]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - net/packet/af_packet.c
82a6f34d39d012fb35d9a0d490503fcc2048e6e2
[~shefty/rdma-dev.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
1463         /*
1464          *      Get and verify the address.
1465          */
1466
1467         if (saddr) {
1468                 if (msg->msg_namelen < sizeof(struct sockaddr))
1469                         return -EINVAL;
1470                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1471                         proto = saddr->spkt_protocol;
1472         } else
1473                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1474
1475         /*
1476          *      Find the device first to size check it
1477          */
1478
1479         saddr->spkt_device[13] = 0;
1480 retry:
1481         rcu_read_lock();
1482         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1483         err = -ENODEV;
1484         if (dev == NULL)
1485                 goto out_unlock;
1486
1487         err = -ENETDOWN;
1488         if (!(dev->flags & IFF_UP))
1489                 goto out_unlock;
1490
1491         /*
1492          * You may not queue a frame bigger than the mtu. This is the lowest level
1493          * raw protocol and you must do your own fragmentation at this level.
1494          */
1495
1496         err = -EMSGSIZE;
1497         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN)
1498                 goto out_unlock;
1499
1500         if (!skb) {
1501                 size_t reserved = LL_RESERVED_SPACE(dev);
1502                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1503
1504                 rcu_read_unlock();
1505                 skb = sock_wmalloc(sk, len + reserved, 0, GFP_KERNEL);
1506                 if (skb == NULL)
1507                         return -ENOBUFS;
1508                 /* FIXME: Save some space for broken drivers that write a hard
1509                  * header at transmission time by themselves. PPP is the notable
1510                  * one here. This should really be fixed at the driver level.
1511                  */
1512                 skb_reserve(skb, reserved);
1513                 skb_reset_network_header(skb);
1514
1515                 /* Try to align data part correctly */
1516                 if (hhlen) {
1517                         skb->data -= hhlen;
1518                         skb->tail -= hhlen;
1519                         if (len < hhlen)
1520                                 skb_reset_network_header(skb);
1521                 }
1522                 err = memcpy_fromiovec(skb_put(skb, len), msg->msg_iov, len);
1523                 if (err)
1524                         goto out_free;
1525                 goto retry;
1526         }
1527
1528         if (len > (dev->mtu + dev->hard_header_len)) {
1529                 /* Earlier code assumed this would be a VLAN pkt,
1530                  * double-check this now that we have the actual
1531                  * packet in hand.
1532                  */
1533                 struct ethhdr *ehdr;
1534                 skb_reset_mac_header(skb);
1535                 ehdr = eth_hdr(skb);
1536                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1537                         err = -EMSGSIZE;
1538                         goto out_unlock;
1539                 }
1540         }
1541
1542         skb->protocol = proto;
1543         skb->dev = dev;
1544         skb->priority = sk->sk_priority;
1545         skb->mark = sk->sk_mark;
1546         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1547         if (err < 0)
1548                 goto out_unlock;
1549
1550         dev_queue_xmit(skb);
1551         rcu_read_unlock();
1552         return len;
1553
1554 out_unlock:
1555         rcu_read_unlock();
1556 out_free:
1557         kfree_skb(skb);
1558         return err;
1559 }
1560
1561 static unsigned int run_filter(const struct sk_buff *skb,
1562                                       const struct sock *sk,
1563                                       unsigned int res)
1564 {
1565         struct sk_filter *filter;
1566
1567         rcu_read_lock();
1568         filter = rcu_dereference(sk->sk_filter);
1569         if (filter != NULL)
1570                 res = SK_RUN_FILTER(filter, skb);
1571         rcu_read_unlock();
1572
1573         return res;
1574 }
1575
1576 /*
1577  * This function makes lazy skb cloning in hope that most of packets
1578  * are discarded by BPF.
1579  *
1580  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1581  * and skb->cb are mangled. It works because (and until) packets
1582  * falling here are owned by current CPU. Output packets are cloned
1583  * by dev_queue_xmit_nit(), input packets are processed by net_bh
1584  * sequencially, so that if we return skb to original state on exit,
1585  * we will not harm anyone.
1586  */
1587
1588 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1589                       struct packet_type *pt, struct net_device *orig_dev)
1590 {
1591         struct sock *sk;
1592         struct sockaddr_ll *sll;
1593         struct packet_sock *po;
1594         u8 *skb_head = skb->data;
1595         int skb_len = skb->len;
1596         unsigned int snaplen, res;
1597
1598         if (skb->pkt_type == PACKET_LOOPBACK)
1599                 goto drop;
1600
1601         sk = pt->af_packet_priv;
1602         po = pkt_sk(sk);
1603
1604         if (!net_eq(dev_net(dev), sock_net(sk)))
1605                 goto drop;
1606
1607         skb->dev = dev;
1608
1609         if (dev->header_ops) {
1610                 /* The device has an explicit notion of ll header,
1611                  * exported to higher levels.
1612                  *
1613                  * Otherwise, the device hides details of its frame
1614                  * structure, so that corresponding packet head is
1615                  * never delivered to user.
1616                  */
1617                 if (sk->sk_type != SOCK_DGRAM)
1618                         skb_push(skb, skb->data - skb_mac_header(skb));
1619                 else if (skb->pkt_type == PACKET_OUTGOING) {
1620                         /* Special case: outgoing packets have ll header at head */
1621                         skb_pull(skb, skb_network_offset(skb));
1622                 }
1623         }
1624
1625         snaplen = skb->len;
1626
1627         res = run_filter(skb, sk, snaplen);
1628         if (!res)
1629                 goto drop_n_restore;
1630         if (snaplen > res)
1631                 snaplen = res;
1632
1633         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1634             (unsigned)sk->sk_rcvbuf)
1635                 goto drop_n_acct;
1636
1637         if (skb_shared(skb)) {
1638                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1639                 if (nskb == NULL)
1640                         goto drop_n_acct;
1641
1642                 if (skb_head != skb->data) {
1643                         skb->data = skb_head;
1644                         skb->len = skb_len;
1645                 }
1646                 kfree_skb(skb);
1647                 skb = nskb;
1648         }
1649
1650         BUILD_BUG_ON(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8 >
1651                      sizeof(skb->cb));
1652
1653         sll = &PACKET_SKB_CB(skb)->sa.ll;
1654         sll->sll_family = AF_PACKET;
1655         sll->sll_hatype = dev->type;
1656         sll->sll_protocol = skb->protocol;
1657         sll->sll_pkttype = skb->pkt_type;
1658         if (unlikely(po->origdev))
1659                 sll->sll_ifindex = orig_dev->ifindex;
1660         else
1661                 sll->sll_ifindex = dev->ifindex;
1662
1663         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1664
1665         PACKET_SKB_CB(skb)->origlen = skb->len;
1666
1667         if (pskb_trim(skb, snaplen))
1668                 goto drop_n_acct;
1669
1670         skb_set_owner_r(skb, sk);
1671         skb->dev = NULL;
1672         skb_dst_drop(skb);
1673
1674         /* drop conntrack reference */
1675         nf_reset(skb);
1676
1677         spin_lock(&sk->sk_receive_queue.lock);
1678         po->stats.tp_packets++;
1679         skb->dropcount = atomic_read(&sk->sk_drops);
1680         __skb_queue_tail(&sk->sk_receive_queue, skb);
1681         spin_unlock(&sk->sk_receive_queue.lock);
1682         sk->sk_data_ready(sk, skb->len);
1683         return 0;
1684
1685 drop_n_acct:
1686         spin_lock(&sk->sk_receive_queue.lock);
1687         po->stats.tp_drops++;
1688         atomic_inc(&sk->sk_drops);
1689         spin_unlock(&sk->sk_receive_queue.lock);
1690
1691 drop_n_restore:
1692         if (skb_head != skb->data && skb_shared(skb)) {
1693                 skb->data = skb_head;
1694                 skb->len = skb_len;
1695         }
1696 drop:
1697         consume_skb(skb);
1698         return 0;
1699 }
1700
1701 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1702                        struct packet_type *pt, struct net_device *orig_dev)
1703 {
1704         struct sock *sk;
1705         struct packet_sock *po;
1706         struct sockaddr_ll *sll;
1707         union {
1708                 struct tpacket_hdr *h1;
1709                 struct tpacket2_hdr *h2;
1710                 struct tpacket3_hdr *h3;
1711                 void *raw;
1712         } h;
1713         u8 *skb_head = skb->data;
1714         int skb_len = skb->len;
1715         unsigned int snaplen, res;
1716         unsigned long status = TP_STATUS_USER;
1717         unsigned short macoff, netoff, hdrlen;
1718         struct sk_buff *copy_skb = NULL;
1719         struct timeval tv;
1720         struct timespec ts;
1721         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
1722
1723         if (skb->pkt_type == PACKET_LOOPBACK)
1724                 goto drop;
1725
1726         sk = pt->af_packet_priv;
1727         po = pkt_sk(sk);
1728
1729         if (!net_eq(dev_net(dev), sock_net(sk)))
1730                 goto drop;
1731
1732         if (dev->header_ops) {
1733                 if (sk->sk_type != SOCK_DGRAM)
1734                         skb_push(skb, skb->data - skb_mac_header(skb));
1735                 else if (skb->pkt_type == PACKET_OUTGOING) {
1736                         /* Special case: outgoing packets have ll header at head */
1737                         skb_pull(skb, skb_network_offset(skb));
1738                 }
1739         }
1740
1741         if (skb->ip_summed == CHECKSUM_PARTIAL)
1742                 status |= TP_STATUS_CSUMNOTREADY;
1743
1744         snaplen = skb->len;
1745
1746         res = run_filter(skb, sk, snaplen);
1747         if (!res)
1748                 goto drop_n_restore;
1749         if (snaplen > res)
1750                 snaplen = res;
1751
1752         if (sk->sk_type == SOCK_DGRAM) {
1753                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1754                                   po->tp_reserve;
1755         } else {
1756                 unsigned maclen = skb_network_offset(skb);
1757                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1758                                        (maclen < 16 ? 16 : maclen)) +
1759                         po->tp_reserve;
1760                 macoff = netoff - maclen;
1761         }
1762         if (po->tp_version <= TPACKET_V2) {
1763                 if (macoff + snaplen > po->rx_ring.frame_size) {
1764                         if (po->copy_thresh &&
1765                                 atomic_read(&sk->sk_rmem_alloc) + skb->truesize
1766                                 < (unsigned)sk->sk_rcvbuf) {
1767                                 if (skb_shared(skb)) {
1768                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
1769                                 } else {
1770                                         copy_skb = skb_get(skb);
1771                                         skb_head = skb->data;
1772                                 }
1773                                 if (copy_skb)
1774                                         skb_set_owner_r(copy_skb, sk);
1775                         }
1776                         snaplen = po->rx_ring.frame_size - macoff;
1777                         if ((int)snaplen < 0)
1778                                 snaplen = 0;
1779                 }
1780         }
1781         spin_lock(&sk->sk_receive_queue.lock);
1782         h.raw = packet_current_rx_frame(po, skb,
1783                                         TP_STATUS_KERNEL, (macoff+snaplen));
1784         if (!h.raw)
1785                 goto ring_is_full;
1786         if (po->tp_version <= TPACKET_V2) {
1787                 packet_increment_rx_head(po, &po->rx_ring);
1788         /*
1789          * LOSING will be reported till you read the stats,
1790          * because it's COR - Clear On Read.
1791          * Anyways, moving it for V1/V2 only as V3 doesn't need this
1792          * at packet level.
1793          */
1794                 if (po->stats.tp_drops)
1795                         status |= TP_STATUS_LOSING;
1796         }
1797         po->stats.tp_packets++;
1798         if (copy_skb) {
1799                 status |= TP_STATUS_COPY;
1800                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1801         }
1802         spin_unlock(&sk->sk_receive_queue.lock);
1803
1804         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1805
1806         switch (po->tp_version) {
1807         case TPACKET_V1:
1808                 h.h1->tp_len = skb->len;
1809                 h.h1->tp_snaplen = snaplen;
1810                 h.h1->tp_mac = macoff;
1811                 h.h1->tp_net = netoff;
1812                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1813                                 && shhwtstamps->syststamp.tv64)
1814                         tv = ktime_to_timeval(shhwtstamps->syststamp);
1815                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1816                                 && shhwtstamps->hwtstamp.tv64)
1817                         tv = ktime_to_timeval(shhwtstamps->hwtstamp);
1818                 else if (skb->tstamp.tv64)
1819                         tv = ktime_to_timeval(skb->tstamp);
1820                 else
1821                         do_gettimeofday(&tv);
1822                 h.h1->tp_sec = tv.tv_sec;
1823                 h.h1->tp_usec = tv.tv_usec;
1824                 hdrlen = sizeof(*h.h1);
1825                 break;
1826         case TPACKET_V2:
1827                 h.h2->tp_len = skb->len;
1828                 h.h2->tp_snaplen = snaplen;
1829                 h.h2->tp_mac = macoff;
1830                 h.h2->tp_net = netoff;
1831                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1832                                 && shhwtstamps->syststamp.tv64)
1833                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1834                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1835                                 && shhwtstamps->hwtstamp.tv64)
1836                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1837                 else if (skb->tstamp.tv64)
1838                         ts = ktime_to_timespec(skb->tstamp);
1839                 else
1840                         getnstimeofday(&ts);
1841                 h.h2->tp_sec = ts.tv_sec;
1842                 h.h2->tp_nsec = ts.tv_nsec;
1843                 if (vlan_tx_tag_present(skb)) {
1844                         h.h2->tp_vlan_tci = vlan_tx_tag_get(skb);
1845                         status |= TP_STATUS_VLAN_VALID;
1846                 } else {
1847                         h.h2->tp_vlan_tci = 0;
1848                 }
1849                 h.h2->tp_padding = 0;
1850                 hdrlen = sizeof(*h.h2);
1851                 break;
1852         case TPACKET_V3:
1853                 /* tp_nxt_offset,vlan are already populated above.
1854                  * So DONT clear those fields here
1855                  */
1856                 h.h3->tp_status |= status;
1857                 h.h3->tp_len = skb->len;
1858                 h.h3->tp_snaplen = snaplen;
1859                 h.h3->tp_mac = macoff;
1860                 h.h3->tp_net = netoff;
1861                 if ((po->tp_tstamp & SOF_TIMESTAMPING_SYS_HARDWARE)
1862                                 && shhwtstamps->syststamp.tv64)
1863                         ts = ktime_to_timespec(shhwtstamps->syststamp);
1864                 else if ((po->tp_tstamp & SOF_TIMESTAMPING_RAW_HARDWARE)
1865                                 && shhwtstamps->hwtstamp.tv64)
1866                         ts = ktime_to_timespec(shhwtstamps->hwtstamp);
1867                 else if (skb->tstamp.tv64)
1868                         ts = ktime_to_timespec(skb->tstamp);
1869                 else
1870                         getnstimeofday(&ts);
1871                 h.h3->tp_sec  = ts.tv_sec;
1872                 h.h3->tp_nsec = ts.tv_nsec;
1873                 hdrlen = sizeof(*h.h3);
1874                 break;
1875         default:
1876                 BUG();
1877         }
1878
1879         sll = h.raw + TPACKET_ALIGN(hdrlen);
1880         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1881         sll->sll_family = AF_PACKET;
1882         sll->sll_hatype = dev->type;
1883         sll->sll_protocol = skb->protocol;
1884         sll->sll_pkttype = skb->pkt_type;
1885         if (unlikely(po->origdev))
1886                 sll->sll_ifindex = orig_dev->ifindex;
1887         else
1888                 sll->sll_ifindex = dev->ifindex;
1889
1890         smp_mb();
1891 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
1892         {
1893                 u8 *start, *end;
1894
1895                 if (po->tp_version <= TPACKET_V2) {
1896                         end = (u8 *)PAGE_ALIGN((unsigned long)h.raw
1897                                 + macoff + snaplen);
1898                         for (start = h.raw; start < end; start += PAGE_SIZE)
1899                                 flush_dcache_page(pgv_to_page(start));
1900                 }
1901                 smp_wmb();
1902         }
1903 #endif
1904         if (po->tp_version <= TPACKET_V2)
1905                 __packet_set_status(po, h.raw, status);
1906         else
1907                 prb_clear_blk_fill_status(&po->rx_ring);
1908
1909         sk->sk_data_ready(sk, 0);
1910
1911 drop_n_restore:
1912         if (skb_head != skb->data && skb_shared(skb)) {
1913                 skb->data = skb_head;
1914                 skb->len = skb_len;
1915         }
1916 drop:
1917         kfree_skb(skb);
1918         return 0;
1919
1920 ring_is_full:
1921         po->stats.tp_drops++;
1922         spin_unlock(&sk->sk_receive_queue.lock);
1923
1924         sk->sk_data_ready(sk, 0);
1925         kfree_skb(copy_skb);
1926         goto drop_n_restore;
1927 }
1928
1929 static void tpacket_destruct_skb(struct sk_buff *skb)
1930 {
1931         struct packet_sock *po = pkt_sk(skb->sk);
1932         void *ph;
1933
1934         if (likely(po->tx_ring.pg_vec)) {
1935                 ph = skb_shinfo(skb)->destructor_arg;
1936                 BUG_ON(__packet_get_status(po, ph) != TP_STATUS_SENDING);
1937                 BUG_ON(atomic_read(&po->tx_ring.pending) == 0);
1938                 atomic_dec(&po->tx_ring.pending);
1939                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE);
1940         }
1941
1942         sock_wfree(skb);
1943 }
1944
1945 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
1946                 void *frame, struct net_device *dev, int size_max,
1947                 __be16 proto, unsigned char *addr)
1948 {
1949         union {
1950                 struct tpacket_hdr *h1;
1951                 struct tpacket2_hdr *h2;
1952                 void *raw;
1953         } ph;
1954         int to_write, offset, len, tp_len, nr_frags, len_max;
1955         struct socket *sock = po->sk.sk_socket;
1956         struct page *page;
1957         void *data;
1958         int err;
1959
1960         ph.raw = frame;
1961
1962         skb->protocol = proto;
1963         skb->dev = dev;
1964         skb->priority = po->sk.sk_priority;
1965         skb->mark = po->sk.sk_mark;
1966         skb_shinfo(skb)->destructor_arg = ph.raw;
1967
1968         switch (po->tp_version) {
1969         case TPACKET_V2:
1970                 tp_len = ph.h2->tp_len;
1971                 break;
1972         default:
1973                 tp_len = ph.h1->tp_len;
1974                 break;
1975         }
1976         if (unlikely(tp_len > size_max)) {
1977                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
1978                 return -EMSGSIZE;
1979         }
1980
1981         skb_reserve(skb, LL_RESERVED_SPACE(dev));
1982         skb_reset_network_header(skb);
1983
1984         data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
1985         to_write = tp_len;
1986
1987         if (sock->type == SOCK_DGRAM) {
1988                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
1989                                 NULL, tp_len);
1990                 if (unlikely(err < 0))
1991                         return -EINVAL;
1992         } else if (dev->hard_header_len) {
1993                 /* net device doesn't like empty head */
1994                 if (unlikely(tp_len <= dev->hard_header_len)) {
1995                         pr_err("packet size is too short (%d < %d)\n",
1996                                tp_len, dev->hard_header_len);
1997                         return -EINVAL;
1998                 }
1999
2000                 skb_push(skb, dev->hard_header_len);
2001                 err = skb_store_bits(skb, 0, data,
2002                                 dev->hard_header_len);
2003                 if (unlikely(err))
2004                         return err;
2005
2006                 data += dev->hard_header_len;
2007                 to_write -= dev->hard_header_len;
2008         }
2009
2010         err = -EFAULT;
2011         offset = offset_in_page(data);
2012         len_max = PAGE_SIZE - offset;
2013         len = ((to_write > len_max) ? len_max : to_write);
2014
2015         skb->data_len = to_write;
2016         skb->len += to_write;
2017         skb->truesize += to_write;
2018         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2019
2020         while (likely(to_write)) {
2021                 nr_frags = skb_shinfo(skb)->nr_frags;
2022
2023                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2024                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2025                                MAX_SKB_FRAGS);
2026                         return -EFAULT;
2027                 }
2028
2029                 page = pgv_to_page(data);
2030                 data += len;
2031                 flush_dcache_page(page);
2032                 get_page(page);
2033                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2034                 to_write -= len;
2035                 offset = 0;
2036                 len_max = PAGE_SIZE;
2037                 len = ((to_write > len_max) ? len_max : to_write);
2038         }
2039
2040         return tp_len;
2041 }
2042
2043 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2044 {
2045         struct sk_buff *skb;
2046         struct net_device *dev;
2047         __be16 proto;
2048         bool need_rls_dev = false;
2049         int err, reserve = 0;
2050         void *ph;
2051         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2052         int tp_len, size_max;
2053         unsigned char *addr;
2054         int len_sum = 0;
2055         int status = 0;
2056
2057         mutex_lock(&po->pg_vec_lock);
2058
2059         err = -EBUSY;
2060         if (saddr == NULL) {
2061                 dev = po->prot_hook.dev;
2062                 proto   = po->num;
2063                 addr    = NULL;
2064         } else {
2065                 err = -EINVAL;
2066                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2067                         goto out;
2068                 if (msg->msg_namelen < (saddr->sll_halen
2069                                         + offsetof(struct sockaddr_ll,
2070                                                 sll_addr)))
2071                         goto out;
2072                 proto   = saddr->sll_protocol;
2073                 addr    = saddr->sll_addr;
2074                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2075                 need_rls_dev = true;
2076         }
2077
2078         err = -ENXIO;
2079         if (unlikely(dev == NULL))
2080                 goto out;
2081
2082         reserve = dev->hard_header_len;
2083
2084         err = -ENETDOWN;
2085         if (unlikely(!(dev->flags & IFF_UP)))
2086                 goto out_put;
2087
2088         size_max = po->tx_ring.frame_size
2089                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2090
2091         if (size_max > dev->mtu + reserve)
2092                 size_max = dev->mtu + reserve;
2093
2094         do {
2095                 ph = packet_current_frame(po, &po->tx_ring,
2096                                 TP_STATUS_SEND_REQUEST);
2097
2098                 if (unlikely(ph == NULL)) {
2099                         schedule();
2100                         continue;
2101                 }
2102
2103                 status = TP_STATUS_SEND_REQUEST;
2104                 skb = sock_alloc_send_skb(&po->sk,
2105                                 LL_ALLOCATED_SPACE(dev)
2106                                 + sizeof(struct sockaddr_ll),
2107                                 0, &err);
2108
2109                 if (unlikely(skb == NULL))
2110                         goto out_status;
2111
2112                 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2113                                 addr);
2114
2115                 if (unlikely(tp_len < 0)) {
2116                         if (po->tp_loss) {
2117                                 __packet_set_status(po, ph,
2118                                                 TP_STATUS_AVAILABLE);
2119                                 packet_increment_head(&po->tx_ring);
2120                                 kfree_skb(skb);
2121                                 continue;
2122                         } else {
2123                                 status = TP_STATUS_WRONG_FORMAT;
2124                                 err = tp_len;
2125                                 goto out_status;
2126                         }
2127                 }
2128
2129                 skb->destructor = tpacket_destruct_skb;
2130                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2131                 atomic_inc(&po->tx_ring.pending);
2132
2133                 status = TP_STATUS_SEND_REQUEST;
2134                 err = dev_queue_xmit(skb);
2135                 if (unlikely(err > 0)) {
2136                         err = net_xmit_errno(err);
2137                         if (err && __packet_get_status(po, ph) ==
2138                                    TP_STATUS_AVAILABLE) {
2139                                 /* skb was destructed already */
2140                                 skb = NULL;
2141                                 goto out_status;
2142                         }
2143                         /*
2144                          * skb was dropped but not destructed yet;
2145                          * let's treat it like congestion or err < 0
2146                          */
2147                         err = 0;
2148                 }
2149                 packet_increment_head(&po->tx_ring);
2150                 len_sum += tp_len;
2151         } while (likely((ph != NULL) ||
2152                         ((!(msg->msg_flags & MSG_DONTWAIT)) &&
2153                          (atomic_read(&po->tx_ring.pending))))
2154                 );
2155
2156         err = len_sum;
2157         goto out_put;
2158
2159 out_status:
2160         __packet_set_status(po, ph, status);
2161         kfree_skb(skb);
2162 out_put:
2163         if (need_rls_dev)
2164                 dev_put(dev);
2165 out:
2166         mutex_unlock(&po->pg_vec_lock);
2167         return err;
2168 }
2169
2170 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2171                                         size_t reserve, size_t len,
2172                                         size_t linear, int noblock,
2173                                         int *err)
2174 {
2175         struct sk_buff *skb;
2176
2177         /* Under a page?  Don't bother with paged skb. */
2178         if (prepad + len < PAGE_SIZE || !linear)
2179                 linear = len;
2180
2181         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2182                                    err);
2183         if (!skb)
2184                 return NULL;
2185
2186         skb_reserve(skb, reserve);
2187         skb_put(skb, linear);
2188         skb->data_len = len - linear;
2189         skb->len += len - linear;
2190
2191         return skb;
2192 }
2193
2194 static int packet_snd(struct socket *sock,
2195                           struct msghdr *msg, size_t len)
2196 {
2197         struct sock *sk = sock->sk;
2198         struct sockaddr_ll *saddr = (struct sockaddr_ll *)msg->msg_name;
2199         struct sk_buff *skb;
2200         struct net_device *dev;
2201         __be16 proto;
2202         bool need_rls_dev = false;
2203         unsigned char *addr;
2204         int err, reserve = 0;
2205         struct virtio_net_hdr vnet_hdr = { 0 };
2206         int offset = 0;
2207         int vnet_hdr_len;
2208         struct packet_sock *po = pkt_sk(sk);
2209         unsigned short gso_type = 0;
2210
2211         /*
2212          *      Get and verify the address.
2213          */
2214
2215         if (saddr == NULL) {
2216                 dev = po->prot_hook.dev;
2217                 proto   = po->num;
2218                 addr    = NULL;
2219         } else {
2220                 err = -EINVAL;
2221                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2222                         goto out;
2223                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2224                         goto out;
2225                 proto   = saddr->sll_protocol;
2226                 addr    = saddr->sll_addr;
2227                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2228                 need_rls_dev = true;
2229         }
2230
2231         err = -ENXIO;
2232         if (dev == NULL)
2233                 goto out_unlock;
2234         if (sock->type == SOCK_RAW)
2235                 reserve = dev->hard_header_len;
2236
2237         err = -ENETDOWN;
2238         if (!(dev->flags & IFF_UP))
2239                 goto out_unlock;
2240
2241         if (po->has_vnet_hdr) {
2242                 vnet_hdr_len = sizeof(vnet_hdr);
2243
2244                 err = -EINVAL;
2245                 if (len < vnet_hdr_len)
2246                         goto out_unlock;
2247
2248                 len -= vnet_hdr_len;
2249
2250                 err = memcpy_fromiovec((void *)&vnet_hdr, msg->msg_iov,
2251                                        vnet_hdr_len);
2252                 if (err < 0)
2253                         goto out_unlock;
2254
2255                 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2256                     (vnet_hdr.csum_start + vnet_hdr.csum_offset + 2 >
2257                       vnet_hdr.hdr_len))
2258                         vnet_hdr.hdr_len = vnet_hdr.csum_start +
2259                                                  vnet_hdr.csum_offset + 2;
2260
2261                 err = -EINVAL;
2262                 if (vnet_hdr.hdr_len > len)
2263                         goto out_unlock;
2264
2265                 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2266                         switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2267                         case VIRTIO_NET_HDR_GSO_TCPV4:
2268                                 gso_type = SKB_GSO_TCPV4;
2269                                 break;
2270                         case VIRTIO_NET_HDR_GSO_TCPV6:
2271                                 gso_type = SKB_GSO_TCPV6;
2272                                 break;
2273                         case VIRTIO_NET_HDR_GSO_UDP:
2274                                 gso_type = SKB_GSO_UDP;
2275                                 break;
2276                         default:
2277                                 goto out_unlock;
2278                         }
2279
2280                         if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2281                                 gso_type |= SKB_GSO_TCP_ECN;
2282
2283                         if (vnet_hdr.gso_size == 0)
2284                                 goto out_unlock;
2285
2286                 }
2287         }
2288
2289         err = -EMSGSIZE;
2290         if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN))
2291                 goto out_unlock;
2292
2293         err = -ENOBUFS;
2294         skb = packet_alloc_skb(sk, LL_ALLOCATED_SPACE(dev),
2295                                LL_RESERVED_SPACE(dev), len, vnet_hdr.hdr_len,
2296                                msg->msg_flags & MSG_DONTWAIT, &err);
2297         if (skb == NULL)
2298                 goto out_unlock;
2299
2300         skb_set_network_header(skb, reserve);
2301
2302         err = -EINVAL;
2303         if (sock->type == SOCK_DGRAM &&
2304             (offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len)) < 0)
2305                 goto out_free;
2306
2307         /* Returns -EFAULT on error */
2308         err = skb_copy_datagram_from_iovec(skb, offset, msg->msg_iov, 0, len);
2309         if (err)
2310                 goto out_free;
2311         err = sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2312         if (err < 0)
2313                 goto out_free;
2314
2315         if (!gso_type && (len > dev->mtu + reserve)) {
2316                 /* Earlier code assumed this would be a VLAN pkt,
2317                  * double-check this now that we have the actual
2318                  * packet in hand.
2319                  */
2320                 struct ethhdr *ehdr;
2321                 skb_reset_mac_header(skb);
2322                 ehdr = eth_hdr(skb);
2323                 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2324                         err = -EMSGSIZE;
2325                         goto out_free;
2326                 }
2327         }
2328
2329         skb->protocol = proto;
2330         skb->dev = dev;
2331         skb->priority = sk->sk_priority;
2332         skb->mark = sk->sk_mark;
2333
2334         if (po->has_vnet_hdr) {
2335                 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2336                         if (!skb_partial_csum_set(skb, vnet_hdr.csum_start,
2337                                                   vnet_hdr.csum_offset)) {
2338                                 err = -EINVAL;
2339                                 goto out_free;
2340                         }
2341                 }
2342
2343                 skb_shinfo(skb)->gso_size = vnet_hdr.gso_size;
2344                 skb_shinfo(skb)->gso_type = gso_type;
2345
2346                 /* Header must be checked, and gso_segs computed. */
2347                 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2348                 skb_shinfo(skb)->gso_segs = 0;
2349
2350                 len += vnet_hdr_len;
2351         }
2352
2353         /*
2354          *      Now send it
2355          */
2356
2357         err = dev_queue_xmit(skb);
2358         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2359                 goto out_unlock;
2360
2361         if (need_rls_dev)
2362                 dev_put(dev);
2363
2364         return len;
2365
2366 out_free:
2367         kfree_skb(skb);
2368 out_unlock:
2369         if (dev && need_rls_dev)
2370                 dev_put(dev);
2371 out:
2372         return err;
2373 }
2374
2375 static int packet_sendmsg(struct kiocb *iocb, struct socket *sock,
2376                 struct msghdr *msg, size_t len)
2377 {
2378         struct sock *sk = sock->sk;
2379         struct packet_sock *po = pkt_sk(sk);
2380         if (po->tx_ring.pg_vec)
2381                 return tpacket_snd(po, msg);
2382         else
2383                 return packet_snd(sock, msg, len);
2384 }
2385
2386 /*
2387  *      Close a PACKET socket. This is fairly simple. We immediately go
2388  *      to 'closed' state and remove our protocol entry in the device list.
2389  */
2390
2391 static int packet_release(struct socket *sock)
2392 {
2393         struct sock *sk = sock->sk;
2394         struct packet_sock *po;
2395         struct net *net;
2396         union tpacket_req_u req_u;
2397
2398         if (!sk)
2399                 return 0;
2400
2401         net = sock_net(sk);
2402         po = pkt_sk(sk);
2403
2404         spin_lock_bh(&net->packet.sklist_lock);
2405         sk_del_node_init_rcu(sk);
2406         sock_prot_inuse_add(net, sk->sk_prot, -1);
2407         spin_unlock_bh(&net->packet.sklist_lock);
2408
2409         spin_lock(&po->bind_lock);
2410         unregister_prot_hook(sk, false);
2411         if (po->prot_hook.dev) {
2412                 dev_put(po->prot_hook.dev);
2413                 po->prot_hook.dev = NULL;
2414         }
2415         spin_unlock(&po->bind_lock);
2416
2417         packet_flush_mclist(sk);
2418
2419         memset(&req_u, 0, sizeof(req_u));
2420
2421         if (po->rx_ring.pg_vec)
2422                 packet_set_ring(sk, &req_u, 1, 0);
2423
2424         if (po->tx_ring.pg_vec)
2425                 packet_set_ring(sk, &req_u, 1, 1);
2426
2427         fanout_release(sk);
2428
2429         synchronize_net();
2430         /*
2431          *      Now the socket is dead. No more input will appear.
2432          */
2433         sock_orphan(sk);
2434         sock->sk = NULL;
2435
2436         /* Purge queues */
2437
2438         skb_queue_purge(&sk->sk_receive_queue);
2439         sk_refcnt_debug_release(sk);
2440
2441         sock_put(sk);
2442         return 0;
2443 }
2444
2445 /*
2446  *      Attach a packet hook.
2447  */
2448
2449 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 protocol)
2450 {
2451         struct packet_sock *po = pkt_sk(sk);
2452
2453         if (po->fanout)
2454                 return -EINVAL;
2455
2456         lock_sock(sk);
2457
2458         spin_lock(&po->bind_lock);
2459         unregister_prot_hook(sk, true);
2460         po->num = protocol;
2461         po->prot_hook.type = protocol;
2462         if (po->prot_hook.dev)
2463                 dev_put(po->prot_hook.dev);
2464         po->prot_hook.dev = dev;
2465
2466         po->ifindex = dev ? dev->ifindex : 0;
2467
2468         if (protocol == 0)
2469                 goto out_unlock;
2470
2471         if (!dev || (dev->flags & IFF_UP)) {
2472                 register_prot_hook(sk);
2473         } else {
2474                 sk->sk_err = ENETDOWN;
2475                 if (!sock_flag(sk, SOCK_DEAD))
2476                         sk->sk_error_report(sk);
2477         }
2478
2479 out_unlock:
2480         spin_unlock(&po->bind_lock);
2481         release_sock(sk);
2482         return 0;
2483 }
2484
2485 /*
2486  *      Bind a packet socket to a device
2487  */
2488
2489 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2490                             int addr_len)
2491 {
2492         struct sock *sk = sock->sk;
2493         char name[15];
2494         struct net_device *dev;
2495         int err = -ENODEV;
2496
2497         /*
2498          *      Check legality
2499          */
2500
2501         if (addr_len != sizeof(struct sockaddr))
2502                 return -EINVAL;
2503         strlcpy(name, uaddr->sa_data, sizeof(name));
2504
2505         dev = dev_get_by_name(sock_net(sk), name);
2506         if (dev)
2507                 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2508         return err;
2509 }
2510
2511 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2512 {
2513         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2514         struct sock *sk = sock->sk;
2515         struct net_device *dev = NULL;
2516         int err;
2517
2518
2519         /*
2520          *      Check legality
2521          */
2522
2523         if (addr_len < sizeof(struct sockaddr_ll))
2524                 return -EINVAL;
2525         if (sll->sll_family != AF_PACKET)
2526                 return -EINVAL;
2527
2528         if (sll->sll_ifindex) {
2529                 err = -ENODEV;
2530                 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2531                 if (dev == NULL)
2532                         goto out;
2533         }
2534         err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2535
2536 out:
2537         return err;
2538 }
2539
2540 static struct proto packet_proto = {
2541         .name     = "PACKET",
2542         .owner    = THIS_MODULE,
2543         .obj_size = sizeof(struct packet_sock),
2544 };
2545
2546 /*
2547  *      Create a packet of type SOCK_PACKET.
2548  */
2549
2550 static int packet_create(struct net *net, struct socket *sock, int protocol,
2551                          int kern)
2552 {
2553         struct sock *sk;
2554         struct packet_sock *po;
2555         __be16 proto = (__force __be16)protocol; /* weird, but documented */
2556         int err;
2557
2558         if (!capable(CAP_NET_RAW))
2559                 return -EPERM;
2560         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2561             sock->type != SOCK_PACKET)
2562                 return -ESOCKTNOSUPPORT;
2563
2564         sock->state = SS_UNCONNECTED;
2565
2566         err = -ENOBUFS;
2567         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2568         if (sk == NULL)
2569                 goto out;
2570
2571         sock->ops = &packet_ops;
2572         if (sock->type == SOCK_PACKET)
2573                 sock->ops = &packet_ops_spkt;
2574
2575         sock_init_data(sock, sk);
2576
2577         po = pkt_sk(sk);
2578         sk->sk_family = PF_PACKET;
2579         po->num = proto;
2580
2581         sk->sk_destruct = packet_sock_destruct;
2582         sk_refcnt_debug_inc(sk);
2583
2584         /*
2585          *      Attach a protocol block
2586          */
2587
2588         spin_lock_init(&po->bind_lock);
2589         mutex_init(&po->pg_vec_lock);
2590         po->prot_hook.func = packet_rcv;
2591
2592         if (sock->type == SOCK_PACKET)
2593                 po->prot_hook.func = packet_rcv_spkt;
2594
2595         po->prot_hook.af_packet_priv = sk;
2596
2597         if (proto) {
2598                 po->prot_hook.type = proto;
2599                 register_prot_hook(sk);
2600         }
2601
2602         spin_lock_bh(&net->packet.sklist_lock);
2603         sk_add_node_rcu(sk, &net->packet.sklist);
2604         sock_prot_inuse_add(net, &packet_proto, 1);
2605         spin_unlock_bh(&net->packet.sklist_lock);
2606
2607         return 0;
2608 out:
2609         return err;
2610 }
2611
2612 static int packet_recv_error(struct sock *sk, struct msghdr *msg, int len)
2613 {
2614         struct sock_exterr_skb *serr;
2615         struct sk_buff *skb, *skb2;
2616         int copied, err;
2617
2618         err = -EAGAIN;
2619         skb = skb_dequeue(&sk->sk_error_queue);
2620         if (skb == NULL)
2621                 goto out;
2622
2623         copied = skb->len;
2624         if (copied > len) {
2625                 msg->msg_flags |= MSG_TRUNC;
2626                 copied = len;
2627         }
2628         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2629         if (err)
2630                 goto out_free_skb;
2631
2632         sock_recv_timestamp(msg, sk, skb);
2633
2634         serr = SKB_EXT_ERR(skb);
2635         put_cmsg(msg, SOL_PACKET, PACKET_TX_TIMESTAMP,
2636                  sizeof(serr->ee), &serr->ee);
2637
2638         msg->msg_flags |= MSG_ERRQUEUE;
2639         err = copied;
2640
2641         /* Reset and regenerate socket error */
2642         spin_lock_bh(&sk->sk_error_queue.lock);
2643         sk->sk_err = 0;
2644         if ((skb2 = skb_peek(&sk->sk_error_queue)) != NULL) {
2645                 sk->sk_err = SKB_EXT_ERR(skb2)->ee.ee_errno;
2646                 spin_unlock_bh(&sk->sk_error_queue.lock);
2647                 sk->sk_error_report(sk);
2648         } else
2649                 spin_unlock_bh(&sk->sk_error_queue.lock);
2650
2651 out_free_skb:
2652         kfree_skb(skb);
2653 out:
2654         return err;
2655 }
2656
2657 /*
2658  *      Pull a packet from our receive queue and hand it to the user.
2659  *      If necessary we block.
2660  */
2661
2662 static int packet_recvmsg(struct kiocb *iocb, struct socket *sock,
2663                           struct msghdr *msg, size_t len, int flags)
2664 {
2665         struct sock *sk = sock->sk;
2666         struct sk_buff *skb;
2667         int copied, err;
2668         struct sockaddr_ll *sll;
2669         int vnet_hdr_len = 0;
2670
2671         err = -EINVAL;
2672         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2673                 goto out;
2674
2675 #if 0
2676         /* What error should we return now? EUNATTACH? */
2677         if (pkt_sk(sk)->ifindex < 0)
2678                 return -ENODEV;
2679 #endif
2680
2681         if (flags & MSG_ERRQUEUE) {
2682                 err = packet_recv_error(sk, msg, len);
2683                 goto out;
2684         }
2685
2686         /*
2687          *      Call the generic datagram receiver. This handles all sorts
2688          *      of horrible races and re-entrancy so we can forget about it
2689          *      in the protocol layers.
2690          *
2691          *      Now it will return ENETDOWN, if device have just gone down,
2692          *      but then it will block.
2693          */
2694
2695         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2696
2697         /*
2698          *      An error occurred so return it. Because skb_recv_datagram()
2699          *      handles the blocking we don't see and worry about blocking
2700          *      retries.
2701          */
2702
2703         if (skb == NULL)
2704                 goto out;
2705
2706         if (pkt_sk(sk)->has_vnet_hdr) {
2707                 struct virtio_net_hdr vnet_hdr = { 0 };
2708
2709                 err = -EINVAL;
2710                 vnet_hdr_len = sizeof(vnet_hdr);
2711                 if (len < vnet_hdr_len)
2712                         goto out_free;
2713
2714                 len -= vnet_hdr_len;
2715
2716                 if (skb_is_gso(skb)) {
2717                         struct skb_shared_info *sinfo = skb_shinfo(skb);
2718
2719                         /* This is a hint as to how much should be linear. */
2720                         vnet_hdr.hdr_len = skb_headlen(skb);
2721                         vnet_hdr.gso_size = sinfo->gso_size;
2722                         if (sinfo->gso_type & SKB_GSO_TCPV4)
2723                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2724                         else if (sinfo->gso_type & SKB_GSO_TCPV6)
2725                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2726                         else if (sinfo->gso_type & SKB_GSO_UDP)
2727                                 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2728                         else if (sinfo->gso_type & SKB_GSO_FCOE)
2729                                 goto out_free;
2730                         else
2731                                 BUG();
2732                         if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2733                                 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2734                 } else
2735                         vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2736
2737                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2738                         vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2739                         vnet_hdr.csum_start = skb_checksum_start_offset(skb);
2740                         vnet_hdr.csum_offset = skb->csum_offset;
2741                 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2742                         vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2743                 } /* else everything is zero */
2744
2745                 err = memcpy_toiovec(msg->msg_iov, (void *)&vnet_hdr,
2746                                      vnet_hdr_len);
2747                 if (err < 0)
2748                         goto out_free;
2749         }
2750
2751         /*
2752          *      If the address length field is there to be filled in, we fill
2753          *      it in now.
2754          */
2755
2756         sll = &PACKET_SKB_CB(skb)->sa.ll;
2757         if (sock->type == SOCK_PACKET)
2758                 msg->msg_namelen = sizeof(struct sockaddr_pkt);
2759         else
2760                 msg->msg_namelen = sll->sll_halen + offsetof(struct sockaddr_ll, sll_addr);
2761
2762         /*
2763          *      You lose any data beyond the buffer you gave. If it worries a
2764          *      user program they can ask the device for its MTU anyway.
2765          */
2766
2767         copied = skb->len;
2768         if (copied > len) {
2769                 copied = len;
2770                 msg->msg_flags |= MSG_TRUNC;
2771         }
2772
2773         err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
2774         if (err)
2775                 goto out_free;
2776
2777         sock_recv_ts_and_drops(msg, sk, skb);
2778
2779         if (msg->msg_name)
2780                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
2781                        msg->msg_namelen);
2782
2783         if (pkt_sk(sk)->auxdata) {
2784                 struct tpacket_auxdata aux;
2785
2786                 aux.tp_status = TP_STATUS_USER;
2787                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2788                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
2789                 aux.tp_len = PACKET_SKB_CB(skb)->origlen;
2790                 aux.tp_snaplen = skb->len;
2791                 aux.tp_mac = 0;
2792                 aux.tp_net = skb_network_offset(skb);
2793                 if (vlan_tx_tag_present(skb)) {
2794                         aux.tp_vlan_tci = vlan_tx_tag_get(skb);
2795                         aux.tp_status |= TP_STATUS_VLAN_VALID;
2796                 } else {
2797                         aux.tp_vlan_tci = 0;
2798                 }
2799                 aux.tp_padding = 0;
2800                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
2801         }
2802
2803         /*
2804          *      Free or return the buffer as appropriate. Again this
2805          *      hides all the races and re-entrancy issues from us.
2806          */
2807         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
2808
2809 out_free:
2810         skb_free_datagram(sk, skb);
2811 out:
2812         return err;
2813 }
2814
2815 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
2816                                int *uaddr_len, int peer)
2817 {
2818         struct net_device *dev;
2819         struct sock *sk = sock->sk;
2820
2821         if (peer)
2822                 return -EOPNOTSUPP;
2823
2824         uaddr->sa_family = AF_PACKET;
2825         rcu_read_lock();
2826         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
2827         if (dev)
2828                 strncpy(uaddr->sa_data, dev->name, 14);
2829         else
2830                 memset(uaddr->sa_data, 0, 14);
2831         rcu_read_unlock();
2832         *uaddr_len = sizeof(*uaddr);
2833
2834         return 0;
2835 }
2836
2837 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
2838                           int *uaddr_len, int peer)
2839 {
2840         struct net_device *dev;
2841         struct sock *sk = sock->sk;
2842         struct packet_sock *po = pkt_sk(sk);
2843         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
2844
2845         if (peer)
2846                 return -EOPNOTSUPP;
2847
2848         sll->sll_family = AF_PACKET;
2849         sll->sll_ifindex = po->ifindex;
2850         sll->sll_protocol = po->num;
2851         sll->sll_pkttype = 0;
2852         rcu_read_lock();
2853         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
2854         if (dev) {
2855                 sll->sll_hatype = dev->type;
2856                 sll->sll_halen = dev->addr_len;
2857                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
2858         } else {
2859                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
2860                 sll->sll_halen = 0;
2861         }
2862         rcu_read_unlock();
2863         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
2864
2865         return 0;
2866 }
2867
2868 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
2869                          int what)
2870 {
2871         switch (i->type) {
2872         case PACKET_MR_MULTICAST:
2873                 if (i->alen != dev->addr_len)
2874                         return -EINVAL;
2875                 if (what > 0)
2876                         return dev_mc_add(dev, i->addr);
2877                 else
2878                         return dev_mc_del(dev, i->addr);
2879                 break;
2880         case PACKET_MR_PROMISC:
2881                 return dev_set_promiscuity(dev, what);
2882                 break;
2883         case PACKET_MR_ALLMULTI:
2884                 return dev_set_allmulti(dev, what);
2885                 break;
2886         case PACKET_MR_UNICAST:
2887                 if (i->alen != dev->addr_len)
2888                         return -EINVAL;
2889                 if (what > 0)
2890                         return dev_uc_add(dev, i->addr);
2891                 else
2892                         return dev_uc_del(dev, i->addr);
2893                 break;
2894         default:
2895                 break;
2896         }
2897         return 0;
2898 }
2899
2900 static void packet_dev_mclist(struct net_device *dev, struct packet_mclist *i, int what)
2901 {
2902         for ( ; i; i = i->next) {
2903                 if (i->ifindex == dev->ifindex)
2904                         packet_dev_mc(dev, i, what);
2905         }
2906 }
2907
2908 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
2909 {
2910         struct packet_sock *po = pkt_sk(sk);
2911         struct packet_mclist *ml, *i;
2912         struct net_device *dev;
2913         int err;
2914
2915         rtnl_lock();
2916
2917         err = -ENODEV;
2918         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
2919         if (!dev)
2920                 goto done;
2921
2922         err = -EINVAL;
2923         if (mreq->mr_alen > dev->addr_len)
2924                 goto done;
2925
2926         err = -ENOBUFS;
2927         i = kmalloc(sizeof(*i), GFP_KERNEL);
2928         if (i == NULL)
2929                 goto done;
2930
2931         err = 0;
2932         for (ml = po->mclist; ml; ml = ml->next) {
2933                 if (ml->ifindex == mreq->mr_ifindex &&
2934                     ml->type == mreq->mr_type &&
2935                     ml->alen == mreq->mr_alen &&
2936                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2937                         ml->count++;
2938                         /* Free the new element ... */
2939                         kfree(i);
2940                         goto done;
2941                 }
2942         }
2943
2944         i->type = mreq->mr_type;
2945         i->ifindex = mreq->mr_ifindex;
2946         i->alen = mreq->mr_alen;
2947         memcpy(i->addr, mreq->mr_address, i->alen);
2948         i->count = 1;
2949         i->next = po->mclist;
2950         po->mclist = i;
2951         err = packet_dev_mc(dev, i, 1);
2952         if (err) {
2953                 po->mclist = i->next;
2954                 kfree(i);
2955         }
2956
2957 done:
2958         rtnl_unlock();
2959         return err;
2960 }
2961
2962 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
2963 {
2964         struct packet_mclist *ml, **mlp;
2965
2966         rtnl_lock();
2967
2968         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
2969                 if (ml->ifindex == mreq->mr_ifindex &&
2970                     ml->type == mreq->mr_type &&
2971                     ml->alen == mreq->mr_alen &&
2972                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
2973                         if (--ml->count == 0) {
2974                                 struct net_device *dev;
2975                                 *mlp = ml->next;
2976                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
2977                                 if (dev)
2978                                         packet_dev_mc(dev, ml, -1);
2979                                 kfree(ml);
2980                         }
2981                         rtnl_unlock();
2982                         return 0;
2983                 }
2984         }
2985         rtnl_unlock();
2986         return -EADDRNOTAVAIL;
2987 }
2988
2989 static void packet_flush_mclist(struct sock *sk)
2990 {
2991         struct packet_sock *po = pkt_sk(sk);
2992         struct packet_mclist *ml;
2993
2994         if (!po->mclist)
2995                 return;
2996
2997         rtnl_lock();
2998         while ((ml = po->mclist) != NULL) {
2999                 struct net_device *dev;
3000
3001                 po->mclist = ml->next;
3002                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3003                 if (dev != NULL)
3004                         packet_dev_mc(dev, ml, -1);
3005                 kfree(ml);
3006         }
3007         rtnl_unlock();
3008 }
3009
3010 static int
3011 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3012 {
3013         struct sock *sk = sock->sk;
3014         struct packet_sock *po = pkt_sk(sk);
3015         int ret;
3016
3017         if (level != SOL_PACKET)
3018                 return -ENOPROTOOPT;
3019
3020         switch (optname) {
3021         case PACKET_ADD_MEMBERSHIP:
3022         case PACKET_DROP_MEMBERSHIP:
3023         {
3024                 struct packet_mreq_max mreq;
3025                 int len = optlen;
3026                 memset(&mreq, 0, sizeof(mreq));
3027                 if (len < sizeof(struct packet_mreq))
3028                         return -EINVAL;
3029                 if (len > sizeof(mreq))
3030                         len = sizeof(mreq);
3031                 if (copy_from_user(&mreq, optval, len))
3032                         return -EFAULT;
3033                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3034                         return -EINVAL;
3035                 if (optname == PACKET_ADD_MEMBERSHIP)
3036                         ret = packet_mc_add(sk, &mreq);
3037                 else
3038                         ret = packet_mc_drop(sk, &mreq);
3039                 return ret;
3040         }
3041
3042         case PACKET_RX_RING:
3043         case PACKET_TX_RING:
3044         {
3045                 union tpacket_req_u req_u;
3046                 int len;
3047
3048                 switch (po->tp_version) {
3049                 case TPACKET_V1:
3050                 case TPACKET_V2:
3051                         len = sizeof(req_u.req);
3052                         break;
3053                 case TPACKET_V3:
3054                 default:
3055                         len = sizeof(req_u.req3);
3056                         break;
3057                 }
3058                 if (optlen < len)
3059                         return -EINVAL;
3060                 if (pkt_sk(sk)->has_vnet_hdr)
3061                         return -EINVAL;
3062                 if (copy_from_user(&req_u.req, optval, len))
3063                         return -EFAULT;
3064                 return packet_set_ring(sk, &req_u, 0,
3065                         optname == PACKET_TX_RING);
3066         }
3067         case PACKET_COPY_THRESH:
3068         {
3069                 int val;
3070
3071                 if (optlen != sizeof(val))
3072                         return -EINVAL;
3073                 if (copy_from_user(&val, optval, sizeof(val)))
3074                         return -EFAULT;
3075
3076                 pkt_sk(sk)->copy_thresh = val;
3077                 return 0;
3078         }
3079         case PACKET_VERSION:
3080         {
3081                 int val;
3082
3083                 if (optlen != sizeof(val))
3084                         return -EINVAL;
3085                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3086                         return -EBUSY;
3087                 if (copy_from_user(&val, optval, sizeof(val)))
3088                         return -EFAULT;
3089                 switch (val) {
3090                 case TPACKET_V1:
3091                 case TPACKET_V2:
3092                 case TPACKET_V3:
3093                         po->tp_version = val;
3094                         return 0;
3095                 default:
3096                         return -EINVAL;
3097                 }
3098         }
3099         case PACKET_RESERVE:
3100         {
3101                 unsigned int val;
3102
3103                 if (optlen != sizeof(val))
3104                         return -EINVAL;
3105                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3106                         return -EBUSY;
3107                 if (copy_from_user(&val, optval, sizeof(val)))
3108                         return -EFAULT;
3109                 po->tp_reserve = val;
3110                 return 0;
3111         }
3112         case PACKET_LOSS:
3113         {
3114                 unsigned int val;
3115
3116                 if (optlen != sizeof(val))
3117                         return -EINVAL;
3118                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3119                         return -EBUSY;
3120                 if (copy_from_user(&val, optval, sizeof(val)))
3121                         return -EFAULT;
3122                 po->tp_loss = !!val;
3123                 return 0;
3124         }
3125         case PACKET_AUXDATA:
3126         {
3127                 int val;
3128
3129                 if (optlen < sizeof(val))
3130                         return -EINVAL;
3131                 if (copy_from_user(&val, optval, sizeof(val)))
3132                         return -EFAULT;
3133
3134                 po->auxdata = !!val;
3135                 return 0;
3136         }
3137         case PACKET_ORIGDEV:
3138         {
3139                 int val;
3140
3141                 if (optlen < sizeof(val))
3142                         return -EINVAL;
3143                 if (copy_from_user(&val, optval, sizeof(val)))
3144                         return -EFAULT;
3145
3146                 po->origdev = !!val;
3147                 return 0;
3148         }
3149         case PACKET_VNET_HDR:
3150         {
3151                 int val;
3152
3153                 if (sock->type != SOCK_RAW)
3154                         return -EINVAL;
3155                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3156                         return -EBUSY;
3157                 if (optlen < sizeof(val))
3158                         return -EINVAL;
3159                 if (copy_from_user(&val, optval, sizeof(val)))
3160                         return -EFAULT;
3161
3162                 po->has_vnet_hdr = !!val;
3163                 return 0;
3164         }
3165         case PACKET_TIMESTAMP:
3166         {
3167                 int val;
3168
3169                 if (optlen != sizeof(val))
3170                         return -EINVAL;
3171                 if (copy_from_user(&val, optval, sizeof(val)))
3172                         return -EFAULT;
3173
3174                 po->tp_tstamp = val;
3175                 return 0;
3176         }
3177         case PACKET_FANOUT:
3178         {
3179                 int val;
3180
3181                 if (optlen != sizeof(val))
3182                         return -EINVAL;
3183                 if (copy_from_user(&val, optval, sizeof(val)))
3184                         return -EFAULT;
3185
3186                 return fanout_add(sk, val & 0xffff, val >> 16);
3187         }
3188         default:
3189                 return -ENOPROTOOPT;
3190         }
3191 }
3192
3193 static int packet_getsockopt(struct socket *sock, int level, int optname,
3194                              char __user *optval, int __user *optlen)
3195 {
3196         int len;
3197         int val;
3198         struct sock *sk = sock->sk;
3199         struct packet_sock *po = pkt_sk(sk);
3200         void *data;
3201         struct tpacket_stats st;
3202         union tpacket_stats_u st_u;
3203
3204         if (level != SOL_PACKET)
3205                 return -ENOPROTOOPT;
3206
3207         if (get_user(len, optlen))
3208                 return -EFAULT;
3209
3210         if (len < 0)
3211                 return -EINVAL;
3212
3213         switch (optname) {
3214         case PACKET_STATISTICS:
3215                 if (po->tp_version == TPACKET_V3) {
3216                         len = sizeof(struct tpacket_stats_v3);
3217                 } else {
3218                         if (len > sizeof(struct tpacket_stats))
3219                                 len = sizeof(struct tpacket_stats);
3220                 }
3221                 spin_lock_bh(&sk->sk_receive_queue.lock);
3222                 if (po->tp_version == TPACKET_V3) {
3223                         memcpy(&st_u.stats3, &po->stats,
3224                         sizeof(struct tpacket_stats));
3225                         st_u.stats3.tp_freeze_q_cnt =
3226                         po->stats_u.stats3.tp_freeze_q_cnt;
3227                         st_u.stats3.tp_packets += po->stats.tp_drops;
3228                         data = &st_u.stats3;
3229                 } else {
3230                         st = po->stats;
3231                         st.tp_packets += st.tp_drops;
3232                         data = &st;
3233                 }
3234                 memset(&po->stats, 0, sizeof(st));
3235                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3236                 break;
3237         case PACKET_AUXDATA:
3238                 if (len > sizeof(int))
3239                         len = sizeof(int);
3240                 val = po->auxdata;
3241
3242                 data = &val;
3243                 break;
3244         case PACKET_ORIGDEV:
3245                 if (len > sizeof(int))
3246                         len = sizeof(int);
3247                 val = po->origdev;
3248
3249                 data = &val;
3250                 break;
3251         case PACKET_VNET_HDR:
3252                 if (len > sizeof(int))
3253                         len = sizeof(int);
3254                 val = po->has_vnet_hdr;
3255
3256                 data = &val;
3257                 break;
3258         case PACKET_VERSION:
3259                 if (len > sizeof(int))
3260                         len = sizeof(int);
3261                 val = po->tp_version;
3262                 data = &val;
3263                 break;
3264         case PACKET_HDRLEN:
3265                 if (len > sizeof(int))
3266                         len = sizeof(int);
3267                 if (copy_from_user(&val, optval, len))
3268                         return -EFAULT;
3269                 switch (val) {
3270                 case TPACKET_V1:
3271                         val = sizeof(struct tpacket_hdr);
3272                         break;
3273                 case TPACKET_V2:
3274                         val = sizeof(struct tpacket2_hdr);
3275                         break;
3276                 case TPACKET_V3:
3277                         val = sizeof(struct tpacket3_hdr);
3278                         break;
3279                 default:
3280                         return -EINVAL;
3281                 }
3282                 data = &val;
3283                 break;
3284         case PACKET_RESERVE:
3285                 if (len > sizeof(unsigned int))
3286                         len = sizeof(unsigned int);
3287                 val = po->tp_reserve;
3288                 data = &val;
3289                 break;
3290         case PACKET_LOSS:
3291                 if (len > sizeof(unsigned int))
3292                         len = sizeof(unsigned int);
3293                 val = po->tp_loss;
3294                 data = &val;
3295                 break;
3296         case PACKET_TIMESTAMP:
3297                 if (len > sizeof(int))
3298                         len = sizeof(int);
3299                 val = po->tp_tstamp;
3300                 data = &val;
3301                 break;
3302         case PACKET_FANOUT:
3303                 if (len > sizeof(int))
3304                         len = sizeof(int);
3305                 val = (po->fanout ?
3306                        ((u32)po->fanout->id |
3307                         ((u32)po->fanout->type << 16)) :
3308                        0);
3309                 data = &val;
3310                 break;
3311         default:
3312                 return -ENOPROTOOPT;
3313         }
3314
3315         if (put_user(len, optlen))
3316                 return -EFAULT;
3317         if (copy_to_user(optval, data, len))
3318                 return -EFAULT;
3319         return 0;
3320 }
3321
3322
3323 static int packet_notifier(struct notifier_block *this, unsigned long msg, void *data)
3324 {
3325         struct sock *sk;
3326         struct hlist_node *node;
3327         struct net_device *dev = data;
3328         struct net *net = dev_net(dev);
3329
3330         rcu_read_lock();
3331         sk_for_each_rcu(sk, node, &net->packet.sklist) {
3332                 struct packet_sock *po = pkt_sk(sk);
3333
3334                 switch (msg) {
3335                 case NETDEV_UNREGISTER:
3336                         if (po->mclist)
3337                                 packet_dev_mclist(dev, po->mclist, -1);
3338                         /* fallthrough */
3339
3340                 case NETDEV_DOWN:
3341                         if (dev->ifindex == po->ifindex) {
3342                                 spin_lock(&po->bind_lock);
3343                                 if (po->running) {
3344                                         __unregister_prot_hook(sk, false);
3345                                         sk->sk_err = ENETDOWN;
3346                                         if (!sock_flag(sk, SOCK_DEAD))
3347                                                 sk->sk_error_report(sk);
3348                                 }
3349                                 if (msg == NETDEV_UNREGISTER) {
3350                                         po->ifindex = -1;
3351                                         if (po->prot_hook.dev)
3352                                                 dev_put(po->prot_hook.dev);
3353                                         po->prot_hook.dev = NULL;
3354                                 }
3355                                 spin_unlock(&po->bind_lock);
3356                         }
3357                         break;
3358                 case NETDEV_UP:
3359                         if (dev->ifindex == po->ifindex) {
3360                                 spin_lock(&po->bind_lock);
3361                                 if (po->num)
3362                                         register_prot_hook(sk);
3363                                 spin_unlock(&po->bind_lock);
3364                         }
3365                         break;
3366                 }
3367         }
3368         rcu_read_unlock();
3369         return NOTIFY_DONE;
3370 }
3371
3372
3373 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3374                         unsigned long arg)
3375 {
3376         struct sock *sk = sock->sk;
3377
3378         switch (cmd) {
3379         case SIOCOUTQ:
3380         {
3381                 int amount = sk_wmem_alloc_get(sk);
3382
3383                 return put_user(amount, (int __user *)arg);
3384         }
3385         case SIOCINQ:
3386         {
3387                 struct sk_buff *skb;
3388                 int amount = 0;
3389
3390                 spin_lock_bh(&sk->sk_receive_queue.lock);
3391                 skb = skb_peek(&sk->sk_receive_queue);
3392                 if (skb)
3393                         amount = skb->len;
3394                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3395                 return put_user(amount, (int __user *)arg);
3396         }
3397         case SIOCGSTAMP:
3398                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3399         case SIOCGSTAMPNS:
3400                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3401
3402 #ifdef CONFIG_INET
3403         case SIOCADDRT:
3404         case SIOCDELRT:
3405         case SIOCDARP:
3406         case SIOCGARP:
3407         case SIOCSARP:
3408         case SIOCGIFADDR:
3409         case SIOCSIFADDR:
3410         case SIOCGIFBRDADDR:
3411         case SIOCSIFBRDADDR:
3412         case SIOCGIFNETMASK:
3413         case SIOCSIFNETMASK:
3414         case SIOCGIFDSTADDR:
3415         case SIOCSIFDSTADDR:
3416         case SIOCSIFFLAGS:
3417                 return inet_dgram_ops.ioctl(sock, cmd, arg);
3418 #endif
3419
3420         default:
3421                 return -ENOIOCTLCMD;
3422         }
3423         return 0;
3424 }
3425
3426 static unsigned int packet_poll(struct file *file, struct socket *sock,
3427                                 poll_table *wait)
3428 {
3429         struct sock *sk = sock->sk;
3430         struct packet_sock *po = pkt_sk(sk);
3431         unsigned int mask = datagram_poll(file, sock, wait);
3432
3433         spin_lock_bh(&sk->sk_receive_queue.lock);
3434         if (po->rx_ring.pg_vec) {
3435                 if (!packet_previous_rx_frame(po, &po->rx_ring,
3436                         TP_STATUS_KERNEL))
3437                         mask |= POLLIN | POLLRDNORM;
3438         }
3439         spin_unlock_bh(&sk->sk_receive_queue.lock);
3440         spin_lock_bh(&sk->sk_write_queue.lock);
3441         if (po->tx_ring.pg_vec) {
3442                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3443                         mask |= POLLOUT | POLLWRNORM;
3444         }
3445         spin_unlock_bh(&sk->sk_write_queue.lock);
3446         return mask;
3447 }
3448
3449
3450 /* Dirty? Well, I still did not learn better way to account
3451  * for user mmaps.
3452  */
3453
3454 static void packet_mm_open(struct vm_area_struct *vma)
3455 {
3456         struct file *file = vma->vm_file;
3457         struct socket *sock = file->private_data;
3458         struct sock *sk = sock->sk;
3459
3460         if (sk)
3461                 atomic_inc(&pkt_sk(sk)->mapped);
3462 }
3463
3464 static void packet_mm_close(struct vm_area_struct *vma)
3465 {
3466         struct file *file = vma->vm_file;
3467         struct socket *sock = file->private_data;
3468         struct sock *sk = sock->sk;
3469
3470         if (sk)
3471                 atomic_dec(&pkt_sk(sk)->mapped);
3472 }
3473
3474 static const struct vm_operations_struct packet_mmap_ops = {
3475         .open   =       packet_mm_open,
3476         .close  =       packet_mm_close,
3477 };
3478
3479 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3480                         unsigned int len)
3481 {
3482         int i;
3483
3484         for (i = 0; i < len; i++) {
3485                 if (likely(pg_vec[i].buffer)) {
3486                         if (is_vmalloc_addr(pg_vec[i].buffer))
3487                                 vfree(pg_vec[i].buffer);
3488                         else
3489                                 free_pages((unsigned long)pg_vec[i].buffer,
3490                                            order);
3491                         pg_vec[i].buffer = NULL;
3492                 }
3493         }
3494         kfree(pg_vec);
3495 }
3496
3497 static char *alloc_one_pg_vec_page(unsigned long order)
3498 {
3499         char *buffer = NULL;
3500         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3501                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3502
3503         buffer = (char *) __get_free_pages(gfp_flags, order);
3504
3505         if (buffer)
3506                 return buffer;
3507
3508         /*
3509          * __get_free_pages failed, fall back to vmalloc
3510          */
3511         buffer = vzalloc((1 << order) * PAGE_SIZE);
3512
3513         if (buffer)
3514                 return buffer;
3515
3516         /*
3517          * vmalloc failed, lets dig into swap here
3518          */
3519         gfp_flags &= ~__GFP_NORETRY;
3520         buffer = (char *)__get_free_pages(gfp_flags, order);
3521         if (buffer)
3522                 return buffer;
3523
3524         /*
3525          * complete and utter failure
3526          */
3527         return NULL;
3528 }
3529
3530 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3531 {
3532         unsigned int block_nr = req->tp_block_nr;
3533         struct pgv *pg_vec;
3534         int i;
3535
3536         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3537         if (unlikely(!pg_vec))
3538                 goto out;
3539
3540         for (i = 0; i < block_nr; i++) {
3541                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3542                 if (unlikely(!pg_vec[i].buffer))
3543                         goto out_free_pgvec;
3544         }
3545
3546 out:
3547         return pg_vec;
3548
3549 out_free_pgvec:
3550         free_pg_vec(pg_vec, order, block_nr);
3551         pg_vec = NULL;
3552         goto out;
3553 }
3554
3555 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3556                 int closing, int tx_ring)
3557 {
3558         struct pgv *pg_vec = NULL;
3559         struct packet_sock *po = pkt_sk(sk);
3560         int was_running, order = 0;
3561         struct packet_ring_buffer *rb;
3562         struct sk_buff_head *rb_queue;
3563         __be16 num;
3564         int err = -EINVAL;
3565         /* Added to avoid minimal code churn */
3566         struct tpacket_req *req = &req_u->req;
3567
3568         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3569         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3570                 WARN(1, "Tx-ring is not supported.\n");
3571                 goto out;
3572         }
3573
3574         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3575         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3576
3577         err = -EBUSY;
3578         if (!closing) {
3579                 if (atomic_read(&po->mapped))
3580                         goto out;
3581                 if (atomic_read(&rb->pending))
3582                         goto out;
3583         }
3584
3585         if (req->tp_block_nr) {
3586                 /* Sanity tests and some calculations */
3587                 err = -EBUSY;
3588                 if (unlikely(rb->pg_vec))
3589                         goto out;
3590
3591                 switch (po->tp_version) {
3592                 case TPACKET_V1:
3593                         po->tp_hdrlen = TPACKET_HDRLEN;
3594                         break;
3595                 case TPACKET_V2:
3596                         po->tp_hdrlen = TPACKET2_HDRLEN;
3597                         break;
3598                 case TPACKET_V3:
3599                         po->tp_hdrlen = TPACKET3_HDRLEN;
3600                         break;
3601                 }
3602
3603                 err = -EINVAL;
3604                 if (unlikely((int)req->tp_block_size <= 0))
3605                         goto out;
3606                 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3607                         goto out;
3608                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3609                                         po->tp_reserve))
3610                         goto out;
3611                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3612                         goto out;
3613
3614                 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3615                 if (unlikely(rb->frames_per_block <= 0))
3616                         goto out;
3617                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3618                                         req->tp_frame_nr))
3619                         goto out;
3620
3621                 err = -ENOMEM;
3622                 order = get_order(req->tp_block_size);
3623                 pg_vec = alloc_pg_vec(req, order);
3624                 if (unlikely(!pg_vec))
3625                         goto out;
3626                 switch (po->tp_version) {
3627                 case TPACKET_V3:
3628                 /* Transmit path is not supported. We checked
3629                  * it above but just being paranoid
3630                  */
3631                         if (!tx_ring)
3632                                 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3633                                 break;
3634                 default:
3635                         break;
3636                 }
3637         }
3638         /* Done */
3639         else {
3640                 err = -EINVAL;
3641                 if (unlikely(req->tp_frame_nr))
3642                         goto out;
3643         }
3644
3645         lock_sock(sk);
3646
3647         /* Detach socket from network */
3648         spin_lock(&po->bind_lock);
3649         was_running = po->running;
3650         num = po->num;
3651         if (was_running) {
3652                 po->num = 0;
3653                 __unregister_prot_hook(sk, false);
3654         }
3655         spin_unlock(&po->bind_lock);
3656
3657         synchronize_net();
3658
3659         err = -EBUSY;
3660         mutex_lock(&po->pg_vec_lock);
3661         if (closing || atomic_read(&po->mapped) == 0) {
3662                 err = 0;
3663                 spin_lock_bh(&rb_queue->lock);
3664                 swap(rb->pg_vec, pg_vec);
3665                 rb->frame_max = (req->tp_frame_nr - 1);
3666                 rb->head = 0;
3667                 rb->frame_size = req->tp_frame_size;
3668                 spin_unlock_bh(&rb_queue->lock);
3669
3670                 swap(rb->pg_vec_order, order);
3671                 swap(rb->pg_vec_len, req->tp_block_nr);
3672
3673                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3674                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3675                                                 tpacket_rcv : packet_rcv;
3676                 skb_queue_purge(rb_queue);
3677                 if (atomic_read(&po->mapped))
3678                         pr_err("packet_mmap: vma is busy: %d\n",
3679                                atomic_read(&po->mapped));
3680         }
3681         mutex_unlock(&po->pg_vec_lock);
3682
3683         spin_lock(&po->bind_lock);
3684         if (was_running) {
3685                 po->num = num;
3686                 register_prot_hook(sk);
3687         }
3688         spin_unlock(&po->bind_lock);
3689         if (closing && (po->tp_version > TPACKET_V2)) {
3690                 /* Because we don't support block-based V3 on tx-ring */
3691                 if (!tx_ring)
3692                         prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3693         }
3694         release_sock(sk);
3695
3696         if (pg_vec)
3697                 free_pg_vec(pg_vec, order, req->tp_block_nr);
3698 out:
3699         return err;
3700 }
3701
3702 static int packet_mmap(struct file *file, struct socket *sock,
3703                 struct vm_area_struct *vma)
3704 {
3705         struct sock *sk = sock->sk;
3706         struct packet_sock *po = pkt_sk(sk);
3707         unsigned long size, expected_size;
3708         struct packet_ring_buffer *rb;
3709         unsigned long start;
3710         int err = -EINVAL;
3711         int i;
3712
3713         if (vma->vm_pgoff)
3714                 return -EINVAL;
3715
3716         mutex_lock(&po->pg_vec_lock);
3717
3718         expected_size = 0;
3719         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3720                 if (rb->pg_vec) {
3721                         expected_size += rb->pg_vec_len
3722                                                 * rb->pg_vec_pages
3723                                                 * PAGE_SIZE;
3724                 }
3725         }
3726
3727         if (expected_size == 0)
3728                 goto out;
3729
3730         size = vma->vm_end - vma->vm_start;
3731         if (size != expected_size)
3732                 goto out;
3733
3734         start = vma->vm_start;
3735         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3736                 if (rb->pg_vec == NULL)
3737                         continue;
3738
3739                 for (i = 0; i < rb->pg_vec_len; i++) {
3740                         struct page *page;
3741                         void *kaddr = rb->pg_vec[i].buffer;
3742                         int pg_num;
3743
3744                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3745                                 page = pgv_to_page(kaddr);
3746                                 err = vm_insert_page(vma, start, page);
3747                                 if (unlikely(err))
3748                                         goto out;
3749                                 start += PAGE_SIZE;
3750                                 kaddr += PAGE_SIZE;
3751                         }
3752                 }
3753         }
3754
3755         atomic_inc(&po->mapped);
3756         vma->vm_ops = &packet_mmap_ops;
3757         err = 0;
3758
3759 out:
3760         mutex_unlock(&po->pg_vec_lock);
3761         return err;
3762 }
3763
3764 static const struct proto_ops packet_ops_spkt = {
3765         .family =       PF_PACKET,
3766         .owner =        THIS_MODULE,
3767         .release =      packet_release,
3768         .bind =         packet_bind_spkt,
3769         .connect =      sock_no_connect,
3770         .socketpair =   sock_no_socketpair,
3771         .accept =       sock_no_accept,
3772         .getname =      packet_getname_spkt,
3773         .poll =         datagram_poll,
3774         .ioctl =        packet_ioctl,
3775         .listen =       sock_no_listen,
3776         .shutdown =     sock_no_shutdown,
3777         .setsockopt =   sock_no_setsockopt,
3778         .getsockopt =   sock_no_getsockopt,
3779         .sendmsg =      packet_sendmsg_spkt,
3780         .recvmsg =      packet_recvmsg,
3781         .mmap =         sock_no_mmap,
3782         .sendpage =     sock_no_sendpage,
3783 };
3784
3785 static const struct proto_ops packet_ops = {
3786         .family =       PF_PACKET,
3787         .owner =        THIS_MODULE,
3788         .release =      packet_release,
3789         .bind =         packet_bind,
3790         .connect =      sock_no_connect,
3791         .socketpair =   sock_no_socketpair,
3792         .accept =       sock_no_accept,
3793         .getname =      packet_getname,
3794         .poll =         packet_poll,
3795         .ioctl =        packet_ioctl,
3796         .listen =       sock_no_listen,
3797         .shutdown =     sock_no_shutdown,
3798         .setsockopt =   packet_setsockopt,
3799         .getsockopt =   packet_getsockopt,
3800         .sendmsg =      packet_sendmsg,
3801         .recvmsg =      packet_recvmsg,
3802         .mmap =         packet_mmap,
3803         .sendpage =     sock_no_sendpage,
3804 };
3805
3806 static const struct net_proto_family packet_family_ops = {
3807         .family =       PF_PACKET,
3808         .create =       packet_create,
3809         .owner  =       THIS_MODULE,
3810 };
3811
3812 static struct notifier_block packet_netdev_notifier = {
3813         .notifier_call =        packet_notifier,
3814 };
3815
3816 #ifdef CONFIG_PROC_FS
3817
3818 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
3819         __acquires(RCU)
3820 {
3821         struct net *net = seq_file_net(seq);
3822
3823         rcu_read_lock();
3824         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
3825 }
3826
3827 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3828 {
3829         struct net *net = seq_file_net(seq);
3830         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
3831 }
3832
3833 static void packet_seq_stop(struct seq_file *seq, void *v)
3834         __releases(RCU)
3835 {
3836         rcu_read_unlock();
3837 }
3838
3839 static int packet_seq_show(struct seq_file *seq, void *v)
3840 {
3841         if (v == SEQ_START_TOKEN)
3842                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
3843         else {
3844                 struct sock *s = sk_entry(v);
3845                 const struct packet_sock *po = pkt_sk(s);
3846
3847                 seq_printf(seq,
3848                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
3849                            s,
3850                            atomic_read(&s->sk_refcnt),
3851                            s->sk_type,
3852                            ntohs(po->num),
3853                            po->ifindex,
3854                            po->running,
3855                            atomic_read(&s->sk_rmem_alloc),
3856                            sock_i_uid(s),
3857                            sock_i_ino(s));
3858         }
3859
3860         return 0;
3861 }
3862
3863 static const struct seq_operations packet_seq_ops = {
3864         .start  = packet_seq_start,
3865         .next   = packet_seq_next,
3866         .stop   = packet_seq_stop,
3867         .show   = packet_seq_show,
3868 };
3869
3870 static int packet_seq_open(struct inode *inode, struct file *file)
3871 {
3872         return seq_open_net(inode, file, &packet_seq_ops,
3873                             sizeof(struct seq_net_private));
3874 }
3875
3876 static const struct file_operations packet_seq_fops = {
3877         .owner          = THIS_MODULE,
3878         .open           = packet_seq_open,
3879         .read           = seq_read,
3880         .llseek         = seq_lseek,
3881         .release        = seq_release_net,
3882 };