net: Add framework to allow sending packets with customized CRC.
[linux-3.10.git] / net / packet / af_packet.c
1 /*
2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
3  *              operating system.  INET is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              PACKET - implements raw packet sockets.
7  *
8  * Authors:     Ross Biro
9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
11  *
12  * Fixes:
13  *              Alan Cox        :       verify_area() now used correctly
14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
15  *              Alan Cox        :       tidied skbuff lists.
16  *              Alan Cox        :       Now uses generic datagram routines I
17  *                                      added. Also fixed the peek/read crash
18  *                                      from all old Linux datagram code.
19  *              Alan Cox        :       Uses the improved datagram code.
20  *              Alan Cox        :       Added NULL's for socket options.
21  *              Alan Cox        :       Re-commented the code.
22  *              Alan Cox        :       Use new kernel side addressing
23  *              Rob Janssen     :       Correct MTU usage.
24  *              Dave Platt      :       Counter leaks caused by incorrect
25  *                                      interrupt locking and some slightly
26  *                                      dubious gcc output. Can you read
27  *                                      compiler: it said _VOLATILE_
28  *      Richard Kooijman        :       Timestamp fixes.
29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
30  *              Alan Cox        :       sendmsg/recvmsg support.
31  *              Alan Cox        :       Protocol setting support
32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
34  *      Michal Ostrowski        :       Module initialization cleanup.
35  *         Ulises Alonso        :       Frame number limit removal and
36  *                                      packet_set_ring memory leak.
37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
38  *                                      The convention is that longer addresses
39  *                                      will simply extend the hardware address
40  *                                      byte arrays at the end of sockaddr_ll
41  *                                      and packet_mreq.
42  *              Johann Baudy    :       Added TX RING.
43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
44  *                                      layer.
45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
46  *
47  *
48  *              This program is free software; you can redistribute it and/or
49  *              modify it under the terms of the GNU General Public License
50  *              as published by the Free Software Foundation; either version
51  *              2 of the License, or (at your option) any later version.
52  *
53  */
54
55 #include <linux/types.h>
56 #include <linux/mm.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
60 #include <linux/in.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
70 #include <net/ip.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
73 #include <net/sock.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/system.h>
77 #include <asm/uaccess.h>
78 #include <asm/ioctls.h>
79 #include <asm/page.h>
80 #include <asm/cacheflush.h>
81 #include <asm/io.h>
82 #include <linux/proc_fs.h>
83 #include <linux/seq_file.h>
84 #include <linux/poll.h>
85 #include <linux/module.h>
86 #include <linux/init.h>
87 #include <linux/mutex.h>
88 #include <linux/if_vlan.h>
89 #include <linux/virtio_net.h>
90 #include <linux/errqueue.h>
91 #include <linux/net_tstamp.h>
92
93 #ifdef CONFIG_INET
94 #include <net/inet_common.h>
95 #endif
96
97 /*
98    Assumptions:
99    - if device has no dev->hard_header routine, it adds and removes ll header
100      inside itself. In this case ll header is invisible outside of device,
101      but higher levels still should reserve dev->hard_header_len.
102      Some devices are enough clever to reallocate skb, when header
103      will not fit to reserved space (tunnel), another ones are silly
104      (PPP).
105    - packet socket receives packets with pulled ll header,
106      so that SOCK_RAW should push it back.
107
108 On receive:
109 -----------
110
111 Incoming, dev->hard_header!=NULL
112    mac_header -> ll header
113    data       -> data
114
115 Outgoing, dev->hard_header!=NULL
116    mac_header -> ll header
117    data       -> ll header
118
119 Incoming, dev->hard_header==NULL
120    mac_header -> UNKNOWN position. It is very likely, that it points to ll
121                  header.  PPP makes it, that is wrong, because introduce
122                  assymetry between rx and tx paths.
123    data       -> data
124
125 Outgoing, dev->hard_header==NULL
126    mac_header -> data. ll header is still not built!
127    data       -> data
128
129 Resume
130   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
131
132
133 On transmit:
134 ------------
135
136 dev->hard_header != NULL
137    mac_header -> ll header
138    data       -> ll header
139
140 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141    mac_header -> data
142    data       -> data
143
144    We should set nh.raw on output to correct posistion,
145    packet classifier depends on it.
146  */
147
148 /* Private packet socket structures. */
149
150 struct packet_mclist {
151         struct packet_mclist    *next;
152         int                     ifindex;
153         int                     count;
154         unsigned short          type;
155         unsigned short          alen;
156         unsigned char           addr[MAX_ADDR_LEN];
157 };
158 /* identical to struct packet_mreq except it has
159  * a longer address field.
160  */
161 struct packet_mreq_max {
162         int             mr_ifindex;
163         unsigned short  mr_type;
164         unsigned short  mr_alen;
165         unsigned char   mr_address[MAX_ADDR_LEN];
166 };
167
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169                 int closing, int tx_ring);
170
171
172 #define V3_ALIGNMENT    (8)
173
174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179 /* kbdq - kernel block descriptor queue */
180 struct tpacket_kbdq_core {
181         struct pgv      *pkbdq;
182         unsigned int    feature_req_word;
183         unsigned int    hdrlen;
184         unsigned char   reset_pending_on_curr_blk;
185         unsigned char   delete_blk_timer;
186         unsigned short  kactive_blk_num;
187         unsigned short  blk_sizeof_priv;
188
189         /* last_kactive_blk_num:
190          * trick to see if user-space has caught up
191          * in order to avoid refreshing timer when every single pkt arrives.
192          */
193         unsigned short  last_kactive_blk_num;
194
195         char            *pkblk_start;
196         char            *pkblk_end;
197         int             kblk_size;
198         unsigned int    knum_blocks;
199         uint64_t        knxt_seq_num;
200         char            *prev;
201         char            *nxt_offset;
202         struct sk_buff  *skb;
203
204         atomic_t        blk_fill_in_prog;
205
206         /* Default is set to 8ms */
207 #define DEFAULT_PRB_RETIRE_TOV  (8)
208
209         unsigned short  retire_blk_tov;
210         unsigned short  version;
211         unsigned long   tov_in_jiffies;
212
213         /* timer to retire an outstanding block */
214         struct timer_list retire_blk_timer;
215 };
216
217 #define PGV_FROM_VMALLOC 1
218 struct pgv {
219         char *buffer;
220 };
221
222 struct packet_ring_buffer {
223         struct pgv              *pg_vec;
224         unsigned int            head;
225         unsigned int            frames_per_block;
226         unsigned int            frame_size;
227         unsigned int            frame_max;
228
229         unsigned int            pg_vec_order;
230         unsigned int            pg_vec_pages;
231         unsigned int            pg_vec_len;
232
233         struct tpacket_kbdq_core        prb_bdqc;
234         atomic_t                pending;
235 };
236
237 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
238 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
239 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
240 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
241 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
242 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
243 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
244
245 struct packet_sock;
246 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
247
248 static void *packet_previous_frame(struct packet_sock *po,
249                 struct packet_ring_buffer *rb,
250                 int status);
251 static void packet_increment_head(struct packet_ring_buffer *buff);
252 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
253                         struct tpacket_block_desc *);
254 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
255                         struct packet_sock *);
256 static void prb_retire_current_block(struct tpacket_kbdq_core *,
257                 struct packet_sock *, unsigned int status);
258 static int prb_queue_frozen(struct tpacket_kbdq_core *);
259 static void prb_open_block(struct tpacket_kbdq_core *,
260                 struct tpacket_block_desc *);
261 static void prb_retire_rx_blk_timer_expired(unsigned long);
262 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
263 static void prb_init_blk_timer(struct packet_sock *,
264                 struct tpacket_kbdq_core *,
265                 void (*func) (unsigned long));
266 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
267 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
268                 struct tpacket3_hdr *);
269 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
270                 struct tpacket3_hdr *);
271 static void packet_flush_mclist(struct sock *sk);
272
273 struct packet_fanout;
274 struct packet_sock {
275         /* struct sock has to be the first member of packet_sock */
276         struct sock             sk;
277         struct packet_fanout    *fanout;
278         struct tpacket_stats    stats;
279         union  tpacket_stats_u  stats_u;
280         struct packet_ring_buffer       rx_ring;
281         struct packet_ring_buffer       tx_ring;
282         int                     copy_thresh;
283         spinlock_t              bind_lock;
284         struct mutex            pg_vec_lock;
285         unsigned int            running:1,      /* prot_hook is attached*/
286                                 auxdata:1,
287                                 origdev:1,
288                                 has_vnet_hdr:1;
289         int                     ifindex;        /* bound device         */
290         __be16                  num;
291         struct packet_mclist    *mclist;
292         atomic_t                mapped;
293         enum tpacket_versions   tp_version;
294         unsigned int            tp_hdrlen;
295         unsigned int            tp_reserve;
296         unsigned int            tp_loss:1;
297         unsigned int            tp_tstamp;
298         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
299 };
300
301 #define PACKET_FANOUT_MAX       256
302
303 struct packet_fanout {
304 #ifdef CONFIG_NET_NS
305         struct net              *net;
306 #endif
307         unsigned int            num_members;
308         u16                     id;
309         u8                      type;
310         u8                      defrag;
311         atomic_t                rr_cur;
312         struct list_head        list;
313         struct sock             *arr[PACKET_FANOUT_MAX];
314         spinlock_t              lock;
315         atomic_t                sk_ref;
316         struct packet_type      prot_hook ____cacheline_aligned_in_smp;
317 };
318
319 struct packet_skb_cb {
320         unsigned int origlen;
321         union {
322                 struct sockaddr_pkt pkt;
323                 struct sockaddr_ll ll;
324         } sa;
325 };
326
327 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
328
329 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
330 #define GET_PBLOCK_DESC(x, bid) \
331         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
332 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
333         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
334 #define GET_NEXT_PRB_BLK_NUM(x) \
335         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
336         ((x)->kactive_blk_num+1) : 0)
337
338 static struct packet_sock *pkt_sk(struct sock *sk)
339 {
340         return (struct packet_sock *)sk;
341 }
342
343 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
344 static void __fanout_link(struct sock *sk, struct packet_sock *po);
345
346 /* register_prot_hook must be invoked with the po->bind_lock held,
347  * or from a context in which asynchronous accesses to the packet
348  * socket is not possible (packet_create()).
349  */
350 static void register_prot_hook(struct sock *sk)
351 {
352         struct packet_sock *po = pkt_sk(sk);
353         if (!po->running) {
354                 if (po->fanout)
355                         __fanout_link(sk, po);
356                 else
357                         dev_add_pack(&po->prot_hook);
358                 sock_hold(sk);
359                 po->running = 1;
360         }
361 }
362
363 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
364  * held.   If the sync parameter is true, we will temporarily drop
365  * the po->bind_lock and do a synchronize_net to make sure no
366  * asynchronous packet processing paths still refer to the elements
367  * of po->prot_hook.  If the sync parameter is false, it is the
368  * callers responsibility to take care of this.
369  */
370 static void __unregister_prot_hook(struct sock *sk, bool sync)
371 {
372         struct packet_sock *po = pkt_sk(sk);
373
374         po->running = 0;
375         if (po->fanout)
376                 __fanout_unlink(sk, po);
377         else
378                 __dev_remove_pack(&po->prot_hook);
379         __sock_put(sk);
380
381         if (sync) {
382                 spin_unlock(&po->bind_lock);
383                 synchronize_net();
384                 spin_lock(&po->bind_lock);
385         }
386 }
387
388 static void unregister_prot_hook(struct sock *sk, bool sync)
389 {
390         struct packet_sock *po = pkt_sk(sk);
391
392         if (po->running)
393                 __unregister_prot_hook(sk, sync);
394 }
395
396 static inline __pure struct page *pgv_to_page(void *addr)
397 {
398         if (is_vmalloc_addr(addr))
399                 return vmalloc_to_page(addr);
400         return virt_to_page(addr);
401 }
402
403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
404 {
405         union {
406                 struct tpacket_hdr *h1;
407                 struct tpacket2_hdr *h2;
408                 void *raw;
409         } h;
410
411         h.raw = frame;
412         switch (po->tp_version) {
413         case TPACKET_V1:
414                 h.h1->tp_status = status;
415                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
416                 break;
417         case TPACKET_V2:
418                 h.h2->tp_status = status;
419                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
420                 break;
421         case TPACKET_V3:
422         default:
423                 WARN(1, "TPACKET version not supported.\n");
424                 BUG();
425         }
426
427         smp_wmb();
428 }
429
430 static int __packet_get_status(struct packet_sock *po, void *frame)
431 {
432         union {
433                 struct tpacket_hdr *h1;
434                 struct tpacket2_hdr *h2;
435                 void *raw;
436         } h;
437
438         smp_rmb();
439
440         h.raw = frame;
441         switch (po->tp_version) {
442         case TPACKET_V1:
443                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
444                 return h.h1->tp_status;
445         case TPACKET_V2:
446                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
447                 return h.h2->tp_status;
448         case TPACKET_V3:
449         default:
450                 WARN(1, "TPACKET version not supported.\n");
451                 BUG();
452                 return 0;
453         }
454 }
455
456 static void *packet_lookup_frame(struct packet_sock *po,
457                 struct packet_ring_buffer *rb,
458                 unsigned int position,
459                 int status)
460 {
461         unsigned int pg_vec_pos, frame_offset;
462         union {
463                 struct tpacket_hdr *h1;
464                 struct tpacket2_hdr *h2;
465                 void *raw;
466         } h;
467
468         pg_vec_pos = position / rb->frames_per_block;
469         frame_offset = position % rb->frames_per_block;
470
471         h.raw = rb->pg_vec[pg_vec_pos].buffer +
472                 (frame_offset * rb->frame_size);
473
474         if (status != __packet_get_status(po, h.raw))
475                 return NULL;
476
477         return h.raw;
478 }
479
480 static void *packet_current_frame(struct packet_sock *po,
481                 struct packet_ring_buffer *rb,
482                 int status)
483 {
484         return packet_lookup_frame(po, rb, rb->head, status);
485 }
486
487 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
488 {
489         del_timer_sync(&pkc->retire_blk_timer);
490 }
491
492 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
493                 int tx_ring,
494                 struct sk_buff_head *rb_queue)
495 {
496         struct tpacket_kbdq_core *pkc;
497
498         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
499
500         spin_lock(&rb_queue->lock);
501         pkc->delete_blk_timer = 1;
502         spin_unlock(&rb_queue->lock);
503
504         prb_del_retire_blk_timer(pkc);
505 }
506
507 static void prb_init_blk_timer(struct packet_sock *po,
508                 struct tpacket_kbdq_core *pkc,
509                 void (*func) (unsigned long))
510 {
511         init_timer(&pkc->retire_blk_timer);
512         pkc->retire_blk_timer.data = (long)po;
513         pkc->retire_blk_timer.function = func;
514         pkc->retire_blk_timer.expires = jiffies;
515 }
516
517 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
518 {
519         struct tpacket_kbdq_core *pkc;
520
521         if (tx_ring)
522                 BUG();
523
524         pkc = tx_ring ? &po->tx_ring.prb_bdqc : &po->rx_ring.prb_bdqc;
525         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
526 }
527
528 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
529                                 int blk_size_in_bytes)
530 {
531         struct net_device *dev;
532         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
533         struct ethtool_cmd ecmd;
534         int err;
535
536         rtnl_lock();
537         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
538         if (unlikely(!dev)) {
539                 rtnl_unlock();
540                 return DEFAULT_PRB_RETIRE_TOV;
541         }
542         err = __ethtool_get_settings(dev, &ecmd);
543         rtnl_unlock();
544         if (!err) {
545                 switch (ecmd.speed) {
546                 case SPEED_10000:
547                         msec = 1;
548                         div = 10000/1000;
549                         break;
550                 case SPEED_1000:
551                         msec = 1;
552                         div = 1000/1000;
553                         break;
554                 /*
555                  * If the link speed is so slow you don't really
556                  * need to worry about perf anyways
557                  */
558                 case SPEED_100:
559                 case SPEED_10:
560                 default:
561                         return DEFAULT_PRB_RETIRE_TOV;
562                 }
563         }
564
565         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
566
567         if (div)
568                 mbits /= div;
569
570         tmo = mbits * msec;
571
572         if (div)
573                 return tmo+1;
574         return tmo;
575 }
576
577 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
578                         union tpacket_req_u *req_u)
579 {
580         p1->feature_req_word = req_u->req3.tp_feature_req_word;
581 }
582
583 static void init_prb_bdqc(struct packet_sock *po,
584                         struct packet_ring_buffer *rb,
585                         struct pgv *pg_vec,
586                         union tpacket_req_u *req_u, int tx_ring)
587 {
588         struct tpacket_kbdq_core *p1 = &rb->prb_bdqc;
589         struct tpacket_block_desc *pbd;
590
591         memset(p1, 0x0, sizeof(*p1));
592
593         p1->knxt_seq_num = 1;
594         p1->pkbdq = pg_vec;
595         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
596         p1->pkblk_start = (char *)pg_vec[0].buffer;
597         p1->kblk_size = req_u->req3.tp_block_size;
598         p1->knum_blocks = req_u->req3.tp_block_nr;
599         p1->hdrlen = po->tp_hdrlen;
600         p1->version = po->tp_version;
601         p1->last_kactive_blk_num = 0;
602         po->stats_u.stats3.tp_freeze_q_cnt = 0;
603         if (req_u->req3.tp_retire_blk_tov)
604                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
605         else
606                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
607                                                 req_u->req3.tp_block_size);
608         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
609         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
610
611         prb_init_ft_ops(p1, req_u);
612         prb_setup_retire_blk_timer(po, tx_ring);
613         prb_open_block(p1, pbd);
614 }
615
616 /*  Do NOT update the last_blk_num first.
617  *  Assumes sk_buff_head lock is held.
618  */
619 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
620 {
621         mod_timer(&pkc->retire_blk_timer,
622                         jiffies + pkc->tov_in_jiffies);
623         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
624 }
625
626 /*
627  * Timer logic:
628  * 1) We refresh the timer only when we open a block.
629  *    By doing this we don't waste cycles refreshing the timer
630  *        on packet-by-packet basis.
631  *
632  * With a 1MB block-size, on a 1Gbps line, it will take
633  * i) ~8 ms to fill a block + ii) memcpy etc.
634  * In this cut we are not accounting for the memcpy time.
635  *
636  * So, if the user sets the 'tmo' to 10ms then the timer
637  * will never fire while the block is still getting filled
638  * (which is what we want). However, the user could choose
639  * to close a block early and that's fine.
640  *
641  * But when the timer does fire, we check whether or not to refresh it.
642  * Since the tmo granularity is in msecs, it is not too expensive
643  * to refresh the timer, lets say every '8' msecs.
644  * Either the user can set the 'tmo' or we can derive it based on
645  * a) line-speed and b) block-size.
646  * prb_calc_retire_blk_tmo() calculates the tmo.
647  *
648  */
649 static void prb_retire_rx_blk_timer_expired(unsigned long data)
650 {
651         struct packet_sock *po = (struct packet_sock *)data;
652         struct tpacket_kbdq_core *pkc = &po->rx_ring.prb_bdqc;
653         unsigned int frozen;
654         struct tpacket_block_desc *pbd;
655
656         spin_lock(&po->sk.sk_receive_queue.lock);
657
658         frozen = prb_queue_frozen(pkc);
659         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
660
661         if (unlikely(pkc->delete_blk_timer))
662                 goto out;
663
664         /* We only need to plug the race when the block is partially filled.
665          * tpacket_rcv:
666          *              lock(); increment BLOCK_NUM_PKTS; unlock()
667          *              copy_bits() is in progress ...
668          *              timer fires on other cpu:
669          *              we can't retire the current block because copy_bits
670          *              is in progress.
671          *
672          */
673         if (BLOCK_NUM_PKTS(pbd)) {
674                 while (atomic_read(&pkc->blk_fill_in_prog)) {
675                         /* Waiting for skb_copy_bits to finish... */
676                         cpu_relax();
677                 }
678         }
679
680         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
681                 if (!frozen) {
682                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
683                         if (!prb_dispatch_next_block(pkc, po))
684                                 goto refresh_timer;
685                         else
686                                 goto out;
687                 } else {
688                         /* Case 1. Queue was frozen because user-space was
689                          *         lagging behind.
690                          */
691                         if (prb_curr_blk_in_use(pkc, pbd)) {
692                                 /*
693                                  * Ok, user-space is still behind.
694                                  * So just refresh the timer.
695                                  */
696                                 goto refresh_timer;
697                         } else {
698                                /* Case 2. queue was frozen,user-space caught up,
699                                 * now the link went idle && the timer fired.
700                                 * We don't have a block to close.So we open this
701                                 * block and restart the timer.
702                                 * opening a block thaws the queue,restarts timer
703                                 * Thawing/timer-refresh is a side effect.
704                                 */
705                                 prb_open_block(pkc, pbd);
706                                 goto out;
707                         }
708                 }
709         }
710
711 refresh_timer:
712         _prb_refresh_rx_retire_blk_timer(pkc);
713
714 out:
715         spin_unlock(&po->sk.sk_receive_queue.lock);
716 }
717
718 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
719                 struct tpacket_block_desc *pbd1, __u32 status)
720 {
721         /* Flush everything minus the block header */
722
723 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
724         u8 *start, *end;
725
726         start = (u8 *)pbd1;
727
728         /* Skip the block header(we know header WILL fit in 4K) */
729         start += PAGE_SIZE;
730
731         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
732         for (; start < end; start += PAGE_SIZE)
733                 flush_dcache_page(pgv_to_page(start));
734
735         smp_wmb();
736 #endif
737
738         /* Now update the block status. */
739
740         BLOCK_STATUS(pbd1) = status;
741
742         /* Flush the block header */
743
744 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
745         start = (u8 *)pbd1;
746         flush_dcache_page(pgv_to_page(start));
747
748         smp_wmb();
749 #endif
750 }
751
752 /*
753  * Side effect:
754  *
755  * 1) flush the block
756  * 2) Increment active_blk_num
757  *
758  * Note:We DONT refresh the timer on purpose.
759  *      Because almost always the next block will be opened.
760  */
761 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
762                 struct tpacket_block_desc *pbd1,
763                 struct packet_sock *po, unsigned int stat)
764 {
765         __u32 status = TP_STATUS_USER | stat;
766
767         struct tpacket3_hdr *last_pkt;
768         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
769
770         if (po->stats.tp_drops)
771                 status |= TP_STATUS_LOSING;
772
773         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
774         last_pkt->tp_next_offset = 0;
775
776         /* Get the ts of the last pkt */
777         if (BLOCK_NUM_PKTS(pbd1)) {
778                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
779                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
780         } else {
781                 /* Ok, we tmo'd - so get the current time */
782                 struct timespec ts;
783                 getnstimeofday(&ts);
784                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
785                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
786         }
787
788         smp_wmb();
789
790         /* Flush the block */
791         prb_flush_block(pkc1, pbd1, status);
792
793         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
794 }
795
796 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
797 {
798         pkc->reset_pending_on_curr_blk = 0;
799 }
800
801 /*
802  * Side effect of opening a block:
803  *
804  * 1) prb_queue is thawed.
805  * 2) retire_blk_timer is refreshed.
806  *
807  */
808 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
809         struct tpacket_block_desc *pbd1)
810 {
811         struct timespec ts;
812         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
813
814         smp_rmb();
815
816         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd1))) {
817
818                 /* We could have just memset this but we will lose the
819                  * flexibility of making the priv area sticky
820                  */
821                 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
822                 BLOCK_NUM_PKTS(pbd1) = 0;
823                 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
824                 getnstimeofday(&ts);
825                 h1->ts_first_pkt.ts_sec = ts.tv_sec;
826                 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827                 pkc1->pkblk_start = (char *)pbd1;
828                 pkc1->nxt_offset = (char *)(pkc1->pkblk_start +
829                 BLK_PLUS_PRIV(pkc1->blk_sizeof_priv));
830                 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831                 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
832                 pbd1->version = pkc1->version;
833                 pkc1->prev = pkc1->nxt_offset;
834                 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
835                 prb_thaw_queue(pkc1);
836                 _prb_refresh_rx_retire_blk_timer(pkc1);
837
838                 smp_wmb();
839
840                 return;
841         }
842
843         WARN(1, "ERROR block:%p is NOT FREE status:%d kactive_blk_num:%d\n",
844                 pbd1, BLOCK_STATUS(pbd1), pkc1->kactive_blk_num);
845         dump_stack();
846         BUG();
847 }
848
849 /*
850  * Queue freeze logic:
851  * 1) Assume tp_block_nr = 8 blocks.
852  * 2) At time 't0', user opens Rx ring.
853  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854  * 4) user-space is either sleeping or processing block '0'.
855  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856  *    it will close block-7,loop around and try to fill block '0'.
857  *    call-flow:
858  *    __packet_lookup_frame_in_block
859  *      prb_retire_current_block()
860  *      prb_dispatch_next_block()
861  *        |->(BLOCK_STATUS == USER) evaluates to true
862  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
863  * 6) Now there are two cases:
864  *    6.1) Link goes idle right after the queue is frozen.
865  *         But remember, the last open_block() refreshed the timer.
866  *         When this timer expires,it will refresh itself so that we can
867  *         re-open block-0 in near future.
868  *    6.2) Link is busy and keeps on receiving packets. This is a simple
869  *         case and __packet_lookup_frame_in_block will check if block-0
870  *         is free and can now be re-used.
871  */
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873                                   struct packet_sock *po)
874 {
875         pkc->reset_pending_on_curr_blk = 1;
876         po->stats_u.stats3.tp_freeze_q_cnt++;
877 }
878
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880
881 /*
882  * If the next block is free then we will dispatch it
883  * and return a good offset.
884  * Else, we will freeze the queue.
885  * So, caller must check the return value.
886  */
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888                 struct packet_sock *po)
889 {
890         struct tpacket_block_desc *pbd;
891
892         smp_rmb();
893
894         /* 1. Get current block num */
895         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896
897         /* 2. If this block is currently in_use then freeze the queue */
898         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899                 prb_freeze_queue(pkc, po);
900                 return NULL;
901         }
902
903         /*
904          * 3.
905          * open this block and return the offset where the first packet
906          * needs to get stored.
907          */
908         prb_open_block(pkc, pbd);
909         return (void *)pkc->nxt_offset;
910 }
911
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913                 struct packet_sock *po, unsigned int status)
914 {
915         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916
917         /* retire/close the current block */
918         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919                 /*
920                  * Plug the case where copy_bits() is in progress on
921                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922                  * have space to copy the pkt in the current block and
923                  * called prb_retire_current_block()
924                  *
925                  * We don't need to worry about the TMO case because
926                  * the timer-handler already handled this case.
927                  */
928                 if (!(status & TP_STATUS_BLK_TMO)) {
929                         while (atomic_read(&pkc->blk_fill_in_prog)) {
930                                 /* Waiting for skb_copy_bits to finish... */
931                                 cpu_relax();
932                         }
933                 }
934                 prb_close_block(pkc, pbd, po, status);
935                 return;
936         }
937
938         WARN(1, "ERROR-pbd[%d]:%p\n", pkc->kactive_blk_num, pbd);
939         dump_stack();
940         BUG();
941 }
942
943 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
944                                       struct tpacket_block_desc *pbd)
945 {
946         return TP_STATUS_USER & BLOCK_STATUS(pbd);
947 }
948
949 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
950 {
951         return pkc->reset_pending_on_curr_blk;
952 }
953
954 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
955 {
956         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
957         atomic_dec(&pkc->blk_fill_in_prog);
958 }
959
960 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
961                         struct tpacket3_hdr *ppd)
962 {
963         ppd->hv1.tp_rxhash = skb_get_rxhash(pkc->skb);
964 }
965
966 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
967                         struct tpacket3_hdr *ppd)
968 {
969         ppd->hv1.tp_rxhash = 0;
970 }
971
972 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
973                         struct tpacket3_hdr *ppd)
974 {
975         if (vlan_tx_tag_present(pkc->skb)) {
976                 ppd->hv1.tp_vlan_tci = vlan_tx_tag_get(pkc->skb);
977                 ppd->tp_status = TP_STATUS_VLAN_VALID;
978         } else {
979                 ppd->hv1.tp_vlan_tci = ppd->tp_status = 0;
980         }
981 }
982
983 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
984                         struct tpacket3_hdr *ppd)
985 {
986         prb_fill_vlan_info(pkc, ppd);
987
988         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989                 prb_fill_rxhash(pkc, ppd);
990         else
991                 prb_clear_rxhash(pkc, ppd);
992 }
993
994 static void prb_fill_curr_block(char *curr,
995                                 struct tpacket_kbdq_core *pkc,
996                                 struct tpacket_block_desc *pbd,
997                                 unsigned int len)
998 {
999         struct tpacket3_hdr *ppd;
1000
1001         ppd  = (struct tpacket3_hdr *)curr;
1002         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1003         pkc->prev = curr;
1004         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006         BLOCK_NUM_PKTS(pbd) += 1;
1007         atomic_inc(&pkc->blk_fill_in_prog);
1008         prb_run_all_ft_ops(pkc, ppd);
1009 }
1010
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013                                             struct sk_buff *skb,
1014                                                 int status,
1015                                             unsigned int len
1016                                             )
1017 {
1018         struct tpacket_kbdq_core *pkc;
1019         struct tpacket_block_desc *pbd;
1020         char *curr, *end;
1021
1022         pkc = GET_PBDQC_FROM_RB(((struct packet_ring_buffer *)&po->rx_ring));
1023         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
1025         /* Queue is frozen when user space is lagging behind */
1026         if (prb_queue_frozen(pkc)) {
1027                 /*
1028                  * Check if that last block which caused the queue to freeze,
1029                  * is still in_use by user-space.
1030                  */
1031                 if (prb_curr_blk_in_use(pkc, pbd)) {
1032                         /* Can't record this packet */
1033                         return NULL;
1034                 } else {
1035                         /*
1036                          * Ok, the block was released by user-space.
1037                          * Now let's open that block.
1038                          * opening a block also thaws the queue.
1039                          * Thawing is a side effect.
1040                          */
1041                         prb_open_block(pkc, pbd);
1042                 }
1043         }
1044
1045         smp_mb();
1046         curr = pkc->nxt_offset;
1047         pkc->skb = skb;
1048         end = (char *) ((char *)pbd + pkc->kblk_size);
1049
1050         /* first try the current block */
1051         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052                 prb_fill_curr_block(curr, pkc, pbd, len);
1053                 return (void *)curr;
1054         }
1055
1056         /* Ok, close the current block */
1057         prb_retire_current_block(pkc, po, 0);
1058
1059         /* Now, try to dispatch the next block */
1060         curr = (char *)prb_dispatch_next_block(pkc, po);
1061         if (curr) {
1062                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063                 prb_fill_curr_block(curr, pkc, pbd, len);
1064                 return (void *)curr;
1065         }
1066
1067         /*
1068          * No free blocks are available.user_space hasn't caught up yet.
1069          * Queue was just frozen and now this packet will get dropped.
1070          */
1071         return NULL;
1072 }
1073
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075                                             struct sk_buff *skb,
1076                                             int status, unsigned int len)
1077 {
1078         char *curr = NULL;
1079         switch (po->tp_version) {
1080         case TPACKET_V1:
1081         case TPACKET_V2:
1082                 curr = packet_lookup_frame(po, &po->rx_ring,
1083                                         po->rx_ring.head, status);
1084                 return curr;
1085         case TPACKET_V3:
1086                 return __packet_lookup_frame_in_block(po, skb, status, len);
1087         default:
1088                 WARN(1, "TPACKET version not supported\n");
1089                 BUG();
1090                 return 0;
1091         }
1092 }
1093
1094 static void *prb_lookup_block(struct packet_sock *po,
1095                                      struct packet_ring_buffer *rb,
1096                                      unsigned int previous,
1097                                      int status)
1098 {
1099         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1100         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, previous);
1101
1102         if (status != BLOCK_STATUS(pbd))
1103                 return NULL;
1104         return pbd;
1105 }
1106
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1108 {
1109         unsigned int prev;
1110         if (rb->prb_bdqc.kactive_blk_num)
1111                 prev = rb->prb_bdqc.kactive_blk_num-1;
1112         else
1113                 prev = rb->prb_bdqc.knum_blocks-1;
1114         return prev;
1115 }
1116
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119                                          struct packet_ring_buffer *rb,
1120                                          int status)
1121 {
1122         unsigned int previous = prb_previous_blk_num(rb);
1123         return prb_lookup_block(po, rb, previous, status);
1124 }
1125
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127                                              struct packet_ring_buffer *rb,
1128                                              int status)
1129 {
1130         if (po->tp_version <= TPACKET_V2)
1131                 return packet_previous_frame(po, rb, status);
1132
1133         return __prb_previous_block(po, rb, status);
1134 }
1135
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137                                             struct packet_ring_buffer *rb)
1138 {
1139         switch (po->tp_version) {
1140         case TPACKET_V1:
1141         case TPACKET_V2:
1142                 return packet_increment_head(rb);
1143         case TPACKET_V3:
1144         default:
1145                 WARN(1, "TPACKET version not supported.\n");
1146                 BUG();
1147                 return;
1148         }
1149 }
1150
1151 static void *packet_previous_frame(struct packet_sock *po,
1152                 struct packet_ring_buffer *rb,
1153                 int status)
1154 {
1155         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156         return packet_lookup_frame(po, rb, previous, status);
1157 }
1158
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1160 {
1161         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162 }
1163
1164 static void packet_sock_destruct(struct sock *sk)
1165 {
1166         skb_queue_purge(&sk->sk_error_queue);
1167
1168         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1169         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1170
1171         if (!sock_flag(sk, SOCK_DEAD)) {
1172                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1173                 return;
1174         }
1175
1176         sk_refcnt_debug_dec(sk);
1177 }
1178
1179 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1180 {
1181         int x = atomic_read(&f->rr_cur) + 1;
1182
1183         if (x >= num)
1184                 x = 0;
1185
1186         return x;
1187 }
1188
1189 static struct sock *fanout_demux_hash(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1190 {
1191         u32 idx, hash = skb->rxhash;
1192
1193         idx = ((u64)hash * num) >> 32;
1194
1195         return f->arr[idx];
1196 }
1197
1198 static struct sock *fanout_demux_lb(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1199 {
1200         int cur, old;
1201
1202         cur = atomic_read(&f->rr_cur);
1203         while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1204                                      fanout_rr_next(f, num))) != cur)
1205                 cur = old;
1206         return f->arr[cur];
1207 }
1208
1209 static struct sock *fanout_demux_cpu(struct packet_fanout *f, struct sk_buff *skb, unsigned int num)
1210 {
1211         unsigned int cpu = smp_processor_id();
1212
1213         return f->arr[cpu % num];
1214 }
1215
1216 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1217                              struct packet_type *pt, struct net_device *orig_dev)
1218 {
1219         struct packet_fanout *f = pt->af_packet_priv;
1220         unsigned int num = f->num_members;
1221         struct packet_sock *po;
1222         struct sock *sk;
1223
1224         if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1225             !num) {
1226                 kfree_skb(skb);
1227                 return 0;
1228         }
1229
1230         switch (f->type) {
1231         case PACKET_FANOUT_HASH:
1232         default:
1233                 if (f->defrag) {
1234                         skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1235                         if (!skb)
1236                                 return 0;
1237                 }
1238                 skb_get_rxhash(skb);
1239                 sk = fanout_demux_hash(f, skb, num);
1240                 break;
1241         case PACKET_FANOUT_LB:
1242                 sk = fanout_demux_lb(f, skb, num);
1243                 break;
1244         case PACKET_FANOUT_CPU:
1245                 sk = fanout_demux_cpu(f, skb, num);
1246                 break;
1247         }
1248
1249         po = pkt_sk(sk);
1250
1251         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1252 }
1253
1254 static DEFINE_MUTEX(fanout_mutex);
1255 static LIST_HEAD(fanout_list);
1256
1257 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1258 {
1259         struct packet_fanout *f = po->fanout;
1260
1261         spin_lock(&f->lock);
1262         f->arr[f->num_members] = sk;
1263         smp_wmb();
1264         f->num_members++;
1265         spin_unlock(&f->lock);
1266 }
1267
1268 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1269 {
1270         struct packet_fanout *f = po->fanout;
1271         int i;
1272
1273         spin_lock(&f->lock);
1274         for (i = 0; i < f->num_members; i++) {
1275                 if (f->arr[i] == sk)
1276                         break;
1277         }
1278         BUG_ON(i >= f->num_members);
1279         f->arr[i] = f->arr[f->num_members - 1];
1280         f->num_members--;
1281         spin_unlock(&f->lock);
1282 }
1283
1284 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1285 {
1286         struct packet_sock *po = pkt_sk(sk);
1287         struct packet_fanout *f, *match;
1288         u8 type = type_flags & 0xff;
1289         u8 defrag = (type_flags & PACKET_FANOUT_FLAG_DEFRAG) ? 1 : 0;
1290         int err;
1291
1292         switch (type) {
1293         case PACKET_FANOUT_HASH:
1294         case PACKET_FANOUT_LB:
1295         case PACKET_FANOUT_CPU:
1296                 break;
1297         default:
1298                 return -EINVAL;
1299         }
1300
1301         if (!po->running)
1302                 return -EINVAL;
1303
1304         if (po->fanout)
1305                 return -EALREADY;
1306
1307         mutex_lock(&fanout_mutex);
1308         match = NULL;
1309         list_for_each_entry(f, &fanout_list, list) {
1310                 if (f->id == id &&
1311                     read_pnet(&f->net) == sock_net(sk)) {
1312                         match = f;
1313                         break;
1314                 }
1315         }
1316         err = -EINVAL;
1317         if (match && match->defrag != defrag)
1318                 goto out;
1319         if (!match) {
1320                 err = -ENOMEM;
1321                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1322                 if (!match)
1323                         goto out;
1324                 write_pnet(&match->net, sock_net(sk));
1325                 match->id = id;
1326                 match->type = type;
1327                 match->defrag = defrag;
1328                 atomic_set(&match->rr_cur, 0);
1329                 INIT_LIST_HEAD(&match->list);
1330                 spin_lock_init(&match->lock);
1331                 atomic_set(&match->sk_ref, 0);
1332                 match->prot_hook.type = po->prot_hook.type;
1333                 match->prot_hook.dev = po->prot_hook.dev;
1334                 match->prot_hook.func = packet_rcv_fanout;
1335                 match->prot_hook.af_packet_priv = match;
1336                 dev_add_pack(&match->prot_hook);
1337                 list_add(&match->list, &fanout_list);
1338         }
1339         err = -EINVAL;
1340         if (match->type == type &&
1341             match->prot_hook.type == po->prot_hook.type &&
1342             match->prot_hook.dev == po->prot_hook.dev) {
1343                 err = -ENOSPC;
1344                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1345                         __dev_remove_pack(&po->prot_hook);
1346                         po->fanout = match;
1347                         atomic_inc(&match->sk_ref);
1348                         __fanout_link(sk, po);
1349                         err = 0;
1350                 }
1351         }
1352 out:
1353         mutex_unlock(&fanout_mutex);
1354         return err;
1355 }
1356
1357 static void fanout_release(struct sock *sk)
1358 {
1359         struct packet_sock *po = pkt_sk(sk);
1360         struct packet_fanout *f;
1361
1362         f = po->fanout;
1363         if (!f)
1364                 return;
1365
1366         po->fanout = NULL;
1367
1368         mutex_lock(&fanout_mutex);
1369         if (atomic_dec_and_test(&f->sk_ref)) {
1370                 list_del(&f->list);
1371                 dev_remove_pack(&f->prot_hook);
1372                 kfree(f);
1373         }
1374         mutex_unlock(&fanout_mutex);
1375 }
1376
1377 static const struct proto_ops packet_ops;
1378
1379 static const struct proto_ops packet_ops_spkt;
1380
1381 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1382                            struct packet_type *pt, struct net_device *orig_dev)
1383 {
1384         struct sock *sk;
1385         struct sockaddr_pkt *spkt;
1386
1387         /*
1388          *      When we registered the protocol we saved the socket in the data
1389          *      field for just this event.
1390          */
1391
1392         sk = pt->af_packet_priv;
1393
1394         /*
1395          *      Yank back the headers [hope the device set this
1396          *      right or kerboom...]
1397          *
1398          *      Incoming packets have ll header pulled,
1399          *      push it back.
1400          *
1401          *      For outgoing ones skb->data == skb_mac_header(skb)
1402          *      so that this procedure is noop.
1403          */
1404
1405         if (skb->pkt_type == PACKET_LOOPBACK)
1406                 goto out;
1407
1408         if (!net_eq(dev_net(dev), sock_net(sk)))
1409                 goto out;
1410
1411         skb = skb_share_check(skb, GFP_ATOMIC);
1412         if (skb == NULL)
1413                 goto oom;
1414
1415         /* drop any routing info */
1416         skb_dst_drop(skb);
1417
1418         /* drop conntrack reference */
1419         nf_reset(skb);
1420
1421         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1422
1423         skb_push(skb, skb->data - skb_mac_header(skb));
1424
1425         /*
1426          *      The SOCK_PACKET socket receives _all_ frames.
1427          */
1428
1429         spkt->spkt_family = dev->type;
1430         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1431         spkt->spkt_protocol = skb->protocol;
1432
1433         /*
1434          *      Charge the memory to the socket. This is done specifically
1435          *      to prevent sockets using all the memory up.
1436          */
1437
1438         if (sock_queue_rcv_skb(sk, skb) == 0)
1439                 return 0;
1440
1441 out:
1442         kfree_skb(skb);
1443 oom:
1444         return 0;
1445 }
1446
1447
1448 /*
1449  *      Output a raw packet to a device layer. This bypasses all the other
1450  *      protocol layers and you must therefore supply it with a complete frame
1451  */
1452
1453 static int packet_sendmsg_spkt(struct kiocb *iocb, struct socket *sock,
1454                                struct msghdr *msg, size_t len)
1455 {
1456         struct sock *sk = sock->sk;
1457         struct sockaddr_pkt *saddr = (struct sockaddr_pkt *)msg->msg_name;
1458         struct sk_buff *skb = NULL;
1459         struct net_device *dev;
1460         __be16 proto = 0;
1461         int err;
1462         int extra_len = 0;
1463
1464         /*
1465          *      Get and verify the address.
1466          */
1467
1468         if (saddr) {
1469                 if (msg->msg_namelen < sizeof(struct sockaddr))
1470                         return -EINVAL;
1471                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1472                         proto = saddr->spkt_protocol;
1473         } else
1474                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1475
1476         /*
1477          *      Find the device first to size check it
1478          */
1479
1480         saddr->spkt_device[13] = 0;
1481 retry:
1482         rcu_read_lock();
1483         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1484         err = -ENODEV;
1485         if (dev == NULL)
1486                 goto out_unlock;
1487
1488         err = -ENETDOWN;
1489         if (!(dev->flags & IFF_UP))
1490                 goto out_unlock;
1491
1492         /*
1493          * You may not queue a frame bigger than the mtu. This is the lowest level
1494          * raw protocol and you must do your own fragmentation at this level.
1495          */
1496
1497         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1498                 if (!netif_supports_nofcs(dev)) {
1499                         err = -EPROTONOSUPPORT;
1500                         goto out_unlock;
1501                 }
1502                 extra_len = 4; /* We're doing our own CRC */
1503         }
1504
1505         err = -EMSGSIZE;
1506         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1507                 goto out_unlock;
1508
1509         if (!skb) {
1510                 size_t reserved = LL_RESERVED_SPACE(dev);
1511                 int tlen = dev->needed_tailroom;
1512                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1513
1514                 rcu_read_unlock();
1515                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1516                 if (skb == NULL)
1517                         return -ENOBUFS;
1518                 /* FIXME: Save some space for broken drivers that write a hard
1519                  * header at transmission time by themselves. PPP is the notable
1520                  * one here. This should really be fixed at the driver level.
1521                  */
1522                 skb_reserve(skb, reserved);
1523                 skb_reset_network_header(skb);
1524
1525                 /* Try to align data part correctly */
1526                 if (hhlen) {
1527                         skb->data -= hhlen;
1528                         skb->tail -= hhlen;
1529                         if (len < hhlen)
1530                                 skb_reset_network_header(skb);
1531                 }
1532                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1533                 if (err)
1534                         goto out_free;
1535                 goto retry;
1536         }
1537
1538         if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1539                 /* Earlier code assumed this would be a VLAN pkt,
1540                  * double-check this now that we have the actual
1541                  * packet in hand.
1542                  */
1543                 struct ethhdr *ehdr;
1544                 skb_reset_mac_header(skb);
1545                 ehdr = eth_hdr(skb);
1546                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1547                         err = -EMSGSIZE;
1548                         goto out_unlock;
1549                 }
1550         }
1551
1552         skb->protocol = proto;
1553         skb->dev = dev;
1554         skb->priority = sk->sk_priority;
1555         skb->mark = sk->sk_mark;
1556         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1557         if (err < 0)
1558                 goto out_unlock;
1559
1560         if (unlikely(extra_len == 4))
1561                 skb->no_fcs = 1;
1562
1563         dev_queue_xmit(skb);
1564         rcu_read_unlock();
1565         return len;
1566
1567 out_unlock:
1568         rcu_read_unlock();
1569 out_free:
1570         kfree_skb(skb);
1571         return err;
1572 }
1573
1574 static unsigned int run_filter(const struct sk_buff *skb,
1575                                       const struct sock *sk,
1576                                       unsigned int res)
1577 {
1578         struct sk_filter *filter;
1579
1580         rcu_read_lock();
1581         filter = rcu_dereference(sk->sk_filter);
1582         if (filter != NULL)
1583                 res = SK_RUN_FILTER(filter, skb);
1584         rcu_read_unlock();
1585
1586         return res;
1587 }
1588
1589 /*
1590  * This function makes lazy skb cloning in hope that most of packets
1591  * are discarded by BPF.
1592  *
1593  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1594  * and skb->cb are mangled. It works because (and until) packets
1595  * falling here are owned by current CPU. Output packets are cloned
1596  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1597  * sequencially, so that if we return skb to original state on exit,
1598  * we will not harm anyone.
1599  */
1600
1601 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1602                       struct packet_type *pt, struct net_device *orig_dev)
1603 {
1604         struct sock *sk;
1605         struct sockaddr_ll *sll;
1606         struct packet_sock *po;
1607         u8 *skb_head = skb->data;
1608         int skb_len = skb->len;
1609         unsigned int snaplen, res;
1610
1611         if (skb->pkt_type == PACKET_LOOPBACK)
1612                 goto drop;
1613
1614         sk = pt->af_packet_priv;
1615         po = pkt_sk(sk);
1616
1617         if (!net_eq(dev_net(dev), sock_net(sk)))
1618                 goto drop;
1619
1620         skb->dev = dev;
1621
1622         if (dev->header_ops) {
1623                 /* The device has an explicit notion of ll header,
1624                  * exported to higher levels.
1625                  *
1626                  * Otherwise, the device hides details of its frame
1627                  * structure, so that corresponding packet head is
1628                  * never delivered to user.
1629                  */
1630                 if (sk->sk_type != SOCK_DGRAM)
1631                         skb_push(skb, skb->data - skb_mac_header(skb));
1632                 else if (skb->pkt_type == PACKET_OUTGOING) {
1633                         /* Special case: outgoing packets have ll header at head */
1634                         skb_pull(skb, skb_network_offset(skb));
1635                 }
1636         }
1637
1638         snaplen = skb->len;
1639
1640         res = run_filter(skb, sk, snaplen);
1641         if (!res)
1642                 goto drop_n_restore;
1643         if (snaplen > res)
1644                 snaplen = res;
1645
1646         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1647                 goto drop_n_acct;
1648
1649         if (skb_shared(skb)) {
1650                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1651                 if (nskb == NULL)
1652                         goto drop_n_acct;
1653
1654                 if (skb_head != skb->data) {
1655                         skb->data = skb_head;
1656                         skb->len = skb_len;
1657                 }
1658                 kfree_skb(skb);
1659                 skb = nskb;
1660         }
1661
1662         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1663                      sizeof(skb->cb));
1664
1665         sll = &PACKET_SKB_CB(skb)->sa.ll;
1666         sll->sll_family = AF_PACKET;
1667         sll->sll_hatype = dev->type;
1668         sll->sll_protocol = skb->protocol;
1669         sll->sll_pkttype = skb->pkt_type;
1670         if (unlikely(po->origdev))
1671                 sll->sll_ifindex = orig_dev->ifindex;
1672         else
1673                 sll->sll_ifindex = dev->ifindex;
1674
1675         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1676
1677         PACKET_SKB_CB(skb)->origlen = skb->len;
1678
1679         if (pskb_trim(skb, snaplen))
1680                 goto drop_n_acct;
1681
1682         skb_set_owner_r(skb, sk);
1683         skb->dev = NULL;
1684         skb_dst_drop(skb);
1685
1686         /* drop conntrack reference */
1687         nf_reset(skb);
1688
1689         spin_lock(&sk->sk_receive_queue.lock);
1690         po->stats.tp_packets++;
1691         skb->dropcount = atomic_read(&sk->sk_drops);
1692         __skb_queue_tail(&sk->sk_receive_queue, skb);
1693         spin_unlock(&sk->sk_receive_queue.lock);
1694         sk->sk_data_ready(sk, skb->len);
1695         return 0;
1696
1697 drop_n_acct:
1698         spin_lock(&sk->sk_receive_queue.lock);
1699         po->stats.tp_drops++;
1700         atomic_inc(&sk->sk_drops);
1701         spin_unlock(&sk->sk_receive_queue.lock);
1702
1703 drop_n_restore:
1704         if (skb_head != skb->data && skb_shared(skb)) {
1705                 skb->data = skb_head;
1706                 skb->len = skb_len;
1707         }
1708 drop:
1709         consume_skb(skb);
1710         return 0;
1711 }
1712
1713 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1714                        struct packet_type *pt, struct net_device *orig_dev)
1715 {
1716         struct sock *sk;
1717         struct packet_sock *po;
1718         struct sockaddr_ll *sll;
1719         union {
1720                 struct tpacket_hdr *h1;
1721                 struct tpacket2_hdr *h2;
1722                 struct tpacket3_hdr *h3;
1723                 void *raw;
1724         } h;
1725         u8 *skb_head = skb->data;
1726         int skb_len = skb->len;
1727         unsigned int snaplen, res;
1728         unsigned long status = TP_STATUS_USER;
1729         unsigned short macoff, netoff, hdrlen;
1730         struct sk_buff *copy_skb = NULL;
1731         struct timeval tv;
1732         struct timespec ts;
1733         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1734
1735         if (skb->pkt_type == PACKET_LOOPBACK)
1736                 goto drop;
1737
1738         sk = pt->af_packet_priv;
1739         po = pkt_sk(sk);
1740
1741         if (!net_eq(dev_net(dev), sock_net(sk)))
1742                 goto drop;
1743
1744         if (dev->header_ops) {
1745                 if (sk->sk_type != SOCK_DGRAM)
1746                         skb_push(skb, skb->data - skb_mac_header(skb));
1747                 else if (skb->pkt_type == PACKET_OUTGOING) {
1748                         /* Special case: outgoing packets have ll header at head */
1749                         skb_pull(skb, skb_network_offset(skb));
1750                 }
1751         }
1752
1753         if (skb->ip_summed == CHECKSUM_PARTIAL)
1754                 status |= TP_STATUS_CSUMNOTREADY;
1755
1756         snaplen = skb->len;
1757
1758         res = run_filter(skb, sk, snaplen);
1759         if (!res)
1760                 goto drop_n_restore;
1761         if (snaplen > res)
1762                 snaplen = res;
1763
1764         if (sk->sk_type == SOCK_DGRAM) {
1765                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1766                                   po->tp_reserve;
1767         } else {
1768                 unsigned maclen = skb_network_offset(skb);
1769                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1770                                        (maclen < 16 ? 16 : maclen)) +
1771                         po->tp_reserve;
1772                 macoff = netoff - maclen;
1773         }
1774         if (po->tp_version <= TPACKET_V2) {
1775                 if (macoff + snaplen > po->rx_ring.frame_size) {
1776                         if (po->copy_thresh &&
1777                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1778                                 if (skb_shared(skb)) {
1779                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1780                                 } else {
1781                                         copy_skb = skb_get(skb);
1782                                         skb_head = skb->data;
1783                                 }
1784                                 if (copy_skb)
1785                                         skb_set_owner_r(copy_skb, sk);
1786                         }
1787                         snaplen = po->rx_ring.frame_size - macoff;
1788                         if ((int)snaplen < 0)
1789                                 snaplen = 0;
1790                 }
1791         }
1792         spin_lock(&sk->sk_receive_queue.lock);
1793         h.raw = packet_current_rx_frame(po, skb,
1794                                         TP_STATUS_KERNEL, (macoff+snaplen));
1795         if (!h.raw)
1796                 goto ring_is_full;
1797         if (po->tp_version <= TPACKET_V2) {
1798                 packet_increment_rx_head(po, &po->rx_ring);
1799         /*
1800          * LOSING will be reported till you read the stats,
1801          * because it's COR - Clear On Read.
1802          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1803          * at packet level.
1804          */
1805                 if (po->stats.tp_drops)
1806                         status |= TP_STATUS_LOSING;
1807         }
1808         po->stats.tp_packets++;
1809         if (copy_skb) {
1810                 status |= TP_STATUS_COPY;
1811                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1812         }
1813         spin_unlock(&sk->sk_receive_queue.lock);
1814
1815         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1816
1817         switch (po->tp_version) {
1818         case TPACKET_V1:
1819                 h.h1->tp_len = skb->len;
1820                 h.h1->tp_snaplen = snaplen;
1821                 h.h1->tp_mac = macoff;
1822                 h.h1->tp_net = netoff;
1823                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1824                                 && shhwtstamps->syststamp.tv64)
1825                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1826                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1827                                 && shhwtstamps->hwtstamp.tv64)
1828                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1829                 else if (skb->tstamp.tv64)
1830                         tv = ktime_to_timeval(skb->tstamp);
1831                 else
1832                         do_gettimeofday(&tv);
1833                 h.h1->tp_sec = tv.tv_sec;
1834                 h.h1->tp_usec = tv.tv_usec;
1835                 hdrlen = sizeof(*h.h1);
1836                 break;
1837         case TPACKET_V2:
1838                 h.h2->tp_len = skb->len;
1839                 h.h2->tp_snaplen = snaplen;
1840                 h.h2->tp_mac = macoff;
1841                 h.h2->tp_net = netoff;
1842                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1843                                 && shhwtstamps->syststamp.tv64)
1844                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1845                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1846                                 && shhwtstamps->hwtstamp.tv64)
1847                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1848                 else if (skb->tstamp.tv64)
1849                         ts = ktime_to_timespec(skb->tstamp);
1850                 else
1851                         getnstimeofday(&ts);
1852                 h.h2->tp_sec = ts.tv_sec;
1853                 h.h2->tp_nsec = ts.tv_nsec;
1854                 if (vlan_tx_tag_present(skb)) {
1855                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1856                         status |= TP_STATUS_VLAN_VALID;
1857                 } else {
1858                         h.h2->tp_vlan_tci = 0;
1859                 }
1860                 h.h2->tp_padding = 0;
1861                 hdrlen = sizeof(*h.h2);
1862                 break;
1863         case TPACKET_V3:
1864                 /* tp_nxt_offset,vlan are already populated above.
1865                  * So DONT clear those fields here
1866                  */
1867                 h.h3->tp_status |= status;
1868                 h.h3->tp_len = skb->len;
1869                 h.h3->tp_snaplen = snaplen;
1870                 h.h3->tp_mac = macoff;
1871                 h.h3->tp_net = netoff;
1872                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1873                                 && shhwtstamps->syststamp.tv64)
1874                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1875                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1876                                 && shhwtstamps->hwtstamp.tv64)
1877                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1878                 else if (skb->tstamp.tv64)
1879                         ts = ktime_to_timespec(skb->tstamp);
1880                 else
1881                         getnstimeofday(&ts);
1882                 h.h3->tp_sec  = ts.tv_sec;
1883                 h.h3->tp_nsec = ts.tv_nsec;
1884                 hdrlen = sizeof(*h.h3);
1885                 break;
1886         default:
1887                 BUG();
1888         }
1889
1890         sll = h.raw + TPACKET_ALIGN(hdrlen);
1891         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1892         sll->sll_family = AF_PACKET;
1893         sll->sll_hatype = dev->type;
1894         sll->sll_protocol = skb->protocol;
1895         sll->sll_pkttype = skb->pkt_type;
1896         if (unlikely(po->origdev))
1897                 sll->sll_ifindex = orig_dev->ifindex;
1898         else
1899                 sll->sll_ifindex = dev->ifindex;
1900
1901         smp_mb();
1902 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1903         {
1904                 u8 *start, *end;
1905
1906                 if (po->tp_version <= TPACKET_V2) {
1907                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1908                                 + macoff + snaplen);
1909                         for (start = h.raw; start < end; start += PAGE_SIZE)
1910                                 flush_dcache_page(pgv_to_page(start));
1911                 }
1912                 smp_wmb();
1913         }
1914 #endif
1915         if (po->tp_version <= TPACKET_V2)
1916                 __packet_set_status(po, h.raw, status);
1917         else
1918                 prb_clear_blk_fill_status(&po->rx_ring);
1919
1920         sk->sk_data_ready(sk, 0);
1921
1922 drop_n_restore:
1923         if (skb_head != skb->data && skb_shared(skb)) {
1924                 skb->data = skb_head;
1925                 skb->len = skb_len;
1926         }
1927 drop:
1928         kfree_skb(skb);
1929         return 0;
1930
1931 ring_is_full:
1932         po->stats.tp_drops++;
1933         spin_unlock(&sk->sk_receive_queue.lock);
1934
1935         sk->sk_data_ready(sk, 0);
1936         kfree_skb(copy_skb);
1937         goto drop_n_restore;
1938 }
1939
1940 static void tpacket_destruct_skb(struct sk_buff *skb)
1941 {
1942         struct packet_sock *po = pkt_sk(skb->sk);
1943         void *ph;
1944
1945         if (likely(po->tx_ring.pg_vec)) {
1946                 ph = skb_shinfo(skb)->destructor_arg;
1947                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
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, int hlen)
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, hlen);
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 int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2055 {
2056         struct sk_buff *skb;
2057         struct net_device *dev;
2058         __be16 proto;
2059         bool need_rls_dev = false;
2060         int err, reserve = 0;
2061         void *ph;
2062         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2063         int tp_len, size_max;
2064         unsigned char *addr;
2065         int len_sum = 0;
2066         int status = 0;
2067         int hlen, tlen;
2068
2069         mutex_lock(&po->pg_vec_lock);
2070
2071         err = -EBUSY;
2072         if (saddr == NULL) {
2073                 dev = po->prot_hook.dev;
2074                 proto   = po->num;
2075                 addr    = NULL;
2076         } else {
2077                 err = -EINVAL;
2078                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2079                         goto out;
2080                 if (msg->msg_namelen < (saddr->sll_halen
2081                                         + offsetof(struct sockaddr_ll,
2082                                                 sll_addr)))
2083                         goto out;
2084                 proto   = saddr->sll_protocol;
2085                 addr    = saddr->sll_addr;
2086                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2087                 need_rls_dev = true;
2088         }
2089
2090         err = -ENXIO;
2091         if (unlikely(dev == NULL))
2092                 goto out;
2093
2094         reserve = dev->hard_header_len;
2095
2096         err = -ENETDOWN;
2097         if (unlikely(!(dev->flags & IFF_UP)))
2098                 goto out_put;
2099
2100         size_max = po->tx_ring.frame_size
2101                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2102
2103         if (size_max > dev->mtu + reserve)
2104                 size_max = dev->mtu + reserve;
2105
2106         do {
2107                 ph = packet_current_frame(po, &po->tx_ring,
2108                                 TP_STATUS_SEND_REQUEST);
2109
2110                 if (unlikely(ph == NULL)) {
2111                         schedule();
2112                         continue;
2113                 }
2114
2115                 status = TP_STATUS_SEND_REQUEST;
2116                 hlen = LL_RESERVED_SPACE(dev);
2117                 tlen = dev->needed_tailroom;
2118                 skb = sock_alloc_send_skb(&po->sk,
2119                                 hlen + tlen + sizeof(struct sockaddr_ll),
2120                                 0, &err);
2121
2122                 if (unlikely(skb == NULL))
2123                         goto out_status;
2124
2125                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2126                                 addr, hlen);
2127
2128                 if (unlikely(tp_len < 0)) {
2129                         if (po->tp_loss) {
2130                                 __packet_set_status(po, ph,
2131                                                 TP_STATUS_AVAILABLE);
2132                                 packet_increment_head(&po->tx_ring);
2133                                 kfree_skb(skb);
2134                                 continue;
2135                         } else {
2136                                 status = TP_STATUS_WRONG_FORMAT;
2137                                 err = tp_len;
2138                                 goto out_status;
2139                         }
2140                 }
2141
2142                 skb->destructor = tpacket_destruct_skb;
2143                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2144                 atomic_inc(&po->tx_ring.pending);
2145
2146                 status = TP_STATUS_SEND_REQUEST;
2147                 err = dev_queue_xmit(skb);
2148                 if (unlikely(err > 0)) {
2149                         err = net_xmit_errno(err);
2150                         if (err && __packet_get_status(po, ph) ==
2151                                    TP_STATUS_AVAILABLE) {
2152                                 /* skb was destructed already */
2153                                 skb = NULL;
2154                                 goto out_status;
2155                         }
2156                         /*
2157                          * skb was dropped but not destructed yet;
2158                          * let's treat it like congestion or err < 0
2159                          */
2160                         err = 0;
2161                 }
2162                 packet_increment_head(&po->tx_ring);
2163                 len_sum += tp_len;
2164         } while (likely((ph != NULL) ||
2165                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2166                          (atomic_read(&po->tx_ring.pending))))
2167                 );
2168
2169         err = len_sum;
2170         goto out_put;
2171
2172 out_status:
2173         __packet_set_status(po, ph, status);
2174         kfree_skb(skb);
2175 out_put:
2176         if (need_rls_dev)
2177                 dev_put(dev);
2178 out:
2179         mutex_unlock(&po->pg_vec_lock);
2180         return err;
2181 }
2182
2183 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2184                                         size_t reserve, size_t len,
2185                                         size_t linear, int noblock,
2186                                         int *err)
2187 {
2188         struct sk_buff *skb;
2189
2190         /* Under a page?  Don't bother with paged skb. */
2191         if (prepad + len < PAGE_SIZE || !linear)
2192                 linear = len;
2193
2194         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2195                                    err);
2196         if (!skb)
2197                 return NULL;
2198
2199         skb_reserve(skb, reserve);
2200         skb_put(skb, linear);
2201         skb->data_len = len - linear;
2202         skb->len += len - linear;
2203
2204         return skb;
2205 }
2206
2207 static int packet_snd(struct socket *sock,
2208                           struct msghdr *msg, size_t len)
2209 {
2210         struct sock *sk = sock->sk;
2211         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2212         struct sk_buff *skb;
2213         struct net_device *dev;
2214         __be16 proto;
2215         bool need_rls_dev = false;
2216         unsigned char *addr;
2217         int err, reserve = 0;
2218         struct virtio_net_hdr vnet_hdr = { 0 };
2219         int offset = 0;
2220         int vnet_hdr_len;
2221         struct packet_sock *po = pkt_sk(sk);
2222         unsigned short gso_type = 0;
2223         int hlen, tlen;
2224         int extra_len = 0;
2225
2226         /*
2227          *      Get and verify the address.
2228          */
2229
2230         if (saddr == NULL) {
2231                 dev = po->prot_hook.dev;
2232                 proto   = po->num;
2233                 addr    = NULL;
2234         } else {
2235                 err = -EINVAL;
2236                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2237                         goto out;
2238                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2239                         goto out;
2240                 proto   = saddr->sll_protocol;
2241                 addr    = saddr->sll_addr;
2242                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2243                 need_rls_dev = true;
2244         }
2245
2246         err = -ENXIO;
2247         if (dev == NULL)
2248                 goto out_unlock;
2249         if (sock->type == SOCK_RAW)
2250                 reserve = dev->hard_header_len;
2251
2252         err = -ENETDOWN;
2253         if (!(dev->flags & IFF_UP))
2254                 goto out_unlock;
2255
2256         if (po->has_vnet_hdr) {
2257                 vnet_hdr_len = sizeof(vnet_hdr);
2258
2259                 err = -EINVAL;
2260                 if (len < vnet_hdr_len)
2261                         goto out_unlock;
2262
2263                 len -= vnet_hdr_len;
2264
2265                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2266                                        vnet_hdr_len);
2267                 if (err < 0)
2268                         goto out_unlock;
2269
2270                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2271                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2272                       vnet_hdr.hdr_len))
2273                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2274                                                  vnet_hdr.csum_offset + 2;
2275
2276                 err = -EINVAL;
2277                 if (vnet_hdr.hdr_len > len)
2278                         goto out_unlock;
2279
2280                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2281                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2282                         case VIRTIO_NET_HDR_GSO_TCPV4:
2283                                 gso_type = SKB_GSO_TCPV4;
2284                                 break;
2285                         case VIRTIO_NET_HDR_GSO_TCPV6:
2286                                 gso_type = SKB_GSO_TCPV6;
2287                                 break;
2288                         case VIRTIO_NET_HDR_GSO_UDP:
2289                                 gso_type = SKB_GSO_UDP;
2290                                 break;
2291                         default:
2292                                 goto out_unlock;
2293                         }
2294
2295                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2296                                 gso_type |= SKB_GSO_TCP_ECN;
2297
2298                         if (vnet_hdr.gso_size == 0)
2299                                 goto out_unlock;
2300
2301                 }
2302         }
2303
2304         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2305                 if (!netif_supports_nofcs(dev)) {
2306                         err = -EPROTONOSUPPORT;
2307                         goto out_unlock;
2308                 }
2309                 extra_len = 4; /* We're doing our own CRC */
2310         }
2311
2312         err = -EMSGSIZE;
2313         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2314                 goto out_unlock;
2315
2316         err = -ENOBUFS;
2317         hlen = LL_RESERVED_SPACE(dev);
2318         tlen = dev->needed_tailroom;
2319         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, vnet_hdr.hdr_len,
2320                                msg->msg_flags & MSG_DONTWAIT, &err);
2321         if (skb == NULL)
2322                 goto out_unlock;
2323
2324         skb_set_network_header(skb, reserve);
2325
2326         err = -EINVAL;
2327         if (sock->type == SOCK_DGRAM &&
2328             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2329                 goto out_free;
2330
2331         /* Returns -EFAULT on error */
2332         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2333         if (err)
2334                 goto out_free;
2335         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2336         if (err < 0)
2337                 goto out_free;
2338
2339         if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2340                 /* Earlier code assumed this would be a VLAN pkt,
2341                  * double-check this now that we have the actual
2342                  * packet in hand.
2343                  */
2344                 struct ethhdr *ehdr;
2345                 skb_reset_mac_header(skb);
2346                 ehdr = eth_hdr(skb);
2347                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2348                         err = -EMSGSIZE;
2349                         goto out_free;
2350                 }
2351         }
2352
2353         skb->protocol = proto;
2354         skb->dev = dev;
2355         skb->priority = sk->sk_priority;
2356         skb->mark = sk->sk_mark;
2357
2358         if (po->has_vnet_hdr) {
2359                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2360                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2361                                                   vnet_hdr.csum_offset)) {
2362                                 err = -EINVAL;
2363                                 goto out_free;
2364                         }
2365                 }
2366
2367                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2368                 skb_shinfo(skb)->gso_type = gso_type;
2369
2370                 /* Header must be checked, and gso_segs computed. */
2371                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2372                 skb_shinfo(skb)->gso_segs = 0;
2373
2374                 len += vnet_hdr_len;
2375         }
2376
2377         if (unlikely(extra_len == 4))
2378                 skb->no_fcs = 1;
2379
2380         /*
2381          *      Now send it
2382          */
2383
2384         err = dev_queue_xmit(skb);
2385         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2386                 goto out_unlock;
2387
2388         if (need_rls_dev)
2389                 dev_put(dev);
2390
2391         return len;
2392
2393 out_free:
2394         kfree_skb(skb);
2395 out_unlock:
2396         if (dev && need_rls_dev)
2397                 dev_put(dev);
2398 out:
2399         return err;
2400 }
2401
2402 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2403                 struct msghdr *msg, size_t len)
2404 {
2405         struct sock *sk = sock->sk;
2406         struct packet_sock *po = pkt_sk(sk);
2407         if (po->tx_ring.pg_vec)
2408                 return tpacket_snd(po, msg);
2409         else
2410                 return packet_snd(sock, msg, len);
2411 }
2412
2413 /*
2414  *      Close a PACKET socket. This is fairly simple. We immediately go
2415  *      to 'closed' state and remove our protocol entry in the device list.
2416  */
2417
2418 static int packet_release(struct socket *sock)
2419 {
2420         struct sock *sk = sock->sk;
2421         struct packet_sock *po;
2422         struct net *net;
2423         union tpacket_req_u req_u;
2424
2425         if (!sk)
2426                 return 0;
2427
2428         net = sock_net(sk);
2429         po = pkt_sk(sk);
2430
2431         spin_lock_bh(&net->packet.sklist_lock);
2432         sk_del_node_init_rcu(sk);
2433         sock_prot_inuse_add(net, sk->sk_prot, -1);
2434         spin_unlock_bh(&net->packet.sklist_lock);
2435
2436         spin_lock(&po->bind_lock);
2437         unregister_prot_hook(sk, false);
2438         if (po->prot_hook.dev) {
2439                 dev_put(po->prot_hook.dev);
2440                 po->prot_hook.dev = NULL;
2441         }
2442         spin_unlock(&po->bind_lock);
2443
2444         packet_flush_mclist(sk);
2445
2446         memset(&req_u, 0, sizeof(req_u));
2447
2448         if (po->rx_ring.pg_vec)
2449                 packet_set_ring(sk, &req_u, 1, 0);
2450
2451         if (po->tx_ring.pg_vec)
2452                 packet_set_ring(sk, &req_u, 1, 1);
2453
2454         fanout_release(sk);
2455
2456         synchronize_net();
2457         /*
2458          *      Now the socket is dead. No more input will appear.
2459          */
2460         sock_orphan(sk);
2461         sock->sk = NULL;
2462
2463         /* Purge queues */
2464
2465         skb_queue_purge(&sk->sk_receive_queue);
2466         sk_refcnt_debug_release(sk);
2467
2468         sock_put(sk);
2469         return 0;
2470 }
2471
2472 /*
2473  *      Attach a packet hook.
2474  */
2475
2476 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2477 {
2478         struct packet_sock *po = pkt_sk(sk);
2479
2480         if (po->fanout) {
2481                 if (dev)
2482                         dev_put(dev);
2483
2484                 return -EINVAL;
2485         }
2486
2487         lock_sock(sk);
2488
2489         spin_lock(&po->bind_lock);
2490         unregister_prot_hook(sk, true);
2491         po->num = protocol;
2492         po->prot_hook.type = protocol;
2493         if (po->prot_hook.dev)
2494                 dev_put(po->prot_hook.dev);
2495         po->prot_hook.dev = dev;
2496
2497         po->ifindex = dev ? dev->ifindex : 0;
2498
2499         if (protocol == 0)
2500                 goto out_unlock;
2501
2502         if (!dev || (dev->flags & IFF_UP)) {
2503                 register_prot_hook(sk);
2504         } else {
2505                 sk->sk_err = ENETDOWN;
2506                 if (!sock_flag(sk, SOCK_DEAD))
2507                         sk->sk_error_report(sk);
2508         }
2509
2510 out_unlock:
2511         spin_unlock(&po->bind_lock);
2512         release_sock(sk);
2513         return 0;
2514 }
2515
2516 /*
2517  *      Bind a packet socket to a device
2518  */
2519
2520 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2521                             int addr_len)
2522 {
2523         struct sock *sk = sock->sk;
2524         char name[15];
2525         struct net_device *dev;
2526         int err = -ENODEV;
2527
2528         /*
2529          *      Check legality
2530          */
2531
2532         if (addr_len != sizeof(struct sockaddr))
2533                 return -EINVAL;
2534         strlcpy(name, uaddr->sa_data, sizeof(name));
2535
2536         dev = dev_get_by_name(sock_net(sk), name);
2537         if (dev)
2538                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2539         return err;
2540 }
2541
2542 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2543 {
2544         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2545         struct sock *sk = sock->sk;
2546         struct net_device *dev = NULL;
2547         int err;
2548
2549
2550         /*
2551          *      Check legality
2552          */
2553
2554         if (addr_len < sizeof(struct sockaddr_ll))
2555                 return -EINVAL;
2556         if (sll->sll_family != AF_PACKET)
2557                 return -EINVAL;
2558
2559         if (sll->sll_ifindex) {
2560                 err = -ENODEV;
2561                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2562                 if (dev == NULL)
2563                         goto out;
2564         }
2565         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2566
2567 out:
2568         return err;
2569 }
2570
2571 static struct proto packet_proto = {
2572         .name     = "PACKET",
2573         .owner    = THIS_MODULE,
2574         .obj_size = sizeof(struct packet_sock),
2575 };
2576
2577 /*
2578  *      Create a packet of type SOCK_PACKET.
2579  */
2580
2581 static int packet_create(struct net *net, struct socket *sock, int protocol,
2582                          int kern)
2583 {
2584         struct sock *sk;
2585         struct packet_sock *po;
2586         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2587         int err;
2588
2589         if (!capable(CAP_NET_RAW))
2590                 return -EPERM;
2591         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2592             sock->type != SOCK_PACKET)
2593                 return -ESOCKTNOSUPPORT;
2594
2595         sock->state = SS_UNCONNECTED;
2596
2597         err = -ENOBUFS;
2598         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2599         if (sk == NULL)
2600                 goto out;
2601
2602         sock->ops = &packet_ops;
2603         if (sock->type == SOCK_PACKET)
2604                 sock->ops = &packet_ops_spkt;
2605
2606         sock_init_data(sock, sk);
2607
2608         po = pkt_sk(sk);
2609         sk->sk_family = PF_PACKET;
2610         po->num = proto;
2611
2612         sk->sk_destruct = packet_sock_destruct;
2613         sk_refcnt_debug_inc(sk);
2614
2615         /*
2616          *      Attach a protocol block
2617          */
2618
2619         spin_lock_init(&po->bind_lock);
2620         mutex_init(&po->pg_vec_lock);
2621         po->prot_hook.func = packet_rcv;
2622
2623         if (sock->type == SOCK_PACKET)
2624                 po->prot_hook.func = packet_rcv_spkt;
2625
2626         po->prot_hook.af_packet_priv = sk;
2627
2628         if (proto) {
2629                 po->prot_hook.type = proto;
2630                 register_prot_hook(sk);
2631         }
2632
2633         spin_lock_bh(&net->packet.sklist_lock);
2634         sk_add_node_rcu(sk, &net->packet.sklist);
2635         sock_prot_inuse_add(net, &packet_proto, 1);
2636         spin_unlock_bh(&net->packet.sklist_lock);
2637
2638         return 0;
2639 out:
2640         return err;
2641 }
2642
2643 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2644 {
2645         struct sock_exterr_skb *serr;
2646         struct sk_buff *skb, *skb2;
2647         int copied, err;
2648
2649         err = -EAGAIN;
2650         skb = skb_dequeue(&sk->sk_error_queue);
2651         if (skb == NULL)
2652                 goto out;
2653
2654         copied = skb->len;
2655         if (copied > len) {
2656                 msg->msg_flags |= MSG_TRUNC;
2657                 copied = len;
2658         }
2659         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2660         if (err)
2661                 goto out_free_skb;
2662
2663         sock_recv_timestamp(msg, sk, skb);
2664
2665         serr = SKB_EXT_ERR(skb);
2666         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2667                  sizeof(serr->ee), &serr->ee);
2668
2669         msg->msg_flags |= MSG_ERRQUEUE;
2670         err = copied;
2671
2672         /* Reset and regenerate socket error */
2673         spin_lock_bh(&sk->sk_error_queue.lock);
2674         sk->sk_err = 0;
2675         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2676                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2677                 spin_unlock_bh(&sk->sk_error_queue.lock);
2678                 sk->sk_error_report(sk);
2679         } else
2680                 spin_unlock_bh(&sk->sk_error_queue.lock);
2681
2682 out_free_skb:
2683         kfree_skb(skb);
2684 out:
2685         return err;
2686 }
2687
2688 /*
2689  *      Pull a packet from our receive queue and hand it to the user.
2690  *      If necessary we block.
2691  */
2692
2693 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2694                           struct msghdr *msg, size_t len, int flags)
2695 {
2696         struct sock *sk = sock->sk;
2697         struct sk_buff *skb;
2698         int copied, err;
2699         struct sockaddr_ll *sll;
2700         int vnet_hdr_len = 0;
2701
2702         err = -EINVAL;
2703         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2704                 goto out;
2705
2706 #if 0
2707         /* What error should we return now? EUNATTACH? */
2708         if (pkt_sk(sk)->ifindex < 0)
2709                 return -ENODEV;
2710 #endif
2711
2712         if (flags & MSG_ERRQUEUE) {
2713                 err = packet_recv_error(sk, msg, len);
2714                 goto out;
2715         }
2716
2717         /*
2718          *      Call the generic datagram receiver. This handles all sorts
2719          *      of horrible races and re-entrancy so we can forget about it
2720          *      in the protocol layers.
2721          *
2722          *      Now it will return ENETDOWN, if device have just gone down,
2723          *      but then it will block.
2724          */
2725
2726         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2727
2728         /*
2729          *      An error occurred so return it. Because skb_recv_datagram()
2730          *      handles the blocking we don't see and worry about blocking
2731          *      retries.
2732          */
2733
2734         if (skb == NULL)
2735                 goto out;
2736
2737         if (pkt_sk(sk)->has_vnet_hdr) {
2738                 struct virtio_net_hdr vnet_hdr = { 0 };
2739
2740                 err = -EINVAL;
2741                 vnet_hdr_len = sizeof(vnet_hdr);
2742                 if (len < vnet_hdr_len)
2743                         goto out_free;
2744
2745                 len -= vnet_hdr_len;
2746
2747                 if (skb_is_gso(skb)) {
2748                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2749
2750                         /* This is a hint as to how much should be linear. */
2751                         vnet_hdr.hdr_len = skb_headlen(skb);
2752                         vnet_hdr.gso_size = sinfo->gso_size;
2753                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2754                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2755                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2756                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2757                         else if (sinfo->gso_type & SKB_GSO_UDP)
2758                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2759                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2760                                 goto out_free;
2761                         else
2762                                 BUG();
2763                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2764                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2765                 } else
2766                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2767
2768                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2769                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2770                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2771                         vnet_hdr.csum_offset = skb->csum_offset;
2772                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2773                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2774                 } /* else everything is zero */
2775
2776                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2777                                      vnet_hdr_len);
2778                 if (err < 0)
2779                         goto out_free;
2780         }
2781
2782         /*
2783          *      If the address length field is there to be filled in, we fill
2784          *      it in now.
2785          */
2786
2787         sll = &PACKET_SKB_CB(skb)->sa.ll;
2788         if (sock->type == SOCK_PACKET)
2789                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2790         else
2791                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2792
2793         /*
2794          *      You lose any data beyond the buffer you gave. If it worries a
2795          *      user program they can ask the device for its MTU anyway.
2796          */
2797
2798         copied = skb->len;
2799         if (copied > len) {
2800                 copied = len;
2801                 msg->msg_flags |= MSG_TRUNC;
2802         }
2803
2804         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2805         if (err)
2806                 goto out_free;
2807
2808         sock_recv_ts_and_drops(msg, sk, skb);
2809
2810         if (msg->msg_name)
2811                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2812                        msg->msg_namelen);
2813
2814         if (pkt_sk(sk)->auxdata) {
2815                 struct tpacket_auxdata aux;
2816
2817                 aux.tp_status = TP_STATUS_USER;
2818                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2819                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2820                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2821                 aux.tp_snaplen = skb->len;
2822                 aux.tp_mac = 0;
2823                 aux.tp_net = skb_network_offset(skb);
2824                 if (vlan_tx_tag_present(skb)) {
2825                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2826                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2827                 } else {
2828                         aux.tp_vlan_tci = 0;
2829                 }
2830                 aux.tp_padding = 0;
2831                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2832         }
2833
2834         /*
2835          *      Free or return the buffer as appropriate. Again this
2836          *      hides all the races and re-entrancy issues from us.
2837          */
2838         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2839
2840 out_free:
2841         skb_free_datagram(sk, skb);
2842 out:
2843         return err;
2844 }
2845
2846 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2847                                int *uaddr_len, int peer)
2848 {
2849         struct net_device *dev;
2850         struct sock *sk = sock->sk;
2851
2852         if (peer)
2853                 return -EOPNOTSUPP;
2854
2855         uaddr->sa_family = AF_PACKET;
2856         rcu_read_lock();
2857         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2858         if (dev)
2859                 strncpy(uaddr->sa_data, dev->name, 14);
2860         else
2861                 memset(uaddr->sa_data, 0, 14);
2862         rcu_read_unlock();
2863         *uaddr_len = sizeof(*uaddr);
2864
2865         return 0;
2866 }
2867
2868 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2869                           int *uaddr_len, int peer)
2870 {
2871         struct net_device *dev;
2872         struct sock *sk = sock->sk;
2873         struct packet_sock *po = pkt_sk(sk);
2874         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2875
2876         if (peer)
2877                 return -EOPNOTSUPP;
2878
2879         sll->sll_family = AF_PACKET;
2880         sll->sll_ifindex = po->ifindex;
2881         sll->sll_protocol = po->num;
2882         sll->sll_pkttype = 0;
2883         rcu_read_lock();
2884         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2885         if (dev) {
2886                 sll->sll_hatype = dev->type;
2887                 sll->sll_halen = dev->addr_len;
2888                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2889         } else {
2890                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2891                 sll->sll_halen = 0;
2892         }
2893         rcu_read_unlock();
2894         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2895
2896         return 0;
2897 }
2898
2899 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2900                          int what)
2901 {
2902         switch (i->type) {
2903         case PACKET_MR_MULTICAST:
2904                 if (i->alen != dev->addr_len)
2905                         return -EINVAL;
2906                 if (what > 0)
2907                         return dev_mc_add(dev, i->addr);
2908                 else
2909                         return dev_mc_del(dev, i->addr);
2910                 break;
2911         case PACKET_MR_PROMISC:
2912                 return dev_set_promiscuity(dev, what);
2913                 break;
2914         case PACKET_MR_ALLMULTI:
2915                 return dev_set_allmulti(dev, what);
2916                 break;
2917         case PACKET_MR_UNICAST:
2918                 if (i->alen != dev->addr_len)
2919                         return -EINVAL;
2920                 if (what > 0)
2921                         return dev_uc_add(dev, i->addr);
2922                 else
2923                         return dev_uc_del(dev, i->addr);
2924                 break;
2925         default:
2926                 break;
2927         }
2928         return 0;
2929 }
2930
2931 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2932 {
2933         for ( ; i; i = i->next) {
2934                 if (i->ifindex == dev->ifindex)
2935                         packet_dev_mc(dev, i, what);
2936         }
2937 }
2938
2939 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2940 {
2941         struct packet_sock *po = pkt_sk(sk);
2942         struct packet_mclist *ml, *i;
2943         struct net_device *dev;
2944         int err;
2945
2946         rtnl_lock();
2947
2948         err = -ENODEV;
2949         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2950         if (!dev)
2951                 goto done;
2952
2953         err = -EINVAL;
2954         if (mreq->mr_alen > dev->addr_len)
2955                 goto done;
2956
2957         err = -ENOBUFS;
2958         i = kmalloc(sizeof(*i), GFP_KERNEL);
2959         if (i == NULL)
2960                 goto done;
2961
2962         err = 0;
2963         for (ml = po->mclist; ml; ml = ml->next) {
2964                 if (ml->ifindex == mreq->mr_ifindex &&
2965                     ml->type == mreq->mr_type &&
2966                     ml->alen == mreq->mr_alen &&
2967                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2968                         ml->count++;
2969                         /* Free the new element ... */
2970                         kfree(i);
2971                         goto done;
2972                 }
2973         }
2974
2975         i->type = mreq->mr_type;
2976         i->ifindex = mreq->mr_ifindex;
2977         i->alen = mreq->mr_alen;
2978         memcpy(i->addr, mreq->mr_address, i->alen);
2979         i->count = 1;
2980         i->next = po->mclist;
2981         po->mclist = i;
2982         err = packet_dev_mc(dev, i, 1);
2983         if (err) {
2984                 po->mclist = i->next;
2985                 kfree(i);
2986         }
2987
2988 done:
2989         rtnl_unlock();
2990         return err;
2991 }
2992
2993 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2994 {
2995         struct packet_mclist *ml, **mlp;
2996
2997         rtnl_lock();
2998
2999         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3000                 if (ml->ifindex == mreq->mr_ifindex &&
3001                     ml->type == mreq->mr_type &&
3002                     ml->alen == mreq->mr_alen &&
3003                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3004                         if (--ml->count == 0) {
3005                                 struct net_device *dev;
3006                                 *mlp = ml->next;
3007                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3008                                 if (dev)
3009                                         packet_dev_mc(dev, ml, -1);
3010                                 kfree(ml);
3011                         }
3012                         rtnl_unlock();
3013                         return 0;
3014                 }
3015         }
3016         rtnl_unlock();
3017         return -EADDRNOTAVAIL;
3018 }
3019
3020 static void packet_flush_mclist(struct sock *sk)
3021 {
3022         struct packet_sock *po = pkt_sk(sk);
3023         struct packet_mclist *ml;
3024
3025         if (!po->mclist)
3026                 return;
3027
3028         rtnl_lock();
3029         while ((ml = po->mclist) != NULL) {
3030                 struct net_device *dev;
3031
3032                 po->mclist = ml->next;
3033                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3034                 if (dev != NULL)
3035                         packet_dev_mc(dev, ml, -1);
3036                 kfree(ml);
3037         }
3038         rtnl_unlock();
3039 }
3040
3041 static int
3042 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3043 {
3044         struct sock *sk = sock->sk;
3045         struct packet_sock *po = pkt_sk(sk);
3046         int ret;
3047
3048         if (level != SOL_PACKET)
3049                 return -ENOPROTOOPT;
3050
3051         switch (optname) {
3052         case PACKET_ADD_MEMBERSHIP:
3053         case PACKET_DROP_MEMBERSHIP:
3054         {
3055                 struct packet_mreq_max mreq;
3056                 int len = optlen;
3057                 memset(&mreq, 0, sizeof(mreq));
3058                 if (len < sizeof(struct packet_mreq))
3059                         return -EINVAL;
3060                 if (len > sizeof(mreq))
3061                         len = sizeof(mreq);
3062                 if (copy_from_user(&mreq, optval, len))
3063                         return -EFAULT;
3064                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3065                         return -EINVAL;
3066                 if (optname == PACKET_ADD_MEMBERSHIP)
3067                         ret = packet_mc_add(sk, &mreq);
3068                 else
3069                         ret = packet_mc_drop(sk, &mreq);
3070                 return ret;
3071         }
3072
3073         case PACKET_RX_RING:
3074         case PACKET_TX_RING:
3075         {
3076                 union tpacket_req_u req_u;
3077                 int len;
3078
3079                 switch (po->tp_version) {
3080                 case TPACKET_V1:
3081                 case TPACKET_V2:
3082                         len = sizeof(req_u.req);
3083                         break;
3084                 case TPACKET_V3:
3085                 default:
3086                         len = sizeof(req_u.req3);
3087                         break;
3088                 }
3089                 if (optlen < len)
3090                         return -EINVAL;
3091                 if (pkt_sk(sk)->has_vnet_hdr)
3092                         return -EINVAL;
3093                 if (copy_from_user(&req_u.req, optval, len))
3094                         return -EFAULT;
3095                 return packet_set_ring(sk, &req_u, 0,
3096                         optname == PACKET_TX_RING);
3097         }
3098         case PACKET_COPY_THRESH:
3099         {
3100                 int val;
3101
3102                 if (optlen != sizeof(val))
3103                         return -EINVAL;
3104                 if (copy_from_user(&val, optval, sizeof(val)))
3105                         return -EFAULT;
3106
3107                 pkt_sk(sk)->copy_thresh = val;
3108                 return 0;
3109         }
3110         case PACKET_VERSION:
3111         {
3112                 int val;
3113
3114                 if (optlen != sizeof(val))
3115                         return -EINVAL;
3116                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3117                         return -EBUSY;
3118                 if (copy_from_user(&val, optval, sizeof(val)))
3119                         return -EFAULT;
3120                 switch (val) {
3121                 case TPACKET_V1:
3122                 case TPACKET_V2:
3123                 case TPACKET_V3:
3124                         po->tp_version = val;
3125                         return 0;
3126                 default:
3127                         return -EINVAL;
3128                 }
3129         }
3130         case PACKET_RESERVE:
3131         {
3132                 unsigned int val;
3133
3134                 if (optlen != sizeof(val))
3135                         return -EINVAL;
3136                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3137                         return -EBUSY;
3138                 if (copy_from_user(&val, optval, sizeof(val)))
3139                         return -EFAULT;
3140                 po->tp_reserve = val;
3141                 return 0;
3142         }
3143         case PACKET_LOSS:
3144         {
3145                 unsigned int val;
3146
3147                 if (optlen != sizeof(val))
3148                         return -EINVAL;
3149                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3150                         return -EBUSY;
3151                 if (copy_from_user(&val, optval, sizeof(val)))
3152                         return -EFAULT;
3153                 po->tp_loss = !!val;
3154                 return 0;
3155         }
3156         case PACKET_AUXDATA:
3157         {
3158                 int val;
3159
3160                 if (optlen < sizeof(val))
3161                         return -EINVAL;
3162                 if (copy_from_user(&val, optval, sizeof(val)))
3163                         return -EFAULT;
3164
3165                 po->auxdata = !!val;
3166                 return 0;
3167         }
3168         case PACKET_ORIGDEV:
3169         {
3170                 int val;
3171
3172                 if (optlen < sizeof(val))
3173                         return -EINVAL;
3174                 if (copy_from_user(&val, optval, sizeof(val)))
3175                         return -EFAULT;
3176
3177                 po->origdev = !!val;
3178                 return 0;
3179         }
3180         case PACKET_VNET_HDR:
3181         {
3182                 int val;
3183
3184                 if (sock->type != SOCK_RAW)
3185                         return -EINVAL;
3186                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3187                         return -EBUSY;
3188                 if (optlen < sizeof(val))
3189                         return -EINVAL;
3190                 if (copy_from_user(&val, optval, sizeof(val)))
3191                         return -EFAULT;
3192
3193                 po->has_vnet_hdr = !!val;
3194                 return 0;
3195         }
3196         case PACKET_TIMESTAMP:
3197         {
3198                 int val;
3199
3200                 if (optlen != sizeof(val))
3201                         return -EINVAL;
3202                 if (copy_from_user(&val, optval, sizeof(val)))
3203                         return -EFAULT;
3204
3205                 po->tp_tstamp = val;
3206                 return 0;
3207         }
3208         case PACKET_FANOUT:
3209         {
3210                 int val;
3211
3212                 if (optlen != sizeof(val))
3213                         return -EINVAL;
3214                 if (copy_from_user(&val, optval, sizeof(val)))
3215                         return -EFAULT;
3216
3217                 return fanout_add(sk, val & 0xffff, val >> 16);
3218         }
3219         default:
3220                 return -ENOPROTOOPT;
3221         }
3222 }
3223
3224 static int packet_getsockopt(struct socket *sock, int level, int optname,
3225                              char __user *optval, int __user *optlen)
3226 {
3227         int len;
3228         int val;
3229         struct sock *sk = sock->sk;
3230         struct packet_sock *po = pkt_sk(sk);
3231         void *data;
3232         struct tpacket_stats st;
3233         union tpacket_stats_u st_u;
3234
3235         if (level != SOL_PACKET)
3236                 return -ENOPROTOOPT;
3237
3238         if (get_user(len, optlen))
3239                 return -EFAULT;
3240
3241         if (len < 0)
3242                 return -EINVAL;
3243
3244         switch (optname) {
3245         case PACKET_STATISTICS:
3246                 if (po->tp_version == TPACKET_V3) {
3247                         len = sizeof(struct tpacket_stats_v3);
3248                 } else {
3249                         if (len > sizeof(struct tpacket_stats))
3250                                 len = sizeof(struct tpacket_stats);
3251                 }
3252                 spin_lock_bh(&sk->sk_receive_queue.lock);
3253                 if (po->tp_version == TPACKET_V3) {
3254                         memcpy(&st_u.stats3, &po->stats,
3255                         sizeof(struct tpacket_stats));
3256                         st_u.stats3.tp_freeze_q_cnt =
3257                         po->stats_u.stats3.tp_freeze_q_cnt;
3258                         st_u.stats3.tp_packets += po->stats.tp_drops;
3259                         data = &st_u.stats3;
3260                 } else {
3261                         st = po->stats;
3262                         st.tp_packets += st.tp_drops;
3263                         data = &st;
3264                 }
3265                 memset(&po->stats, 0, sizeof(st));
3266                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3267                 break;
3268         case PACKET_AUXDATA:
3269                 if (len > sizeof(int))
3270                         len = sizeof(int);
3271                 val = po->auxdata;
3272
3273                 data = &val;
3274                 break;
3275         case PACKET_ORIGDEV:
3276                 if (len > sizeof(int))
3277                         len = sizeof(int);
3278                 val = po->origdev;
3279
3280                 data = &val;
3281                 break;
3282         case PACKET_VNET_HDR:
3283                 if (len > sizeof(int))
3284                         len = sizeof(int);
3285                 val = po->has_vnet_hdr;
3286
3287                 data = &val;
3288                 break;
3289         case PACKET_VERSION:
3290                 if (len > sizeof(int))
3291                         len = sizeof(int);
3292                 val = po->tp_version;
3293                 data = &val;
3294                 break;
3295         case PACKET_HDRLEN:
3296                 if (len > sizeof(int))
3297                         len = sizeof(int);
3298                 if (copy_from_user(&val, optval, len))
3299                         return -EFAULT;
3300                 switch (val) {
3301                 case TPACKET_V1:
3302                         val = sizeof(struct tpacket_hdr);
3303                         break;
3304                 case TPACKET_V2:
3305                         val = sizeof(struct tpacket2_hdr);
3306                         break;
3307                 case TPACKET_V3:
3308                         val = sizeof(struct tpacket3_hdr);
3309                         break;
3310                 default:
3311                         return -EINVAL;
3312                 }
3313                 data = &val;
3314                 break;
3315         case PACKET_RESERVE:
3316                 if (len > sizeof(unsigned int))
3317                         len = sizeof(unsigned int);
3318                 val = po->tp_reserve;
3319                 data = &val;
3320                 break;
3321         case PACKET_LOSS:
3322                 if (len > sizeof(unsigned int))
3323                         len = sizeof(unsigned int);
3324                 val = po->tp_loss;
3325                 data = &val;
3326                 break;
3327         case PACKET_TIMESTAMP:
3328                 if (len > sizeof(int))
3329                         len = sizeof(int);
3330                 val = po->tp_tstamp;
3331                 data = &val;
3332                 break;
3333         case PACKET_FANOUT:
3334                 if (len > sizeof(int))
3335                         len = sizeof(int);
3336                 val = (po->fanout ?
3337                        ((u32)po->fanout->id |
3338                         ((u32)po->fanout->type << 16)) :
3339                        0);
3340                 data = &val;
3341                 break;
3342         default:
3343                 return -ENOPROTOOPT;
3344         }
3345
3346         if (put_user(len, optlen))
3347                 return -EFAULT;
3348         if (copy_to_user(optval, data, len))
3349                 return -EFAULT;
3350         return 0;
3351 }
3352
3353
3354 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3355 {
3356         struct sock *sk;
3357         struct hlist_node *node;
3358         struct net_device *dev = data;
3359         struct net *net = dev_net(dev);
3360
3361         rcu_read_lock();
3362         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3363                 struct packet_sock *po = pkt_sk(sk);
3364
3365                 switch (msg) {
3366                 case NETDEV_UNREGISTER:
3367                         if (po->mclist)
3368                                 packet_dev_mclist(dev, po->mclist, -1);
3369                         /* fallthrough */
3370
3371                 case NETDEV_DOWN:
3372                         if (dev->ifindex == po->ifindex) {
3373                                 spin_lock(&po->bind_lock);
3374                                 if (po->running) {
3375                                         __unregister_prot_hook(sk, false);
3376                                         sk->sk_err = ENETDOWN;
3377                                         if (!sock_flag(sk, SOCK_DEAD))
3378                                                 sk->sk_error_report(sk);
3379                                 }
3380                                 if (msg == NETDEV_UNREGISTER) {
3381                                         po->ifindex = -1;
3382                                         if (po->prot_hook.dev)
3383                                                 dev_put(po->prot_hook.dev);
3384                                         po->prot_hook.dev = NULL;
3385                                 }
3386                                 spin_unlock(&po->bind_lock);
3387                         }
3388                         break;
3389                 case NETDEV_UP:
3390                         if (dev->ifindex == po->ifindex) {
3391                                 spin_lock(&po->bind_lock);
3392                                 if (po->num)
3393                                         register_prot_hook(sk);
3394                                 spin_unlock(&po->bind_lock);
3395                         }
3396                         break;
3397                 }
3398         }
3399         rcu_read_unlock();
3400         return NOTIFY_DONE;
3401 }
3402
3403
3404 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3405                         unsigned long arg)
3406 {
3407         struct sock *sk = sock->sk;
3408
3409         switch (cmd) {
3410         case SIOCOUTQ:
3411         {
3412                 int amount = sk_wmem_alloc_get(sk);
3413
3414                 return put_user(amount, (int __user *)arg);
3415         }
3416         case SIOCINQ:
3417         {
3418                 struct sk_buff *skb;
3419                 int amount = 0;
3420
3421                 spin_lock_bh(&sk->sk_receive_queue.lock);
3422                 skb = skb_peek(&sk->sk_receive_queue);
3423                 if (skb)
3424                         amount = skb->len;
3425                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3426                 return put_user(amount, (int __user *)arg);
3427         }
3428         case SIOCGSTAMP:
3429                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3430         case SIOCGSTAMPNS:
3431                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3432
3433 #ifdef CONFIG_INET
3434         case SIOCADDRT:
3435         case SIOCDELRT:
3436         case SIOCDARP:
3437         case SIOCGARP:
3438         case SIOCSARP:
3439         case SIOCGIFADDR:
3440         case SIOCSIFADDR:
3441         case SIOCGIFBRDADDR:
3442         case SIOCSIFBRDADDR:
3443         case SIOCGIFNETMASK:
3444         case SIOCSIFNETMASK:
3445         case SIOCGIFDSTADDR:
3446         case SIOCSIFDSTADDR:
3447         case SIOCSIFFLAGS:
3448                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3449 #endif
3450
3451         default:
3452                 return -ENOIOCTLCMD;
3453         }
3454         return 0;
3455 }
3456
3457 static unsigned int packet_poll(struct file *file, struct socket *sock,
3458                                 poll_table *wait)
3459 {
3460         struct sock *sk = sock->sk;
3461         struct packet_sock *po = pkt_sk(sk);
3462         unsigned int mask = datagram_poll(file, sock, wait);
3463
3464         spin_lock_bh(&sk->sk_receive_queue.lock);
3465         if (po->rx_ring.pg_vec) {
3466                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3467                         TP_STATUS_KERNEL))
3468                         mask |= POLLIN | POLLRDNORM;
3469         }
3470         spin_unlock_bh(&sk->sk_receive_queue.lock);
3471         spin_lock_bh(&sk->sk_write_queue.lock);
3472         if (po->tx_ring.pg_vec) {
3473                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3474                         mask |= POLLOUT | POLLWRNORM;
3475         }
3476         spin_unlock_bh(&sk->sk_write_queue.lock);
3477         return mask;
3478 }
3479
3480
3481 /* Dirty? Well, I still did not learn better way to account
3482  * for user mmaps.
3483  */
3484
3485 static void packet_mm_open(struct vm_area_struct *vma)
3486 {
3487         struct file *file = vma->vm_file;
3488         struct socket *sock = file->private_data;
3489         struct sock *sk = sock->sk;
3490
3491         if (sk)
3492                 atomic_inc(&pkt_sk(sk)->mapped);
3493 }
3494
3495 static void packet_mm_close(struct vm_area_struct *vma)
3496 {
3497         struct file *file = vma->vm_file;
3498         struct socket *sock = file->private_data;
3499         struct sock *sk = sock->sk;
3500
3501         if (sk)
3502                 atomic_dec(&pkt_sk(sk)->mapped);
3503 }
3504
3505 static const struct vm_operations_struct packet_mmap_ops = {
3506         .open   =       packet_mm_open,
3507         .close  =       packet_mm_close,
3508 };
3509
3510 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3511                         unsigned int len)
3512 {
3513         int i;
3514
3515         for (i = 0; i < len; i++) {
3516                 if (likely(pg_vec[i].buffer)) {
3517                         if (is_vmalloc_addr(pg_vec[i].buffer))
3518                                 vfree(pg_vec[i].buffer);
3519                         else
3520                                 free_pages((unsigned long)pg_vec[i].buffer,
3521                                            order);
3522                         pg_vec[i].buffer = NULL;
3523                 }
3524         }
3525         kfree(pg_vec);
3526 }
3527
3528 static char *alloc_one_pg_vec_page(unsigned long order)
3529 {
3530         char *buffer = NULL;
3531         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3532                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3533
3534         buffer = (char *) __get_free_pages(gfp_flags, order);
3535
3536         if (buffer)
3537                 return buffer;
3538
3539         /*
3540          * __get_free_pages failed, fall back to vmalloc
3541          */
3542         buffer = vzalloc((1 << order) * PAGE_SIZE);
3543
3544         if (buffer)
3545                 return buffer;
3546
3547         /*
3548          * vmalloc failed, lets dig into swap here
3549          */
3550         gfp_flags &= ~__GFP_NORETRY;
3551         buffer = (char *)__get_free_pages(gfp_flags, order);
3552         if (buffer)
3553                 return buffer;
3554
3555         /*
3556          * complete and utter failure
3557          */
3558         return NULL;
3559 }
3560
3561 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3562 {
3563         unsigned int block_nr = req->tp_block_nr;
3564         struct pgv *pg_vec;
3565         int i;
3566
3567         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3568         if (unlikely(!pg_vec))
3569                 goto out;
3570
3571         for (i = 0; i < block_nr; i++) {
3572                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3573                 if (unlikely(!pg_vec[i].buffer))
3574                         goto out_free_pgvec;
3575         }
3576
3577 out:
3578         return pg_vec;
3579
3580 out_free_pgvec:
3581         free_pg_vec(pg_vec, order, block_nr);
3582         pg_vec = NULL;
3583         goto out;
3584 }
3585
3586 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3587                 int closing, int tx_ring)
3588 {
3589         struct pgv *pg_vec = NULL;
3590         struct packet_sock *po = pkt_sk(sk);
3591         int was_running, order = 0;
3592         struct packet_ring_buffer *rb;
3593         struct sk_buff_head *rb_queue;
3594         __be16 num;
3595         int err = -EINVAL;
3596         /* Added to avoid minimal code churn */
3597         struct tpacket_req *req = &req_u->req;
3598
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 (unlikely(req->tp_frame_size < po->tp_hdrlen +
3640                                         po->tp_reserve))
3641                         goto out;
3642                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3643                         goto out;
3644
3645                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3646                 if (unlikely(rb->frames_per_block <= 0))
3647                         goto out;
3648                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3649                                         req->tp_frame_nr))
3650                         goto out;
3651
3652                 err = -ENOMEM;
3653                 order = get_order(req->tp_block_size);
3654                 pg_vec = alloc_pg_vec(req, order);
3655                 if (unlikely(!pg_vec))
3656                         goto out;
3657                 switch (po->tp_version) {
3658                 case TPACKET_V3:
3659                 /* Transmit path is not supported. We checked
3660                  * it above but just being paranoid
3661                  */
3662                         if (!tx_ring)
3663                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3664                                 break;
3665                 default:
3666                         break;
3667                 }
3668         }
3669         /* Done */
3670         else {
3671                 err = -EINVAL;
3672                 if (unlikely(req->tp_frame_nr))
3673                         goto out;
3674         }
3675
3676         lock_sock(sk);
3677
3678         /* Detach socket from network */
3679         spin_lock(&po->bind_lock);
3680         was_running = po->running;
3681         num = po->num;
3682         if (was_running) {
3683                 po->num = 0;
3684                 __unregister_prot_hook(sk, false);
3685         }
3686         spin_unlock(&po->bind_lock);
3687
3688         synchronize_net();
3689
3690         err = -EBUSY;
3691         mutex_lock(&po->pg_vec_lock);
3692         if (closing || atomic_read(&po->mapped) == 0) {
3693                 err = 0;
3694                 spin_lock_bh(&rb_queue->lock);
3695                 swap(rb->pg_vec, pg_vec);
3696                 rb->frame_max = (req->tp_frame_nr - 1);
3697                 rb->head = 0;
3698                 rb->frame_size = req->tp_frame_size;
3699                 spin_unlock_bh(&rb_queue->lock);
3700
3701                 swap(rb->pg_vec_order, order);
3702                 swap(rb->pg_vec_len, req->tp_block_nr);
3703
3704                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3705                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3706                                                 tpacket_rcv : packet_rcv;
3707                 skb_queue_purge(rb_queue);
3708                 if (atomic_read(&po->mapped))
3709                         pr_err("packet_mmap: vma is busy: %d\n",
3710                                atomic_read(&po->mapped));
3711         }
3712         mutex_unlock(&po->pg_vec_lock);
3713
3714         spin_lock(&po->bind_lock);
3715         if (was_running) {
3716                 po->num = num;
3717                 register_prot_hook(sk);
3718         }
3719         spin_unlock(&po->bind_lock);
3720         if (closing && (po->tp_version > TPACKET_V2)) {
3721                 /* Because we don't support block-based V3 on tx-ring */
3722                 if (!tx_ring)
3723                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3724         }
3725         release_sock(sk);
3726
3727         if (pg_vec)
3728                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3729 out:
3730         return err;
3731 }
3732
3733 static int packet_mmap(struct file *file, struct socket *sock,
3734                 struct vm_area_struct *vma)
3735 {
3736         struct sock *sk = sock->sk;
3737         struct packet_sock *po = pkt_sk(sk);
3738         unsigned long size, expected_size;
3739         struct packet_ring_buffer *rb;
3740         unsigned long start;
3741         int err = -EINVAL;
3742         int i;
3743
3744         if (vma->vm_pgoff)
3745                 return -EINVAL;
3746
3747         mutex_lock(&po->pg_vec_lock);
3748
3749         expected_size = 0;
3750         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3751                 if (rb->pg_vec) {
3752                         expected_size += rb->pg_vec_len
3753                                                 * rb->pg_vec_pages
3754                                                 * PAGE_SIZE;
3755                 }
3756         }
3757
3758         if (expected_size == 0)
3759                 goto out;
3760
3761         size = vma->vm_end - vma->vm_start;
3762         if (size != expected_size)
3763                 goto out;
3764
3765         start = vma->vm_start;
3766         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3767                 if (rb->pg_vec == NULL)
3768                         continue;
3769
3770                 for (i = 0; i < rb->pg_vec_len; i++) {
3771                         struct page *page;
3772                         void *kaddr = rb->pg_vec[i].buffer;
3773                         int pg_num;
3774
3775                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3776                                 page = pgv_to_page(kaddr);
3777                                 err = vm_insert_page(vma, start, page);
3778                                 if (unlikely(err))
3779                                         goto out;
3780                                 start += PAGE_SIZE;
3781                                 kaddr += PAGE_SIZE;
3782                         }
3783                 }
3784         }
3785
3786         atomic_inc(&po->mapped);
3787         vma->vm_ops = &packet_mmap_ops;
3788         err = 0;
3789
3790 out:
3791         mutex_unlock(&po->pg_vec_lock);
3792         return err;
3793 }
3794
3795 static const struct proto_ops packet_ops_spkt = {
3796         .family =       PF_PACKET,
3797         .owner =        THIS_MODULE,
3798         .release =      packet_release,
3799         .bind =         packet_bind_spkt,
3800         .connect =      sock_no_connect,
3801         .socketpair =   sock_no_socketpair,
3802         .accept =       sock_no_accept,
3803         .getname =      packet_getname_spkt,
3804         .poll =         datagram_poll,
3805         .ioctl =        packet_ioctl,
3806         .listen =       sock_no_listen,
3807         .shutdown =     sock_no_shutdown,
3808         .setsockopt =   sock_no_setsockopt,
3809         .getsockopt =   sock_no_getsockopt,
3810         .sendmsg =      packet_sendmsg_spkt,
3811         .recvmsg =      packet_recvmsg,
3812         .mmap =         sock_no_mmap,
3813         .sendpage =     sock_no_sendpage,
3814 };
3815
3816 static const struct proto_ops packet_ops = {
3817         .family =       PF_PACKET,
3818         .owner =        THIS_MODULE,
3819         .release =      packet_release,
3820         .bind =         packet_bind,
3821         .connect =      sock_no_connect,
3822         .socketpair =   sock_no_socketpair,
3823         .accept =       sock_no_accept,
3824         .getname =      packet_getname,
3825         .poll =         packet_poll,
3826         .ioctl =        packet_ioctl,
3827         .listen =       sock_no_listen,
3828         .shutdown =     sock_no_shutdown,
3829         .setsockopt =   packet_setsockopt,
3830         .getsockopt =   packet_getsockopt,
3831         .sendmsg =      packet_sendmsg,
3832         .recvmsg =      packet_recvmsg,
3833         .mmap =         packet_mmap,
3834         .sendpage =     sock_no_sendpage,
3835 };
3836
3837 static const struct net_proto_family packet_family_ops = {
3838         .family =       PF_PACKET,
3839         .create =       packet_create,
3840         .owner  =       THIS_MODULE,
3841 };
3842
3843 static struct notifier_block packet_netdev_notifier = {
3844         .notifier_call =        packet_notifier,
3845 };
3846
3847 #ifdef CONFIG_PROC_FS
3848
3849 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3850         __acquires(RCU)
3851 {
3852         struct net *net = seq_file_net(seq);
3853
3854         rcu_read_lock();
3855         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3856 }
3857
3858 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3859 {
3860         struct net *net = seq_file_net(seq);
3861         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3862 }
3863
3864 static void packet_seq_stop(struct seq_file *seq, void *v)
3865         __releases(RCU)
3866 {
3867         rcu_read_unlock();
3868 }
3869
3870 static int packet_seq_show(struct seq_file *seq, void *v)
3871 {
3872         if (v == SEQ_START_TOKEN)
3873                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3874         else {
3875                 struct sock *s = sk_entry(v);
3876                 const struct packet_sock *po = pkt_sk(s);
3877
3878                 seq_printf(seq,
3879                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3880                            s,
3881                            atomic_read(&s->sk_refcnt),
3882                            s->sk_type,
3883                            ntohs(po->num),
3884                            po->ifindex,
3885                            po->running,
3886                            atomic_read(&s->sk_rmem_alloc),
3887                            sock_i_uid(s),
3888                            sock_i_ino(s));
3889         }
3890
3891         return 0;
3892 }
3893
3894 static const struct seq_operations packet_seq_ops = {
3895         .start  = packet_seq_start,
3896         .next   = packet_seq_next,
3897         .stop   = packet_seq_stop,
3898         .show   = packet_seq_show,
3899 };
3900
3901 static int packet_seq_open(struct inode *inode, struct file *file)
3902 {
3903         return seq_open_net(inode, file, &packet_seq_ops,
3904                             sizeof(struct seq_net_private));
3905 }
3906
3907 static const struct file_operations packet_seq_fops = {
3908         .owner          = THIS_MODULE,
3909         .open           = packet_seq_open,
3910         .read           = seq_read,
3911         .llseek         = seq_lseek,
3912         .release        = seq_release_net,
3913 };
3914
3915 #endif
3916
3917 static int __net_init packet_net_init(struct net *net)
3918 {
3919         spin_lock_init(&net->packet.sklist_lock);
3920         INIT_HLIST_HEAD(&net->packet.sklist);
3921
3922         if (!proc_net_fops_create(net, "packet", 0, &packet_seq_fops))
3923                 return -ENOMEM;
3924
3925         return 0;
3926 }
3927
3928 static void __net_exit packet_net_exit(struct net *net)
3929 {
3930         proc_net_remove(net, "packet");
3931 }
3932
3933 static struct pernet_operations packet_net_ops = {
3934         .init = packet_net_init,
3935         .exit = packet_net_exit,
3936 };
3937
3938
3939 static void __exit packet_exit(void)
3940 {
3941         unregister_netdevice_notifier(&packet_netdev_notifier);
3942         unregister_pernet_subsys(&packet_net_ops);
3943         sock_unregister(PF_PACKET);
3944         proto_unregister(&packet_proto);
3945 }
3946
3947 static int __init packet_init(void)
3948 {
3949         int rc = proto_register(&packet_proto, 0);
3950
3951         if (rc != 0)
3952                 goto out;
3953
3954         sock_register(&packet_family_ops);
3955         register_pernet_subsys(&packet_net_ops);
3956         register_netdevice_notifier(&packet_netdev_notifier);
3957 out:
3958         return rc;
3959 }
3960
3961 module_init(packet_init);
3962 module_exit(packet_exit);
3963 MODULE_LICENSE("GPL");
3964 MODULE_ALIAS_NETPROTO(PF_PACKET);