struct packet_sock *po = pkt_sk(s);
2707
2708                 seq_printf(seq,
2709                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
2710                            s,
2711                            atomic_read(&s->sk_refcnt),
2712                            s->sk_type,
2713                            ntohs(po->num),
2714                            po->ifindex,
2715                            po->running,
2716                            atomic_read(&s->sk_rmem_alloc),
2717                            sock_i_uid(s),
2718                            sock_i_ino(s));
2719         }
2720
2721         return 0;
2722 }
2723
2724 static const struct seq_operations packet_seq_ops = {
2725         .start  = packet_seq_start,
2726         .next   = packet_seq_next,
2727         .stop   = packet_seq_stop,
2728         .show   = packet_seq_show,
2729 };
2730
2731 static int packet_seq_open(struct inode *inode, struct file *file)
2732 {
2733         return seq_open_net(inode, file, &packet_seq_ops,
2734                             sizeof(struct seq_net_private));
2735 }
2736
2737 static const struct file_operations packet_seq_fops = {
2738         .owner          = THIS_MODULE,
2739         .open           = packet_seq_open,
2740         .read           = seq_read,
2741         .llseek         = seq_lseek,
2742         .release        = seq_release_net,
2743 };
2744
2745 #endif
2746
2747 static int __net_init packet_net_init(struct net *net)
2748 {
2749         spin_lock_init(&net->packet.sklist_lock);
2750         INIT_HLIST_HEAD(&net->packet.sklist);
2751
2752         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
2753                 return -ENOMEM;
2754
2755         return 0;
2756 }
2757
2758 static void __net_exit packet_net_exit(struct net *net)
2759 {
2760         proc_net_remove(net, "packet");
2761 }
2762
2763 static struct pernet_operations packet_net_ops = {
2764         .init = packet_net_init,
2765         .exit = packet_net_exit,
2766 };
2767
2768
2769 static void __exit packet_exit(void)
2770 {
2771         unregister_netdevice_notifier(&packet_netdev_notifier);
2772         unregister_pernet_subsys(&packet_net_ops);
2773         sock_unregister(PF_PACKET);
2774         proto_unregister(&packet_proto);
2775 }
2776
2777 static int __init packet_init(void)
2778 {
2779         int rc = proto_register(&packet_proto, 0);
2780
2781         if (rc != 0)
2782                 goto out;
2783
2784         sock_register(&packet_family_ops);
2785         register_pernet_subsys(&packet_net_ops);
2786         register_netdevice_notifier(&packet_netdev_notifier);
2787 out:
2788         return rc;
2789 }
2790
2791 module_init(packet_init);
2792 module_exit(packet_exit);
2793 MODULE_LICENSE("GPL");
2794 MODULE_ALIAS_NETPROTO(PF_PACKET);