3a5a58f8baa291bdde96d49583d4ee96a5cc0fba
[pandora-kernel.git] / net / packet / af_packet.c
1 /*
2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
3  *              operating system.  INET is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              PACKET - implements raw packet sockets.
7  *
8  * Authors:     Ross Biro
9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
11  *
12  * Fixes:
13  *              Alan Cox        :       verify_area() now used correctly
14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
15  *              Alan Cox        :       tidied skbuff lists.
16  *              Alan Cox        :       Now uses generic datagram routines I
17  *                                      added. Also fixed the peek/read crash
18  *                                      from all old Linux datagram code.
19  *              Alan Cox        :       Uses the improved datagram code.
20  *              Alan Cox        :       Added NULL's for socket options.
21  *              Alan Cox        :       Re-commented the code.
22  *              Alan Cox        :       Use new kernel side addressing
23  *              Rob Janssen     :       Correct MTU usage.
24  *              Dave Platt      :       Counter leaks caused by incorrect
25  *                                      interrupt locking and some slightly
26  *                                      dubious gcc output. Can you read
27  *                                      compiler: it said _VOLATILE_
28  *      Richard Kooijman        :       Timestamp fixes.
29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
30  *              Alan Cox        :       sendmsg/recvmsg support.
31  *              Alan Cox        :       Protocol setting support
32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
34  *      Michal Ostrowski        :       Module initialization cleanup.
35  *         Ulises Alonso        :       Frame number limit removal and
36  *                                      packet_set_ring memory leak.
37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
38  *                                      The convention is that longer addresses
39  *                                      will simply extend the hardware address
40  *                                      byte arrays at the end of sockaddr_ll
41  *                                      and packet_mreq.
42  *              Johann Baudy    :       Added TX RING.
43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
44  *                                      layer.
45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
46  *
47  *
48  *              This program is free software; you can redistribute it and/or
49  *              modify it under the terms of the GNU General Public License
50  *              as published by the Free Software Foundation; either version
51  *              2 of the License, or (at your option) any later version.
52  *
53  */
54
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/system.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <asm/page.h>
80 #include <asm/cacheflush.h>
81 #include <asm/io.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
92
93 #ifdef CONFIG_INET
94 #include <net/inet_common.h>
95 #endif
96
97 /*
98    Assumptions:
99    - if device has no dev->hard_header routine, it adds and removes ll header
100      inside itself. In this case ll header is invisible outside of device,
101      but higher levels still should reserve dev->hard_header_len.
102      Some devices are enough clever to reallocate skb, when header
103      will not fit to reserved space (tunnel), another ones are silly
104      (PPP).
105    - packet socket receives packets with pulled ll header,
106      so that SOCK_RAW should push it back.
107
108 On receive:
109 -----------
110
111 Incoming, dev->hard_header!=NULL
112    mac_header -> ll header
113    data       -> data
114
115 Outgoing, dev->hard_header!=NULL
116    mac_header -> ll header
117    data       -> ll header
118
119 Incoming, dev->hard_header==NULL
120    mac_header -> UNKNOWN position. It is very likely, that it points to ll
121                  header.  PPP makes it, that is wrong, because introduce
122                  assymetry between rx and tx paths.
123    data       -> data
124
125 Outgoing, dev->hard_header==NULL
126    mac_header -> data. ll header is still not built!
127    data       -> data
128
129 Resume
130   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
131
132
133 On transmit:
134 ------------
135
136 dev->hard_header != NULL
137    mac_header -> ll header
138    data       -> ll header
139
140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141    mac_header -> data
142    data       -> data
143
144    We should set nh.raw on output to correct posistion,
145    packet classifier depends on it.
146  */
147
148 /* Private packet socket structures. */
149
150 struct packet_mclist {
151         struct packet_mclist    *next;
152         int                     ifindex;
153         int                     count;
154         unsigned short          type;
155         unsigned short          alen;
156         unsigned char           addr[MAX_ADDR_LEN];
157 };
158 /* identical to struct packet_mreq except it has
159  * a longer address field.
160  */
161 struct packet_mreq_max {
162         int             mr_ifindex;
163         unsigned short  mr_type;
164         unsigned short  mr_alen;
165         unsigned char   mr_address[MAX_ADDR_LEN];
166 };
167
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169                 int closing, int tx_ring);
170
171
172 #define V3_ALIGNMENT    (8)
173
174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179 /* kbdq - kernel block descriptor queue */
180 struct tpacket_kbdq_core {
181         struct pgv      *pkbdq;
182         unsigned int    feature_req_word;
183         unsigned int    hdrlen;
184         unsigned char   reset_pending_on_curr_blk;
185         unsigned char   delete_blk_timer;
186         unsigned short  kactive_blk_num;
187         unsigned short  blk_sizeof_priv;
188
189         /* last_kactive_blk_num:
190          * trick to see if user-space has caught up
191          * in order to avoid refreshing timer when every single pkt arrives.
192          */
193         unsigned short  last_kactive_blk_num;
194
195         char            *pkblk_start;
196         char            *pkblk_end;
197         int             kblk_size;
198         unsigned int    max_frame_len;
199         unsigned int    knum_blocks;
200         uint64_t        knxt_seq_num;
201         char            *prev;
202         char            *nxt_offset;
203         struct sk_buff  *skb;
204
205         atomic_t        blk_fill_in_prog;
206
207         /* Default is set to 8ms */
208 #define DEFAULT_PRB_RETIRE_TOV  (8)
209
210         unsigned short  retire_blk_tov;
211         unsigned short  version;
212         unsigned long   tov_in_jiffies;
213
214         /* timer to retire an outstanding block */
215         struct timer_list retire_blk_timer;
216 };
217
218 #define PGV_FROM_VMALLOC 1
219 struct pgv {
220         char *buffer;
221 };
222
223 struct packet_ring_buffer {
224         struct pgv              *pg_vec;
225         unsigned int            head;
226         unsigned int            frames_per_block;
227         unsigned int            frame_size;
228         unsigned int            frame_max;
229
230         unsigned int            pg_vec_order;
231         unsigned int            pg_vec_pages;
232         unsigned int            pg_vec_len;
233
234         struct tpacket_kbdq_core        prb_bdqc;
235         atomic_t                pending;
236 };
237
238 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
239 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
240 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
241 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
242 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
243 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
244 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
245
246 struct packet_sock;
247 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
248
249 static void *packet_previous_frame(struct packet_sock *po,
250                 struct packet_ring_buffer *rb,
251                 int status);
252 static void packet_increment_head(struct packet_ring_buffer *buff);
253 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
254                         struct tpacket_block_desc *);
255 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
256                         struct packet_sock *);
257 static void prb_retire_current_block(struct tpacket_kbdq_core *,
258                 struct packet_sock *, unsigned int status);
259 static int prb_queue_frozen(struct tpacket_kbdq_core *);
260 static void prb_open_block(struct tpacket_kbdq_core *,
261                 struct tpacket_block_desc *);
262 static void prb_retire_rx_blk_timer_expired(unsigned long);
263 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
264 static void prb_init_blk_timer(struct packet_sock *,
265                 struct tpacket_kbdq_core *,
266                 void (*func) (unsigned long));
267 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
268 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
269                 struct tpacket3_hdr *);
270 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
271                 struct tpacket3_hdr *);
272 static void packet_flush_mclist(struct sock *sk);
273
274 struct packet_fanout;
275 struct packet_sock {
276         /* struct sock has to be the first member of packet_sock */
277         struct sock             sk;
278         struct packet_fanout    *fanout;
279         struct tpacket_stats    stats;
280         union  tpacket_stats_u  stats_u;
281         struct packet_ring_buffer       rx_ring;
282         struct packet_ring_buffer       tx_ring;
283         int                     copy_thresh;
284         spinlock_t              bind_lock;
285         struct mutex            pg_vec_lock;
286         unsigned int            running:1,      /* prot_hook is attached*/
287                                 auxdata:1,
288                                 origdev:1,
289                                 has_vnet_hdr:1;
290         int                     ifindex;        /* bound device         */
291         __be16                  num;
292         struct packet_mclist    *mclist;
293         atomic_t                mapped;
294         enum tpacket_versions   tp_version;
295         unsigned int            tp_hdrlen;
296         unsigned int            tp_reserve;
297         unsigned int            tp_loss:1;
298         unsigned int            tp_tstamp;
299         struct net_device __rcu *cached_dev;
300         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
301 };
302
303 #define PACKET_FANOUT_MAX       256
304
305 struct packet_fanout {
306 #ifdef CONFIG_NET_NS
307         struct net              *net;
308 #endif
309         unsigned int            num_members;
310         u16                     id;
311         u8                      type;
312         u8                      defrag;
313         atomic_t                rr_cur;
314         struct list_head        list;
315         struct sock             *arr[PACKET_FANOUT_MAX];
316         spinlock_t              lock;
317         atomic_t                sk_ref;
318         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
319 };
320
321 struct packet_skb_cb {
322         unsigned int origlen;
323         union {
324                 struct sockaddr_pkt pkt;
325                 struct sockaddr_ll ll;
326         } sa;
327 };
328
329 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
330
331 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
332 #define GET_PBLOCK_DESC(x, bid) \
333         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
334 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
335         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
336 #define GET_NEXT_PRB_BLK_NUM(x) \
337         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
338         ((x)->kactive_blk_num+1) : 0)
339
340 static struct packet_sock *pkt_sk(struct sock *sk)
341 {
342         return (struct packet_sock *)sk;
343 }
344
345 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
346 static void __fanout_link(struct sock *sk, struct packet_sock *po);
347
348 /* register_prot_hook must be invoked with the po->bind_lock held,
349  * or from a context in which asynchronous accesses to the packet
350  * socket is not possible (packet_create()).
351  */
352 static void register_prot_hook(struct sock *sk)
353 {
354         struct packet_sock *po = pkt_sk(sk);
355
356         if (!po->running) {
357                 if (po->fanout) {
358                         __fanout_link(sk, po);
359                 } else {
360                         dev_add_pack(&po->prot_hook);
361                         rcu_assign_pointer(po->cached_dev, po->prot_hook.dev);
362                 }
363
364                 sock_hold(sk);
365                 po->running = 1;
366         }
367 }
368
369 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
370  * held.   If the sync parameter is true, we will temporarily drop
371  * the po->bind_lock and do a synchronize_net to make sure no
372  * asynchronous packet processing paths still refer to the elements
373  * of po->prot_hook.  If the sync parameter is false, it is the
374  * callers responsibility to take care of this.
375  */
376 static void __unregister_prot_hook(struct sock *sk, bool sync)
377 {
378         struct packet_sock *po = pkt_sk(sk);
379
380         po->running = 0;
381         if (po->fanout) {
382                 __fanout_unlink(sk, po);
383         } else {
384                 __dev_remove_pack(&po->prot_hook);
385                 RCU_INIT_POINTER(po->cached_dev, NULL);
386         }
387
388         __sock_put(sk);
389
390         if (sync) {
391                 spin_unlock(&po->bind_lock);
392                 synchronize_net();
393                 spin_lock(&po->bind_lock);
394         }
395 }
396
397 static void unregister_prot_hook(struct sock *sk, bool sync)
398 {
399         struct packet_sock *po = pkt_sk(sk);
400
401         if (po->running)
402                 __unregister_prot_hook(sk, sync);
403 }
404
405 static inline __pure struct page *pgv_to_page(void *addr)
406 {
407         if (is_vmalloc_addr(addr))
408                 return vmalloc_to_page(addr);
409         return virt_to_page(addr);
410 }
411
412 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
413 {
414         union {
415                 struct tpacket_hdr *h1;
416                 struct tpacket2_hdr *h2;
417                 void *raw;
418         } h;
419
420         h.raw = frame;
421         switch (po->tp_version) {
422         case TPACKET_V1:
423                 h.h1->tp_status = status;
424                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
425                 break;
426         case TPACKET_V2:
427                 h.h2->tp_status = status;
428                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
429                 break;
430         case TPACKET_V3:
431         default:
432                 WARN(1, "TPACKET version not supported.\n");
433                 BUG();
434         }
435
436         smp_wmb();
437 }
438
439 static int __packet_get_status(struct packet_sock *po, void *frame)
440 {
441         union {
442                 struct tpacket_hdr *h1;
443                 struct tpacket2_hdr *h2;
444                 void *raw;
445         } h;
446
447         smp_rmb();
448
449         h.raw = frame;
450         switch (po->tp_version) {
451         case TPACKET_V1:
452                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
453                 return h.h1->tp_status;
454         case TPACKET_V2:
455                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
456                 return h.h2->tp_status;
457         case TPACKET_V3:
458         default:
459                 WARN(1, "TPACKET version not supported.\n");
460                 BUG();
461                 return 0;
462         }
463 }
464
465 static void *packet_lookup_frame(struct packet_sock *po,
466                 struct packet_ring_buffer *rb,
467                 unsigned int position,
468                 int status)
469 {
470         unsigned int pg_vec_pos, frame_offset;
471         union {
472                 struct tpacket_hdr *h1;
473                 struct tpacket2_hdr *h2;
474                 void *raw;
475         } h;
476
477         pg_vec_pos = position / rb->frames_per_block;
478         frame_offset = position % rb->frames_per_block;
479
480         h.raw = rb->pg_vec[pg_vec_pos].buffer +
481                 (frame_offset * rb->frame_size);
482
483         if (status != __packet_get_status(po, h.raw))
484                 return NULL;
485
486         return h.raw;
487 }
488
489 static void *packet_current_frame(struct packet_sock *po,
490                 struct packet_ring_buffer *rb,
491                 int status)
492 {
493         return packet_lookup_frame(po, rb, rb->head, status);
494 }
495
496 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
497 {
498         del_timer_sync(&pkc->retire_blk_timer);
499 }
500
501 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
502                 int tx_ring,
503                 struct sk_buff_head *rb_queue)
504 {
505         struct tpacket_kbdq_core *pkc;
506
507         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
508
509         spin_lock_bh(&rb_queue->lock);
510         pkc->delete_blk_timer = 1;
511         spin_unlock_bh(&rb_queue->lock);
512
513         prb_del_retire_blk_timer(pkc);
514 }
515
516 static void prb_init_blk_timer(struct packet_sock *po,
517                 struct tpacket_kbdq_core *pkc,
518                 void (*func) (unsigned long))
519 {
520         init_timer(&pkc->retire_blk_timer);
521         pkc->retire_blk_timer.data = (long)po;
522         pkc->retire_blk_timer.function = func;
523         pkc->retire_blk_timer.expires = jiffies;
524 }
525
526 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
527 {
528         struct tpacket_kbdq_core *pkc;
529
530         if (tx_ring)
531                 BUG();
532
533         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
534         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
535 }
536
537 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
538                                 int blk_size_in_bytes)
539 {
540         struct net_device *dev;
541         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
542         struct ethtool_cmd ecmd;
543         int err;
544
545         rtnl_lock();
546         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
547         if (unlikely(!dev)) {
548                 rtnl_unlock();
549                 return DEFAULT_PRB_RETIRE_TOV;
550         }
551         err = __ethtool_get_settings(dev, &ecmd);
552         rtnl_unlock();
553         if (!err) {
554                 switch (ecmd.speed) {
555                 case SPEED_10000:
556                         msec = 1;
557                         div = 10000/1000;
558                         break;
559                 case SPEED_1000:
560                         msec = 1;
561                         div = 1000/1000;
562                         break;
563                 /*
564                  * If the link speed is so slow you don't really
565                  * need to worry about perf anyways
566                  */
567                 case SPEED_100:
568                 case SPEED_10:
569                 default:
570                         return DEFAULT_PRB_RETIRE_TOV;
571                 }
572         }
573
574         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
575
576         if (div)
577                 mbits /= div;
578
579         tmo = mbits * msec;
580
581         if (div)
582                 return tmo+1;
583         return tmo;
584 }
585
586 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
587                         union tpacket_req_u *req_u)
588 {
589         p1->feature_req_word = req_u->req3.tp_feature_req_word;
590 }
591
592 static void init_prb_bdqc(struct packet_sock *po,
593                         struct packet_ring_buffer *rb,
594                         struct pgv *pg_vec,
595                         union tpacket_req_u *req_u, int tx_ring)
596 {
597         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
598         struct tpacket_block_desc *pbd;
599
600         memset(p1, 0x0, sizeof(*p1));
601
602         p1->knxt_seq_num = 1;
603         p1->pkbdq = pg_vec;
604         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
605         p1->pkblk_start = (char *)pg_vec[0].buffer;
606         p1->kblk_size = req_u->req3.tp_block_size;
607         p1->knum_blocks = req_u->req3.tp_block_nr;
608         p1->hdrlen = po->tp_hdrlen;
609         p1->version = po->tp_version;
610         p1->last_kactive_blk_num = 0;
611         po->stats_u.stats3.tp_freeze_q_cnt = 0;
612         if (req_u->req3.tp_retire_blk_tov)
613                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
614         else
615                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
616                                                 req_u->req3.tp_block_size);
617         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
618         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
619
620         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
621         prb_init_ft_ops(p1, req_u);
622         prb_setup_retire_blk_timer(po, tx_ring);
623         prb_open_block(p1, pbd);
624 }
625
626 /*  Do NOT update the last_blk_num first.
627  *  Assumes sk_buff_head lock is held.
628  */
629 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
630 {
631         mod_timer(&pkc->retire_blk_timer,
632                         jiffies + pkc->tov_in_jiffies);
633         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
634 }
635
636 /*
637  * Timer logic:
638  * 1) We refresh the timer only when we open a block.
639  *    By doing this we don't waste cycles refreshing the timer
640  *        on packet-by-packet basis.
641  *
642  * With a 1MB block-size, on a 1Gbps line, it will take
643  * i) ~8 ms to fill a block + ii) memcpy etc.
644  * In this cut we are not accounting for the memcpy time.
645  *
646  * So, if the user sets the 'tmo' to 10ms then the timer
647  * will never fire while the block is still getting filled
648  * (which is what we want). However, the user could choose
649  * to close a block early and that's fine.
650  *
651  * But when the timer does fire, we check whether or not to refresh it.
652  * Since the tmo granularity is in msecs, it is not too expensive
653  * to refresh the timer, lets say every '8' msecs.
654  * Either the user can set the 'tmo' or we can derive it based on
655  * a) line-speed and b) block-size.
656  * prb_calc_retire_blk_tmo() calculates the tmo.
657  *
658  */
659 static void prb_retire_rx_blk_timer_expired(unsigned long data)
660 {
661         struct packet_sock *po = (struct packet_sock *)data;
662         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
663         unsigned int frozen;
664         struct tpacket_block_desc *pbd;
665
666         spin_lock(&po->sk.sk_receive_queue.lock);
667
668         frozen = prb_queue_frozen(pkc);
669         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
670
671         if (unlikely(pkc->delete_blk_timer))
672                 goto out;
673
674         /* We only need to plug the race when the block is partially filled.
675          * tpacket_rcv:
676          *              lock(); increment BLOCK_NUM_PKTS; unlock()
677          *              copy_bits() is in progress ...
678          *              timer fires on other cpu:
679          *              we can't retire the current block because copy_bits
680          *              is in progress.
681          *
682          */
683         if (BLOCK_NUM_PKTS(pbd)) {
684                 while (atomic_read(&pkc->blk_fill_in_prog)) {
685                         /* Waiting for skb_copy_bits to finish... */
686                         cpu_relax();
687                 }
688         }
689
690         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
691                 if (!frozen) {
692                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
693                         if (!prb_dispatch_next_block(pkc, po))
694                                 goto refresh_timer;
695                         else
696                                 goto out;
697                 } else {
698                         /* Case 1. Queue was frozen because user-space was
699                          *         lagging behind.
700                          */
701                         if (prb_curr_blk_in_use(pkc, pbd)) {
702                                 /*
703                                  * Ok, user-space is still behind.
704                                  * So just refresh the timer.
705                                  */
706                                 goto refresh_timer;
707                         } else {
708                                /* Case 2. queue was frozen,user-space caught up,
709                                 * now the link went idle && the timer fired.
710                                 * We don't have a block to close.So we open this
711                                 * block and restart the timer.
712                                 * opening a block thaws the queue,restarts timer
713                                 * Thawing/timer-refresh is a side effect.
714                                 */
715                                 prb_open_block(pkc, pbd);
716                                 goto out;
717                         }
718                 }
719         }
720
721 refresh_timer:
722         _prb_refresh_rx_retire_blk_timer(pkc);
723
724 out:
725         spin_unlock(&po->sk.sk_receive_queue.lock);
726 }
727
728 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
729                 struct tpacket_block_desc *pbd1, __u32 status)
730 {
731         /* Flush everything minus the block header */
732
733 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
734         u8 *start, *end;
735
736         start = (u8 *)pbd1;
737
738         /* Skip the block header(we know header WILL fit in 4K) */
739         start += PAGE_SIZE;
740
741         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
742         for (; start < end; start += PAGE_SIZE)
743                 flush_dcache_page(pgv_to_page(start));
744
745         smp_wmb();
746 #endif
747
748         /* Now update the block status. */
749
750         BLOCK_STATUS(pbd1) = status;
751
752         /* Flush the block header */
753
754 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
755         start = (u8 *)pbd1;
756         flush_dcache_page(pgv_to_page(start));
757
758         smp_wmb();
759 #endif
760 }
761
762 /*
763  * Side effect:
764  *
765  * 1) flush the block
766  * 2) Increment active_blk_num
767  *
768  * Note:We DONT refresh the timer on purpose.
769  *      Because almost always the next block will be opened.
770  */
771 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
772                 struct tpacket_block_desc *pbd1,
773                 struct packet_sock *po, unsigned int stat)
774 {
775         __u32 status = TP_STATUS_USER | stat;
776
777         struct tpacket3_hdr *last_pkt;
778         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
779
780         if (po->stats.tp_drops)
781                 status |= TP_STATUS_LOSING;
782
783         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
784         last_pkt->tp_next_offset = 0;
785
786         /* Get the ts of the last pkt */
787         if (BLOCK_NUM_PKTS(pbd1)) {
788                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
789                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
790         } else {
791                 /* Ok, we tmo'd - so get the current time */
792                 struct timespec ts;
793                 getnstimeofday(&ts);
794                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
795                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
796         }
797
798         smp_wmb();
799
800         /* Flush the block */
801         prb_flush_block(pkc1, pbd1, status);
802
803         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
804 }
805
806 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
807 {
808         pkc->reset_pending_on_curr_blk = 0;
809 }
810
811 /*
812  * Side effect of opening a block:
813  *
814  * 1) prb_queue is thawed.
815  * 2) retire_blk_timer is refreshed.
816  *
817  */
818 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
819         struct tpacket_block_desc *pbd1)
820 {
821         struct timespec ts;
822         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
823
824         smp_rmb();
825
826         /* We could have just memset this but we will lose the
827          * flexibility of making the priv area sticky
828          */
829         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
830         BLOCK_NUM_PKTS(pbd1) = 0;
831         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
832         getnstimeofday(&ts);
833         h1->ts_first_pkt.ts_sec = ts.tv_sec;
834         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
835         pkc1->pkblk_start = (char *)pbd1;
836         pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
837                                     BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
838         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
839         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
840         pbd1->version = pkc1->version;
841         pkc1->prev = pkc1->nxt_offset;
842         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
843         prb_thaw_queue(pkc1);
844         _prb_refresh_rx_retire_blk_timer(pkc1);
845
846         smp_wmb();
847 }
848
849 /*
850  * Queue freeze logic:
851  * 1) Assume tp_block_nr = 8 blocks.
852  * 2) At time 't0', user opens Rx ring.
853  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854  * 4) user-space is either sleeping or processing block '0'.
855  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856  *    it will close block-7,loop around and try to fill block '0'.
857  *    call-flow:
858  *    __packet_lookup_frame_in_block
859  *      prb_retire_current_block()
860  *      prb_dispatch_next_block()
861  *        |->(BLOCK_STATUS == USER) evaluates to true
862  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
863  * 6) Now there are two cases:
864  *    6.1) Link goes idle right after the queue is frozen.
865  *         But remember, the last open_block() refreshed the timer.
866  *         When this timer expires,it will refresh itself so that we can
867  *         re-open block-0 in near future.
868  *    6.2) Link is busy and keeps on receiving packets. This is a simple
869  *         case and __packet_lookup_frame_in_block will check if block-0
870  *         is free and can now be re-used.
871  */
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873                                   struct packet_sock *po)
874 {
875         pkc->reset_pending_on_curr_blk = 1;
876         po->stats_u.stats3.tp_freeze_q_cnt++;
877 }
878
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880
881 /*
882  * If the next block is free then we will dispatch it
883  * and return a good offset.
884  * Else, we will freeze the queue.
885  * So, caller must check the return value.
886  */
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888                 struct packet_sock *po)
889 {
890         struct tpacket_block_desc *pbd;
891
892         smp_rmb();
893
894         /* 1. Get current block num */
895         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896
897         /* 2. If this block is currently in_use then freeze the queue */
898         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899                 prb_freeze_queue(pkc, po);
900                 return NULL;
901         }
902
903         /*
904          * 3.
905          * open this block and return the offset where the first packet
906          * needs to get stored.
907          */
908         prb_open_block(pkc, pbd);
909         return (void *)pkc->nxt_offset;
910 }
911
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913                 struct packet_sock *po, unsigned int status)
914 {
915         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916
917         /* retire/close the current block */
918         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919                 /*
920                  * Plug the case where copy_bits() is in progress on
921                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922                  * have space to copy the pkt in the current block and
923                  * called prb_retire_current_block()
924                  *
925                  * We don't need to worry about the TMO case because
926                  * the timer-handler already handled this case.
927                  */
928                 if (!(status & TP_STATUS_BLK_TMO)) {
929                         while (atomic_read(&pkc->blk_fill_in_prog)) {
930                                 /* Waiting for skb_copy_bits to finish... */
931                                 cpu_relax();
932                         }
933                 }
934                 prb_close_block(pkc, pbd, po, status);
935                 return;
936         }
937 }
938
939 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
940                                       struct tpacket_block_desc *pbd)
941 {
942         return TP_STATUS_USER & BLOCK_STATUS(pbd);
943 }
944
945 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
946 {
947         return pkc->reset_pending_on_curr_blk;
948 }
949
950 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
951 {
952         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
953         atomic_dec(&pkc->blk_fill_in_prog);
954 }
955
956 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
957                         struct tpacket3_hdr *ppd)
958 {
959         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
960 }
961
962 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
963                         struct tpacket3_hdr *ppd)
964 {
965         ppd->hv1.tp_rxhash = 0;
966 }
967
968 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
969                         struct tpacket3_hdr *ppd)
970 {
971         if (vlan_tx_tag_present(pkc->skb)) {
972                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
973                 ppd->tp_status = TP_STATUS_VLAN_VALID;
974         } else {
975                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
976         }
977 }
978
979 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
980                         struct tpacket3_hdr *ppd)
981 {
982         prb_fill_vlan_info(pkc, ppd);
983
984         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
985                 prb_fill_rxhash(pkc, ppd);
986         else
987                 prb_clear_rxhash(pkc, ppd);
988 }
989
990 static void prb_fill_curr_block(char *curr,
991                                 struct tpacket_kbdq_core *pkc,
992                                 struct tpacket_block_desc *pbd,
993                                 unsigned int len)
994 {
995         struct tpacket3_hdr *ppd;
996
997         ppd  = (struct tpacket3_hdr *)curr;
998         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
999         pkc->prev = curr;
1000         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1001         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1002         BLOCK_NUM_PKTS(pbd) += 1;
1003         atomic_inc(&pkc->blk_fill_in_prog);
1004         prb_run_all_ft_ops(pkc, ppd);
1005 }
1006
1007 /* Assumes caller has the sk->rx_queue.lock */
1008 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1009                                             struct sk_buff *skb,
1010                                                 int status,
1011                                             unsigned int len
1012                                             )
1013 {
1014         struct tpacket_kbdq_core *pkc;
1015         struct tpacket_block_desc *pbd;
1016         char *curr, *end;
1017
1018         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1019         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1020
1021         /* Queue is frozen when user space is lagging behind */
1022         if (prb_queue_frozen(pkc)) {
1023                 /*
1024                  * Check if that last block which caused the queue to freeze,
1025                  * is still in_use by user-space.
1026                  */
1027                 if (prb_curr_blk_in_use(pkc, pbd)) {
1028                         /* Can't record this packet */
1029                         return NULL;
1030                 } else {
1031                         /*
1032                          * Ok, the block was released by user-space.
1033                          * Now let's open that block.
1034                          * opening a block also thaws the queue.
1035                          * Thawing is a side effect.
1036                          */
1037                         prb_open_block(pkc, pbd);
1038                 }
1039         }
1040
1041         smp_mb();
1042         curr = pkc->nxt_offset;
1043         pkc->skb = skb;
1044         end = (char *) ((char *)pbd + pkc->kblk_size);
1045
1046         /* first try the current block */
1047         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1048                 prb_fill_curr_block(curr, pkc, pbd, len);
1049                 return (void *)curr;
1050         }
1051
1052         /* Ok, close the current block */
1053         prb_retire_current_block(pkc, po, 0);
1054
1055         /* Now, try to dispatch the next block */
1056         curr = (char *)prb_dispatch_next_block(pkc, po);
1057         if (curr) {
1058                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1059                 prb_fill_curr_block(curr, pkc, pbd, len);
1060                 return (void *)curr;
1061         }
1062
1063         /*
1064          * No free blocks are available.user_space hasn't caught up yet.
1065          * Queue was just frozen and now this packet will get dropped.
1066          */
1067         return NULL;
1068 }
1069
1070 static void *packet_current_rx_frame(struct packet_sock *po,
1071                                             struct sk_buff *skb,
1072                                             int status, unsigned int len)
1073 {
1074         char *curr = NULL;
1075         switch (po->tp_version) {
1076         case TPACKET_V1:
1077         case TPACKET_V2:
1078                 curr = packet_lookup_frame(po, &po->rx_ring,
1079                                         po->rx_ring.head, status);
1080                 return curr;
1081         case TPACKET_V3:
1082                 return __packet_lookup_frame_in_block(po, skb, status, len);
1083         default:
1084                 WARN(1, "TPACKET version not supported\n");
1085                 BUG();
1086                 return 0;
1087         }
1088 }
1089
1090 static void *prb_lookup_block(struct packet_sock *po,
1091                                      struct packet_ring_buffer *rb,
1092                                      unsigned int previous,
1093                                      int status)
1094 {
1095         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1096         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1097
1098         if (status != BLOCK_STATUS(pbd))
1099                 return NULL;
1100         return pbd;
1101 }
1102
1103 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1104 {
1105         unsigned int prev;
1106         if (rb->prb_bdqc.kactive_blk_num)
1107                 prev = rb->prb_bdqc.kactive_blk_num-1;
1108         else
1109                 prev = rb->prb_bdqc.knum_blocks-1;
1110         return prev;
1111 }
1112
1113 /* Assumes caller has held the rx_queue.lock */
1114 static void *__prb_previous_block(struct packet_sock *po,
1115                                          struct packet_ring_buffer *rb,
1116                                          int status)
1117 {
1118         unsigned int previous = prb_previous_blk_num(rb);
1119         return prb_lookup_block(po, rb, previous, status);
1120 }
1121
1122 static void *packet_previous_rx_frame(struct packet_sock *po,
1123                                              struct packet_ring_buffer *rb,
1124                                              int status)
1125 {
1126         if (po->tp_version <= TPACKET_V2)
1127                 return packet_previous_frame(po, rb, status);
1128
1129         return __prb_previous_block(po, rb, status);
1130 }
1131
1132 static void packet_increment_rx_head(struct packet_sock *po,
1133                                             struct packet_ring_buffer *rb)
1134 {
1135         switch (po->tp_version) {
1136         case TPACKET_V1:
1137         case TPACKET_V2:
1138                 return packet_increment_head(rb);
1139         case TPACKET_V3:
1140         default:
1141                 WARN(1, "TPACKET version not supported.\n");
1142                 BUG();
1143                 return;
1144         }
1145 }
1146
1147 static void *packet_previous_frame(struct packet_sock *po,
1148                 struct packet_ring_buffer *rb,
1149                 int status)
1150 {
1151         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1152         return packet_lookup_frame(po, rb, previous, status);
1153 }
1154
1155 static void packet_increment_head(struct packet_ring_buffer *buff)
1156 {
1157         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1158 }
1159
1160 static void packet_sock_destruct(struct sock *sk)
1161 {
1162         skb_queue_purge(&sk->sk_error_queue);
1163
1164         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1165         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1166
1167         if (!sock_flag(sk, SOCK_DEAD)) {
1168                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1169                 return;
1170         }
1171
1172         sk_refcnt_debug_dec(sk);
1173 }
1174
1175 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1176 {
1177         u32 idx, hash = skb->rxhash;
1178
1179         idx = ((u64)hash * num) >> 32;
1180
1181         return f->arr[idx];
1182 }
1183
1184 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1185 {
1186         unsigned int val = atomic_inc_return(&f->rr_cur);
1187
1188         return f->arr[val % num];
1189 }
1190
1191 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1192 {
1193         unsigned int cpu = smp_processor_id();
1194
1195         return f->arr[cpu % num];
1196 }
1197
1198 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1199                              struct packet_type *pt, struct net_device *orig_dev)
1200 {
1201         struct packet_fanout *f = pt->af_packet_priv;
1202         unsigned int num = ACCESS_ONCE(f->num_members);
1203         struct packet_sock *po;
1204         struct sock *sk;
1205
1206         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1207             !num) {
1208                 kfree_skb(skb);
1209                 return 0;
1210         }
1211
1212         switch (f->type) {
1213         case PACKET_FANOUT_HASH:
1214         default:
1215                 if (f->defrag) {
1216                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1217                         if (!skb)
1218                                 return 0;
1219                 }
1220                 skb_get_rxhash(skb);
1221                 sk = fanout_demux_hash(f, skb, num);
1222                 break;
1223         case PACKET_FANOUT_LB:
1224                 sk = fanout_demux_lb(f, skb, num);
1225                 break;
1226         case PACKET_FANOUT_CPU:
1227                 sk = fanout_demux_cpu(f, skb, num);
1228                 break;
1229         }
1230
1231         po = pkt_sk(sk);
1232
1233         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1234 }
1235
1236 static DEFINE_MUTEX(fanout_mutex);
1237 static LIST_HEAD(fanout_list);
1238
1239 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1240 {
1241         struct packet_fanout *f = po->fanout;
1242
1243         spin_lock(&f->lock);
1244         f->arr[f->num_members] = sk;
1245         smp_wmb();
1246         f->num_members++;
1247         if (f->num_members == 1)
1248                 dev_add_pack(&f->prot_hook);
1249         spin_unlock(&f->lock);
1250 }
1251
1252 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1253 {
1254         struct packet_fanout *f = po->fanout;
1255         int i;
1256
1257         spin_lock(&f->lock);
1258         for (i = 0; i < f->num_members; i++) {
1259                 if (f->arr[i] == sk)
1260                         break;
1261         }
1262         BUG_ON(i >= f->num_members);
1263         f->arr[i] = f->arr[f->num_members - 1];
1264         f->num_members--;
1265         if (f->num_members == 0)
1266                 __dev_remove_pack(&f->prot_hook);
1267         spin_unlock(&f->lock);
1268 }
1269
1270 bool match_fanout_group(struct packet_type *ptype, struct sock * sk)
1271 {
1272         if (sk->sk_family != PF_PACKET)
1273                 return false;
1274
1275         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1276 }
1277
1278 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1279 {
1280         struct packet_sock *po = pkt_sk(sk);
1281         struct packet_fanout *f, *match;
1282         u8 type = type_flags & 0xff;
1283         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1284         int err;
1285
1286         switch (type) {
1287         case PACKET_FANOUT_HASH:
1288         case PACKET_FANOUT_LB:
1289         case PACKET_FANOUT_CPU:
1290                 break;
1291         default:
1292                 return -EINVAL;
1293         }
1294
1295         mutex_lock(&fanout_mutex);
1296
1297         err = -EINVAL;
1298         if (!po->running)
1299                 goto out;
1300
1301         err = -EALREADY;
1302         if (po->fanout)
1303                 goto out;
1304
1305         match = NULL;
1306         list_for_each_entry(f, &fanout_list, list) {
1307                 if (f->id == id &&
1308                     read_pnet(&f->net) == sock_net(sk)) {
1309                         match = f;
1310                         break;
1311                 }
1312         }
1313         err = -EINVAL;
1314         if (match && match->defrag != defrag)
1315                 goto out;
1316         if (!match) {
1317                 err = -ENOMEM;
1318                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1319                 if (!match)
1320                         goto out;
1321                 write_pnet(&match->net, sock_net(sk));
1322                 match->id = id;
1323                 match->type = type;
1324                 match->defrag = defrag;
1325                 atomic_set(&match->rr_cur, 0);
1326                 INIT_LIST_HEAD(&match->list);
1327                 spin_lock_init(&match->lock);
1328                 atomic_set(&match->sk_ref, 0);
1329                 match->prot_hook.type = po->prot_hook.type;
1330                 match->prot_hook.dev = po->prot_hook.dev;
1331                 match->prot_hook.func = packet_rcv_fanout;
1332                 match->prot_hook.af_packet_priv = match;
1333                 match->prot_hook.id_match = match_fanout_group;
1334                 list_add(&match->list, &fanout_list);
1335         }
1336         err = -EINVAL;
1337         if (match->type == type &&
1338             match->prot_hook.type == po->prot_hook.type &&
1339             match->prot_hook.dev == po->prot_hook.dev) {
1340                 err = -ENOSPC;
1341                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1342                         __dev_remove_pack(&po->prot_hook);
1343                         po->fanout = match;
1344                         atomic_inc(&match->sk_ref);
1345                         __fanout_link(sk, po);
1346                         err = 0;
1347                 }
1348         }
1349 out:
1350         mutex_unlock(&fanout_mutex);
1351         return err;
1352 }
1353
1354 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1355  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1356  * It is the responsibility of the caller to call fanout_release_data() and
1357  * free the returned packet_fanout (after synchronize_net())
1358  */
1359 static struct packet_fanout *fanout_release(struct sock *sk)
1360 {
1361         struct packet_sock *po = pkt_sk(sk);
1362         struct packet_fanout *f;
1363
1364         mutex_lock(&fanout_mutex);
1365         f = po->fanout;
1366         if (f) {
1367                 po->fanout = NULL;
1368
1369                 if (atomic_dec_and_test(&f->sk_ref))
1370                         list_del(&f->list);
1371                 else
1372                         f = NULL;
1373         }
1374         mutex_unlock(&fanout_mutex);
1375
1376         return f;
1377 }
1378
1379 static const struct proto_ops packet_ops;
1380
1381 static const struct proto_ops packet_ops_spkt;
1382
1383 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1384                            struct packet_type *pt, struct net_device *orig_dev)
1385 {
1386         struct sock *sk;
1387         struct sockaddr_pkt *spkt;
1388
1389         /*
1390          *      When we registered the protocol we saved the socket in the data
1391          *      field for just this event.
1392          */
1393
1394         sk = pt->af_packet_priv;
1395
1396         /*
1397          *      Yank back the headers [hope the device set this
1398          *      right or kerboom...]
1399          *
1400          *      Incoming packets have ll header pulled,
1401          *      push it back.
1402          *
1403          *      For outgoing ones skb->data == skb_mac_header(skb)
1404          *      so that this procedure is noop.
1405          */
1406
1407         if (skb->pkt_type == PACKET_LOOPBACK)
1408                 goto out;
1409
1410         if (!net_eq(dev_net(dev), sock_net(sk)))
1411                 goto out;
1412
1413         skb = skb_share_check(skb, GFP_ATOMIC);
1414         if (skb == NULL)
1415                 goto oom;
1416
1417         /* drop any routing info */
1418         skb_dst_drop(skb);
1419
1420         /* drop conntrack reference */
1421         nf_reset(skb);
1422
1423         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1424
1425         skb_push(skb, skb->data - skb_mac_header(skb));
1426
1427         /*
1428          *      The SOCK_PACKET socket receives _all_ frames.
1429          */
1430
1431         spkt->spkt_family = dev->type;
1432         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1433         spkt->spkt_protocol = skb->protocol;
1434
1435         /*
1436          *      Charge the memory to the socket. This is done specifically
1437          *      to prevent sockets using all the memory up.
1438          */
1439
1440         if (sock_queue_rcv_skb(sk, skb) == 0)
1441                 return 0;
1442
1443 out:
1444         kfree_skb(skb);
1445 oom:
1446         return 0;
1447 }
1448
1449
1450 /*
1451  *      Output a raw packet to a device layer. This bypasses all the other
1452  *      protocol layers and you must therefore supply it with a complete frame
1453  */
1454
1455 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1456                                struct msghdr *msg, size_t len)
1457 {
1458         struct sock *sk = sock->sk;
1459         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1460         struct sk_buff *skb = NULL;
1461         struct net_device *dev;
1462         __be16 proto = 0;
1463         int err;
1464
1465         /*
1466          *      Get and verify the address.
1467          */
1468
1469         if (saddr) {
1470                 if (msg->msg_namelen < sizeof(struct sockaddr))
1471                         return -EINVAL;
1472                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1473                         proto = saddr->spkt_protocol;
1474         } else
1475                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1476
1477         /*
1478          *      Find the device first to size check it
1479          */
1480
1481         saddr->spkt_device[13] = 0;
1482 retry:
1483         rcu_read_lock();
1484         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1485         err = -ENODEV;
1486         if (dev == NULL)
1487                 goto out_unlock;
1488
1489         err = -ENETDOWN;
1490         if (!(dev->flags & IFF_UP))
1491                 goto out_unlock;
1492
1493         /*
1494          * You may not queue a frame bigger than the mtu. This is the lowest level
1495          * raw protocol and you must do your own fragmentation at this level.
1496          */
1497
1498         err = -EMSGSIZE;
1499         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1500                 goto out_unlock;
1501
1502         if (!skb) {
1503                 size_t reserved = LL_RESERVED_SPACE(dev);
1504                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1505
1506                 rcu_read_unlock();
1507                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1508                 if (skb == NULL)
1509                         return -ENOBUFS;
1510                 /* FIXME: Save some space for broken drivers that write a hard
1511                  * header at transmission time by themselves. PPP is the notable
1512                  * one here. This should really be fixed at the driver level.
1513                  */
1514                 skb_reserve(skb, reserved);
1515                 skb_reset_network_header(skb);
1516
1517                 /* Try to align data part correctly */
1518                 if (hhlen) {
1519                         skb->data -= hhlen;
1520                         skb->tail -= hhlen;
1521                         if (len < hhlen)
1522                                 skb_reset_network_header(skb);
1523                 }
1524                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1525                 if (err)
1526                         goto out_free;
1527                 goto retry;
1528         }
1529
1530         if (len > (dev->mtu + dev->hard_header_len)) {
1531                 /* Earlier code assumed this would be a VLAN pkt,
1532                  * double-check this now that we have the actual
1533                  * packet in hand.
1534                  */
1535                 struct ethhdr *ehdr;
1536                 skb_reset_mac_header(skb);
1537                 ehdr = eth_hdr(skb);
1538                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1539                         err = -EMSGSIZE;
1540                         goto out_unlock;
1541                 }
1542         }
1543
1544         skb->protocol = proto;
1545         skb->dev = dev;
1546         skb->priority = sk->sk_priority;
1547         skb->mark = sk->sk_mark;
1548         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1549         if (err < 0)
1550                 goto out_unlock;
1551
1552         dev_queue_xmit(skb);
1553         rcu_read_unlock();
1554         return len;
1555
1556 out_unlock:
1557         rcu_read_unlock();
1558 out_free:
1559         kfree_skb(skb);
1560         return err;
1561 }
1562
1563 static unsigned int run_filter(const struct sk_buff *skb,
1564                                       const struct sock *sk,
1565                                       unsigned int res)
1566 {
1567         struct sk_filter *filter;
1568
1569         rcu_read_lock();
1570         filter = rcu_dereference(sk->sk_filter);
1571         if (filter != NULL)
1572                 res = SK_RUN_FILTER(filter, skb);
1573         rcu_read_unlock();
1574
1575         return res;
1576 }
1577
1578 /*
1579  * This function makes lazy skb cloning in hope that most of packets
1580  * are discarded by BPF.
1581  *
1582  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1583  * and skb->cb are mangled. It works because (and until) packets
1584  * falling here are owned by current CPU. Output packets are cloned
1585  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1586  * sequencially, so that if we return skb to original state on exit,
1587  * we will not harm anyone.
1588  */
1589
1590 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1591                       struct packet_type *pt, struct net_device *orig_dev)
1592 {
1593         struct sock *sk;
1594         struct sockaddr_ll *sll;
1595         struct packet_sock *po;
1596         u8 *skb_head = skb->data;
1597         int skb_len = skb->len;
1598         unsigned int snaplen, res;
1599
1600         if (skb->pkt_type == PACKET_LOOPBACK)
1601                 goto drop;
1602
1603         sk = pt->af_packet_priv;
1604         po = pkt_sk(sk);
1605
1606         if (!net_eq(dev_net(dev), sock_net(sk)))
1607                 goto drop;
1608
1609         skb->dev = dev;
1610
1611         if (dev->header_ops) {
1612                 /* The device has an explicit notion of ll header,
1613                  * exported to higher levels.
1614                  *
1615                  * Otherwise, the device hides details of its frame
1616                  * structure, so that corresponding packet head is
1617                  * never delivered to user.
1618                  */
1619                 if (sk->sk_type != SOCK_DGRAM)
1620                         skb_push(skb, skb->data - skb_mac_header(skb));
1621                 else if (skb->pkt_type == PACKET_OUTGOING) {
1622                         /* Special case: outgoing packets have ll header at head */
1623                         skb_pull(skb, skb_network_offset(skb));
1624                 }
1625         }
1626
1627         snaplen = skb->len;
1628
1629         res = run_filter(skb, sk, snaplen);
1630         if (!res)
1631                 goto drop_n_restore;
1632         if (snaplen > res)
1633                 snaplen = res;
1634
1635         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1636                 goto drop_n_acct;
1637
1638         if (skb_shared(skb)) {
1639                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1640                 if (nskb == NULL)
1641                         goto drop_n_acct;
1642
1643                 if (skb_head != skb->data) {
1644                         skb->data = skb_head;
1645                         skb->len = skb_len;
1646                 }
1647                 kfree_skb(skb);
1648                 skb = nskb;
1649         }
1650
1651         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1652                      sizeof(skb->cb));
1653
1654         sll = &PACKET_SKB_CB(skb)->sa.ll;
1655         sll->sll_family = AF_PACKET;
1656         sll->sll_hatype = dev->type;
1657         sll->sll_protocol = skb->protocol;
1658         sll->sll_pkttype = skb->pkt_type;
1659         if (unlikely(po->origdev))
1660                 sll->sll_ifindex = orig_dev->ifindex;
1661         else
1662                 sll->sll_ifindex = dev->ifindex;
1663
1664         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1665
1666         PACKET_SKB_CB(skb)->origlen = skb->len;
1667
1668         if (pskb_trim(skb, snaplen))
1669                 goto drop_n_acct;
1670
1671         skb_set_owner_r(skb, sk);
1672         skb->dev = NULL;
1673         skb_dst_drop(skb);
1674
1675         /* drop conntrack reference */
1676         nf_reset(skb);
1677
1678         spin_lock(&sk->sk_receive_queue.lock);
1679         po->stats.tp_packets++;
1680         skb->dropcount = atomic_read(&sk->sk_drops);
1681         __skb_queue_tail(&sk->sk_receive_queue, skb);
1682         spin_unlock(&sk->sk_receive_queue.lock);
1683         sk->sk_data_ready(sk, skb->len);
1684         return 0;
1685
1686 drop_n_acct:
1687         spin_lock(&sk->sk_receive_queue.lock);
1688         po->stats.tp_drops++;
1689         atomic_inc(&sk->sk_drops);
1690         spin_unlock(&sk->sk_receive_queue.lock);
1691
1692 drop_n_restore:
1693         if (skb_head != skb->data && skb_shared(skb)) {
1694                 skb->data = skb_head;
1695                 skb->len = skb_len;
1696         }
1697 drop:
1698         consume_skb(skb);
1699         return 0;
1700 }
1701
1702 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1703                        struct packet_type *pt, struct net_device *orig_dev)
1704 {
1705         struct sock *sk;
1706         struct packet_sock *po;
1707         struct sockaddr_ll *sll;
1708         union {
1709                 struct tpacket_hdr *h1;
1710                 struct tpacket2_hdr *h2;
1711                 struct tpacket3_hdr *h3;
1712                 void *raw;
1713         } h;
1714         u8 *skb_head = skb->data;
1715         int skb_len = skb->len;
1716         unsigned int snaplen, res;
1717         unsigned long status = TP_STATUS_USER;
1718         unsigned short macoff, netoff, hdrlen;
1719         struct sk_buff *copy_skb = NULL;
1720         struct timeval tv;
1721         struct timespec ts;
1722         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1723
1724         if (skb->pkt_type == PACKET_LOOPBACK)
1725                 goto drop;
1726
1727         sk = pt->af_packet_priv;
1728         po = pkt_sk(sk);
1729
1730         if (!net_eq(dev_net(dev), sock_net(sk)))
1731                 goto drop;
1732
1733         if (dev->header_ops) {
1734                 if (sk->sk_type != SOCK_DGRAM)
1735                         skb_push(skb, skb->data - skb_mac_header(skb));
1736                 else if (skb->pkt_type == PACKET_OUTGOING) {
1737                         /* Special case: outgoing packets have ll header at head */
1738                         skb_pull(skb, skb_network_offset(skb));
1739                 }
1740         }
1741
1742         if (skb->ip_summed == CHECKSUM_PARTIAL)
1743                 status |= TP_STATUS_CSUMNOTREADY;
1744
1745         snaplen = skb->len;
1746
1747         res = run_filter(skb, sk, snaplen);
1748         if (!res)
1749                 goto drop_n_restore;
1750         if (snaplen > res)
1751                 snaplen = res;
1752
1753         if (sk->sk_type == SOCK_DGRAM) {
1754                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1755                                   po->tp_reserve;
1756         } else {
1757                 unsigned maclen = skb_network_offset(skb);
1758                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1759                                        (maclen < 16 ? 16 : maclen)) +
1760                         po->tp_reserve;
1761                 macoff = netoff - maclen;
1762         }
1763         if (po->tp_version <= TPACKET_V2) {
1764                 if (macoff + snaplen > po->rx_ring.frame_size) {
1765                         if (po->copy_thresh &&
1766                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1767                                 if (skb_shared(skb)) {
1768                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1769                                 } else {
1770                                         copy_skb = skb_get(skb);
1771                                         skb_head = skb->data;
1772                                 }
1773                                 if (copy_skb)
1774                                         skb_set_owner_r(copy_skb, sk);
1775                         }
1776                         snaplen = po->rx_ring.frame_size - macoff;
1777                         if ((int)snaplen < 0)
1778                                 snaplen = 0;
1779                 }
1780         } else if (unlikely(macoff + snaplen >
1781                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1782                 u32 nval;
1783
1784                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1785                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1786                             snaplen, nval, macoff);
1787                 snaplen = nval;
1788                 if (unlikely((int)snaplen < 0)) {
1789                         snaplen = 0;
1790                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1791                 }
1792         }
1793         spin_lock(&sk->sk_receive_queue.lock);
1794         h.raw = packet_current_rx_frame(po, skb,
1795                                         TP_STATUS_KERNEL, (macoff+snaplen));
1796         if (!h.raw)
1797                 goto ring_is_full;
1798         if (po->tp_version <= TPACKET_V2) {
1799                 packet_increment_rx_head(po, &po->rx_ring);
1800         /*
1801          * LOSING will be reported till you read the stats,
1802          * because it's COR - Clear On Read.
1803          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1804          * at packet level.
1805          */
1806                 if (po->stats.tp_drops)
1807                         status |= TP_STATUS_LOSING;
1808         }
1809         po->stats.tp_packets++;
1810         if (copy_skb) {
1811                 status |= TP_STATUS_COPY;
1812                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1813         }
1814         spin_unlock(&sk->sk_receive_queue.lock);
1815
1816         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1817
1818         switch (po->tp_version) {
1819         case TPACKET_V1:
1820                 h.h1->tp_len = skb->len;
1821                 h.h1->tp_snaplen = snaplen;
1822                 h.h1->tp_mac = macoff;
1823                 h.h1->tp_net = netoff;
1824                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1825                                 && shhwtstamps->syststamp.tv64)
1826                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1827                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1828                                 && shhwtstamps->hwtstamp.tv64)
1829                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1830                 else if (skb->tstamp.tv64)
1831                         tv = ktime_to_timeval(skb->tstamp);
1832                 else
1833                         do_gettimeofday(&tv);
1834                 h.h1->tp_sec = tv.tv_sec;
1835                 h.h1->tp_usec = tv.tv_usec;
1836                 hdrlen = sizeof(*h.h1);
1837                 break;
1838         case TPACKET_V2:
1839                 h.h2->tp_len = skb->len;
1840                 h.h2->tp_snaplen = snaplen;
1841                 h.h2->tp_mac = macoff;
1842                 h.h2->tp_net = netoff;
1843                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1844                                 && shhwtstamps->syststamp.tv64)
1845                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1846                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1847                                 && shhwtstamps->hwtstamp.tv64)
1848                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1849                 else if (skb->tstamp.tv64)
1850                         ts = ktime_to_timespec(skb->tstamp);
1851                 else
1852                         getnstimeofday(&ts);
1853                 h.h2->tp_sec = ts.tv_sec;
1854                 h.h2->tp_nsec = ts.tv_nsec;
1855                 if (vlan_tx_tag_present(skb)) {
1856                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1857                         status |= TP_STATUS_VLAN_VALID;
1858                 } else {
1859                         h.h2->tp_vlan_tci = 0;
1860                 }
1861                 h.h2->tp_padding = 0;
1862                 hdrlen = sizeof(*h.h2);
1863                 break;
1864         case TPACKET_V3:
1865                 /* tp_nxt_offset,vlan are already populated above.
1866                  * So DONT clear those fields here
1867                  */
1868                 h.h3->tp_status |= status;
1869                 h.h3->tp_len = skb->len;
1870                 h.h3->tp_snaplen = snaplen;
1871                 h.h3->tp_mac = macoff;
1872                 h.h3->tp_net = netoff;
1873                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1874                                 && shhwtstamps->syststamp.tv64)
1875                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1876                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1877                                 && shhwtstamps->hwtstamp.tv64)
1878                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1879                 else if (skb->tstamp.tv64)
1880                         ts = ktime_to_timespec(skb->tstamp);
1881                 else
1882                         getnstimeofday(&ts);
1883                 h.h3->tp_sec  = ts.tv_sec;
1884                 h.h3->tp_nsec = ts.tv_nsec;
1885                 hdrlen = sizeof(*h.h3);
1886                 break;
1887         default:
1888                 BUG();
1889         }
1890
1891         sll = h.raw + TPACKET_ALIGN(hdrlen);
1892         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1893         sll->sll_family = AF_PACKET;
1894         sll->sll_hatype = dev->type;
1895         sll->sll_protocol = skb->protocol;
1896         sll->sll_pkttype = skb->pkt_type;
1897         if (unlikely(po->origdev))
1898                 sll->sll_ifindex = orig_dev->ifindex;
1899         else
1900                 sll->sll_ifindex = dev->ifindex;
1901
1902         smp_mb();
1903 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1904         {
1905                 u8 *start, *end;
1906
1907                 if (po->tp_version <= TPACKET_V2) {
1908                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1909                                 + macoff + snaplen);
1910                         for (start = h.raw; start < end; start += PAGE_SIZE)
1911                                 flush_dcache_page(pgv_to_page(start));
1912                 }
1913                 smp_wmb();
1914         }
1915 #endif
1916         if (po->tp_version <= TPACKET_V2)
1917                 __packet_set_status(po, h.raw, status);
1918         else
1919                 prb_clear_blk_fill_status(&po->rx_ring);
1920
1921         sk->sk_data_ready(sk, 0);
1922
1923 drop_n_restore:
1924         if (skb_head != skb->data && skb_shared(skb)) {
1925                 skb->data = skb_head;
1926                 skb->len = skb_len;
1927         }
1928 drop:
1929         kfree_skb(skb);
1930         return 0;
1931
1932 ring_is_full:
1933         po->stats.tp_drops++;
1934         spin_unlock(&sk->sk_receive_queue.lock);
1935
1936         sk->sk_data_ready(sk, 0);
1937         kfree_skb(copy_skb);
1938         goto drop_n_restore;
1939 }
1940
1941 static void tpacket_destruct_skb(struct sk_buff *skb)
1942 {
1943         struct packet_sock *po = pkt_sk(skb->sk);
1944         void *ph;
1945
1946         if (likely(po->tx_ring.pg_vec)) {
1947                 ph = skb_shinfo(skb)->destructor_arg;
1948                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1949                 atomic_dec(&po->tx_ring.pending);
1950                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1951         }
1952
1953         sock_wfree(skb);
1954 }
1955
1956 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1957                 void *frame, struct net_device *dev, int size_max,
1958                 __be16 proto, unsigned char *addr)
1959 {
1960         union {
1961                 struct tpacket_hdr *h1;
1962                 struct tpacket2_hdr *h2;
1963                 void *raw;
1964         } ph;
1965         int to_write, offset, len, tp_len, nr_frags, len_max;
1966         struct socket *sock = po->sk.sk_socket;
1967         struct page *page;
1968         void *data;
1969         int err;
1970
1971         ph.raw = frame;
1972
1973         skb->protocol = proto;
1974         skb->dev = dev;
1975         skb->priority = po->sk.sk_priority;
1976         skb->mark = po->sk.sk_mark;
1977         skb_shinfo(skb)->destructor_arg = ph.raw;
1978
1979         switch (po->tp_version) {
1980         case TPACKET_V2:
1981                 tp_len = ph.h2->tp_len;
1982                 break;
1983         default:
1984                 tp_len = ph.h1->tp_len;
1985                 break;
1986         }
1987         if (unlikely(tp_len > size_max)) {
1988                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1989                 return -EMSGSIZE;
1990         }
1991
1992         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1993         skb_reset_network_header(skb);
1994
1995         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1996         to_write = tp_len;
1997
1998         if (sock->type == SOCK_DGRAM) {
1999                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2000                                 NULL, tp_len);
2001                 if (unlikely(err < 0))
2002                         return -EINVAL;
2003         } else if (dev->hard_header_len) {
2004                 /* net device doesn't like empty head */
2005                 if (unlikely(tp_len <= dev->hard_header_len)) {
2006                         pr_err("packet size is too short (%d < %d)\n",
2007                                tp_len, dev->hard_header_len);
2008                         return -EINVAL;
2009                 }
2010
2011                 skb_push(skb, dev->hard_header_len);
2012                 err = skb_store_bits(skb, 0, data,
2013                                 dev->hard_header_len);
2014                 if (unlikely(err))
2015                         return err;
2016
2017                 data += dev->hard_header_len;
2018                 to_write -= dev->hard_header_len;
2019         }
2020
2021         err = -EFAULT;
2022         offset = offset_in_page(data);
2023         len_max = PAGE_SIZE - offset;
2024         len = ((to_write > len_max) ? len_max : to_write);
2025
2026         skb->data_len = to_write;
2027         skb->len += to_write;
2028         skb->truesize += to_write;
2029         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2030
2031         while (likely(to_write)) {
2032                 nr_frags = skb_shinfo(skb)->nr_frags;
2033
2034                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2035                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2036                                MAX_SKB_FRAGS);
2037                         return -EFAULT;
2038                 }
2039
2040                 page = pgv_to_page(data);
2041                 data += len;
2042                 flush_dcache_page(page);
2043                 get_page(page);
2044                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2045                 to_write -= len;
2046                 offset = 0;
2047                 len_max = PAGE_SIZE;
2048                 len = ((to_write > len_max) ? len_max : to_write);
2049         }
2050
2051         return tp_len;
2052 }
2053
2054 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
2055 {
2056         struct net_device *dev;
2057
2058         rcu_read_lock();
2059         dev = rcu_dereference(po->cached_dev);
2060         if (dev)
2061                 dev_hold(dev);
2062         rcu_read_unlock();
2063
2064         return dev;
2065 }
2066
2067 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2068 {
2069         struct sk_buff *skb;
2070         struct net_device *dev;
2071         __be16 proto;
2072         int err, reserve = 0;
2073         void *ph;
2074         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2075         int tp_len, size_max;
2076         unsigned char *addr;
2077         int len_sum = 0;
2078         int status = 0;
2079
2080         mutex_lock(&po->pg_vec_lock);
2081
2082         err = -EBUSY;
2083         if (saddr == NULL) {
2084                 dev     = packet_cached_dev_get(po);
2085                 proto   = po->num;
2086                 addr    = NULL;
2087         } else {
2088                 err = -EINVAL;
2089                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2090                         goto out;
2091                 if (msg->msg_namelen < (saddr->sll_halen
2092                                         + offsetof(struct sockaddr_ll,
2093                                                 sll_addr)))
2094                         goto out;
2095                 proto   = saddr->sll_protocol;
2096                 addr    = saddr->sll_addr;
2097                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2098         }
2099
2100         err = -ENXIO;
2101         if (unlikely(dev == NULL))
2102                 goto out;
2103         err = -ENETDOWN;
2104         if (unlikely(!(dev->flags & IFF_UP)))
2105                 goto out_put;
2106
2107         reserve = dev->hard_header_len;
2108
2109         size_max = po->tx_ring.frame_size
2110                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2111
2112         if (size_max > dev->mtu + reserve)
2113                 size_max = dev->mtu + reserve;
2114
2115         do {
2116                 ph = packet_current_frame(po, &po->tx_ring,
2117                                 TP_STATUS_SEND_REQUEST);
2118
2119                 if (unlikely(ph == NULL)) {
2120                         schedule();
2121                         continue;
2122                 }
2123
2124                 status = TP_STATUS_SEND_REQUEST;
2125                 skb = sock_alloc_send_skb(&po->sk,
2126                                 LL_ALLOCATED_SPACE(dev)
2127                                 + sizeof(struct sockaddr_ll),
2128                                 0, &err);
2129
2130                 if (unlikely(skb == NULL))
2131                         goto out_status;
2132
2133                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2134                                 addr);
2135
2136                 if (unlikely(tp_len < 0)) {
2137                         if (po->tp_loss) {
2138                                 __packet_set_status(po, ph,
2139                                                 TP_STATUS_AVAILABLE);
2140                                 packet_increment_head(&po->tx_ring);
2141                                 kfree_skb(skb);
2142                                 continue;
2143                         } else {
2144                                 status = TP_STATUS_WRONG_FORMAT;
2145                                 err = tp_len;
2146                                 goto out_status;
2147                         }
2148                 }
2149
2150                 skb->destructor = tpacket_destruct_skb;
2151                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2152                 atomic_inc(&po->tx_ring.pending);
2153
2154                 status = TP_STATUS_SEND_REQUEST;
2155                 err = dev_queue_xmit(skb);
2156                 if (unlikely(err > 0)) {
2157                         err = net_xmit_errno(err);
2158                         if (err && __packet_get_status(po, ph) ==
2159                                    TP_STATUS_AVAILABLE) {
2160                                 /* skb was destructed already */
2161                                 skb = NULL;
2162                                 goto out_status;
2163                         }
2164                         /*
2165                          * skb was dropped but not destructed yet;
2166                          * let's treat it like congestion or err < 0
2167                          */
2168                         err = 0;
2169                 }
2170                 packet_increment_head(&po->tx_ring);
2171                 len_sum += tp_len;
2172         } while (likely((ph != NULL) ||
2173                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2174                          (atomic_read(&po->tx_ring.pending))))
2175                 );
2176
2177         err = len_sum;
2178         goto out_put;
2179
2180 out_status:
2181         __packet_set_status(po, ph, status);
2182         kfree_skb(skb);
2183 out_put:
2184         dev_put(dev);
2185 out:
2186         mutex_unlock(&po->pg_vec_lock);
2187         return err;
2188 }
2189
2190 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2191                                         size_t reserve, size_t len,
2192                                         size_t linear, int noblock,
2193                                         int *err)
2194 {
2195         struct sk_buff *skb;
2196
2197         /* Under a page?  Don't bother with paged skb. */
2198         if (prepad + len < PAGE_SIZE || !linear)
2199                 linear = len;
2200
2201         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2202                                    err);
2203         if (!skb)
2204                 return NULL;
2205
2206         skb_reserve(skb, reserve);
2207         skb_put(skb, linear);
2208         skb->data_len = len - linear;
2209         skb->len += len - linear;
2210
2211         return skb;
2212 }
2213
2214 static int packet_snd(struct socket *sock,
2215                           struct msghdr *msg, size_t len)
2216 {
2217         struct sock *sk = sock->sk;
2218         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2219         struct sk_buff *skb;
2220         struct net_device *dev;
2221         __be16 proto;
2222         unsigned char *addr;
2223         int err, reserve = 0;
2224         struct virtio_net_hdr vnet_hdr = { 0 };
2225         int offset = 0;
2226         int vnet_hdr_len;
2227         struct packet_sock *po = pkt_sk(sk);
2228         unsigned short gso_type = 0;
2229
2230         /*
2231          *      Get and verify the address.
2232          */
2233
2234         if (saddr == NULL) {
2235                 dev     = packet_cached_dev_get(po);
2236                 proto   = po->num;
2237                 addr    = NULL;
2238         } else {
2239                 err = -EINVAL;
2240                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2241                         goto out;
2242                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2243                         goto out;
2244                 proto   = saddr->sll_protocol;
2245                 addr    = saddr->sll_addr;
2246                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2247         }
2248
2249         err = -ENXIO;
2250         if (unlikely(dev == NULL))
2251                 goto out_unlock;
2252         err = -ENETDOWN;
2253         if (unlikely(!(dev->flags & IFF_UP)))
2254                 goto out_unlock;
2255
2256         if (sock->type == SOCK_RAW)
2257                 reserve = dev->hard_header_len;
2258         if (po->has_vnet_hdr) {
2259                 vnet_hdr_len = sizeof(vnet_hdr);
2260
2261                 err = -EINVAL;
2262                 if (len < vnet_hdr_len)
2263                         goto out_unlock;
2264
2265                 len -= vnet_hdr_len;
2266
2267                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2268                                        vnet_hdr_len);
2269                 if (err < 0)
2270                         goto out_unlock;
2271
2272                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2273                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2274                       vnet_hdr.hdr_len))
2275                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2276                                                  vnet_hdr.csum_offset + 2;
2277
2278                 err = -EINVAL;
2279                 if (vnet_hdr.hdr_len > len)
2280                         goto out_unlock;
2281
2282                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2283                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2284                         case VIRTIO_NET_HDR_GSO_TCPV4:
2285                                 gso_type = SKB_GSO_TCPV4;
2286                                 break;
2287                         case VIRTIO_NET_HDR_GSO_TCPV6:
2288                                 gso_type = SKB_GSO_TCPV6;
2289                                 break;
2290                         case VIRTIO_NET_HDR_GSO_UDP:
2291                                 gso_type = SKB_GSO_UDP;
2292                                 break;
2293                         default:
2294                                 goto out_unlock;
2295                         }
2296
2297                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2298                                 gso_type |= SKB_GSO_TCP_ECN;
2299
2300                         if (vnet_hdr.gso_size == 0)
2301                                 goto out_unlock;
2302
2303                 }
2304         }
2305
2306         err = -EMSGSIZE;
2307         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2308                 goto out_unlock;
2309
2310         err = -ENOBUFS;
2311         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2312                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2313                                msg->msg_flags & MSG_DONTWAIT, &err);
2314         if (skb == NULL)
2315                 goto out_unlock;
2316
2317         skb_set_network_header(skb, reserve);
2318
2319         err = -EINVAL;
2320         if (sock->type == SOCK_DGRAM &&
2321             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2322                 goto out_free;
2323
2324         /* Returns -EFAULT on error */
2325         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2326         if (err)
2327                 goto out_free;
2328         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2329         if (err < 0)
2330                 goto out_free;
2331
2332         if (!gso_type && (len > dev->mtu + reserve)) {
2333                 /* Earlier code assumed this would be a VLAN pkt,
2334                  * double-check this now that we have the actual
2335                  * packet in hand.
2336                  */
2337                 struct ethhdr *ehdr;
2338                 skb_reset_mac_header(skb);
2339                 ehdr = eth_hdr(skb);
2340                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2341                         err = -EMSGSIZE;
2342                         goto out_free;
2343                 }
2344         }
2345
2346         skb->protocol = proto;
2347         skb->dev = dev;
2348         skb->priority = sk->sk_priority;
2349         skb->mark = sk->sk_mark;
2350
2351         if (po->has_vnet_hdr) {
2352                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2353                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2354                                                   vnet_hdr.csum_offset)) {
2355                                 err = -EINVAL;
2356                                 goto out_free;
2357                         }
2358                 }
2359
2360                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2361                 skb_shinfo(skb)->gso_type = gso_type;
2362
2363                 /* Header must be checked, and gso_segs computed. */
2364                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2365                 skb_shinfo(skb)->gso_segs = 0;
2366
2367                 len += vnet_hdr_len;
2368         }
2369
2370         /*
2371          *      Now send it
2372          */
2373
2374         err = dev_queue_xmit(skb);
2375         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2376                 goto out_unlock;
2377
2378         dev_put(dev);
2379
2380         return len;
2381
2382 out_free:
2383         kfree_skb(skb);
2384 out_unlock:
2385         if (dev)
2386                 dev_put(dev);
2387 out:
2388         return err;
2389 }
2390
2391 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2392                 struct msghdr *msg, size_t len)
2393 {
2394         struct sock *sk = sock->sk;
2395         struct packet_sock *po = pkt_sk(sk);
2396         if (po->tx_ring.pg_vec)
2397                 return tpacket_snd(po, msg);
2398         else
2399                 return packet_snd(sock, msg, len);
2400 }
2401
2402 /*
2403  *      Close a PACKET socket. This is fairly simple. We immediately go
2404  *      to 'closed' state and remove our protocol entry in the device list.
2405  */
2406
2407 static int packet_release(struct socket *sock)
2408 {
2409         struct sock *sk = sock->sk;
2410         struct packet_sock *po;
2411         struct packet_fanout *f;
2412         struct net *net;
2413         union tpacket_req_u req_u;
2414
2415         if (!sk)
2416                 return 0;
2417
2418         net = sock_net(sk);
2419         po = pkt_sk(sk);
2420
2421         spin_lock_bh(&net->packet.sklist_lock);
2422         sk_del_node_init_rcu(sk);
2423         sock_prot_inuse_add(net, sk->sk_prot, -1);
2424         spin_unlock_bh(&net->packet.sklist_lock);
2425
2426         spin_lock(&po->bind_lock);
2427         unregister_prot_hook(sk, false);
2428         if (po->prot_hook.dev) {
2429                 dev_put(po->prot_hook.dev);
2430                 po->prot_hook.dev = NULL;
2431         }
2432         spin_unlock(&po->bind_lock);
2433
2434         packet_flush_mclist(sk);
2435
2436         if (po->rx_ring.pg_vec) {
2437                 memset(&req_u, 0, sizeof(req_u));
2438                 packet_set_ring(sk, &req_u, 1, 0);
2439         }
2440
2441         if (po->tx_ring.pg_vec) {
2442                 memset(&req_u, 0, sizeof(req_u));
2443                 packet_set_ring(sk, &req_u, 1, 1);
2444         }
2445
2446         f = fanout_release(sk);
2447
2448         synchronize_net();
2449
2450         kfree(f);
2451
2452         /*
2453          *      Now the socket is dead. No more input will appear.
2454          */
2455         sock_orphan(sk);
2456         sock->sk = NULL;
2457
2458         /* Purge queues */
2459
2460         skb_queue_purge(&sk->sk_receive_queue);
2461         sk_refcnt_debug_release(sk);
2462
2463         sock_put(sk);
2464         return 0;
2465 }
2466
2467 /*
2468  *      Attach a packet hook.
2469  */
2470
2471 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2472 {
2473         struct packet_sock *po = pkt_sk(sk);
2474
2475         if (po->fanout) {
2476                 if (dev)
2477                         dev_put(dev);
2478
2479                 return -EINVAL;
2480         }
2481
2482         lock_sock(sk);
2483
2484         spin_lock(&po->bind_lock);
2485         unregister_prot_hook(sk, true);
2486         po->num = protocol;
2487         po->prot_hook.type = protocol;
2488         if (po->prot_hook.dev)
2489                 dev_put(po->prot_hook.dev);
2490         po->prot_hook.dev = dev;
2491
2492         po->ifindex = dev ? dev->ifindex : 0;
2493
2494         if (protocol == 0)
2495                 goto out_unlock;
2496
2497         if (!dev || (dev->flags & IFF_UP)) {
2498                 register_prot_hook(sk);
2499         } else {
2500                 sk->sk_err = ENETDOWN;
2501                 if (!sock_flag(sk, SOCK_DEAD))
2502                         sk->sk_error_report(sk);
2503         }
2504
2505 out_unlock:
2506         spin_unlock(&po->bind_lock);
2507         release_sock(sk);
2508         return 0;
2509 }
2510
2511 /*
2512  *      Bind a packet socket to a device
2513  */
2514
2515 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2516                             int addr_len)
2517 {
2518         struct sock *sk = sock->sk;
2519         char name[15];
2520         struct net_device *dev;
2521         int err = -ENODEV;
2522
2523         /*
2524          *      Check legality
2525          */
2526
2527         if (addr_len != sizeof(struct sockaddr))
2528                 return -EINVAL;
2529         strlcpy(name, uaddr->sa_data, sizeof(name));
2530
2531         dev = dev_get_by_name(sock_net(sk), name);
2532         if (dev)
2533                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2534         return err;
2535 }
2536
2537 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2538 {
2539         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2540         struct sock *sk = sock->sk;
2541         struct net_device *dev = NULL;
2542         int err;
2543
2544
2545         /*
2546          *      Check legality
2547          */
2548
2549         if (addr_len < sizeof(struct sockaddr_ll))
2550                 return -EINVAL;
2551         if (sll->sll_family != AF_PACKET)
2552                 return -EINVAL;
2553
2554         if (sll->sll_ifindex) {
2555                 err = -ENODEV;
2556                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2557                 if (dev == NULL)
2558                         goto out;
2559         }
2560         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2561
2562 out:
2563         return err;
2564 }
2565
2566 static struct proto packet_proto = {
2567         .name     = "PACKET",
2568         .owner    = THIS_MODULE,
2569         .obj_size = sizeof(struct packet_sock),
2570 };
2571
2572 /*
2573  *      Create a packet of type SOCK_PACKET.
2574  */
2575
2576 static int packet_create(struct net *net, struct socket *sock, int protocol,
2577                          int kern)
2578 {
2579         struct sock *sk;
2580         struct packet_sock *po;
2581         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2582         int err;
2583
2584         if (!capable(CAP_NET_RAW))
2585                 return -EPERM;
2586         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2587             sock->type != SOCK_PACKET)
2588                 return -ESOCKTNOSUPPORT;
2589
2590         sock->state = SS_UNCONNECTED;
2591
2592         err = -ENOBUFS;
2593         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2594         if (sk == NULL)
2595                 goto out;
2596
2597         sock->ops = &packet_ops;
2598         if (sock->type == SOCK_PACKET)
2599                 sock->ops = &packet_ops_spkt;
2600
2601         sock_init_data(sock, sk);
2602
2603         po = pkt_sk(sk);
2604         sk->sk_family = PF_PACKET;
2605         po->num = proto;
2606         RCU_INIT_POINTER(po->cached_dev, NULL);
2607
2608         sk->sk_destruct = packet_sock_destruct;
2609         sk_refcnt_debug_inc(sk);
2610
2611         /*
2612          *      Attach a protocol block
2613          */
2614
2615         spin_lock_init(&po->bind_lock);
2616         mutex_init(&po->pg_vec_lock);
2617         po->prot_hook.func = packet_rcv;
2618
2619         if (sock->type == SOCK_PACKET)
2620                 po->prot_hook.func = packet_rcv_spkt;
2621
2622         po->prot_hook.af_packet_priv = sk;
2623
2624         if (proto) {
2625                 po->prot_hook.type = proto;
2626                 register_prot_hook(sk);
2627         }
2628
2629         spin_lock_bh(&net->packet.sklist_lock);
2630         sk_add_node_rcu(sk, &net->packet.sklist);
2631         sock_prot_inuse_add(net, &packet_proto, 1);
2632         spin_unlock_bh(&net->packet.sklist_lock);
2633
2634         return 0;
2635 out:
2636         return err;
2637 }
2638
2639 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2640 {
2641         struct sock_exterr_skb *serr;
2642         struct sk_buff *skb, *skb2;
2643         int copied, err;
2644
2645         err = -EAGAIN;
2646         skb = skb_dequeue(&sk->sk_error_queue);
2647         if (skb == NULL)
2648                 goto out;
2649
2650         copied = skb->len;
2651         if (copied > len) {
2652                 msg->msg_flags |= MSG_TRUNC;
2653                 copied = len;
2654         }
2655         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2656         if (err)
2657                 goto out_free_skb;
2658
2659         sock_recv_timestamp(msg, sk, skb);
2660
2661         serr = SKB_EXT_ERR(skb);
2662         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2663                  sizeof(serr->ee), &serr->ee);
2664
2665         msg->msg_flags |= MSG_ERRQUEUE;
2666         err = copied;
2667
2668         /* Reset and regenerate socket error */
2669         spin_lock_bh(&sk->sk_error_queue.lock);
2670         sk->sk_err = 0;
2671         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2672                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2673                 spin_unlock_bh(&sk->sk_error_queue.lock);
2674                 sk->sk_error_report(sk);
2675         } else
2676                 spin_unlock_bh(&sk->sk_error_queue.lock);
2677
2678 out_free_skb:
2679         kfree_skb(skb);
2680 out:
2681         return err;
2682 }
2683
2684 /*
2685  *      Pull a packet from our receive queue and hand it to the user.
2686  *      If necessary we block.
2687  */
2688
2689 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2690                           struct msghdr *msg, size_t len, int flags)
2691 {
2692         struct sock *sk = sock->sk;
2693         struct sk_buff *skb;
2694         int copied, err;
2695         int vnet_hdr_len = 0;
2696
2697         err = -EINVAL;
2698         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2699                 goto out;
2700
2701 #if 0
2702         /* What error should we return now? EUNATTACH? */
2703         if (pkt_sk(sk)->ifindex < 0)
2704                 return -ENODEV;
2705 #endif
2706
2707         if (flags & MSG_ERRQUEUE) {
2708                 err = packet_recv_error(sk, msg, len);
2709                 goto out;
2710         }
2711
2712         /*
2713          *      Call the generic datagram receiver. This handles all sorts
2714          *      of horrible races and re-entrancy so we can forget about it
2715          *      in the protocol layers.
2716          *
2717          *      Now it will return ENETDOWN, if device have just gone down,
2718          *      but then it will block.
2719          */
2720
2721         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2722
2723         /*
2724          *      An error occurred so return it. Because skb_recv_datagram()
2725          *      handles the blocking we don't see and worry about blocking
2726          *      retries.
2727          */
2728
2729         if (skb == NULL)
2730                 goto out;
2731
2732         if (pkt_sk(sk)->has_vnet_hdr) {
2733                 struct virtio_net_hdr vnet_hdr = { 0 };
2734
2735                 err = -EINVAL;
2736                 vnet_hdr_len = sizeof(vnet_hdr);
2737                 if (len < vnet_hdr_len)
2738                         goto out_free;
2739
2740                 len -= vnet_hdr_len;
2741
2742                 if (skb_is_gso(skb)) {
2743                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2744
2745                         /* This is a hint as to how much should be linear. */
2746                         vnet_hdr.hdr_len = skb_headlen(skb);
2747                         vnet_hdr.gso_size = sinfo->gso_size;
2748                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2749                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2750                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2751                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2752                         else if (sinfo->gso_type & SKB_GSO_UDP)
2753                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2754                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2755                                 goto out_free;
2756                         else
2757                                 BUG();
2758                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2759                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2760                 } else
2761                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2762
2763                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2764                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2765                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2766                         vnet_hdr.csum_offset = skb->csum_offset;
2767                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2768                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2769                 } /* else everything is zero */
2770
2771                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2772                                      vnet_hdr_len);
2773                 if (err < 0)
2774                         goto out_free;
2775         }
2776
2777         /* You lose any data beyond the buffer you gave. If it worries
2778          * a user program they can ask the device for its MTU
2779          * anyway.
2780          */
2781         copied = skb->len;
2782         if (copied > len) {
2783                 copied = len;
2784                 msg->msg_flags |= MSG_TRUNC;
2785         }
2786
2787         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2788         if (err)
2789                 goto out_free;
2790
2791         sock_recv_ts_and_drops(msg, sk, skb);
2792
2793         if (msg->msg_name) {
2794                 /* If the address length field is there to be filled
2795                  * in, we fill it in now.
2796                  */
2797                 if (sock->type == SOCK_PACKET) {
2798                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
2799                 } else {
2800                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2801                         msg->msg_namelen = sll->sll_halen +
2802                                 offsetof(struct sockaddr_ll, sll_addr);
2803                 }
2804                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2805                        msg->msg_namelen);
2806         }
2807
2808         if (pkt_sk(sk)->auxdata) {
2809                 struct tpacket_auxdata aux;
2810
2811                 aux.tp_status = TP_STATUS_USER;
2812                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2813                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2814                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2815                 aux.tp_snaplen = skb->len;
2816                 aux.tp_mac = 0;
2817                 aux.tp_net = skb_network_offset(skb);
2818                 if (vlan_tx_tag_present(skb)) {
2819                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2820                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2821                 } else {
2822                         aux.tp_vlan_tci = 0;
2823                 }
2824                 aux.tp_padding = 0;
2825                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2826         }
2827
2828         /*
2829          *      Free or return the buffer as appropriate. Again this
2830          *      hides all the races and re-entrancy issues from us.
2831          */
2832         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2833
2834 out_free:
2835         skb_free_datagram(sk, skb);
2836 out:
2837         return err;
2838 }
2839
2840 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2841                                int *uaddr_len, int peer)
2842 {
2843         struct net_device *dev;
2844         struct sock *sk = sock->sk;
2845
2846         if (peer)
2847                 return -EOPNOTSUPP;
2848
2849         uaddr->sa_family = AF_PACKET;
2850         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
2851         rcu_read_lock();
2852         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2853         if (dev)
2854                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
2855         rcu_read_unlock();
2856         *uaddr_len = sizeof(*uaddr);
2857
2858         return 0;
2859 }
2860
2861 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2862                           int *uaddr_len, int peer)
2863 {
2864         struct net_device *dev;
2865         struct sock *sk = sock->sk;
2866         struct packet_sock *po = pkt_sk(sk);
2867         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2868
2869         if (peer)
2870                 return -EOPNOTSUPP;
2871
2872         sll->sll_family = AF_PACKET;
2873         sll->sll_ifindex = po->ifindex;
2874         sll->sll_protocol = po->num;
2875         sll->sll_pkttype = 0;
2876         rcu_read_lock();
2877         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2878         if (dev) {
2879                 sll->sll_hatype = dev->type;
2880                 sll->sll_halen = dev->addr_len;
2881                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2882         } else {
2883                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2884                 sll->sll_halen = 0;
2885         }
2886         rcu_read_unlock();
2887         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2888
2889         return 0;
2890 }
2891
2892 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2893                          int what)
2894 {
2895         switch (i->type) {
2896         case PACKET_MR_MULTICAST:
2897                 if (i->alen != dev->addr_len)
2898                         return -EINVAL;
2899                 if (what > 0)
2900                         return dev_mc_add(dev, i->addr);
2901                 else
2902                         return dev_mc_del(dev, i->addr);
2903                 break;
2904         case PACKET_MR_PROMISC:
2905                 return dev_set_promiscuity(dev, what);
2906                 break;
2907         case PACKET_MR_ALLMULTI:
2908                 return dev_set_allmulti(dev, what);
2909                 break;
2910         case PACKET_MR_UNICAST:
2911                 if (i->alen != dev->addr_len)
2912                         return -EINVAL;
2913                 if (what > 0)
2914                         return dev_uc_add(dev, i->addr);
2915                 else
2916                         return dev_uc_del(dev, i->addr);
2917                 break;
2918         default:
2919                 break;
2920         }
2921         return 0;
2922 }
2923
2924 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2925 {
2926         for ( ; i; i = i->next) {
2927                 if (i->ifindex == dev->ifindex)
2928                         packet_dev_mc(dev, i, what);
2929         }
2930 }
2931
2932 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2933 {
2934         struct packet_sock *po = pkt_sk(sk);
2935         struct packet_mclist *ml, *i;
2936         struct net_device *dev;
2937         int err;
2938
2939         rtnl_lock();
2940
2941         err = -ENODEV;
2942         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2943         if (!dev)
2944                 goto done;
2945
2946         err = -EINVAL;
2947         if (mreq->mr_alen > dev->addr_len)
2948                 goto done;
2949
2950         err = -ENOBUFS;
2951         i = kmalloc(sizeof(*i), GFP_KERNEL);
2952         if (i == NULL)
2953                 goto done;
2954
2955         err = 0;
2956         for (ml = po->mclist; ml; ml = ml->next) {
2957                 if (ml->ifindex == mreq->mr_ifindex &&
2958                     ml->type == mreq->mr_type &&
2959                     ml->alen == mreq->mr_alen &&
2960                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2961                         ml->count++;
2962                         /* Free the new element ... */
2963                         kfree(i);
2964                         goto done;
2965                 }
2966         }
2967
2968         i->type = mreq->mr_type;
2969         i->ifindex = mreq->mr_ifindex;
2970         i->alen = mreq->mr_alen;
2971         memcpy(i->addr, mreq->mr_address, i->alen);
2972         i->count = 1;
2973         i->next = po->mclist;
2974         po->mclist = i;
2975         err = packet_dev_mc(dev, i, 1);
2976         if (err) {
2977                 po->mclist = i->next;
2978                 kfree(i);
2979         }
2980
2981 done:
2982         rtnl_unlock();
2983         return err;
2984 }
2985
2986 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2987 {
2988         struct packet_mclist *ml, **mlp;
2989
2990         rtnl_lock();
2991
2992         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2993                 if (ml->ifindex == mreq->mr_ifindex &&
2994                     ml->type == mreq->mr_type &&
2995                     ml->alen == mreq->mr_alen &&
2996                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2997                         if (--ml->count == 0) {
2998                                 struct net_device *dev;
2999                                 *mlp = ml->next;
3000                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3001                                 if (dev)
3002                                         packet_dev_mc(dev, ml, -1);
3003                                 kfree(ml);
3004                         }
3005                         rtnl_unlock();
3006                         return 0;
3007                 }
3008         }
3009         rtnl_unlock();
3010         return -EADDRNOTAVAIL;
3011 }
3012
3013 static void packet_flush_mclist(struct sock *sk)
3014 {
3015         struct packet_sock *po = pkt_sk(sk);
3016         struct packet_mclist *ml;
3017
3018         if (!po->mclist)
3019                 return;
3020
3021         rtnl_lock();
3022         while ((ml = po->mclist) != NULL) {
3023                 struct net_device *dev;
3024
3025                 po->mclist = ml->next;
3026                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3027                 if (dev != NULL)
3028                         packet_dev_mc(dev, ml, -1);
3029                 kfree(ml);
3030         }
3031         rtnl_unlock();
3032 }
3033
3034 static int
3035 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3036 {
3037         struct sock *sk = sock->sk;
3038         struct packet_sock *po = pkt_sk(sk);
3039         int ret;
3040
3041         if (level != SOL_PACKET)
3042                 return -ENOPROTOOPT;
3043
3044         switch (optname) {
3045         case PACKET_ADD_MEMBERSHIP:
3046         case PACKET_DROP_MEMBERSHIP:
3047         {
3048                 struct packet_mreq_max mreq;
3049                 int len = optlen;
3050                 memset(&mreq, 0, sizeof(mreq));
3051                 if (len < sizeof(struct packet_mreq))
3052                         return -EINVAL;
3053                 if (len > sizeof(mreq))
3054                         len = sizeof(mreq);
3055                 if (copy_from_user(&mreq, optval, len))
3056                         return -EFAULT;
3057                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3058                         return -EINVAL;
3059                 if (optname == PACKET_ADD_MEMBERSHIP)
3060                         ret = packet_mc_add(sk, &mreq);
3061                 else
3062                         ret = packet_mc_drop(sk, &mreq);
3063                 return ret;
3064         }
3065
3066         case PACKET_RX_RING:
3067         case PACKET_TX_RING:
3068         {
3069                 union tpacket_req_u req_u;
3070                 int len;
3071
3072                 switch (po->tp_version) {
3073                 case TPACKET_V1:
3074                 case TPACKET_V2:
3075                         len = sizeof(req_u.req);
3076                         break;
3077                 case TPACKET_V3:
3078                 default:
3079                         len = sizeof(req_u.req3);
3080                         break;
3081                 }
3082                 if (optlen < len)
3083                         return -EINVAL;
3084                 if (pkt_sk(sk)->has_vnet_hdr)
3085                         return -EINVAL;
3086                 if (copy_from_user(&req_u.req, optval, len))
3087                         return -EFAULT;
3088                 return packet_set_ring(sk, &req_u, 0,
3089                         optname == PACKET_TX_RING);
3090         }
3091         case PACKET_COPY_THRESH:
3092         {
3093                 int val;
3094
3095                 if (optlen != sizeof(val))
3096                         return -EINVAL;
3097                 if (copy_from_user(&val, optval, sizeof(val)))
3098                         return -EFAULT;
3099
3100                 pkt_sk(sk)->copy_thresh = val;
3101                 return 0;
3102         }
3103         case PACKET_VERSION:
3104         {
3105                 int val;
3106
3107                 if (optlen != sizeof(val))
3108                         return -EINVAL;
3109                 if (copy_from_user(&val, optval, sizeof(val)))
3110                         return -EFAULT;
3111                 switch (val) {
3112                 case TPACKET_V1:
3113                 case TPACKET_V2:
3114                 case TPACKET_V3:
3115                         break;
3116                 default:
3117                         return -EINVAL;
3118                 }
3119                 lock_sock(sk);
3120                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3121                         ret = -EBUSY;
3122                 } else {
3123                         po->tp_version = val;
3124                         ret = 0;
3125                 }
3126                 release_sock(sk);
3127                 return ret;
3128         }
3129         case PACKET_RESERVE:
3130         {
3131                 unsigned int val;
3132
3133                 if (optlen != sizeof(val))
3134                         return -EINVAL;
3135                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3136                         return -EBUSY;
3137                 if (copy_from_user(&val, optval, sizeof(val)))
3138                         return -EFAULT;
3139                 po->tp_reserve = val;
3140                 return 0;
3141         }
3142         case PACKET_LOSS:
3143         {
3144                 unsigned int val;
3145
3146                 if (optlen != sizeof(val))
3147                         return -EINVAL;
3148                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3149                         return -EBUSY;
3150                 if (copy_from_user(&val, optval, sizeof(val)))
3151                         return -EFAULT;
3152                 po->tp_loss = !!val;
3153                 return 0;
3154         }
3155         case PACKET_AUXDATA:
3156         {
3157                 int val;
3158
3159                 if (optlen < sizeof(val))
3160                         return -EINVAL;
3161                 if (copy_from_user(&val, optval, sizeof(val)))
3162                         return -EFAULT;
3163
3164                 po->auxdata = !!val;
3165                 return 0;
3166         }
3167         case PACKET_ORIGDEV:
3168         {
3169                 int val;
3170
3171                 if (optlen < sizeof(val))
3172                         return -EINVAL;
3173                 if (copy_from_user(&val, optval, sizeof(val)))
3174                         return -EFAULT;
3175
3176                 po->origdev = !!val;
3177                 return 0;
3178         }
3179         case PACKET_VNET_HDR:
3180         {
3181                 int val;
3182
3183                 if (sock->type != SOCK_RAW)
3184                         return -EINVAL;
3185                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3186                         return -EBUSY;
3187                 if (optlen < sizeof(val))
3188                         return -EINVAL;
3189                 if (copy_from_user(&val, optval, sizeof(val)))
3190                         return -EFAULT;
3191
3192                 po->has_vnet_hdr = !!val;
3193                 return 0;
3194         }
3195         case PACKET_TIMESTAMP:
3196         {
3197                 int val;
3198
3199                 if (optlen != sizeof(val))
3200                         return -EINVAL;
3201                 if (copy_from_user(&val, optval, sizeof(val)))
3202                         return -EFAULT;
3203
3204                 po->tp_tstamp = val;
3205                 return 0;
3206         }
3207         case PACKET_FANOUT:
3208         {
3209                 int val;
3210
3211                 if (optlen != sizeof(val))
3212                         return -EINVAL;
3213                 if (copy_from_user(&val, optval, sizeof(val)))
3214                         return -EFAULT;
3215
3216                 return fanout_add(sk, val & 0xffff, val >> 16);
3217         }
3218         default:
3219                 return -ENOPROTOOPT;
3220         }
3221 }
3222
3223 static int packet_getsockopt(struct socket *sock, int level, int optname,
3224                              char __user *optval, int __user *optlen)
3225 {
3226         int len;
3227         int val;
3228         struct sock *sk = sock->sk;
3229         struct packet_sock *po = pkt_sk(sk);
3230         void *data;
3231         struct tpacket_stats st;
3232         union tpacket_stats_u st_u;
3233
3234         if (level != SOL_PACKET)
3235                 return -ENOPROTOOPT;
3236
3237         if (get_user(len, optlen))
3238                 return -EFAULT;
3239
3240         if (len < 0)
3241                 return -EINVAL;
3242
3243         switch (optname) {
3244         case PACKET_STATISTICS:
3245                 if (po->tp_version == TPACKET_V3) {
3246                         len = sizeof(struct tpacket_stats_v3);
3247                 } else {
3248                         if (len > sizeof(struct tpacket_stats))
3249                                 len = sizeof(struct tpacket_stats);
3250                 }
3251                 spin_lock_bh(&sk->sk_receive_queue.lock);
3252                 if (po->tp_version == TPACKET_V3) {
3253                         memcpy(&st_u.stats3, &po->stats,
3254                         sizeof(struct tpacket_stats));
3255                         st_u.stats3.tp_freeze_q_cnt =
3256                         po->stats_u.stats3.tp_freeze_q_cnt;
3257                         st_u.stats3.tp_packets += po->stats.tp_drops;
3258                         data = &st_u.stats3;
3259                 } else {
3260                         st = po->stats;
3261                         st.tp_packets += st.tp_drops;
3262                         data = &st;
3263                 }
3264                 memset(&po->stats, 0, sizeof(st));
3265                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3266                 break;
3267         case PACKET_AUXDATA:
3268                 if (len > sizeof(int))
3269                         len = sizeof(int);
3270                 val = po->auxdata;
3271
3272                 data = &val;
3273                 break;
3274         case PACKET_ORIGDEV:
3275                 if (len > sizeof(int))
3276                         len = sizeof(int);
3277                 val = po->origdev;
3278
3279                 data = &val;
3280                 break;
3281         case PACKET_VNET_HDR:
3282                 if (len > sizeof(int))
3283                         len = sizeof(int);
3284                 val = po->has_vnet_hdr;
3285
3286                 data = &val;
3287                 break;
3288         case PACKET_VERSION:
3289                 if (len > sizeof(int))
3290                         len = sizeof(int);
3291                 val = po->tp_version;
3292                 data = &val;
3293                 break;
3294         case PACKET_HDRLEN:
3295                 if (len > sizeof(int))
3296                         len = sizeof(int);
3297                 if (copy_from_user(&val, optval, len))
3298                         return -EFAULT;
3299                 switch (val) {
3300                 case TPACKET_V1:
3301                         val = sizeof(struct tpacket_hdr);
3302                         break;
3303                 case TPACKET_V2:
3304                         val = sizeof(struct tpacket2_hdr);
3305                         break;
3306                 case TPACKET_V3:
3307                         val = sizeof(struct tpacket3_hdr);
3308                         break;
3309                 default:
3310                         return -EINVAL;
3311                 }
3312                 data = &val;
3313                 break;
3314         case PACKET_RESERVE:
3315                 if (len > sizeof(unsigned int))
3316                         len = sizeof(unsigned int);
3317                 val = po->tp_reserve;
3318                 data = &val;
3319                 break;
3320         case PACKET_LOSS:
3321                 if (len > sizeof(unsigned int))
3322                         len = sizeof(unsigned int);
3323                 val = po->tp_loss;
3324                 data = &val;
3325                 break;
3326         case PACKET_TIMESTAMP:
3327                 if (len > sizeof(int))
3328                         len = sizeof(int);
3329                 val = po->tp_tstamp;
3330                 data = &val;
3331                 break;
3332         case PACKET_FANOUT:
3333                 if (len > sizeof(int))
3334                         len = sizeof(int);
3335                 val = (po->fanout ?
3336                        ((u32)po->fanout->id |
3337                         ((u32)po->fanout->type << 16)) :
3338                        0);
3339                 data = &val;
3340                 break;
3341         default:
3342                 return -ENOPROTOOPT;
3343         }
3344
3345         if (put_user(len, optlen))
3346                 return -EFAULT;
3347         if (copy_to_user(optval, data, len))
3348                 return -EFAULT;
3349         return 0;
3350 }
3351
3352
3353 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3354 {
3355         struct sock *sk;
3356         struct hlist_node *node;
3357         struct net_device *dev = data;
3358         struct net *net = dev_net(dev);
3359
3360         rcu_read_lock();
3361         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3362                 struct packet_sock *po = pkt_sk(sk);
3363
3364                 switch (msg) {
3365                 case NETDEV_UNREGISTER:
3366                         if (po->mclist)
3367                                 packet_dev_mclist(dev, po->mclist, -1);
3368                         /* fallthrough */
3369
3370                 case NETDEV_DOWN:
3371                         if (dev->ifindex == po->ifindex) {
3372                                 spin_lock(&po->bind_lock);
3373                                 if (po->running) {
3374                                         __unregister_prot_hook(sk, false);
3375                                         sk->sk_err = ENETDOWN;
3376                                         if (!sock_flag(sk, SOCK_DEAD))
3377                                                 sk->sk_error_report(sk);
3378                                 }
3379                                 if (msg == NETDEV_UNREGISTER) {
3380                                         po->ifindex = -1;
3381                                         if (po->prot_hook.dev)
3382                                                 dev_put(po->prot_hook.dev);
3383                                         po->prot_hook.dev = NULL;
3384                                 }
3385                                 spin_unlock(&po->bind_lock);
3386                         }
3387                         break;
3388                 case NETDEV_UP:
3389                         if (dev->ifindex == po->ifindex) {
3390                                 spin_lock(&po->bind_lock);
3391                                 if (po->num)
3392                                         register_prot_hook(sk);
3393                                 spin_unlock(&po->bind_lock);
3394                         }
3395                         break;
3396                 }
3397         }
3398         rcu_read_unlock();
3399         return NOTIFY_DONE;
3400 }
3401
3402
3403 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3404                         unsigned long arg)
3405 {
3406         struct sock *sk = sock->sk;
3407
3408         switch (cmd) {
3409         case SIOCOUTQ:
3410         {
3411                 int amount = sk_wmem_alloc_get(sk);
3412
3413                 return put_user(amount, (int __user *)arg);
3414         }
3415         case SIOCINQ:
3416         {
3417                 struct sk_buff *skb;
3418                 int amount = 0;
3419
3420                 spin_lock_bh(&sk->sk_receive_queue.lock);
3421                 skb = skb_peek(&sk->sk_receive_queue);
3422                 if (skb)
3423                         amount = skb->len;
3424                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3425                 return put_user(amount, (int __user *)arg);
3426         }
3427         case SIOCGSTAMP:
3428                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3429         case SIOCGSTAMPNS:
3430                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3431
3432 #ifdef CONFIG_INET
3433         case SIOCADDRT:
3434         case SIOCDELRT:
3435         case SIOCDARP:
3436         case SIOCGARP:
3437         case SIOCSARP:
3438         case SIOCGIFADDR:
3439         case SIOCSIFADDR:
3440         case SIOCGIFBRDADDR:
3441         case SIOCSIFBRDADDR:
3442         case SIOCGIFNETMASK:
3443         case SIOCSIFNETMASK:
3444         case SIOCGIFDSTADDR:
3445         case SIOCSIFDSTADDR:
3446         case SIOCSIFFLAGS:
3447                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3448 #endif
3449
3450         default:
3451                 return -ENOIOCTLCMD;
3452         }
3453         return 0;
3454 }
3455
3456 static unsigned int packet_poll(struct file *file, struct socket *sock,
3457                                 poll_table *wait)
3458 {
3459         struct sock *sk = sock->sk;
3460         struct packet_sock *po = pkt_sk(sk);
3461         unsigned int mask = datagram_poll(file, sock, wait);
3462
3463         spin_lock_bh(&sk->sk_receive_queue.lock);
3464         if (po->rx_ring.pg_vec) {
3465                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3466                         TP_STATUS_KERNEL))
3467                         mask |= POLLIN | POLLRDNORM;
3468         }
3469         spin_unlock_bh(&sk->sk_receive_queue.lock);
3470         spin_lock_bh(&sk->sk_write_queue.lock);
3471         if (po->tx_ring.pg_vec) {
3472                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3473                         mask |= POLLOUT | POLLWRNORM;
3474         }
3475         spin_unlock_bh(&sk->sk_write_queue.lock);
3476         return mask;
3477 }
3478
3479
3480 /* Dirty? Well, I still did not learn better way to account
3481  * for user mmaps.
3482  */
3483
3484 static void packet_mm_open(struct vm_area_struct *vma)
3485 {
3486         struct file *file = vma->vm_file;
3487         struct socket *sock = file->private_data;
3488         struct sock *sk = sock->sk;
3489
3490         if (sk)
3491                 atomic_inc(&pkt_sk(sk)->mapped);
3492 }
3493
3494 static void packet_mm_close(struct vm_area_struct *vma)
3495 {
3496         struct file *file = vma->vm_file;
3497         struct socket *sock = file->private_data;
3498         struct sock *sk = sock->sk;
3499
3500         if (sk)
3501                 atomic_dec(&pkt_sk(sk)->mapped);
3502 }
3503
3504 static const struct vm_operations_struct packet_mmap_ops = {
3505         .open   =       packet_mm_open,
3506         .close  =       packet_mm_close,
3507 };
3508
3509 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3510                         unsigned int len)
3511 {
3512         int i;
3513
3514         for (i = 0; i < len; i++) {
3515                 if (likely(pg_vec[i].buffer)) {
3516                         if (is_vmalloc_addr(pg_vec[i].buffer))
3517                                 vfree(pg_vec[i].buffer);
3518                         else
3519                                 free_pages((unsigned long)pg_vec[i].buffer,
3520                                            order);
3521                         pg_vec[i].buffer = NULL;
3522                 }
3523         }
3524         kfree(pg_vec);
3525 }
3526
3527 static char *alloc_one_pg_vec_page(unsigned long order)
3528 {
3529         char *buffer = NULL;
3530         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3531                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3532
3533         buffer = (char *) __get_free_pages(gfp_flags, order);
3534
3535         if (buffer)
3536                 return buffer;
3537
3538         /*
3539          * __get_free_pages failed, fall back to vmalloc
3540          */
3541         buffer = vzalloc((1 << order) * PAGE_SIZE);
3542
3543         if (buffer)
3544                 return buffer;
3545
3546         /*
3547          * vmalloc failed, lets dig into swap here
3548          */
3549         gfp_flags &= ~__GFP_NORETRY;
3550         buffer = (char *)__get_free_pages(gfp_flags, order);
3551         if (buffer)
3552                 return buffer;
3553
3554         /*
3555          * complete and utter failure
3556          */
3557         return NULL;
3558 }
3559
3560 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3561 {
3562         unsigned int block_nr = req->tp_block_nr;
3563         struct pgv *pg_vec;
3564         int i;
3565
3566         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3567         if (unlikely(!pg_vec))
3568                 goto out;
3569
3570         for (i = 0; i < block_nr; i++) {
3571                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3572                 if (unlikely(!pg_vec[i].buffer))
3573                         goto out_free_pgvec;
3574         }
3575
3576 out:
3577         return pg_vec;
3578
3579 out_free_pgvec:
3580         free_pg_vec(pg_vec, order, block_nr);
3581         pg_vec = NULL;
3582         goto out;
3583 }
3584
3585 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3586                 int closing, int tx_ring)
3587 {
3588         struct pgv *pg_vec = NULL;
3589         struct packet_sock *po = pkt_sk(sk);
3590         int was_running, order = 0;
3591         struct packet_ring_buffer *rb;
3592         struct sk_buff_head *rb_queue;
3593         __be16 num;
3594         int err = -EINVAL;
3595         /* Added to avoid minimal code churn */
3596         struct tpacket_req *req = &req_u->req;
3597
3598         lock_sock(sk);
3599         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3600         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3601                 WARN(1, "Tx-ring is not supported.\n");
3602                 goto out;
3603         }
3604
3605         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3606         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3607
3608         err = -EBUSY;
3609         if (!closing) {
3610                 if (atomic_read(&po->mapped))
3611                         goto out;
3612                 if (atomic_read(&rb->pending))
3613                         goto out;
3614         }
3615
3616         if (req->tp_block_nr) {
3617                 /* Sanity tests and some calculations */
3618                 err = -EBUSY;
3619                 if (unlikely(rb->pg_vec))
3620                         goto out;
3621
3622                 switch (po->tp_version) {
3623                 case TPACKET_V1:
3624                         po->tp_hdrlen = TPACKET_HDRLEN;
3625                         break;
3626                 case TPACKET_V2:
3627                         po->tp_hdrlen = TPACKET2_HDRLEN;
3628                         break;
3629                 case TPACKET_V3:
3630                         po->tp_hdrlen = TPACKET3_HDRLEN;
3631                         break;
3632                 }
3633
3634                 err = -EINVAL;
3635                 if (unlikely((int)req->tp_block_size <= 0))
3636                         goto out;
3637                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3638                         goto out;
3639                 if (po->tp_version >= TPACKET_V3 &&
3640                     req->tp_block_size <=
3641                           BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
3642                         goto out;
3643                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3644                                         po->tp_reserve))
3645                         goto out;
3646                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3647                         goto out;
3648
3649                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3650                 if (unlikely(rb->frames_per_block <= 0))
3651                         goto out;
3652                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3653                                         req->tp_frame_nr))
3654                         goto out;
3655
3656                 err = -ENOMEM;
3657                 order = get_order(req->tp_block_size);
3658                 pg_vec = alloc_pg_vec(req, order);
3659                 if (unlikely(!pg_vec))
3660                         goto out;
3661                 switch (po->tp_version) {
3662                 case TPACKET_V3:
3663                 /* Transmit path is not supported. We checked