[PACKET]: Don't truncate non-linear skbs with mmaped IO
[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  * Version:     $Id: af_packet.c,v 1.61 2002/02/08 03:57:19 davem Exp $
9  *
10  * Authors:     Ross Biro
11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
13  *
14  * Fixes:       
15  *              Alan Cox        :       verify_area() now used correctly
16  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
17  *              Alan Cox        :       tidied skbuff lists.
18  *              Alan Cox        :       Now uses generic datagram routines I
19  *                                      added. Also fixed the peek/read crash
20  *                                      from all old Linux datagram code.
21  *              Alan Cox        :       Uses the improved datagram code.
22  *              Alan Cox        :       Added NULL's for socket options.
23  *              Alan Cox        :       Re-commented the code.
24  *              Alan Cox        :       Use new kernel side addressing
25  *              Rob Janssen     :       Correct MTU usage.
26  *              Dave Platt      :       Counter leaks caused by incorrect
27  *                                      interrupt locking and some slightly
28  *                                      dubious gcc output. Can you read
29  *                                      compiler: it said _VOLATILE_
30  *      Richard Kooijman        :       Timestamp fixes.
31  *              Alan Cox        :       New buffers. Use sk->mac.raw.
32  *              Alan Cox        :       sendmsg/recvmsg support.
33  *              Alan Cox        :       Protocol setting support
34  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
35  *      Cyrus Durgin            :       Fixed kerneld for kmod.
36  *      Michal Ostrowski        :       Module initialization cleanup.
37  *         Ulises Alonso        :       Frame number limit removal and 
38  *                                      packet_set_ring memory leak.
39  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
40  *                                      The convention is that longer addresses
41  *                                      will simply extend the hardware address
42  *                                      byte arrays at the end of sockaddr_ll 
43  *                                      and packet_mreq.
44  *
45  *              This program is free software; you can redistribute it and/or
46  *              modify it under the terms of the GNU General Public License
47  *              as published by the Free Software Foundation; either version
48  *              2 of the License, or (at your option) any later version.
49  *
50  */
51  
52 #include <linux/types.h>
53 #include <linux/sched.h>
54 #include <linux/mm.h>
55 #include <linux/capability.h>
56 #include <linux/fcntl.h>
57 #include <linux/socket.h>
58 #include <linux/in.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/if_packet.h>
62 #include <linux/wireless.h>
63 #include <linux/kmod.h>
64 #include <net/ip.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/errno.h>
69 #include <linux/timer.h>
70 #include <asm/system.h>
71 #include <asm/uaccess.h>
72 #include <asm/ioctls.h>
73 #include <asm/page.h>
74 #include <asm/io.h>
75 #include <linux/proc_fs.h>
76 #include <linux/seq_file.h>
77 #include <linux/poll.h>
78 #include <linux/module.h>
79 #include <linux/init.h>
80
81 #ifdef CONFIG_INET
82 #include <net/inet_common.h>
83 #endif
84
85 #define CONFIG_SOCK_PACKET      1
86
87 /*
88    Proposed replacement for SIOC{ADD,DEL}MULTI and
89    IFF_PROMISC, IFF_ALLMULTI flags.
90
91    It is more expensive, but I believe,
92    it is really correct solution: reentereble, safe and fault tolerant.
93
94    IFF_PROMISC/IFF_ALLMULTI/SIOC{ADD/DEL}MULTI are faked by keeping
95    reference count and global flag, so that real status is
96    (gflag|(count != 0)), so that we can use obsolete faulty interface
97    not harming clever users.
98  */
99 #define CONFIG_PACKET_MULTICAST 1
100
101 /*
102    Assumptions:
103    - if device has no dev->hard_header routine, it adds and removes ll header
104      inside itself. In this case ll header is invisible outside of device,
105      but higher levels still should reserve dev->hard_header_len.
106      Some devices are enough clever to reallocate skb, when header
107      will not fit to reserved space (tunnel), another ones are silly
108      (PPP).
109    - packet socket receives packets with pulled ll header,
110      so that SOCK_RAW should push it back.
111
112 On receive:
113 -----------
114
115 Incoming, dev->hard_header!=NULL
116    mac.raw -> ll header
117    data    -> data
118
119 Outgoing, dev->hard_header!=NULL
120    mac.raw -> ll header
121    data    -> ll header
122
123 Incoming, dev->hard_header==NULL
124    mac.raw -> UNKNOWN position. It is very likely, that it points to ll header.
125               PPP makes it, that is wrong, because introduce assymetry
126               between rx and tx paths.
127    data    -> data
128
129 Outgoing, dev->hard_header==NULL
130    mac.raw -> data. ll header is still not built!
131    data    -> data
132
133 Resume
134   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
135
136
137 On transmit:
138 ------------
139
140 dev->hard_header != NULL
141    mac.raw -> ll header
142    data    -> ll header
143
144 dev->hard_header == NULL (ll header is added by device, we cannot control it)
145    mac.raw -> data
146    data -> data
147
148    We should set nh.raw on output to correct posistion,
149    packet classifier depends on it.
150  */
151
152 /* List of all packet sockets. */
153 static HLIST_HEAD(packet_sklist);
154 static DEFINE_RWLOCK(packet_sklist_lock);
155
156 static atomic_t packet_socks_nr;
157
158
159 /* Private packet socket structures. */
160
161 #ifdef CONFIG_PACKET_MULTICAST
162 struct packet_mclist
163 {
164         struct packet_mclist    *next;
165         int                     ifindex;
166         int                     count;
167         unsigned short          type;
168         unsigned short          alen;
169         unsigned char           addr[MAX_ADDR_LEN];
170 };
171 /* identical to struct packet_mreq except it has
172  * a longer address field.
173  */
174 struct packet_mreq_max
175 {
176         int             mr_ifindex;
177         unsigned short  mr_type;
178         unsigned short  mr_alen;
179         unsigned char   mr_address[MAX_ADDR_LEN];
180 };
181 #endif
182 #ifdef CONFIG_PACKET_MMAP
183 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing);
184 #endif
185
186 static void packet_flush_mclist(struct sock *sk);
187
188 struct packet_sock {
189         /* struct sock has to be the first member of packet_sock */
190         struct sock             sk;
191         struct tpacket_stats    stats;
192 #ifdef CONFIG_PACKET_MMAP
193         char *                  *pg_vec;
194         unsigned int            head;
195         unsigned int            frames_per_block;
196         unsigned int            frame_size;
197         unsigned int            frame_max;
198         int                     copy_thresh;
199 #endif
200         struct packet_type      prot_hook;
201         spinlock_t              bind_lock;
202         char                    running;        /* prot_hook is attached*/
203         int                     ifindex;        /* bound device         */
204         unsigned short          num;
205 #ifdef CONFIG_PACKET_MULTICAST
206         struct packet_mclist    *mclist;
207 #endif
208 #ifdef CONFIG_PACKET_MMAP
209         atomic_t                mapped;
210         unsigned int            pg_vec_order;
211         unsigned int            pg_vec_pages;
212         unsigned int            pg_vec_len;
213 #endif
214 };
215
216 #ifdef CONFIG_PACKET_MMAP
217
218 static inline char *packet_lookup_frame(struct packet_sock *po, unsigned int position)
219 {
220         unsigned int pg_vec_pos, frame_offset;
221         char *frame;
222
223         pg_vec_pos = position / po->frames_per_block;
224         frame_offset = position % po->frames_per_block;
225
226         frame = po->pg_vec[pg_vec_pos] + (frame_offset * po->frame_size);
227         
228         return frame;
229 }
230 #endif
231
232 static inline struct packet_sock *pkt_sk(struct sock *sk)
233 {
234         return (struct packet_sock *)sk;
235 }
236
237 static void packet_sock_destruct(struct sock *sk)
238 {
239         BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
240         BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
241
242         if (!sock_flag(sk, SOCK_DEAD)) {
243                 printk("Attempt to release alive packet socket: %p\n", sk);
244                 return;
245         }
246
247         atomic_dec(&packet_socks_nr);
248 #ifdef PACKET_REFCNT_DEBUG
249         printk(KERN_DEBUG "PACKET socket %p is free, %d are alive\n", sk, atomic_read(&packet_socks_nr));
250 #endif
251 }
252
253
254 static const struct proto_ops packet_ops;
255
256 #ifdef CONFIG_SOCK_PACKET
257 static const struct proto_ops packet_ops_spkt;
258
259 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,  struct packet_type *pt, struct net_device *orig_dev)
260 {
261         struct sock *sk;
262         struct sockaddr_pkt *spkt;
263
264         /*
265          *      When we registered the protocol we saved the socket in the data
266          *      field for just this event.
267          */
268
269         sk = pt->af_packet_priv;
270         
271         /*
272          *      Yank back the headers [hope the device set this
273          *      right or kerboom...]
274          *
275          *      Incoming packets have ll header pulled,
276          *      push it back.
277          *
278          *      For outgoing ones skb->data == skb->mac.raw
279          *      so that this procedure is noop.
280          */
281
282         if (skb->pkt_type == PACKET_LOOPBACK)
283                 goto out;
284
285         if ((skb = skb_share_check(skb, GFP_ATOMIC)) == NULL)
286                 goto oom;
287
288         /* drop any routing info */
289         dst_release(skb->dst);
290         skb->dst = NULL;
291
292         /* drop conntrack reference */
293         nf_reset(skb);
294
295         spkt = (struct sockaddr_pkt*)skb->cb;
296
297         skb_push(skb, skb->data-skb->mac.raw);
298
299         /*
300          *      The SOCK_PACKET socket receives _all_ frames.
301          */
302
303         spkt->spkt_family = dev->type;
304         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
305         spkt->spkt_protocol = skb->protocol;
306
307         /*
308          *      Charge the memory to the socket. This is done specifically
309          *      to prevent sockets using all the memory up.
310          */
311
312         if (sock_queue_rcv_skb(sk,skb) == 0)
313                 return 0;
314
315 out:
316         kfree_skb(skb);
317 oom:
318         return 0;
319 }
320
321
322 /*
323  *      Output a raw packet to a device layer. This bypasses all the other
324  *      protocol layers and you must therefore supply it with a complete frame
325  */
326  
327 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
328                                struct msghdr *msg, size_t len)
329 {
330         struct sock *sk = sock->sk;
331         struct sockaddr_pkt *saddr=(struct sockaddr_pkt *)msg->msg_name;
332         struct sk_buff *skb;
333         struct net_device *dev;
334         unsigned short proto=0;
335         int err;
336         
337         /*
338          *      Get and verify the address. 
339          */
340
341         if (saddr)
342         {
343                 if (msg->msg_namelen < sizeof(struct sockaddr))
344                         return(-EINVAL);
345                 if (msg->msg_namelen==sizeof(struct sockaddr_pkt))
346                         proto=saddr->spkt_protocol;
347         }
348         else
349                 return(-ENOTCONN);      /* SOCK_PACKET must be sent giving an address */
350
351         /*
352          *      Find the device first to size check it 
353          */
354
355         saddr->spkt_device[13] = 0;
356         dev = dev_get_by_name(saddr->spkt_device);
357         err = -ENODEV;
358         if (dev == NULL)
359                 goto out_unlock;
360         
361         /*
362          *      You may not queue a frame bigger than the mtu. This is the lowest level
363          *      raw protocol and you must do your own fragmentation at this level.
364          */
365          
366         err = -EMSGSIZE;
367         if (len > dev->mtu + dev->hard_header_len)
368                 goto out_unlock;
369
370         err = -ENOBUFS;
371         skb = sock_wmalloc(sk, len + LL_RESERVED_SPACE(dev), 0, GFP_KERNEL);
372
373         /*
374          *      If the write buffer is full, then tough. At this level the user gets to
375          *      deal with the problem - do your own algorithmic backoffs. That's far
376          *      more flexible.
377          */
378          
379         if (skb == NULL) 
380                 goto out_unlock;
381
382         /*
383          *      Fill it in 
384          */
385          
386         /* FIXME: Save some space for broken drivers that write a
387          * hard header at transmission time by themselves. PPP is the
388          * notable one here. This should really be fixed at the driver level.
389          */
390         skb_reserve(skb, LL_RESERVED_SPACE(dev));
391         skb->nh.raw = skb->data;
392
393         /* Try to align data part correctly */
394         if (dev->hard_header) {
395                 skb->data -= dev->hard_header_len;
396                 skb->tail -= dev->hard_header_len;
397                 if (len < dev->hard_header_len)
398                         skb->nh.raw = skb->data;
399         }
400
401         /* Returns -EFAULT on error */
402         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
403         skb->protocol = proto;
404         skb->dev = dev;
405         skb->priority = sk->sk_priority;
406         if (err)
407                 goto out_free;
408
409         err = -ENETDOWN;
410         if (!(dev->flags & IFF_UP))
411                 goto out_free;
412
413         /*
414          *      Now send it
415          */
416
417         dev_queue_xmit(skb);
418         dev_put(dev);
419         return(len);
420
421 out_free:
422         kfree_skb(skb);
423 out_unlock:
424         if (dev)
425                 dev_put(dev);
426         return err;
427 }
428 #endif
429
430 static inline unsigned run_filter(struct sk_buff *skb, struct sock *sk, unsigned res)
431 {
432         struct sk_filter *filter;
433
434         bh_lock_sock(sk);
435         filter = sk->sk_filter;
436         /*
437          * Our caller already checked that filter != NULL but we need to
438          * verify that under bh_lock_sock() to be safe
439          */
440         if (likely(filter != NULL))
441                 res = sk_run_filter(skb, filter->insns, filter->len);
442         bh_unlock_sock(sk);
443
444         return res;
445 }
446
447 /*
448    This function makes lazy skb cloning in hope that most of packets
449    are discarded by BPF.
450
451    Note tricky part: we DO mangle shared skb! skb->data, skb->len
452    and skb->cb are mangled. It works because (and until) packets
453    falling here are owned by current CPU. Output packets are cloned
454    by dev_queue_xmit_nit(), input packets are processed by net_bh
455    sequencially, so that if we return skb to original state on exit,
456    we will not harm anyone.
457  */
458
459 static int packet_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
460 {
461         struct sock *sk;
462         struct sockaddr_ll *sll;
463         struct packet_sock *po;
464         u8 * skb_head = skb->data;
465         int skb_len = skb->len;
466         unsigned snaplen;
467
468         if (skb->pkt_type == PACKET_LOOPBACK)
469                 goto drop;
470
471         sk = pt->af_packet_priv;
472         po = pkt_sk(sk);
473
474         skb->dev = dev;
475
476         if (dev->hard_header) {
477                 /* The device has an explicit notion of ll header,
478                    exported to higher levels.
479
480                    Otherwise, the device hides datails of it frame
481                    structure, so that corresponding packet head
482                    never delivered to user.
483                  */
484                 if (sk->sk_type != SOCK_DGRAM)
485                         skb_push(skb, skb->data - skb->mac.raw);
486                 else if (skb->pkt_type == PACKET_OUTGOING) {
487                         /* Special case: outgoing packets have ll header at head */
488                         skb_pull(skb, skb->nh.raw - skb->data);
489                 }
490         }
491
492         snaplen = skb->len;
493
494         if (sk->sk_filter) {
495                 unsigned res = run_filter(skb, sk, snaplen);
496                 if (res == 0)
497                         goto drop_n_restore;
498                 if (snaplen > res)
499                         snaplen = res;
500         }
501
502         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
503             (unsigned)sk->sk_rcvbuf)
504                 goto drop_n_acct;
505
506         if (skb_shared(skb)) {
507                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
508                 if (nskb == NULL)
509                         goto drop_n_acct;
510
511                 if (skb_head != skb->data) {
512                         skb->data = skb_head;
513                         skb->len = skb_len;
514                 }
515                 kfree_skb(skb);
516                 skb = nskb;
517         }
518
519         sll = (struct sockaddr_ll*)skb->cb;
520         sll->sll_family = AF_PACKET;
521         sll->sll_hatype = dev->type;
522         sll->sll_protocol = skb->protocol;
523         sll->sll_pkttype = skb->pkt_type;
524         sll->sll_ifindex = dev->ifindex;
525         sll->sll_halen = 0;
526
527         if (dev->hard_header_parse)
528                 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
529
530         if (pskb_trim(skb, snaplen))
531                 goto drop_n_acct;
532
533         skb_set_owner_r(skb, sk);
534         skb->dev = NULL;
535         dst_release(skb->dst);
536         skb->dst = NULL;
537
538         /* drop conntrack reference */
539         nf_reset(skb);
540
541         spin_lock(&sk->sk_receive_queue.lock);
542         po->stats.tp_packets++;
543         __skb_queue_tail(&sk->sk_receive_queue, skb);
544         spin_unlock(&sk->sk_receive_queue.lock);
545         sk->sk_data_ready(sk, skb->len);
546         return 0;
547
548 drop_n_acct:
549         spin_lock(&sk->sk_receive_queue.lock);
550         po->stats.tp_drops++;
551         spin_unlock(&sk->sk_receive_queue.lock);
552
553 drop_n_restore:
554         if (skb_head != skb->data && skb_shared(skb)) {
555                 skb->data = skb_head;
556                 skb->len = skb_len;
557         }
558 drop:
559         kfree_skb(skb);
560         return 0;
561 }
562
563 #ifdef CONFIG_PACKET_MMAP
564 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
565 {
566         struct sock *sk;
567         struct packet_sock *po;
568         struct sockaddr_ll *sll;
569         struct tpacket_hdr *h;
570         u8 * skb_head = skb->data;
571         int skb_len = skb->len;
572         unsigned snaplen;
573         unsigned long status = TP_STATUS_LOSING|TP_STATUS_USER;
574         unsigned short macoff, netoff;
575         struct sk_buff *copy_skb = NULL;
576
577         if (skb->pkt_type == PACKET_LOOPBACK)
578                 goto drop;
579
580         sk = pt->af_packet_priv;
581         po = pkt_sk(sk);
582
583         if (dev->hard_header) {
584                 if (sk->sk_type != SOCK_DGRAM)
585                         skb_push(skb, skb->data - skb->mac.raw);
586                 else if (skb->pkt_type == PACKET_OUTGOING) {
587                         /* Special case: outgoing packets have ll header at head */
588                         skb_pull(skb, skb->nh.raw - skb->data);
589                         if (skb->ip_summed == CHECKSUM_HW)
590                                 status |= TP_STATUS_CSUMNOTREADY;
591                 }
592         }
593
594         snaplen = skb->len;
595
596         if (sk->sk_filter) {
597                 unsigned res = run_filter(skb, sk, snaplen);
598                 if (res == 0)
599                         goto drop_n_restore;
600                 if (snaplen > res)
601                         snaplen = res;
602         }
603
604         if (sk->sk_type == SOCK_DGRAM) {
605                 macoff = netoff = TPACKET_ALIGN(TPACKET_HDRLEN) + 16;
606         } else {
607                 unsigned maclen = skb->nh.raw - skb->data;
608                 netoff = TPACKET_ALIGN(TPACKET_HDRLEN + (maclen < 16 ? 16 : maclen));
609                 macoff = netoff - maclen;
610         }
611
612         if (macoff + snaplen > po->frame_size) {
613                 if (po->copy_thresh &&
614                     atomic_read(&sk->sk_rmem_alloc) + skb->truesize <
615                     (unsigned)sk->sk_rcvbuf) {
616                         if (skb_shared(skb)) {
617                                 copy_skb = skb_clone(skb, GFP_ATOMIC);
618                         } else {
619                                 copy_skb = skb_get(skb);
620                                 skb_head = skb->data;
621                         }
622                         if (copy_skb)
623                                 skb_set_owner_r(copy_skb, sk);
624                 }
625                 snaplen = po->frame_size - macoff;
626                 if ((int)snaplen < 0)
627                         snaplen = 0;
628         }
629
630         spin_lock(&sk->sk_receive_queue.lock);
631         h = (struct tpacket_hdr *)packet_lookup_frame(po, po->head);
632         
633         if (h->tp_status)
634                 goto ring_is_full;
635         po->head = po->head != po->frame_max ? po->head+1 : 0;
636         po->stats.tp_packets++;
637         if (copy_skb) {
638                 status |= TP_STATUS_COPY;
639                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
640         }
641         if (!po->stats.tp_drops)
642                 status &= ~TP_STATUS_LOSING;
643         spin_unlock(&sk->sk_receive_queue.lock);
644
645         skb_copy_bits(skb, 0, (u8*)h + macoff, snaplen);
646
647         h->tp_len = skb->len;
648         h->tp_snaplen = snaplen;
649         h->tp_mac = macoff;
650         h->tp_net = netoff;
651         if (skb->tstamp.off_sec == 0) { 
652                 __net_timestamp(skb);
653                 sock_enable_timestamp(sk);
654         }
655         h->tp_sec = skb->tstamp.off_sec;
656         h->tp_usec = skb->tstamp.off_usec;
657
658         sll = (struct sockaddr_ll*)((u8*)h + TPACKET_ALIGN(sizeof(*h)));
659         sll->sll_halen = 0;
660         if (dev->hard_header_parse)
661                 sll->sll_halen = dev->hard_header_parse(skb, sll->sll_addr);
662         sll->sll_family = AF_PACKET;
663         sll->sll_hatype = dev->type;
664         sll->sll_protocol = skb->protocol;
665         sll->sll_pkttype = skb->pkt_type;
666         sll->sll_ifindex = dev->ifindex;
667
668         h->tp_status = status;
669         mb();
670
671         {
672                 struct page *p_start, *p_end;
673                 u8 *h_end = (u8 *)h + macoff + snaplen - 1;
674
675                 p_start = virt_to_page(h);
676                 p_end = virt_to_page(h_end);
677                 while (p_start <= p_end) {
678                         flush_dcache_page(p_start);
679                         p_start++;
680                 }
681         }
682
683         sk->sk_data_ready(sk, 0);
684
685 drop_n_restore:
686         if (skb_head != skb->data && skb_shared(skb)) {
687                 skb->data = skb_head;
688                 skb->len = skb_len;
689         }
690 drop:
691         kfree_skb(skb);
692         return 0;
693
694 ring_is_full:
695         po->stats.tp_drops++;
696         spin_unlock(&sk->sk_receive_queue.lock);
697
698         sk->sk_data_ready(sk, 0);
699         if (copy_skb)
700                 kfree_skb(copy_skb);
701         goto drop_n_restore;
702 }
703
704 #endif
705
706
707 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
708                           struct msghdr *msg, size_t len)
709 {
710         struct sock *sk = sock->sk;
711         struct sockaddr_ll *saddr=(struct sockaddr_ll *)msg->msg_name;
712         struct sk_buff *skb;
713         struct net_device *dev;
714         unsigned short proto;
715         unsigned char *addr;
716         int ifindex, err, reserve = 0;
717
718         /*
719          *      Get and verify the address. 
720          */
721          
722         if (saddr == NULL) {
723                 struct packet_sock *po = pkt_sk(sk);
724
725                 ifindex = po->ifindex;
726                 proto   = po->num;
727                 addr    = NULL;
728         } else {
729                 err = -EINVAL;
730                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
731                         goto out;
732                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
733                         goto out;
734                 ifindex = saddr->sll_ifindex;
735                 proto   = saddr->sll_protocol;
736                 addr    = saddr->sll_addr;
737         }
738
739
740         dev = dev_get_by_index(ifindex);
741         err = -ENXIO;
742         if (dev == NULL)
743                 goto out_unlock;
744         if (sock->type == SOCK_RAW)
745                 reserve = dev->hard_header_len;
746
747         err = -EMSGSIZE;
748         if (len > dev->mtu+reserve)
749                 goto out_unlock;
750
751         skb = sock_alloc_send_skb(sk, len + LL_RESERVED_SPACE(dev),
752                                 msg->msg_flags & MSG_DONTWAIT, &err);
753         if (skb==NULL)
754                 goto out_unlock;
755
756         skb_reserve(skb, LL_RESERVED_SPACE(dev));
757         skb->nh.raw = skb->data;
758
759         if (dev->hard_header) {
760                 int res;
761                 err = -EINVAL;
762                 res = dev->hard_header(skb, dev, ntohs(proto), addr, NULL, len);
763                 if (sock->type != SOCK_DGRAM) {
764                         skb->tail = skb->data;
765                         skb->len = 0;
766                 } else if (res < 0)
767                         goto out_free;
768         }
769
770         /* Returns -EFAULT on error */
771         err = memcpy_fromiovec(skb_put(skb,len), msg->msg_iov, len);
772         if (err)
773                 goto out_free;
774
775         skb->protocol = proto;
776         skb->dev = dev;
777         skb->priority = sk->sk_priority;
778
779         err = -ENETDOWN;
780         if (!(dev->flags & IFF_UP))
781                 goto out_free;
782
783         /*
784          *      Now send it
785          */
786
787         err = dev_queue_xmit(skb);
788         if (err > 0 && (err = net_xmit_errno(err)) != 0)
789                 goto out_unlock;
790
791         dev_put(dev);
792
793         return(len);
794
795 out_free:
796         kfree_skb(skb);
797 out_unlock:
798         if (dev)
799                 dev_put(dev);
800 out:
801         return err;
802 }
803
804 /*
805  *      Close a PACKET socket. This is fairly simple. We immediately go
806  *      to 'closed' state and remove our protocol entry in the device list.
807  */
808
809 static int packet_release(struct socket *sock)
810 {
811         struct sock *sk = sock->sk;
812         struct packet_sock *po;
813
814         if (!sk)
815                 return 0;
816
817         po = pkt_sk(sk);
818
819         write_lock_bh(&packet_sklist_lock);
820         sk_del_node_init(sk);
821         write_unlock_bh(&packet_sklist_lock);
822
823         /*
824          *      Unhook packet receive handler.
825          */
826
827         if (po->running) {
828                 /*
829                  *      Remove the protocol hook
830                  */
831                 dev_remove_pack(&po->prot_hook);
832                 po->running = 0;
833                 po->num = 0;
834                 __sock_put(sk);
835         }
836
837 #ifdef CONFIG_PACKET_MULTICAST
838         packet_flush_mclist(sk);
839 #endif
840
841 #ifdef CONFIG_PACKET_MMAP
842         if (po->pg_vec) {
843                 struct tpacket_req req;
844                 memset(&req, 0, sizeof(req));
845                 packet_set_ring(sk, &req, 1);
846         }
847 #endif
848
849         /*
850          *      Now the socket is dead. No more input will appear.
851          */
852
853         sock_orphan(sk);
854         sock->sk = NULL;
855
856         /* Purge queues */
857
858         skb_queue_purge(&sk->sk_receive_queue);
859
860         sock_put(sk);
861         return 0;
862 }
863
864 /*
865  *      Attach a packet hook.
866  */
867
868 static int packet_do_bind(struct sock *sk, struct net_device *dev, int protocol)
869 {
870         struct packet_sock *po = pkt_sk(sk);
871         /*
872          *      Detach an existing hook if present.
873          */
874
875         lock_sock(sk);
876
877         spin_lock(&po->bind_lock);
878         if (po->running) {
879                 __sock_put(sk);
880                 po->running = 0;
881                 po->num = 0;
882                 spin_unlock(&po->bind_lock);
883                 dev_remove_pack(&po->prot_hook);
884                 spin_lock(&po->bind_lock);
885         }
886
887         po->num = protocol;
888         po->prot_hook.type = protocol;
889         po->prot_hook.dev = dev;
890
891         po->ifindex = dev ? dev->ifindex : 0;
892
893         if (protocol == 0)
894                 goto out_unlock;
895
896         if (dev) {
897                 if (dev->flags&IFF_UP) {
898                         dev_add_pack(&po->prot_hook);
899                         sock_hold(sk);
900                         po->running = 1;
901                 } else {
902                         sk->sk_err = ENETDOWN;
903                         if (!sock_flag(sk, SOCK_DEAD))
904                                 sk->sk_error_report(sk);
905                 }
906         } else {
907                 dev_add_pack(&po->prot_hook);
908                 sock_hold(sk);
909                 po->running = 1;
910         }
911
912 out_unlock:
913         spin_unlock(&po->bind_lock);
914         release_sock(sk);
915         return 0;
916 }
917
918 /*
919  *      Bind a packet socket to a device
920  */
921
922 #ifdef CONFIG_SOCK_PACKET
923
924 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr, int addr_len)
925 {
926         struct sock *sk=sock->sk;
927         char name[15];
928         struct net_device *dev;
929         int err = -ENODEV;
930         
931         /*
932          *      Check legality
933          */
934          
935         if (addr_len != sizeof(struct sockaddr))
936                 return -EINVAL;
937         strlcpy(name,uaddr->sa_data,sizeof(name));
938
939         dev = dev_get_by_name(name);
940         if (dev) {
941                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
942                 dev_put(dev);
943         }
944         return err;
945 }
946 #endif
947
948 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
949 {
950         struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
951         struct sock *sk=sock->sk;
952         struct net_device *dev = NULL;
953         int err;
954
955
956         /*
957          *      Check legality
958          */
959          
960         if (addr_len < sizeof(struct sockaddr_ll))
961                 return -EINVAL;
962         if (sll->sll_family != AF_PACKET)
963                 return -EINVAL;
964
965         if (sll->sll_ifindex) {
966                 err = -ENODEV;
967                 dev = dev_get_by_index(sll->sll_ifindex);
968                 if (dev == NULL)
969                         goto out;
970         }
971         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
972         if (dev)
973                 dev_put(dev);
974
975 out:
976         return err;
977 }
978
979 static struct proto packet_proto = {
980         .name     = "PACKET",
981         .owner    = THIS_MODULE,
982         .obj_size = sizeof(struct packet_sock),
983 };
984
985 /*
986  *      Create a packet of type SOCK_PACKET. 
987  */
988
989 static int packet_create(struct socket *sock, int protocol)
990 {
991         struct sock *sk;
992         struct packet_sock *po;
993         int err;
994
995         if (!capable(CAP_NET_RAW))
996                 return -EPERM;
997         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW
998 #ifdef CONFIG_SOCK_PACKET
999             && sock->type != SOCK_PACKET
1000 #endif
1001             )
1002                 return -ESOCKTNOSUPPORT;
1003
1004         sock->state = SS_UNCONNECTED;
1005
1006         err = -ENOBUFS;
1007         sk = sk_alloc(PF_PACKET, GFP_KERNEL, &packet_proto, 1);
1008         if (sk == NULL)
1009                 goto out;
1010
1011         sock->ops = &packet_ops;
1012 #ifdef CONFIG_SOCK_PACKET
1013         if (sock->type == SOCK_PACKET)
1014                 sock->ops = &packet_ops_spkt;
1015 #endif
1016         sock_init_data(sock, sk);
1017
1018         po = pkt_sk(sk);
1019         sk->sk_family = PF_PACKET;
1020         po->num = protocol;
1021
1022         sk->sk_destruct = packet_sock_destruct;
1023         atomic_inc(&packet_socks_nr);
1024
1025         /*
1026          *      Attach a protocol block
1027          */
1028
1029         spin_lock_init(&po->bind_lock);
1030         po->prot_hook.func = packet_rcv;
1031 #ifdef CONFIG_SOCK_PACKET
1032         if (sock->type == SOCK_PACKET)
1033                 po->prot_hook.func = packet_rcv_spkt;
1034 #endif
1035         po->prot_hook.af_packet_priv = sk;
1036
1037         if (protocol) {
1038                 po->prot_hook.type = protocol;
1039                 dev_add_pack(&po->prot_hook);
1040                 sock_hold(sk);
1041                 po->running = 1;
1042         }
1043
1044         write_lock_bh(&packet_sklist_lock);
1045         sk_add_node(sk, &packet_sklist);
1046         write_unlock_bh(&packet_sklist_lock);
1047         return(0);
1048 out:
1049         return err;
1050 }
1051
1052 /*
1053  *      Pull a packet from our receive queue and hand it to the user.
1054  *      If necessary we block.
1055  */
1056
1057 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
1058                           struct msghdr *msg, size_t len, int flags)
1059 {
1060         struct sock *sk = sock->sk;
1061         struct sk_buff *skb;
1062         int copied, err;
1063         struct sockaddr_ll *sll;
1064
1065         err = -EINVAL;
1066         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT))
1067                 goto out;
1068
1069 #if 0
1070         /* What error should we return now? EUNATTACH? */
1071         if (pkt_sk(sk)->ifindex < 0)
1072                 return -ENODEV;
1073 #endif
1074
1075         /*
1076          *      Call the generic datagram receiver. This handles all sorts
1077          *      of horrible races and re-entrancy so we can forget about it
1078          *      in the protocol layers.
1079          *
1080          *      Now it will return ENETDOWN, if device have just gone down,
1081          *      but then it will block.
1082          */
1083
1084         skb=skb_recv_datagram(sk,flags,flags&MSG_DONTWAIT,&err);
1085
1086         /*
1087          *      An error occurred so return it. Because skb_recv_datagram() 
1088          *      handles the blocking we don't see and worry about blocking
1089          *      retries.
1090          */
1091
1092         if (skb == NULL)
1093                 goto out;
1094
1095         /*
1096          *      If the address length field is there to be filled in, we fill
1097          *      it in now.
1098          */
1099
1100         sll = (struct sockaddr_ll*)skb->cb;
1101         if (sock->type == SOCK_PACKET)
1102                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
1103         else
1104                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
1105
1106         /*
1107          *      You lose any data beyond the buffer you gave. If it worries a
1108          *      user program they can ask the device for its MTU anyway.
1109          */
1110
1111         copied = skb->len;
1112         if (copied > len)
1113         {
1114                 copied=len;
1115                 msg->msg_flags|=MSG_TRUNC;
1116         }
1117
1118         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
1119         if (err)
1120                 goto out_free;
1121
1122         sock_recv_timestamp(msg, sk, skb);
1123
1124         if (msg->msg_name)
1125                 memcpy(msg->msg_name, skb->cb, msg->msg_namelen);
1126
1127         /*
1128          *      Free or return the buffer as appropriate. Again this
1129          *      hides all the races and re-entrancy issues from us.
1130          */
1131         err = (flags&MSG_TRUNC) ? skb->len : copied;
1132
1133 out_free:
1134         skb_free_datagram(sk, skb);
1135 out:
1136         return err;
1137 }
1138
1139 #ifdef CONFIG_SOCK_PACKET
1140 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
1141                                int *uaddr_len, int peer)
1142 {
1143         struct net_device *dev;
1144         struct sock *sk = sock->sk;
1145
1146         if (peer)
1147                 return -EOPNOTSUPP;
1148
1149         uaddr->sa_family = AF_PACKET;
1150         dev = dev_get_by_index(pkt_sk(sk)->ifindex);
1151         if (dev) {
1152                 strlcpy(uaddr->sa_data, dev->name, 15);
1153                 dev_put(dev);
1154         } else
1155                 memset(uaddr->sa_data, 0, 14);
1156         *uaddr_len = sizeof(*uaddr);
1157
1158         return 0;
1159 }
1160 #endif
1161
1162 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
1163                           int *uaddr_len, int peer)
1164 {
1165         struct net_device *dev;
1166         struct sock *sk = sock->sk;
1167         struct packet_sock *po = pkt_sk(sk);
1168         struct sockaddr_ll *sll = (struct sockaddr_ll*)uaddr;
1169
1170         if (peer)
1171                 return -EOPNOTSUPP;
1172
1173         sll->sll_family = AF_PACKET;
1174         sll->sll_ifindex = po->ifindex;
1175         sll->sll_protocol = po->num;
1176         dev = dev_get_by_index(po->ifindex);
1177         if (dev) {
1178                 sll->sll_hatype = dev->type;
1179                 sll->sll_halen = dev->addr_len;
1180                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
1181                 dev_put(dev);
1182         } else {
1183                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
1184                 sll->sll_halen = 0;
1185         }
1186         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
1187
1188         return 0;
1189 }
1190
1191 #ifdef CONFIG_PACKET_MULTICAST
1192 static void packet_dev_mc(struct net_device *dev, struct packet_mclist *i, int what)
1193 {
1194         switch (i->type) {
1195         case PACKET_MR_MULTICAST:
1196                 if (what > 0)
1197                         dev_mc_add(dev, i->addr, i->alen, 0);
1198                 else
1199                         dev_mc_delete(dev, i->addr, i->alen, 0);
1200                 break;
1201         case PACKET_MR_PROMISC:
1202                 dev_set_promiscuity(dev, what);
1203                 break;
1204         case PACKET_MR_ALLMULTI:
1205                 dev_set_allmulti(dev, what);
1206                 break;
1207         default:;
1208         }
1209 }
1210
1211 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
1212 {
1213         for ( ; i; i=i->next) {
1214                 if (i->ifindex == dev->ifindex)
1215                         packet_dev_mc(dev, i, what);
1216         }
1217 }
1218
1219 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
1220 {
1221         struct packet_sock *po = pkt_sk(sk);
1222         struct packet_mclist *ml, *i;
1223         struct net_device *dev;
1224         int err;
1225
1226         rtnl_lock();
1227
1228         err = -ENODEV;
1229         dev = __dev_get_by_index(mreq->mr_ifindex);
1230         if (!dev)
1231                 goto done;
1232
1233         err = -EINVAL;
1234         if (mreq->mr_alen > dev->addr_len)
1235                 goto done;
1236
1237         err = -ENOBUFS;
1238         i = kmalloc(sizeof(*i), GFP_KERNEL);
1239         if (i == NULL)
1240                 goto done;
1241
1242         err = 0;
1243         for (ml = po->mclist; ml; ml = ml->next) {
1244                 if (ml->ifindex == mreq->mr_ifindex &&
1245                     ml->type == mreq->mr_type &&
1246                     ml->alen == mreq->mr_alen &&
1247                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1248                         ml->count++;
1249                         /* Free the new element ... */
1250                         kfree(i);
1251                         goto done;
1252                 }
1253         }
1254
1255         i->type = mreq->mr_type;
1256         i->ifindex = mreq->mr_ifindex;
1257         i->alen = mreq->mr_alen;
1258         memcpy(i->addr, mreq->mr_address, i->alen);
1259         i->count = 1;
1260         i->next = po->mclist;
1261         po->mclist = i;
1262         packet_dev_mc(dev, i, +1);
1263
1264 done:
1265         rtnl_unlock();
1266         return err;
1267 }
1268
1269 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
1270 {
1271         struct packet_mclist *ml, **mlp;
1272
1273         rtnl_lock();
1274
1275         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
1276                 if (ml->ifindex == mreq->mr_ifindex &&
1277                     ml->type == mreq->mr_type &&
1278                     ml->alen == mreq->mr_alen &&
1279                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
1280                         if (--ml->count == 0) {
1281                                 struct net_device *dev;
1282                                 *mlp = ml->next;
1283                                 dev = dev_get_by_index(ml->ifindex);
1284                                 if (dev) {
1285                                         packet_dev_mc(dev, ml, -1);
1286                                         dev_put(dev);
1287                                 }
1288                                 kfree(ml);
1289                         }
1290                         rtnl_unlock();
1291                         return 0;
1292                 }
1293         }
1294         rtnl_unlock();
1295         return -EADDRNOTAVAIL;
1296 }
1297
1298 static void packet_flush_mclist(struct sock *sk)
1299 {
1300         struct packet_sock *po = pkt_sk(sk);
1301         struct packet_mclist *ml;
1302
1303         if (!po->mclist)
1304                 return;
1305
1306         rtnl_lock();
1307         while ((ml = po->mclist) != NULL) {
1308                 struct net_device *dev;
1309
1310                 po->mclist = ml->next;
1311                 if ((dev = dev_get_by_index(ml->ifindex)) != NULL) {
1312                         packet_dev_mc(dev, ml, -1);
1313                         dev_put(dev);
1314                 }
1315                 kfree(ml);
1316         }
1317         rtnl_unlock();
1318 }
1319 #endif
1320
1321 static int
1322 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, int optlen)
1323 {
1324         struct sock *sk = sock->sk;
1325         int ret;
1326
1327         if (level != SOL_PACKET)
1328                 return -ENOPROTOOPT;
1329
1330         switch(optname) {
1331 #ifdef CONFIG_PACKET_MULTICAST
1332         case PACKET_ADD_MEMBERSHIP:     
1333         case PACKET_DROP_MEMBERSHIP:
1334         {
1335                 struct packet_mreq_max mreq;
1336                 int len = optlen;
1337                 memset(&mreq, 0, sizeof(mreq));
1338                 if (len < sizeof(struct packet_mreq))
1339                         return -EINVAL;
1340                 if (len > sizeof(mreq))
1341                         len = sizeof(mreq);
1342                 if (copy_from_user(&mreq,optval,len))
1343                         return -EFAULT;
1344                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
1345                         return -EINVAL;
1346                 if (optname == PACKET_ADD_MEMBERSHIP)
1347                         ret = packet_mc_add(sk, &mreq);
1348                 else
1349                         ret = packet_mc_drop(sk, &mreq);
1350                 return ret;
1351         }
1352 #endif
1353 #ifdef CONFIG_PACKET_MMAP
1354         case PACKET_RX_RING:
1355         {
1356                 struct tpacket_req req;
1357
1358                 if (optlen<sizeof(req))
1359                         return -EINVAL;
1360                 if (copy_from_user(&req,optval,sizeof(req)))
1361                         return -EFAULT;
1362                 return packet_set_ring(sk, &req, 0);
1363         }
1364         case PACKET_COPY_THRESH:
1365         {
1366                 int val;
1367
1368                 if (optlen!=sizeof(val))
1369                         return -EINVAL;
1370                 if (copy_from_user(&val,optval,sizeof(val)))
1371                         return -EFAULT;
1372
1373                 pkt_sk(sk)->copy_thresh = val;
1374                 return 0;
1375         }
1376 #endif
1377         default:
1378                 return -ENOPROTOOPT;
1379         }
1380 }
1381
1382 static int packet_getsockopt(struct socket *sock, int level, int optname,
1383                              char __user *optval, int __user *optlen)
1384 {
1385         int len;
1386         struct sock *sk = sock->sk;
1387         struct packet_sock *po = pkt_sk(sk);
1388
1389         if (level != SOL_PACKET)
1390                 return -ENOPROTOOPT;
1391
1392         if (get_user(len, optlen))
1393                 return -EFAULT;
1394
1395         if (len < 0)
1396                 return -EINVAL;
1397                 
1398         switch(optname) {
1399         case PACKET_STATISTICS:
1400         {
1401                 struct tpacket_stats st;
1402
1403                 if (len > sizeof(struct tpacket_stats))
1404                         len = sizeof(struct tpacket_stats);
1405                 spin_lock_bh(&sk->sk_receive_queue.lock);
1406                 st = po->stats;
1407                 memset(&po->stats, 0, sizeof(st));
1408                 spin_unlock_bh(&sk->sk_receive_queue.lock);
1409                 st.tp_packets += st.tp_drops;
1410
1411                 if (copy_to_user(optval, &st, len))
1412                         return -EFAULT;
1413                 break;
1414         }
1415         default:
1416                 return -ENOPROTOOPT;
1417         }
1418
1419         if (put_user(len, optlen))
1420                 return -EFAULT;
1421         return 0;
1422 }
1423
1424
1425 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
1426 {
1427         struct sock *sk;
1428         struct hlist_node *node;
1429         struct net_device *dev = (struct net_device*)data;
1430
1431         read_lock(&packet_sklist_lock);
1432         sk_for_each(sk, node, &packet_sklist) {
1433                 struct packet_sock *po = pkt_sk(sk);
1434
1435                 switch (msg) {
1436                 case NETDEV_UNREGISTER:
1437 #ifdef CONFIG_PACKET_MULTICAST
1438                         if (po->mclist)
1439                                 packet_dev_mclist(dev, po->mclist, -1);
1440                         // fallthrough
1441 #endif
1442                 case NETDEV_DOWN:
1443                         if (dev->ifindex == po->ifindex) {
1444                                 spin_lock(&po->bind_lock);
1445                                 if (po->running) {
1446                                         __dev_remove_pack(&po->prot_hook);
1447                                         __sock_put(sk);
1448                                         po->running = 0;
1449                                         sk->sk_err = ENETDOWN;
1450                                         if (!sock_flag(sk, SOCK_DEAD))
1451                                                 sk->sk_error_report(sk);
1452                                 }
1453                                 if (msg == NETDEV_UNREGISTER) {
1454                                         po->ifindex = -1;
1455                                         po->prot_hook.dev = NULL;
1456                                 }
1457                                 spin_unlock(&po->bind_lock);
1458                         }
1459                         break;
1460                 case NETDEV_UP:
1461                         spin_lock(&po->bind_lock);
1462                         if (dev->ifindex == po->ifindex && po->num &&
1463                             !po->running) {
1464                                 dev_add_pack(&po->prot_hook);
1465                                 sock_hold(sk);
1466                                 po->running = 1;
1467                         }
1468                         spin_unlock(&po->bind_lock);
1469                         break;
1470                 }
1471         }
1472         read_unlock(&packet_sklist_lock);
1473         return NOTIFY_DONE;
1474 }
1475
1476
1477 static int packet_ioctl(struct socket *sock, unsigned int cmd,
1478                         unsigned long arg)
1479 {
1480         struct sock *sk = sock->sk;
1481
1482         switch(cmd) {
1483                 case SIOCOUTQ:
1484                 {
1485                         int amount = atomic_read(&sk->sk_wmem_alloc);
1486                         return put_user(amount, (int __user *)arg);
1487                 }
1488                 case SIOCINQ:
1489                 {
1490                         struct sk_buff *skb;
1491                         int amount = 0;
1492
1493                         spin_lock_bh(&sk->sk_receive_queue.lock);
1494                         skb = skb_peek(&sk->sk_receive_queue);
1495                         if (skb)
1496                                 amount = skb->len;
1497                         spin_unlock_bh(&sk->sk_receive_queue.lock);
1498                         return put_user(amount, (int __user *)arg);
1499                 }
1500                 case SIOCGSTAMP:
1501                         return sock_get_timestamp(sk, (struct timeval __user *)arg);
1502                         
1503 #ifdef CONFIG_INET
1504                 case SIOCADDRT:
1505                 case SIOCDELRT:
1506                 case SIOCDARP:
1507                 case SIOCGARP:
1508                 case SIOCSARP:
1509                 case SIOCGIFADDR:
1510                 case SIOCSIFADDR:
1511                 case SIOCGIFBRDADDR:
1512                 case SIOCSIFBRDADDR:
1513                 case SIOCGIFNETMASK:
1514                 case SIOCSIFNETMASK:
1515                 case SIOCGIFDSTADDR:
1516                 case SIOCSIFDSTADDR:
1517                 case SIOCSIFFLAGS:
1518                         return inet_dgram_ops.ioctl(sock, cmd, arg);
1519 #endif
1520
1521                 default:
1522                         return -ENOIOCTLCMD;
1523         }
1524         return 0;
1525 }
1526
1527 #ifndef CONFIG_PACKET_MMAP
1528 #define packet_mmap sock_no_mmap
1529 #define packet_poll datagram_poll
1530 #else
1531
1532 static unsigned int packet_poll(struct file * file, struct socket *sock,
1533                                 poll_table *wait)
1534 {
1535         struct sock *sk = sock->sk;
1536         struct packet_sock *po = pkt_sk(sk);
1537         unsigned int mask = datagram_poll(file, sock, wait);
1538
1539         spin_lock_bh(&sk->sk_receive_queue.lock);
1540         if (po->pg_vec) {
1541                 unsigned last = po->head ? po->head-1 : po->frame_max;
1542                 struct tpacket_hdr *h;
1543
1544                 h = (struct tpacket_hdr *)packet_lookup_frame(po, last);
1545
1546                 if (h->tp_status)
1547                         mask |= POLLIN | POLLRDNORM;
1548         }
1549         spin_unlock_bh(&sk->sk_receive_queue.lock);
1550         return mask;
1551 }
1552
1553
1554 /* Dirty? Well, I still did not learn better way to account
1555  * for user mmaps.
1556  */
1557
1558 static void packet_mm_open(struct vm_area_struct *vma)
1559 {
1560         struct file *file = vma->vm_file;
1561         struct socket * sock = file->private_data;
1562         struct sock *sk = sock->sk;
1563         
1564         if (sk)
1565                 atomic_inc(&pkt_sk(sk)->mapped);
1566 }
1567
1568 static void packet_mm_close(struct vm_area_struct *vma)
1569 {
1570         struct file *file = vma->vm_file;
1571         struct socket * sock = file->private_data;
1572         struct sock *sk = sock->sk;
1573         
1574         if (sk)
1575                 atomic_dec(&pkt_sk(sk)->mapped);
1576 }
1577
1578 static struct vm_operations_struct packet_mmap_ops = {
1579         .open = packet_mm_open,
1580         .close =packet_mm_close,
1581 };
1582
1583 static inline struct page *pg_vec_endpage(char *one_pg_vec, unsigned int order)
1584 {
1585         return virt_to_page(one_pg_vec + (PAGE_SIZE << order) - 1);
1586 }
1587
1588 static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len)
1589 {
1590         int i;
1591
1592         for (i = 0; i < len; i++) {
1593                 if (likely(pg_vec[i]))
1594                         free_pages((unsigned long) pg_vec[i], order);
1595         }
1596         kfree(pg_vec);
1597 }
1598
1599 static inline char *alloc_one_pg_vec_page(unsigned long order)
1600 {
1601         return (char *) __get_free_pages(GFP_KERNEL | __GFP_COMP | __GFP_ZERO,
1602                                          order);
1603 }
1604
1605 static char **alloc_pg_vec(struct tpacket_req *req, int order)
1606 {
1607         unsigned int block_nr = req->tp_block_nr;
1608         char **pg_vec;
1609         int i;
1610
1611         pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL);
1612         if (unlikely(!pg_vec))
1613                 goto out;
1614
1615         for (i = 0; i < block_nr; i++) {
1616                 pg_vec[i] = alloc_one_pg_vec_page(order);
1617                 if (unlikely(!pg_vec[i]))
1618                         goto out_free_pgvec;
1619         }
1620
1621 out:
1622         return pg_vec;
1623
1624 out_free_pgvec:
1625         free_pg_vec(pg_vec, order, block_nr);
1626         pg_vec = NULL;
1627         goto out;
1628 }
1629
1630 static int packet_set_ring(struct sock *sk, struct tpacket_req *req, int closing)
1631 {
1632         char **pg_vec = NULL;
1633         struct packet_sock *po = pkt_sk(sk);
1634         int was_running, num, order = 0;
1635         int err = 0;
1636         
1637         if (req->tp_block_nr) {
1638                 int i, l;
1639
1640                 /* Sanity tests and some calculations */
1641
1642                 if (unlikely(po->pg_vec))
1643                         return -EBUSY;
1644
1645                 if (unlikely((int)req->tp_block_size <= 0))
1646                         return -EINVAL;
1647                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
1648                         return -EINVAL;
1649                 if (unlikely(req->tp_frame_size < TPACKET_HDRLEN))
1650                         return -EINVAL;
1651                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
1652                         return -EINVAL;
1653
1654                 po->frames_per_block = req->tp_block_size/req->tp_frame_size;
1655                 if (unlikely(po->frames_per_block <= 0))
1656                         return -EINVAL;
1657                 if (unlikely((po->frames_per_block * req->tp_block_nr) !=
1658                              req->tp_frame_nr))
1659                         return -EINVAL;
1660
1661                 err = -ENOMEM;
1662                 order = get_order(req->tp_block_size);
1663                 pg_vec = alloc_pg_vec(req, order);
1664                 if (unlikely(!pg_vec))
1665                         goto out;
1666
1667                 l = 0;
1668                 for (i = 0; i < req->tp_block_nr; i++) {
1669                         char *ptr = pg_vec[i];
1670                         struct tpacket_hdr *header;
1671                         int k;
1672
1673                         for (k = 0; k < po->frames_per_block; k++) {
1674                                 header = (struct tpacket_hdr *) ptr;
1675                                 header->tp_status = TP_STATUS_KERNEL;
1676                                 ptr += req->tp_frame_size;
1677                         }
1678                 }
1679                 /* Done */
1680         } else {
1681                 if (unlikely(req->tp_frame_nr))
1682                         return -EINVAL;
1683         }
1684
1685         lock_sock(sk);
1686
1687         /* Detach socket from network */
1688         spin_lock(&po->bind_lock);
1689         was_running = po->running;
1690         num = po->num;
1691         if (was_running) {
1692                 __dev_remove_pack(&po->prot_hook);
1693                 po->num = 0;
1694                 po->running = 0;
1695                 __sock_put(sk);
1696         }
1697         spin_unlock(&po->bind_lock);
1698                 
1699         synchronize_net();
1700
1701         err = -EBUSY;
1702         if (closing || atomic_read(&po->mapped) == 0) {
1703                 err = 0;
1704 #define XC(a, b) ({ __typeof__ ((a)) __t; __t = (a); (a) = (b); __t; })
1705
1706                 spin_lock_bh(&sk->sk_receive_queue.lock);
1707                 pg_vec = XC(po->pg_vec, pg_vec);
1708                 po->frame_max = (req->tp_frame_nr - 1);
1709                 po->head = 0;
1710                 po->frame_size = req->tp_frame_size;
1711                 spin_unlock_bh(&sk->sk_receive_queue.lock);
1712
1713                 order = XC(po->pg_vec_order, order);
1714                 req->tp_block_nr = XC(po->pg_vec_len, req->tp_block_nr);
1715
1716                 po->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
1717                 po->prot_hook.func = po->pg_vec ? tpacket_rcv : packet_rcv;
1718                 skb_queue_purge(&sk->sk_receive_queue);
1719 #undef XC
1720                 if (atomic_read(&po->mapped))
1721                         printk(KERN_DEBUG "packet_mmap: vma is busy: %d\n", atomic_read(&po->mapped));
1722         }
1723
1724         spin_lock(&po->bind_lock);
1725         if (was_running && !po->running) {
1726                 sock_hold(sk);
1727                 po->running = 1;
1728                 po->num = num;
1729                 dev_add_pack(&po->prot_hook);
1730         }
1731         spin_unlock(&po->bind_lock);
1732
1733         release_sock(sk);
1734
1735         if (pg_vec)
1736                 free_pg_vec(pg_vec, order, req->tp_block_nr);
1737 out:
1738         return err;
1739 }
1740
1741 static int packet_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1742 {
1743         struct sock *sk = sock->sk;
1744         struct packet_sock *po = pkt_sk(sk);
1745         unsigned long size;
1746         unsigned long start;
1747         int err = -EINVAL;
1748         int i;
1749
1750         if (vma->vm_pgoff)
1751                 return -EINVAL;
1752
1753         size = vma->vm_end - vma->vm_start;
1754
1755         lock_sock(sk);
1756         if (po->pg_vec == NULL)
1757                 goto out;
1758         if (size != po->pg_vec_len*po->pg_vec_pages*PAGE_SIZE)
1759                 goto out;
1760
1761         start = vma->vm_start;
1762         for (i = 0; i < po->pg_vec_len; i++) {
1763                 struct page *page = virt_to_page(po->pg_vec[i]);
1764                 int pg_num;
1765
1766                 for (pg_num = 0; pg_num < po->pg_vec_pages; pg_num++, page++) {
1767                         err = vm_insert_page(vma, start, page);
1768                         if (unlikely(err))
1769                                 goto out;
1770                         start += PAGE_SIZE;
1771                 }
1772         }
1773         atomic_inc(&po->mapped);
1774         vma->vm_ops = &packet_mmap_ops;
1775         err = 0;
1776
1777 out:
1778         release_sock(sk);
1779         return err;
1780 }
1781 #endif
1782
1783
1784 #ifdef CONFIG_SOCK_PACKET
1785 static const struct proto_ops packet_ops_spkt = {
1786         .family =       PF_PACKET,
1787         .owner =        THIS_MODULE,
1788         .release =      packet_release,
1789         .bind =         packet_bind_spkt,
1790         .connect =      sock_no_connect,
1791         .socketpair =   sock_no_socketpair,
1792         .accept =       sock_no_accept,
1793         .getname =      packet_getname_spkt,
1794         .poll =         datagram_poll,
1795         .ioctl =        packet_ioctl,
1796         .listen =       sock_no_listen,
1797         .shutdown =     sock_no_shutdown,
1798         .setsockopt =   sock_no_setsockopt,
1799         .getsockopt =   sock_no_getsockopt,
1800         .sendmsg =      packet_sendmsg_spkt,
1801         .recvmsg =      packet_recvmsg,
1802         .mmap =         sock_no_mmap,
1803         .sendpage =     sock_no_sendpage,
1804 };
1805 #endif
1806
1807 static const struct proto_ops packet_ops = {
1808         .family =       PF_PACKET,
1809         .owner =        THIS_MODULE,
1810         .release =      packet_release,
1811         .bind =         packet_bind,
1812         .connect =      sock_no_connect,
1813         .socketpair =   sock_no_socketpair,
1814         .accept =       sock_no_accept,
1815         .getname =      packet_getname, 
1816         .poll =         packet_poll,
1817         .ioctl =        packet_ioctl,
1818         .listen =       sock_no_listen,
1819         .shutdown =     sock_no_shutdown,
1820         .setsockopt =   packet_setsockopt,
1821         .getsockopt =   packet_getsockopt,
1822         .sendmsg =      packet_sendmsg,
1823         .recvmsg =      packet_recvmsg,
1824         .mmap =         packet_mmap,
1825         .sendpage =     sock_no_sendpage,
1826 };
1827
1828 static struct net_proto_family packet_family_ops = {
1829         .family =       PF_PACKET,
1830         .create =       packet_create,
1831         .owner  =       THIS_MODULE,
1832 };
1833
1834 static struct notifier_block packet_netdev_notifier = {
1835         .notifier_call =packet_notifier,
1836 };
1837
1838 #ifdef CONFIG_PROC_FS
1839 static inline struct sock *packet_seq_idx(loff_t off)
1840 {
1841         struct sock *s;
1842         struct hlist_node *node;
1843
1844         sk_for_each(s, node, &packet_sklist) {
1845                 if (!off--)
1846                         return s;
1847         }
1848         return NULL;
1849 }
1850
1851 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
1852 {
1853         read_lock(&packet_sklist_lock);
1854         return *pos ? packet_seq_idx(*pos - 1) : SEQ_START_TOKEN;
1855 }
1856
1857 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1858 {
1859         ++*pos;
1860         return  (v == SEQ_START_TOKEN) 
1861                 ? sk_head(&packet_sklist) 
1862                 : sk_next((struct sock*)v) ;
1863 }
1864
1865 static void packet_seq_stop(struct seq_file *seq, void *v)
1866 {
1867         read_unlock(&packet_sklist_lock);               
1868 }
1869
1870 static int packet_seq_show(struct seq_file *seq, void *v) 
1871 {
1872         if (v == SEQ_START_TOKEN)
1873                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
1874         else {
1875                 struct sock *s = v;
1876                 const struct packet_sock *po = pkt_sk(s);
1877
1878                 seq_printf(seq,
1879                            "%p %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
1880                            s,
1881                            atomic_read(&s->sk_refcnt),
1882                            s->sk_type,
1883                            ntohs(po->num),
1884                            po->ifindex,
1885                            po->running,
1886                            atomic_read(&s->sk_rmem_alloc),
1887                            sock_i_uid(s),
1888                            sock_i_ino(s) );
1889         }
1890
1891         return 0;
1892 }
1893
1894 static struct seq_operations packet_seq_ops = {
1895         .start  = packet_seq_start,
1896         .next   = packet_seq_next,
1897         .stop   = packet_seq_stop,
1898         .show   = packet_seq_show,
1899 };
1900
1901 static int packet_seq_open(struct inode *inode, struct file *file)
1902 {
1903         return seq_open(file, &packet_seq_ops);
1904 }
1905
1906 static struct file_operations packet_seq_fops = {
1907         .owner          = THIS_MODULE,
1908         .open           = packet_seq_open,
1909         .read           = seq_read,
1910         .llseek         = seq_lseek,
1911         .release        = seq_release,
1912 };
1913
1914 #endif
1915
1916 static void __exit packet_exit(void)
1917 {
1918         proc_net_remove("packet");
1919         unregister_netdevice_notifier(&packet_netdev_notifier);
1920         sock_unregister(PF_PACKET);
1921         proto_unregister(&packet_proto);
1922 }
1923
1924 static int __init packet_init(void)
1925 {
1926         int rc = proto_register(&packet_proto, 0);
1927
1928         if (rc != 0)
1929                 goto out;
1930
1931         sock_register(&packet_family_ops);
1932         register_netdevice_notifier(&packet_netdev_notifier);
1933         proc_net_fops_create("packet", 0, &packet_seq_fops);
1934 out:
1935         return rc;
1936 }
1937
1938 module_init(packet_init);
1939 module_exit(packet_exit);
1940 MODULE_LICENSE("GPL");
1941 MODULE_ALIAS_NETPROTO(PF_PACKET);