packet: fix races in fanout_add()
[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    knum_blocks;
199         uint64_t        knxt_seq_num;
200         char            *prev;
201         char            *nxt_offset;
202         struct sk_buff  *skb;
203
204         atomic_t        blk_fill_in_prog;
205
206         /* Default is set to 8ms */
207 #define DEFAULT_PRB_RETIRE_TOV  (8)
208
209         unsigned short  retire_blk_tov;
210         unsigned short  version;
211         unsigned long   tov_in_jiffies;
212
213         /* timer to retire an outstanding block */
214         struct timer_list retire_blk_timer;
215 };
216
217 #define PGV_FROM_VMALLOC 1
218 struct pgv {
219         char *buffer;
220 };
221
222 struct packet_ring_buffer {
223         struct pgv              *pg_vec;
224         unsigned int            head;
225         unsigned int            frames_per_block;
226         unsigned int            frame_size;
227         unsigned int            frame_max;
228
229         unsigned int            pg_vec_order;
230         unsigned int            pg_vec_pages;
231         unsigned int            pg_vec_len;
232
233         struct tpacket_kbdq_core        prb_bdqc;
234         atomic_t                pending;
235 };
236
237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
238 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
239 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
240 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
241 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
243 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
244
245 struct packet_sock;
246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
247
248 static void *packet_previous_frame(struct packet_sock *po,
249                 struct packet_ring_buffer *rb,
250                 int status);
251 static void packet_increment_head(struct packet_ring_buffer *buff);
252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
253                         struct tpacket_block_desc *);
254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
255                         struct packet_sock *);
256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
257                 struct packet_sock *, unsigned int status);
258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
259 static void prb_open_block(struct tpacket_kbdq_core *,
260                 struct tpacket_block_desc *);
261 static void prb_retire_rx_blk_timer_expired(unsigned long);
262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
263 static void prb_init_blk_timer(struct packet_sock *,
264                 struct tpacket_kbdq_core *,
265                 void (*func) (unsigned long));
266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
268                 struct tpacket3_hdr *);
269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
270                 struct tpacket3_hdr *);
271 static void packet_flush_mclist(struct sock *sk);
272
273 struct packet_fanout;
274 struct packet_sock {
275         /* struct sock has to be the first member of packet_sock */
276         struct sock             sk;
277         struct packet_fanout    *fanout;
278         struct tpacket_stats    stats;
279         union  tpacket_stats_u  stats_u;
280         struct packet_ring_buffer       rx_ring;
281         struct packet_ring_buffer       tx_ring;
282         int                     copy_thresh;
283         spinlock_t              bind_lock;
284         struct mutex            pg_vec_lock;
285         unsigned int            running:1,      /* prot_hook is attached*/
286                                 auxdata:1,
287                                 origdev:1,
288                                 has_vnet_hdr:1;
289         int                     ifindex;        /* bound device         */
290         __be16                  num;
291         struct packet_mclist    *mclist;
292         atomic_t                mapped;
293         enum tpacket_versions   tp_version;
294         unsigned int            tp_hdrlen;
295         unsigned int            tp_reserve;
296         unsigned int            tp_loss:1;
297         unsigned int            tp_tstamp;
298         struct net_device __rcu *cached_dev;
299         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
300 };
301
302 #define PACKET_FANOUT_MAX       256
303
304 struct packet_fanout {
305 #ifdef CONFIG_NET_NS
306         struct net              *net;
307 #endif
308         unsigned int            num_members;
309         u16                     id;
310         u8                      type;
311         u8                      defrag;
312         atomic_t                rr_cur;
313         struct list_head        list;
314         struct sock             *arr[PACKET_FANOUT_MAX];
315         spinlock_t              lock;
316         atomic_t                sk_ref;
317         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
318 };
319
320 struct packet_skb_cb {
321         unsigned int origlen;
322         union {
323                 struct sockaddr_pkt pkt;
324                 struct sockaddr_ll ll;
325         } sa;
326 };
327
328 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
329
330 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
331 #define GET_PBLOCK_DESC(x, bid) \
332         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
333 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
334         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
335 #define GET_NEXT_PRB_BLK_NUM(x) \
336         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
337         ((x)->kactive_blk_num+1) : 0)
338
339 static struct packet_sock *pkt_sk(struct sock *sk)
340 {
341         return (struct packet_sock *)sk;
342 }
343
344 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
345 static void __fanout_link(struct sock *sk, struct packet_sock *po);
346
347 /* register_prot_hook must be invoked with the po->bind_lock held,
348  * or from a context in which asynchronous accesses to the packet
349  * socket is not possible (packet_create()).
350  */
351 static void register_prot_hook(struct sock *sk)
352 {
353         struct packet_sock *po = pkt_sk(sk);
354
355         if (!po->running) {
356                 if (po->fanout) {
357                         __fanout_link(sk, po);
358                 } else {
359                         dev_add_pack(&po->prot_hook);
360                         rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
361                 }
362
363                 sock_hold(sk);
364                 po->running = 1;
365         }
366 }
367
368 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
369  * held.   If the sync parameter is true, we will temporarily drop
370  * the po->bind_lock and do a synchronize_net to make sure no
371  * asynchronous packet processing paths still refer to the elements
372  * of po->prot_hook.  If the sync parameter is false, it is the
373  * callers responsibility to take care of this.
374  */
375 static void __unregister_prot_hook(struct sock *sk, bool sync)
376 {
377         struct packet_sock *po = pkt_sk(sk);
378
379         po->running = 0;
380         if (po->fanout) {
381                 __fanout_unlink(sk, po);
382         } else {
383                 __dev_remove_pack(&po->prot_hook);
384                 RCU_INIT_POINTER(po->cached_dev, NULL);
385         }
386
387         __sock_put(sk);
388
389         if (sync) {
390                 spin_unlock(&po->bind_lock);
391                 synchronize_net();
392                 spin_lock(&po->bind_lock);
393         }
394 }
395
396 static void unregister_prot_hook(struct sock *sk, bool sync)
397 {
398         struct packet_sock *po = pkt_sk(sk);
399
400         if (po->running)
401                 __unregister_prot_hook(sk, sync);
402 }
403
404 static inline __pure struct page *pgv_to_page(void *addr)
405 {
406         if (is_vmalloc_addr(addr))
407                 return vmalloc_to_page(addr);
408         return virt_to_page(addr);
409 }
410
411 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
412 {
413         union {
414                 struct tpacket_hdr *h1;
415                 struct tpacket2_hdr *h2;
416                 void *raw;
417         } h;
418
419         h.raw = frame;
420         switch (po->tp_version) {
421         case TPACKET_V1:
422                 h.h1->tp_status = status;
423                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
424                 break;
425         case TPACKET_V2:
426                 h.h2->tp_status = status;
427                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
428                 break;
429         case TPACKET_V3:
430         default:
431                 WARN(1, "TPACKET version not supported.\n");
432                 BUG();
433         }
434
435         smp_wmb();
436 }
437
438 static int __packet_get_status(struct packet_sock *po, void *frame)
439 {
440         union {
441                 struct tpacket_hdr *h1;
442                 struct tpacket2_hdr *h2;
443                 void *raw;
444         } h;
445
446         smp_rmb();
447
448         h.raw = frame;
449         switch (po->tp_version) {
450         case TPACKET_V1:
451                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
452                 return h.h1->tp_status;
453         case TPACKET_V2:
454                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
455                 return h.h2->tp_status;
456         case TPACKET_V3:
457         default:
458                 WARN(1, "TPACKET version not supported.\n");
459                 BUG();
460                 return 0;
461         }
462 }
463
464 static void *packet_lookup_frame(struct packet_sock *po,
465                 struct packet_ring_buffer *rb,
466                 unsigned int position,
467                 int status)
468 {
469         unsigned int pg_vec_pos, frame_offset;
470         union {
471                 struct tpacket_hdr *h1;
472                 struct tpacket2_hdr *h2;
473                 void *raw;
474         } h;
475
476         pg_vec_pos = position / rb->frames_per_block;
477         frame_offset = position % rb->frames_per_block;
478
479         h.raw = rb->pg_vec[pg_vec_pos].buffer +
480                 (frame_offset * rb->frame_size);
481
482         if (status != __packet_get_status(po, h.raw))
483                 return NULL;
484
485         return h.raw;
486 }
487
488 static void *packet_current_frame(struct packet_sock *po,
489                 struct packet_ring_buffer *rb,
490                 int status)
491 {
492         return packet_lookup_frame(po, rb, rb->head, status);
493 }
494
495 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
496 {
497         del_timer_sync(&pkc->retire_blk_timer);
498 }
499
500 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
501                 int tx_ring,
502                 struct sk_buff_head *rb_queue)
503 {
504         struct tpacket_kbdq_core *pkc;
505
506         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
507
508         spin_lock_bh(&rb_queue->lock);
509         pkc->delete_blk_timer = 1;
510         spin_unlock_bh(&rb_queue->lock);
511
512         prb_del_retire_blk_timer(pkc);
513 }
514
515 static void prb_init_blk_timer(struct packet_sock *po,
516                 struct tpacket_kbdq_core *pkc,
517                 void (*func) (unsigned long))
518 {
519         init_timer(&pkc->retire_blk_timer);
520         pkc->retire_blk_timer.data = (long)po;
521         pkc->retire_blk_timer.function = func;
522         pkc->retire_blk_timer.expires = jiffies;
523 }
524
525 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
526 {
527         struct tpacket_kbdq_core *pkc;
528
529         if (tx_ring)
530                 BUG();
531
532         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
533         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
534 }
535
536 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
537                                 int blk_size_in_bytes)
538 {
539         struct net_device *dev;
540         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
541         struct ethtool_cmd ecmd;
542         int err;
543
544         rtnl_lock();
545         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
546         if (unlikely(!dev)) {
547                 rtnl_unlock();
548                 return DEFAULT_PRB_RETIRE_TOV;
549         }
550         err = __ethtool_get_settings(dev, &ecmd);
551         rtnl_unlock();
552         if (!err) {
553                 switch (ecmd.speed) {
554                 case SPEED_10000:
555                         msec = 1;
556                         div = 10000/1000;
557                         break;
558                 case SPEED_1000:
559                         msec = 1;
560                         div = 1000/1000;
561                         break;
562                 /*
563                  * If the link speed is so slow you don't really
564                  * need to worry about perf anyways
565                  */
566                 case SPEED_100:
567                 case SPEED_10:
568                 default:
569                         return DEFAULT_PRB_RETIRE_TOV;
570                 }
571         }
572
573         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
574
575         if (div)
576                 mbits /= div;
577
578         tmo = mbits * msec;
579
580         if (div)
581                 return tmo+1;
582         return tmo;
583 }
584
585 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
586                         union tpacket_req_u *req_u)
587 {
588         p1->feature_req_word = req_u->req3.tp_feature_req_word;
589 }
590
591 static void init_prb_bdqc(struct packet_sock *po,
592                         struct packet_ring_buffer *rb,
593                         struct pgv *pg_vec,
594                         union tpacket_req_u *req_u, int tx_ring)
595 {
596         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
597         struct tpacket_block_desc *pbd;
598
599         memset(p1, 0x0, sizeof(*p1));
600
601         p1->knxt_seq_num = 1;
602         p1->pkbdq = pg_vec;
603         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
604         p1->pkblk_start = (char *)pg_vec[0].buffer;
605         p1->kblk_size = req_u->req3.tp_block_size;
606         p1->knum_blocks = req_u->req3.tp_block_nr;
607         p1->hdrlen = po->tp_hdrlen;
608         p1->version = po->tp_version;
609         p1->last_kactive_blk_num = 0;
610         po->stats_u.stats3.tp_freeze_q_cnt = 0;
611         if (req_u->req3.tp_retire_blk_tov)
612                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
613         else
614                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
615                                                 req_u->req3.tp_block_size);
616         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
617         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
618
619         prb_init_ft_ops(p1, req_u);
620         prb_setup_retire_blk_timer(po, tx_ring);
621         prb_open_block(p1, pbd);
622 }
623
624 /*  Do NOT update the last_blk_num first.
625  *  Assumes sk_buff_head lock is held.
626  */
627 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
628 {
629         mod_timer(&pkc->retire_blk_timer,
630                         jiffies + pkc->tov_in_jiffies);
631         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
632 }
633
634 /*
635  * Timer logic:
636  * 1) We refresh the timer only when we open a block.
637  *    By doing this we don't waste cycles refreshing the timer
638  *        on packet-by-packet basis.
639  *
640  * With a 1MB block-size, on a 1Gbps line, it will take
641  * i) ~8 ms to fill a block + ii) memcpy etc.
642  * In this cut we are not accounting for the memcpy time.
643  *
644  * So, if the user sets the 'tmo' to 10ms then the timer
645  * will never fire while the block is still getting filled
646  * (which is what we want). However, the user could choose
647  * to close a block early and that's fine.
648  *
649  * But when the timer does fire, we check whether or not to refresh it.
650  * Since the tmo granularity is in msecs, it is not too expensive
651  * to refresh the timer, lets say every '8' msecs.
652  * Either the user can set the 'tmo' or we can derive it based on
653  * a) line-speed and b) block-size.
654  * prb_calc_retire_blk_tmo() calculates the tmo.
655  *
656  */
657 static void prb_retire_rx_blk_timer_expired(unsigned long data)
658 {
659         struct packet_sock *po = (struct packet_sock *)data;
660         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
661         unsigned int frozen;
662         struct tpacket_block_desc *pbd;
663
664         spin_lock(&po->sk.sk_receive_queue.lock);
665
666         frozen = prb_queue_frozen(pkc);
667         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
668
669         if (unlikely(pkc->delete_blk_timer))
670                 goto out;
671
672         /* We only need to plug the race when the block is partially filled.
673          * tpacket_rcv:
674          *              lock(); increment BLOCK_NUM_PKTS; unlock()
675          *              copy_bits() is in progress ...
676          *              timer fires on other cpu:
677          *              we can't retire the current block because copy_bits
678          *              is in progress.
679          *
680          */
681         if (BLOCK_NUM_PKTS(pbd)) {
682                 while (atomic_read(&pkc->blk_fill_in_prog)) {
683                         /* Waiting for skb_copy_bits to finish... */
684                         cpu_relax();
685                 }
686         }
687
688         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
689                 if (!frozen) {
690                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
691                         if (!prb_dispatch_next_block(pkc, po))
692                                 goto refresh_timer;
693                         else
694                                 goto out;
695                 } else {
696                         /* Case 1. Queue was frozen because user-space was
697                          *         lagging behind.
698                          */
699                         if (prb_curr_blk_in_use(pkc, pbd)) {
700                                 /*
701                                  * Ok, user-space is still behind.
702                                  * So just refresh the timer.
703                                  */
704                                 goto refresh_timer;
705                         } else {
706                                /* Case 2. queue was frozen,user-space caught up,
707                                 * now the link went idle && the timer fired.
708                                 * We don't have a block to close.So we open this
709                                 * block and restart the timer.
710                                 * opening a block thaws the queue,restarts timer
711                                 * Thawing/timer-refresh is a side effect.
712                                 */
713                                 prb_open_block(pkc, pbd);
714                                 goto out;
715                         }
716                 }
717         }
718
719 refresh_timer:
720         _prb_refresh_rx_retire_blk_timer(pkc);
721
722 out:
723         spin_unlock(&po->sk.sk_receive_queue.lock);
724 }
725
726 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
727                 struct tpacket_block_desc *pbd1, __u32 status)
728 {
729         /* Flush everything minus the block header */
730
731 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
732         u8 *start, *end;
733
734         start = (u8 *)pbd1;
735
736         /* Skip the block header(we know header WILL fit in 4K) */
737         start += PAGE_SIZE;
738
739         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
740         for (; start < end; start += PAGE_SIZE)
741                 flush_dcache_page(pgv_to_page(start));
742
743         smp_wmb();
744 #endif
745
746         /* Now update the block status. */
747
748         BLOCK_STATUS(pbd1) = status;
749
750         /* Flush the block header */
751
752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
753         start = (u8 *)pbd1;
754         flush_dcache_page(pgv_to_page(start));
755
756         smp_wmb();
757 #endif
758 }
759
760 /*
761  * Side effect:
762  *
763  * 1) flush the block
764  * 2) Increment active_blk_num
765  *
766  * Note:We DONT refresh the timer on purpose.
767  *      Because almost always the next block will be opened.
768  */
769 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
770                 struct tpacket_block_desc *pbd1,
771                 struct packet_sock *po, unsigned int stat)
772 {
773         __u32 status = TP_STATUS_USER | stat;
774
775         struct tpacket3_hdr *last_pkt;
776         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
777
778         if (po->stats.tp_drops)
779                 status |= TP_STATUS_LOSING;
780
781         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
782         last_pkt->tp_next_offset = 0;
783
784         /* Get the ts of the last pkt */
785         if (BLOCK_NUM_PKTS(pbd1)) {
786                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
787                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
788         } else {
789                 /* Ok, we tmo'd - so get the current time */
790                 struct timespec ts;
791                 getnstimeofday(&ts);
792                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
793                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
794         }
795
796         smp_wmb();
797
798         /* Flush the block */
799         prb_flush_block(pkc1, pbd1, status);
800
801         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
802 }
803
804 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
805 {
806         pkc->reset_pending_on_curr_blk = 0;
807 }
808
809 /*
810  * Side effect of opening a block:
811  *
812  * 1) prb_queue is thawed.
813  * 2) retire_blk_timer is refreshed.
814  *
815  */
816 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
817         struct tpacket_block_desc *pbd1)
818 {
819         struct timespec ts;
820         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
821
822         smp_rmb();
823
824         /* We could have just memset this but we will lose the
825          * flexibility of making the priv area sticky
826          */
827         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
828         BLOCK_NUM_PKTS(pbd1) = 0;
829         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830         getnstimeofday(&ts);
831         h1->ts_first_pkt.ts_sec = ts.tv_sec;
832         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
833         pkc1->pkblk_start = (char *)pbd1;
834         pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
835                                     BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
836         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
838         pbd1->version = pkc1->version;
839         pkc1->prev = pkc1->nxt_offset;
840         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
841         prb_thaw_queue(pkc1);
842         _prb_refresh_rx_retire_blk_timer(pkc1);
843
844         smp_wmb();
845 }
846
847 /*
848  * Queue freeze logic:
849  * 1) Assume tp_block_nr = 8 blocks.
850  * 2) At time 't0', user opens Rx ring.
851  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
852  * 4) user-space is either sleeping or processing block '0'.
853  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
854  *    it will close block-7,loop around and try to fill block '0'.
855  *    call-flow:
856  *    __packet_lookup_frame_in_block
857  *      prb_retire_current_block()
858  *      prb_dispatch_next_block()
859  *        |->(BLOCK_STATUS == USER) evaluates to true
860  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
861  * 6) Now there are two cases:
862  *    6.1) Link goes idle right after the queue is frozen.
863  *         But remember, the last open_block() refreshed the timer.
864  *         When this timer expires,it will refresh itself so that we can
865  *         re-open block-0 in near future.
866  *    6.2) Link is busy and keeps on receiving packets. This is a simple
867  *         case and __packet_lookup_frame_in_block will check if block-0
868  *         is free and can now be re-used.
869  */
870 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
871                                   struct packet_sock *po)
872 {
873         pkc->reset_pending_on_curr_blk = 1;
874         po->stats_u.stats3.tp_freeze_q_cnt++;
875 }
876
877 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
878
879 /*
880  * If the next block is free then we will dispatch it
881  * and return a good offset.
882  * Else, we will freeze the queue.
883  * So, caller must check the return value.
884  */
885 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
886                 struct packet_sock *po)
887 {
888         struct tpacket_block_desc *pbd;
889
890         smp_rmb();
891
892         /* 1. Get current block num */
893         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
894
895         /* 2. If this block is currently in_use then freeze the queue */
896         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
897                 prb_freeze_queue(pkc, po);
898                 return NULL;
899         }
900
901         /*
902          * 3.
903          * open this block and return the offset where the first packet
904          * needs to get stored.
905          */
906         prb_open_block(pkc, pbd);
907         return (void *)pkc->nxt_offset;
908 }
909
910 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
911                 struct packet_sock *po, unsigned int status)
912 {
913         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
914
915         /* retire/close the current block */
916         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
917                 /*
918                  * Plug the case where copy_bits() is in progress on
919                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
920                  * have space to copy the pkt in the current block and
921                  * called prb_retire_current_block()
922                  *
923                  * We don't need to worry about the TMO case because
924                  * the timer-handler already handled this case.
925                  */
926                 if (!(status & TP_STATUS_BLK_TMO)) {
927                         while (atomic_read(&pkc->blk_fill_in_prog)) {
928                                 /* Waiting for skb_copy_bits to finish... */
929                                 cpu_relax();
930                         }
931                 }
932                 prb_close_block(pkc, pbd, po, status);
933                 return;
934         }
935 }
936
937 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
938                                       struct tpacket_block_desc *pbd)
939 {
940         return TP_STATUS_USER & BLOCK_STATUS(pbd);
941 }
942
943 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
944 {
945         return pkc->reset_pending_on_curr_blk;
946 }
947
948 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
949 {
950         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
951         atomic_dec(&pkc->blk_fill_in_prog);
952 }
953
954 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
955                         struct tpacket3_hdr *ppd)
956 {
957         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
958 }
959
960 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
961                         struct tpacket3_hdr *ppd)
962 {
963         ppd->hv1.tp_rxhash = 0;
964 }
965
966 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
967                         struct tpacket3_hdr *ppd)
968 {
969         if (vlan_tx_tag_present(pkc->skb)) {
970                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
971                 ppd->tp_status = TP_STATUS_VLAN_VALID;
972         } else {
973                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
974         }
975 }
976
977 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
978                         struct tpacket3_hdr *ppd)
979 {
980         prb_fill_vlan_info(pkc, ppd);
981
982         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
983                 prb_fill_rxhash(pkc, ppd);
984         else
985                 prb_clear_rxhash(pkc, ppd);
986 }
987
988 static void prb_fill_curr_block(char *curr,
989                                 struct tpacket_kbdq_core *pkc,
990                                 struct tpacket_block_desc *pbd,
991                                 unsigned int len)
992 {
993         struct tpacket3_hdr *ppd;
994
995         ppd  = (struct tpacket3_hdr *)curr;
996         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
997         pkc->prev = curr;
998         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
999         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1000         BLOCK_NUM_PKTS(pbd) += 1;
1001         atomic_inc(&pkc->blk_fill_in_prog);
1002         prb_run_all_ft_ops(pkc, ppd);
1003 }
1004
1005 /* Assumes caller has the sk->rx_queue.lock */
1006 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1007                                             struct sk_buff *skb,
1008                                                 int status,
1009                                             unsigned int len
1010                                             )
1011 {
1012         struct tpacket_kbdq_core *pkc;
1013         struct tpacket_block_desc *pbd;
1014         char *curr, *end;
1015
1016         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1017         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1018
1019         /* Queue is frozen when user space is lagging behind */
1020         if (prb_queue_frozen(pkc)) {
1021                 /*
1022                  * Check if that last block which caused the queue to freeze,
1023                  * is still in_use by user-space.
1024                  */
1025                 if (prb_curr_blk_in_use(pkc, pbd)) {
1026                         /* Can't record this packet */
1027                         return NULL;
1028                 } else {
1029                         /*
1030                          * Ok, the block was released by user-space.
1031                          * Now let's open that block.
1032                          * opening a block also thaws the queue.
1033                          * Thawing is a side effect.
1034                          */
1035                         prb_open_block(pkc, pbd);
1036                 }
1037         }
1038
1039         smp_mb();
1040         curr = pkc->nxt_offset;
1041         pkc->skb = skb;
1042         end = (char *) ((char *)pbd + pkc->kblk_size);
1043
1044         /* first try the current block */
1045         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1046                 prb_fill_curr_block(curr, pkc, pbd, len);
1047                 return (void *)curr;
1048         }
1049
1050         /* Ok, close the current block */
1051         prb_retire_current_block(pkc, po, 0);
1052
1053         /* Now, try to dispatch the next block */
1054         curr = (char *)prb_dispatch_next_block(pkc, po);
1055         if (curr) {
1056                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1057                 prb_fill_curr_block(curr, pkc, pbd, len);
1058                 return (void *)curr;
1059         }
1060
1061         /*
1062          * No free blocks are available.user_space hasn't caught up yet.
1063          * Queue was just frozen and now this packet will get dropped.
1064          */
1065         return NULL;
1066 }
1067
1068 static void *packet_current_rx_frame(struct packet_sock *po,
1069                                             struct sk_buff *skb,
1070                                             int status, unsigned int len)
1071 {
1072         char *curr = NULL;
1073         switch (po->tp_version) {
1074         case TPACKET_V1:
1075         case TPACKET_V2:
1076                 curr = packet_lookup_frame(po, &po->rx_ring,
1077                                         po->rx_ring.head, status);
1078                 return curr;
1079         case TPACKET_V3:
1080                 return __packet_lookup_frame_in_block(po, skb, status, len);
1081         default:
1082                 WARN(1, "TPACKET version not supported\n");
1083                 BUG();
1084                 return 0;
1085         }
1086 }
1087
1088 static void *prb_lookup_block(struct packet_sock *po,
1089                                      struct packet_ring_buffer *rb,
1090                                      unsigned int previous,
1091                                      int status)
1092 {
1093         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1094         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1095
1096         if (status != BLOCK_STATUS(pbd))
1097                 return NULL;
1098         return pbd;
1099 }
1100
1101 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1102 {
1103         unsigned int prev;
1104         if (rb->prb_bdqc.kactive_blk_num)
1105                 prev = rb->prb_bdqc.kactive_blk_num-1;
1106         else
1107                 prev = rb->prb_bdqc.knum_blocks-1;
1108         return prev;
1109 }
1110
1111 /* Assumes caller has held the rx_queue.lock */
1112 static void *__prb_previous_block(struct packet_sock *po,
1113                                          struct packet_ring_buffer *rb,
1114                                          int status)
1115 {
1116         unsigned int previous = prb_previous_blk_num(rb);
1117         return prb_lookup_block(po, rb, previous, status);
1118 }
1119
1120 static void *packet_previous_rx_frame(struct packet_sock *po,
1121                                              struct packet_ring_buffer *rb,
1122                                              int status)
1123 {
1124         if (po->tp_version <= TPACKET_V2)
1125                 return packet_previous_frame(po, rb, status);
1126
1127         return __prb_previous_block(po, rb, status);
1128 }
1129
1130 static void packet_increment_rx_head(struct packet_sock *po,
1131                                             struct packet_ring_buffer *rb)
1132 {
1133         switch (po->tp_version) {
1134         case TPACKET_V1:
1135         case TPACKET_V2:
1136                 return packet_increment_head(rb);
1137         case TPACKET_V3:
1138         default:
1139                 WARN(1, "TPACKET version not supported.\n");
1140                 BUG();
1141                 return;
1142         }
1143 }
1144
1145 static void *packet_previous_frame(struct packet_sock *po,
1146                 struct packet_ring_buffer *rb,
1147                 int status)
1148 {
1149         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1150         return packet_lookup_frame(po, rb, previous, status);
1151 }
1152
1153 static void packet_increment_head(struct packet_ring_buffer *buff)
1154 {
1155         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1156 }
1157
1158 static void packet_sock_destruct(struct sock *sk)
1159 {
1160         skb_queue_purge(&sk->sk_error_queue);
1161
1162         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1163         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1164
1165         if (!sock_flag(sk, SOCK_DEAD)) {
1166                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1167                 return;
1168         }
1169
1170         sk_refcnt_debug_dec(sk);
1171 }
1172
1173 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1174 {
1175         u32 idx, hash = skb->rxhash;
1176
1177         idx = ((u64)hash * num) >> 32;
1178
1179         return f->arr[idx];
1180 }
1181
1182 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1183 {
1184         unsigned int val = atomic_inc_return(&f->rr_cur);
1185
1186         return f->arr[val % num];
1187 }
1188
1189 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1190 {
1191         unsigned int cpu = smp_processor_id();
1192
1193         return f->arr[cpu % num];
1194 }
1195
1196 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1197                              struct packet_type *pt, struct net_device *orig_dev)
1198 {
1199         struct packet_fanout *f = pt->af_packet_priv;
1200         unsigned int num = ACCESS_ONCE(f->num_members);
1201         struct packet_sock *po;
1202         struct sock *sk;
1203
1204         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1205             !num) {
1206                 kfree_skb(skb);
1207                 return 0;
1208         }
1209
1210         switch (f->type) {
1211         case PACKET_FANOUT_HASH:
1212         default:
1213                 if (f->defrag) {
1214                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1215                         if (!skb)
1216                                 return 0;
1217                 }
1218                 skb_get_rxhash(skb);
1219                 sk = fanout_demux_hash(f, skb, num);
1220                 break;
1221         case PACKET_FANOUT_LB:
1222                 sk = fanout_demux_lb(f, skb, num);
1223                 break;
1224         case PACKET_FANOUT_CPU:
1225                 sk = fanout_demux_cpu(f, skb, num);
1226                 break;
1227         }
1228
1229         po = pkt_sk(sk);
1230
1231         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1232 }
1233
1234 static DEFINE_MUTEX(fanout_mutex);
1235 static LIST_HEAD(fanout_list);
1236
1237 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1238 {
1239         struct packet_fanout *f = po->fanout;
1240
1241         spin_lock(&f->lock);
1242         f->arr[f->num_members] = sk;
1243         smp_wmb();
1244         f->num_members++;
1245         spin_unlock(&f->lock);
1246 }
1247
1248 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1249 {
1250         struct packet_fanout *f = po->fanout;
1251         int i;
1252
1253         spin_lock(&f->lock);
1254         for (i = 0; i < f->num_members; i++) {
1255                 if (f->arr[i] == sk)
1256                         break;
1257         }
1258         BUG_ON(i >= f->num_members);
1259         f->arr[i] = f->arr[f->num_members - 1];
1260         f->num_members--;
1261         spin_unlock(&f->lock);
1262 }
1263
1264 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1265 {
1266         if (sk->sk_family != PF_PACKET)
1267                 return false;
1268
1269         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1270 }
1271
1272 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1273 {
1274         struct packet_sock *po = pkt_sk(sk);
1275         struct packet_fanout *f, *match;
1276         u8 type = type_flags & 0xff;
1277         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1278         int err;
1279
1280         switch (type) {
1281         case PACKET_FANOUT_HASH:
1282         case PACKET_FANOUT_LB:
1283         case PACKET_FANOUT_CPU:
1284                 break;
1285         default:
1286                 return -EINVAL;
1287         }
1288
1289         mutex_lock(&fanout_mutex);
1290
1291         err = -EINVAL;
1292         if (!po->running)
1293                 goto out;
1294
1295         err = -EALREADY;
1296         if (po->fanout)
1297                 goto out;
1298
1299         match = NULL;
1300         list_for_each_entry(f, &fanout_list, list) {
1301                 if (f->id == id &&
1302                     read_pnet(&f->net) == sock_net(sk)) {
1303                         match = f;
1304                         break;
1305                 }
1306         }
1307         err = -EINVAL;
1308         if (match && match->defrag != defrag)
1309                 goto out;
1310         if (!match) {
1311                 err = -ENOMEM;
1312                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1313                 if (!match)
1314                         goto out;
1315                 write_pnet(&match->net, sock_net(sk));
1316                 match->id = id;
1317                 match->type = type;
1318                 match->defrag = defrag;
1319                 atomic_set(&match->rr_cur, 0);
1320                 INIT_LIST_HEAD(&match->list);
1321                 spin_lock_init(&match->lock);
1322                 atomic_set(&match->sk_ref, 0);
1323                 match->prot_hook.type = po->prot_hook.type;
1324                 match->prot_hook.dev = po->prot_hook.dev;
1325                 match->prot_hook.func = packet_rcv_fanout;
1326                 match->prot_hook.af_packet_priv = match;
1327                 match->prot_hook.id_match = match_fanout_group;
1328                 dev_add_pack(&match->prot_hook);
1329                 list_add(&match->list, &fanout_list);
1330         }
1331         err = -EINVAL;
1332         if (match->type == type &&
1333             match->prot_hook.type == po->prot_hook.type &&
1334             match->prot_hook.dev == po->prot_hook.dev) {
1335                 err = -ENOSPC;
1336                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1337                         __dev_remove_pack(&po->prot_hook);
1338                         po->fanout = match;
1339                         atomic_inc(&match->sk_ref);
1340                         __fanout_link(sk, po);
1341                         err = 0;
1342                 }
1343         }
1344 out:
1345         mutex_unlock(&fanout_mutex);
1346         return err;
1347 }
1348
1349 static void fanout_release(struct sock *sk)
1350 {
1351         struct packet_sock *po = pkt_sk(sk);
1352         struct packet_fanout *f;
1353
1354         mutex_lock(&fanout_mutex);
1355         f = po->fanout;
1356         if (f) {
1357                 po->fanout = NULL;
1358
1359                 if (atomic_dec_and_test(&f->sk_ref)) {
1360                         list_del(&f->list);
1361                         dev_remove_pack(&f->prot_hook);
1362                         kfree(f);
1363                 }
1364         }
1365         mutex_unlock(&fanout_mutex);
1366 }
1367
1368 static const struct proto_ops packet_ops;
1369
1370 static const struct proto_ops packet_ops_spkt;
1371
1372 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1373                            struct packet_type *pt, struct net_device *orig_dev)
1374 {
1375         struct sock *sk;
1376         struct sockaddr_pkt *spkt;
1377
1378         /*
1379          *      When we registered the protocol we saved the socket in the data
1380          *      field for just this event.
1381          */
1382
1383         sk = pt->af_packet_priv;
1384
1385         /*
1386          *      Yank back the headers [hope the device set this
1387          *      right or kerboom...]
1388          *
1389          *      Incoming packets have ll header pulled,
1390          *      push it back.
1391          *
1392          *      For outgoing ones skb->data == skb_mac_header(skb)
1393          *      so that this procedure is noop.
1394          */
1395
1396         if (skb->pkt_type == PACKET_LOOPBACK)
1397                 goto out;
1398
1399         if (!net_eq(dev_net(dev), sock_net(sk)))
1400                 goto out;
1401
1402         skb = skb_share_check(skb, GFP_ATOMIC);
1403         if (skb == NULL)
1404                 goto oom;
1405
1406         /* drop any routing info */
1407         skb_dst_drop(skb);
1408
1409         /* drop conntrack reference */
1410         nf_reset(skb);
1411
1412         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1413
1414         skb_push(skb, skb->data - skb_mac_header(skb));
1415
1416         /*
1417          *      The SOCK_PACKET socket receives _all_ frames.
1418          */
1419
1420         spkt->spkt_family = dev->type;
1421         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1422         spkt->spkt_protocol = skb->protocol;
1423
1424         /*
1425          *      Charge the memory to the socket. This is done specifically
1426          *      to prevent sockets using all the memory up.
1427          */
1428
1429         if (sock_queue_rcv_skb(sk, skb) == 0)
1430                 return 0;
1431
1432 out:
1433         kfree_skb(skb);
1434 oom:
1435         return 0;
1436 }
1437
1438
1439 /*
1440  *      Output a raw packet to a device layer. This bypasses all the other
1441  *      protocol layers and you must therefore supply it with a complete frame
1442  */
1443
1444 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1445                                struct msghdr *msg, size_t len)
1446 {
1447         struct sock *sk = sock->sk;
1448         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1449         struct sk_buff *skb = NULL;
1450         struct net_device *dev;
1451         __be16 proto = 0;
1452         int err;
1453
1454         /*
1455          *      Get and verify the address.
1456          */
1457
1458         if (saddr) {
1459                 if (msg->msg_namelen < sizeof(struct sockaddr))
1460                         return -EINVAL;
1461                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1462                         proto = saddr->spkt_protocol;
1463         } else
1464                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1465
1466         /*
1467          *      Find the device first to size check it
1468          */
1469
1470         saddr->spkt_device[13] = 0;
1471 retry:
1472         rcu_read_lock();
1473         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1474         err = -ENODEV;
1475         if (dev == NULL)
1476                 goto out_unlock;
1477
1478         err = -ENETDOWN;
1479         if (!(dev->flags & IFF_UP))
1480                 goto out_unlock;
1481
1482         /*
1483          * You may not queue a frame bigger than the mtu. This is the lowest level
1484          * raw protocol and you must do your own fragmentation at this level.
1485          */
1486
1487         err = -EMSGSIZE;
1488         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1489                 goto out_unlock;
1490
1491         if (!skb) {
1492                 size_t reserved = LL_RESERVED_SPACE(dev);
1493                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1494
1495                 rcu_read_unlock();
1496                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1497                 if (skb == NULL)
1498                         return -ENOBUFS;
1499                 /* FIXME: Save some space for broken drivers that write a hard
1500                  * header at transmission time by themselves. PPP is the notable
1501                  * one here. This should really be fixed at the driver level.
1502                  */
1503                 skb_reserve(skb, reserved);
1504                 skb_reset_network_header(skb);
1505
1506                 /* Try to align data part correctly */
1507                 if (hhlen) {
1508                         skb->data -= hhlen;
1509                         skb->tail -= hhlen;
1510                         if (len < hhlen)
1511                                 skb_reset_network_header(skb);
1512                 }
1513                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1514                 if (err)
1515                         goto out_free;
1516                 goto retry;
1517         }
1518
1519         if (len > (dev->mtu + dev->hard_header_len)) {
1520                 /* Earlier code assumed this would be a VLAN pkt,
1521                  * double-check this now that we have the actual
1522                  * packet in hand.
1523                  */
1524                 struct ethhdr *ehdr;
1525                 skb_reset_mac_header(skb);
1526                 ehdr = eth_hdr(skb);
1527                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1528                         err = -EMSGSIZE;
1529                         goto out_unlock;
1530                 }
1531         }
1532
1533         skb->protocol = proto;
1534         skb->dev = dev;
1535         skb->priority = sk->sk_priority;
1536         skb->mark = sk->sk_mark;
1537         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1538         if (err < 0)
1539                 goto out_unlock;
1540
1541         dev_queue_xmit(skb);
1542         rcu_read_unlock();
1543         return len;
1544
1545 out_unlock:
1546         rcu_read_unlock();
1547 out_free:
1548         kfree_skb(skb);
1549         return err;
1550 }
1551
1552 static unsigned int run_filter(const struct sk_buff *skb,
1553                                       const struct sock *sk,
1554                                       unsigned int res)
1555 {
1556         struct sk_filter *filter;
1557
1558         rcu_read_lock();
1559         filter = rcu_dereference(sk->sk_filter);
1560         if (filter != NULL)
1561                 res = SK_RUN_FILTER(filter, skb);
1562         rcu_read_unlock();
1563
1564         return res;
1565 }
1566
1567 /*
1568  * This function makes lazy skb cloning in hope that most of packets
1569  * are discarded by BPF.
1570  *
1571  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1572  * and skb->cb are mangled. It works because (and until) packets
1573  * falling here are owned by current CPU. Output packets are cloned
1574  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1575  * sequencially, so that if we return skb to original state on exit,
1576  * we will not harm anyone.
1577  */
1578
1579 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1580                       struct packet_type *pt, struct net_device *orig_dev)
1581 {
1582         struct sock *sk;
1583         struct sockaddr_ll *sll;
1584         struct packet_sock *po;
1585         u8 *skb_head = skb->data;
1586         int skb_len = skb->len;
1587         unsigned int snaplen, res;
1588
1589         if (skb->pkt_type == PACKET_LOOPBACK)
1590                 goto drop;
1591
1592         sk = pt->af_packet_priv;
1593         po = pkt_sk(sk);
1594
1595         if (!net_eq(dev_net(dev), sock_net(sk)))
1596                 goto drop;
1597
1598         skb->dev = dev;
1599
1600         if (dev->header_ops) {
1601                 /* The device has an explicit notion of ll header,
1602                  * exported to higher levels.
1603                  *
1604                  * Otherwise, the device hides details of its frame
1605                  * structure, so that corresponding packet head is
1606                  * never delivered to user.
1607                  */
1608                 if (sk->sk_type != SOCK_DGRAM)
1609                         skb_push(skb, skb->data - skb_mac_header(skb));
1610                 else if (skb->pkt_type == PACKET_OUTGOING) {
1611                         /* Special case: outgoing packets have ll header at head */
1612                         skb_pull(skb, skb_network_offset(skb));
1613                 }
1614         }
1615
1616         snaplen = skb->len;
1617
1618         res = run_filter(skb, sk, snaplen);
1619         if (!res)
1620                 goto drop_n_restore;
1621         if (snaplen > res)
1622                 snaplen = res;
1623
1624         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1625                 goto drop_n_acct;
1626
1627         if (skb_shared(skb)) {
1628                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1629                 if (nskb == NULL)
1630                         goto drop_n_acct;
1631
1632                 if (skb_head != skb->data) {
1633                         skb->data = skb_head;
1634                         skb->len = skb_len;
1635                 }
1636                 kfree_skb(skb);
1637                 skb = nskb;
1638         }
1639
1640         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1641                      sizeof(skb->cb));
1642
1643         sll = &PACKET_SKB_CB(skb)->sa.ll;
1644         sll->sll_family = AF_PACKET;
1645         sll->sll_hatype = dev->type;
1646         sll->sll_protocol = skb->protocol;
1647         sll->sll_pkttype = skb->pkt_type;
1648         if (unlikely(po->origdev))
1649                 sll->sll_ifindex = orig_dev->ifindex;
1650         else
1651                 sll->sll_ifindex = dev->ifindex;
1652
1653         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1654
1655         PACKET_SKB_CB(skb)->origlen = skb->len;
1656
1657         if (pskb_trim(skb, snaplen))
1658                 goto drop_n_acct;
1659
1660         skb_set_owner_r(skb, sk);
1661         skb->dev = NULL;
1662         skb_dst_drop(skb);
1663
1664         /* drop conntrack reference */
1665         nf_reset(skb);
1666
1667         spin_lock(&sk->sk_receive_queue.lock);
1668         po->stats.tp_packets++;
1669         skb->dropcount = atomic_read(&sk->sk_drops);
1670         __skb_queue_tail(&sk->sk_receive_queue, skb);
1671         spin_unlock(&sk->sk_receive_queue.lock);
1672         sk->sk_data_ready(sk, skb->len);
1673         return 0;
1674
1675 drop_n_acct:
1676         spin_lock(&sk->sk_receive_queue.lock);
1677         po->stats.tp_drops++;
1678         atomic_inc(&sk->sk_drops);
1679         spin_unlock(&sk->sk_receive_queue.lock);
1680
1681 drop_n_restore:
1682         if (skb_head != skb->data && skb_shared(skb)) {
1683                 skb->data = skb_head;
1684                 skb->len = skb_len;
1685         }
1686 drop:
1687         consume_skb(skb);
1688         return 0;
1689 }
1690
1691 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1692                        struct packet_type *pt, struct net_device *orig_dev)
1693 {
1694         struct sock *sk;
1695         struct packet_sock *po;
1696         struct sockaddr_ll *sll;
1697         union {
1698                 struct tpacket_hdr *h1;
1699                 struct tpacket2_hdr *h2;
1700                 struct tpacket3_hdr *h3;
1701                 void *raw;
1702         } h;
1703         u8 *skb_head = skb->data;
1704         int skb_len = skb->len;
1705         unsigned int snaplen, res;
1706         unsigned long status = TP_STATUS_USER;
1707         unsigned short macoff, netoff, hdrlen;
1708         struct sk_buff *copy_skb = NULL;
1709         struct timeval tv;
1710         struct timespec ts;
1711         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1712
1713         if (skb->pkt_type == PACKET_LOOPBACK)
1714                 goto drop;
1715
1716         sk = pt->af_packet_priv;
1717         po = pkt_sk(sk);
1718
1719         if (!net_eq(dev_net(dev), sock_net(sk)))
1720                 goto drop;
1721
1722         if (dev->header_ops) {
1723                 if (sk->sk_type != SOCK_DGRAM)
1724                         skb_push(skb, skb->data - skb_mac_header(skb));
1725                 else if (skb->pkt_type == PACKET_OUTGOING) {
1726                         /* Special case: outgoing packets have ll header at head */
1727                         skb_pull(skb, skb_network_offset(skb));
1728                 }
1729         }
1730
1731         if (skb->ip_summed == CHECKSUM_PARTIAL)
1732                 status |= TP_STATUS_CSUMNOTREADY;
1733
1734         snaplen = skb->len;
1735
1736         res = run_filter(skb, sk, snaplen);
1737         if (!res)
1738                 goto drop_n_restore;
1739         if (snaplen > res)
1740                 snaplen = res;
1741
1742         if (sk->sk_type == SOCK_DGRAM) {
1743                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1744                                   po->tp_reserve;
1745         } else {
1746                 unsigned maclen = skb_network_offset(skb);
1747                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1748                                        (maclen < 16 ? 16 : maclen)) +
1749                         po->tp_reserve;
1750                 macoff = netoff - maclen;
1751         }
1752         if (po->tp_version <= TPACKET_V2) {
1753                 if (macoff + snaplen > po->rx_ring.frame_size) {
1754                         if (po->copy_thresh &&
1755                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1756                                 if (skb_shared(skb)) {
1757                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1758                                 } else {
1759                                         copy_skb = skb_get(skb);
1760                                         skb_head = skb->data;
1761                                 }
1762                                 if (copy_skb)
1763                                         skb_set_owner_r(copy_skb, sk);
1764                         }
1765                         snaplen = po->rx_ring.frame_size - macoff;
1766                         if ((int)snaplen < 0)
1767                                 snaplen = 0;
1768                 }
1769         }
1770         spin_lock(&sk->sk_receive_queue.lock);
1771         h.raw = packet_current_rx_frame(po, skb,
1772                                         TP_STATUS_KERNEL, (macoff+snaplen));
1773         if (!h.raw)
1774                 goto ring_is_full;
1775         if (po->tp_version <= TPACKET_V2) {
1776                 packet_increment_rx_head(po, &po->rx_ring);
1777         /*
1778          * LOSING will be reported till you read the stats,
1779          * because it's COR - Clear On Read.
1780          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1781          * at packet level.
1782          */
1783                 if (po->stats.tp_drops)
1784                         status |= TP_STATUS_LOSING;
1785         }
1786         po->stats.tp_packets++;
1787         if (copy_skb) {
1788                 status |= TP_STATUS_COPY;
1789                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1790         }
1791         spin_unlock(&sk->sk_receive_queue.lock);
1792
1793         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1794
1795         switch (po->tp_version) {
1796         case TPACKET_V1:
1797                 h.h1->tp_len = skb->len;
1798                 h.h1->tp_snaplen = snaplen;
1799                 h.h1->tp_mac = macoff;
1800                 h.h1->tp_net = netoff;
1801                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1802                                 && shhwtstamps->syststamp.tv64)
1803                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1804                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1805                                 && shhwtstamps->hwtstamp.tv64)
1806                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1807                 else if (skb->tstamp.tv64)
1808                         tv = ktime_to_timeval(skb->tstamp);
1809                 else
1810                         do_gettimeofday(&tv);
1811                 h.h1->tp_sec = tv.tv_sec;
1812                 h.h1->tp_usec = tv.tv_usec;
1813                 hdrlen = sizeof(*h.h1);
1814                 break;
1815         case TPACKET_V2:
1816                 h.h2->tp_len = skb->len;
1817                 h.h2->tp_snaplen = snaplen;
1818                 h.h2->tp_mac = macoff;
1819                 h.h2->tp_net = netoff;
1820                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1821                                 && shhwtstamps->syststamp.tv64)
1822                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1823                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1824                                 && shhwtstamps->hwtstamp.tv64)
1825                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1826                 else if (skb->tstamp.tv64)
1827                         ts = ktime_to_timespec(skb->tstamp);
1828                 else
1829                         getnstimeofday(&ts);
1830                 h.h2->tp_sec = ts.tv_sec;
1831                 h.h2->tp_nsec = ts.tv_nsec;
1832                 if (vlan_tx_tag_present(skb)) {
1833                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1834                         status |= TP_STATUS_VLAN_VALID;
1835                 } else {
1836                         h.h2->tp_vlan_tci = 0;
1837                 }
1838                 h.h2->tp_padding = 0;
1839                 hdrlen = sizeof(*h.h2);
1840                 break;
1841         case TPACKET_V3:
1842                 /* tp_nxt_offset,vlan are already populated above.
1843                  * So DONT clear those fields here
1844                  */
1845                 h.h3->tp_status |= status;
1846                 h.h3->tp_len = skb->len;
1847                 h.h3->tp_snaplen = snaplen;
1848                 h.h3->tp_mac = macoff;
1849                 h.h3->tp_net = netoff;
1850                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1851                                 && shhwtstamps->syststamp.tv64)
1852                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1853                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1854                                 && shhwtstamps->hwtstamp.tv64)
1855                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1856                 else if (skb->tstamp.tv64)
1857                         ts = ktime_to_timespec(skb->tstamp);
1858                 else
1859                         getnstimeofday(&ts);
1860                 h.h3->tp_sec  = ts.tv_sec;
1861                 h.h3->tp_nsec = ts.tv_nsec;
1862                 hdrlen = sizeof(*h.h3);
1863                 break;
1864         default:
1865                 BUG();
1866         }
1867
1868         sll = h.raw + TPACKET_ALIGN(hdrlen);
1869         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1870         sll->sll_family = AF_PACKET;
1871         sll->sll_hatype = dev->type;
1872         sll->sll_protocol = skb->protocol;
1873         sll->sll_pkttype = skb->pkt_type;
1874         if (unlikely(po->origdev))
1875                 sll->sll_ifindex = orig_dev->ifindex;
1876         else
1877                 sll->sll_ifindex = dev->ifindex;
1878
1879         smp_mb();
1880 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1881         {
1882                 u8 *start, *end;
1883
1884                 if (po->tp_version <= TPACKET_V2) {
1885                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1886                                 + macoff + snaplen);
1887                         for (start = h.raw; start < end; start += PAGE_SIZE)
1888                                 flush_dcache_page(pgv_to_page(start));
1889                 }
1890                 smp_wmb();
1891         }
1892 #endif
1893         if (po->tp_version <= TPACKET_V2)
1894                 __packet_set_status(po, h.raw, status);
1895         else
1896                 prb_clear_blk_fill_status(&po->rx_ring);
1897
1898         sk->sk_data_ready(sk, 0);
1899
1900 drop_n_restore:
1901         if (skb_head != skb->data && skb_shared(skb)) {
1902                 skb->data = skb_head;
1903                 skb->len = skb_len;
1904         }
1905 drop:
1906         kfree_skb(skb);
1907         return 0;
1908
1909 ring_is_full:
1910         po->stats.tp_drops++;
1911         spin_unlock(&sk->sk_receive_queue.lock);
1912
1913         sk->sk_data_ready(sk, 0);
1914         kfree_skb(copy_skb);
1915         goto drop_n_restore;
1916 }
1917
1918 static void tpacket_destruct_skb(struct sk_buff *skb)
1919 {
1920         struct packet_sock *po = pkt_sk(skb->sk);
1921         void *ph;
1922
1923         if (likely(po->tx_ring.pg_vec)) {
1924                 ph = skb_shinfo(skb)->destructor_arg;
1925                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1926                 atomic_dec(&po->tx_ring.pending);
1927                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1928         }
1929
1930         sock_wfree(skb);
1931 }
1932
1933 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1934                 void *frame, struct net_device *dev, int size_max,
1935                 __be16 proto, unsigned char *addr)
1936 {
1937         union {
1938                 struct tpacket_hdr *h1;
1939                 struct tpacket2_hdr *h2;
1940                 void *raw;
1941         } ph;
1942         int to_write, offset, len, tp_len, nr_frags, len_max;
1943         struct socket *sock = po->sk.sk_socket;
1944         struct page *page;
1945         void *data;
1946         int err;
1947
1948         ph.raw = frame;
1949
1950         skb->protocol = proto;
1951         skb->dev = dev;
1952         skb->priority = po->sk.sk_priority;
1953         skb->mark = po->sk.sk_mark;
1954         skb_shinfo(skb)->destructor_arg = ph.raw;
1955
1956         switch (po->tp_version) {
1957         case TPACKET_V2:
1958                 tp_len = ph.h2->tp_len;
1959                 break;
1960         default:
1961                 tp_len = ph.h1->tp_len;
1962                 break;
1963         }
1964         if (unlikely(tp_len > size_max)) {
1965                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1966                 return -EMSGSIZE;
1967         }
1968
1969         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1970         skb_reset_network_header(skb);
1971
1972         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1973         to_write = tp_len;
1974
1975         if (sock->type == SOCK_DGRAM) {
1976                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1977                                 NULL, tp_len);
1978                 if (unlikely(err < 0))
1979                         return -EINVAL;
1980         } else if (dev->hard_header_len) {
1981                 /* net device doesn't like empty head */
1982                 if (unlikely(tp_len <= dev->hard_header_len)) {
1983                         pr_err("packet size is too short (%d < %d)\n",
1984                                tp_len, dev->hard_header_len);
1985                         return -EINVAL;
1986                 }
1987
1988                 skb_push(skb, dev->hard_header_len);
1989                 err = skb_store_bits(skb, 0, data,
1990                                 dev->hard_header_len);
1991                 if (unlikely(err))
1992                         return err;
1993
1994                 data += dev->hard_header_len;
1995                 to_write -= dev->hard_header_len;
1996         }
1997
1998         err = -EFAULT;
1999         offset = offset_in_page(data);
2000         len_max = PAGE_SIZE - offset;
2001         len = ((to_write > len_max) ? len_max : to_write);
2002
2003         skb->data_len = to_write;
2004         skb->len += to_write;
2005         skb->truesize += to_write;
2006         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2007
2008         while (likely(to_write)) {
2009                 nr_frags = skb_shinfo(skb)->nr_frags;
2010
2011                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2012                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2013                                MAX_SKB_FRAGS);
2014                         return -EFAULT;
2015                 }
2016
2017                 page = pgv_to_page(data);
2018                 data += len;
2019                 flush_dcache_page(page);
2020                 get_page(page);
2021                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2022                 to_write -= len;
2023                 offset = 0;
2024                 len_max = PAGE_SIZE;
2025                 len = ((to_write > len_max) ? len_max : to_write);
2026         }
2027
2028         return tp_len;
2029 }
2030
2031 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2032 {
2033         struct net_device *dev;
2034
2035         rcu_read_lock();
2036         dev = rcu_dereference(po->cached_dev);
2037         if (dev)
2038                 dev_hold(dev);
2039         rcu_read_unlock();
2040
2041         return dev;
2042 }
2043
2044 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2045 {
2046         struct sk_buff *skb;
2047         struct net_device *dev;
2048         __be16 proto;
2049         int err, reserve = 0;
2050         void *ph;
2051         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2052         int tp_len, size_max;
2053         unsigned char *addr;
2054         int len_sum = 0;
2055         int status = 0;
2056
2057         mutex_lock(&po->pg_vec_lock);
2058
2059         err = -EBUSY;
2060         if (saddr == NULL) {
2061                 dev     = packet_cached_dev_get(po);
2062                 proto   = po->num;
2063                 addr    = NULL;
2064         } else {
2065                 err = -EINVAL;
2066                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2067                         goto out;
2068                 if (msg->msg_namelen < (saddr->sll_halen
2069                                         + offsetof(struct sockaddr_ll,
2070                                                 sll_addr)))
2071                         goto out;
2072                 proto   = saddr->sll_protocol;
2073                 addr    = saddr->sll_addr;
2074                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2075         }
2076
2077         err = -ENXIO;
2078         if (unlikely(dev == NULL))
2079                 goto out;
2080         err = -ENETDOWN;
2081         if (unlikely(!(dev->flags & IFF_UP)))
2082                 goto out_put;
2083
2084         reserve = dev->hard_header_len;
2085
2086         size_max = po->tx_ring.frame_size
2087                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2088
2089         if (size_max > dev->mtu + reserve)
2090                 size_max = dev->mtu + reserve;
2091
2092         do {
2093                 ph = packet_current_frame(po, &po->tx_ring,
2094                                 TP_STATUS_SEND_REQUEST);
2095
2096                 if (unlikely(ph == NULL)) {
2097                         schedule();
2098                         continue;
2099                 }
2100
2101                 status = TP_STATUS_SEND_REQUEST;
2102                 skb = sock_alloc_send_skb(&po->sk,
2103                                 LL_ALLOCATED_SPACE(dev)
2104                                 + sizeof(struct sockaddr_ll),
2105                                 0, &err);
2106
2107                 if (unlikely(skb == NULL))
2108                         goto out_status;
2109
2110                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2111                                 addr);
2112
2113                 if (unlikely(tp_len < 0)) {
2114                         if (po->tp_loss) {
2115                                 __packet_set_status(po, ph,
2116                                                 TP_STATUS_AVAILABLE);
2117                                 packet_increment_head(&po->tx_ring);
2118                                 kfree_skb(skb);
2119                                 continue;
2120                         } else {
2121                                 status = TP_STATUS_WRONG_FORMAT;
2122                                 err = tp_len;
2123                                 goto out_status;
2124                         }
2125                 }
2126
2127                 skb->destructor = tpacket_destruct_skb;
2128                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2129                 atomic_inc(&po->tx_ring.pending);
2130
2131                 status = TP_STATUS_SEND_REQUEST;
2132                 err = dev_queue_xmit(skb);
2133                 if (unlikely(err > 0)) {
2134                         err = net_xmit_errno(err);
2135                         if (err && __packet_get_status(po, ph) ==
2136                                    TP_STATUS_AVAILABLE) {
2137                                 /* skb was destructed already */
2138                                 skb = NULL;
2139                                 goto out_status;
2140                         }
2141                         /*
2142                          * skb was dropped but not destructed yet;
2143                          * let's treat it like congestion or err < 0
2144                          */
2145                         err = 0;
2146                 }
2147                 packet_increment_head(&po->tx_ring);
2148                 len_sum += tp_len;
2149         } while (likely((ph != NULL) ||
2150                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2151                          (atomic_read(&po->tx_ring.pending))))
2152                 );
2153
2154         err = len_sum;
2155         goto out_put;
2156
2157 out_status:
2158         __packet_set_status(po, ph, status);
2159         kfree_skb(skb);
2160 out_put:
2161         dev_put(dev);
2162 out:
2163         mutex_unlock(&po->pg_vec_lock);
2164         return err;
2165 }
2166
2167 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2168                                         size_t reserve, size_t len,
2169                                         size_t linear, int noblock,
2170                                         int *err)
2171 {
2172         struct sk_buff *skb;
2173
2174         /* Under a page?  Don't bother with paged skb. */
2175         if (prepad + len < PAGE_SIZE || !linear)
2176                 linear = len;
2177
2178         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2179                                    err);
2180         if (!skb)
2181                 return NULL;
2182
2183         skb_reserve(skb, reserve);
2184         skb_put(skb, linear);
2185         skb->data_len = len - linear;
2186         skb->len += len - linear;
2187
2188         return skb;
2189 }
2190
2191 static int packet_snd(struct socket *sock,
2192                           struct msghdr *msg, size_t len)
2193 {
2194         struct sock *sk = sock->sk;
2195         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2196         struct sk_buff *skb;
2197         struct net_device *dev;
2198         __be16 proto;
2199         unsigned char *addr;
2200         int err, reserve = 0;
2201         struct virtio_net_hdr vnet_hdr = { 0 };
2202         int offset = 0;
2203         int vnet_hdr_len;
2204         struct packet_sock *po = pkt_sk(sk);
2205         unsigned short gso_type = 0;
2206
2207         /*
2208          *      Get and verify the address.
2209          */
2210
2211         if (saddr == NULL) {
2212                 dev     = packet_cached_dev_get(po);
2213                 proto   = po->num;
2214                 addr    = NULL;
2215         } else {
2216                 err = -EINVAL;
2217                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2218                         goto out;
2219                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2220                         goto out;
2221                 proto   = saddr->sll_protocol;
2222                 addr    = saddr->sll_addr;
2223                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2224         }
2225
2226         err = -ENXIO;
2227         if (unlikely(dev == NULL))
2228                 goto out_unlock;
2229         err = -ENETDOWN;
2230         if (unlikely(!(dev->flags & IFF_UP)))
2231                 goto out_unlock;
2232
2233         if (sock->type == SOCK_RAW)
2234                 reserve = dev->hard_header_len;
2235         if (po->has_vnet_hdr) {
2236                 vnet_hdr_len = sizeof(vnet_hdr);
2237
2238                 err = -EINVAL;
2239                 if (len < vnet_hdr_len)
2240                         goto out_unlock;
2241
2242                 len -= vnet_hdr_len;
2243
2244                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2245                                        vnet_hdr_len);
2246                 if (err < 0)
2247                         goto out_unlock;
2248
2249                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2250                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2251                       vnet_hdr.hdr_len))
2252                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2253                                                  vnet_hdr.csum_offset + 2;
2254
2255                 err = -EINVAL;
2256                 if (vnet_hdr.hdr_len > len)
2257                         goto out_unlock;
2258
2259                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2260                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2261                         case VIRTIO_NET_HDR_GSO_TCPV4:
2262                                 gso_type = SKB_GSO_TCPV4;
2263                                 break;
2264                         case VIRTIO_NET_HDR_GSO_TCPV6:
2265                                 gso_type = SKB_GSO_TCPV6;
2266                                 break;
2267                         case VIRTIO_NET_HDR_GSO_UDP:
2268                                 gso_type = SKB_GSO_UDP;
2269                                 break;
2270                         default:
2271                                 goto out_unlock;
2272                         }
2273
2274                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2275                                 gso_type |= SKB_GSO_TCP_ECN;
2276
2277                         if (vnet_hdr.gso_size == 0)
2278                                 goto out_unlock;
2279
2280                 }
2281         }
2282
2283         err = -EMSGSIZE;
2284         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2285                 goto out_unlock;
2286
2287         err = -ENOBUFS;
2288         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2289                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2290                                msg->msg_flags & MSG_DONTWAIT, &err);
2291         if (skb == NULL)
2292                 goto out_unlock;
2293
2294         skb_set_network_header(skb, reserve);
2295
2296         err = -EINVAL;
2297         if (sock->type == SOCK_DGRAM &&
2298             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2299                 goto out_free;
2300
2301         /* Returns -EFAULT on error */
2302         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2303         if (err)
2304                 goto out_free;
2305         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2306         if (err < 0)
2307                 goto out_free;
2308
2309         if (!gso_type && (len > dev->mtu + reserve)) {
2310                 /* Earlier code assumed this would be a VLAN pkt,
2311                  * double-check this now that we have the actual
2312                  * packet in hand.
2313                  */
2314                 struct ethhdr *ehdr;
2315                 skb_reset_mac_header(skb);
2316                 ehdr = eth_hdr(skb);
2317                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2318                         err = -EMSGSIZE;
2319                         goto out_free;
2320                 }
2321         }
2322
2323         skb->protocol = proto;
2324         skb->dev = dev;
2325         skb->priority = sk->sk_priority;
2326         skb->mark = sk->sk_mark;
2327
2328         if (po->has_vnet_hdr) {
2329                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2330                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2331                                                   vnet_hdr.csum_offset)) {
2332                                 err = -EINVAL;
2333                                 goto out_free;
2334                         }
2335                 }
2336
2337                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2338                 skb_shinfo(skb)->gso_type = gso_type;
2339
2340                 /* Header must be checked, and gso_segs computed. */
2341                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2342                 skb_shinfo(skb)->gso_segs = 0;
2343
2344                 len += vnet_hdr_len;
2345         }
2346
2347         /*
2348          *      Now send it
2349          */
2350
2351         err = dev_queue_xmit(skb);
2352         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2353                 goto out_unlock;
2354
2355         dev_put(dev);
2356
2357         return len;
2358
2359 out_free:
2360         kfree_skb(skb);
2361 out_unlock:
2362         if (dev)
2363                 dev_put(dev);
2364 out:
2365         return err;
2366 }
2367
2368 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2369                 struct msghdr *msg, size_t len)
2370 {
2371         struct sock *sk = sock->sk;
2372         struct packet_sock *po = pkt_sk(sk);
2373         if (po->tx_ring.pg_vec)
2374                 return tpacket_snd(po, msg);
2375         else
2376                 return packet_snd(sock, msg, len);
2377 }
2378
2379 /*
2380  *      Close a PACKET socket. This is fairly simple. We immediately go
2381  *      to 'closed' state and remove our protocol entry in the device list.
2382  */
2383
2384 static int packet_release(struct socket *sock)
2385 {
2386         struct sock *sk = sock->sk;
2387         struct packet_sock *po;
2388         struct net *net;
2389         union tpacket_req_u req_u;
2390
2391         if (!sk)
2392                 return 0;
2393
2394         net = sock_net(sk);
2395         po = pkt_sk(sk);
2396
2397         spin_lock_bh(&net->packet.sklist_lock);
2398         sk_del_node_init_rcu(sk);
2399         sock_prot_inuse_add(net, sk->sk_prot, -1);
2400         spin_unlock_bh(&net->packet.sklist_lock);
2401
2402         spin_lock(&po->bind_lock);
2403         unregister_prot_hook(sk, false);
2404         if (po->prot_hook.dev) {
2405                 dev_put(po->prot_hook.dev);
2406                 po->prot_hook.dev = NULL;
2407         }
2408         spin_unlock(&po->bind_lock);
2409
2410         packet_flush_mclist(sk);
2411
2412         if (po->rx_ring.pg_vec) {
2413                 memset(&req_u, 0, sizeof(req_u));
2414                 packet_set_ring(sk, &req_u, 1, 0);
2415         }
2416
2417         if (po->tx_ring.pg_vec) {
2418                 memset(&req_u, 0, sizeof(req_u));
2419                 packet_set_ring(sk, &req_u, 1, 1);
2420         }
2421
2422         fanout_release(sk);
2423
2424         synchronize_net();
2425         /*
2426          *      Now the socket is dead. No more input will appear.
2427          */
2428         sock_orphan(sk);
2429         sock->sk = NULL;
2430
2431         /* Purge queues */
2432
2433         skb_queue_purge(&sk->sk_receive_queue);
2434         sk_refcnt_debug_release(sk);
2435
2436         sock_put(sk);
2437         return 0;
2438 }
2439
2440 /*
2441  *      Attach a packet hook.
2442  */
2443
2444 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2445 {
2446         struct packet_sock *po = pkt_sk(sk);
2447
2448         if (po->fanout) {
2449                 if (dev)
2450                         dev_put(dev);
2451
2452                 return -EINVAL;
2453         }
2454
2455         lock_sock(sk);
2456
2457         spin_lock(&po->bind_lock);
2458         unregister_prot_hook(sk, true);
2459         po->num = protocol;
2460         po->prot_hook.type = protocol;
2461         if (po->prot_hook.dev)
2462                 dev_put(po->prot_hook.dev);
2463         po->prot_hook.dev = dev;
2464
2465         po->ifindex = dev ? dev->ifindex : 0;
2466
2467         if (protocol == 0)
2468                 goto out_unlock;
2469
2470         if (!dev || (dev->flags & IFF_UP)) {
2471                 register_prot_hook(sk);
2472         } else {
2473                 sk->sk_err = ENETDOWN;
2474                 if (!sock_flag(sk, SOCK_DEAD))
2475                         sk->sk_error_report(sk);
2476         }
2477
2478 out_unlock:
2479         spin_unlock(&po->bind_lock);
2480         release_sock(sk);
2481         return 0;
2482 }
2483
2484 /*
2485  *      Bind a packet socket to a device
2486  */
2487
2488 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2489                             int addr_len)
2490 {
2491         struct sock *sk = sock->sk;
2492         char name[15];
2493         struct net_device *dev;
2494         int err = -ENODEV;
2495
2496         /*
2497          *      Check legality
2498          */
2499
2500         if (addr_len != sizeof(struct sockaddr))
2501                 return -EINVAL;
2502         strlcpy(name, uaddr->sa_data, sizeof(name));
2503
2504         dev = dev_get_by_name(sock_net(sk), name);
2505         if (dev)
2506                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2507         return err;
2508 }
2509
2510 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2511 {
2512         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2513         struct sock *sk = sock->sk;
2514         struct net_device *dev = NULL;
2515         int err;
2516
2517
2518         /*
2519          *      Check legality
2520          */
2521
2522         if (addr_len < sizeof(struct sockaddr_ll))
2523                 return -EINVAL;
2524         if (sll->sll_family != AF_PACKET)
2525                 return -EINVAL;
2526
2527         if (sll->sll_ifindex) {
2528                 err = -ENODEV;
2529                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2530                 if (dev == NULL)
2531                         goto out;
2532         }
2533         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2534
2535 out:
2536         return err;
2537 }
2538
2539 static struct proto packet_proto = {
2540         .name     = "PACKET",
2541         .owner    = THIS_MODULE,
2542         .obj_size = sizeof(struct packet_sock),
2543 };
2544
2545 /*
2546  *      Create a packet of type SOCK_PACKET.
2547  */
2548
2549 static int packet_create(struct net *net, struct socket *sock, int protocol,
2550                          int kern)
2551 {
2552         struct sock *sk;
2553         struct packet_sock *po;
2554         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2555         int err;
2556
2557         if (!capable(CAP_NET_RAW))
2558                 return -EPERM;
2559         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2560             sock->type != SOCK_PACKET)
2561                 return -ESOCKTNOSUPPORT;
2562
2563         sock->state = SS_UNCONNECTED;
2564
2565         err = -ENOBUFS;
2566         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2567         if (sk == NULL)
2568                 goto out;
2569
2570         sock->ops = &packet_ops;
2571         if (sock->type == SOCK_PACKET)
2572                 sock->ops = &packet_ops_spkt;
2573
2574         sock_init_data(sock, sk);
2575
2576         po = pkt_sk(sk);
2577         sk->sk_family = PF_PACKET;
2578         po->num = proto;
2579         RCU_INIT_POINTER(po->cached_dev, NULL);
2580
2581         sk->sk_destruct = packet_sock_destruct;
2582         sk_refcnt_debug_inc(sk);
2583
2584         /*
2585          *      Attach a protocol block
2586          */
2587
2588         spin_lock_init(&po->bind_lock);
2589         mutex_init(&po->pg_vec_lock);
2590         po->prot_hook.func = packet_rcv;
2591
2592         if (sock->type == SOCK_PACKET)
2593                 po->prot_hook.func = packet_rcv_spkt;
2594
2595         po->prot_hook.af_packet_priv = sk;
2596
2597         if (proto) {
2598                 po->prot_hook.type = proto;
2599                 register_prot_hook(sk);
2600         }
2601
2602         spin_lock_bh(&net->packet.sklist_lock);
2603         sk_add_node_rcu(sk, &net->packet.sklist);
2604         sock_prot_inuse_add(net, &packet_proto, 1);
2605         spin_unlock_bh(&net->packet.sklist_lock);
2606
2607         return 0;
2608 out:
2609         return err;
2610 }
2611
2612 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2613 {
2614         struct sock_exterr_skb *serr;
2615         struct sk_buff *skb, *skb2;
2616         int copied, err;
2617
2618         err = -EAGAIN;
2619         skb = skb_dequeue(&sk->sk_error_queue);
2620         if (skb == NULL)
2621                 goto out;
2622
2623         copied = skb->len;
2624         if (copied > len) {
2625                 msg->msg_flags |= MSG_TRUNC;
2626                 copied = len;
2627         }
2628         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2629         if (err)
2630                 goto out_free_skb;
2631
2632         sock_recv_timestamp(msg, sk, skb);
2633
2634         serr = SKB_EXT_ERR(skb);
2635         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2636                  sizeof(serr->ee), &serr->ee);
2637
2638         msg->msg_flags |= MSG_ERRQUEUE;
2639         err = copied;
2640
2641         /* Reset and regenerate socket error */
2642         spin_lock_bh(&sk->sk_error_queue.lock);
2643         sk->sk_err = 0;
2644         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2645                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2646                 spin_unlock_bh(&sk->sk_error_queue.lock);
2647                 sk->sk_error_report(sk);
2648         } else
2649                 spin_unlock_bh(&sk->sk_error_queue.lock);
2650
2651 out_free_skb:
2652         kfree_skb(skb);
2653 out:
2654         return err;
2655 }
2656
2657 /*
2658  *      Pull a packet from our receive queue and hand it to the user.
2659  *      If necessary we block.
2660  */
2661
2662 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2663                           struct msghdr *msg, size_t len, int flags)
2664 {
2665         struct sock *sk = sock->sk;
2666         struct sk_buff *skb;
2667         int copied, err;
2668         int vnet_hdr_len = 0;
2669
2670         err = -EINVAL;
2671         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2672                 goto out;
2673
2674 #if 0
2675         /* What error should we return now? EUNATTACH? */
2676         if (pkt_sk(sk)->ifindex < 0)
2677                 return -ENODEV;
2678 #endif
2679
2680         if (flags & MSG_ERRQUEUE) {
2681                 err = packet_recv_error(sk, msg, len);
2682                 goto out;
2683         }
2684
2685         /*
2686          *      Call the generic datagram receiver. This handles all sorts
2687          *      of horrible races and re-entrancy so we can forget about it
2688          *      in the protocol layers.
2689          *
2690          *      Now it will return ENETDOWN, if device have just gone down,
2691          *      but then it will block.
2692          */
2693
2694         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2695
2696         /*
2697          *      An error occurred so return it. Because skb_recv_datagram()
2698          *      handles the blocking we don't see and worry about blocking
2699          *      retries.
2700          */
2701
2702         if (skb == NULL)
2703                 goto out;
2704
2705         if (pkt_sk(sk)->has_vnet_hdr) {
2706                 struct virtio_net_hdr vnet_hdr = { 0 };
2707
2708                 err = -EINVAL;
2709                 vnet_hdr_len = sizeof(vnet_hdr);
2710                 if (len < vnet_hdr_len)
2711                         goto out_free;
2712
2713                 len -= vnet_hdr_len;
2714
2715                 if (skb_is_gso(skb)) {
2716                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2717
2718                         /* This is a hint as to how much should be linear. */
2719                         vnet_hdr.hdr_len = skb_headlen(skb);
2720                         vnet_hdr.gso_size = sinfo->gso_size;
2721                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2722                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2723                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2724                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2725                         else if (sinfo->gso_type & SKB_GSO_UDP)
2726                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2727                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2728                                 goto out_free;
2729                         else
2730                                 BUG();
2731                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2732                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2733                 } else
2734                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2735
2736                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2737                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2738                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2739                         vnet_hdr.csum_offset = skb->csum_offset;
2740                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2741                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2742                 } /* else everything is zero */
2743
2744                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2745                                      vnet_hdr_len);
2746                 if (err < 0)
2747                         goto out_free;
2748         }
2749
2750         /* You lose any data beyond the buffer you gave. If it worries
2751          * a user program they can ask the device for its MTU
2752          * anyway.
2753          */
2754         copied = skb->len;
2755         if (copied > len) {
2756                 copied = len;
2757                 msg->msg_flags |= MSG_TRUNC;
2758         }
2759
2760         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2761         if (err)
2762                 goto out_free;
2763
2764         sock_recv_ts_and_drops(msg, sk, skb);
2765
2766         if (msg->msg_name) {
2767                 /* If the address length field is there to be filled
2768                  * in, we fill it in now.
2769                  */
2770                 if (sock->type == SOCK_PACKET) {
2771                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2772                 } else {
2773                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2774                         msg->msg_namelen = sll->sll_halen +
2775                                 offsetof(struct sockaddr_ll, sll_addr);
2776                 }
2777                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2778                        msg->msg_namelen);
2779         }
2780
2781         if (pkt_sk(sk)->auxdata) {
2782                 struct tpacket_auxdata aux;
2783
2784                 aux.tp_status = TP_STATUS_USER;
2785                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2786                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2787                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2788                 aux.tp_snaplen = skb->len;
2789                 aux.tp_mac = 0;
2790                 aux.tp_net = skb_network_offset(skb);
2791                 if (vlan_tx_tag_present(skb)) {
2792                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2793                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2794                 } else {
2795                         aux.tp_vlan_tci = 0;
2796                 }
2797                 aux.tp_padding = 0;
2798                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2799         }
2800
2801         /*
2802          *      Free or return the buffer as appropriate. Again this
2803          *      hides all the races and re-entrancy issues from us.
2804          */
2805         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2806
2807 out_free:
2808         skb_free_datagram(sk, skb);
2809 out:
2810         return err;
2811 }
2812
2813 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2814                                int *uaddr_len, int peer)
2815 {
2816         struct net_device *dev;
2817         struct sock *sk = sock->sk;
2818
2819         if (peer)
2820                 return -EOPNOTSUPP;
2821
2822         uaddr->sa_family = AF_PACKET;
2823         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2824         rcu_read_lock();
2825         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2826         if (dev)
2827                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2828         rcu_read_unlock();
2829         *uaddr_len = sizeof(*uaddr);
2830
2831         return 0;
2832 }
2833
2834 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2835                           int *uaddr_len, int peer)
2836 {
2837         struct net_device *dev;
2838         struct sock *sk = sock->sk;
2839         struct packet_sock *po = pkt_sk(sk);
2840         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2841
2842         if (peer)
2843                 return -EOPNOTSUPP;
2844
2845         sll->sll_family = AF_PACKET;
2846         sll->sll_ifindex = po->ifindex;
2847         sll->sll_protocol = po->num;
2848         sll->sll_pkttype = 0;
2849         rcu_read_lock();
2850         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2851         if (dev) {
2852                 sll->sll_hatype = dev->type;
2853                 sll->sll_halen = dev->addr_len;
2854                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2855         } else {
2856                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2857                 sll->sll_halen = 0;
2858         }
2859         rcu_read_unlock();
2860         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2861
2862         return 0;
2863 }
2864
2865 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2866                          int what)
2867 {
2868         switch (i->type) {
2869         case PACKET_MR_MULTICAST:
2870                 if (i->alen != dev->addr_len)
2871                         return -EINVAL;
2872                 if (what > 0)
2873                         return dev_mc_add(dev, i->addr);
2874                 else
2875                         return dev_mc_del(dev, i->addr);
2876                 break;
2877         case PACKET_MR_PROMISC:
2878                 return dev_set_promiscuity(dev, what);
2879                 break;
2880         case PACKET_MR_ALLMULTI:
2881                 return dev_set_allmulti(dev, what);
2882                 break;
2883         case PACKET_MR_UNICAST:
2884                 if (i->alen != dev->addr_len)
2885                         return -EINVAL;
2886                 if (what > 0)
2887                         return dev_uc_add(dev, i->addr);
2888                 else
2889                         return dev_uc_del(dev, i->addr);
2890                 break;
2891         default:
2892                 break;
2893         }
2894         return 0;
2895 }
2896
2897 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2898 {
2899         for ( ; i; i = i->next) {
2900                 if (i->ifindex == dev->ifindex)
2901                         packet_dev_mc(dev, i, what);
2902         }
2903 }
2904
2905 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2906 {
2907         struct packet_sock *po = pkt_sk(sk);
2908         struct packet_mclist *ml, *i;
2909         struct net_device *dev;
2910         int err;
2911
2912         rtnl_lock();
2913
2914         err = -ENODEV;
2915         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2916         if (!dev)
2917                 goto done;
2918
2919         err = -EINVAL;
2920         if (mreq->mr_alen > dev->addr_len)
2921                 goto done;
2922
2923         err = -ENOBUFS;
2924         i = kmalloc(sizeof(*i), GFP_KERNEL);
2925         if (i == NULL)
2926                 goto done;
2927
2928         err = 0;
2929         for (ml = po->mclist; ml; ml = ml->next) {
2930                 if (ml->ifindex == mreq->mr_ifindex &&
2931                     ml->type == mreq->mr_type &&
2932                     ml->alen == mreq->mr_alen &&
2933                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2934                         ml->count++;
2935                         /* Free the new element ... */
2936                         kfree(i);
2937                         goto done;
2938                 }
2939         }
2940
2941         i->type = mreq->mr_type;
2942         i->ifindex = mreq->mr_ifindex;
2943         i->alen = mreq->mr_alen;
2944         memcpy(i->addr, mreq->mr_address, i->alen);
2945         i->count = 1;
2946         i->next = po->mclist;
2947         po->mclist = i;
2948         err = packet_dev_mc(dev, i, 1);
2949         if (err) {
2950                 po->mclist = i->next;
2951                 kfree(i);
2952         }
2953
2954 done:
2955         rtnl_unlock();
2956         return err;
2957 }
2958
2959 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2960 {
2961         struct packet_mclist *ml, **mlp;
2962
2963         rtnl_lock();
2964
2965         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2966                 if (ml->ifindex == mreq->mr_ifindex &&
2967                     ml->type == mreq->mr_type &&
2968                     ml->alen == mreq->mr_alen &&
2969                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2970                         if (--ml->count == 0) {
2971                                 struct net_device *dev;
2972                                 *mlp = ml->next;
2973                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2974                                 if (dev)
2975                                         packet_dev_mc(dev, ml, -1);
2976                                 kfree(ml);
2977                         }
2978                         rtnl_unlock();
2979                         return 0;
2980                 }
2981         }
2982         rtnl_unlock();
2983         return -EADDRNOTAVAIL;
2984 }
2985
2986 static void packet_flush_mclist(struct sock *sk)
2987 {
2988         struct packet_sock *po = pkt_sk(sk);
2989         struct packet_mclist *ml;
2990
2991         if (!po->mclist)
2992                 return;
2993
2994         rtnl_lock();
2995         while ((ml = po->mclist) != NULL) {
2996                 struct net_device *dev;
2997
2998                 po->mclist = ml->next;
2999                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3000                 if (dev != NULL)
3001                         packet_dev_mc(dev, ml, -1);
3002                 kfree(ml);
3003         }
3004         rtnl_unlock();
3005 }
3006
3007 static int
3008 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3009 {
3010         struct sock *sk = sock->sk;
3011         struct packet_sock *po = pkt_sk(sk);
3012         int ret;
3013
3014         if (level != SOL_PACKET)
3015                 return -ENOPROTOOPT;
3016
3017         switch (optname) {
3018         case PACKET_ADD_MEMBERSHIP:
3019         case PACKET_DROP_MEMBERSHIP:
3020         {
3021                 struct packet_mreq_max mreq;
3022                 int len = optlen;
3023                 memset(&mreq, 0, sizeof(mreq));
3024                 if (len < sizeof(struct packet_mreq))
3025                         return -EINVAL;
3026                 if (len > sizeof(mreq))
3027                         len = sizeof(mreq);
3028                 if (copy_from_user(&mreq, optval, len))
3029                         return -EFAULT;
3030                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3031                         return -EINVAL;
3032                 if (optname == PACKET_ADD_MEMBERSHIP)
3033                         ret = packet_mc_add(sk, &mreq);
3034                 else
3035                         ret = packet_mc_drop(sk, &mreq);
3036                 return ret;
3037         }
3038
3039         case PACKET_RX_RING:
3040         case PACKET_TX_RING:
3041         {
3042                 union tpacket_req_u req_u;
3043                 int len;
3044
3045                 switch (po->tp_version) {
3046                 case TPACKET_V1:
3047                 case TPACKET_V2:
3048                         len = sizeof(req_u.req);
3049                         break;
3050                 case TPACKET_V3:
3051                 default:
3052                         len = sizeof(req_u.req3);
3053                         break;
3054                 }
3055                 if (optlen < len)
3056                         return -EINVAL;
3057                 if (pkt_sk(sk)->has_vnet_hdr)
3058                         return -EINVAL;
3059                 if (copy_from_user(&req_u.req, optval, len))
3060                         return -EFAULT;
3061                 return packet_set_ring(sk, &req_u, 0,
3062                         optname == PACKET_TX_RING);
3063         }
3064         case PACKET_COPY_THRESH:
3065         {
3066                 int val;
3067
3068                 if (optlen != sizeof(val))
3069                         return -EINVAL;
3070                 if (copy_from_user(&val, optval, sizeof(val)))
3071                         return -EFAULT;
3072
3073                 pkt_sk(sk)->copy_thresh = val;
3074                 return 0;
3075         }
3076         case PACKET_VERSION:
3077         {
3078                 int val;
3079
3080                 if (optlen != sizeof(val))
3081                         return -EINVAL;
3082                 if (copy_from_user(&val, optval, sizeof(val)))
3083                         return -EFAULT;
3084                 switch (val) {
3085                 case TPACKET_V1:
3086                 case TPACKET_V2:
3087                 case TPACKET_V3:
3088                         break;
3089                 default:
3090                         return -EINVAL;
3091                 }
3092                 lock_sock(sk);
3093                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3094                         ret = -EBUSY;
3095                 } else {
3096                         po->tp_version = val;
3097                         ret = 0;
3098                 }
3099                 release_sock(sk);
3100                 return ret;
3101         }
3102         case PACKET_RESERVE:
3103         {
3104                 unsigned int val;
3105
3106                 if (optlen != sizeof(val))
3107                         return -EINVAL;
3108                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3109                         return -EBUSY;
3110                 if (copy_from_user(&val, optval, sizeof(val)))
3111                         return -EFAULT;
3112                 po->tp_reserve = val;
3113                 return 0;
3114         }
3115         case PACKET_LOSS:
3116         {
3117                 unsigned int val;
3118
3119                 if (optlen != sizeof(val))
3120                         return -EINVAL;
3121                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3122                         return -EBUSY;
3123                 if (copy_from_user(&val, optval, sizeof(val)))
3124                         return -EFAULT;
3125                 po->tp_loss = !!val;
3126                 return 0;
3127         }
3128         case PACKET_AUXDATA:
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                 po->auxdata = !!val;
3138                 return 0;
3139         }
3140         case PACKET_ORIGDEV:
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
3149                 po->origdev = !!val;
3150                 return 0;
3151         }
3152         case PACKET_VNET_HDR:
3153         {
3154                 int val;
3155
3156                 if (sock->type != SOCK_RAW)
3157                         return -EINVAL;
3158                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3159                         return -EBUSY;
3160                 if (optlen < sizeof(val))
3161                         return -EINVAL;
3162                 if (copy_from_user(&val, optval, sizeof(val)))
3163                         return -EFAULT;
3164
3165                 po->has_vnet_hdr = !!val;
3166                 return 0;
3167         }
3168         case PACKET_TIMESTAMP:
3169         {
3170                 int val;
3171
3172                 if (optlen != sizeof(val))
3173                         return -EINVAL;
3174                 if (copy_from_user(&val, optval, sizeof(val)))
3175                         return -EFAULT;
3176
3177                 po->tp_tstamp = val;
3178                 return 0;
3179         }
3180         case PACKET_FANOUT:
3181         {
3182                 int val;
3183
3184                 if (optlen != sizeof(val))
3185                         return -EINVAL;