Merge branch 'master' of github.com:davem330/net
[linux-3.10.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 packet_type      prot_hook ____cacheline_aligned_in_smp;
299 };
300
301 #define PACKET_FANOUT_MAX       256
302
303 struct packet_fanout {
304 #ifdef CONFIG_NET_NS
305         struct net              *net;
306 #endif
307         unsigned int            num_members;
308         u16                     id;
309         u8                      type;
310         u8                      defrag;
311         atomic_t                rr_cur;
312         struct list_head        list;
313         struct sock             *arr[PACKET_FANOUT_MAX];
314         spinlock_t              lock;
315         atomic_t                sk_ref;
316         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
317 };
318
319 struct packet_skb_cb {
320         unsigned int origlen;
321         union {
322                 struct sockaddr_pkt pkt;
323                 struct sockaddr_ll ll;
324         } sa;
325 };
326
327 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
328
329 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
330 #define GET_PBLOCK_DESC(x, bid) \
331         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
333         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
334 #define GET_NEXT_PRB_BLK_NUM(x) \
335         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
336         ((x)->kactive_blk_num+1) : 0)
337
338 static inline struct packet_sock *pkt_sk(struct sock *sk)
339 {
340         return (struct packet_sock *)sk;
341 }
342
343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
345
346 /* register_prot_hook must be invoked with the po->bind_lock held,
347  * or from a context in which asynchronous accesses to the packet
348  * socket is not possible (packet_create()).
349  */
350 static void register_prot_hook(struct sock *sk)
351 {
352         struct packet_sock *po = pkt_sk(sk);
353         if (!po->running) {
354                 if (po->fanout)
355                         __fanout_link(sk, po);
356                 else
357                         dev_add_pack(&po->prot_hook);
358                 sock_hold(sk);
359                 po->running = 1;
360         }
361 }
362
363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
364  * held.   If the sync parameter is true, we will temporarily drop
365  * the po->bind_lock and do a synchronize_net to make sure no
366  * asynchronous packet processing paths still refer to the elements
367  * of po->prot_hook.  If the sync parameter is false, it is the
368  * callers responsibility to take care of this.
369  */
370 static void __unregister_prot_hook(struct sock *sk, bool sync)
371 {
372         struct packet_sock *po = pkt_sk(sk);
373
374         po->running = 0;
375         if (po->fanout)
376                 __fanout_unlink(sk, po);
377         else
378                 __dev_remove_pack(&po->prot_hook);
379         __sock_put(sk);
380
381         if (sync) {
382                 spin_unlock(&po->bind_lock);
383                 synchronize_net();
384                 spin_lock(&po->bind_lock);
385         }
386 }
387
388 static void unregister_prot_hook(struct sock *sk, bool sync)
389 {
390         struct packet_sock *po = pkt_sk(sk);
391
392         if (po->running)
393                 __unregister_prot_hook(sk, sync);
394 }
395
396 static inline __pure struct page *pgv_to_page(void *addr)
397 {
398         if (is_vmalloc_addr(addr))
399                 return vmalloc_to_page(addr);
400         return virt_to_page(addr);
401 }
402
403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
404 {
405         union {
406                 struct tpacket_hdr *h1;
407                 struct tpacket2_hdr *h2;
408                 void *raw;
409         } h;
410
411         h.raw = frame;
412         switch (po->tp_version) {
413         case TPACKET_V1:
414                 h.h1->tp_status = status;
415                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
416                 break;
417         case TPACKET_V2:
418                 h.h2->tp_status = status;
419                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
420                 break;
421         case TPACKET_V3:
422         default:
423                 WARN(1, "TPACKET version not supported.\n");
424                 BUG();
425         }
426
427         smp_wmb();
428 }
429
430 static int __packet_get_status(struct packet_sock *po, void *frame)
431 {
432         union {
433                 struct tpacket_hdr *h1;
434                 struct tpacket2_hdr *h2;
435                 void *raw;
436         } h;
437
438         smp_rmb();
439
440         h.raw = frame;
441         switch (po->tp_version) {
442         case TPACKET_V1:
443                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
444                 return h.h1->tp_status;
445         case TPACKET_V2:
446                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
447                 return h.h2->tp_status;
448         case TPACKET_V3:
449         default:
450                 WARN(1, "TPACKET version not supported.\n");
451                 BUG();
452                 return 0;
453         }
454 }
455
456 static void *packet_lookup_frame(struct packet_sock *po,
457                 struct packet_ring_buffer *rb,
458                 unsigned int position,
459                 int status)
460 {
461         unsigned int pg_vec_pos, frame_offset;
462         union {
463                 struct tpacket_hdr *h1;
464                 struct tpacket2_hdr *h2;
465                 void *raw;
466         } h;
467
468         pg_vec_pos = position / rb->frames_per_block;
469         frame_offset = position % rb->frames_per_block;
470
471         h.raw = rb->pg_vec[pg_vec_pos].buffer +
472                 (frame_offset * rb->frame_size);
473
474         if (status != __packet_get_status(po, h.raw))
475                 return NULL;
476
477         return h.raw;
478 }
479
480 static inline void *packet_current_frame(struct packet_sock *po,
481                 struct packet_ring_buffer *rb,
482                 int status)
483 {
484         return packet_lookup_frame(po, rb, rb->head, status);
485 }
486
487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
488 {
489         del_timer_sync(&pkc->retire_blk_timer);
490 }
491
492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
493                 int tx_ring,
494                 struct sk_buff_head *rb_queue)
495 {
496         struct tpacket_kbdq_core *pkc;
497
498         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
499
500         spin_lock(&rb_queue->lock);
501         pkc->delete_blk_timer = 1;
502         spin_unlock(&rb_queue->lock);
503
504         prb_del_retire_blk_timer(pkc);
505 }
506
507 static void prb_init_blk_timer(struct packet_sock *po,
508                 struct tpacket_kbdq_core *pkc,
509                 void (*func) (unsigned long))
510 {
511         init_timer(&pkc->retire_blk_timer);
512         pkc->retire_blk_timer.data = (long)po;
513         pkc->retire_blk_timer.function = func;
514         pkc->retire_blk_timer.expires = jiffies;
515 }
516
517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
518 {
519         struct tpacket_kbdq_core *pkc;
520
521         if (tx_ring)
522                 BUG();
523
524         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
525         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
526 }
527
528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
529                                 int blk_size_in_bytes)
530 {
531         struct net_device *dev;
532         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
533         struct ethtool_cmd ecmd;
534         int err;
535
536         rtnl_lock();
537         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
538         if (unlikely(!dev)) {
539                 rtnl_unlock();
540                 return DEFAULT_PRB_RETIRE_TOV;
541         }
542         err = __ethtool_get_settings(dev, &ecmd);
543         rtnl_unlock();
544         if (!err) {
545                 switch (ecmd.speed) {
546                 case SPEED_10000:
547                         msec = 1;
548                         div = 10000/1000;
549                         break;
550                 case SPEED_1000:
551                         msec = 1;
552                         div = 1000/1000;
553                         break;
554                 /*
555                  * If the link speed is so slow you don't really
556                  * need to worry about perf anyways
557                  */
558                 case SPEED_100:
559                 case SPEED_10:
560                 default:
561                         return DEFAULT_PRB_RETIRE_TOV;
562                 }
563         }
564
565         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
566
567         if (div)
568                 mbits /= div;
569
570         tmo = mbits * msec;
571
572         if (div)
573                 return tmo+1;
574         return tmo;
575 }
576
577 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
578                         union tpacket_req_u *req_u)
579 {
580         p1->feature_req_word = req_u->req3.tp_feature_req_word;
581 }
582
583 static void init_prb_bdqc(struct packet_sock *po,
584                         struct packet_ring_buffer *rb,
585                         struct pgv *pg_vec,
586                         union tpacket_req_u *req_u, int tx_ring)
587 {
588         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
589         struct tpacket_block_desc *pbd;
590
591         memset(p1, 0x0, sizeof(*p1));
592
593         p1->knxt_seq_num = 1;
594         p1->pkbdq = pg_vec;
595         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
596         p1->pkblk_start = (char *)pg_vec[0].buffer;
597         p1->kblk_size = req_u->req3.tp_block_size;
598         p1->knum_blocks = req_u->req3.tp_block_nr;
599         p1->hdrlen = po->tp_hdrlen;
600         p1->version = po->tp_version;
601         p1->last_kactive_blk_num = 0;
602         po->stats_u.stats3.tp_freeze_q_cnt = 0;
603         if (req_u->req3.tp_retire_blk_tov)
604                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
605         else
606                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
607                                                 req_u->req3.tp_block_size);
608         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
609         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
610
611         prb_init_ft_ops(p1, req_u);
612         prb_setup_retire_blk_timer(po, tx_ring);
613         prb_open_block(p1, pbd);
614 }
615
616 /*  Do NOT update the last_blk_num first.
617  *  Assumes sk_buff_head lock is held.
618  */
619 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
620 {
621         mod_timer(&pkc->retire_blk_timer,
622                         jiffies + pkc->tov_in_jiffies);
623         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
624 }
625
626 /*
627  * Timer logic:
628  * 1) We refresh the timer only when we open a block.
629  *    By doing this we don't waste cycles refreshing the timer
630  *        on packet-by-packet basis.
631  *
632  * With a 1MB block-size, on a 1Gbps line, it will take
633  * i) ~8 ms to fill a block + ii) memcpy etc.
634  * In this cut we are not accounting for the memcpy time.
635  *
636  * So, if the user sets the 'tmo' to 10ms then the timer
637  * will never fire while the block is still getting filled
638  * (which is what we want). However, the user could choose
639  * to close a block early and that's fine.
640  *
641  * But when the timer does fire, we check whether or not to refresh it.
642  * Since the tmo granularity is in msecs, it is not too expensive
643  * to refresh the timer, lets say every '8' msecs.
644  * Either the user can set the 'tmo' or we can derive it based on
645  * a) line-speed and b) block-size.
646  * prb_calc_retire_blk_tmo() calculates the tmo.
647  *
648  */
649 static void prb_retire_rx_blk_timer_expired(unsigned long data)
650 {
651         struct packet_sock *po = (struct packet_sock *)data;
652         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
653         unsigned int frozen;
654         struct tpacket_block_desc *pbd;
655
656         spin_lock(&po->sk.sk_receive_queue.lock);
657
658         frozen = prb_queue_frozen(pkc);
659         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
660
661         if (unlikely(pkc->delete_blk_timer))
662                 goto out;
663
664         /* We only need to plug the race when the block is partially filled.
665          * tpacket_rcv:
666          *              lock(); increment BLOCK_NUM_PKTS; unlock()
667          *              copy_bits() is in progress ...
668          *              timer fires on other cpu:
669          *              we can't retire the current block because copy_bits
670          *              is in progress.
671          *
672          */
673         if (BLOCK_NUM_PKTS(pbd)) {
674                 while (atomic_read(&pkc->blk_fill_in_prog)) {
675                         /* Waiting for skb_copy_bits to finish... */
676                         cpu_relax();
677                 }
678         }
679
680         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
681                 if (!frozen) {
682                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
683                         if (!prb_dispatch_next_block(pkc, po))
684                                 goto refresh_timer;
685                         else
686                                 goto out;
687                 } else {
688                         /* Case 1. Queue was frozen because user-space was
689                          *         lagging behind.
690                          */
691                         if (prb_curr_blk_in_use(pkc, pbd)) {
692                                 /*
693                                  * Ok, user-space is still behind.
694                                  * So just refresh the timer.
695                                  */
696                                 goto refresh_timer;
697                         } else {
698                                /* Case 2. queue was frozen,user-space caught up,
699                                 * now the link went idle && the timer fired.
700                                 * We don't have a block to close.So we open this
701                                 * block and restart the timer.
702                                 * opening a block thaws the queue,restarts timer
703                                 * Thawing/timer-refresh is a side effect.
704                                 */
705                                 prb_open_block(pkc, pbd);
706                                 goto out;
707                         }
708                 }
709         }
710
711 refresh_timer:
712         _prb_refresh_rx_retire_blk_timer(pkc);
713
714 out:
715         spin_unlock(&po->sk.sk_receive_queue.lock);
716 }
717
718 static inline void prb_flush_block(struct tpacket_kbdq_core *pkc1,
719                 struct tpacket_block_desc *pbd1, __u32 status)
720 {
721         /* Flush everything minus the block header */
722
723 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
724         u8 *start, *end;
725
726         start = (u8 *)pbd1;
727
728         /* Skip the block header(we know header WILL fit in 4K) */
729         start += PAGE_SIZE;
730
731         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
732         for (; start < end; start += PAGE_SIZE)
733                 flush_dcache_page(pgv_to_page(start));
734
735         smp_wmb();
736 #endif
737
738         /* Now update the block status. */
739
740         BLOCK_STATUS(pbd1) = status;
741
742         /* Flush the block header */
743
744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
745         start = (u8 *)pbd1;
746         flush_dcache_page(pgv_to_page(start));
747
748         smp_wmb();
749 #endif
750 }
751
752 /*
753  * Side effect:
754  *
755  * 1) flush the block
756  * 2) Increment active_blk_num
757  *
758  * Note:We DONT refresh the timer on purpose.
759  *      Because almost always the next block will be opened.
760  */
761 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
762                 struct tpacket_block_desc *pbd1,
763                 struct packet_sock *po, unsigned int stat)
764 {
765         __u32 status = TP_STATUS_USER | stat;
766
767         struct tpacket3_hdr *last_pkt;
768         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
769
770         if (po->stats.tp_drops)
771                 status |= TP_STATUS_LOSING;
772
773         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
774         last_pkt->tp_next_offset = 0;
775
776         /* Get the ts of the last pkt */
777         if (BLOCK_NUM_PKTS(pbd1)) {
778                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
779                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
780         } else {
781                 /* Ok, we tmo'd - so get the current time */
782                 struct timespec ts;
783                 getnstimeofday(&ts);
784                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
785                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
786         }
787
788         smp_wmb();
789
790         /* Flush the block */
791         prb_flush_block(pkc1, pbd1, status);
792
793         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
794 }
795
796 static inline void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
797 {
798         pkc->reset_pending_on_curr_blk = 0;
799 }
800
801 /*
802  * Side effect of opening a block:
803  *
804  * 1) prb_queue is thawed.
805  * 2) retire_blk_timer is refreshed.
806  *
807  */
808 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
809         struct tpacket_block_desc *pbd1)
810 {
811         struct timespec ts;
812         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
813
814         smp_rmb();
815
816         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
817
818                 /* We could have just memset this but we will lose the
819                  * flexibility of making the priv area sticky
820                  */
821                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
822                 BLOCK_NUM_PKTS(pbd1) = 0;
823                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
824                 getnstimeofday(&ts);
825                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827                 pkc1->pkblk_start = (char *)pbd1;
828                 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
829                 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
830                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
832                 pbd1->version = pkc1->version;
833                 pkc1->prev = pkc1->nxt_offset;
834                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
835                 prb_thaw_queue(pkc1);
836                 _prb_refresh_rx_retire_blk_timer(pkc1);
837
838                 smp_wmb();
839
840                 return;
841         }
842
843         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
844                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
845         dump_stack();
846         BUG();
847 }
848
849 /*
850  * Queue freeze logic:
851  * 1) Assume tp_block_nr = 8 blocks.
852  * 2) At time 't0', user opens Rx ring.
853  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854  * 4) user-space is either sleeping or processing block '0'.
855  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856  *    it will close block-7,loop around and try to fill block '0'.
857  *    call-flow:
858  *    __packet_lookup_frame_in_block
859  *      prb_retire_current_block()
860  *      prb_dispatch_next_block()
861  *        |->(BLOCK_STATUS == USER) evaluates to true
862  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
863  * 6) Now there are two cases:
864  *    6.1) Link goes idle right after the queue is frozen.
865  *         But remember, the last open_block() refreshed the timer.
866  *         When this timer expires,it will refresh itself so that we can
867  *         re-open block-0 in near future.
868  *    6.2) Link is busy and keeps on receiving packets. This is a simple
869  *         case and __packet_lookup_frame_in_block will check if block-0
870  *         is free and can now be re-used.
871  */
872 static inline void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873                                   struct packet_sock *po)
874 {
875         pkc->reset_pending_on_curr_blk = 1;
876         po->stats_u.stats3.tp_freeze_q_cnt++;
877 }
878
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880
881 /*
882  * If the next block is free then we will dispatch it
883  * and return a good offset.
884  * Else, we will freeze the queue.
885  * So, caller must check the return value.
886  */
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888                 struct packet_sock *po)
889 {
890         struct tpacket_block_desc *pbd;
891
892         smp_rmb();
893
894         /* 1. Get current block num */
895         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896
897         /* 2. If this block is currently in_use then freeze the queue */
898         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899                 prb_freeze_queue(pkc, po);
900                 return NULL;
901         }
902
903         /*
904          * 3.
905          * open this block and return the offset where the first packet
906          * needs to get stored.
907          */
908         prb_open_block(pkc, pbd);
909         return (void *)pkc->nxt_offset;
910 }
911
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913                 struct packet_sock *po, unsigned int status)
914 {
915         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916
917         /* retire/close the current block */
918         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919                 /*
920                  * Plug the case where copy_bits() is in progress on
921                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922                  * have space to copy the pkt in the current block and
923                  * called prb_retire_current_block()
924                  *
925                  * We don't need to worry about the TMO case because
926                  * the timer-handler already handled this case.
927                  */
928                 if (!(status & TP_STATUS_BLK_TMO)) {
929                         while (atomic_read(&pkc->blk_fill_in_prog)) {
930                                 /* Waiting for skb_copy_bits to finish... */
931                                 cpu_relax();
932                         }
933                 }
934                 prb_close_block(pkc, pbd, po, status);
935                 return;
936         }
937
938         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
939         dump_stack();
940         BUG();
941 }
942
943 static inline int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
944                                       struct tpacket_block_desc *pbd)
945 {
946         return TP_STATUS_USER & BLOCK_STATUS(pbd);
947 }
948
949 static inline int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
950 {
951         return pkc->reset_pending_on_curr_blk;
952 }
953
954 static inline void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
955 {
956         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
957         atomic_dec(&pkc->blk_fill_in_prog);
958 }
959
960 static inline void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
961                         struct tpacket3_hdr *ppd)
962 {
963         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
964 }
965
966 static inline void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
967                         struct tpacket3_hdr *ppd)
968 {
969         ppd->hv1.tp_rxhash = 0;
970 }
971
972 static inline void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
973                         struct tpacket3_hdr *ppd)
974 {
975         if (vlan_tx_tag_present(pkc->skb)) {
976                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
977                 ppd->tp_status = TP_STATUS_VLAN_VALID;
978         } else {
979                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
980         }
981 }
982
983 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
984                         struct tpacket3_hdr *ppd)
985 {
986         prb_fill_vlan_info(pkc, ppd);
987
988         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989                 prb_fill_rxhash(pkc, ppd);
990         else
991                 prb_clear_rxhash(pkc, ppd);
992 }
993
994 static inline void prb_fill_curr_block(char *curr,
995                                 struct tpacket_kbdq_core *pkc,
996                                 struct tpacket_block_desc *pbd,
997                                 unsigned int len)
998 {
999         struct tpacket3_hdr *ppd;
1000
1001         ppd  = (struct tpacket3_hdr *)curr;
1002         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1003         pkc->prev = curr;
1004         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006         BLOCK_NUM_PKTS(pbd) += 1;
1007         atomic_inc(&pkc->blk_fill_in_prog);
1008         prb_run_all_ft_ops(pkc, ppd);
1009 }
1010
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013                                             struct sk_buff *skb,
1014                                                 int status,
1015                                             unsigned int len
1016                                             )
1017 {
1018         struct tpacket_kbdq_core *pkc;
1019         struct tpacket_block_desc *pbd;
1020         char *curr, *end;
1021
1022         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1023         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
1025         /* Queue is frozen when user space is lagging behind */
1026         if (prb_queue_frozen(pkc)) {
1027                 /*
1028                  * Check if that last block which caused the queue to freeze,
1029                  * is still in_use by user-space.
1030                  */
1031                 if (prb_curr_blk_in_use(pkc, pbd)) {
1032                         /* Can't record this packet */
1033                         return NULL;
1034                 } else {
1035                         /*
1036                          * Ok, the block was released by user-space.
1037                          * Now let's open that block.
1038                          * opening a block also thaws the queue.
1039                          * Thawing is a side effect.
1040                          */
1041                         prb_open_block(pkc, pbd);
1042                 }
1043         }
1044
1045         smp_mb();
1046         curr = pkc->nxt_offset;
1047         pkc->skb = skb;
1048         end = (char *) ((char *)pbd + pkc->kblk_size);
1049
1050         /* first try the current block */
1051         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052                 prb_fill_curr_block(curr, pkc, pbd, len);
1053                 return (void *)curr;
1054         }
1055
1056         /* Ok, close the current block */
1057         prb_retire_current_block(pkc, po, 0);
1058
1059         /* Now, try to dispatch the next block */
1060         curr = (char *)prb_dispatch_next_block(pkc, po);
1061         if (curr) {
1062                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063                 prb_fill_curr_block(curr, pkc, pbd, len);
1064                 return (void *)curr;
1065         }
1066
1067         /*
1068          * No free blocks are available.user_space hasn't caught up yet.
1069          * Queue was just frozen and now this packet will get dropped.
1070          */
1071         return NULL;
1072 }
1073
1074 static inline void *packet_current_rx_frame(struct packet_sock *po,
1075                                             struct sk_buff *skb,
1076                                             int status, unsigned int len)
1077 {
1078         char *curr = NULL;
1079         switch (po->tp_version) {
1080         case TPACKET_V1:
1081         case TPACKET_V2:
1082                 curr = packet_lookup_frame(po, &po->rx_ring,
1083                                         po->rx_ring.head, status);
1084                 return curr;
1085         case TPACKET_V3:
1086                 return __packet_lookup_frame_in_block(po, skb, status, len);
1087         default:
1088                 WARN(1, "TPACKET version not supported\n");
1089                 BUG();
1090                 return 0;
1091         }
1092 }
1093
1094 static inline void *prb_lookup_block(struct packet_sock *po,
1095                                      struct packet_ring_buffer *rb,
1096                                      unsigned int previous,
1097                                      int status)
1098 {
1099         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1100         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1101
1102         if (status != BLOCK_STATUS(pbd))
1103                 return NULL;
1104         return pbd;
1105 }
1106
1107 static inline int prb_previous_blk_num(struct packet_ring_buffer *rb)
1108 {
1109         unsigned int prev;
1110         if (rb->prb_bdqc.kactive_blk_num)
1111                 prev = rb->prb_bdqc.kactive_blk_num-1;
1112         else
1113                 prev = rb->prb_bdqc.knum_blocks-1;
1114         return prev;
1115 }
1116
1117 /* Assumes caller has held the rx_queue.lock */
1118 static inline void *__prb_previous_block(struct packet_sock *po,
1119                                          struct packet_ring_buffer *rb,
1120                                          int status)
1121 {
1122         unsigned int previous = prb_previous_blk_num(rb);
1123         return prb_lookup_block(po, rb, previous, status);
1124 }
1125
1126 static inline void *packet_previous_rx_frame(struct packet_sock *po,
1127                                              struct packet_ring_buffer *rb,
1128                                              int status)
1129 {
1130         if (po->tp_version <= TPACKET_V2)
1131                 return packet_previous_frame(po, rb, status);
1132
1133         return __prb_previous_block(po, rb, status);
1134 }
1135
1136 static inline void packet_increment_rx_head(struct packet_sock *po,
1137                                             struct packet_ring_buffer *rb)
1138 {
1139         switch (po->tp_version) {
1140         case TPACKET_V1:
1141         case TPACKET_V2:
1142                 return packet_increment_head(rb);
1143         case TPACKET_V3:
1144         default:
1145                 WARN(1, "TPACKET version not supported.\n");
1146                 BUG();
1147                 return;
1148         }
1149 }
1150
1151 static inline void *packet_previous_frame(struct packet_sock *po,
1152                 struct packet_ring_buffer *rb,
1153                 int status)
1154 {
1155         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156         return packet_lookup_frame(po, rb, previous, status);
1157 }
1158
1159 static inline void packet_increment_head(struct packet_ring_buffer *buff)
1160 {
1161         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162 }
1163
1164 static void packet_sock_destruct(struct sock *sk)
1165 {
1166         skb_queue_purge(&sk->sk_error_queue);
1167
1168         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1169         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1170
1171         if (!sock_flag(sk, SOCK_DEAD)) {
1172                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1173                 return;
1174         }
1175
1176         sk_refcnt_debug_dec(sk);
1177 }
1178
1179 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1180 {
1181         int x = atomic_read(&f->rr_cur) + 1;
1182
1183         if (x >= num)
1184                 x = 0;
1185
1186         return x;
1187 }
1188
1189 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1190 {
1191         u32 idx, hash = skb->rxhash;
1192
1193         idx = ((u64)hash * num) >> 32;
1194
1195         return f->arr[idx];
1196 }
1197
1198 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1199 {
1200         int cur, old;
1201
1202         cur = atomic_read(&f->rr_cur);
1203         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1204                                      fanout_rr_next(f, num))) != cur)
1205                 cur = old;
1206         return f->arr[cur];
1207 }
1208
1209 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1210 {
1211         unsigned int cpu = smp_processor_id();
1212
1213         return f->arr[cpu % num];
1214 }
1215
1216 static struct sk_buff *fanout_check_defrag(struct sk_buff *skb)
1217 {
1218 #ifdef CONFIG_INET
1219         const struct iphdr *iph;
1220         u32 len;
1221
1222         if (skb->protocol != htons(ETH_P_IP))
1223                 return skb;
1224
1225         if (!pskb_may_pull(skb, sizeof(struct iphdr)))
1226                 return skb;
1227
1228         iph = ip_hdr(skb);
1229         if (iph->ihl < 5 || iph->version != 4)
1230                 return skb;
1231         if (!pskb_may_pull(skb, iph->ihl*4))
1232                 return skb;
1233         iph = ip_hdr(skb);
1234         len = ntohs(iph->tot_len);
1235         if (skb->len < len || len < (iph->ihl * 4))
1236                 return skb;
1237
1238         if (ip_is_fragment(ip_hdr(skb))) {
1239                 skb = skb_share_check(skb, GFP_ATOMIC);
1240                 if (skb) {
1241                         if (pskb_trim_rcsum(skb, len))
1242                                 return skb;
1243                         memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
1244                         if (ip_defrag(skb, IP_DEFRAG_AF_PACKET))
1245                                 return NULL;
1246                         skb->rxhash = 0;
1247                 }
1248         }
1249 #endif
1250         return skb;
1251 }
1252
1253 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1254                              struct packet_type *pt, struct net_device *orig_dev)
1255 {
1256         struct packet_fanout *f = pt->af_packet_priv;
1257         unsigned int num = f->num_members;
1258         struct packet_sock *po;
1259         struct sock *sk;
1260
1261         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1262             !num) {
1263                 kfree_skb(skb);
1264                 return 0;
1265         }
1266
1267         switch (f->type) {
1268         case PACKET_FANOUT_HASH:
1269         default:
1270                 if (f->defrag) {
1271                         skb = fanout_check_defrag(skb);
1272                         if (!skb)
1273                                 return 0;
1274                 }
1275                 skb_get_rxhash(skb);
1276                 sk = fanout_demux_hash(f, skb, num);
1277                 break;
1278         case PACKET_FANOUT_LB:
1279                 sk = fanout_demux_lb(f, skb, num);
1280                 break;
1281         case PACKET_FANOUT_CPU:
1282                 sk = fanout_demux_cpu(f, skb, num);
1283                 break;
1284         }
1285
1286         po = pkt_sk(sk);
1287
1288         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1289 }
1290
1291 static DEFINE_MUTEX(fanout_mutex);
1292 static LIST_HEAD(fanout_list);
1293
1294 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1295 {
1296         struct packet_fanout *f = po->fanout;
1297
1298         spin_lock(&f->lock);
1299         f->arr[f->num_members] = sk;
1300         smp_wmb();
1301         f->num_members++;
1302         spin_unlock(&f->lock);
1303 }
1304
1305 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1306 {
1307         struct packet_fanout *f = po->fanout;
1308         int i;
1309
1310         spin_lock(&f->lock);
1311         for (i = 0; i < f->num_members; i++) {
1312                 if (f->arr[i] == sk)
1313                         break;
1314         }
1315         BUG_ON(i >= f->num_members);
1316         f->arr[i] = f->arr[f->num_members - 1];
1317         f->num_members--;
1318         spin_unlock(&f->lock);
1319 }
1320
1321 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1322 {
1323         struct packet_sock *po = pkt_sk(sk);
1324         struct packet_fanout *f, *match;
1325         u8 type = type_flags & 0xff;
1326         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1327         int err;
1328
1329         switch (type) {
1330         case PACKET_FANOUT_HASH:
1331         case PACKET_FANOUT_LB:
1332         case PACKET_FANOUT_CPU:
1333                 break;
1334         default:
1335                 return -EINVAL;
1336         }
1337
1338         if (!po->running)
1339                 return -EINVAL;
1340
1341         if (po->fanout)
1342                 return -EALREADY;
1343
1344         mutex_lock(&fanout_mutex);
1345         match = NULL;
1346         list_for_each_entry(f, &fanout_list, list) {
1347                 if (f->id == id &&
1348                     read_pnet(&f->net) == sock_net(sk)) {
1349                         match = f;
1350                         break;
1351                 }
1352         }
1353         err = -EINVAL;
1354         if (match && match->defrag != defrag)
1355                 goto out;
1356         if (!match) {
1357                 err = -ENOMEM;
1358                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1359                 if (!match)
1360                         goto out;
1361                 write_pnet(&match->net, sock_net(sk));
1362                 match->id = id;
1363                 match->type = type;
1364                 match->defrag = defrag;
1365                 atomic_set(&match->rr_cur, 0);
1366                 INIT_LIST_HEAD(&match->list);
1367                 spin_lock_init(&match->lock);
1368                 atomic_set(&match->sk_ref, 0);
1369                 match->prot_hook.type = po->prot_hook.type;
1370                 match->prot_hook.dev = po->prot_hook.dev;
1371                 match->prot_hook.func = packet_rcv_fanout;
1372                 match->prot_hook.af_packet_priv = match;
1373                 dev_add_pack(&match->prot_hook);
1374                 list_add(&match->list, &fanout_list);
1375         }
1376         err = -EINVAL;
1377         if (match->type == type &&
1378             match->prot_hook.type == po->prot_hook.type &&
1379             match->prot_hook.dev == po->prot_hook.dev) {
1380                 err = -ENOSPC;
1381                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1382                         __dev_remove_pack(&po->prot_hook);
1383                         po->fanout = match;
1384                         atomic_inc(&match->sk_ref);
1385                         __fanout_link(sk, po);
1386                         err = 0;
1387                 }
1388         }
1389 out:
1390         mutex_unlock(&fanout_mutex);
1391         return err;
1392 }
1393
1394 static void fanout_release(struct sock *sk)
1395 {
1396         struct packet_sock *po = pkt_sk(sk);
1397         struct packet_fanout *f;
1398
1399         f = po->fanout;
1400         if (!f)
1401                 return;
1402
1403         po->fanout = NULL;
1404
1405         mutex_lock(&fanout_mutex);
1406         if (atomic_dec_and_test(&f->sk_ref)) {
1407                 list_del(&f->list);
1408                 dev_remove_pack(&f->prot_hook);
1409                 kfree(f);
1410         }
1411         mutex_unlock(&fanout_mutex);
1412 }
1413
1414 static const struct proto_ops packet_ops;
1415
1416 static const struct proto_ops packet_ops_spkt;
1417
1418 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1419                            struct packet_type *pt, struct net_device *orig_dev)
1420 {
1421         struct sock *sk;
1422         struct sockaddr_pkt *spkt;
1423
1424         /*
1425          *      When we registered the protocol we saved the socket in the data
1426          *      field for just this event.
1427          */
1428
1429         sk = pt->af_packet_priv;
1430
1431         /*
1432          *      Yank back the headers [hope the device set this
1433          *      right or kerboom...]
1434          *
1435          *      Incoming packets have ll header pulled,
1436          *      push it back.
1437          *
1438          *      For outgoing ones skb->data == skb_mac_header(skb)
1439          *      so that this procedure is noop.
1440          */
1441
1442         if (skb->pkt_type == PACKET_LOOPBACK)
1443                 goto out;
1444
1445         if (!net_eq(dev_net(dev), sock_net(sk)))
1446                 goto out;
1447
1448         skb = skb_share_check(skb, GFP_ATOMIC);
1449         if (skb == NULL)
1450                 goto oom;
1451
1452         /* drop any routing info */
1453         skb_dst_drop(skb);
1454
1455         /* drop conntrack reference */
1456         nf_reset(skb);
1457
1458         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1459
1460         skb_push(skb, skb->data - skb_mac_header(skb));
1461
1462         /*
1463          *      The SOCK_PACKET socket receives _all_ frames.
1464          */
1465
1466         spkt->spkt_family = dev->type;
1467         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1468         spkt->spkt_protocol = skb->protocol;
1469
1470         /*
1471          *      Charge the memory to the socket. This is done specifically
1472          *      to prevent sockets using all the memory up.
1473          */
1474
1475         if (sock_queue_rcv_skb(sk, skb) == 0)
1476                 return 0;
1477
1478 out:
1479         kfree_skb(skb);
1480 oom:
1481         return 0;
1482 }
1483
1484
1485 /*
1486  *      Output a raw packet to a device layer. This bypasses all the other
1487  *      protocol layers and you must therefore supply it with a complete frame
1488  */
1489
1490 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1491                                struct msghdr *msg, size_t len)
1492 {
1493         struct sock *sk = sock->sk;
1494         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1495         struct sk_buff *skb = NULL;
1496         struct net_device *dev;
1497         __be16 proto = 0;
1498         int err;
1499
1500         /*
1501          *      Get and verify the address.
1502          */
1503
1504         if (saddr) {
1505                 if (msg->msg_namelen < sizeof(struct sockaddr))
1506                         return -EINVAL;
1507                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1508                         proto = saddr->spkt_protocol;
1509         } else
1510                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1511
1512         /*
1513          *      Find the device first to size check it
1514          */
1515
1516         saddr->spkt_device[13] = 0;
1517 retry:
1518         rcu_read_lock();
1519         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1520         err = -ENODEV;
1521         if (dev == NULL)
1522                 goto out_unlock;
1523
1524         err = -ENETDOWN;
1525         if (!(dev->flags & IFF_UP))
1526                 goto out_unlock;
1527
1528         /*
1529          * You may not queue a frame bigger than the mtu. This is the lowest level
1530          * raw protocol and you must do your own fragmentation at this level.
1531          */
1532
1533         err = -EMSGSIZE;
1534         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1535                 goto out_unlock;
1536
1537         if (!skb) {
1538                 size_t reserved = LL_RESERVED_SPACE(dev);
1539                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1540
1541                 rcu_read_unlock();
1542                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1543                 if (skb == NULL)
1544                         return -ENOBUFS;
1545                 /* FIXME: Save some space for broken drivers that write a hard
1546                  * header at transmission time by themselves. PPP is the notable
1547                  * one here. This should really be fixed at the driver level.
1548                  */
1549                 skb_reserve(skb, reserved);
1550                 skb_reset_network_header(skb);
1551
1552                 /* Try to align data part correctly */
1553                 if (hhlen) {
1554                         skb->data -= hhlen;
1555                         skb->tail -= hhlen;
1556                         if (len < hhlen)
1557                                 skb_reset_network_header(skb);
1558                 }
1559                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1560                 if (err)
1561                         goto out_free;
1562                 goto retry;
1563         }
1564
1565         if (len > (dev->mtu + dev->hard_header_len)) {
1566                 /* Earlier code assumed this would be a VLAN pkt,
1567                  * double-check this now that we have the actual
1568                  * packet in hand.
1569                  */
1570                 struct ethhdr *ehdr;
1571                 skb_reset_mac_header(skb);
1572                 ehdr = eth_hdr(skb);
1573                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1574                         err = -EMSGSIZE;
1575                         goto out_unlock;
1576                 }
1577         }
1578
1579         skb->protocol = proto;
1580         skb->dev = dev;
1581         skb->priority = sk->sk_priority;
1582         skb->mark = sk->sk_mark;
1583         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1584         if (err < 0)
1585                 goto out_unlock;
1586
1587         dev_queue_xmit(skb);
1588         rcu_read_unlock();
1589         return len;
1590
1591 out_unlock:
1592         rcu_read_unlock();
1593 out_free:
1594         kfree_skb(skb);
1595         return err;
1596 }
1597
1598 static inline unsigned int run_filter(const struct sk_buff *skb,
1599                                       const struct sock *sk,
1600                                       unsigned int res)
1601 {
1602         struct sk_filter *filter;
1603
1604         rcu_read_lock();
1605         filter = rcu_dereference(sk->sk_filter);
1606         if (filter != NULL)
1607                 res = SK_RUN_FILTER(filter, skb);
1608         rcu_read_unlock();
1609
1610         return res;
1611 }
1612
1613 /*
1614  * This function makes lazy skb cloning in hope that most of packets
1615  * are discarded by BPF.
1616  *
1617  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1618  * and skb->cb are mangled. It works because (and until) packets
1619  * falling here are owned by current CPU. Output packets are cloned
1620  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1621  * sequencially, so that if we return skb to original state on exit,
1622  * we will not harm anyone.
1623  */
1624
1625 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1626                       struct packet_type *pt, struct net_device *orig_dev)
1627 {
1628         struct sock *sk;
1629         struct sockaddr_ll *sll;
1630         struct packet_sock *po;
1631         u8 *skb_head = skb->data;
1632         int skb_len = skb->len;
1633         unsigned int snaplen, res;
1634
1635         if (skb->pkt_type == PACKET_LOOPBACK)
1636                 goto drop;
1637
1638         sk = pt->af_packet_priv;
1639         po = pkt_sk(sk);
1640
1641         if (!net_eq(dev_net(dev), sock_net(sk)))
1642                 goto drop;
1643
1644         skb->dev = dev;
1645
1646         if (dev->header_ops) {
1647                 /* The device has an explicit notion of ll header,
1648                  * exported to higher levels.
1649                  *
1650                  * Otherwise, the device hides details of its frame
1651                  * structure, so that corresponding packet head is
1652                  * never delivered to user.
1653                  */
1654                 if (sk->sk_type != SOCK_DGRAM)
1655                         skb_push(skb, skb->data - skb_mac_header(skb));
1656                 else if (skb->pkt_type == PACKET_OUTGOING) {
1657                         /* Special case: outgoing packets have ll header at head */
1658                         skb_pull(skb, skb_network_offset(skb));
1659                 }
1660         }
1661
1662         snaplen = skb->len;
1663
1664         res = run_filter(skb, sk, snaplen);
1665         if (!res)
1666                 goto drop_n_restore;
1667         if (snaplen > res)
1668                 snaplen = res;
1669
1670         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1671             (unsigned)sk->sk_rcvbuf)
1672                 goto drop_n_acct;
1673
1674         if (skb_shared(skb)) {
1675                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1676                 if (nskb == NULL)
1677                         goto drop_n_acct;
1678
1679                 if (skb_head != skb->data) {
1680                         skb->data = skb_head;
1681                         skb->len = skb_len;
1682                 }
1683                 kfree_skb(skb);
1684                 skb = nskb;
1685         }
1686
1687         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1688                      sizeof(skb->cb));
1689
1690         sll = &PACKET_SKB_CB(skb)->sa.ll;
1691         sll->sll_family = AF_PACKET;
1692         sll->sll_hatype = dev->type;
1693         sll->sll_protocol = skb->protocol;
1694         sll->sll_pkttype = skb->pkt_type;
1695         if (unlikely(po->origdev))
1696                 sll->sll_ifindex = orig_dev->ifindex;
1697         else
1698                 sll->sll_ifindex = dev->ifindex;
1699
1700         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1701
1702         PACKET_SKB_CB(skb)->origlen = skb->len;
1703
1704         if (pskb_trim(skb, snaplen))
1705                 goto drop_n_acct;
1706
1707         skb_set_owner_r(skb, sk);
1708         skb->dev = NULL;
1709         skb_dst_drop(skb);
1710
1711         /* drop conntrack reference */
1712         nf_reset(skb);
1713
1714         spin_lock(&sk->sk_receive_queue.lock);
1715         po->stats.tp_packets++;
1716         skb->dropcount = atomic_read(&sk->sk_drops);
1717         __skb_queue_tail(&sk->sk_receive_queue, skb);
1718         spin_unlock(&sk->sk_receive_queue.lock);
1719         sk->sk_data_ready(sk, skb->len);
1720         return 0;
1721
1722 drop_n_acct:
1723         spin_lock(&sk->sk_receive_queue.lock);
1724         po->stats.tp_drops++;
1725         atomic_inc(&sk->sk_drops);
1726         spin_unlock(&sk->sk_receive_queue.lock);
1727
1728 drop_n_restore:
1729         if (skb_head != skb->data && skb_shared(skb)) {
1730                 skb->data = skb_head;
1731                 skb->len = skb_len;
1732         }
1733 drop:
1734         consume_skb(skb);
1735         return 0;
1736 }
1737
1738 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1739                        struct packet_type *pt, struct net_device *orig_dev)
1740 {
1741         struct sock *sk;
1742         struct packet_sock *po;
1743         struct sockaddr_ll *sll;
1744         union {
1745                 struct tpacket_hdr *h1;
1746                 struct tpacket2_hdr *h2;
1747                 struct tpacket3_hdr *h3;
1748                 void *raw;
1749         } h;
1750         u8 *skb_head = skb->data;
1751         int skb_len = skb->len;
1752         unsigned int snaplen, res;
1753         unsigned long status = TP_STATUS_USER;
1754         unsigned short macoff, netoff, hdrlen;
1755         struct sk_buff *copy_skb = NULL;
1756         struct timeval tv;
1757         struct timespec ts;
1758         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1759
1760         if (skb->pkt_type == PACKET_LOOPBACK)
1761                 goto drop;
1762
1763         sk = pt->af_packet_priv;
1764         po = pkt_sk(sk);
1765
1766         if (!net_eq(dev_net(dev), sock_net(sk)))
1767                 goto drop;
1768
1769         if (dev->header_ops) {
1770                 if (sk->sk_type != SOCK_DGRAM)
1771                         skb_push(skb, skb->data - skb_mac_header(skb));
1772                 else if (skb->pkt_type == PACKET_OUTGOING) {
1773                         /* Special case: outgoing packets have ll header at head */
1774                         skb_pull(skb, skb_network_offset(skb));
1775                 }
1776         }
1777
1778         if (skb->ip_summed == CHECKSUM_PARTIAL)
1779                 status |= TP_STATUS_CSUMNOTREADY;
1780
1781         snaplen = skb->len;
1782
1783         res = run_filter(skb, sk, snaplen);
1784         if (!res)
1785                 goto drop_n_restore;
1786         if (snaplen > res)
1787                 snaplen = res;
1788
1789         if (sk->sk_type == SOCK_DGRAM) {
1790                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1791                                   po->tp_reserve;
1792         } else {
1793                 unsigned maclen = skb_network_offset(skb);
1794                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1795                                        (maclen < 16 ? 16 : maclen)) +
1796                         po->tp_reserve;
1797                 macoff = netoff - maclen;
1798         }
1799         if (po->tp_version <= TPACKET_V2) {
1800                 if (macoff + snaplen > po->rx_ring.frame_size) {
1801                         if (po->copy_thresh &&
1802                                 atomic_read(&sk->sk_rmem_alloc) + skb->truesize
1803                                 < (unsigned)sk->sk_rcvbuf) {
1804                                 if (skb_shared(skb)) {
1805                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1806                                 } else {
1807                                         copy_skb = skb_get(skb);
1808                                         skb_head = skb->data;
1809                                 }
1810                                 if (copy_skb)
1811                                         skb_set_owner_r(copy_skb, sk);
1812                         }
1813                         snaplen = po->rx_ring.frame_size - macoff;
1814                         if ((int)snaplen < 0)
1815                                 snaplen = 0;
1816                 }
1817         }
1818         spin_lock(&sk->sk_receive_queue.lock);
1819         h.raw = packet_current_rx_frame(po, skb,
1820                                         TP_STATUS_KERNEL, (macoff+snaplen));
1821         if (!h.raw)
1822                 goto ring_is_full;
1823         if (po->tp_version <= TPACKET_V2) {
1824                 packet_increment_rx_head(po, &po->rx_ring);
1825         /*
1826          * LOSING will be reported till you read the stats,
1827          * because it's COR - Clear On Read.
1828          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1829          * at packet level.
1830          */
1831                 if (po->stats.tp_drops)
1832                         status |= TP_STATUS_LOSING;
1833         }
1834         po->stats.tp_packets++;
1835         if (copy_skb) {
1836                 status |= TP_STATUS_COPY;
1837                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1838         }
1839         spin_unlock(&sk->sk_receive_queue.lock);
1840
1841         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1842
1843         switch (po->tp_version) {
1844         case TPACKET_V1:
1845                 h.h1->tp_len = skb->len;
1846                 h.h1->tp_snaplen = snaplen;
1847                 h.h1->tp_mac = macoff;
1848                 h.h1->tp_net = netoff;
1849                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1850                                 && shhwtstamps->syststamp.tv64)
1851                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1852                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1853                                 && shhwtstamps->hwtstamp.tv64)
1854                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1855                 else if (skb->tstamp.tv64)
1856                         tv = ktime_to_timeval(skb->tstamp);
1857                 else
1858                         do_gettimeofday(&tv);
1859                 h.h1->tp_sec = tv.tv_sec;
1860                 h.h1->tp_usec = tv.tv_usec;
1861                 hdrlen = sizeof(*h.h1);
1862                 break;
1863         case TPACKET_V2:
1864                 h.h2->tp_len = skb->len;
1865                 h.h2->tp_snaplen = snaplen;
1866                 h.h2->tp_mac = macoff;
1867                 h.h2->tp_net = netoff;
1868                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1869                                 && shhwtstamps->syststamp.tv64)
1870                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1871                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1872                                 && shhwtstamps->hwtstamp.tv64)
1873                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1874                 else if (skb->tstamp.tv64)
1875                         ts = ktime_to_timespec(skb->tstamp);
1876                 else
1877                         getnstimeofday(&ts);
1878                 h.h2->tp_sec = ts.tv_sec;
1879                 h.h2->tp_nsec = ts.tv_nsec;
1880                 if (vlan_tx_tag_present(skb)) {
1881                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1882                         status |= TP_STATUS_VLAN_VALID;
1883                 } else {
1884                         h.h2->tp_vlan_tci = 0;
1885                 }
1886                 h.h2->tp_padding = 0;
1887                 hdrlen = sizeof(*h.h2);
1888                 break;
1889         case TPACKET_V3:
1890                 /* tp_nxt_offset,vlan are already populated above.
1891                  * So DONT clear those fields here
1892                  */
1893                 h.h3->tp_status |= status;
1894                 h.h3->tp_len = skb->len;
1895                 h.h3->tp_snaplen = snaplen;
1896                 h.h3->tp_mac = macoff;
1897                 h.h3->tp_net = netoff;
1898                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1899                                 && shhwtstamps->syststamp.tv64)
1900                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1901                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1902                                 && shhwtstamps->hwtstamp.tv64)
1903                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1904                 else if (skb->tstamp.tv64)
1905                         ts = ktime_to_timespec(skb->tstamp);
1906                 else
1907                         getnstimeofday(&ts);
1908                 h.h3->tp_sec  = ts.tv_sec;
1909                 h.h3->tp_nsec = ts.tv_nsec;
1910                 hdrlen = sizeof(*h.h3);
1911                 break;
1912         default:
1913                 BUG();
1914         }
1915
1916         sll = h.raw + TPACKET_ALIGN(hdrlen);
1917         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1918         sll->sll_family = AF_PACKET;
1919         sll->sll_hatype = dev->type;
1920         sll->sll_protocol = skb->protocol;
1921         sll->sll_pkttype = skb->pkt_type;
1922         if (unlikely(po->origdev))
1923                 sll->sll_ifindex = orig_dev->ifindex;
1924         else
1925                 sll->sll_ifindex = dev->ifindex;
1926
1927         smp_mb();
1928 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1929         {
1930                 u8 *start, *end;
1931
1932                 if (po->tp_version <= TPACKET_V2) {
1933                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1934                                 + macoff + snaplen);
1935                         for (start = h.raw; start < end; start += PAGE_SIZE)
1936                                 flush_dcache_page(pgv_to_page(start));
1937                 }
1938                 smp_wmb();
1939         }
1940 #endif
1941         if (po->tp_version <= TPACKET_V2)
1942                 __packet_set_status(po, h.raw, status);
1943         else
1944                 prb_clear_blk_fill_status(&po->rx_ring);
1945
1946         sk->sk_data_ready(sk, 0);
1947
1948 drop_n_restore:
1949         if (skb_head != skb->data && skb_shared(skb)) {
1950                 skb->data = skb_head;
1951                 skb->len = skb_len;
1952         }
1953 drop:
1954         kfree_skb(skb);
1955         return 0;
1956
1957 ring_is_full:
1958         po->stats.tp_drops++;
1959         spin_unlock(&sk->sk_receive_queue.lock);
1960
1961         sk->sk_data_ready(sk, 0);
1962         kfree_skb(copy_skb);
1963         goto drop_n_restore;
1964 }
1965
1966 static void tpacket_destruct_skb(struct sk_buff *skb)
1967 {
1968         struct packet_sock *po = pkt_sk(skb->sk);
1969         void *ph;
1970
1971         BUG_ON(skb == NULL);
1972
1973         if (likely(po->tx_ring.pg_vec)) {
1974                 ph = skb_shinfo(skb)->destructor_arg;
1975                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1976                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1977                 atomic_dec(&po->tx_ring.pending);
1978                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1979         }
1980
1981         sock_wfree(skb);
1982 }
1983
1984 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1985                 void *frame, struct net_device *dev, int size_max,
1986                 __be16 proto, unsigned char *addr)
1987 {
1988         union {
1989                 struct tpacket_hdr *h1;
1990                 struct tpacket2_hdr *h2;
1991                 void *raw;
1992         } ph;
1993         int to_write, offset, len, tp_len, nr_frags, len_max;
1994         struct socket *sock = po->sk.sk_socket;
1995         struct page *page;
1996         void *data;
1997         int err;
1998
1999         ph.raw = frame;
2000
2001         skb->protocol = proto;
2002         skb->dev = dev;
2003         skb->priority = po->sk.sk_priority;
2004         skb->mark = po->sk.sk_mark;
2005         skb_shinfo(skb)->destructor_arg = ph.raw;
2006
2007         switch (po->tp_version) {
2008         case TPACKET_V2:
2009                 tp_len = ph.h2->tp_len;
2010                 break;
2011         default:
2012                 tp_len = ph.h1->tp_len;
2013                 break;
2014         }
2015         if (unlikely(tp_len > size_max)) {
2016                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2017                 return -EMSGSIZE;
2018         }
2019
2020         skb_reserve(skb, LL_RESERVED_SPACE(dev));
2021         skb_reset_network_header(skb);
2022
2023         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2024         to_write = tp_len;
2025
2026         if (sock->type == SOCK_DGRAM) {
2027                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2028                                 NULL, tp_len);
2029                 if (unlikely(err < 0))
2030                         return -EINVAL;
2031         } else if (dev->hard_header_len) {
2032                 /* net device doesn't like empty head */
2033                 if (unlikely(tp_len <= dev->hard_header_len)) {
2034                         pr_err("packet size is too short (%d < %d)\n",
2035                                tp_len, dev->hard_header_len);
2036                         return -EINVAL;
2037                 }
2038
2039                 skb_push(skb, dev->hard_header_len);
2040                 err = skb_store_bits(skb, 0, data,
2041                                 dev->hard_header_len);
2042                 if (unlikely(err))
2043                         return err;
2044
2045                 data += dev->hard_header_len;
2046                 to_write -= dev->hard_header_len;
2047         }
2048
2049         err = -EFAULT;
2050         offset = offset_in_page(data);
2051         len_max = PAGE_SIZE - offset;
2052         len = ((to_write > len_max) ? len_max : to_write);
2053
2054         skb->data_len = to_write;
2055         skb->len += to_write;
2056         skb->truesize += to_write;
2057         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2058
2059         while (likely(to_write)) {
2060                 nr_frags = skb_shinfo(skb)->nr_frags;
2061
2062                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2063                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2064                                MAX_SKB_FRAGS);
2065                         return -EFAULT;
2066                 }
2067
2068                 page = pgv_to_page(data);
2069                 data += len;
2070                 flush_dcache_page(page);
2071                 get_page(page);
2072                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2073                 to_write -= len;
2074                 offset = 0;
2075                 len_max = PAGE_SIZE;
2076                 len = ((to_write > len_max) ? len_max : to_write);
2077         }
2078
2079         return tp_len;
2080 }
2081
2082 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2083 {
2084         struct sk_buff *skb;
2085         struct net_device *dev;
2086         __be16 proto;
2087         bool need_rls_dev = false;
2088         int err, reserve = 0;
2089         void *ph;
2090         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2091         int tp_len, size_max;
2092         unsigned char *addr;
2093         int len_sum = 0;
2094         int status = 0;
2095
2096         mutex_lock(&po->pg_vec_lock);
2097
2098         err = -EBUSY;
2099         if (saddr == NULL) {
2100                 dev = po->prot_hook.dev;
2101                 proto   = po->num;
2102                 addr    = NULL;
2103         } else {
2104                 err = -EINVAL;
2105                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2106                         goto out;
2107                 if (msg->msg_namelen < (saddr->sll_halen
2108                                         + offsetof(struct sockaddr_ll,
2109                                                 sll_addr)))
2110                         goto out;
2111                 proto   = saddr->sll_protocol;
2112                 addr    = saddr->sll_addr;
2113                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2114                 need_rls_dev = true;
2115         }
2116
2117         err = -ENXIO;
2118         if (unlikely(dev == NULL))
2119                 goto out;
2120
2121         reserve = dev->hard_header_len;
2122
2123         err = -ENETDOWN;
2124         if (unlikely(!(dev->flags & IFF_UP)))
2125                 goto out_put;
2126
2127         size_max = po->tx_ring.frame_size
2128                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2129
2130         if (size_max > dev->mtu + reserve)
2131                 size_max = dev->mtu + reserve;
2132
2133         do {
2134                 ph = packet_current_frame(po, &po->tx_ring,
2135                                 TP_STATUS_SEND_REQUEST);
2136
2137                 if (unlikely(ph == NULL)) {
2138                         schedule();
2139                         continue;
2140                 }
2141
2142                 status = TP_STATUS_SEND_REQUEST;
2143                 skb = sock_alloc_send_skb(&po->sk,
2144                                 LL_ALLOCATED_SPACE(dev)
2145                                 + sizeof(struct sockaddr_ll),
2146                                 0, &err);
2147
2148                 if (unlikely(skb == NULL))
2149                         goto out_status;
2150
2151                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2152                                 addr);
2153
2154                 if (unlikely(tp_len < 0)) {
2155                         if (po->tp_loss) {
2156                                 __packet_set_status(po, ph,
2157                                                 TP_STATUS_AVAILABLE);
2158                                 packet_increment_head(&po->tx_ring);
2159                                 kfree_skb(skb);
2160                                 continue;
2161                         } else {
2162                                 status = TP_STATUS_WRONG_FORMAT;
2163                                 err = tp_len;
2164                                 goto out_status;
2165                         }
2166                 }
2167
2168                 skb->destructor = tpacket_destruct_skb;
2169                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2170                 atomic_inc(&po->tx_ring.pending);
2171
2172                 status = TP_STATUS_SEND_REQUEST;
2173                 err = dev_queue_xmit(skb);
2174                 if (unlikely(err > 0)) {
2175                         err = net_xmit_errno(err);
2176                         if (err && __packet_get_status(po, ph) ==
2177                                    TP_STATUS_AVAILABLE) {
2178                                 /* skb was destructed already */
2179                                 skb = NULL;
2180                                 goto out_status;
2181                         }
2182                         /*
2183                          * skb was dropped but not destructed yet;
2184                          * let's treat it like congestion or err < 0
2185                          */
2186                         err = 0;
2187                 }
2188                 packet_increment_head(&po->tx_ring);
2189                 len_sum += tp_len;
2190         } while (likely((ph != NULL) ||
2191                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2192                          (atomic_read(&po->tx_ring.pending))))
2193                 );
2194
2195         err = len_sum;
2196         goto out_put;
2197
2198 out_status:
2199         __packet_set_status(po, ph, status);
2200         kfree_skb(skb);
2201 out_put:
2202         if (need_rls_dev)
2203                 dev_put(dev);
2204 out:
2205         mutex_unlock(&po->pg_vec_lock);
2206         return err;
2207 }
2208
2209 static inline struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2210                                                size_t reserve, size_t len,
2211                                                size_t linear, int noblock,
2212                                                int *err)
2213 {
2214         struct sk_buff *skb;
2215
2216         /* Under a page?  Don't bother with paged skb. */
2217         if (prepad + len < PAGE_SIZE || !linear)
2218                 linear = len;
2219
2220         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2221                                    err);
2222         if (!skb)
2223                 return NULL;
2224
2225         skb_reserve(skb, reserve);
2226         skb_put(skb, linear);
2227         skb->data_len = len - linear;
2228         skb->len += len - linear;
2229
2230         return skb;
2231 }
2232
2233 static int packet_snd(struct socket *sock,
2234                           struct msghdr *msg, size_t len)
2235 {
2236         struct sock *sk = sock->sk;
2237         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2238         struct sk_buff *skb;
2239         struct net_device *dev;
2240         __be16 proto;
2241         bool need_rls_dev = false;
2242         unsigned char *addr;
2243         int err, reserve = 0;
2244         struct virtio_net_hdr vnet_hdr = { 0 };
2245         int offset = 0;
2246         int vnet_hdr_len;
2247         struct packet_sock *po = pkt_sk(sk);
2248         unsigned short gso_type = 0;
2249
2250         /*
2251          *      Get and verify the address.
2252          */
2253
2254         if (saddr == NULL) {
2255                 dev = po->prot_hook.dev;
2256                 proto   = po->num;
2257                 addr    = NULL;
2258         } else {
2259                 err = -EINVAL;
2260                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2261                         goto out;
2262                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2263                         goto out;
2264                 proto   = saddr->sll_protocol;
2265                 addr    = saddr->sll_addr;
2266                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2267                 need_rls_dev = true;
2268         }
2269
2270         err = -ENXIO;
2271         if (dev == NULL)
2272                 goto out_unlock;
2273         if (sock->type == SOCK_RAW)
2274                 reserve = dev->hard_header_len;
2275
2276         err = -ENETDOWN;
2277         if (!(dev->flags & IFF_UP))
2278                 goto out_unlock;
2279
2280         if (po->has_vnet_hdr) {
2281                 vnet_hdr_len = sizeof(vnet_hdr);
2282
2283                 err = -EINVAL;
2284                 if (len < vnet_hdr_len)
2285                         goto out_unlock;
2286
2287                 len -= vnet_hdr_len;
2288
2289                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2290                                        vnet_hdr_len);
2291                 if (err < 0)
2292                         goto out_unlock;
2293
2294                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2295                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2296                       vnet_hdr.hdr_len))
2297                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2298                                                  vnet_hdr.csum_offset + 2;
2299
2300                 err = -EINVAL;
2301                 if (vnet_hdr.hdr_len > len)
2302                         goto out_unlock;
2303
2304                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2305                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2306                         case VIRTIO_NET_HDR_GSO_TCPV4:
2307                                 gso_type = SKB_GSO_TCPV4;
2308                                 break;
2309                         case VIRTIO_NET_HDR_GSO_TCPV6:
2310                                 gso_type = SKB_GSO_TCPV6;
2311                                 break;
2312                         case VIRTIO_NET_HDR_GSO_UDP:
2313                                 gso_type = SKB_GSO_UDP;
2314                                 break;
2315                         default:
2316                                 goto out_unlock;
2317                         }
2318
2319                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2320                                 gso_type |= SKB_GSO_TCP_ECN;
2321
2322                         if (vnet_hdr.gso_size == 0)
2323                                 goto out_unlock;
2324
2325                 }
2326         }
2327
2328         err = -EMSGSIZE;
2329         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2330                 goto out_unlock;
2331
2332         err = -ENOBUFS;
2333         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2334                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2335                                msg->msg_flags & MSG_DONTWAIT, &err);
2336         if (skb == NULL)
2337                 goto out_unlock;
2338
2339         skb_set_network_header(skb, reserve);
2340
2341         err = -EINVAL;
2342         if (sock->type == SOCK_DGRAM &&
2343             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2344                 goto out_free;
2345
2346         /* Returns -EFAULT on error */
2347         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2348         if (err)
2349                 goto out_free;
2350         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2351         if (err < 0)
2352                 goto out_free;
2353
2354         if (!gso_type && (len > dev->mtu + reserve)) {
2355                 /* Earlier code assumed this would be a VLAN pkt,
2356                  * double-check this now that we have the actual
2357                  * packet in hand.
2358                  */
2359                 struct ethhdr *ehdr;
2360                 skb_reset_mac_header(skb);
2361                 ehdr = eth_hdr(skb);
2362                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2363                         err = -EMSGSIZE;
2364                         goto out_free;
2365                 }
2366         }
2367
2368         skb->protocol = proto;
2369         skb->dev = dev;
2370         skb->priority = sk->sk_priority;
2371         skb->mark = sk->sk_mark;
2372
2373         if (po->has_vnet_hdr) {
2374                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2375                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2376                                                   vnet_hdr.csum_offset)) {
2377                                 err = -EINVAL;
2378                                 goto out_free;
2379                         }
2380                 }
2381
2382                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2383                 skb_shinfo(skb)->gso_type = gso_type;
2384
2385                 /* Header must be checked, and gso_segs computed. */
2386                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2387                 skb_shinfo(skb)->gso_segs = 0;
2388
2389                 len += vnet_hdr_len;
2390         }
2391
2392         /*
2393          *      Now send it
2394          */
2395
2396         err = dev_queue_xmit(skb);
2397         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2398                 goto out_unlock;
2399
2400         if (need_rls_dev)
2401                 dev_put(dev);
2402
2403         return len;
2404
2405 out_free:
2406         kfree_skb(skb);
2407 out_unlock:
2408         if (dev && need_rls_dev)
2409                 dev_put(dev);
2410 out:
2411         return err;
2412 }
2413
2414 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2415                 struct msghdr *msg, size_t len)
2416 {
2417         struct sock *sk = sock->sk;
2418         struct packet_sock *po = pkt_sk(sk);
2419         if (po->tx_ring.pg_vec)
2420                 return tpacket_snd(po, msg);
2421         else
2422                 return packet_snd(sock, msg, len);
2423 }
2424
2425 /*
2426  *      Close a PACKET socket. This is fairly simple. We immediately go
2427  *      to 'closed' state and remove our protocol entry in the device list.
2428  */
2429
2430 static int packet_release(struct socket *sock)
2431 {
2432         struct sock *sk = sock->sk;
2433         struct packet_sock *po;
2434         struct net *net;
2435         union tpacket_req_u req_u;
2436
2437         if (!sk)
2438                 return 0;
2439
2440         net = sock_net(sk);
2441         po = pkt_sk(sk);
2442
2443         spin_lock_bh(&net->packet.sklist_lock);
2444         sk_del_node_init_rcu(sk);
2445         sock_prot_inuse_add(net, sk->sk_prot, -1);
2446         spin_unlock_bh(&net->packet.sklist_lock);
2447
2448         spin_lock(&po->bind_lock);
2449         unregister_prot_hook(sk, false);
2450         if (po->prot_hook.dev) {
2451                 dev_put(po->prot_hook.dev);
2452                 po->prot_hook.dev = NULL;
2453         }
2454         spin_unlock(&po->bind_lock);
2455
2456         packet_flush_mclist(sk);
2457
2458         memset(&req_u, 0, sizeof(req_u));
2459
2460         if (po->rx_ring.pg_vec)
2461                 packet_set_ring(sk, &req_u, 1, 0);
2462
2463         if (po->tx_ring.pg_vec)
2464                 packet_set_ring(sk, &req_u, 1, 1);
2465
2466         fanout_release(sk);
2467
2468         synchronize_net();
2469         /*
2470          *      Now the socket is dead. No more input will appear.
2471          */
2472         sock_orphan(sk);
2473         sock->sk = NULL;
2474
2475         /* Purge queues */
2476
2477         skb_queue_purge(&sk->sk_receive_queue);
2478         sk_refcnt_debug_release(sk);
2479
2480         sock_put(sk);
2481         return 0;
2482 }
2483
2484 /*
2485  *      Attach a packet hook.
2486  */
2487
2488 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2489 {
2490         struct packet_sock *po = pkt_sk(sk);
2491
2492         if (po->fanout)
2493                 return -EINVAL;
2494
2495         lock_sock(sk);
2496
2497         spin_lock(&po->bind_lock);
2498         unregister_prot_hook(sk, true);
2499         po->num = protocol;
2500         po->prot_hook.type = protocol;
2501         if (po->prot_hook.dev)
2502                 dev_put(po->prot_hook.dev);
2503         po->prot_hook.dev = dev;
2504
2505         po->ifindex = dev ? dev->ifindex : 0;
2506
2507         if (protocol == 0)
2508                 goto out_unlock;
2509
2510         if (!dev || (dev->flags & IFF_UP)) {
2511                 register_prot_hook(sk);
2512         } else {
2513                 sk->sk_err = ENETDOWN;
2514                 if (!sock_flag(sk, SOCK_DEAD))
2515                         sk->sk_error_report(sk);
2516         }
2517
2518 out_unlock:
2519         spin_unlock(&po->bind_lock);
2520         release_sock(sk);
2521         return 0;
2522 }
2523
2524 /*
2525  *      Bind a packet socket to a device
2526  */
2527
2528 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2529                             int addr_len)
2530 {
2531         struct sock *sk = sock->sk;
2532         char name[15];
2533         struct net_device *dev;
2534         int err = -ENODEV;
2535
2536         /*
2537          *      Check legality
2538          */
2539
2540         if (addr_len != sizeof(struct sockaddr))
2541                 return -EINVAL;
2542         strlcpy(name, uaddr->sa_data, sizeof(name));
2543
2544         dev = dev_get_by_name(sock_net(sk), name);
2545         if (dev)
2546                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2547         return err;
2548 }
2549
2550 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2551 {
2552         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2553         struct sock *sk = sock->sk;
2554         struct net_device *dev = NULL;
2555         int err;
2556
2557
2558         /*
2559          *      Check legality
2560          */
2561
2562         if (addr_len < sizeof(struct sockaddr_ll))
2563                 return -EINVAL;
2564         if (sll->sll_family != AF_PACKET)
2565                 return -EINVAL;
2566
2567         if (sll->sll_ifindex) {
2568                 err = -ENODEV;
2569                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2570                 if (dev == NULL)
2571                         goto out;
2572         }
2573         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2574
2575 out:
2576         return err;
2577 }
2578
2579 static struct proto packet_proto = {
2580         .name     = "PACKET",
2581         .owner    = THIS_MODULE,
2582         .obj_size = sizeof(struct packet_sock),
2583 };
2584
2585 /*
2586  *      Create a packet of type SOCK_PACKET.
2587  */
2588
2589 static int packet_create(struct net *net, struct socket *sock, int protocol,
2590                          int kern)
2591 {
2592         struct sock *sk;
2593         struct packet_sock *po;
2594         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2595         int err;
2596
2597         if (!capable(CAP_NET_RAW))
2598                 return -EPERM;
2599         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2600             sock->type != SOCK_PACKET)
2601                 return -ESOCKTNOSUPPORT;
2602
2603         sock->state = SS_UNCONNECTED;
2604
2605         err = -ENOBUFS;
2606         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2607         if (sk == NULL)
2608                 goto out;
2609
2610         sock->ops = &packet_ops;
2611         if (sock->type == SOCK_PACKET)
2612                 sock->ops = &packet_ops_spkt;
2613
2614         sock_init_data(sock, sk);
2615
2616         po = pkt_sk(sk);
2617         sk->sk_family = PF_PACKET;
2618         po->num = proto;
2619
2620         sk->sk_destruct = packet_sock_destruct;
2621         sk_refcnt_debug_inc(sk);
2622
2623         /*
2624          *      Attach a protocol block
2625          */
2626
2627         spin_lock_init(&po->bind_lock);
2628         mutex_init(&po->pg_vec_lock);
2629         po->prot_hook.func = packet_rcv;
2630
2631         if (sock->type == SOCK_PACKET)
2632                 po->prot_hook.func = packet_rcv_spkt;
2633
2634         po->prot_hook.af_packet_priv = sk;
2635
2636         if (proto) {
2637                 po->prot_hook.type = proto;
2638                 register_prot_hook(sk);
2639         }
2640
2641         spin_lock_bh(&net->packet.sklist_lock);
2642         sk_add_node_rcu(sk, &net->packet.sklist);
2643         sock_prot_inuse_add(net, &packet_proto, 1);
2644         spin_unlock_bh(&net->packet.sklist_lock);
2645
2646         return 0;
2647 out:
2648         return err;
2649 }
2650
2651 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2652 {
2653         struct sock_exterr_skb *serr;
2654         struct sk_buff *skb, *skb2;
2655         int copied, err;
2656
2657         err = -EAGAIN;
2658         skb = skb_dequeue(&sk->sk_error_queue);
2659         if (skb == NULL)
2660                 goto out;
2661
2662         copied = skb->len;
2663         if (copied > len) {
2664                 msg->msg_flags |= MSG_TRUNC;
2665                 copied = len;
2666         }
2667         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2668         if (err)
2669                 goto out_free_skb;
2670
2671         sock_recv_timestamp(msg, sk, skb);
2672
2673         serr = SKB_EXT_ERR(skb);
2674         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2675                  sizeof(serr->ee), &serr->ee);
2676
2677         msg->msg_flags |= MSG_ERRQUEUE;
2678         err = copied;
2679
2680         /* Reset and regenerate socket error */
2681         spin_lock_bh(&sk->sk_error_queue.lock);
2682         sk->sk_err = 0;
2683         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2684                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2685                 spin_unlock_bh(&sk->sk_error_queue.lock);
2686                 sk->sk_error_report(sk);
2687         } else
2688                 spin_unlock_bh(&sk->sk_error_queue.lock);
2689
2690 out_free_skb:
2691         kfree_skb(skb);
2692 out:
2693         return err;
2694 }
2695
2696 /*
2697  *      Pull a packet from our receive queue and hand it to the user.
2698  *      If necessary we block.
2699  */
2700
2701 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2702                           struct msghdr *msg, size_t len, int flags)
2703 {
2704         struct sock *sk = sock->sk;
2705         struct sk_buff *skb;
2706         int copied, err;
2707         struct sockaddr_ll *sll;
2708         int vnet_hdr_len = 0;
2709
2710         err = -EINVAL;
2711         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2712                 goto out;
2713
2714 #if 0
2715         /* What error should we return now? EUNATTACH? */
2716         if (pkt_sk(sk)->ifindex < 0)
2717                 return -ENODEV;
2718 #endif
2719
2720         if (flags & MSG_ERRQUEUE) {
2721                 err = packet_recv_error(sk, msg, len);
2722                 goto out;
2723         }
2724
2725         /*
2726          *      Call the generic datagram receiver. This handles all sorts
2727          *      of horrible races and re-entrancy so we can forget about it
2728          *      in the protocol layers.
2729          *
2730          *      Now it will return ENETDOWN, if device have just gone down,
2731          *      but then it will block.
2732          */
2733
2734         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2735
2736         /*
2737          *      An error occurred so return it. Because skb_recv_datagram()
2738          *      handles the blocking we don't see and worry about blocking
2739          *      retries.
2740          */
2741
2742         if (skb == NULL)
2743                 goto out;
2744
2745         if (pkt_sk(sk)->has_vnet_hdr) {
2746                 struct virtio_net_hdr vnet_hdr = { 0 };
2747
2748                 err = -EINVAL;
2749                 vnet_hdr_len = sizeof(vnet_hdr);
2750                 if (len < vnet_hdr_len)
2751                         goto out_free;
2752
2753                 len -= vnet_hdr_len;
2754
2755                 if (skb_is_gso(skb)) {
2756                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2757
2758                         /* This is a hint as to how much should be linear. */
2759                         vnet_hdr.hdr_len = skb_headlen(skb);
2760                         vnet_hdr.gso_size = sinfo->gso_size;
2761                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2762                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2763                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2764                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2765                         else if (sinfo->gso_type & SKB_GSO_UDP)
2766                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2767                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2768                                 goto out_free;
2769                         else
2770                                 BUG();
2771                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2772                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2773                 } else
2774                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2775
2776                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2777                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2778                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2779                         vnet_hdr.csum_offset = skb->csum_offset;
2780                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2781                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2782                 } /* else everything is zero */
2783
2784                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2785                                      vnet_hdr_len);
2786                 if (err < 0)
2787                         goto out_free;
2788         }
2789
2790         /*
2791          *      If the address length field is there to be filled in, we fill
2792          *      it in now.
2793          */
2794
2795         sll = &PACKET_SKB_CB(skb)->sa.ll;
2796         if (sock->type == SOCK_PACKET)
2797                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2798         else
2799                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2800
2801         /*
2802          *      You lose any data beyond the buffer you gave. If it worries a
2803          *      user program they can ask the device for its MTU anyway.
2804          */
2805
2806         copied = skb->len;
2807         if (copied > len) {
2808                 copied = len;
2809                 msg->msg_flags |= MSG_TRUNC;
2810         }
2811
2812         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2813         if (err)
2814                 goto out_free;
2815
2816         sock_recv_ts_and_drops(msg, sk, skb);
2817
2818         if (msg->msg_name)
2819                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2820                        msg->msg_namelen);
2821
2822         if (pkt_sk(sk)->auxdata) {
2823                 struct tpacket_auxdata aux;
2824
2825                 aux.tp_status = TP_STATUS_USER;
2826                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2827                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2828                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2829                 aux.tp_snaplen = skb->len;
2830                 aux.tp_mac = 0;
2831                 aux.tp_net = skb_network_offset(skb);
2832                 if (vlan_tx_tag_present(skb)) {
2833                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2834                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2835                 } else {
2836                         aux.tp_vlan_tci = 0;
2837                 }
2838                 aux.tp_padding = 0;
2839                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2840         }
2841
2842         /*
2843          *      Free or return the buffer as appropriate. Again this
2844          *      hides all the races and re-entrancy issues from us.
2845          */
2846         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2847
2848 out_free:
2849         skb_free_datagram(sk, skb);
2850 out:
2851         return err;
2852 }
2853
2854 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2855                                int *uaddr_len, int peer)
2856 {
2857         struct net_device *dev;
2858         struct sock *sk = sock->sk;
2859
2860         if (peer)
2861                 return -EOPNOTSUPP;
2862
2863         uaddr->sa_family = AF_PACKET;
2864         rcu_read_lock();
2865         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2866         if (dev)
2867                 strncpy(uaddr->sa_data, dev->name, 14);
2868         else
2869                 memset(uaddr->sa_data, 0, 14);
2870         rcu_read_unlock();
2871         *uaddr_len = sizeof(*uaddr);
2872
2873         return 0;
2874 }
2875
2876 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2877                           int *uaddr_len, int peer)
2878 {
2879         struct net_device *dev;
2880         struct sock *sk = sock->sk;
2881         struct packet_sock *po = pkt_sk(sk);
2882         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2883
2884         if (peer)
2885                 return -EOPNOTSUPP;
2886
2887         sll->sll_family = AF_PACKET;
2888         sll->sll_ifindex = po->ifindex;
2889         sll->sll_protocol = po->num;
2890         sll->sll_pkttype = 0;
2891         rcu_read_lock();
2892         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2893         if (dev) {
2894                 sll->sll_hatype = dev->type;
2895                 sll->sll_halen = dev->addr_len;
2896                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2897         } else {
2898                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2899                 sll->sll_halen = 0;
2900         }
2901         rcu_read_unlock();
2902         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2903
2904         return 0;
2905 }
2906
2907 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2908                          int what)
2909 {
2910         switch (i->type) {
2911         case PACKET_MR_MULTICAST:
2912                 if (i->alen != dev->addr_len)
2913                         return -EINVAL;
2914                 if (what > 0)
2915                         return dev_mc_add(dev, i->addr);
2916                 else
2917                         return dev_mc_del(dev, i->addr);
2918                 break;
2919         case PACKET_MR_PROMISC:
2920                 return dev_set_promiscuity(dev, what);
2921                 break;
2922         case PACKET_MR_ALLMULTI:
2923                 return dev_set_allmulti(dev, what);
2924                 break;
2925         case PACKET_MR_UNICAST:
2926                 if (i->alen != dev->addr_len)
2927                         return -EINVAL;
2928                 if (what > 0)
2929                         return dev_uc_add(dev, i->addr);
2930                 else
2931                         return dev_uc_del(dev, i->addr);
2932                 break;
2933         default:
2934                 break;
2935         }
2936         return 0;
2937 }
2938
2939 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2940 {
2941         for ( ; i; i = i->next) {
2942                 if (i->ifindex == dev->ifindex)
2943                         packet_dev_mc(dev, i, what);
2944         }
2945 }
2946
2947 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2948 {
2949         struct packet_sock *po = pkt_sk(sk);
2950         struct packet_mclist *ml, *i;
2951         struct net_device *dev;
2952         int err;
2953
2954         rtnl_lock();
2955
2956         err = -ENODEV;
2957         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2958         if (!dev)
2959                 goto done;
2960
2961         err = -EINVAL;
2962         if (mreq->mr_alen > dev->addr_len)
2963                 goto done;
2964
2965         err = -ENOBUFS;
2966         i = kmalloc(sizeof(*i), GFP_KERNEL);
2967         if (i == NULL)
2968                 goto done;
2969
2970         err = 0;
2971         for (ml = po->mclist; ml; ml = ml->next) {
2972                 if (ml->ifindex == mreq->mr_ifindex &&
2973                     ml->type == mreq->mr_type &&
2974                     ml->alen == mreq->mr_alen &&
2975                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2976                         ml->count++;
2977                         /* Free the new element ... */
2978                         kfree(i);
2979                         goto done;
2980                 }
2981         }
2982
2983         i->type = mreq->mr_type;
2984         i->ifindex = mreq->mr_ifindex;
2985         i->alen = mreq->mr_alen;
2986         memcpy(i->addr, mreq->mr_address, i->alen);
2987         i->count = 1;
2988         i->next = po->mclist;
2989         po->mclist = i;
2990         err = packet_dev_mc(dev, i, 1);
2991         if (err) {
2992                 po->mclist = i->next;
2993                 kfree(i);
2994         }
2995
2996 done:
2997         rtnl_unlock();
2998         return err;
2999 }
3000
3001 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3002 {
3003         struct packet_mclist *ml, **mlp;
3004
3005         rtnl_lock();
3006
3007         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3008                 if (ml->ifindex == mreq->mr_ifindex &&
3009                     ml->type == mreq->mr_type &&
3010                     ml->alen == mreq->mr_alen &&
3011                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3012                         if (--ml->count == 0) {
3013                                 struct net_device *dev;
3014                                 *mlp = ml->next;
3015                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3016                                 if (dev)
3017                                         packet_dev_mc(dev, ml, -1);
3018                                 kfree(ml);
3019                         }
3020                         rtnl_unlock();
3021                         return 0;
3022                 }
3023         }
3024         rtnl_unlock();
3025         return -EADDRNOTAVAIL;
3026 }
3027
3028 static void packet_flush_mclist(struct sock *sk)
3029 {
3030         struct packet_sock *po = pkt_sk(sk);
3031         struct packet_mclist *ml;
3032
3033         if (!po->mclist)
3034                 return;
3035
3036         rtnl_lock();
3037         while ((ml = po->mclist) != NULL) {
3038                 struct net_device *dev;
3039
3040                 po->mclist = ml->next;
3041                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3042                 if (dev != NULL)
3043                         packet_dev_mc(dev, ml, -1);
3044                 kfree(ml);
3045         }
3046         rtnl_unlock();
3047 }
3048
3049 static int
3050 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3051 {
3052         struct sock *sk = sock->sk;
3053         struct packet_sock *po = pkt_sk(sk);
3054         int ret;
3055
3056         if (level != SOL_PACKET)
3057                 return -ENOPROTOOPT;
3058
3059         switch (optname) {
3060         case PACKET_ADD_MEMBERSHIP:
3061         case PACKET_DROP_MEMBERSHIP:
3062         {
3063                 struct packet_mreq_max mreq;
3064                 int len = optlen;
3065                 memset(&mreq, 0, sizeof(mreq));
3066                 if (len < sizeof(struct packet_mreq))
3067                         return -EINVAL;
3068                 if (len > sizeof(mreq))
3069                         len = sizeof(mreq);
3070                 if (copy_from_user(&mreq, optval, len))
3071                         return -EFAULT;
3072                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3073                         return -EINVAL;
3074                 if (optname == PACKET_ADD_MEMBERSHIP)
3075                         ret = packet_mc_add(sk, &mreq);
3076                 else
3077                         ret = packet_mc_drop(sk, &mreq);
3078                 return ret;
3079         }
3080
3081         case PACKET_RX_RING:
3082         case PACKET_TX_RING:
3083         {
3084                 union tpacket_req_u req_u;
3085                 int len;
3086
3087                 switch (po->tp_version) {
3088                 case TPACKET_V1:
3089                 case TPACKET_V2:
3090                         len = sizeof(req_u.req);
3091                         break;
3092                 case TPACKET_V3:
3093                 default:
3094                         len = sizeof(req_u.req3);
3095                         break;
3096                 }
3097                 if (optlen < len)
3098                         return -EINVAL;
3099                 if (pkt_sk(sk)->has_vnet_hdr)
3100                         return -EINVAL;
3101                 if (copy_from_user(&req_u.req, optval, len))
3102                         return -EFAULT;
3103                 return packet_set_ring(sk, &req_u, 0,
3104                         optname == PACKET_TX_RING);
3105         }
3106         case PACKET_COPY_THRESH:
3107         {
3108                 int val;
3109
3110                 if (optlen != sizeof(val))
3111                         return -EINVAL;
3112                 if (copy_from_user(&val, optval, sizeof(val)))
3113                         return -EFAULT;
3114
3115                 pkt_sk(sk)->copy_thresh = val;
3116                 return 0;
3117         }
3118         case PACKET_VERSION:
3119         {
3120                 int val;
3121
3122                 if (optlen != sizeof(val))
3123                         return -EINVAL;
3124                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3125                         return -EBUSY;
3126                 if (copy_from_user(&val, optval, sizeof(val)))
3127                         return -EFAULT;
3128                 switch (val) {
3129                 case TPACKET_V1:
3130                 case TPACKET_V2:
3131                 case TPACKET_V3:
3132                         po->tp_version = val;
3133                         return 0;
3134                 default:
3135                         return -EINVAL;
3136                 }
3137         }
3138         case PACKET_RESERVE:
3139         {
3140                 unsigned int val;
3141
3142                 if (optlen != sizeof(val))
3143                         return -EINVAL;
3144                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3145                         return -EBUSY;
3146                 if (copy_from_user(&val, optval, sizeof(val)))
3147                         return -EFAULT;
3148                 po->tp_reserve = val;
3149                 return 0;
3150         }
3151         case PACKET_LOSS:
3152         {
3153                 unsigned int val;
3154
3155                 if (optlen != sizeof(val))
3156                         return -EINVAL;
3157                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3158                         return -EBUSY;
3159                 if (copy_from_user(&val, optval, sizeof(val)))
3160                         return -EFAULT;
3161                 po->tp_loss = !!val;
3162                 return 0;
3163         }
3164         case PACKET_AUXDATA:
3165         {
3166                 int val;
3167
3168                 if (optlen < sizeof(val))
3169                         return -EINVAL;
3170                 if (copy_from_user(&val, optval, sizeof(val)))
3171                         return -EFAULT;
3172
3173                 po->auxdata = !!val;
3174                 return 0;
3175         }
3176         case PACKET_ORIGDEV:
3177         {
3178                 int val;
3179
3180                 if (optlen < sizeof(val))
3181                         return -EINVAL;
3182                 if (copy_from_user(&val, optval, sizeof(val)))
3183                         return -EFAULT;
3184
3185                 po->origdev = !!val;
3186                 return 0;
3187         }
3188         case PACKET_VNET_HDR:
3189         {
3190                 int val;
3191
3192                 if (sock->type != SOCK_RAW)
3193                         return -EINVAL;
3194                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3195                         return -EBUSY;
3196                 if (optlen < sizeof(val))
3197                         return -EINVAL;
3198                 if (copy_from_user(&val, optval, sizeof(val)))
3199                         return -EFAULT;
3200
3201                 po->has_vnet_hdr = !!val;
3202                 return 0;
3203         }
3204         case PACKET_TIMESTAMP:
3205         {
3206                 int val;
3207
3208                 if (optlen != sizeof(val))
3209                         return -EINVAL;
3210                 if (copy_from_user(&val, optval, sizeof(val)))
3211                         return -EFAULT;
3212
3213                 po->tp_tstamp = val;
3214                 return 0;
3215         }
3216         case PACKET_FANOUT:
3217         {
3218                 int val;
3219
3220                 if (optlen != sizeof(val))
3221                         return -EINVAL;
3222                 if (copy_from_user(&val, optval, sizeof(val)))
3223                         return -EFAULT;
3224
3225                 return fanout_add(sk, val & 0xffff, val >> 16);
3226         }
3227         default:
3228                 return -ENOPROTOOPT;
3229         }
3230 }
3231
3232 static int packet_getsockopt(struct socket *sock, int level, int optname,
3233                              char __user *optval, int __user *optlen)
3234 {
3235         int len;
3236         int val;
3237         struct sock *sk = sock->sk;
3238         struct packet_sock *po = pkt_sk(sk);
3239         void *data;
3240         struct tpacket_stats st;
3241         union tpacket_stats_u st_u;
3242
3243         if (level != SOL_PACKET)
3244                 return -ENOPROTOOPT;
3245
3246         if (get_user(len, optlen))
3247                 return -EFAULT;
3248
3249         if (len < 0)
3250                 return -EINVAL;
3251
3252         switch (optname) {
3253         case PACKET_STATISTICS:
3254                 if (po->tp_version == TPACKET_V3) {
3255                         len = sizeof(struct tpacket_stats_v3);
3256                 } else {
3257                         if (len > sizeof(struct tpacket_stats))
3258                                 len = sizeof(struct tpacket_stats);
3259                 }
3260                 spin_lock_bh(&sk->sk_receive_queue.lock);
3261                 if (po->tp_version == TPACKET_V3) {
3262                         memcpy(&st_u.stats3, &po->stats,
3263                         sizeof(struct tpacket_stats));
3264                         st_u.stats3.tp_freeze_q_cnt =
3265                         po->stats_u.stats3.tp_freeze_q_cnt;
3266                         st_u.stats3.tp_packets += po->stats.tp_drops;
3267                         data = &st_u.stats3;
3268                 } else {
3269                         st = po->stats;
3270                         st.tp_packets += st.tp_drops;
3271                         data = &st;
3272                 }
3273                 memset(&po->stats, 0, sizeof(st));
3274                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3275                 break;
3276         case PACKET_AUXDATA:
3277                 if (len > sizeof(int))
3278                         len = sizeof(int);
3279                 val = po->auxdata;
3280
3281                 data = &val;
3282                 break;
3283         case PACKET_ORIGDEV:
3284                 if (len > sizeof(int))
3285                         len = sizeof(int);
3286                 val = po->origdev;
3287
3288                 data = &val;
3289                 break;
3290         case PACKET_VNET_HDR:
3291                 if (len > sizeof(int))
3292                         len = sizeof(int);
3293                 val = po->has_vnet_hdr;
3294
3295                 data = &val;
3296                 break;
3297         case PACKET_VERSION:
3298                 if (len > sizeof(int))
3299                         len = sizeof(int);
3300                 val = po->tp_version;
3301                 data = &val;
3302                 break;
3303         case PACKET_HDRLEN:
3304                 if (len > sizeof(int))
3305                         len = sizeof(int);
3306                 if (copy_from_user(&val, optval, len))
3307                         return -EFAULT;
3308                 switch (val) {
3309                 case TPACKET_V1:
3310                         val = sizeof(struct tpacket_hdr);
3311                         break;
3312                 case TPACKET_V2:
3313                         val = sizeof(struct tpacket2_hdr);
3314                         break;
3315                 case TPACKET_V3:
3316                         val = sizeof(struct tpacket3_hdr);
3317                         break;
3318                 default:
3319                         return -EINVAL;
3320                 }
3321                 data = &val;
3322                 break;
3323         case PACKET_RESERVE:
3324                 if (len > sizeof(unsigned int))
3325                         len = sizeof(unsigned int);
3326                 val = po->tp_reserve;
3327                 data = &val;
3328                 break;
3329         case PACKET_LOSS:
3330                 if (len > sizeof(unsigned int))
3331                         len = sizeof(unsigned int);
3332                 val = po->tp_loss;
3333                 data = &val;
3334                 break;
3335         case PACKET_TIMESTAMP:
3336                 if (len > sizeof(int))
3337                         len = sizeof(int);
3338                 val = po->tp_tstamp;
3339                 data = &val;
3340                 break;
3341         case PACKET_FANOUT:
3342                 if (len > sizeof(int))
3343                         len = sizeof(int);
3344                 val = (po->fanout ?
3345                        ((u32)po->fanout->id |
3346                         ((u32)po->fanout->type << 16)) :
3347                        0);
3348                 data = &val;
3349                 break;
3350         default:
3351                 return -ENOPROTOOPT;
3352         }
3353
3354         if (put_user(len, optlen))
3355                 return -EFAULT;
3356         if (copy_to_user(optval, data, len))
3357                 return -EFAULT;
3358         return 0;
3359 }
3360
3361
3362 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3363 {
3364         struct sock *sk;
3365         struct hlist_node *node;
3366         struct net_device *dev = data;
3367         struct net *net = dev_net(dev);
3368
3369         rcu_read_lock();
3370         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3371                 struct packet_sock *po = pkt_sk(sk);
3372
3373                 switch (msg) {
3374                 case NETDEV_UNREGISTER:
3375                         if (po->mclist)
3376                                 packet_dev_mclist(dev, po->mclist, -1);
3377                         /* fallthrough */
3378
3379                 case NETDEV_DOWN:
3380                         if (dev->ifindex == po->ifindex) {
3381                                 spin_lock(&po->bind_lock);
3382                                 if (po->running) {
3383                                         __unregister_prot_hook(sk, false);
3384                                         sk->sk_err = ENETDOWN;
3385                                         if (!sock_flag(sk, SOCK_DEAD))
3386                                                 sk->sk_error_report(sk);
3387                                 }
3388                                 if (msg == NETDEV_UNREGISTER) {
3389                                         po->ifindex = -1;
3390                                         if (po->prot_hook.dev)
3391                                                 dev_put(po->prot_hook.dev);
3392                                         po->prot_hook.dev = NULL;
3393                                 }
3394                                 spin_unlock(&po->bind_lock);
3395                         }
3396                         break;
3397                 case NETDEV_UP:
3398                         if (dev->ifindex == po->ifindex) {
3399                                 spin_lock(&po->bind_lock);
3400                                 if (po->num)
3401                                         register_prot_hook(sk);
3402                                 spin_unlock(&po->bind_lock);
3403                         }
3404                         break;
3405                 }
3406         }
3407         rcu_read_unlock();
3408         return NOTIFY_DONE;
3409 }
3410
3411
3412 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3413                         unsigned long arg)
3414 {
3415         struct sock *sk = sock->sk;
3416
3417         switch (cmd) {
3418         case SIOCOUTQ:
3419         {
3420                 int amount = sk_wmem_alloc_get(sk);
3421
3422                 return put_user(amount, (int __user *)arg);
3423         }
3424         case SIOCINQ:
3425         {
3426                 struct sk_buff *skb;
3427                 int amount = 0;
3428
3429                 spin_lock_bh(&sk->sk_receive_queue.lock);
3430                 skb = skb_peek(&sk->sk_receive_queue);
3431                 if (skb)
3432                         amount = skb->len;
3433                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3434                 return put_user(amount, (int __user *)arg);
3435         }
3436         case SIOCGSTAMP:
3437                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3438         case SIOCGSTAMPNS:
3439                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3440
3441 #ifdef CONFIG_INET
3442         case SIOCADDRT:
3443         case SIOCDELRT:
3444         case SIOCDARP:
3445         case SIOCGARP:
3446         case SIOCSARP:
3447         case SIOCGIFADDR:
3448         case SIOCSIFADDR:
3449         case SIOCGIFBRDADDR:
3450         case SIOCSIFBRDADDR:
3451         case SIOCGIFNETMASK:
3452         case SIOCSIFNETMASK:
3453         case SIOCGIFDSTADDR:
3454         case SIOCSIFDSTADDR:
3455         case SIOCSIFFLAGS:
3456                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3457 #endif
3458
3459         default:
3460                 return -ENOIOCTLCMD;
3461         }
3462         return 0;
3463 }
3464
3465 static unsigned int packet_poll(struct file *file, struct socket *sock,
3466                                 poll_table *wait)
3467 {
3468         struct sock *sk = sock->sk;
3469         struct packet_sock *po = pkt_sk(sk);
3470         unsigned int mask = datagram_poll(file, sock, wait);
3471
3472         spin_lock_bh(&sk->sk_receive_queue.lock);
3473         if (po->rx_ring.pg_vec) {
3474                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3475                         TP_STATUS_KERNEL))
3476                         mask |= POLLIN | POLLRDNORM;
3477         }
3478         spin_unlock_bh(&sk->sk_receive_queue.lock);
3479         spin_lock_bh(&sk->sk_write_queue.lock);
3480         if (po->tx_ring.pg_vec) {
3481                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3482                         mask |= POLLOUT | POLLWRNORM;
3483         }
3484         spin_unlock_bh(&sk->sk_write_queue.lock);
3485         return mask;
3486 }
3487
3488
3489 /* Dirty? Well, I still did not learn better way to account
3490  * for user mmaps.
3491  */
3492
3493 static void packet_mm_open(struct vm_area_struct *vma)
3494 {
3495         struct file *file = vma->vm_file;
3496         struct socket *sock = file->private_data;
3497         struct sock *sk = sock->sk;
3498
3499         if (sk)
3500                 atomic_inc(&pkt_sk(sk)->mapped);
3501 }
3502
3503 static void packet_mm_close(struct vm_area_struct *vma)
3504 {
3505         struct file *file = vma->vm_file;
3506         struct socket *sock = file->private_data;
3507         struct sock *sk = sock->sk;
3508
3509         if (sk)
3510                 atomic_dec(&pkt_sk(sk)->mapped);
3511 }
3512
3513 static const struct vm_operations_struct packet_mmap_ops = {
3514         .open   =       packet_mm_open,
3515         .close  =       packet_mm_close,
3516 };
3517
3518 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3519                         unsigned int len)
3520 {
3521         int i;
3522
3523         for (i = 0; i < len; i++) {
3524                 if (likely(pg_vec[i].buffer)) {
3525                         if (is_vmalloc_addr(pg_vec[i].buffer))
3526                                 vfree(pg_vec[i].buffer);
3527                         else
3528                                 free_pages((unsigned long)pg_vec[i].buffer,
3529                                            order);
3530                         pg_vec[i].buffer = NULL;
3531                 }
3532         }
3533         kfree(pg_vec);
3534 }
3535
3536 static inline char *alloc_one_pg_vec_page(unsigned long order)
3537 {
3538         char *buffer = NULL;
3539         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3540                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3541
3542         buffer = (char *) __get_free_pages(gfp_flags, order);
3543
3544         if (buffer)
3545                 return buffer;
3546
3547         /*
3548          * __get_free_pages failed, fall back to vmalloc
3549          */
3550         buffer = vzalloc((1 << order) * PAGE_SIZE);
3551
3552         if (buffer)
3553                 return buffer;
3554
3555         /*
3556          * vmalloc failed, lets dig into swap here
3557          */
3558         gfp_flags &= ~__GFP_NORETRY;
3559         buffer = (char *)__get_free_pages(gfp_flags, order);
3560         if (buffer)
3561                 return buffer;
3562
3563         /*
3564          * complete and utter failure
3565          */
3566         return NULL;
3567 }
3568
3569 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3570 {
3571         unsigned int block_nr = req->tp_block_nr;
3572         struct pgv *pg_vec;
3573         int i;
3574
3575         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3576         if (unlikely(!pg_vec))
3577                 goto out;
3578
3579         for (i = 0; i < block_nr; i++) {
3580                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3581                 if (unlikely(!pg_vec[i].buffer))
3582                         goto out_free_pgvec;
3583         }
3584
3585 out:
3586         return pg_vec;
3587
3588 out_free_pgvec:
3589         free_pg_vec(pg_vec, order, block_nr);
3590         pg_vec = NULL;
3591         goto out;
3592 }
3593
3594 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3595                 int closing, int tx_ring)
3596 {
3597         struct pgv *pg_vec = NULL;
3598         struct packet_sock *po = pkt_sk(sk);
3599         int was_running, order = 0;
3600         struct packet_ring_buffer *rb;
3601         struct sk_buff_head *rb_queue;
3602         __be16 num;
3603         int err = -EINVAL;
3604         /* Added to avoid minimal code churn */
3605         struct tpacket_req *req = &req_u->req;
3606
3607         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3608         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3609                 WARN(1, "Tx-ring is not supported.\n");
3610                 goto out;
3611         }
3612
3613         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3614         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3615
3616         err = -EBUSY;
3617         if (!closing) {
3618                 if (atomic_read(&po->mapped))
3619                         goto out;
3620                 if (atomic_read(&rb->pending))
3621                         goto out;
3622         }
3623
3624         if (req->tp_block_nr) {
3625                 /* Sanity tests and some calculations */
3626                 err = -EBUSY;
3627                 if (unlikely(rb->pg_vec))
3628                         goto out;
3629
3630                 switch (po->tp_version) {
3631                 case TPACKET_V1:
3632                         po->tp_hdrlen = TPACKET_HDRLEN;
3633                         break;
3634                 case TPACKET_V2:
3635                         po->tp_hdrlen = TPACKET2_HDRLEN;
3636                         break;
3637                 case TPACKET_V3:
3638                         po->tp_hdrlen = TPACKET3_HDRLEN;
3639                         break;
3640                 }
3641
3642                 err = -EINVAL;
3643                 if (unlikely((int)req->tp_block_size <= 0))
3644                         goto out;
3645                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3646                         goto out;
3647                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3648                                         po->tp_reserve))
3649                         goto out;
3650                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3651                         goto out;
3652
3653                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3654                 if (unlikely(rb->frames_per_block <= 0))
3655                         goto out;
3656                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3657                                         req->tp_frame_nr))
3658                         goto out;
3659
3660                 err = -ENOMEM;
3661                 order = get_order(req->tp_block_size);
3662                 pg_vec = alloc_pg_vec(req, order);
3663                 if (unlikely(!pg_vec))
3664                         goto out;
3665                 switch (po->tp_version) {
3666                 case TPACKET_V3:
3667                 /* Transmit path is not supported. We checked
3668                  * it above but just being paranoid
3669                  */
3670                         if (!tx_ring)
3671                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3672                                 break;
3673                 default:
3674                         break;
3675                 }
3676         }
3677         /* Done */
3678         else {
3679                 err = -EINVAL;
3680                 if (unlikely(req->tp_frame_nr))
3681                         goto out;
3682         }
3683
3684         lock_sock(sk);
3685
3686         /* Detach socket from network */
3687         spin_lock(&po->bind_lock);
3688         was_running = po->running;
3689         num = po->num;
3690         if (was_running) {
3691                 po->num = 0;
3692                 __unregister_prot_hook(sk, false);
3693         }
3694         spin_unlock(&po->bind_lock);
3695
3696         synchronize_net();
3697
3698         err = -EBUSY;
3699         mutex_lock(&po->pg_vec_lock);
3700         if (closing || atomic_read(&po->mapped) == 0) {
3701                 err = 0;
3702                 spin_lock_bh(&rb_queue->lock);
3703                 swap(rb->pg_vec, pg_vec);
3704                 rb->frame_max = (req->tp_frame_nr - 1);
3705                 rb->head = 0;
3706                 rb->frame_size = req->tp_frame_size;
3707                 spin_unlock_bh(&rb_queue->lock);
3708
3709                 swap(rb->pg_vec_order, order);
3710                 swap(rb->pg_vec_len, req->tp_block_nr);
3711
3712                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3713                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3714                                                 tpacket_rcv : packet_rcv;
3715                 skb_queue_purge(rb_queue);
3716                 if (atomic_read(&po->mapped))
3717                         pr_err("packet_mmap: vma is busy: %d\n",
3718                                atomic_read(&po->mapped));
3719         }
3720         mutex_unlock(&po->pg_vec_lock);
3721
3722         spin_lock(&po->bind_lock);
3723         if (was_running) {
3724                 po->num = num;
3725                 register_prot_hook(sk);
3726         }
3727         spin_unlock(&po->bind_lock);
3728         if (closing && (po->tp_version > TPACKET_V2)) {
3729                 /* Because we don't support block-based V3 on tx-ring */
3730                 if (!tx_ring)
3731                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3732         }
3733         release_sock(sk);
3734
3735         if (pg_vec)
3736                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3737 out:
3738         return err;
3739 }
3740
3741 static int packet_mmap(struct file *file, struct socket *sock,
3742                 struct vm_area_struct *vma)
3743 {
3744         struct sock *sk = sock->sk;
3745         struct packet_sock *po = pkt_sk(sk);
3746         unsigned long size, expected_size;
3747         struct packet_ring_buffer *rb;
3748         unsigned long start;
3749         int err = -EINVAL;
3750         int i;
3751
3752         if (vma->vm_pgoff)
3753                 return -EINVAL;
3754
3755         mutex_lock(&po->pg_vec_lock);
3756
3757         expected_size = 0;
3758         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3759                 if (rb->pg_vec) {
3760                         expected_size += rb->pg_vec_len
3761                                                 * rb->pg_vec_pages
3762                                                 * PAGE_SIZE;
3763                 }
3764         }
3765
3766         if (expected_size == 0)
3767                 goto out;
3768
3769         size = vma->vm_end - vma->vm_start;
3770         if (size != expected_size)
3771                 goto out;
3772
3773         start = vma->vm_start;
3774         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3775                 if (rb->pg_vec == NULL)
3776                         continue;
3777
3778                 for (i = 0; i < rb->pg_vec_len; i++) {
3779                         struct page *page;
3780                         void *kaddr = rb->pg_vec[i].buffer;
3781                         int pg_num;
3782
3783                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3784                                 page = pgv_to_page(kaddr);
3785                                 err = vm_insert_page(vma, start, page);
3786                                 if (unlikely(err))
3787                                         goto out;
3788                                 start += PAGE_SIZE;
3789                                 kaddr += PAGE_SIZE;
3790                         }
3791                 }
3792         }
3793
3794         atomic_inc(&po->mapped);
3795         vma->vm_ops = &packet_mmap_ops;
3796         err = 0;
3797
3798 out:
3799         mutex_unlock(&po->pg_vec_lock);
3800         return err;
3801 }
3802
3803 static const struct proto_ops packet_ops_spkt = {
3804         .family =       PF_PACKET,
3805         .owner =        THIS_MODULE,
3806         .release =      packet_release,
3807         .bind =         packet_bind_spkt,
3808         .connect =      sock_no_connect,
3809         .socketpair =   sock_no_socketpair,
3810         .accept =       sock_no_accept,
3811         .getname =      packet_getname_spkt,
3812         .poll =         datagram_poll,
3813         .ioctl =        packet_ioctl,
3814         .listen =       sock_no_listen,
3815         .shutdown =     sock_no_shutdown,
3816         .setsockopt =   sock_no_setsockopt,
3817         .getsockopt =   sock_no_getsockopt,
3818         .sendmsg =      packet_sendmsg_spkt,
3819         .recvmsg =      packet_recvmsg,
3820         .mmap =         sock_no_mmap,
3821         .sendpage =     sock_no_sendpage,
3822 };
3823
3824 static const struct proto_ops packet_ops = {
3825         .family =       PF_PACKET,
3826         .owner =        THIS_MODULE,
3827         .release =      packet_release,
3828         .bind =         packet_bind,
3829         .connect =      sock_no_connect,
3830         .socketpair =   sock_no_socketpair,
3831         .accept =       sock_no_accept,
3832         .getname =      packet_getname,
3833         .poll =         packet_poll,
3834         .ioctl =        packet_ioctl,
3835         .listen =       sock_no_listen,
3836         .shutdown =     sock_no_shutdown,
3837         .setsockopt =   packet_setsockopt,
3838         .getsockopt =   packet_getsockopt,
3839         .sendmsg =      packet_sendmsg,
3840         .recvmsg =      packet_recvmsg,
3841         .mmap =         packet_mmap,
3842         .sendpage =     sock_no_sendpage,
3843 };
3844
3845 static const struct net_proto_family packet_family_ops = {
3846         .family =       PF_PACKET,
3847         .create =       packet_create,
3848         .owner  =       THIS_MODULE,
3849 };
3850
3851 static struct notifier_block packet_netdev_notifier = {
3852         .notifier_call =        packet_notifier,
3853 };
3854
3855 #ifdef CONFIG_PROC_FS
3856
3857 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3858         __acquires(RCU)
3859 {
3860         struct net *net = seq_file_net(seq);
3861
3862         rcu_read_lock();
3863         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3864 }
3865
3866 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3867 {
3868         struct net *net = seq_file_net(seq);
3869         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3870 }
3871
3872 static void packet_seq_stop(struct seq_file *seq, void *v)
3873         __releases(RCU)
3874 {
3875         rcu_read_unlock();
3876 }
3877
3878 static int packet_seq_show(struct seq_file *seq, void *v)
3879 {
3880         if (v == SEQ_START_TOKEN)
3881                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3882         else {
3883                 struct sock *s = sk_entry(v);
3884                 const struct packet_sock *po = pkt_sk(s);
3885
3886                 seq_printf(seq,
3887                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3888                            s,
3889                            atomic_read(&s->sk_refcnt),
3890                            s->sk_type,
3891                            ntohs(po->num),
3892                            po->ifindex,
3893                            po->running,
3894                            atomic_read(&s->sk_rmem_alloc),
3895                            sock_i_uid(s),
3896                            sock_i_ino(s));
3897         }
3898
3899         return 0;
3900 }
3901
3902 static const struct seq_operations packet_seq_ops = {
3903         .start  = packet_seq_start,
3904         .next   = packet_seq_next,
3905         .stop   = packet_seq_stop,
3906         .show   = packet_seq_show,
3907 };
3908
3909 static int packet_seq_open(struct inode *inode, struct file *file)
3910 {
3911         return seq_open_net(inode, file, &packet_seq_ops,
3912                             sizeof(struct seq_net_private));
3913 }
3914
3915 static const struct file_operations packet_seq_fops = {
3916         .owner          = THIS_MODULE,
3917         .open           = packet_seq_open,
3918         .read           = seq_read,
3919         .llseek         = seq_lseek,
3920         .release        = seq_release_net,
3921 };
3922
3923 #endif
3924
3925 static int __net_init packet_net_init(struct net *net)
3926 {
3927         spin_lock_init(&net->packet.sklist_lock);
3928         INIT_HLIST_HEAD(&net->packet.sklist);
3929
3930         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3931                 return -ENOMEM;
3932
3933         return 0;
3934 }
3935
3936 static void __net_exit packet_net_exit(struct net *net)
3937 {
3938         proc_net_remove(net, "packet");
3939 }
3940
3941 static struct pernet_operations packet_net_ops = {
3942         .init = packet_net_init,
3943         .exit = packet_net_exit,
3944 };
3945
3946
3947 static void __exit packet_exit(void)
3948 {
3949         unregister_netdevice_notifier(&packet_netdev_notifier);
3950         unregister_pernet_subsys(&packet_net_ops);
3951         sock_unregister(PF_PACKET);
3952         proto_unregister(&packet_proto);
3953 }
3954
3955 static int __init packet_init(void)
3956 {
3957         int rc = proto_register(&packet_proto, 0);
3958
3959         if (rc != 0)
3960                 goto out;
3961
3962         sock_register(&packet_family_ops);
3963         register_pernet_subsys(&packet_net_ops);
3964         register_netdevice_notifier(&packet_netdev_notifier);
3965 out:
3966         return rc;
3967 }
3968
3969 module_init(packet_init);
3970 module_exit(packet_exit);
3971 MODULE_LICENSE("GPL");
3972 MODULE_ALIAS_NETPROTO(PF_PACKET);