net/packet: fix a race in packet_bind() and packet_notifier()
[pandora-kernel.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  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
44  *                                      layer.
45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
46  *
47  *
48  *              This program is free software; you can redistribute it and/or
49  *              modify it under the terms of the GNU General Public License
50  *              as published by the Free Software Foundation; either version
51  *              2 of the License, or (at your option) any later version.
52  *
53  */
54
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/system.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <asm/page.h>
80 #include <asm/cacheflush.h>
81 #include <asm/io.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
92
93 #ifdef CONFIG_INET
94 #include <net/inet_common.h>
95 #endif
96
97 /*
98    Assumptions:
99    - if device has no dev->hard_header routine, it adds and removes ll header
100      inside itself. In this case ll header is invisible outside of device,
101      but higher levels still should reserve dev->hard_header_len.
102      Some devices are enough clever to reallocate skb, when header
103      will not fit to reserved space (tunnel), another ones are silly
104      (PPP).
105    - packet socket receives packets with pulled ll header,
106      so that SOCK_RAW should push it back.
107
108 On receive:
109 -----------
110
111 Incoming, dev->hard_header!=NULL
112    mac_header -> ll header
113    data       -> data
114
115 Outgoing, dev->hard_header!=NULL
116    mac_header -> ll header
117    data       -> ll header
118
119 Incoming, dev->hard_header==NULL
120    mac_header -> UNKNOWN position. It is very likely, that it points to ll
121                  header.  PPP makes it, that is wrong, because introduce
122                  assymetry between rx and tx paths.
123    data       -> data
124
125 Outgoing, dev->hard_header==NULL
126    mac_header -> data. ll header is still not built!
127    data       -> data
128
129 Resume
130   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
131
132
133 On transmit:
134 ------------
135
136 dev->hard_header != NULL
137    mac_header -> ll header
138    data       -> ll header
139
140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141    mac_header -> data
142    data       -> data
143
144    We should set nh.raw on output to correct posistion,
145    packet classifier depends on it.
146  */
147
148 /* Private packet socket structures. */
149
150 struct packet_mclist {
151         struct packet_mclist    *next;
152         int                     ifindex;
153         int                     count;
154         unsigned short          type;
155         unsigned short          alen;
156         unsigned char           addr[MAX_ADDR_LEN];
157 };
158 /* identical to struct packet_mreq except it has
159  * a longer address field.
160  */
161 struct packet_mreq_max {
162         int             mr_ifindex;
163         unsigned short  mr_type;
164         unsigned short  mr_alen;
165         unsigned char   mr_address[MAX_ADDR_LEN];
166 };
167
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169                 int closing, int tx_ring);
170
171
172 #define V3_ALIGNMENT    (8)
173
174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179 /* kbdq - kernel block descriptor queue */
180 struct tpacket_kbdq_core {
181         struct pgv      *pkbdq;
182         unsigned int    feature_req_word;
183         unsigned int    hdrlen;
184         unsigned char   reset_pending_on_curr_blk;
185         unsigned char   delete_blk_timer;
186         unsigned short  kactive_blk_num;
187         unsigned short  blk_sizeof_priv;
188
189         /* last_kactive_blk_num:
190          * trick to see if user-space has caught up
191          * in order to avoid refreshing timer when every single pkt arrives.
192          */
193         unsigned short  last_kactive_blk_num;
194
195         char            *pkblk_start;
196         char            *pkblk_end;
197         int             kblk_size;
198         unsigned int    max_frame_len;
199         unsigned int    knum_blocks;
200         uint64_t        knxt_seq_num;
201         char            *prev;
202         char            *nxt_offset;
203         struct sk_buff  *skb;
204
205         atomic_t        blk_fill_in_prog;
206
207         /* Default is set to 8ms */
208 #define DEFAULT_PRB_RETIRE_TOV  (8)
209
210         unsigned short  retire_blk_tov;
211         unsigned short  version;
212         unsigned long   tov_in_jiffies;
213
214         /* timer to retire an outstanding block */
215         struct timer_list retire_blk_timer;
216 };
217
218 #define PGV_FROM_VMALLOC 1
219 struct pgv {
220         char *buffer;
221 };
222
223 struct packet_ring_buffer {
224         struct pgv              *pg_vec;
225         unsigned int            head;
226         unsigned int            frames_per_block;
227         unsigned int            frame_size;
228         unsigned int            frame_max;
229
230         unsigned int            pg_vec_order;
231         unsigned int            pg_vec_pages;
232         unsigned int            pg_vec_len;
233
234         struct tpacket_kbdq_core        prb_bdqc;
235         atomic_t                pending;
236 };
237
238 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
239 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
240 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
241 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
242 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
243 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
244 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
245
246 struct packet_sock;
247 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
248
249 static void *packet_previous_frame(struct packet_sock *po,
250                 struct packet_ring_buffer *rb,
251                 int status);
252 static void packet_increment_head(struct packet_ring_buffer *buff);
253 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
254                         struct tpacket_block_desc *);
255 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
256                         struct packet_sock *);
257 static void prb_retire_current_block(struct tpacket_kbdq_core *,
258                 struct packet_sock *, unsigned int status);
259 static int prb_queue_frozen(struct tpacket_kbdq_core *);
260 static void prb_open_block(struct tpacket_kbdq_core *,
261                 struct tpacket_block_desc *);
262 static void prb_retire_rx_blk_timer_expired(unsigned long);
263 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
264 static void prb_init_blk_timer(struct packet_sock *,
265                 struct tpacket_kbdq_core *,
266                 void (*func) (unsigned long));
267 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
268 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
269                 struct tpacket3_hdr *);
270 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
271                 struct tpacket3_hdr *);
272 static void packet_flush_mclist(struct sock *sk);
273
274 struct packet_fanout;
275 struct packet_sock {
276         /* struct sock has to be the first member of packet_sock */
277         struct sock             sk;
278         struct packet_fanout    *fanout;
279         struct tpacket_stats    stats;
280         union  tpacket_stats_u  stats_u;
281         struct packet_ring_buffer       rx_ring;
282         struct packet_ring_buffer       tx_ring;
283         int                     copy_thresh;
284         spinlock_t              bind_lock;
285         struct mutex            pg_vec_lock;
286         unsigned int            running:1,      /* prot_hook is attached*/
287                                 auxdata:1,
288                                 origdev:1,
289                                 has_vnet_hdr:1;
290         int                     ifindex;        /* bound device         */
291         __be16                  num;
292         struct packet_mclist    *mclist;
293         atomic_t                mapped;
294         enum tpacket_versions   tp_version;
295         unsigned int            tp_hdrlen;
296         unsigned int            tp_reserve;
297         unsigned int            tp_loss:1;
298         unsigned int            tp_tstamp;
299         struct net_device __rcu *cached_dev;
300         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
301 };
302
303 #define PACKET_FANOUT_MAX       256
304
305 struct packet_fanout {
306 #ifdef CONFIG_NET_NS
307         struct net              *net;
308 #endif
309         unsigned int            num_members;
310         u16                     id;
311         u8                      type;
312         u8                      defrag;
313         atomic_t                rr_cur;
314         struct list_head        list;
315         struct sock             *arr[PACKET_FANOUT_MAX];
316         spinlock_t              lock;
317         atomic_t                sk_ref;
318         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
319 };
320
321 struct packet_skb_cb {
322         unsigned int origlen;
323         union {
324                 struct sockaddr_pkt pkt;
325                 struct sockaddr_ll ll;
326         } sa;
327 };
328
329 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
330
331 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
332 #define GET_PBLOCK_DESC(x, bid) \
333         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
334 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
335         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
336 #define GET_NEXT_PRB_BLK_NUM(x) \
337         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
338         ((x)->kactive_blk_num+1) : 0)
339
340 static struct packet_sock *pkt_sk(struct sock *sk)
341 {
342         return (struct packet_sock *)sk;
343 }
344
345 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
346 static void __fanout_link(struct sock *sk, struct packet_sock *po);
347
348 /* register_prot_hook must be invoked with the po->bind_lock held,
349  * or from a context in which asynchronous accesses to the packet
350  * socket is not possible (packet_create()).
351  */
352 static void register_prot_hook(struct sock *sk)
353 {
354         struct packet_sock *po = pkt_sk(sk);
355
356         if (!po->running) {
357                 if (po->fanout) {
358                         __fanout_link(sk, po);
359                 } else {
360                         dev_add_pack(&po->prot_hook);
361                         rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
362                 }
363
364                 sock_hold(sk);
365                 po->running = 1;
366         }
367 }
368
369 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
370  * held.   If the sync parameter is true, we will temporarily drop
371  * the po->bind_lock and do a synchronize_net to make sure no
372  * asynchronous packet processing paths still refer to the elements
373  * of po->prot_hook.  If the sync parameter is false, it is the
374  * callers responsibility to take care of this.
375  */
376 static void __unregister_prot_hook(struct sock *sk, bool sync)
377 {
378         struct packet_sock *po = pkt_sk(sk);
379
380         po->running = 0;
381         if (po->fanout) {
382                 __fanout_unlink(sk, po);
383         } else {
384                 __dev_remove_pack(&po->prot_hook);
385                 RCU_INIT_POINTER(po->cached_dev, NULL);
386         }
387
388         __sock_put(sk);
389
390         if (sync) {
391                 spin_unlock(&po->bind_lock);
392                 synchronize_net();
393                 spin_lock(&po->bind_lock);
394         }
395 }
396
397 static void unregister_prot_hook(struct sock *sk, bool sync)
398 {
399         struct packet_sock *po = pkt_sk(sk);
400
401         if (po->running)
402                 __unregister_prot_hook(sk, sync);
403 }
404
405 static inline __pure struct page *pgv_to_page(void *addr)
406 {
407         if (is_vmalloc_addr(addr))
408                 return vmalloc_to_page(addr);
409         return virt_to_page(addr);
410 }
411
412 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
413 {
414         union {
415                 struct tpacket_hdr *h1;
416                 struct tpacket2_hdr *h2;
417                 void *raw;
418         } h;
419
420         h.raw = frame;
421         switch (po->tp_version) {
422         case TPACKET_V1:
423                 h.h1->tp_status = status;
424                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
425                 break;
426         case TPACKET_V2:
427                 h.h2->tp_status = status;
428                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
429                 break;
430         case TPACKET_V3:
431         default:
432                 WARN(1, "TPACKET version not supported.\n");
433                 BUG();
434         }
435
436         smp_wmb();
437 }
438
439 static int __packet_get_status(struct packet_sock *po, void *frame)
440 {
441         union {
442                 struct tpacket_hdr *h1;
443                 struct tpacket2_hdr *h2;
444                 void *raw;
445         } h;
446
447         smp_rmb();
448
449         h.raw = frame;
450         switch (po->tp_version) {
451         case TPACKET_V1:
452                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
453                 return h.h1->tp_status;
454         case TPACKET_V2:
455                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
456                 return h.h2->tp_status;
457         case TPACKET_V3:
458         default:
459                 WARN(1, "TPACKET version not supported.\n");
460                 BUG();
461                 return 0;
462         }
463 }
464
465 static void *packet_lookup_frame(struct packet_sock *po,
466                 struct packet_ring_buffer *rb,
467                 unsigned int position,
468                 int status)
469 {
470         unsigned int pg_vec_pos, frame_offset;
471         union {
472                 struct tpacket_hdr *h1;
473                 struct tpacket2_hdr *h2;
474                 void *raw;
475         } h;
476
477         pg_vec_pos = position / rb->frames_per_block;
478         frame_offset = position % rb->frames_per_block;
479
480         h.raw = rb->pg_vec[pg_vec_pos].buffer +
481                 (frame_offset * rb->frame_size);
482
483         if (status != __packet_get_status(po, h.raw))
484                 return NULL;
485
486         return h.raw;
487 }
488
489 static void *packet_current_frame(struct packet_sock *po,
490                 struct packet_ring_buffer *rb,
491                 int status)
492 {
493         return packet_lookup_frame(po, rb, rb->head, status);
494 }
495
496 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
497 {
498         del_timer_sync(&pkc->retire_blk_timer);
499 }
500
501 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
502                 int tx_ring,
503                 struct sk_buff_head *rb_queue)
504 {
505         struct tpacket_kbdq_core *pkc;
506
507         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
508
509         spin_lock_bh(&rb_queue->lock);
510         pkc->delete_blk_timer = 1;
511         spin_unlock_bh(&rb_queue->lock);
512
513         prb_del_retire_blk_timer(pkc);
514 }
515
516 static void prb_init_blk_timer(struct packet_sock *po,
517                 struct tpacket_kbdq_core *pkc,
518                 void (*func) (unsigned long))
519 {
520         init_timer(&pkc->retire_blk_timer);
521         pkc->retire_blk_timer.data = (long)po;
522         pkc->retire_blk_timer.function = func;
523         pkc->retire_blk_timer.expires = jiffies;
524 }
525
526 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
527 {
528         struct tpacket_kbdq_core *pkc;
529
530         if (tx_ring)
531                 BUG();
532
533         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
534         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
535 }
536
537 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
538                                 int blk_size_in_bytes)
539 {
540         struct net_device *dev;
541         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
542         struct ethtool_cmd ecmd;
543         int err;
544
545         rtnl_lock();
546         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
547         if (unlikely(!dev)) {
548                 rtnl_unlock();
549                 return DEFAULT_PRB_RETIRE_TOV;
550         }
551         err = __ethtool_get_settings(dev, &ecmd);
552         rtnl_unlock();
553         if (!err) {
554                 switch (ecmd.speed) {
555                 case SPEED_10000:
556                         msec = 1;
557                         div = 10000/1000;
558                         break;
559                 case SPEED_1000:
560                         msec = 1;
561                         div = 1000/1000;
562                         break;
563                 /*
564                  * If the link speed is so slow you don't really
565                  * need to worry about perf anyways
566                  */
567                 case SPEED_100:
568                 case SPEED_10:
569                 default:
570                         return DEFAULT_PRB_RETIRE_TOV;
571                 }
572         }
573
574         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
575
576         if (div)
577                 mbits /= div;
578
579         tmo = mbits * msec;
580
581         if (div)
582                 return tmo+1;
583         return tmo;
584 }
585
586 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
587                         union tpacket_req_u *req_u)
588 {
589         p1->feature_req_word = req_u->req3.tp_feature_req_word;
590 }
591
592 static void init_prb_bdqc(struct packet_sock *po,
593                         struct packet_ring_buffer *rb,
594                         struct pgv *pg_vec,
595                         union tpacket_req_u *req_u, int tx_ring)
596 {
597         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
598         struct tpacket_block_desc *pbd;
599
600         memset(p1, 0x0, sizeof(*p1));
601
602         p1->knxt_seq_num = 1;
603         p1->pkbdq = pg_vec;
604         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
605         p1->pkblk_start = (char *)pg_vec[0].buffer;
606         p1->kblk_size = req_u->req3.tp_block_size;
607         p1->knum_blocks = req_u->req3.tp_block_nr;
608         p1->hdrlen = po->tp_hdrlen;
609         p1->version = po->tp_version;
610         p1->last_kactive_blk_num = 0;
611         po->stats_u.stats3.tp_freeze_q_cnt = 0;
612         if (req_u->req3.tp_retire_blk_tov)
613                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
614         else
615                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
616                                                 req_u->req3.tp_block_size);
617         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
618         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
619
620         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
621         prb_init_ft_ops(p1, req_u);
622         prb_setup_retire_blk_timer(po, tx_ring);
623         prb_open_block(p1, pbd);
624 }
625
626 /*  Do NOT update the last_blk_num first.
627  *  Assumes sk_buff_head lock is held.
628  */
629 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
630 {
631         mod_timer(&pkc->retire_blk_timer,
632                         jiffies + pkc->tov_in_jiffies);
633         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
634 }
635
636 /*
637  * Timer logic:
638  * 1) We refresh the timer only when we open a block.
639  *    By doing this we don't waste cycles refreshing the timer
640  *        on packet-by-packet basis.
641  *
642  * With a 1MB block-size, on a 1Gbps line, it will take
643  * i) ~8 ms to fill a block + ii) memcpy etc.
644  * In this cut we are not accounting for the memcpy time.
645  *
646  * So, if the user sets the 'tmo' to 10ms then the timer
647  * will never fire while the block is still getting filled
648  * (which is what we want). However, the user could choose
649  * to close a block early and that's fine.
650  *
651  * But when the timer does fire, we check whether or not to refresh it.
652  * Since the tmo granularity is in msecs, it is not too expensive
653  * to refresh the timer, lets say every '8' msecs.
654  * Either the user can set the 'tmo' or we can derive it based on
655  * a) line-speed and b) block-size.
656  * prb_calc_retire_blk_tmo() calculates the tmo.
657  *
658  */
659 static void prb_retire_rx_blk_timer_expired(unsigned long data)
660 {
661         struct packet_sock *po = (struct packet_sock *)data;
662         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
663         unsigned int frozen;
664         struct tpacket_block_desc *pbd;
665
666         spin_lock(&po->sk.sk_receive_queue.lock);
667
668         frozen = prb_queue_frozen(pkc);
669         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
670
671         if (unlikely(pkc->delete_blk_timer))
672                 goto out;
673
674         /* We only need to plug the race when the block is partially filled.
675          * tpacket_rcv:
676          *              lock(); increment BLOCK_NUM_PKTS; unlock()
677          *              copy_bits() is in progress ...
678          *              timer fires on other cpu:
679          *              we can't retire the current block because copy_bits
680          *              is in progress.
681          *
682          */
683         if (BLOCK_NUM_PKTS(pbd)) {
684                 while (atomic_read(&pkc->blk_fill_in_prog)) {
685                         /* Waiting for skb_copy_bits to finish... */
686                         cpu_relax();
687                 }
688         }
689
690         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
691                 if (!frozen) {
692                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
693                         if (!prb_dispatch_next_block(pkc, po))
694                                 goto refresh_timer;
695                         else
696                                 goto out;
697                 } else {
698                         /* Case 1. Queue was frozen because user-space was
699                          *         lagging behind.
700                          */
701                         if (prb_curr_blk_in_use(pkc, pbd)) {
702                                 /*
703                                  * Ok, user-space is still behind.
704                                  * So just refresh the timer.
705                                  */
706                                 goto refresh_timer;
707                         } else {
708                                /* Case 2. queue was frozen,user-space caught up,
709                                 * now the link went idle && the timer fired.
710                                 * We don't have a block to close.So we open this
711                                 * block and restart the timer.
712                                 * opening a block thaws the queue,restarts timer
713                                 * Thawing/timer-refresh is a side effect.
714                                 */
715                                 prb_open_block(pkc, pbd);
716                                 goto out;
717                         }
718                 }
719         }
720
721 refresh_timer:
722         _prb_refresh_rx_retire_blk_timer(pkc);
723
724 out:
725         spin_unlock(&po->sk.sk_receive_queue.lock);
726 }
727
728 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
729                 struct tpacket_block_desc *pbd1, __u32 status)
730 {
731         /* Flush everything minus the block header */
732
733 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
734         u8 *start, *end;
735
736         start = (u8 *)pbd1;
737
738         /* Skip the block header(we know header WILL fit in 4K) */
739         start += PAGE_SIZE;
740
741         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
742         for (; start < end; start += PAGE_SIZE)
743                 flush_dcache_page(pgv_to_page(start));
744
745         smp_wmb();
746 #endif
747
748         /* Now update the block status. */
749
750         BLOCK_STATUS(pbd1) = status;
751
752         /* Flush the block header */
753
754 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
755         start = (u8 *)pbd1;
756         flush_dcache_page(pgv_to_page(start));
757
758         smp_wmb();
759 #endif
760 }
761
762 /*
763  * Side effect:
764  *
765  * 1) flush the block
766  * 2) Increment active_blk_num
767  *
768  * Note:We DONT refresh the timer on purpose.
769  *      Because almost always the next block will be opened.
770  */
771 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
772                 struct tpacket_block_desc *pbd1,
773                 struct packet_sock *po, unsigned int stat)
774 {
775         __u32 status = TP_STATUS_USER | stat;
776
777         struct tpacket3_hdr *last_pkt;
778         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
779
780         if (po->stats.tp_drops)
781                 status |= TP_STATUS_LOSING;
782
783         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
784         last_pkt->tp_next_offset = 0;
785
786         /* Get the ts of the last pkt */
787         if (BLOCK_NUM_PKTS(pbd1)) {
788                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
789                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
790         } else {
791                 /* Ok, we tmo'd - so get the current time */
792                 struct timespec ts;
793                 getnstimeofday(&ts);
794                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
795                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
796         }
797
798         smp_wmb();
799
800         /* Flush the block */
801         prb_flush_block(pkc1, pbd1, status);
802
803         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
804 }
805
806 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
807 {
808         pkc->reset_pending_on_curr_blk = 0;
809 }
810
811 /*
812  * Side effect of opening a block:
813  *
814  * 1) prb_queue is thawed.
815  * 2) retire_blk_timer is refreshed.
816  *
817  */
818 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
819         struct tpacket_block_desc *pbd1)
820 {
821         struct timespec ts;
822         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
823
824         smp_rmb();
825
826         /* We could have just memset this but we will lose the
827          * flexibility of making the priv area sticky
828          */
829         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
830         BLOCK_NUM_PKTS(pbd1) = 0;
831         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
832         getnstimeofday(&ts);
833         h1->ts_first_pkt.ts_sec = ts.tv_sec;
834         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
835         pkc1->pkblk_start = (char *)pbd1;
836         pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
837                                     BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
838         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
839         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
840         pbd1->version = pkc1->version;
841         pkc1->prev = pkc1->nxt_offset;
842         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
843         prb_thaw_queue(pkc1);
844         _prb_refresh_rx_retire_blk_timer(pkc1);
845
846         smp_wmb();
847 }
848
849 /*
850  * Queue freeze logic:
851  * 1) Assume tp_block_nr = 8 blocks.
852  * 2) At time 't0', user opens Rx ring.
853  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854  * 4) user-space is either sleeping or processing block '0'.
855  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856  *    it will close block-7,loop around and try to fill block '0'.
857  *    call-flow:
858  *    __packet_lookup_frame_in_block
859  *      prb_retire_current_block()
860  *      prb_dispatch_next_block()
861  *        |->(BLOCK_STATUS == USER) evaluates to true
862  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
863  * 6) Now there are two cases:
864  *    6.1) Link goes idle right after the queue is frozen.
865  *         But remember, the last open_block() refreshed the timer.
866  *         When this timer expires,it will refresh itself so that we can
867  *         re-open block-0 in near future.
868  *    6.2) Link is busy and keeps on receiving packets. This is a simple
869  *         case and __packet_lookup_frame_in_block will check if block-0
870  *         is free and can now be re-used.
871  */
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873                                   struct packet_sock *po)
874 {
875         pkc->reset_pending_on_curr_blk = 1;
876         po->stats_u.stats3.tp_freeze_q_cnt++;
877 }
878
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880
881 /*
882  * If the next block is free then we will dispatch it
883  * and return a good offset.
884  * Else, we will freeze the queue.
885  * So, caller must check the return value.
886  */
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888                 struct packet_sock *po)
889 {
890         struct tpacket_block_desc *pbd;
891
892         smp_rmb();
893
894         /* 1. Get current block num */
895         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896
897         /* 2. If this block is currently in_use then freeze the queue */
898         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899                 prb_freeze_queue(pkc, po);
900                 return NULL;
901         }
902
903         /*
904          * 3.
905          * open this block and return the offset where the first packet
906          * needs to get stored.
907          */
908         prb_open_block(pkc, pbd);
909         return (void *)pkc->nxt_offset;
910 }
911
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913                 struct packet_sock *po, unsigned int status)
914 {
915         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916
917         /* retire/close the current block */
918         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919                 /*
920                  * Plug the case where copy_bits() is in progress on
921                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922                  * have space to copy the pkt in the current block and
923                  * called prb_retire_current_block()
924                  *
925                  * We don't need to worry about the TMO case because
926                  * the timer-handler already handled this case.
927                  */
928                 if (!(status & TP_STATUS_BLK_TMO)) {
929                         while (atomic_read(&pkc->blk_fill_in_prog)) {
930                                 /* Waiting for skb_copy_bits to finish... */
931                                 cpu_relax();
932                         }
933                 }
934                 prb_close_block(pkc, pbd, po, status);
935                 return;
936         }
937 }
938
939 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
940                                       struct tpacket_block_desc *pbd)
941 {
942         return TP_STATUS_USER & BLOCK_STATUS(pbd);
943 }
944
945 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
946 {
947         return pkc->reset_pending_on_curr_blk;
948 }
949
950 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
951 {
952         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
953         atomic_dec(&pkc->blk_fill_in_prog);
954 }
955
956 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
957                         struct tpacket3_hdr *ppd)
958 {
959         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
960 }
961
962 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
963                         struct tpacket3_hdr *ppd)
964 {
965         ppd->hv1.tp_rxhash = 0;
966 }
967
968 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
969                         struct tpacket3_hdr *ppd)
970 {
971         if (vlan_tx_tag_present(pkc->skb)) {
972                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
973                 ppd->tp_status = TP_STATUS_VLAN_VALID;
974         } else {
975                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
976         }
977 }
978
979 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
980                         struct tpacket3_hdr *ppd)
981 {
982         prb_fill_vlan_info(pkc, ppd);
983
984         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
985                 prb_fill_rxhash(pkc, ppd);
986         else
987                 prb_clear_rxhash(pkc, ppd);
988 }
989
990 static void prb_fill_curr_block(char *curr,
991                                 struct tpacket_kbdq_core *pkc,
992                                 struct tpacket_block_desc *pbd,
993                                 unsigned int len)
994 {
995         struct tpacket3_hdr *ppd;
996
997         ppd  = (struct tpacket3_hdr *)curr;
998         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
999         pkc->prev = curr;
1000         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1001         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1002         BLOCK_NUM_PKTS(pbd) += 1;
1003         atomic_inc(&pkc->blk_fill_in_prog);
1004         prb_run_all_ft_ops(pkc, ppd);
1005 }
1006
1007 /* Assumes caller has the sk->rx_queue.lock */
1008 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1009                                             struct sk_buff *skb,
1010                                                 int status,
1011                                             unsigned int len
1012                                             )
1013 {
1014         struct tpacket_kbdq_core *pkc;
1015         struct tpacket_block_desc *pbd;
1016         char *curr, *end;
1017
1018         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1019         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1020
1021         /* Queue is frozen when user space is lagging behind */
1022         if (prb_queue_frozen(pkc)) {
1023                 /*
1024                  * Check if that last block which caused the queue to freeze,
1025                  * is still in_use by user-space.
1026                  */
1027                 if (prb_curr_blk_in_use(pkc, pbd)) {
1028                         /* Can't record this packet */
1029                         return NULL;
1030                 } else {
1031                         /*
1032                          * Ok, the block was released by user-space.
1033                          * Now let's open that block.
1034                          * opening a block also thaws the queue.
1035                          * Thawing is a side effect.
1036                          */
1037                         prb_open_block(pkc, pbd);
1038                 }
1039         }
1040
1041         smp_mb();
1042         curr = pkc->nxt_offset;
1043         pkc->skb = skb;
1044         end = (char *) ((char *)pbd + pkc->kblk_size);
1045
1046         /* first try the current block */
1047         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1048                 prb_fill_curr_block(curr, pkc, pbd, len);
1049                 return (void *)curr;
1050         }
1051
1052         /* Ok, close the current block */
1053         prb_retire_current_block(pkc, po, 0);
1054
1055         /* Now, try to dispatch the next block */
1056         curr = (char *)prb_dispatch_next_block(pkc, po);
1057         if (curr) {
1058                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1059                 prb_fill_curr_block(curr, pkc, pbd, len);
1060                 return (void *)curr;
1061         }
1062
1063         /*
1064          * No free blocks are available.user_space hasn't caught up yet.
1065          * Queue was just frozen and now this packet will get dropped.
1066          */
1067         return NULL;
1068 }
1069
1070 static void *packet_current_rx_frame(struct packet_sock *po,
1071                                             struct sk_buff *skb,
1072                                             int status, unsigned int len)
1073 {
1074         char *curr = NULL;
1075         switch (po->tp_version) {
1076         case TPACKET_V1:
1077         case TPACKET_V2:
1078                 curr = packet_lookup_frame(po, &po->rx_ring,
1079                                         po->rx_ring.head, status);
1080                 return curr;
1081         case TPACKET_V3:
1082                 return __packet_lookup_frame_in_block(po, skb, status, len);
1083         default:
1084                 WARN(1, "TPACKET version not supported\n");
1085                 BUG();
1086                 return 0;
1087         }
1088 }
1089
1090 static void *prb_lookup_block(struct packet_sock *po,
1091                                      struct packet_ring_buffer *rb,
1092                                      unsigned int previous,
1093                                      int status)
1094 {
1095         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1096         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1097
1098         if (status != BLOCK_STATUS(pbd))
1099                 return NULL;
1100         return pbd;
1101 }
1102
1103 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1104 {
1105         unsigned int prev;
1106         if (rb->prb_bdqc.kactive_blk_num)
1107                 prev = rb->prb_bdqc.kactive_blk_num-1;
1108         else
1109                 prev = rb->prb_bdqc.knum_blocks-1;
1110         return prev;
1111 }
1112
1113 /* Assumes caller has held the rx_queue.lock */
1114 static void *__prb_previous_block(struct packet_sock *po,
1115                                          struct packet_ring_buffer *rb,
1116                                          int status)
1117 {
1118         unsigned int previous = prb_previous_blk_num(rb);
1119         return prb_lookup_block(po, rb, previous, status);
1120 }
1121
1122 static void *packet_previous_rx_frame(struct packet_sock *po,
1123                                              struct packet_ring_buffer *rb,
1124                                              int status)
1125 {
1126         if (po->tp_version <= TPACKET_V2)
1127                 return packet_previous_frame(po, rb, status);
1128
1129         return __prb_previous_block(po, rb, status);
1130 }
1131
1132 static void packet_increment_rx_head(struct packet_sock *po,
1133                                             struct packet_ring_buffer *rb)
1134 {
1135         switch (po->tp_version) {
1136         case TPACKET_V1:
1137         case TPACKET_V2:
1138                 return packet_increment_head(rb);
1139         case TPACKET_V3:
1140         default:
1141                 WARN(1, "TPACKET version not supported.\n");
1142                 BUG();
1143                 return;
1144         }
1145 }
1146
1147 static void *packet_previous_frame(struct packet_sock *po,
1148                 struct packet_ring_buffer *rb,
1149                 int status)
1150 {
1151         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1152         return packet_lookup_frame(po, rb, previous, status);
1153 }
1154
1155 static void packet_increment_head(struct packet_ring_buffer *buff)
1156 {
1157         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1158 }
1159
1160 static void packet_sock_destruct(struct sock *sk)
1161 {
1162         skb_queue_purge(&sk->sk_error_queue);
1163
1164         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1165         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1166
1167         if (!sock_flag(sk, SOCK_DEAD)) {
1168                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1169                 return;
1170         }
1171
1172         sk_refcnt_debug_dec(sk);
1173 }
1174
1175 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1176 {
1177         u32 idx, hash = skb->rxhash;
1178
1179         idx = ((u64)hash * num) >> 32;
1180
1181         return f->arr[idx];
1182 }
1183
1184 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1185 {
1186         unsigned int val = atomic_inc_return(&f->rr_cur);
1187
1188         return f->arr[val % num];
1189 }
1190
1191 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1192 {
1193         unsigned int cpu = smp_processor_id();
1194
1195         return f->arr[cpu % num];
1196 }
1197
1198 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1199                              struct packet_type *pt, struct net_device *orig_dev)
1200 {
1201         struct packet_fanout *f = pt->af_packet_priv;
1202         unsigned int num = ACCESS_ONCE(f->num_members);
1203         struct packet_sock *po;
1204         struct sock *sk;
1205
1206         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1207             !num) {
1208                 kfree_skb(skb);
1209                 return 0;
1210         }
1211
1212         switch (f->type) {
1213         case PACKET_FANOUT_HASH:
1214         default:
1215                 if (f->defrag) {
1216                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1217                         if (!skb)
1218                                 return 0;
1219                 }
1220                 skb_get_rxhash(skb);
1221                 sk = fanout_demux_hash(f, skb, num);
1222                 break;
1223         case PACKET_FANOUT_LB:
1224                 sk = fanout_demux_lb(f, skb, num);
1225                 break;
1226         case PACKET_FANOUT_CPU:
1227                 sk = fanout_demux_cpu(f, skb, num);
1228                 break;
1229         }
1230
1231         po = pkt_sk(sk);
1232
1233         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1234 }
1235
1236 static DEFINE_MUTEX(fanout_mutex);
1237 static LIST_HEAD(fanout_list);
1238
1239 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1240 {
1241         struct packet_fanout *f = po->fanout;
1242
1243         spin_lock(&f->lock);
1244         f->arr[f->num_members] = sk;
1245         smp_wmb();
1246         f->num_members++;
1247         if (f->num_members == 1)
1248                 dev_add_pack(&f->prot_hook);
1249         spin_unlock(&f->lock);
1250 }
1251
1252 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1253 {
1254         struct packet_fanout *f = po->fanout;
1255         int i;
1256
1257         spin_lock(&f->lock);
1258         for (i = 0; i < f->num_members; i++) {
1259                 if (f->arr[i] == sk)
1260                         break;
1261         }
1262         BUG_ON(i >= f->num_members);
1263         f->arr[i] = f->arr[f->num_members - 1];
1264         f->num_members--;
1265         if (f->num_members == 0)
1266                 __dev_remove_pack(&f->prot_hook);
1267         spin_unlock(&f->lock);
1268 }
1269
1270 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1271 {
1272         if (sk->sk_family != PF_PACKET)
1273                 return false;
1274
1275         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1276 }
1277
1278 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1279 {
1280         struct packet_sock *po = pkt_sk(sk);
1281         struct packet_fanout *f, *match;
1282         u8 type = type_flags & 0xff;
1283         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1284         int err;
1285
1286         switch (type) {
1287         case PACKET_FANOUT_HASH:
1288         case PACKET_FANOUT_LB:
1289         case PACKET_FANOUT_CPU:
1290                 break;
1291         default:
1292                 return -EINVAL;
1293         }
1294
1295         mutex_lock(&fanout_mutex);
1296
1297         err = -EALREADY;
1298         if (po->fanout)
1299                 goto out;
1300
1301         match = NULL;
1302         list_for_each_entry(f, &fanout_list, list) {
1303                 if (f->id == id &&
1304                     read_pnet(&f->net) == sock_net(sk)) {
1305                         match = f;
1306                         break;
1307                 }
1308         }
1309         err = -EINVAL;
1310         if (match && match->defrag != defrag)
1311                 goto out;
1312         if (!match) {
1313                 err = -ENOMEM;
1314                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1315                 if (!match)
1316                         goto out;
1317                 write_pnet(&match->net, sock_net(sk));
1318                 match->id = id;
1319                 match->type = type;
1320                 match->defrag = defrag;
1321                 atomic_set(&match->rr_cur, 0);
1322                 INIT_LIST_HEAD(&match->list);
1323                 spin_lock_init(&match->lock);
1324                 atomic_set(&match->sk_ref, 0);
1325                 match->prot_hook.type = po->prot_hook.type;
1326                 match->prot_hook.dev = po->prot_hook.dev;
1327                 match->prot_hook.func = packet_rcv_fanout;
1328                 match->prot_hook.af_packet_priv = match;
1329                 match->prot_hook.id_match = match_fanout_group;
1330                 list_add(&match->list, &fanout_list);
1331         }
1332         err = -EINVAL;
1333
1334         spin_lock(&po->bind_lock);
1335         if (po->running &&
1336             match->type == type &&
1337             match->prot_hook.type == po->prot_hook.type &&
1338             match->prot_hook.dev == po->prot_hook.dev) {
1339                 err = -ENOSPC;
1340                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1341                         __dev_remove_pack(&po->prot_hook);
1342                         po->fanout = match;
1343                         atomic_inc(&match->sk_ref);
1344                         __fanout_link(sk, po);
1345                         err = 0;
1346                 }
1347         }
1348         spin_unlock(&po->bind_lock);
1349
1350         if (err && !atomic_read(&match->sk_ref)) {
1351                 list_del(&match->list);
1352                 kfree(match);
1353         }
1354
1355 out:
1356         mutex_unlock(&fanout_mutex);
1357         return err;
1358 }
1359
1360 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1361  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1362  * It is the responsibility of the caller to call fanout_release_data() and
1363  * free the returned packet_fanout (after synchronize_net())
1364  */
1365 static struct packet_fanout *fanout_release(struct sock *sk)
1366 {
1367         struct packet_sock *po = pkt_sk(sk);
1368         struct packet_fanout *f;
1369
1370         mutex_lock(&fanout_mutex);
1371         f = po->fanout;
1372         if (f) {
1373                 po->fanout = NULL;
1374
1375                 if (atomic_dec_and_test(&f->sk_ref))
1376                         list_del(&f->list);
1377                 else
1378                         f = NULL;
1379         }
1380         mutex_unlock(&fanout_mutex);
1381
1382         return f;
1383 }
1384
1385 static const struct proto_ops packet_ops;
1386
1387 static const struct proto_ops packet_ops_spkt;
1388
1389 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1390                            struct packet_type *pt, struct net_device *orig_dev)
1391 {
1392         struct sock *sk;
1393         struct sockaddr_pkt *spkt;
1394
1395         /*
1396          *      When we registered the protocol we saved the socket in the data
1397          *      field for just this event.
1398          */
1399
1400         sk = pt->af_packet_priv;
1401
1402         /*
1403          *      Yank back the headers [hope the device set this
1404          *      right or kerboom...]
1405          *
1406          *      Incoming packets have ll header pulled,
1407          *      push it back.
1408          *
1409          *      For outgoing ones skb->data == skb_mac_header(skb)
1410          *      so that this procedure is noop.
1411          */
1412
1413         if (skb->pkt_type == PACKET_LOOPBACK)
1414                 goto out;
1415
1416         if (!net_eq(dev_net(dev), sock_net(sk)))
1417                 goto out;
1418
1419         skb = skb_share_check(skb, GFP_ATOMIC);
1420         if (skb == NULL)
1421                 goto oom;
1422
1423         /* drop any routing info */
1424         skb_dst_drop(skb);
1425
1426         /* drop conntrack reference */
1427         nf_reset(skb);
1428
1429         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1430
1431         skb_push(skb, skb->data - skb_mac_header(skb));
1432
1433         /*
1434          *      The SOCK_PACKET socket receives _all_ frames.
1435          */
1436
1437         spkt->spkt_family = dev->type;
1438         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1439         spkt->spkt_protocol = skb->protocol;
1440
1441         /*
1442          *      Charge the memory to the socket. This is done specifically
1443          *      to prevent sockets using all the memory up.
1444          */
1445
1446         if (sock_queue_rcv_skb(sk, skb) == 0)
1447                 return 0;
1448
1449 out:
1450         kfree_skb(skb);
1451 oom:
1452         return 0;
1453 }
1454
1455
1456 /*
1457  *      Output a raw packet to a device layer. This bypasses all the other
1458  *      protocol layers and you must therefore supply it with a complete frame
1459  */
1460
1461 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1462                                struct msghdr *msg, size_t len)
1463 {
1464         struct sock *sk = sock->sk;
1465         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1466         struct sk_buff *skb = NULL;
1467         struct net_device *dev;
1468         __be16 proto = 0;
1469         int err;
1470
1471         /*
1472          *      Get and verify the address.
1473          */
1474
1475         if (saddr) {
1476                 if (msg->msg_namelen < sizeof(struct sockaddr))
1477                         return -EINVAL;
1478                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1479                         proto = saddr->spkt_protocol;
1480         } else
1481                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1482
1483         /*
1484          *      Find the device first to size check it
1485          */
1486
1487         saddr->spkt_device[13] = 0;
1488 retry:
1489         rcu_read_lock();
1490         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1491         err = -ENODEV;
1492         if (dev == NULL)
1493                 goto out_unlock;
1494
1495         err = -ENETDOWN;
1496         if (!(dev->flags & IFF_UP))
1497                 goto out_unlock;
1498
1499         /*
1500          * You may not queue a frame bigger than the mtu. This is the lowest level
1501          * raw protocol and you must do your own fragmentation at this level.
1502          */
1503
1504         err = -EMSGSIZE;
1505         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1506                 goto out_unlock;
1507
1508         if (!skb) {
1509                 size_t reserved = LL_RESERVED_SPACE(dev);
1510                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1511
1512                 rcu_read_unlock();
1513                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1514                 if (skb == NULL)
1515                         return -ENOBUFS;
1516                 /* FIXME: Save some space for broken drivers that write a hard
1517                  * header at transmission time by themselves. PPP is the notable
1518                  * one here. This should really be fixed at the driver level.
1519                  */
1520                 skb_reserve(skb, reserved);
1521                 skb_reset_network_header(skb);
1522
1523                 /* Try to align data part correctly */
1524                 if (hhlen) {
1525                         skb->data -= hhlen;
1526                         skb->tail -= hhlen;
1527                         if (len < hhlen)
1528                                 skb_reset_network_header(skb);
1529                 }
1530                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1531                 if (err)
1532                         goto out_free;
1533                 goto retry;
1534         }
1535
1536         if (len > (dev->mtu + dev->hard_header_len)) {
1537                 /* Earlier code assumed this would be a VLAN pkt,
1538                  * double-check this now that we have the actual
1539                  * packet in hand.
1540                  */
1541                 struct ethhdr *ehdr;
1542                 skb_reset_mac_header(skb);
1543                 ehdr = eth_hdr(skb);
1544                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1545                         err = -EMSGSIZE;
1546                         goto out_unlock;
1547                 }
1548         }
1549
1550         skb->protocol = proto;
1551         skb->dev = dev;
1552         skb->priority = sk->sk_priority;
1553         skb->mark = sk->sk_mark;
1554         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1555         if (err < 0)
1556                 goto out_unlock;
1557
1558         dev_queue_xmit(skb);
1559         rcu_read_unlock();
1560         return len;
1561
1562 out_unlock:
1563         rcu_read_unlock();
1564 out_free:
1565         kfree_skb(skb);
1566         return err;
1567 }
1568
1569 static unsigned int run_filter(const struct sk_buff *skb,
1570                                       const struct sock *sk,
1571                                       unsigned int res)
1572 {
1573         struct sk_filter *filter;
1574
1575         rcu_read_lock();
1576         filter = rcu_dereference(sk->sk_filter);
1577         if (filter != NULL)
1578                 res = SK_RUN_FILTER(filter, skb);
1579         rcu_read_unlock();
1580
1581         return res;
1582 }
1583
1584 /*
1585  * This function makes lazy skb cloning in hope that most of packets
1586  * are discarded by BPF.
1587  *
1588  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1589  * and skb->cb are mangled. It works because (and until) packets
1590  * falling here are owned by current CPU. Output packets are cloned
1591  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1592  * sequencially, so that if we return skb to original state on exit,
1593  * we will not harm anyone.
1594  */
1595
1596 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1597                       struct packet_type *pt, struct net_device *orig_dev)
1598 {
1599         struct sock *sk;
1600         struct sockaddr_ll *sll;
1601         struct packet_sock *po;
1602         u8 *skb_head = skb->data;
1603         int skb_len = skb->len;
1604         unsigned int snaplen, res;
1605
1606         if (skb->pkt_type == PACKET_LOOPBACK)
1607                 goto drop;
1608
1609         sk = pt->af_packet_priv;
1610         po = pkt_sk(sk);
1611
1612         if (!net_eq(dev_net(dev), sock_net(sk)))
1613                 goto drop;
1614
1615         skb->dev = dev;
1616
1617         if (dev->header_ops) {
1618                 /* The device has an explicit notion of ll header,
1619                  * exported to higher levels.
1620                  *
1621                  * Otherwise, the device hides details of its frame
1622                  * structure, so that corresponding packet head is
1623                  * never delivered to user.
1624                  */
1625                 if (sk->sk_type != SOCK_DGRAM)
1626                         skb_push(skb, skb->data - skb_mac_header(skb));
1627                 else if (skb->pkt_type == PACKET_OUTGOING) {
1628                         /* Special case: outgoing packets have ll header at head */
1629                         skb_pull(skb, skb_network_offset(skb));
1630                 }
1631         }
1632
1633         snaplen = skb->len;
1634
1635         res = run_filter(skb, sk, snaplen);
1636         if (!res)
1637                 goto drop_n_restore;
1638         if (snaplen > res)
1639                 snaplen = res;
1640
1641         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1642                 goto drop_n_acct;
1643
1644         if (skb_shared(skb)) {
1645                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1646                 if (nskb == NULL)
1647                         goto drop_n_acct;
1648
1649                 if (skb_head != skb->data) {
1650                         skb->data = skb_head;
1651                         skb->len = skb_len;
1652                 }
1653                 kfree_skb(skb);
1654                 skb = nskb;
1655         }
1656
1657         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1658                      sizeof(skb->cb));
1659
1660         sll = &PACKET_SKB_CB(skb)->sa.ll;
1661         sll->sll_family = AF_PACKET;
1662         sll->sll_hatype = dev->type;
1663         sll->sll_protocol = skb->protocol;
1664         sll->sll_pkttype = skb->pkt_type;
1665         if (unlikely(po->origdev))
1666                 sll->sll_ifindex = orig_dev->ifindex;
1667         else
1668                 sll->sll_ifindex = dev->ifindex;
1669
1670         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1671
1672         PACKET_SKB_CB(skb)->origlen = skb->len;
1673
1674         if (pskb_trim(skb, snaplen))
1675                 goto drop_n_acct;
1676
1677         skb_set_owner_r(skb, sk);
1678         skb->dev = NULL;
1679         skb_dst_drop(skb);
1680
1681         /* drop conntrack reference */
1682         nf_reset(skb);
1683
1684         spin_lock(&sk->sk_receive_queue.lock);
1685         po->stats.tp_packets++;
1686         skb->dropcount = atomic_read(&sk->sk_drops);
1687         __skb_queue_tail(&sk->sk_receive_queue, skb);
1688         spin_unlock(&sk->sk_receive_queue.lock);
1689         sk->sk_data_ready(sk, skb->len);
1690         return 0;
1691
1692 drop_n_acct:
1693         spin_lock(&sk->sk_receive_queue.lock);
1694         po->stats.tp_drops++;
1695         atomic_inc(&sk->sk_drops);
1696         spin_unlock(&sk->sk_receive_queue.lock);
1697
1698 drop_n_restore:
1699         if (skb_head != skb->data && skb_shared(skb)) {
1700                 skb->data = skb_head;
1701                 skb->len = skb_len;
1702         }
1703 drop:
1704         consume_skb(skb);
1705         return 0;
1706 }
1707
1708 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1709                        struct packet_type *pt, struct net_device *orig_dev)
1710 {
1711         struct sock *sk;
1712         struct packet_sock *po;
1713         struct sockaddr_ll *sll;
1714         union {
1715                 struct tpacket_hdr *h1;
1716                 struct tpacket2_hdr *h2;
1717                 struct tpacket3_hdr *h3;
1718                 void *raw;
1719         } h;
1720         u8 *skb_head = skb->data;
1721         int skb_len = skb->len;
1722         unsigned int snaplen, res;
1723         unsigned long status = TP_STATUS_USER;
1724         unsigned short macoff, netoff, hdrlen;
1725         struct sk_buff *copy_skb = NULL;
1726         struct timeval tv;
1727         struct timespec ts;
1728         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1729
1730         if (skb->pkt_type == PACKET_LOOPBACK)
1731                 goto drop;
1732
1733         sk = pt->af_packet_priv;
1734         po = pkt_sk(sk);
1735
1736         if (!net_eq(dev_net(dev), sock_net(sk)))
1737                 goto drop;
1738
1739         if (dev->header_ops) {
1740                 if (sk->sk_type != SOCK_DGRAM)
1741                         skb_push(skb, skb->data - skb_mac_header(skb));
1742                 else if (skb->pkt_type == PACKET_OUTGOING) {
1743                         /* Special case: outgoing packets have ll header at head */
1744                         skb_pull(skb, skb_network_offset(skb));
1745                 }
1746         }
1747
1748         if (skb->ip_summed == CHECKSUM_PARTIAL)
1749                 status |= TP_STATUS_CSUMNOTREADY;
1750
1751         snaplen = skb->len;
1752
1753         res = run_filter(skb, sk, snaplen);
1754         if (!res)
1755                 goto drop_n_restore;
1756         if (snaplen > res)
1757                 snaplen = res;
1758
1759         if (sk->sk_type == SOCK_DGRAM) {
1760                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1761                                   po->tp_reserve;
1762         } else {
1763                 unsigned maclen = skb_network_offset(skb);
1764                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1765                                        (maclen < 16 ? 16 : maclen)) +
1766                         po->tp_reserve;
1767                 macoff = netoff - maclen;
1768         }
1769         if (po->tp_version <= TPACKET_V2) {
1770                 if (macoff + snaplen > po->rx_ring.frame_size) {
1771                         if (po->copy_thresh &&
1772                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1773                                 if (skb_shared(skb)) {
1774                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1775                                 } else {
1776                                         copy_skb = skb_get(skb);
1777                                         skb_head = skb->data;
1778                                 }
1779                                 if (copy_skb)
1780                                         skb_set_owner_r(copy_skb, sk);
1781                         }
1782                         snaplen = po->rx_ring.frame_size - macoff;
1783                         if ((int)snaplen < 0)
1784                                 snaplen = 0;
1785                 }
1786         } else if (unlikely(macoff + snaplen >
1787                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1788                 u32 nval;
1789
1790                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1791                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1792                             snaplen, nval, macoff);
1793                 snaplen = nval;
1794                 if (unlikely((int)snaplen < 0)) {
1795                         snaplen = 0;
1796                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1797                 }
1798         }
1799         spin_lock(&sk->sk_receive_queue.lock);
1800         h.raw = packet_current_rx_frame(po, skb,
1801                                         TP_STATUS_KERNEL, (macoff+snaplen));
1802         if (!h.raw)
1803                 goto ring_is_full;
1804         if (po->tp_version <= TPACKET_V2) {
1805                 packet_increment_rx_head(po, &po->rx_ring);
1806         /*
1807          * LOSING will be reported till you read the stats,
1808          * because it's COR - Clear On Read.
1809          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1810          * at packet level.
1811          */
1812                 if (po->stats.tp_drops)
1813                         status |= TP_STATUS_LOSING;
1814         }
1815         po->stats.tp_packets++;
1816         if (copy_skb) {
1817                 status |= TP_STATUS_COPY;
1818                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1819         }
1820         spin_unlock(&sk->sk_receive_queue.lock);
1821
1822         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1823
1824         switch (po->tp_version) {
1825         case TPACKET_V1:
1826                 h.h1->tp_len = skb->len;
1827                 h.h1->tp_snaplen = snaplen;
1828                 h.h1->tp_mac = macoff;
1829                 h.h1->tp_net = netoff;
1830                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1831                                 && shhwtstamps->syststamp.tv64)
1832                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1833                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1834                                 && shhwtstamps->hwtstamp.tv64)
1835                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1836                 else if (skb->tstamp.tv64)
1837                         tv = ktime_to_timeval(skb->tstamp);
1838                 else
1839                         do_gettimeofday(&tv);
1840                 h.h1->tp_sec = tv.tv_sec;
1841                 h.h1->tp_usec = tv.tv_usec;
1842                 hdrlen = sizeof(*h.h1);
1843                 break;
1844         case TPACKET_V2:
1845                 h.h2->tp_len = skb->len;
1846                 h.h2->tp_snaplen = snaplen;
1847                 h.h2->tp_mac = macoff;
1848                 h.h2->tp_net = netoff;
1849                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1850                                 && shhwtstamps->syststamp.tv64)
1851                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1852                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1853                                 && shhwtstamps->hwtstamp.tv64)
1854                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1855                 else if (skb->tstamp.tv64)
1856                         ts = ktime_to_timespec(skb->tstamp);
1857                 else
1858                         getnstimeofday(&ts);
1859                 h.h2->tp_sec = ts.tv_sec;
1860                 h.h2->tp_nsec = ts.tv_nsec;
1861                 if (vlan_tx_tag_present(skb)) {
1862                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1863                         status |= TP_STATUS_VLAN_VALID;
1864                 } else {
1865                         h.h2->tp_vlan_tci = 0;
1866                 }
1867                 h.h2->tp_padding = 0;
1868                 hdrlen = sizeof(*h.h2);
1869                 break;
1870         case TPACKET_V3:
1871                 /* tp_nxt_offset,vlan are already populated above.
1872                  * So DONT clear those fields here
1873                  */
1874                 h.h3->tp_status |= status;
1875                 h.h3->tp_len = skb->len;
1876                 h.h3->tp_snaplen = snaplen;
1877                 h.h3->tp_mac = macoff;
1878                 h.h3->tp_net = netoff;
1879                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1880                                 && shhwtstamps->syststamp.tv64)
1881                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1882                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1883                                 && shhwtstamps->hwtstamp.tv64)
1884                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1885                 else if (skb->tstamp.tv64)
1886                         ts = ktime_to_timespec(skb->tstamp);
1887                 else
1888                         getnstimeofday(&ts);
1889                 h.h3->tp_sec  = ts.tv_sec;
1890                 h.h3->tp_nsec = ts.tv_nsec;
1891                 hdrlen = sizeof(*h.h3);
1892                 break;
1893         default:
1894                 BUG();
1895         }
1896
1897         sll = h.raw + TPACKET_ALIGN(hdrlen);
1898         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1899         sll->sll_family = AF_PACKET;
1900         sll->sll_hatype = dev->type;
1901         sll->sll_protocol = skb->protocol;
1902         sll->sll_pkttype = skb->pkt_type;
1903         if (unlikely(po->origdev))
1904                 sll->sll_ifindex = orig_dev->ifindex;
1905         else
1906                 sll->sll_ifindex = dev->ifindex;
1907
1908         smp_mb();
1909 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1910         {
1911                 u8 *start, *end;
1912
1913                 if (po->tp_version <= TPACKET_V2) {
1914                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1915                                 + macoff + snaplen);
1916                         for (start = h.raw; start < end; start += PAGE_SIZE)
1917                                 flush_dcache_page(pgv_to_page(start));
1918                 }
1919                 smp_wmb();
1920         }
1921 #endif
1922         if (po->tp_version <= TPACKET_V2)
1923                 __packet_set_status(po, h.raw, status);
1924         else
1925                 prb_clear_blk_fill_status(&po->rx_ring);
1926
1927         sk->sk_data_ready(sk, 0);
1928
1929 drop_n_restore:
1930         if (skb_head != skb->data && skb_shared(skb)) {
1931                 skb->data = skb_head;
1932                 skb->len = skb_len;
1933         }
1934 drop:
1935         kfree_skb(skb);
1936         return 0;
1937
1938 ring_is_full:
1939         po->stats.tp_drops++;
1940         spin_unlock(&sk->sk_receive_queue.lock);
1941
1942         sk->sk_data_ready(sk, 0);
1943         kfree_skb(copy_skb);
1944         goto drop_n_restore;
1945 }
1946
1947 static void tpacket_destruct_skb(struct sk_buff *skb)
1948 {
1949         struct packet_sock *po = pkt_sk(skb->sk);
1950         void *ph;
1951
1952         if (likely(po->tx_ring.pg_vec)) {
1953                 ph = skb_shinfo(skb)->destructor_arg;
1954                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1955                 atomic_dec(&po->tx_ring.pending);
1956                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1957         }
1958
1959         sock_wfree(skb);
1960 }
1961
1962 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1963                 void *frame, struct net_device *dev, int size_max,
1964                 __be16 proto, unsigned char *addr)
1965 {
1966         union {
1967                 struct tpacket_hdr *h1;
1968                 struct tpacket2_hdr *h2;
1969                 void *raw;
1970         } ph;
1971         int to_write, offset, len, tp_len, nr_frags, len_max;
1972         struct socket *sock = po->sk.sk_socket;
1973         struct page *page;
1974         void *data;
1975         int err;
1976
1977         ph.raw = frame;
1978
1979         skb->protocol = proto;
1980         skb->dev = dev;
1981         skb->priority = po->sk.sk_priority;
1982         skb->mark = po->sk.sk_mark;
1983         skb_shinfo(skb)->destructor_arg = ph.raw;
1984
1985         switch (po->tp_version) {
1986         case TPACKET_V2:
1987                 tp_len = ph.h2->tp_len;
1988                 break;
1989         default:
1990                 tp_len = ph.h1->tp_len;
1991                 break;
1992         }
1993         if (unlikely(tp_len > size_max)) {
1994                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1995                 return -EMSGSIZE;
1996         }
1997
1998         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1999         skb_reset_network_header(skb);
2000
2001         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2002         to_write = tp_len;
2003
2004         if (sock->type == SOCK_DGRAM) {
2005                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2006                                 NULL, tp_len);
2007                 if (unlikely(err < 0))
2008                         return -EINVAL;
2009         } else if (dev->hard_header_len) {
2010                 /* net device doesn't like empty head */
2011                 if (unlikely(tp_len <= dev->hard_header_len)) {
2012                         pr_err("packet size is too short (%d < %d)\n",
2013                                tp_len, dev->hard_header_len);
2014                         return -EINVAL;
2015                 }
2016
2017                 skb_push(skb, dev->hard_header_len);
2018                 err = skb_store_bits(skb, 0, data,
2019                                 dev->hard_header_len);
2020                 if (unlikely(err))
2021                         return err;
2022
2023                 data += dev->hard_header_len;
2024                 to_write -= dev->hard_header_len;
2025         }
2026
2027         err = -EFAULT;
2028         offset = offset_in_page(data);
2029         len_max = PAGE_SIZE - offset;
2030         len = ((to_write > len_max) ? len_max : to_write);
2031
2032         skb->data_len = to_write;
2033         skb->len += to_write;
2034         skb->truesize += to_write;
2035         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2036
2037         while (likely(to_write)) {
2038                 nr_frags = skb_shinfo(skb)->nr_frags;
2039
2040                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2041                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2042                                MAX_SKB_FRAGS);
2043                         return -EFAULT;
2044                 }
2045
2046                 page = pgv_to_page(data);
2047                 data += len;
2048                 flush_dcache_page(page);
2049                 get_page(page);
2050                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2051                 to_write -= len;
2052                 offset = 0;
2053                 len_max = PAGE_SIZE;
2054                 len = ((to_write > len_max) ? len_max : to_write);
2055         }
2056
2057         return tp_len;
2058 }
2059
2060 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2061 {
2062         struct net_device *dev;
2063
2064         rcu_read_lock();
2065         dev = rcu_dereference(po->cached_dev);
2066         if (dev)
2067                 dev_hold(dev);
2068         rcu_read_unlock();
2069
2070         return dev;
2071 }
2072
2073 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2074 {
2075         struct sk_buff *skb;
2076         struct net_device *dev;
2077         __be16 proto;
2078         int err, reserve = 0;
2079         void *ph;
2080         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2081         int tp_len, size_max;
2082         unsigned char *addr;
2083         int len_sum = 0;
2084         int status = 0;
2085
2086         mutex_lock(&po->pg_vec_lock);
2087
2088         err = -EBUSY;
2089         if (saddr == NULL) {
2090                 dev     = packet_cached_dev_get(po);
2091                 proto   = po->num;
2092                 addr    = NULL;
2093         } else {
2094                 err = -EINVAL;
2095                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2096                         goto out;
2097                 if (msg->msg_namelen < (saddr->sll_halen
2098                                         + offsetof(struct sockaddr_ll,
2099                                                 sll_addr)))
2100                         goto out;
2101                 proto   = saddr->sll_protocol;
2102                 addr    = saddr->sll_addr;
2103                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2104         }
2105
2106         err = -ENXIO;
2107         if (unlikely(dev == NULL))
2108                 goto out;
2109         err = -ENETDOWN;
2110         if (unlikely(!(dev->flags & IFF_UP)))
2111                 goto out_put;
2112
2113         reserve = dev->hard_header_len;
2114
2115         size_max = po->tx_ring.frame_size
2116                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2117
2118         if (size_max > dev->mtu + reserve)
2119                 size_max = dev->mtu + reserve;
2120
2121         do {
2122                 ph = packet_current_frame(po, &po->tx_ring,
2123                                 TP_STATUS_SEND_REQUEST);
2124
2125                 if (unlikely(ph == NULL)) {
2126                         schedule();
2127                         continue;
2128                 }
2129
2130                 status = TP_STATUS_SEND_REQUEST;
2131                 skb = sock_alloc_send_skb(&po->sk,
2132                                 LL_ALLOCATED_SPACE(dev)
2133                                 + sizeof(struct sockaddr_ll),
2134                                 0, &err);
2135
2136                 if (unlikely(skb == NULL))
2137                         goto out_status;
2138
2139                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2140                                 addr);
2141
2142                 if (unlikely(tp_len < 0)) {
2143                         if (po->tp_loss) {
2144                                 __packet_set_status(po, ph,
2145                                                 TP_STATUS_AVAILABLE);
2146                                 packet_increment_head(&po->tx_ring);
2147                                 kfree_skb(skb);
2148                                 continue;
2149                         } else {
2150                                 status = TP_STATUS_WRONG_FORMAT;
2151                                 err = tp_len;
2152                                 goto out_status;
2153                         }
2154                 }
2155
2156                 skb->destructor = tpacket_destruct_skb;
2157                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2158                 atomic_inc(&po->tx_ring.pending);
2159
2160                 status = TP_STATUS_SEND_REQUEST;
2161                 err = dev_queue_xmit(skb);
2162                 if (unlikely(err > 0)) {
2163                         err = net_xmit_errno(err);
2164                         if (err && __packet_get_status(po, ph) ==
2165                                    TP_STATUS_AVAILABLE) {
2166                                 /* skb was destructed already */
2167                                 skb = NULL;
2168                                 goto out_status;
2169                         }
2170                         /*
2171                          * skb was dropped but not destructed yet;
2172                          * let's treat it like congestion or err < 0
2173                          */
2174                         err = 0;
2175                 }
2176                 packet_increment_head(&po->tx_ring);
2177                 len_sum += tp_len;
2178         } while (likely((ph != NULL) ||
2179                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2180                          (atomic_read(&po->tx_ring.pending))))
2181                 );
2182
2183         err = len_sum;
2184         goto out_put;
2185
2186 out_status:
2187         __packet_set_status(po, ph, status);
2188         kfree_skb(skb);
2189 out_put:
2190         dev_put(dev);
2191 out:
2192         mutex_unlock(&po->pg_vec_lock);
2193         return err;
2194 }
2195
2196 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2197                                         size_t reserve, size_t len,
2198                                         size_t linear, int noblock,
2199                                         int *err)
2200 {
2201         struct sk_buff *skb;
2202
2203         /* Under a page?  Don't bother with paged skb. */
2204         if (prepad + len < PAGE_SIZE || !linear)
2205                 linear = len;
2206
2207         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2208                                    err);
2209         if (!skb)
2210                 return NULL;
2211
2212         skb_reserve(skb, reserve);
2213         skb_put(skb, linear);
2214         skb->data_len = len - linear;
2215         skb->len += len - linear;
2216
2217         return skb;
2218 }
2219
2220 static int packet_snd(struct socket *sock,
2221                           struct msghdr *msg, size_t len)
2222 {
2223         struct sock *sk = sock->sk;
2224         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2225         struct sk_buff *skb;
2226         struct net_device *dev;
2227         __be16 proto;
2228         unsigned char *addr;
2229         int err, reserve = 0;
2230         struct virtio_net_hdr vnet_hdr = { 0 };
2231         int offset = 0;
2232         int vnet_hdr_len;
2233         struct packet_sock *po = pkt_sk(sk);
2234         bool has_vnet_hdr = false;
2235         unsigned short gso_type = 0;
2236
2237         /*
2238          *      Get and verify the address.
2239          */
2240
2241         if (saddr == NULL) {
2242                 dev     = packet_cached_dev_get(po);
2243                 proto   = po->num;
2244                 addr    = NULL;
2245         } else {
2246                 err = -EINVAL;
2247                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2248                         goto out;
2249                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2250                         goto out;
2251                 proto   = saddr->sll_protocol;
2252                 addr    = saddr->sll_addr;
2253                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2254         }
2255
2256         err = -ENXIO;
2257         if (unlikely(dev == NULL))
2258                 goto out_unlock;
2259         err = -ENETDOWN;
2260         if (unlikely(!(dev->flags & IFF_UP)))
2261                 goto out_unlock;
2262
2263         if (sock->type == SOCK_RAW)
2264                 reserve = dev->hard_header_len;
2265         if (po->has_vnet_hdr) {
2266                 vnet_hdr_len = sizeof(vnet_hdr);
2267                 has_vnet_hdr = true;
2268
2269                 err = -EINVAL;
2270                 if (len < vnet_hdr_len)
2271                         goto out_unlock;
2272
2273                 len -= vnet_hdr_len;
2274
2275                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2276                                        vnet_hdr_len);
2277                 if (err < 0)
2278                         goto out_unlock;
2279
2280                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2281                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2282                       vnet_hdr.hdr_len))
2283                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2284                                                  vnet_hdr.csum_offset + 2;
2285
2286                 err = -EINVAL;
2287                 if (vnet_hdr.hdr_len > len)
2288                         goto out_unlock;
2289
2290                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2291                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2292                         case VIRTIO_NET_HDR_GSO_TCPV4:
2293                                 gso_type = SKB_GSO_TCPV4;
2294                                 break;
2295                         case VIRTIO_NET_HDR_GSO_TCPV6:
2296                                 gso_type = SKB_GSO_TCPV6;
2297                                 break;
2298                         case VIRTIO_NET_HDR_GSO_UDP:
2299                                 gso_type = SKB_GSO_UDP;
2300                                 break;
2301                         default:
2302                                 goto out_unlock;
2303                         }
2304
2305                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2306                                 gso_type |= SKB_GSO_TCP_ECN;
2307
2308                         if (vnet_hdr.gso_size == 0)
2309                                 goto out_unlock;
2310
2311                 }
2312         }
2313
2314         err = -EMSGSIZE;
2315         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2316                 goto out_unlock;
2317
2318         err = -ENOBUFS;
2319         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2320                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2321                                msg->msg_flags & MSG_DONTWAIT, &err);
2322         if (skb == NULL)
2323                 goto out_unlock;
2324
2325         skb_set_network_header(skb, reserve);
2326
2327         err = -EINVAL;
2328         if (sock->type == SOCK_DGRAM &&
2329             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2330                 goto out_free;
2331
2332         /* Returns -EFAULT on error */
2333         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2334         if (err)
2335                 goto out_free;
2336         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2337         if (err < 0)
2338                 goto out_free;
2339
2340         if (!gso_type && (len > dev->mtu + reserve)) {
2341                 /* Earlier code assumed this would be a VLAN pkt,
2342                  * double-check this now that we have the actual
2343                  * packet in hand.
2344                  */
2345                 struct ethhdr *ehdr;
2346                 skb_reset_mac_header(skb);
2347                 ehdr = eth_hdr(skb);
2348                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2349                         err = -EMSGSIZE;
2350                         goto out_free;
2351                 }
2352         }
2353
2354         skb->protocol = proto;
2355         skb->dev = dev;
2356         skb->priority = sk->sk_priority;
2357         skb->mark = sk->sk_mark;
2358
2359         if (has_vnet_hdr) {
2360                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2361                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2362                                                   vnet_hdr.csum_offset)) {
2363                                 err = -EINVAL;
2364                                 goto out_free;
2365                         }
2366                 }
2367
2368                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2369                 skb_shinfo(skb)->gso_type = gso_type;
2370
2371                 /* Header must be checked, and gso_segs computed. */
2372                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2373                 skb_shinfo(skb)->gso_segs = 0;
2374
2375                 len += vnet_hdr_len;
2376         }
2377
2378         /*
2379          *      Now send it
2380          */
2381
2382         err = dev_queue_xmit(skb);
2383         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2384                 goto out_unlock;
2385
2386         dev_put(dev);
2387
2388         return len;
2389
2390 out_free:
2391         kfree_skb(skb);
2392 out_unlock:
2393         if (dev)
2394                 dev_put(dev);
2395 out:
2396         return err;
2397 }
2398
2399 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2400                 struct msghdr *msg, size_t len)
2401 {
2402         struct sock *sk = sock->sk;
2403         struct packet_sock *po = pkt_sk(sk);
2404         if (po->tx_ring.pg_vec)
2405                 return tpacket_snd(po, msg);
2406         else
2407                 return packet_snd(sock, msg, len);
2408 }
2409
2410 /*
2411  *      Close a PACKET socket. This is fairly simple. We immediately go
2412  *      to 'closed' state and remove our protocol entry in the device list.
2413  */
2414
2415 static int packet_release(struct socket *sock)
2416 {
2417         struct sock *sk = sock->sk;
2418         struct packet_sock *po;
2419         struct packet_fanout *f;
2420         struct net *net;
2421         union tpacket_req_u req_u;
2422
2423         if (!sk)
2424                 return 0;
2425
2426         net = sock_net(sk);
2427         po = pkt_sk(sk);
2428
2429         spin_lock_bh(&net->packet.sklist_lock);
2430         sk_del_node_init_rcu(sk);
2431         sock_prot_inuse_add(net, sk->sk_prot, -1);
2432         spin_unlock_bh(&net->packet.sklist_lock);
2433
2434         spin_lock(&po->bind_lock);
2435         unregister_prot_hook(sk, false);
2436         if (po->prot_hook.dev) {
2437                 dev_put(po->prot_hook.dev);
2438                 po->prot_hook.dev = NULL;
2439         }
2440         spin_unlock(&po->bind_lock);
2441
2442         packet_flush_mclist(sk);
2443
2444         if (po->rx_ring.pg_vec) {
2445                 memset(&req_u, 0, sizeof(req_u));
2446                 packet_set_ring(sk, &req_u, 1, 0);
2447         }
2448
2449         if (po->tx_ring.pg_vec) {
2450                 memset(&req_u, 0, sizeof(req_u));
2451                 packet_set_ring(sk, &req_u, 1, 1);
2452         }
2453
2454         f = fanout_release(sk);
2455
2456         synchronize_net();
2457
2458         kfree(f);
2459
2460         /*
2461          *      Now the socket is dead. No more input will appear.
2462          */
2463         sock_orphan(sk);
2464         sock->sk = NULL;
2465
2466         /* Purge queues */
2467
2468         skb_queue_purge(&sk->sk_receive_queue);
2469         sk_refcnt_debug_release(sk);
2470
2471         sock_put(sk);
2472         return 0;
2473 }
2474
2475 /*
2476  *      Attach a packet hook.
2477  */
2478
2479 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2480                           __be16 protocol)
2481 {
2482         struct packet_sock *po = pkt_sk(sk);
2483         struct net_device *dev_curr;
2484         struct net_device *dev = NULL;
2485         int ret = 0;
2486         bool unlisted = false;
2487
2488         lock_sock(sk);
2489
2490         spin_lock(&po->bind_lock);
2491         rcu_read_lock();
2492
2493         if (po->fanout) {
2494                 ret = -EINVAL;
2495                 goto out_unlock;
2496         }
2497
2498         if (name) {
2499                 dev = dev_get_by_name_rcu(sock_net(sk), name);
2500                 if (!dev) {
2501                         ret = -ENODEV;
2502                         goto out_unlock;
2503                 }
2504         } else if (ifindex) {
2505                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
2506                 if (!dev) {
2507                         ret = -ENODEV;
2508                         goto out_unlock;
2509                 }
2510         }
2511
2512         if (dev)
2513                 dev_hold(dev);
2514
2515         dev_curr = po->prot_hook.dev;
2516
2517         if (po->running) {
2518                 rcu_read_unlock();
2519                 /* prevents packet_notifier() from calling
2520                  * register_prot_hook()
2521                  */
2522                 po->num = 0;
2523                 __unregister_prot_hook(sk, true);
2524                 rcu_read_lock();
2525                 dev_curr = po->prot_hook.dev;
2526                 if (dev)
2527                         unlisted = !dev_get_by_index_rcu(sock_net(sk),
2528                                                          dev->ifindex);
2529         }
2530         BUG_ON(po->running);
2531         po->num = protocol;
2532         po->prot_hook.type = protocol;
2533
2534         if (unlikely(unlisted)) {
2535                 dev_put(dev);
2536                 po->prot_hook.dev = NULL;
2537                 po->ifindex = -1;
2538         } else {
2539                 po->prot_hook.dev = dev;
2540                 po->ifindex = dev ? dev->ifindex : 0;
2541         }
2542
2543         if (dev_curr)
2544                 dev_put(dev_curr);
2545
2546         if (protocol == 0)
2547                 goto out_unlock;
2548
2549         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
2550                 register_prot_hook(sk);
2551         } else {
2552                 sk->sk_err = ENETDOWN;
2553                 if (!sock_flag(sk, SOCK_DEAD))
2554                         sk->sk_error_report(sk);
2555         }
2556
2557 out_unlock:
2558         rcu_read_unlock();
2559         spin_unlock(&po->bind_lock);
2560         release_sock(sk);
2561         return ret;
2562 }
2563
2564 /*
2565  *      Bind a packet socket to a device
2566  */
2567
2568 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2569                             int addr_len)
2570 {
2571         struct sock *sk = sock->sk;
2572         char name[15];
2573
2574         /*
2575          *      Check legality
2576          */
2577
2578         if (addr_len != sizeof(struct sockaddr))
2579                 return -EINVAL;
2580         strlcpy(name, uaddr->sa_data, sizeof(name));
2581
2582         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
2583 }
2584
2585 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2586 {
2587         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2588         struct sock *sk = sock->sk;
2589
2590         /*
2591          *      Check legality
2592          */
2593
2594         if (addr_len < sizeof(struct sockaddr_ll))
2595                 return -EINVAL;
2596         if (sll->sll_family != AF_PACKET)
2597                 return -EINVAL;
2598
2599         return packet_do_bind(sk, NULL, sll->sll_ifindex,
2600                               sll->sll_protocol ? : pkt_sk(sk)->num);
2601 }
2602
2603 static struct proto packet_proto = {
2604         .name     = "PACKET",
2605         .owner    = THIS_MODULE,
2606         .obj_size = sizeof(struct packet_sock),
2607 };
2608
2609 /*
2610  *      Create a packet of type SOCK_PACKET.
2611  */
2612
2613 static int packet_create(struct net *net, struct socket *sock, int protocol,
2614                          int kern)
2615 {
2616         struct sock *sk;
2617         struct packet_sock *po;
2618         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2619         int err;
2620
2621         if (!capable(CAP_NET_RAW))
2622                 return -EPERM;
2623         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2624             sock->type != SOCK_PACKET)
2625                 return -ESOCKTNOSUPPORT;
2626
2627         sock->state = SS_UNCONNECTED;
2628
2629         err = -ENOBUFS;
2630         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2631         if (sk == NULL)
2632                 goto out;
2633
2634         sock->ops = &packet_ops;
2635         if (sock->type == SOCK_PACKET)
2636                 sock->ops = &packet_ops_spkt;
2637
2638         sock_init_data(sock, sk);
2639
2640         po = pkt_sk(sk);
2641         sk->sk_family = PF_PACKET;
2642         po->num = proto;
2643         RCU_INIT_POINTER(po->cached_dev, NULL);
2644
2645         sk->sk_destruct = packet_sock_destruct;
2646         sk_refcnt_debug_inc(sk);
2647
2648         /*
2649          *      Attach a protocol block
2650          */
2651
2652         spin_lock_init(&po->bind_lock);
2653         mutex_init(&po->pg_vec_lock);
2654         po->prot_hook.func = packet_rcv;
2655
2656         if (sock->type == SOCK_PACKET)
2657                 po->prot_hook.func = packet_rcv_spkt;
2658
2659         po->prot_hook.af_packet_priv = sk;
2660
2661         if (proto) {
2662                 po->prot_hook.type = proto;
2663                 register_prot_hook(sk);
2664         }
2665
2666         spin_lock_bh(&net->packet.sklist_lock);
2667         sk_add_node_rcu(sk, &net->packet.sklist);
2668         sock_prot_inuse_add(net, &packet_proto, 1);
2669         spin_unlock_bh(&net->packet.sklist_lock);
2670
2671         return 0;
2672 out:
2673         return err;
2674 }
2675
2676 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2677 {
2678         struct sock_exterr_skb *serr;
2679         struct sk_buff *skb, *skb2;
2680         int copied, err;
2681
2682         err = -EAGAIN;
2683         skb = skb_dequeue(&sk->sk_error_queue);
2684         if (skb == NULL)
2685                 goto out;
2686
2687         copied = skb->len;
2688         if (copied > len) {
2689                 msg->msg_flags |= MSG_TRUNC;
2690                 copied = len;
2691         }
2692         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2693         if (err)
2694                 goto out_free_skb;
2695
2696         sock_recv_timestamp(msg, sk, skb);
2697
2698         serr = SKB_EXT_ERR(skb);
2699         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2700                  sizeof(serr->ee), &serr->ee);
2701
2702         msg->msg_flags |= MSG_ERRQUEUE;
2703         err = copied;
2704
2705         /* Reset and regenerate socket error */
2706         spin_lock_bh(&sk->sk_error_queue.lock);
2707         sk->sk_err = 0;
2708         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2709                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2710                 spin_unlock_bh(&sk->sk_error_queue.lock);
2711                 sk->sk_error_report(sk);
2712         } else
2713                 spin_unlock_bh(&sk->sk_error_queue.lock);
2714
2715 out_free_skb:
2716         kfree_skb(skb);
2717 out:
2718         return err;
2719 }
2720
2721 /*
2722  *      Pull a packet from our receive queue and hand it to the user.
2723  *      If necessary we block.
2724  */
2725
2726 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2727                           struct msghdr *msg, size_t len, int flags)
2728 {
2729         struct sock *sk = sock->sk;
2730         struct sk_buff *skb;
2731         int copied, err;
2732         int vnet_hdr_len = 0;
2733
2734         err = -EINVAL;
2735         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2736                 goto out;
2737
2738 #if 0
2739         /* What error should we return now? EUNATTACH? */
2740         if (pkt_sk(sk)->ifindex < 0)
2741                 return -ENODEV;
2742 #endif
2743
2744         if (flags & MSG_ERRQUEUE) {
2745                 err = packet_recv_error(sk, msg, len);
2746                 goto out;
2747         }
2748
2749         /*
2750          *      Call the generic datagram receiver. This handles all sorts
2751          *      of horrible races and re-entrancy so we can forget about it
2752          *      in the protocol layers.
2753          *
2754          *      Now it will return ENETDOWN, if device have just gone down,
2755          *      but then it will block.
2756          */
2757
2758         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2759
2760         /*
2761          *      An error occurred so return it. Because skb_recv_datagram()
2762          *      handles the blocking we don't see and worry about blocking
2763          *      retries.
2764          */
2765
2766         if (skb == NULL)
2767                 goto out;
2768
2769         if (pkt_sk(sk)->has_vnet_hdr) {
2770                 struct virtio_net_hdr vnet_hdr = { 0 };
2771
2772                 err = -EINVAL;
2773                 vnet_hdr_len = sizeof(vnet_hdr);
2774                 if (len < vnet_hdr_len)
2775                         goto out_free;
2776
2777                 len -= vnet_hdr_len;
2778
2779                 if (skb_is_gso(skb)) {
2780                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2781
2782                         /* This is a hint as to how much should be linear. */
2783                         vnet_hdr.hdr_len = skb_headlen(skb);
2784                         vnet_hdr.gso_size = sinfo->gso_size;
2785                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2786                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2787                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2788                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2789                         else if (sinfo->gso_type & SKB_GSO_UDP)
2790                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2791                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2792                                 goto out_free;
2793                         else
2794                                 BUG();
2795                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2796                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2797                 } else
2798                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2799
2800                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2801                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2802                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2803                         vnet_hdr.csum_offset = skb->csum_offset;
2804                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2805                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2806                 } /* else everything is zero */
2807
2808                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2809                                      vnet_hdr_len);
2810                 if (err < 0)
2811                         goto out_free;
2812         }
2813
2814         /* You lose any data beyond the buffer you gave. If it worries
2815          * a user program they can ask the device for its MTU
2816          * anyway.
2817          */
2818         copied = skb->len;
2819         if (copied > len) {
2820                 copied = len;
2821                 msg->msg_flags |= MSG_TRUNC;
2822         }
2823
2824         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2825         if (err)
2826                 goto out_free;
2827
2828         sock_recv_ts_and_drops(msg, sk, skb);
2829
2830         if (msg->msg_name) {
2831                 /* If the address length field is there to be filled
2832                  * in, we fill it in now.
2833                  */
2834                 if (sock->type == SOCK_PACKET) {
2835                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2836                 } else {
2837                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2838                         msg->msg_namelen = sll->sll_halen +
2839                                 offsetof(struct sockaddr_ll, sll_addr);
2840                 }
2841                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2842                        msg->msg_namelen);
2843         }
2844
2845         if (pkt_sk(sk)->auxdata) {
2846                 struct tpacket_auxdata aux;
2847
2848                 aux.tp_status = TP_STATUS_USER;
2849                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2850                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2851                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2852                 aux.tp_snaplen = skb->len;
2853                 aux.tp_mac = 0;
2854                 aux.tp_net = skb_network_offset(skb);
2855                 if (vlan_tx_tag_present(skb)) {
2856                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2857                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2858                 } else {
2859                         aux.tp_vlan_tci = 0;
2860                 }
2861                 aux.tp_padding = 0;
2862                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2863         }
2864
2865         /*
2866          *      Free or return the buffer as appropriate. Again this
2867          *      hides all the races and re-entrancy issues from us.
2868          */
2869         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2870
2871 out_free:
2872         skb_free_datagram(sk, skb);
2873 out:
2874         return err;
2875 }
2876
2877 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2878                                int *uaddr_len, int peer)
2879 {
2880         struct net_device *dev;
2881         struct sock *sk = sock->sk;
2882
2883         if (peer)
2884                 return -EOPNOTSUPP;
2885
2886         uaddr->sa_family = AF_PACKET;
2887         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2888         rcu_read_lock();
2889         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2890         if (dev)
2891                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2892         rcu_read_unlock();
2893         *uaddr_len = sizeof(*uaddr);
2894
2895         return 0;
2896 }
2897
2898 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2899                           int *uaddr_len, int peer)
2900 {
2901         struct net_device *dev;
2902         struct sock *sk = sock->sk;
2903         struct packet_sock *po = pkt_sk(sk);
2904         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2905
2906         if (peer)
2907                 return -EOPNOTSUPP;
2908
2909         sll->sll_family = AF_PACKET;
2910         sll->sll_ifindex = po->ifindex;
2911         sll->sll_protocol = po->num;
2912         sll->sll_pkttype = 0;
2913         rcu_read_lock();
2914         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2915         if (dev) {
2916                 sll->sll_hatype = dev->type;
2917                 sll->sll_halen = dev->addr_len;
2918                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2919         } else {
2920                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2921                 sll->sll_halen = 0;
2922         }
2923         rcu_read_unlock();
2924         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2925
2926         return 0;
2927 }
2928
2929 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2930                          int what)
2931 {
2932         switch (i->type) {
2933         case PACKET_MR_MULTICAST:
2934                 if (i->alen != dev->addr_len)
2935                         return -EINVAL;
2936                 if (what > 0)
2937                         return dev_mc_add(dev, i->addr);
2938                 else
2939                         return dev_mc_del(dev, i->addr);
2940                 break;
2941         case PACKET_MR_PROMISC:
2942                 return dev_set_promiscuity(dev, what);
2943                 break;
2944         case PACKET_MR_ALLMULTI:
2945                 return dev_set_allmulti(dev, what);
2946                 break;
2947         case PACKET_MR_UNICAST:
2948                 if (i->alen != dev->addr_len)
2949                         return -EINVAL;
2950                 if (what > 0)
2951                         return dev_uc_add(dev, i->addr);
2952                 else
2953                         return dev_uc_del(dev, i->addr);
2954                 break;
2955         default:
2956                 break;
2957         }
2958         return 0;
2959 }
2960
2961 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2962 {
2963         for ( ; i; i = i->next) {
2964                 if (i->ifindex == dev->ifindex)
2965                         packet_dev_mc(dev, i, what);
2966         }
2967 }
2968
2969 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2970 {
2971         struct packet_sock *po = pkt_sk(sk);
2972         struct packet_mclist *ml, *i;
2973         struct net_device *dev;
2974         int err;
2975
2976         rtnl_lock();
2977
2978         err = -ENODEV;
2979         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2980         if (!dev)
2981                 goto done;
2982
2983         err = -EINVAL;
2984         if (mreq->mr_alen > dev->addr_len)
2985                 goto done;
2986
2987         err = -ENOBUFS;
2988         i = kmalloc(sizeof(*i), GFP_KERNEL);
2989         if (i == NULL)
2990                 goto done;
2991
2992         err = 0;
2993         for (ml = po->mclist; ml; ml = ml->next) {
2994                 if (ml->ifindex == mreq->mr_ifindex &&
2995                     ml->type == mreq->mr_type &&
2996                     ml->alen == mreq->mr_alen &&
2997                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2998                         ml->count++;
2999                         /* Free the new element ... */
3000                         kfree(i);
3001                         goto done;
3002                 }
3003         }
3004
3005         i->type = mreq->mr_type;
3006         i->ifindex = mreq->mr_ifindex;
3007         i->alen = mreq->mr_alen;
3008         memcpy(i->addr, mreq->mr_address, i->alen);
3009         i->count = 1;
3010         i->next = po->mclist;
3011         po->mclist = i;
3012         err = packet_dev_mc(dev, i, 1);
3013         if (err) {
3014                 po->mclist = i->next;
3015                 kfree(i);
3016         }
3017
3018 done:
3019         rtnl_unlock();
3020         return err;
3021 }
3022
3023 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3024 {
3025         struct packet_mclist *ml, **mlp;
3026
3027         rtnl_lock();
3028
3029         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3030                 if (ml->ifindex == mreq->mr_ifindex &&
3031                     ml->type == mreq->mr_type &&
3032                     ml->alen == mreq->mr_alen &&
3033                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3034                         if (--ml->count == 0) {
3035                                 struct net_device *dev;
3036                                 *mlp = ml->next;
3037                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3038                                 if (dev)
3039                                         packet_dev_mc(dev, ml, -1);
3040                                 kfree(ml);
3041                         }
3042                         rtnl_unlock();
3043                         return 0;
3044                 }
3045         }
3046         rtnl_unlock();
3047         return -EADDRNOTAVAIL;
3048 }
3049
3050 static void packet_flush_mclist(struct sock *sk)
3051 {
3052         struct packet_sock *po = pkt_sk(sk);
3053         struct packet_mclist *ml;
3054
3055         if (!po->mclist)
3056                 return;
3057
3058         rtnl_lock();
3059         while ((ml = po->mclist) != NULL) {
3060                 struct net_device *dev;
3061
3062                 po->mclist = ml->next;
3063                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3064                 if (dev != NULL)
3065                         packet_dev_mc(dev, ml, -1);
3066                 kfree(ml);
3067         }
3068         rtnl_unlock();
3069 }
3070
3071 static int
3072 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3073 {
3074         struct sock *sk = sock->sk;
3075         struct packet_sock *po = pkt_sk(sk);
3076         int ret;
3077
3078         if (level != SOL_PACKET)
3079                 return -ENOPROTOOPT;
3080
3081         switch (optname) {
3082         case PACKET_ADD_MEMBERSHIP:
3083         case PACKET_DROP_MEMBERSHIP:
3084         {
3085                 struct packet_mreq_max mreq;
3086                 int len = optlen;
3087                 memset(&mreq, 0, sizeof(mreq));
3088                 if (len < sizeof(struct packet_mreq))
3089                         return -EINVAL;
3090                 if (len > sizeof(mreq))
3091                         len = sizeof(mreq);
3092                 if (copy_from_user(&mreq, optval, len))
3093                         return -EFAULT;
3094                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3095                         return -EINVAL;
3096                 if (optname == PACKET_ADD_MEMBERSHIP)
3097                         ret = packet_mc_add(sk, &mreq);
3098                 else
3099                         ret = packet_mc_drop(sk, &mreq);
3100                 return ret;
3101         }
3102
3103         case PACKET_RX_RING:
3104         case PACKET_TX_RING:
3105         {
3106                 union tpacket_req_u req_u;
3107                 int len;
3108
3109                 switch (po->tp_version) {
3110                 case TPACKET_V1:
3111                 case TPACKET_V2:
3112                         len = sizeof(req_u.req);
3113                         break;
3114                 case TPACKET_V3:
3115                 default:
3116                         len = sizeof(req_u.req3);
3117                         break;
3118                 }
3119                 if (optlen < len)
3120                         return -EINVAL;
3121                 if (pkt_sk(sk)->has_vnet_hdr)
3122                         return -EINVAL;
3123                 if (copy_from_user(&req_u.req, optval, len))
3124                         return -EFAULT;
3125                 return packet_set_ring(sk, &req_u, 0,
3126                         optname == PACKET_TX_RING);
3127         }
3128         case PACKET_COPY_THRESH:
3129         {
3130                 int val;
3131
3132                 if (optlen != sizeof(val))
3133                         return -EINVAL;
3134                 if (copy_from_user(&val, optval, sizeof(val)))
3135                         return -EFAULT;
3136
3137                 pkt_sk(sk)->copy_thresh = val;
3138                 return 0;
3139         }
3140         case PACKET_VERSION:
3141         {
3142                 int val;
3143
3144                 if (optlen != sizeof(val))
3145                         return -EINVAL;
3146                 if (copy_from_user(&val, optval, sizeof(val)))
3147                         return -EFAULT;
3148                 switch (val) {
3149                 case TPACKET_V1:
3150                 case TPACKET_V2:
3151                 case TPACKET_V3:
3152                         break;
3153                 default:
3154                         return -EINVAL;
3155                 }
3156                 lock_sock(sk);
3157                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3158                         ret = -EBUSY;
3159                 } else {
3160                         po->tp_version = val;
3161                         ret = 0;
3162                 }
3163                 release_sock(sk);
3164                 return ret;
3165         }
3166         case PACKET_RESERVE:
3167         {
3168                 unsigned int val;
3169
3170                 if (optlen != sizeof(val))
3171                         return -EINVAL;
3172                 if (copy_from_user(&val, optval, sizeof(val)))
3173                         return -EFAULT;
3174                 if (val > INT_MAX)
3175                         return -EINVAL;
3176                 lock_sock(sk);
3177                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3178                         ret = -EBUSY;
3179                 } else {
3180                         po->tp_reserve = val;
3181                         ret = 0;
3182                 }