Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[linux-block.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 <linux/uaccess.h>
77 #include <asm/ioctls.h>
78 #include <asm/page.h>
79 #include <asm/cacheflush.h>
80 #include <asm/io.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
92 #ifdef CONFIG_INET
93 #include <net/inet_common.h>
94 #endif
95 #include <linux/bpf.h>
96 #include <net/compat.h>
97
98 #include "internal.h"
99
100 /*
101    Assumptions:
102    - if device has no dev->hard_header routine, it adds and removes ll header
103      inside itself. In this case ll header is invisible outside of device,
104      but higher levels still should reserve dev->hard_header_len.
105      Some devices are enough clever to reallocate skb, when header
106      will not fit to reserved space (tunnel), another ones are silly
107      (PPP).
108    - packet socket receives packets with pulled ll header,
109      so that SOCK_RAW should push it back.
110
111 On receive:
112 -----------
113
114 Incoming, dev->hard_header!=NULL
115    mac_header -> ll header
116    data       -> data
117
118 Outgoing, dev->hard_header!=NULL
119    mac_header -> ll header
120    data       -> ll header
121
122 Incoming, dev->hard_header==NULL
123    mac_header -> UNKNOWN position. It is very likely, that it points to ll
124                  header.  PPP makes it, that is wrong, because introduce
125                  assymetry between rx and tx paths.
126    data       -> data
127
128 Outgoing, dev->hard_header==NULL
129    mac_header -> data. ll header is still not built!
130    data       -> data
131
132 Resume
133   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
134
135
136 On transmit:
137 ------------
138
139 dev->hard_header != NULL
140    mac_header -> ll header
141    data       -> ll header
142
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
144    mac_header -> data
145    data       -> data
146
147    We should set nh.raw on output to correct posistion,
148    packet classifier depends on it.
149  */
150
151 /* Private packet socket structures. */
152
153 /* identical to struct packet_mreq except it has
154  * a longer address field.
155  */
156 struct packet_mreq_max {
157         int             mr_ifindex;
158         unsigned short  mr_type;
159         unsigned short  mr_alen;
160         unsigned char   mr_address[MAX_ADDR_LEN];
161 };
162
163 union tpacket_uhdr {
164         struct tpacket_hdr  *h1;
165         struct tpacket2_hdr *h2;
166         struct tpacket3_hdr *h3;
167         void *raw;
168 };
169
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171                 int closing, int tx_ring);
172
173 #define V3_ALIGNMENT    (8)
174
175 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
187
188 struct packet_sock;
189 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
190                        struct packet_type *pt, struct net_device *orig_dev);
191
192 static void *packet_previous_frame(struct packet_sock *po,
193                 struct packet_ring_buffer *rb,
194                 int status);
195 static void packet_increment_head(struct packet_ring_buffer *buff);
196 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
198                         struct packet_sock *);
199 static void prb_retire_current_block(struct tpacket_kbdq_core *,
200                 struct packet_sock *, unsigned int status);
201 static int prb_queue_frozen(struct tpacket_kbdq_core *);
202 static void prb_open_block(struct tpacket_kbdq_core *,
203                 struct tpacket_block_desc *);
204 static void prb_retire_rx_blk_timer_expired(unsigned long);
205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
206 static void prb_init_blk_timer(struct packet_sock *,
207                 struct tpacket_kbdq_core *,
208                 void (*func) (unsigned long));
209 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
210 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
211                 struct tpacket3_hdr *);
212 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
213                 struct tpacket3_hdr *);
214 static void packet_flush_mclist(struct sock *sk);
215 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb);
216
217 struct packet_skb_cb {
218         union {
219                 struct sockaddr_pkt pkt;
220                 union {
221                         /* Trick: alias skb original length with
222                          * ll.sll_family and ll.protocol in order
223                          * to save room.
224                          */
225                         unsigned int origlen;
226                         struct sockaddr_ll ll;
227                 };
228         } sa;
229 };
230
231 #define vio_le() virtio_legacy_is_little_endian()
232
233 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
234
235 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
236 #define GET_PBLOCK_DESC(x, bid) \
237         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
238 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
239         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
240 #define GET_NEXT_PRB_BLK_NUM(x) \
241         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
242         ((x)->kactive_blk_num+1) : 0)
243
244 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
245 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246
247 static int packet_direct_xmit(struct sk_buff *skb)
248 {
249         struct net_device *dev = skb->dev;
250         struct sk_buff *orig_skb = skb;
251         struct netdev_queue *txq;
252         int ret = NETDEV_TX_BUSY;
253
254         if (unlikely(!netif_running(dev) ||
255                      !netif_carrier_ok(dev)))
256                 goto drop;
257
258         skb = validate_xmit_skb_list(skb, dev);
259         if (skb != orig_skb)
260                 goto drop;
261
262         packet_pick_tx_queue(dev, skb);
263         txq = skb_get_tx_queue(dev, skb);
264
265         local_bh_disable();
266
267         HARD_TX_LOCK(dev, txq, smp_processor_id());
268         if (!netif_xmit_frozen_or_drv_stopped(txq))
269                 ret = netdev_start_xmit(skb, dev, txq, false);
270         HARD_TX_UNLOCK(dev, txq);
271
272         local_bh_enable();
273
274         if (!dev_xmit_complete(ret))
275                 kfree_skb(skb);
276
277         return ret;
278 drop:
279         atomic_long_inc(&dev->tx_dropped);
280         kfree_skb_list(skb);
281         return NET_XMIT_DROP;
282 }
283
284 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
285 {
286         struct net_device *dev;
287
288         rcu_read_lock();
289         dev = rcu_dereference(po->cached_dev);
290         if (likely(dev))
291                 dev_hold(dev);
292         rcu_read_unlock();
293
294         return dev;
295 }
296
297 static void packet_cached_dev_assign(struct packet_sock *po,
298                                      struct net_device *dev)
299 {
300         rcu_assign_pointer(po->cached_dev, dev);
301 }
302
303 static void packet_cached_dev_reset(struct packet_sock *po)
304 {
305         RCU_INIT_POINTER(po->cached_dev, NULL);
306 }
307
308 static bool packet_use_direct_xmit(const struct packet_sock *po)
309 {
310         return po->xmit == packet_direct_xmit;
311 }
312
313 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
314 {
315         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
316 }
317
318 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
319 {
320         const struct net_device_ops *ops = dev->netdev_ops;
321         u16 queue_index;
322
323         if (ops->ndo_select_queue) {
324                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
325                                                     __packet_pick_tx_queue);
326                 queue_index = netdev_cap_txqueue(dev, queue_index);
327         } else {
328                 queue_index = __packet_pick_tx_queue(dev, skb);
329         }
330
331         skb_set_queue_mapping(skb, queue_index);
332 }
333
334 /* register_prot_hook must be invoked with the po->bind_lock held,
335  * or from a context in which asynchronous accesses to the packet
336  * socket is not possible (packet_create()).
337  */
338 static void register_prot_hook(struct sock *sk)
339 {
340         struct packet_sock *po = pkt_sk(sk);
341
342         if (!po->running) {
343                 if (po->fanout)
344                         __fanout_link(sk, po);
345                 else
346                         dev_add_pack(&po->prot_hook);
347
348                 sock_hold(sk);
349                 po->running = 1;
350         }
351 }
352
353 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
354  * held.   If the sync parameter is true, we will temporarily drop
355  * the po->bind_lock and do a synchronize_net to make sure no
356  * asynchronous packet processing paths still refer to the elements
357  * of po->prot_hook.  If the sync parameter is false, it is the
358  * callers responsibility to take care of this.
359  */
360 static void __unregister_prot_hook(struct sock *sk, bool sync)
361 {
362         struct packet_sock *po = pkt_sk(sk);
363
364         po->running = 0;
365
366         if (po->fanout)
367                 __fanout_unlink(sk, po);
368         else
369                 __dev_remove_pack(&po->prot_hook);
370
371         __sock_put(sk);
372
373         if (sync) {
374                 spin_unlock(&po->bind_lock);
375                 synchronize_net();
376                 spin_lock(&po->bind_lock);
377         }
378 }
379
380 static void unregister_prot_hook(struct sock *sk, bool sync)
381 {
382         struct packet_sock *po = pkt_sk(sk);
383
384         if (po->running)
385                 __unregister_prot_hook(sk, sync);
386 }
387
388 static inline struct page * __pure pgv_to_page(void *addr)
389 {
390         if (is_vmalloc_addr(addr))
391                 return vmalloc_to_page(addr);
392         return virt_to_page(addr);
393 }
394
395 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
396 {
397         union tpacket_uhdr h;
398
399         h.raw = frame;
400         switch (po->tp_version) {
401         case TPACKET_V1:
402                 h.h1->tp_status = status;
403                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
404                 break;
405         case TPACKET_V2:
406                 h.h2->tp_status = status;
407                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
408                 break;
409         case TPACKET_V3:
410                 h.h3->tp_status = status;
411                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
412                 break;
413         default:
414                 WARN(1, "TPACKET version not supported.\n");
415                 BUG();
416         }
417
418         smp_wmb();
419 }
420
421 static int __packet_get_status(struct packet_sock *po, void *frame)
422 {
423         union tpacket_uhdr h;
424
425         smp_rmb();
426
427         h.raw = frame;
428         switch (po->tp_version) {
429         case TPACKET_V1:
430                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
431                 return h.h1->tp_status;
432         case TPACKET_V2:
433                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
434                 return h.h2->tp_status;
435         case TPACKET_V3:
436                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
437                 return h.h3->tp_status;
438         default:
439                 WARN(1, "TPACKET version not supported.\n");
440                 BUG();
441                 return 0;
442         }
443 }
444
445 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
446                                    unsigned int flags)
447 {
448         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
449
450         if (shhwtstamps &&
451             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
452             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
453                 return TP_STATUS_TS_RAW_HARDWARE;
454
455         if (ktime_to_timespec_cond(skb->tstamp, ts))
456                 return TP_STATUS_TS_SOFTWARE;
457
458         return 0;
459 }
460
461 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
462                                     struct sk_buff *skb)
463 {
464         union tpacket_uhdr h;
465         struct timespec ts;
466         __u32 ts_status;
467
468         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
469                 return 0;
470
471         h.raw = frame;
472         switch (po->tp_version) {
473         case TPACKET_V1:
474                 h.h1->tp_sec = ts.tv_sec;
475                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
476                 break;
477         case TPACKET_V2:
478                 h.h2->tp_sec = ts.tv_sec;
479                 h.h2->tp_nsec = ts.tv_nsec;
480                 break;
481         case TPACKET_V3:
482                 h.h3->tp_sec = ts.tv_sec;
483                 h.h3->tp_nsec = ts.tv_nsec;
484                 break;
485         default:
486                 WARN(1, "TPACKET version not supported.\n");
487                 BUG();
488         }
489
490         /* one flush is safe, as both fields always lie on the same cacheline */
491         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
492         smp_wmb();
493
494         return ts_status;
495 }
496
497 static void *packet_lookup_frame(struct packet_sock *po,
498                 struct packet_ring_buffer *rb,
499                 unsigned int position,
500                 int status)
501 {
502         unsigned int pg_vec_pos, frame_offset;
503         union tpacket_uhdr h;
504
505         pg_vec_pos = position / rb->frames_per_block;
506         frame_offset = position % rb->frames_per_block;
507
508         h.raw = rb->pg_vec[pg_vec_pos].buffer +
509                 (frame_offset * rb->frame_size);
510
511         if (status != __packet_get_status(po, h.raw))
512                 return NULL;
513
514         return h.raw;
515 }
516
517 static void *packet_current_frame(struct packet_sock *po,
518                 struct packet_ring_buffer *rb,
519                 int status)
520 {
521         return packet_lookup_frame(po, rb, rb->head, status);
522 }
523
524 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
525 {
526         del_timer_sync(&pkc->retire_blk_timer);
527 }
528
529 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
530                 struct sk_buff_head *rb_queue)
531 {
532         struct tpacket_kbdq_core *pkc;
533
534         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
535
536         spin_lock_bh(&rb_queue->lock);
537         pkc->delete_blk_timer = 1;
538         spin_unlock_bh(&rb_queue->lock);
539
540         prb_del_retire_blk_timer(pkc);
541 }
542
543 static void prb_init_blk_timer(struct packet_sock *po,
544                 struct tpacket_kbdq_core *pkc,
545                 void (*func) (unsigned long))
546 {
547         setup_timer(&pkc->retire_blk_timer, func, (long)po);
548         pkc->retire_blk_timer.expires = jiffies;
549 }
550
551 static void prb_setup_retire_blk_timer(struct packet_sock *po)
552 {
553         struct tpacket_kbdq_core *pkc;
554
555         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
556         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
557 }
558
559 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
560                                 int blk_size_in_bytes)
561 {
562         struct net_device *dev;
563         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
564         struct ethtool_link_ksettings ecmd;
565         int err;
566
567         rtnl_lock();
568         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
569         if (unlikely(!dev)) {
570                 rtnl_unlock();
571                 return DEFAULT_PRB_RETIRE_TOV;
572         }
573         err = __ethtool_get_link_ksettings(dev, &ecmd);
574         rtnl_unlock();
575         if (!err) {
576                 /*
577                  * If the link speed is so slow you don't really
578                  * need to worry about perf anyways
579                  */
580                 if (ecmd.base.speed < SPEED_1000 ||
581                     ecmd.base.speed == SPEED_UNKNOWN) {
582                         return DEFAULT_PRB_RETIRE_TOV;
583                 } else {
584                         msec = 1;
585                         div = ecmd.base.speed / 1000;
586                 }
587         }
588
589         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
590
591         if (div)
592                 mbits /= div;
593
594         tmo = mbits * msec;
595
596         if (div)
597                 return tmo+1;
598         return tmo;
599 }
600
601 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
602                         union tpacket_req_u *req_u)
603 {
604         p1->feature_req_word = req_u->req3.tp_feature_req_word;
605 }
606
607 static void init_prb_bdqc(struct packet_sock *po,
608                         struct packet_ring_buffer *rb,
609                         struct pgv *pg_vec,
610                         union tpacket_req_u *req_u)
611 {
612         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
613         struct tpacket_block_desc *pbd;
614
615         memset(p1, 0x0, sizeof(*p1));
616
617         p1->knxt_seq_num = 1;
618         p1->pkbdq = pg_vec;
619         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
620         p1->pkblk_start = pg_vec[0].buffer;
621         p1->kblk_size = req_u->req3.tp_block_size;
622         p1->knum_blocks = req_u->req3.tp_block_nr;
623         p1->hdrlen = po->tp_hdrlen;
624         p1->version = po->tp_version;
625         p1->last_kactive_blk_num = 0;
626         po->stats.stats3.tp_freeze_q_cnt = 0;
627         if (req_u->req3.tp_retire_blk_tov)
628                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
629         else
630                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
631                                                 req_u->req3.tp_block_size);
632         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
633         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
634
635         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
636         prb_init_ft_ops(p1, req_u);
637         prb_setup_retire_blk_timer(po);
638         prb_open_block(p1, pbd);
639 }
640
641 /*  Do NOT update the last_blk_num first.
642  *  Assumes sk_buff_head lock is held.
643  */
644 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
645 {
646         mod_timer(&pkc->retire_blk_timer,
647                         jiffies + pkc->tov_in_jiffies);
648         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
649 }
650
651 /*
652  * Timer logic:
653  * 1) We refresh the timer only when we open a block.
654  *    By doing this we don't waste cycles refreshing the timer
655  *        on packet-by-packet basis.
656  *
657  * With a 1MB block-size, on a 1Gbps line, it will take
658  * i) ~8 ms to fill a block + ii) memcpy etc.
659  * In this cut we are not accounting for the memcpy time.
660  *
661  * So, if the user sets the 'tmo' to 10ms then the timer
662  * will never fire while the block is still getting filled
663  * (which is what we want). However, the user could choose
664  * to close a block early and that's fine.
665  *
666  * But when the timer does fire, we check whether or not to refresh it.
667  * Since the tmo granularity is in msecs, it is not too expensive
668  * to refresh the timer, lets say every '8' msecs.
669  * Either the user can set the 'tmo' or we can derive it based on
670  * a) line-speed and b) block-size.
671  * prb_calc_retire_blk_tmo() calculates the tmo.
672  *
673  */
674 static void prb_retire_rx_blk_timer_expired(unsigned long data)
675 {
676         struct packet_sock *po = (struct packet_sock *)data;
677         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
678         unsigned int frozen;
679         struct tpacket_block_desc *pbd;
680
681         spin_lock(&po->sk.sk_receive_queue.lock);
682
683         frozen = prb_queue_frozen(pkc);
684         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
685
686         if (unlikely(pkc->delete_blk_timer))
687                 goto out;
688
689         /* We only need to plug the race when the block is partially filled.
690          * tpacket_rcv:
691          *              lock(); increment BLOCK_NUM_PKTS; unlock()
692          *              copy_bits() is in progress ...
693          *              timer fires on other cpu:
694          *              we can't retire the current block because copy_bits
695          *              is in progress.
696          *
697          */
698         if (BLOCK_NUM_PKTS(pbd)) {
699                 while (atomic_read(&pkc->blk_fill_in_prog)) {
700                         /* Waiting for skb_copy_bits to finish... */
701                         cpu_relax();
702                 }
703         }
704
705         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
706                 if (!frozen) {
707                         if (!BLOCK_NUM_PKTS(pbd)) {
708                                 /* An empty block. Just refresh the timer. */
709                                 goto refresh_timer;
710                         }
711                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
712                         if (!prb_dispatch_next_block(pkc, po))
713                                 goto refresh_timer;
714                         else
715                                 goto out;
716                 } else {
717                         /* Case 1. Queue was frozen because user-space was
718                          *         lagging behind.
719                          */
720                         if (prb_curr_blk_in_use(pbd)) {
721                                 /*
722                                  * Ok, user-space is still behind.
723                                  * So just refresh the timer.
724                                  */
725                                 goto refresh_timer;
726                         } else {
727                                /* Case 2. queue was frozen,user-space caught up,
728                                 * now the link went idle && the timer fired.
729                                 * We don't have a block to close.So we open this
730                                 * block and restart the timer.
731                                 * opening a block thaws the queue,restarts timer
732                                 * Thawing/timer-refresh is a side effect.
733                                 */
734                                 prb_open_block(pkc, pbd);
735                                 goto out;
736                         }
737                 }
738         }
739
740 refresh_timer:
741         _prb_refresh_rx_retire_blk_timer(pkc);
742
743 out:
744         spin_unlock(&po->sk.sk_receive_queue.lock);
745 }
746
747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
748                 struct tpacket_block_desc *pbd1, __u32 status)
749 {
750         /* Flush everything minus the block header */
751
752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
753         u8 *start, *end;
754
755         start = (u8 *)pbd1;
756
757         /* Skip the block header(we know header WILL fit in 4K) */
758         start += PAGE_SIZE;
759
760         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
761         for (; start < end; start += PAGE_SIZE)
762                 flush_dcache_page(pgv_to_page(start));
763
764         smp_wmb();
765 #endif
766
767         /* Now update the block status. */
768
769         BLOCK_STATUS(pbd1) = status;
770
771         /* Flush the block header */
772
773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
774         start = (u8 *)pbd1;
775         flush_dcache_page(pgv_to_page(start));
776
777         smp_wmb();
778 #endif
779 }
780
781 /*
782  * Side effect:
783  *
784  * 1) flush the block
785  * 2) Increment active_blk_num
786  *
787  * Note:We DONT refresh the timer on purpose.
788  *      Because almost always the next block will be opened.
789  */
790 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
791                 struct tpacket_block_desc *pbd1,
792                 struct packet_sock *po, unsigned int stat)
793 {
794         __u32 status = TP_STATUS_USER | stat;
795
796         struct tpacket3_hdr *last_pkt;
797         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
798         struct sock *sk = &po->sk;
799
800         if (po->stats.stats3.tp_drops)
801                 status |= TP_STATUS_LOSING;
802
803         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
804         last_pkt->tp_next_offset = 0;
805
806         /* Get the ts of the last pkt */
807         if (BLOCK_NUM_PKTS(pbd1)) {
808                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
809                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
810         } else {
811                 /* Ok, we tmo'd - so get the current time.
812                  *
813                  * It shouldn't really happen as we don't close empty
814                  * blocks. See prb_retire_rx_blk_timer_expired().
815                  */
816                 struct timespec ts;
817                 getnstimeofday(&ts);
818                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
819                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
820         }
821
822         smp_wmb();
823
824         /* Flush the block */
825         prb_flush_block(pkc1, pbd1, status);
826
827         sk->sk_data_ready(sk);
828
829         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
830 }
831
832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
833 {
834         pkc->reset_pending_on_curr_blk = 0;
835 }
836
837 /*
838  * Side effect of opening a block:
839  *
840  * 1) prb_queue is thawed.
841  * 2) retire_blk_timer is refreshed.
842  *
843  */
844 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
845         struct tpacket_block_desc *pbd1)
846 {
847         struct timespec ts;
848         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
849
850         smp_rmb();
851
852         /* We could have just memset this but we will lose the
853          * flexibility of making the priv area sticky
854          */
855
856         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
857         BLOCK_NUM_PKTS(pbd1) = 0;
858         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
859
860         getnstimeofday(&ts);
861
862         h1->ts_first_pkt.ts_sec = ts.tv_sec;
863         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
864
865         pkc1->pkblk_start = (char *)pbd1;
866         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
867
868         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
869         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
870
871         pbd1->version = pkc1->version;
872         pkc1->prev = pkc1->nxt_offset;
873         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
874
875         prb_thaw_queue(pkc1);
876         _prb_refresh_rx_retire_blk_timer(pkc1);
877
878         smp_wmb();
879 }
880
881 /*
882  * Queue freeze logic:
883  * 1) Assume tp_block_nr = 8 blocks.
884  * 2) At time 't0', user opens Rx ring.
885  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
886  * 4) user-space is either sleeping or processing block '0'.
887  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
888  *    it will close block-7,loop around and try to fill block '0'.
889  *    call-flow:
890  *    __packet_lookup_frame_in_block
891  *      prb_retire_current_block()
892  *      prb_dispatch_next_block()
893  *        |->(BLOCK_STATUS == USER) evaluates to true
894  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
895  * 6) Now there are two cases:
896  *    6.1) Link goes idle right after the queue is frozen.
897  *         But remember, the last open_block() refreshed the timer.
898  *         When this timer expires,it will refresh itself so that we can
899  *         re-open block-0 in near future.
900  *    6.2) Link is busy and keeps on receiving packets. This is a simple
901  *         case and __packet_lookup_frame_in_block will check if block-0
902  *         is free and can now be re-used.
903  */
904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
905                                   struct packet_sock *po)
906 {
907         pkc->reset_pending_on_curr_blk = 1;
908         po->stats.stats3.tp_freeze_q_cnt++;
909 }
910
911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
912
913 /*
914  * If the next block is free then we will dispatch it
915  * and return a good offset.
916  * Else, we will freeze the queue.
917  * So, caller must check the return value.
918  */
919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
920                 struct packet_sock *po)
921 {
922         struct tpacket_block_desc *pbd;
923
924         smp_rmb();
925
926         /* 1. Get current block num */
927         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
928
929         /* 2. If this block is currently in_use then freeze the queue */
930         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
931                 prb_freeze_queue(pkc, po);
932                 return NULL;
933         }
934
935         /*
936          * 3.
937          * open this block and return the offset where the first packet
938          * needs to get stored.
939          */
940         prb_open_block(pkc, pbd);
941         return (void *)pkc->nxt_offset;
942 }
943
944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
945                 struct packet_sock *po, unsigned int status)
946 {
947         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
948
949         /* retire/close the current block */
950         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
951                 /*
952                  * Plug the case where copy_bits() is in progress on
953                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
954                  * have space to copy the pkt in the current block and
955                  * called prb_retire_current_block()
956                  *
957                  * We don't need to worry about the TMO case because
958                  * the timer-handler already handled this case.
959                  */
960                 if (!(status & TP_STATUS_BLK_TMO)) {
961                         while (atomic_read(&pkc->blk_fill_in_prog)) {
962                                 /* Waiting for skb_copy_bits to finish... */
963                                 cpu_relax();
964                         }
965                 }
966                 prb_close_block(pkc, pbd, po, status);
967                 return;
968         }
969 }
970
971 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
972 {
973         return TP_STATUS_USER & BLOCK_STATUS(pbd);
974 }
975
976 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
977 {
978         return pkc->reset_pending_on_curr_blk;
979 }
980
981 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
982 {
983         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
984         atomic_dec(&pkc->blk_fill_in_prog);
985 }
986
987 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
988                         struct tpacket3_hdr *ppd)
989 {
990         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
991 }
992
993 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
994                         struct tpacket3_hdr *ppd)
995 {
996         ppd->hv1.tp_rxhash = 0;
997 }
998
999 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1000                         struct tpacket3_hdr *ppd)
1001 {
1002         if (skb_vlan_tag_present(pkc->skb)) {
1003                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1004                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1005                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1006         } else {
1007                 ppd->hv1.tp_vlan_tci = 0;
1008                 ppd->hv1.tp_vlan_tpid = 0;
1009                 ppd->tp_status = TP_STATUS_AVAILABLE;
1010         }
1011 }
1012
1013 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1014                         struct tpacket3_hdr *ppd)
1015 {
1016         ppd->hv1.tp_padding = 0;
1017         prb_fill_vlan_info(pkc, ppd);
1018
1019         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1020                 prb_fill_rxhash(pkc, ppd);
1021         else
1022                 prb_clear_rxhash(pkc, ppd);
1023 }
1024
1025 static void prb_fill_curr_block(char *curr,
1026                                 struct tpacket_kbdq_core *pkc,
1027                                 struct tpacket_block_desc *pbd,
1028                                 unsigned int len)
1029 {
1030         struct tpacket3_hdr *ppd;
1031
1032         ppd  = (struct tpacket3_hdr *)curr;
1033         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1034         pkc->prev = curr;
1035         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1036         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1037         BLOCK_NUM_PKTS(pbd) += 1;
1038         atomic_inc(&pkc->blk_fill_in_prog);
1039         prb_run_all_ft_ops(pkc, ppd);
1040 }
1041
1042 /* Assumes caller has the sk->rx_queue.lock */
1043 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1044                                             struct sk_buff *skb,
1045                                                 int status,
1046                                             unsigned int len
1047                                             )
1048 {
1049         struct tpacket_kbdq_core *pkc;
1050         struct tpacket_block_desc *pbd;
1051         char *curr, *end;
1052
1053         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1054         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1055
1056         /* Queue is frozen when user space is lagging behind */
1057         if (prb_queue_frozen(pkc)) {
1058                 /*
1059                  * Check if that last block which caused the queue to freeze,
1060                  * is still in_use by user-space.
1061                  */
1062                 if (prb_curr_blk_in_use(pbd)) {
1063                         /* Can't record this packet */
1064                         return NULL;
1065                 } else {
1066                         /*
1067                          * Ok, the block was released by user-space.
1068                          * Now let's open that block.
1069                          * opening a block also thaws the queue.
1070                          * Thawing is a side effect.
1071                          */
1072                         prb_open_block(pkc, pbd);
1073                 }
1074         }
1075
1076         smp_mb();
1077         curr = pkc->nxt_offset;
1078         pkc->skb = skb;
1079         end = (char *)pbd + pkc->kblk_size;
1080
1081         /* first try the current block */
1082         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1083                 prb_fill_curr_block(curr, pkc, pbd, len);
1084                 return (void *)curr;
1085         }
1086
1087         /* Ok, close the current block */
1088         prb_retire_current_block(pkc, po, 0);
1089
1090         /* Now, try to dispatch the next block */
1091         curr = (char *)prb_dispatch_next_block(pkc, po);
1092         if (curr) {
1093                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1094                 prb_fill_curr_block(curr, pkc, pbd, len);
1095                 return (void *)curr;
1096         }
1097
1098         /*
1099          * No free blocks are available.user_space hasn't caught up yet.
1100          * Queue was just frozen and now this packet will get dropped.
1101          */
1102         return NULL;
1103 }
1104
1105 static void *packet_current_rx_frame(struct packet_sock *po,
1106                                             struct sk_buff *skb,
1107                                             int status, unsigned int len)
1108 {
1109         char *curr = NULL;
1110         switch (po->tp_version) {
1111         case TPACKET_V1:
1112         case TPACKET_V2:
1113                 curr = packet_lookup_frame(po, &po->rx_ring,
1114                                         po->rx_ring.head, status);
1115                 return curr;
1116         case TPACKET_V3:
1117                 return __packet_lookup_frame_in_block(po, skb, status, len);
1118         default:
1119                 WARN(1, "TPACKET version not supported\n");
1120                 BUG();
1121                 return NULL;
1122         }
1123 }
1124
1125 static void *prb_lookup_block(struct packet_sock *po,
1126                                      struct packet_ring_buffer *rb,
1127                                      unsigned int idx,
1128                                      int status)
1129 {
1130         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1131         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1132
1133         if (status != BLOCK_STATUS(pbd))
1134                 return NULL;
1135         return pbd;
1136 }
1137
1138 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1139 {
1140         unsigned int prev;
1141         if (rb->prb_bdqc.kactive_blk_num)
1142                 prev = rb->prb_bdqc.kactive_blk_num-1;
1143         else
1144                 prev = rb->prb_bdqc.knum_blocks-1;
1145         return prev;
1146 }
1147
1148 /* Assumes caller has held the rx_queue.lock */
1149 static void *__prb_previous_block(struct packet_sock *po,
1150                                          struct packet_ring_buffer *rb,
1151                                          int status)
1152 {
1153         unsigned int previous = prb_previous_blk_num(rb);
1154         return prb_lookup_block(po, rb, previous, status);
1155 }
1156
1157 static void *packet_previous_rx_frame(struct packet_sock *po,
1158                                              struct packet_ring_buffer *rb,
1159                                              int status)
1160 {
1161         if (po->tp_version <= TPACKET_V2)
1162                 return packet_previous_frame(po, rb, status);
1163
1164         return __prb_previous_block(po, rb, status);
1165 }
1166
1167 static void packet_increment_rx_head(struct packet_sock *po,
1168                                             struct packet_ring_buffer *rb)
1169 {
1170         switch (po->tp_version) {
1171         case TPACKET_V1:
1172         case TPACKET_V2:
1173                 return packet_increment_head(rb);
1174         case TPACKET_V3:
1175         default:
1176                 WARN(1, "TPACKET version not supported.\n");
1177                 BUG();
1178                 return;
1179         }
1180 }
1181
1182 static void *packet_previous_frame(struct packet_sock *po,
1183                 struct packet_ring_buffer *rb,
1184                 int status)
1185 {
1186         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1187         return packet_lookup_frame(po, rb, previous, status);
1188 }
1189
1190 static void packet_increment_head(struct packet_ring_buffer *buff)
1191 {
1192         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1193 }
1194
1195 static void packet_inc_pending(struct packet_ring_buffer *rb)
1196 {
1197         this_cpu_inc(*rb->pending_refcnt);
1198 }
1199
1200 static void packet_dec_pending(struct packet_ring_buffer *rb)
1201 {
1202         this_cpu_dec(*rb->pending_refcnt);
1203 }
1204
1205 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1206 {
1207         unsigned int refcnt = 0;
1208         int cpu;
1209
1210         /* We don't use pending refcount in rx_ring. */
1211         if (rb->pending_refcnt == NULL)
1212                 return 0;
1213
1214         for_each_possible_cpu(cpu)
1215                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1216
1217         return refcnt;
1218 }
1219
1220 static int packet_alloc_pending(struct packet_sock *po)
1221 {
1222         po->rx_ring.pending_refcnt = NULL;
1223
1224         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1225         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1226                 return -ENOBUFS;
1227
1228         return 0;
1229 }
1230
1231 static void packet_free_pending(struct packet_sock *po)
1232 {
1233         free_percpu(po->tx_ring.pending_refcnt);
1234 }
1235
1236 #define ROOM_POW_OFF    2
1237 #define ROOM_NONE       0x0
1238 #define ROOM_LOW        0x1
1239 #define ROOM_NORMAL     0x2
1240
1241 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1242 {
1243         int idx, len;
1244
1245         len = po->rx_ring.frame_max + 1;
1246         idx = po->rx_ring.head;
1247         if (pow_off)
1248                 idx += len >> pow_off;
1249         if (idx >= len)
1250                 idx -= len;
1251         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1252 }
1253
1254 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1255 {
1256         int idx, len;
1257
1258         len = po->rx_ring.prb_bdqc.knum_blocks;
1259         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1260         if (pow_off)
1261                 idx += len >> pow_off;
1262         if (idx >= len)
1263                 idx -= len;
1264         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1265 }
1266
1267 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1268 {
1269         struct sock *sk = &po->sk;
1270         int ret = ROOM_NONE;
1271
1272         if (po->prot_hook.func != tpacket_rcv) {
1273                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1274                                           - (skb ? skb->truesize : 0);
1275                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1276                         return ROOM_NORMAL;
1277                 else if (avail > 0)
1278                         return ROOM_LOW;
1279                 else
1280                         return ROOM_NONE;
1281         }
1282
1283         if (po->tp_version == TPACKET_V3) {
1284                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1285                         ret = ROOM_NORMAL;
1286                 else if (__tpacket_v3_has_room(po, 0))
1287                         ret = ROOM_LOW;
1288         } else {
1289                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1290                         ret = ROOM_NORMAL;
1291                 else if (__tpacket_has_room(po, 0))
1292                         ret = ROOM_LOW;
1293         }
1294
1295         return ret;
1296 }
1297
1298 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1299 {
1300         int ret;
1301         bool has_room;
1302
1303         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1304         ret = __packet_rcv_has_room(po, skb);
1305         has_room = ret == ROOM_NORMAL;
1306         if (po->pressure == has_room)
1307                 po->pressure = !has_room;
1308         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1309
1310         return ret;
1311 }
1312
1313 static void packet_sock_destruct(struct sock *sk)
1314 {
1315         skb_queue_purge(&sk->sk_error_queue);
1316
1317         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1318         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1319
1320         if (!sock_flag(sk, SOCK_DEAD)) {
1321                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1322                 return;
1323         }
1324
1325         sk_refcnt_debug_dec(sk);
1326 }
1327
1328 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1329 {
1330         u32 rxhash;
1331         int i, count = 0;
1332
1333         rxhash = skb_get_hash(skb);
1334         for (i = 0; i < ROLLOVER_HLEN; i++)
1335                 if (po->rollover->history[i] == rxhash)
1336                         count++;
1337
1338         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1339         return count > (ROLLOVER_HLEN >> 1);
1340 }
1341
1342 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1343                                       struct sk_buff *skb,
1344                                       unsigned int num)
1345 {
1346         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1347 }
1348
1349 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1350                                     struct sk_buff *skb,
1351                                     unsigned int num)
1352 {
1353         unsigned int val = atomic_inc_return(&f->rr_cur);
1354
1355         return val % num;
1356 }
1357
1358 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1359                                      struct sk_buff *skb,
1360                                      unsigned int num)
1361 {
1362         return smp_processor_id() % num;
1363 }
1364
1365 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1366                                      struct sk_buff *skb,
1367                                      unsigned int num)
1368 {
1369         return prandom_u32_max(num);
1370 }
1371
1372 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1373                                           struct sk_buff *skb,
1374                                           unsigned int idx, bool try_self,
1375                                           unsigned int num)
1376 {
1377         struct packet_sock *po, *po_next, *po_skip = NULL;
1378         unsigned int i, j, room = ROOM_NONE;
1379
1380         po = pkt_sk(f->arr[idx]);
1381
1382         if (try_self) {
1383                 room = packet_rcv_has_room(po, skb);
1384                 if (room == ROOM_NORMAL ||
1385                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1386                         return idx;
1387                 po_skip = po;
1388         }
1389
1390         i = j = min_t(int, po->rollover->sock, num - 1);
1391         do {
1392                 po_next = pkt_sk(f->arr[i]);
1393                 if (po_next != po_skip && !po_next->pressure &&
1394                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1395                         if (i != j)
1396                                 po->rollover->sock = i;
1397                         atomic_long_inc(&po->rollover->num);
1398                         if (room == ROOM_LOW)
1399                                 atomic_long_inc(&po->rollover->num_huge);
1400                         return i;
1401                 }
1402
1403                 if (++i == num)
1404                         i = 0;
1405         } while (i != j);
1406
1407         atomic_long_inc(&po->rollover->num_failed);
1408         return idx;
1409 }
1410
1411 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1412                                     struct sk_buff *skb,
1413                                     unsigned int num)
1414 {
1415         return skb_get_queue_mapping(skb) % num;
1416 }
1417
1418 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1419                                      struct sk_buff *skb,
1420                                      unsigned int num)
1421 {
1422         struct bpf_prog *prog;
1423         unsigned int ret = 0;
1424
1425         rcu_read_lock();
1426         prog = rcu_dereference(f->bpf_prog);
1427         if (prog)
1428                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1429         rcu_read_unlock();
1430
1431         return ret;
1432 }
1433
1434 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1435 {
1436         return f->flags & (flag >> 8);
1437 }
1438
1439 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1440                              struct packet_type *pt, struct net_device *orig_dev)
1441 {
1442         struct packet_fanout *f = pt->af_packet_priv;
1443         unsigned int num = READ_ONCE(f->num_members);
1444         struct net *net = read_pnet(&f->net);
1445         struct packet_sock *po;
1446         unsigned int idx;
1447
1448         if (!net_eq(dev_net(dev), net) || !num) {
1449                 kfree_skb(skb);
1450                 return 0;
1451         }
1452
1453         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1454                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1455                 if (!skb)
1456                         return 0;
1457         }
1458         switch (f->type) {
1459         case PACKET_FANOUT_HASH:
1460         default:
1461                 idx = fanout_demux_hash(f, skb, num);
1462                 break;
1463         case PACKET_FANOUT_LB:
1464                 idx = fanout_demux_lb(f, skb, num);
1465                 break;
1466         case PACKET_FANOUT_CPU:
1467                 idx = fanout_demux_cpu(f, skb, num);
1468                 break;
1469         case PACKET_FANOUT_RND:
1470                 idx = fanout_demux_rnd(f, skb, num);
1471                 break;
1472         case PACKET_FANOUT_QM:
1473                 idx = fanout_demux_qm(f, skb, num);
1474                 break;
1475         case PACKET_FANOUT_ROLLOVER:
1476                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1477                 break;
1478         case PACKET_FANOUT_CBPF:
1479         case PACKET_FANOUT_EBPF:
1480                 idx = fanout_demux_bpf(f, skb, num);
1481                 break;
1482         }
1483
1484         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1485                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1486
1487         po = pkt_sk(f->arr[idx]);
1488         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1489 }
1490
1491 DEFINE_MUTEX(fanout_mutex);
1492 EXPORT_SYMBOL_GPL(fanout_mutex);
1493 static LIST_HEAD(fanout_list);
1494 static u16 fanout_next_id;
1495
1496 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1497 {
1498         struct packet_fanout *f = po->fanout;
1499
1500         spin_lock(&f->lock);
1501         f->arr[f->num_members] = sk;
1502         smp_wmb();
1503         f->num_members++;
1504         if (f->num_members == 1)
1505                 dev_add_pack(&f->prot_hook);
1506         spin_unlock(&f->lock);
1507 }
1508
1509 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1510 {
1511         struct packet_fanout *f = po->fanout;
1512         int i;
1513
1514         spin_lock(&f->lock);
1515         for (i = 0; i < f->num_members; i++) {
1516                 if (f->arr[i] == sk)
1517                         break;
1518         }
1519         BUG_ON(i >= f->num_members);
1520         f->arr[i] = f->arr[f->num_members - 1];
1521         f->num_members--;
1522         if (f->num_members == 0)
1523                 __dev_remove_pack(&f->prot_hook);
1524         spin_unlock(&f->lock);
1525 }
1526
1527 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1528 {
1529         if (sk->sk_family != PF_PACKET)
1530                 return false;
1531
1532         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1533 }
1534
1535 static void fanout_init_data(struct packet_fanout *f)
1536 {
1537         switch (f->type) {
1538         case PACKET_FANOUT_LB:
1539                 atomic_set(&f->rr_cur, 0);
1540                 break;
1541         case PACKET_FANOUT_CBPF:
1542         case PACKET_FANOUT_EBPF:
1543                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1544                 break;
1545         }
1546 }
1547
1548 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1549 {
1550         struct bpf_prog *old;
1551
1552         spin_lock(&f->lock);
1553         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1554         rcu_assign_pointer(f->bpf_prog, new);
1555         spin_unlock(&f->lock);
1556
1557         if (old) {
1558                 synchronize_net();
1559                 bpf_prog_destroy(old);
1560         }
1561 }
1562
1563 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1564                                 unsigned int len)
1565 {
1566         struct bpf_prog *new;
1567         struct sock_fprog fprog;
1568         int ret;
1569
1570         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1571                 return -EPERM;
1572         if (len != sizeof(fprog))
1573                 return -EINVAL;
1574         if (copy_from_user(&fprog, data, len))
1575                 return -EFAULT;
1576
1577         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1578         if (ret)
1579                 return ret;
1580
1581         __fanout_set_data_bpf(po->fanout, new);
1582         return 0;
1583 }
1584
1585 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1586                                 unsigned int len)
1587 {
1588         struct bpf_prog *new;
1589         u32 fd;
1590
1591         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1592                 return -EPERM;
1593         if (len != sizeof(fd))
1594                 return -EINVAL;
1595         if (copy_from_user(&fd, data, len))
1596                 return -EFAULT;
1597
1598         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1599         if (IS_ERR(new))
1600                 return PTR_ERR(new);
1601
1602         __fanout_set_data_bpf(po->fanout, new);
1603         return 0;
1604 }
1605
1606 static int fanout_set_data(struct packet_sock *po, char __user *data,
1607                            unsigned int len)
1608 {
1609         switch (po->fanout->type) {
1610         case PACKET_FANOUT_CBPF:
1611                 return fanout_set_data_cbpf(po, data, len);
1612         case PACKET_FANOUT_EBPF:
1613                 return fanout_set_data_ebpf(po, data, len);
1614         default:
1615                 return -EINVAL;
1616         };
1617 }
1618
1619 static void fanout_release_data(struct packet_fanout *f)
1620 {
1621         switch (f->type) {
1622         case PACKET_FANOUT_CBPF:
1623         case PACKET_FANOUT_EBPF:
1624                 __fanout_set_data_bpf(f, NULL);
1625         };
1626 }
1627
1628 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1629 {
1630         struct packet_fanout *f;
1631
1632         list_for_each_entry(f, &fanout_list, list) {
1633                 if (f->id == candidate_id &&
1634                     read_pnet(&f->net) == sock_net(sk)) {
1635                         return false;
1636                 }
1637         }
1638         return true;
1639 }
1640
1641 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1642 {
1643         u16 id = fanout_next_id;
1644
1645         do {
1646                 if (__fanout_id_is_free(sk, id)) {
1647                         *new_id = id;
1648                         fanout_next_id = id + 1;
1649                         return true;
1650                 }
1651
1652                 id++;
1653         } while (id != fanout_next_id);
1654
1655         return false;
1656 }
1657
1658 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1659 {
1660         struct packet_rollover *rollover = NULL;
1661         struct packet_sock *po = pkt_sk(sk);
1662         struct packet_fanout *f, *match;
1663         u8 type = type_flags & 0xff;
1664         u8 flags = type_flags >> 8;
1665         int err;
1666
1667         switch (type) {
1668         case PACKET_FANOUT_ROLLOVER:
1669                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1670                         return -EINVAL;
1671         case PACKET_FANOUT_HASH:
1672         case PACKET_FANOUT_LB:
1673         case PACKET_FANOUT_CPU:
1674         case PACKET_FANOUT_RND:
1675         case PACKET_FANOUT_QM:
1676         case PACKET_FANOUT_CBPF:
1677         case PACKET_FANOUT_EBPF:
1678                 break;
1679         default:
1680                 return -EINVAL;
1681         }
1682
1683         mutex_lock(&fanout_mutex);
1684
1685         err = -EALREADY;
1686         if (po->fanout)
1687                 goto out;
1688
1689         if (type == PACKET_FANOUT_ROLLOVER ||
1690             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1691                 err = -ENOMEM;
1692                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1693                 if (!rollover)
1694                         goto out;
1695                 atomic_long_set(&rollover->num, 0);
1696                 atomic_long_set(&rollover->num_huge, 0);
1697                 atomic_long_set(&rollover->num_failed, 0);
1698                 po->rollover = rollover;
1699         }
1700
1701         if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1702                 if (id != 0) {
1703                         err = -EINVAL;
1704                         goto out;
1705                 }
1706                 if (!fanout_find_new_id(sk, &id)) {
1707                         err = -ENOMEM;
1708                         goto out;
1709                 }
1710                 /* ephemeral flag for the first socket in the group: drop it */
1711                 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1712         }
1713
1714         match = NULL;
1715         list_for_each_entry(f, &fanout_list, list) {
1716                 if (f->id == id &&
1717                     read_pnet(&f->net) == sock_net(sk)) {
1718                         match = f;
1719                         break;
1720                 }
1721         }
1722         err = -EINVAL;
1723         if (match && match->flags != flags)
1724                 goto out;
1725         if (!match) {
1726                 err = -ENOMEM;
1727                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1728                 if (!match)
1729                         goto out;
1730                 write_pnet(&match->net, sock_net(sk));
1731                 match->id = id;
1732                 match->type = type;
1733                 match->flags = flags;
1734                 INIT_LIST_HEAD(&match->list);
1735                 spin_lock_init(&match->lock);
1736                 refcount_set(&match->sk_ref, 0);
1737                 fanout_init_data(match);
1738                 match->prot_hook.type = po->prot_hook.type;
1739                 match->prot_hook.dev = po->prot_hook.dev;
1740                 match->prot_hook.func = packet_rcv_fanout;
1741                 match->prot_hook.af_packet_priv = match;
1742                 match->prot_hook.id_match = match_fanout_group;
1743                 list_add(&match->list, &fanout_list);
1744         }
1745         err = -EINVAL;
1746
1747         spin_lock(&po->bind_lock);
1748         if (po->running &&
1749             match->type == type &&
1750             match->prot_hook.type == po->prot_hook.type &&
1751             match->prot_hook.dev == po->prot_hook.dev) {
1752                 err = -ENOSPC;
1753                 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1754                         __dev_remove_pack(&po->prot_hook);
1755                         po->fanout = match;
1756                         refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1757                         __fanout_link(sk, po);
1758                         err = 0;
1759                 }
1760         }
1761         spin_unlock(&po->bind_lock);
1762
1763         if (err && !refcount_read(&match->sk_ref)) {
1764                 list_del(&match->list);
1765                 kfree(match);
1766         }
1767
1768 out:
1769         if (err && rollover) {
1770                 kfree_rcu(rollover, rcu);
1771                 po->rollover = NULL;
1772         }
1773         mutex_unlock(&fanout_mutex);
1774         return err;
1775 }
1776
1777 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1778  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1779  * It is the responsibility of the caller to call fanout_release_data() and
1780  * free the returned packet_fanout (after synchronize_net())
1781  */
1782 static struct packet_fanout *fanout_release(struct sock *sk)
1783 {
1784         struct packet_sock *po = pkt_sk(sk);
1785         struct packet_fanout *f;
1786
1787         mutex_lock(&fanout_mutex);
1788         f = po->fanout;
1789         if (f) {
1790                 po->fanout = NULL;
1791
1792                 if (refcount_dec_and_test(&f->sk_ref))
1793                         list_del(&f->list);
1794                 else
1795                         f = NULL;
1796
1797                 if (po->rollover) {
1798                         kfree_rcu(po->rollover, rcu);
1799                         po->rollover = NULL;
1800                 }
1801         }
1802         mutex_unlock(&fanout_mutex);
1803
1804         return f;
1805 }
1806
1807 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1808                                           struct sk_buff *skb)
1809 {
1810         /* Earlier code assumed this would be a VLAN pkt, double-check
1811          * this now that we have the actual packet in hand. We can only
1812          * do this check on Ethernet devices.
1813          */
1814         if (unlikely(dev->type != ARPHRD_ETHER))
1815                 return false;
1816
1817         skb_reset_mac_header(skb);
1818         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1819 }
1820
1821 static const struct proto_ops packet_ops;
1822
1823 static const struct proto_ops packet_ops_spkt;
1824
1825 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1826                            struct packet_type *pt, struct net_device *orig_dev)
1827 {
1828         struct sock *sk;
1829         struct sockaddr_pkt *spkt;
1830
1831         /*
1832          *      When we registered the protocol we saved the socket in the data
1833          *      field for just this event.
1834          */
1835
1836         sk = pt->af_packet_priv;
1837
1838         /*
1839          *      Yank back the headers [hope the device set this
1840          *      right or kerboom...]
1841          *
1842          *      Incoming packets have ll header pulled,
1843          *      push it back.
1844          *
1845          *      For outgoing ones skb->data == skb_mac_header(skb)
1846          *      so that this procedure is noop.
1847          */
1848
1849         if (skb->pkt_type == PACKET_LOOPBACK)
1850                 goto out;
1851
1852         if (!net_eq(dev_net(dev), sock_net(sk)))
1853                 goto out;
1854
1855         skb = skb_share_check(skb, GFP_ATOMIC);
1856         if (skb == NULL)
1857                 goto oom;
1858
1859         /* drop any routing info */
1860         skb_dst_drop(skb);
1861
1862         /* drop conntrack reference */
1863         nf_reset(skb);
1864
1865         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1866
1867         skb_push(skb, skb->data - skb_mac_header(skb));
1868
1869         /*
1870          *      The SOCK_PACKET socket receives _all_ frames.
1871          */
1872
1873         spkt->spkt_family = dev->type;
1874         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1875         spkt->spkt_protocol = skb->protocol;
1876
1877         /*
1878          *      Charge the memory to the socket. This is done specifically
1879          *      to prevent sockets using all the memory up.
1880          */
1881
1882         if (sock_queue_rcv_skb(sk, skb) == 0)
1883                 return 0;
1884
1885 out:
1886         kfree_skb(skb);
1887 oom:
1888         return 0;
1889 }
1890
1891
1892 /*
1893  *      Output a raw packet to a device layer. This bypasses all the other
1894  *      protocol layers and you must therefore supply it with a complete frame
1895  */
1896
1897 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1898                                size_t len)
1899 {
1900         struct sock *sk = sock->sk;
1901         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1902         struct sk_buff *skb = NULL;
1903         struct net_device *dev;
1904         struct sockcm_cookie sockc;
1905         __be16 proto = 0;
1906         int err;
1907         int extra_len = 0;
1908
1909         /*
1910          *      Get and verify the address.
1911          */
1912
1913         if (saddr) {
1914                 if (msg->msg_namelen < sizeof(struct sockaddr))
1915                         return -EINVAL;
1916                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1917                         proto = saddr->spkt_protocol;
1918         } else
1919                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1920
1921         /*
1922          *      Find the device first to size check it
1923          */
1924
1925         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1926 retry:
1927         rcu_read_lock();
1928         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1929         err = -ENODEV;
1930         if (dev == NULL)
1931                 goto out_unlock;
1932
1933         err = -ENETDOWN;
1934         if (!(dev->flags & IFF_UP))
1935                 goto out_unlock;
1936
1937         /*
1938          * You may not queue a frame bigger than the mtu. This is the lowest level
1939          * raw protocol and you must do your own fragmentation at this level.
1940          */
1941
1942         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1943                 if (!netif_supports_nofcs(dev)) {
1944                         err = -EPROTONOSUPPORT;
1945                         goto out_unlock;
1946                 }
1947                 extra_len = 4; /* We're doing our own CRC */
1948         }
1949
1950         err = -EMSGSIZE;
1951         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1952                 goto out_unlock;
1953
1954         if (!skb) {
1955                 size_t reserved = LL_RESERVED_SPACE(dev);
1956                 int tlen = dev->needed_tailroom;
1957                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1958
1959                 rcu_read_unlock();
1960                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1961                 if (skb == NULL)
1962                         return -ENOBUFS;
1963                 /* FIXME: Save some space for broken drivers that write a hard
1964                  * header at transmission time by themselves. PPP is the notable
1965                  * one here. This should really be fixed at the driver level.
1966                  */
1967                 skb_reserve(skb, reserved);
1968                 skb_reset_network_header(skb);
1969
1970                 /* Try to align data part correctly */
1971                 if (hhlen) {
1972                         skb->data -= hhlen;
1973                         skb->tail -= hhlen;
1974                         if (len < hhlen)
1975                                 skb_reset_network_header(skb);
1976                 }
1977                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1978                 if (err)
1979                         goto out_free;
1980                 goto retry;
1981         }
1982
1983         if (!dev_validate_header(dev, skb->data, len)) {
1984                 err = -EINVAL;
1985                 goto out_unlock;
1986         }
1987         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1988             !packet_extra_vlan_len_allowed(dev, skb)) {
1989                 err = -EMSGSIZE;
1990                 goto out_unlock;
1991         }
1992
1993         sockc.tsflags = sk->sk_tsflags;
1994         if (msg->msg_controllen) {
1995                 err = sock_cmsg_send(sk, msg, &sockc);
1996                 if (unlikely(err))
1997                         goto out_unlock;
1998         }
1999
2000         skb->protocol = proto;
2001         skb->dev = dev;
2002         skb->priority = sk->sk_priority;
2003         skb->mark = sk->sk_mark;
2004
2005         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2006
2007         if (unlikely(extra_len == 4))
2008                 skb->no_fcs = 1;
2009
2010         skb_probe_transport_header(skb, 0);
2011
2012         dev_queue_xmit(skb);
2013         rcu_read_unlock();
2014         return len;
2015
2016 out_unlock:
2017         rcu_read_unlock();
2018 out_free:
2019         kfree_skb(skb);
2020         return err;
2021 }
2022
2023 static unsigned int run_filter(struct sk_buff *skb,
2024                                const struct sock *sk,
2025                                unsigned int res)
2026 {
2027         struct sk_filter *filter;
2028
2029         rcu_read_lock();
2030         filter = rcu_dereference(sk->sk_filter);
2031         if (filter != NULL)
2032                 res = bpf_prog_run_clear_cb(filter->prog, skb);
2033         rcu_read_unlock();
2034
2035         return res;
2036 }
2037
2038 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2039                            size_t *len)
2040 {
2041         struct virtio_net_hdr vnet_hdr;
2042
2043         if (*len < sizeof(vnet_hdr))
2044                 return -EINVAL;
2045         *len -= sizeof(vnet_hdr);
2046
2047         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
2048                 return -EINVAL;
2049
2050         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2051 }
2052
2053 /*
2054  * This function makes lazy skb cloning in hope that most of packets
2055  * are discarded by BPF.
2056  *
2057  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2058  * and skb->cb are mangled. It works because (and until) packets
2059  * falling here are owned by current CPU. Output packets are cloned
2060  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2061  * sequencially, so that if we return skb to original state on exit,
2062  * we will not harm anyone.
2063  */
2064
2065 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2066                       struct packet_type *pt, struct net_device *orig_dev)
2067 {
2068         struct sock *sk;
2069         struct sockaddr_ll *sll;
2070         struct packet_sock *po;
2071         u8 *skb_head = skb->data;
2072         int skb_len = skb->len;
2073         unsigned int snaplen, res;
2074         bool is_drop_n_account = false;
2075
2076         if (skb->pkt_type == PACKET_LOOPBACK)
2077                 goto drop;
2078
2079         sk = pt->af_packet_priv;
2080         po = pkt_sk(sk);
2081
2082         if (!net_eq(dev_net(dev), sock_net(sk)))
2083                 goto drop;
2084
2085         skb->dev = dev;
2086
2087         if (dev->header_ops) {
2088                 /* The device has an explicit notion of ll header,
2089                  * exported to higher levels.
2090                  *
2091                  * Otherwise, the device hides details of its frame
2092                  * structure, so that corresponding packet head is
2093                  * never delivered to user.
2094                  */
2095                 if (sk->sk_type != SOCK_DGRAM)
2096                         skb_push(skb, skb->data - skb_mac_header(skb));
2097                 else if (skb->pkt_type == PACKET_OUTGOING) {
2098                         /* Special case: outgoing packets have ll header at head */
2099                         skb_pull(skb, skb_network_offset(skb));
2100                 }
2101         }
2102
2103         snaplen = skb->len;
2104
2105         res = run_filter(skb, sk, snaplen);
2106         if (!res)
2107                 goto drop_n_restore;
2108         if (snaplen > res)
2109                 snaplen = res;
2110
2111         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2112                 goto drop_n_acct;
2113
2114         if (skb_shared(skb)) {
2115                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2116                 if (nskb == NULL)
2117                         goto drop_n_acct;
2118
2119                 if (skb_head != skb->data) {
2120                         skb->data = skb_head;
2121                         skb->len = skb_len;
2122                 }
2123                 consume_skb(skb);
2124                 skb = nskb;
2125         }
2126
2127         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2128
2129         sll = &PACKET_SKB_CB(skb)->sa.ll;
2130         sll->sll_hatype = dev->type;
2131         sll->sll_pkttype = skb->pkt_type;
2132         if (unlikely(po->origdev))
2133                 sll->sll_ifindex = orig_dev->ifindex;
2134         else
2135                 sll->sll_ifindex = dev->ifindex;
2136
2137         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2138
2139         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2140          * Use their space for storing the original skb length.
2141          */
2142         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2143
2144         if (pskb_trim(skb, snaplen))
2145                 goto drop_n_acct;
2146
2147         skb_set_owner_r(skb, sk);
2148         skb->dev = NULL;
2149         skb_dst_drop(skb);
2150
2151         /* drop conntrack reference */
2152         nf_reset(skb);
2153
2154         spin_lock(&sk->sk_receive_queue.lock);
2155         po->stats.stats1.tp_packets++;
2156         sock_skb_set_dropcount(sk, skb);
2157         __skb_queue_tail(&sk->sk_receive_queue, skb);
2158         spin_unlock(&sk->sk_receive_queue.lock);
2159         sk->sk_data_ready(sk);
2160         return 0;
2161
2162 drop_n_acct:
2163         is_drop_n_account = true;
2164         spin_lock(&sk->sk_receive_queue.lock);
2165         po->stats.stats1.tp_drops++;
2166         atomic_inc(&sk->sk_drops);
2167         spin_unlock(&sk->sk_receive_queue.lock);
2168
2169 drop_n_restore:
2170         if (skb_head != skb->data && skb_shared(skb)) {
2171                 skb->data = skb_head;
2172                 skb->len = skb_len;
2173         }
2174 drop:
2175         if (!is_drop_n_account)
2176                 consume_skb(skb);
2177         else
2178                 kfree_skb(skb);
2179         return 0;
2180 }
2181
2182 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2183                        struct packet_type *pt, struct net_device *orig_dev)
2184 {
2185         struct sock *sk;
2186         struct packet_sock *po;
2187         struct sockaddr_ll *sll;
2188         union tpacket_uhdr h;
2189         u8 *skb_head = skb->data;
2190         int skb_len = skb->len;
2191         unsigned int snaplen, res;
2192         unsigned long status = TP_STATUS_USER;
2193         unsigned short macoff, netoff, hdrlen;
2194         struct sk_buff *copy_skb = NULL;
2195         struct timespec ts;
2196         __u32 ts_status;
2197         bool is_drop_n_account = false;
2198         bool do_vnet = false;
2199
2200         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2201          * We may add members to them until current aligned size without forcing
2202          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2203          */
2204         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2205         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2206
2207         if (skb->pkt_type == PACKET_LOOPBACK)
2208                 goto drop;
2209
2210         sk = pt->af_packet_priv;
2211         po = pkt_sk(sk);
2212
2213         if (!net_eq(dev_net(dev), sock_net(sk)))
2214                 goto drop;
2215
2216         if (dev->header_ops) {
2217                 if (sk->sk_type != SOCK_DGRAM)
2218                         skb_push(skb, skb->data - skb_mac_header(skb));
2219                 else if (skb->pkt_type == PACKET_OUTGOING) {
2220                         /* Special case: outgoing packets have ll header at head */
2221                         skb_pull(skb, skb_network_offset(skb));
2222                 }
2223         }
2224
2225         snaplen = skb->len;
2226
2227         res = run_filter(skb, sk, snaplen);
2228         if (!res)
2229                 goto drop_n_restore;
2230
2231         if (skb->ip_summed == CHECKSUM_PARTIAL)
2232                 status |= TP_STATUS_CSUMNOTREADY;
2233         else if (skb->pkt_type != PACKET_OUTGOING &&
2234                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2235                   skb_csum_unnecessary(skb)))
2236                 status |= TP_STATUS_CSUM_VALID;
2237
2238         if (snaplen > res)
2239                 snaplen = res;
2240
2241         if (sk->sk_type == SOCK_DGRAM) {
2242                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2243                                   po->tp_reserve;
2244         } else {
2245                 unsigned int maclen = skb_network_offset(skb);
2246                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2247                                        (maclen < 16 ? 16 : maclen)) +
2248                                        po->tp_reserve;
2249                 if (po->has_vnet_hdr) {
2250                         netoff += sizeof(struct virtio_net_hdr);
2251                         do_vnet = true;
2252                 }
2253                 macoff = netoff - maclen;
2254         }
2255         if (po->tp_version <= TPACKET_V2) {
2256                 if (macoff + snaplen > po->rx_ring.frame_size) {
2257                         if (po->copy_thresh &&
2258                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2259                                 if (skb_shared(skb)) {
2260                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2261                                 } else {
2262                                         copy_skb = skb_get(skb);
2263                                         skb_head = skb->data;
2264                                 }
2265                                 if (copy_skb)
2266                                         skb_set_owner_r(copy_skb, sk);
2267                         }
2268                         snaplen = po->rx_ring.frame_size - macoff;
2269                         if ((int)snaplen < 0) {
2270                                 snaplen = 0;
2271                                 do_vnet = false;
2272                         }
2273                 }
2274         } else if (unlikely(macoff + snaplen >
2275                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2276                 u32 nval;
2277
2278                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2279                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2280                             snaplen, nval, macoff);
2281                 snaplen = nval;
2282                 if (unlikely((int)snaplen < 0)) {
2283                         snaplen = 0;
2284                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2285                         do_vnet = false;
2286                 }
2287         }
2288         spin_lock(&sk->sk_receive_queue.lock);
2289         h.raw = packet_current_rx_frame(po, skb,
2290                                         TP_STATUS_KERNEL, (macoff+snaplen));
2291         if (!h.raw)
2292                 goto drop_n_account;
2293         if (po->tp_version <= TPACKET_V2) {
2294                 packet_increment_rx_head(po, &po->rx_ring);
2295         /*
2296          * LOSING will be reported till you read the stats,
2297          * because it's COR - Clear On Read.
2298          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2299          * at packet level.
2300          */
2301                 if (po->stats.stats1.tp_drops)
2302                         status |= TP_STATUS_LOSING;
2303         }
2304         po->stats.stats1.tp_packets++;
2305         if (copy_skb) {
2306                 status |= TP_STATUS_COPY;
2307                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2308         }
2309         spin_unlock(&sk->sk_receive_queue.lock);
2310
2311         if (do_vnet) {
2312                 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2313                                             sizeof(struct virtio_net_hdr),
2314                                             vio_le(), true)) {
2315                         spin_lock(&sk->sk_receive_queue.lock);
2316                         goto drop_n_account;
2317                 }
2318         }
2319
2320         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2321
2322         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2323                 getnstimeofday(&ts);
2324
2325         status |= ts_status;
2326
2327         switch (po->tp_version) {
2328         case TPACKET_V1:
2329                 h.h1->tp_len = skb->len;
2330                 h.h1->tp_snaplen = snaplen;
2331                 h.h1->tp_mac = macoff;
2332                 h.h1->tp_net = netoff;
2333                 h.h1->tp_sec = ts.tv_sec;
2334                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2335                 hdrlen = sizeof(*h.h1);
2336                 break;
2337         case TPACKET_V2:
2338                 h.h2->tp_len = skb->len;
2339                 h.h2->tp_snaplen = snaplen;
2340                 h.h2->tp_mac = macoff;
2341                 h.h2->tp_net = netoff;
2342                 h.h2->tp_sec = ts.tv_sec;
2343                 h.h2->tp_nsec = ts.tv_nsec;
2344                 if (skb_vlan_tag_present(skb)) {
2345                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2346                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2347                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2348                 } else {
2349                         h.h2->tp_vlan_tci = 0;
2350                         h.h2->tp_vlan_tpid = 0;
2351                 }
2352                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2353                 hdrlen = sizeof(*h.h2);
2354                 break;
2355         case TPACKET_V3:
2356                 /* tp_nxt_offset,vlan are already populated above.
2357                  * So DONT clear those fields here
2358                  */
2359                 h.h3->tp_status |= status;
2360                 h.h3->tp_len = skb->len;
2361                 h.h3->tp_snaplen = snaplen;
2362                 h.h3->tp_mac = macoff;
2363                 h.h3->tp_net = netoff;
2364                 h.h3->tp_sec  = ts.tv_sec;
2365                 h.h3->tp_nsec = ts.tv_nsec;
2366                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2367                 hdrlen = sizeof(*h.h3);
2368                 break;
2369         default:
2370                 BUG();
2371         }
2372
2373         sll = h.raw + TPACKET_ALIGN(hdrlen);
2374         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2375         sll->sll_family = AF_PACKET;
2376         sll->sll_hatype = dev->type;
2377         sll->sll_protocol = skb->protocol;
2378         sll->sll_pkttype = skb->pkt_type;
2379         if (unlikely(po->origdev))
2380                 sll->sll_ifindex = orig_dev->ifindex;
2381         else
2382                 sll->sll_ifindex = dev->ifindex;
2383
2384         smp_mb();
2385
2386 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2387         if (po->tp_version <= TPACKET_V2) {
2388                 u8 *start, *end;
2389
2390                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2391                                         macoff + snaplen);
2392
2393                 for (start = h.raw; start < end; start += PAGE_SIZE)
2394                         flush_dcache_page(pgv_to_page(start));
2395         }
2396         smp_wmb();
2397 #endif
2398
2399         if (po->tp_version <= TPACKET_V2) {
2400                 __packet_set_status(po, h.raw, status);
2401                 sk->sk_data_ready(sk);
2402         } else {
2403                 prb_clear_blk_fill_status(&po->rx_ring);
2404         }
2405
2406 drop_n_restore:
2407         if (skb_head != skb->data && skb_shared(skb)) {
2408                 skb->data = skb_head;
2409                 skb->len = skb_len;
2410         }
2411 drop:
2412         if (!is_drop_n_account)
2413                 consume_skb(skb);
2414         else
2415                 kfree_skb(skb);
2416         return 0;
2417
2418 drop_n_account:
2419         is_drop_n_account = true;
2420         po->stats.stats1.tp_drops++;
2421         spin_unlock(&sk->sk_receive_queue.lock);
2422
2423         sk->sk_data_ready(sk);
2424         kfree_skb(copy_skb);
2425         goto drop_n_restore;
2426 }
2427
2428 static void tpacket_destruct_skb(struct sk_buff *skb)
2429 {
2430         struct packet_sock *po = pkt_sk(skb->sk);
2431
2432         if (likely(po->tx_ring.pg_vec)) {
2433                 void *ph;
2434                 __u32 ts;
2435
2436                 ph = skb_shinfo(skb)->destructor_arg;
2437                 packet_dec_pending(&po->tx_ring);
2438
2439                 ts = __packet_set_timestamp(po, ph, skb);
2440                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2441         }
2442
2443         sock_wfree(skb);
2444 }
2445
2446 static void tpacket_set_protocol(const struct net_device *dev,
2447                                  struct sk_buff *skb)
2448 {
2449         if (dev->type == ARPHRD_ETHER) {
2450                 skb_reset_mac_header(skb);
2451                 skb->protocol = eth_hdr(skb)->h_proto;
2452         }
2453 }
2454
2455 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2456 {
2457         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2458             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2459              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2460               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2461                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2462                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2463                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2464
2465         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2466                 return -EINVAL;
2467
2468         return 0;
2469 }
2470
2471 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2472                                  struct virtio_net_hdr *vnet_hdr)
2473 {
2474         if (*len < sizeof(*vnet_hdr))
2475                 return -EINVAL;
2476         *len -= sizeof(*vnet_hdr);
2477
2478         if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2479                 return -EFAULT;
2480
2481         return __packet_snd_vnet_parse(vnet_hdr, *len);
2482 }
2483
2484 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2485                 void *frame, struct net_device *dev, void *data, int tp_len,
2486                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2487                 const struct sockcm_cookie *sockc)
2488 {
2489         union tpacket_uhdr ph;
2490         int to_write, offset, len, nr_frags, len_max;
2491         struct socket *sock = po->sk.sk_socket;
2492         struct page *page;
2493         int err;
2494
2495         ph.raw = frame;
2496
2497         skb->protocol = proto;
2498         skb->dev = dev;
2499         skb->priority = po->sk.sk_priority;
2500         skb->mark = po->sk.sk_mark;
2501         sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2502         skb_shinfo(skb)->destructor_arg = ph.raw;
2503
2504         skb_reserve(skb, hlen);
2505         skb_reset_network_header(skb);
2506
2507         to_write = tp_len;
2508
2509         if (sock->type == SOCK_DGRAM) {
2510                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2511                                 NULL, tp_len);
2512                 if (unlikely(err < 0))
2513                         return -EINVAL;
2514         } else if (copylen) {
2515                 int hdrlen = min_t(int, copylen, tp_len);
2516
2517                 skb_push(skb, dev->hard_header_len);
2518                 skb_put(skb, copylen - dev->hard_header_len);
2519                 err = skb_store_bits(skb, 0, data, hdrlen);
2520                 if (unlikely(err))
2521                         return err;
2522                 if (!dev_validate_header(dev, skb->data, hdrlen))
2523                         return -EINVAL;
2524                 if (!skb->protocol)
2525                         tpacket_set_protocol(dev, skb);
2526
2527                 data += hdrlen;
2528                 to_write -= hdrlen;
2529         }
2530
2531         offset = offset_in_page(data);
2532         len_max = PAGE_SIZE - offset;
2533         len = ((to_write > len_max) ? len_max : to_write);
2534
2535         skb->data_len = to_write;
2536         skb->len += to_write;
2537         skb->truesize += to_write;
2538         refcount_add(to_write, &po->sk.sk_wmem_alloc);
2539
2540         while (likely(to_write)) {
2541                 nr_frags = skb_shinfo(skb)->nr_frags;
2542
2543                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2544                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2545                                MAX_SKB_FRAGS);
2546                         return -EFAULT;
2547                 }
2548
2549                 page = pgv_to_page(data);
2550                 data += len;
2551                 flush_dcache_page(page);
2552                 get_page(page);
2553                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2554                 to_write -= len;
2555                 offset = 0;
2556                 len_max = PAGE_SIZE;
2557                 len = ((to_write > len_max) ? len_max : to_write);
2558         }
2559
2560         skb_probe_transport_header(skb, 0);
2561
2562         return tp_len;
2563 }
2564
2565 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2566                                 int size_max, void **data)
2567 {
2568         union tpacket_uhdr ph;
2569         int tp_len, off;
2570
2571         ph.raw = frame;
2572
2573         switch (po->tp_version) {
2574         case TPACKET_V3:
2575                 if (ph.h3->tp_next_offset != 0) {
2576                         pr_warn_once("variable sized slot not supported");
2577                         return -EINVAL;
2578                 }
2579                 tp_len = ph.h3->tp_len;
2580                 break;
2581         case TPACKET_V2:
2582                 tp_len = ph.h2->tp_len;
2583                 break;
2584         default:
2585                 tp_len = ph.h1->tp_len;
2586                 break;
2587         }
2588         if (unlikely(tp_len > size_max)) {
2589                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2590                 return -EMSGSIZE;
2591         }
2592
2593         if (unlikely(po->tp_tx_has_off)) {
2594                 int off_min, off_max;
2595
2596                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2597                 off_max = po->tx_ring.frame_size - tp_len;
2598                 if (po->sk.sk_type == SOCK_DGRAM) {
2599                         switch (po->tp_version) {
2600                         case TPACKET_V3:
2601                                 off = ph.h3->tp_net;
2602                                 break;
2603                         case TPACKET_V2:
2604                                 off = ph.h2->tp_net;
2605                                 break;
2606                         default:
2607                                 off = ph.h1->tp_net;
2608                                 break;
2609                         }
2610                 } else {
2611                         switch (po->tp_version) {
2612                         case TPACKET_V3:
2613                                 off = ph.h3->tp_mac;
2614                                 break;
2615                         case TPACKET_V2:
2616                                 off = ph.h2->tp_mac;
2617                                 break;
2618                         default:
2619                                 off = ph.h1->tp_mac;
2620                                 break;
2621                         }
2622                 }
2623                 if (unlikely((off < off_min) || (off_max < off)))
2624                         return -EINVAL;
2625         } else {
2626                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2627         }
2628
2629         *data = frame + off;
2630         return tp_len;
2631 }
2632
2633 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2634 {
2635         struct sk_buff *skb;
2636         struct net_device *dev;
2637         struct virtio_net_hdr *vnet_hdr = NULL;
2638         struct sockcm_cookie sockc;
2639         __be16 proto;
2640         int err, reserve = 0;
2641         void *ph;
2642         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2643         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2644         int tp_len, size_max;
2645         unsigned char *addr;
2646         void *data;
2647         int len_sum = 0;
2648         int status = TP_STATUS_AVAILABLE;
2649         int hlen, tlen, copylen = 0;
2650
2651         mutex_lock(&po->pg_vec_lock);
2652
2653         if (likely(saddr == NULL)) {
2654                 dev     = packet_cached_dev_get(po);
2655                 proto   = po->num;
2656                 addr    = NULL;
2657         } else {
2658                 err = -EINVAL;
2659                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2660                         goto out;
2661                 if (msg->msg_namelen < (saddr->sll_halen
2662                                         + offsetof(struct sockaddr_ll,
2663                                                 sll_addr)))
2664                         goto out;
2665                 proto   = saddr->sll_protocol;
2666                 addr    = saddr->sll_addr;
2667                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2668         }
2669
2670         err = -ENXIO;
2671         if (unlikely(dev == NULL))
2672                 goto out;
2673         err = -ENETDOWN;
2674         if (unlikely(!(dev->flags & IFF_UP)))
2675                 goto out_put;
2676
2677         sockc.tsflags = po->sk.sk_tsflags;
2678         if (msg->msg_controllen) {
2679                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2680                 if (unlikely(err))
2681                         goto out_put;
2682         }
2683
2684         if (po->sk.sk_socket->type == SOCK_RAW)
2685                 reserve = dev->hard_header_len;
2686         size_max = po->tx_ring.frame_size
2687                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2688
2689         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2690                 size_max = dev->mtu + reserve + VLAN_HLEN;
2691
2692         do {
2693                 ph = packet_current_frame(po, &po->tx_ring,
2694                                           TP_STATUS_SEND_REQUEST);
2695                 if (unlikely(ph == NULL)) {
2696                         if (need_wait && need_resched())
2697                                 schedule();
2698                         continue;
2699                 }
2700
2701                 skb = NULL;
2702                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2703                 if (tp_len < 0)
2704                         goto tpacket_error;
2705
2706                 status = TP_STATUS_SEND_REQUEST;
2707                 hlen = LL_RESERVED_SPACE(dev);
2708                 tlen = dev->needed_tailroom;
2709                 if (po->has_vnet_hdr) {
2710                         vnet_hdr = data;
2711                         data += sizeof(*vnet_hdr);
2712                         tp_len -= sizeof(*vnet_hdr);
2713                         if (tp_len < 0 ||
2714                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2715                                 tp_len = -EINVAL;
2716                                 goto tpacket_error;
2717                         }
2718                         copylen = __virtio16_to_cpu(vio_le(),
2719                                                     vnet_hdr->hdr_len);
2720                 }
2721                 copylen = max_t(int, copylen, dev->hard_header_len);
2722                 skb = sock_alloc_send_skb(&po->sk,
2723                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2724                                 (copylen - dev->hard_header_len),
2725                                 !need_wait, &err);
2726
2727                 if (unlikely(skb == NULL)) {
2728                         /* we assume the socket was initially writeable ... */
2729                         if (likely(len_sum > 0))
2730                                 err = len_sum;
2731                         goto out_status;
2732                 }
2733                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2734                                           addr, hlen, copylen, &sockc);
2735                 if (likely(tp_len >= 0) &&
2736                     tp_len > dev->mtu + reserve &&
2737                     !po->has_vnet_hdr &&
2738                     !packet_extra_vlan_len_allowed(dev, skb))
2739                         tp_len = -EMSGSIZE;
2740
2741                 if (unlikely(tp_len < 0)) {
2742 tpacket_error:
2743                         if (po->tp_loss) {
2744                                 __packet_set_status(po, ph,
2745                                                 TP_STATUS_AVAILABLE);
2746                                 packet_increment_head(&po->tx_ring);
2747                                 kfree_skb(skb);
2748                                 continue;
2749                         } else {
2750                                 status = TP_STATUS_WRONG_FORMAT;
2751                                 err = tp_len;
2752                                 goto out_status;
2753                         }
2754                 }
2755
2756                 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2757                                                               vio_le())) {
2758                         tp_len = -EINVAL;
2759                         goto tpacket_error;
2760                 }
2761
2762                 skb->destructor = tpacket_destruct_skb;
2763                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2764                 packet_inc_pending(&po->tx_ring);
2765
2766                 status = TP_STATUS_SEND_REQUEST;
2767                 err = po->xmit(skb);
2768                 if (unlikely(err > 0)) {
2769                         err = net_xmit_errno(err);
2770                         if (err && __packet_get_status(po, ph) ==
2771                                    TP_STATUS_AVAILABLE) {
2772                                 /* skb was destructed already */
2773                                 skb = NULL;
2774                                 goto out_status;
2775                         }
2776                         /*
2777                          * skb was dropped but not destructed yet;
2778                          * let's treat it like congestion or err < 0
2779                          */
2780                         err = 0;
2781                 }
2782                 packet_increment_head(&po->tx_ring);
2783                 len_sum += tp_len;
2784         } while (likely((ph != NULL) ||
2785                 /* Note: packet_read_pending() might be slow if we have
2786                  * to call it as it's per_cpu variable, but in fast-path
2787                  * we already short-circuit the loop with the first
2788                  * condition, and luckily don't have to go that path
2789                  * anyway.
2790                  */
2791                  (need_wait && packet_read_pending(&po->tx_ring))));
2792
2793         err = len_sum;
2794         goto out_put;
2795
2796 out_status:
2797         __packet_set_status(po, ph, status);
2798         kfree_skb(skb);
2799 out_put:
2800         dev_put(dev);
2801 out:
2802         mutex_unlock(&po->pg_vec_lock);
2803         return err;
2804 }
2805
2806 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2807                                         size_t reserve, size_t len,
2808                                         size_t linear, int noblock,
2809                                         int *err)
2810 {
2811         struct sk_buff *skb;
2812
2813         /* Under a page?  Don't bother with paged skb. */
2814         if (prepad + len < PAGE_SIZE || !linear)
2815                 linear = len;
2816
2817         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2818                                    err, 0);
2819         if (!skb)
2820                 return NULL;
2821
2822         skb_reserve(skb, reserve);
2823         skb_put(skb, linear);
2824         skb->data_len = len - linear;
2825         skb->len += len - linear;
2826
2827         return skb;
2828 }
2829
2830 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2831 {
2832         struct sock *sk = sock->sk;
2833         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2834         struct sk_buff *skb;
2835         struct net_device *dev;
2836         __be16 proto;
2837         unsigned char *addr;
2838         int err, reserve = 0;
2839         struct sockcm_cookie sockc;
2840         struct virtio_net_hdr vnet_hdr = { 0 };
2841         int offset = 0;
2842         struct packet_sock *po = pkt_sk(sk);
2843         bool has_vnet_hdr = false;
2844         int hlen, tlen, linear;
2845         int extra_len = 0;
2846
2847         /*
2848          *      Get and verify the address.
2849          */
2850
2851         if (likely(saddr == NULL)) {
2852                 dev     = packet_cached_dev_get(po);
2853                 proto   = po->num;
2854                 addr    = NULL;
2855         } else {
2856                 err = -EINVAL;
2857                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2858                         goto out;
2859                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2860                         goto out;
2861                 proto   = saddr->sll_protocol;
2862                 addr    = saddr->sll_addr;
2863                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2864         }
2865
2866         err = -ENXIO;
2867         if (unlikely(dev == NULL))
2868                 goto out_unlock;
2869         err = -ENETDOWN;
2870         if (unlikely(!(dev->flags & IFF_UP)))
2871                 goto out_unlock;
2872
2873         sockc.tsflags = sk->sk_tsflags;
2874         sockc.mark = sk->sk_mark;
2875         if (msg->msg_controllen) {
2876                 err = sock_cmsg_send(sk, msg, &sockc);
2877                 if (unlikely(err))
2878                         goto out_unlock;
2879         }
2880
2881         if (sock->type == SOCK_RAW)
2882                 reserve = dev->hard_header_len;
2883         if (po->has_vnet_hdr) {
2884                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2885                 if (err)
2886                         goto out_unlock;
2887                 has_vnet_hdr = true;
2888         }
2889
2890         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2891                 if (!netif_supports_nofcs(dev)) {
2892                         err = -EPROTONOSUPPORT;
2893                         goto out_unlock;
2894                 }
2895                 extra_len = 4; /* We're doing our own CRC */
2896         }
2897
2898         err = -EMSGSIZE;
2899         if (!vnet_hdr.gso_type &&
2900             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2901                 goto out_unlock;
2902
2903         err = -ENOBUFS;
2904         hlen = LL_RESERVED_SPACE(dev);
2905         tlen = dev->needed_tailroom;
2906         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2907         linear = max(linear, min_t(int, len, dev->hard_header_len));
2908         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2909                                msg->msg_flags & MSG_DONTWAIT, &err);
2910         if (skb == NULL)
2911                 goto out_unlock;
2912
2913         skb_set_network_header(skb, reserve);
2914
2915         err = -EINVAL;
2916         if (sock->type == SOCK_DGRAM) {
2917                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2918                 if (unlikely(offset < 0))
2919                         goto out_free;
2920         }
2921
2922         /* Returns -EFAULT on error */
2923         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2924         if (err)
2925                 goto out_free;
2926
2927         if (sock->type == SOCK_RAW &&
2928             !dev_validate_header(dev, skb->data, len)) {
2929                 err = -EINVAL;
2930                 goto out_free;
2931         }
2932
2933         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2934
2935         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2936             !packet_extra_vlan_len_allowed(dev, skb)) {
2937                 err = -EMSGSIZE;
2938                 goto out_free;
2939         }
2940
2941         skb->protocol = proto;
2942         skb->dev = dev;
2943         skb->priority = sk->sk_priority;
2944         skb->mark = sockc.mark;
2945
2946         if (has_vnet_hdr) {
2947                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2948                 if (err)
2949                         goto out_free;
2950                 len += sizeof(vnet_hdr);
2951         }
2952
2953         skb_probe_transport_header(skb, reserve);
2954
2955         if (unlikely(extra_len == 4))
2956                 skb->no_fcs = 1;
2957
2958         err = po->xmit(skb);
2959         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2960                 goto out_unlock;
2961
2962         dev_put(dev);
2963
2964         return len;
2965
2966 out_free:
2967         kfree_skb(skb);
2968 out_unlock:
2969         if (dev)
2970                 dev_put(dev);
2971 out:
2972         return err;
2973 }
2974
2975 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2976 {
2977         struct sock *sk = sock->sk;
2978         struct packet_sock *po = pkt_sk(sk);
2979
2980         if (po->tx_ring.pg_vec)
2981                 return tpacket_snd(po, msg);
2982         else
2983                 return packet_snd(sock, msg, len);
2984 }
2985
2986 /*
2987  *      Close a PACKET socket. This is fairly simple. We immediately go
2988  *      to 'closed' state and remove our protocol entry in the device list.
2989  */
2990
2991 static int packet_release(struct socket *sock)
2992 {
2993         struct sock *sk = sock->sk;
2994         struct packet_sock *po;
2995         struct packet_fanout *f;
2996         struct net *net;
2997         union tpacket_req_u req_u;
2998
2999         if (!sk)
3000                 return 0;
3001
3002         net = sock_net(sk);
3003         po = pkt_sk(sk);
3004
3005         mutex_lock(&net->packet.sklist_lock);
3006         sk_del_node_init_rcu(sk);
3007         mutex_unlock(&net->packet.sklist_lock);
3008
3009         preempt_disable();
3010         sock_prot_inuse_add(net, sk->sk_prot, -1);
3011         preempt_enable();
3012
3013         spin_lock(&po->bind_lock);
3014         unregister_prot_hook(sk, false);
3015         packet_cached_dev_reset(po);
3016
3017         if (po->prot_hook.dev) {
3018                 dev_put(po->prot_hook.dev);
3019                 po->prot_hook.dev = NULL;
3020         }
3021         spin_unlock(&po->bind_lock);
3022
3023         packet_flush_mclist(sk);
3024
3025         if (po->rx_ring.pg_vec) {
3026                 memset(&req_u, 0, sizeof(req_u));
3027                 packet_set_ring(sk, &req_u, 1, 0);
3028         }
3029
3030         if (po->tx_ring.pg_vec) {
3031                 memset(&req_u, 0, sizeof(req_u));
3032                 packet_set_ring(sk, &req_u, 1, 1);
3033         }
3034
3035         f = fanout_release(sk);
3036
3037         synchronize_net();
3038
3039         if (f) {
3040                 fanout_release_data(f);
3041                 kfree(f);
3042         }
3043         /*
3044          *      Now the socket is dead. No more input will appear.
3045          */
3046         sock_orphan(sk);
3047         sock->sk = NULL;
3048
3049         /* Purge queues */
3050
3051         skb_queue_purge(&sk->sk_receive_queue);
3052         packet_free_pending(po);
3053         sk_refcnt_debug_release(sk);
3054
3055         sock_put(sk);
3056         return 0;
3057 }
3058
3059 /*
3060  *      Attach a packet hook.
3061  */
3062
3063 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3064                           __be16 proto)
3065 {
3066         struct packet_sock *po = pkt_sk(sk);
3067         struct net_device *dev_curr;
3068         __be16 proto_curr;
3069         bool need_rehook;
3070         struct net_device *dev = NULL;
3071         int ret = 0;
3072         bool unlisted = false;
3073
3074         lock_sock(sk);
3075         spin_lock(&po->bind_lock);
3076         rcu_read_lock();
3077
3078         if (po->fanout) {
3079                 ret = -EINVAL;
3080                 goto out_unlock;
3081         }
3082
3083         if (name) {
3084                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3085                 if (!dev) {
3086                         ret = -ENODEV;
3087                         goto out_unlock;
3088                 }
3089         } else if (ifindex) {
3090                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3091                 if (!dev) {
3092                         ret = -ENODEV;
3093                         goto out_unlock;
3094                 }
3095         }
3096
3097         if (dev)
3098                 dev_hold(dev);
3099
3100         proto_curr = po->prot_hook.type;
3101         dev_curr = po->prot_hook.dev;
3102
3103         need_rehook = proto_curr != proto || dev_curr != dev;
3104
3105         if (need_rehook) {
3106                 if (po->running) {
3107                         rcu_read_unlock();
3108                         __unregister_prot_hook(sk, true);
3109                         rcu_read_lock();
3110                         dev_curr = po->prot_hook.dev;
3111                         if (dev)
3112                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3113                                                                  dev->ifindex);
3114                 }
3115
3116                 po->num = proto;
3117                 po->prot_hook.type = proto;
3118
3119                 if (unlikely(unlisted)) {
3120                         dev_put(dev);
3121                         po->prot_hook.dev = NULL;
3122                         po->ifindex = -1;
3123                         packet_cached_dev_reset(po);
3124                 } else {
3125                         po->prot_hook.dev = dev;
3126                         po->ifindex = dev ? dev->ifindex : 0;
3127                         packet_cached_dev_assign(po, dev);
3128                 }
3129         }
3130         if (dev_curr)
3131                 dev_put(dev_curr);
3132
3133         if (proto == 0 || !need_rehook)
3134                 goto out_unlock;
3135
3136         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3137                 register_prot_hook(sk);
3138         } else {
3139                 sk->sk_err = ENETDOWN;
3140                 if (!sock_flag(sk, SOCK_DEAD))
3141                         sk->sk_error_report(sk);
3142         }
3143
3144 out_unlock:
3145         rcu_read_unlock();
3146         spin_unlock(&po->bind_lock);
3147         release_sock(sk);
3148         return ret;
3149 }
3150
3151 /*
3152  *      Bind a packet socket to a device
3153  */
3154
3155 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3156                             int addr_len)
3157 {
3158         struct sock *sk = sock->sk;
3159         char name[sizeof(uaddr->sa_data) + 1];
3160
3161         /*
3162          *      Check legality
3163          */
3164
3165         if (addr_len != sizeof(struct sockaddr))
3166                 return -EINVAL;
3167         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3168          * zero-terminated.
3169          */
3170         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3171         name[sizeof(uaddr->sa_data)] = 0;
3172
3173         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3174 }
3175
3176 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3177 {
3178         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3179         struct sock *sk = sock->sk;
3180
3181         /*
3182          *      Check legality
3183          */
3184
3185         if (addr_len < sizeof(struct sockaddr_ll))
3186                 return -EINVAL;
3187         if (sll->sll_family != AF_PACKET)
3188                 return -EINVAL;
3189
3190         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3191                               sll->sll_protocol ? : pkt_sk(sk)->num);
3192 }
3193
3194 static struct proto packet_proto = {
3195         .name     = "PACKET",
3196         .owner    = THIS_MODULE,
3197         .obj_size = sizeof(struct packet_sock),
3198 };
3199
3200 /*
3201  *      Create a packet of type SOCK_PACKET.
3202  */
3203
3204 static int packet_create(struct net *net, struct socket *sock, int protocol,
3205                          int kern)
3206 {
3207         struct sock *sk;
3208         struct packet_sock *po;
3209         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3210         int err;
3211
3212         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3213                 return -EPERM;
3214         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3215             sock->type != SOCK_PACKET)
3216                 return -ESOCKTNOSUPPORT;
3217
3218         sock->state = SS_UNCONNECTED;
3219
3220         err = -ENOBUFS;
3221         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3222         if (sk == NULL)
3223                 goto out;
3224
3225         sock->ops = &packet_ops;
3226         if (sock->type == SOCK_PACKET)
3227                 sock->ops = &packet_ops_spkt;
3228
3229         sock_init_data(sock, sk);
3230
3231         po = pkt_sk(sk);
3232         sk->sk_family = PF_PACKET;
3233         po->num = proto;
3234         po->xmit = dev_queue_xmit;
3235
3236         err = packet_alloc_pending(po);
3237         if (err)
3238                 goto out2;
3239
3240         packet_cached_dev_reset(po);
3241
3242         sk->sk_destruct = packet_sock_destruct;
3243         sk_refcnt_debug_inc(sk);
3244
3245         /*
3246          *      Attach a protocol block
3247          */
3248
3249         spin_lock_init(&po->bind_lock);
3250         mutex_init(&po->pg_vec_lock);
3251         po->rollover = NULL;
3252         po->prot_hook.func = packet_rcv;
3253
3254         if (sock->type == SOCK_PACKET)
3255                 po->prot_hook.func = packet_rcv_spkt;
3256
3257         po->prot_hook.af_packet_priv = sk;
3258
3259         if (proto) {
3260                 po->prot_hook.type = proto;
3261                 register_prot_hook(sk);
3262         }
3263
3264         mutex_lock(&net->packet.sklist_lock);
3265         sk_add_node_rcu(sk, &net->packet.sklist);
3266         mutex_unlock(&net->packet.sklist_lock);
3267
3268         preempt_disable();
3269         sock_prot_inuse_add(net, &packet_proto, 1);
3270         preempt_enable();
3271
3272         return 0;
3273 out2:
3274         sk_free(sk);
3275 out:
3276         return err;
3277 }
3278
3279 /*
3280  *      Pull a packet from our receive queue and hand it to the user.
3281  *      If necessary we block.
3282  */
3283
3284 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3285                           int flags)
3286 {
3287         struct sock *sk = sock->sk;
3288         struct sk_buff *skb;
3289         int copied, err;
3290         int vnet_hdr_len = 0;
3291         unsigned int origlen = 0;
3292
3293         err = -EINVAL;
3294         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3295                 goto out;
3296
3297 #if 0
3298         /* What error should we return now? EUNATTACH? */
3299         if (pkt_sk(sk)->ifindex < 0)
3300                 return -ENODEV;
3301 #endif
3302
3303         if (flags & MSG_ERRQUEUE) {
3304                 err = sock_recv_errqueue(sk, msg, len,
3305                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3306                 goto out;
3307         }
3308
3309         /*
3310          *      Call the generic datagram receiver. This handles all sorts
3311          *      of horrible races and re-entrancy so we can forget about it
3312          *      in the protocol layers.
3313          *
3314          *      Now it will return ENETDOWN, if device have just gone down,
3315          *      but then it will block.
3316          */
3317
3318         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3319
3320         /*
3321          *      An error occurred so return it. Because skb_recv_datagram()
3322          *      handles the blocking we don't see and worry about blocking
3323          *      retries.
3324          */
3325
3326         if (skb == NULL)
3327                 goto out;
3328
3329         if (pkt_sk(sk)->pressure)
3330                 packet_rcv_has_room(pkt_sk(sk), NULL);
3331
3332         if (pkt_sk(sk)->has_vnet_hdr) {
3333                 err = packet_rcv_vnet(msg, skb, &len);
3334                 if (err)
3335                         goto out_free;
3336                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3337         }
3338
3339         /* You lose any data beyond the buffer you gave. If it worries
3340          * a user program they can ask the device for its MTU
3341          * anyway.
3342          */
3343         copied = skb->len;
3344         if (copied > len) {
3345                 copied = len;
3346                 msg->msg_flags |= MSG_TRUNC;
3347         }
3348
3349         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3350         if (err)
3351                 goto out_free;
3352
3353         if (sock->type != SOCK_PACKET) {
3354                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3355
3356                 /* Original length was stored in sockaddr_ll fields */
3357                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3358                 sll->sll_family = AF_PACKET;
3359                 sll->sll_protocol = skb->protocol;
3360         }
3361
3362         sock_recv_ts_and_drops(msg, sk, skb);
3363
3364         if (msg->msg_name) {
3365                 /* If the address length field is there to be filled
3366                  * in, we fill it in now.
3367                  */
3368                 if (sock->type == SOCK_PACKET) {
3369                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3370                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3371                 } else {
3372                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3373
3374                         msg->msg_namelen = sll->sll_halen +
3375                                 offsetof(struct sockaddr_ll, sll_addr);
3376                 }
3377                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3378                        msg->msg_namelen);
3379         }
3380
3381         if (pkt_sk(sk)->auxdata) {
3382                 struct tpacket_auxdata aux;
3383
3384                 aux.tp_status = TP_STATUS_USER;
3385                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3386                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3387                 else if (skb->pkt_type != PACKET_OUTGOING &&
3388                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3389                           skb_csum_unnecessary(skb)))
3390                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3391
3392                 aux.tp_len = origlen;
3393                 aux.tp_snaplen = skb->len;
3394                 aux.tp_mac = 0;
3395                 aux.tp_net = skb_network_offset(skb);
3396                 if (skb_vlan_tag_present(skb)) {
3397                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3398                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3399                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3400                 } else {
3401                         aux.tp_vlan_tci = 0;
3402                         aux.tp_vlan_tpid = 0;
3403                 }
3404                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3405         }
3406
3407         /*
3408          *      Free or return the buffer as appropriate. Again this
3409          *      hides all the races and re-entrancy issues from us.
3410          */
3411         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3412
3413 out_free:
3414         skb_free_datagram(sk, skb);
3415 out:
3416         return err;
3417 }
3418
3419 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3420                                int *uaddr_len, int peer)
3421 {
3422         struct net_device *dev;
3423         struct sock *sk = sock->sk;
3424
3425         if (peer)
3426                 return -EOPNOTSUPP;
3427
3428         uaddr->sa_family = AF_PACKET;
3429         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3430         rcu_read_lock();
3431         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3432         if (dev)
3433                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3434         rcu_read_unlock();
3435         *uaddr_len = sizeof(*uaddr);
3436
3437         return 0;
3438 }
3439
3440 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3441                           int *uaddr_len, int peer)
3442 {
3443         struct net_device *dev;
3444         struct sock *sk = sock->sk;
3445         struct packet_sock *po = pkt_sk(sk);
3446         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3447
3448         if (peer)
3449                 return -EOPNOTSUPP;
3450
3451         sll->sll_family = AF_PACKET;
3452         sll->sll_ifindex = po->ifindex;
3453         sll->sll_protocol = po->num;
3454         sll->sll_pkttype = 0;
3455         rcu_read_lock();
3456         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3457         if (dev) {
3458                 sll->sll_hatype = dev->type;
3459                 sll->sll_halen = dev->addr_len;
3460                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3461         } else {
3462                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3463                 sll->sll_halen = 0;
3464         }
3465         rcu_read_unlock();
3466         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3467
3468         return 0;
3469 }
3470
3471 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3472                          int what)
3473 {
3474         switch (i->type) {
3475         case PACKET_MR_MULTICAST:
3476                 if (i->alen != dev->addr_len)
3477                         return -EINVAL;
3478                 if (what > 0)
3479                         return dev_mc_add(dev, i->addr);
3480                 else
3481                         return dev_mc_del(dev, i->addr);
3482                 break;
3483         case PACKET_MR_PROMISC:
3484                 return dev_set_promiscuity(dev, what);
3485         case PACKET_MR_ALLMULTI:
3486                 return dev_set_allmulti(dev, what);
3487         case PACKET_MR_UNICAST:
3488                 if (i->alen != dev->addr_len)
3489                         return -EINVAL;
3490                 if (what > 0)
3491                         return dev_uc_add(dev, i->addr);
3492                 else
3493                         return dev_uc_del(dev, i->addr);
3494                 break;
3495         default:
3496                 break;
3497         }
3498         return 0;
3499 }
3500
3501 static void packet_dev_mclist_delete(struct net_device *dev,
3502                                      struct packet_mclist **mlp)
3503 {
3504         struct packet_mclist *ml;
3505
3506         while ((ml = *mlp) != NULL) {
3507                 if (ml->ifindex == dev->ifindex) {
3508                         packet_dev_mc(dev, ml, -1);
3509                         *mlp = ml->next;
3510                         kfree(ml);
3511                 } else
3512                         mlp = &ml->next;
3513         }
3514 }
3515
3516 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3517 {
3518         struct packet_sock *po = pkt_sk(sk);
3519         struct packet_mclist *ml, *i;
3520         struct net_device *dev;
3521         int err;
3522
3523         rtnl_lock();
3524
3525         err = -ENODEV;
3526         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3527         if (!dev)
3528                 goto done;
3529
3530         err = -EINVAL;
3531         if (mreq->mr_alen > dev->addr_len)
3532                 goto done;
3533
3534         err = -ENOBUFS;
3535         i = kmalloc(sizeof(*i), GFP_KERNEL);
3536         if (i == NULL)
3537                 goto done;
3538
3539         err = 0;
3540         for (ml = po->mclist; ml; ml = ml->next) {
3541                 if (ml->ifindex == mreq->mr_ifindex &&
3542                     ml->type == mreq->mr_type &&
3543                     ml->alen == mreq->mr_alen &&
3544                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3545                         ml->count++;
3546                         /* Free the new element ... */
3547                         kfree(i);
3548                         goto done;
3549                 }
3550         }
3551
3552         i->type = mreq->mr_type;
3553         i->ifindex = mreq->mr_ifindex;
3554         i->alen = mreq->mr_alen;
3555         memcpy(i->addr, mreq->mr_address, i->alen);
3556         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3557         i->count = 1;
3558         i->next = po->mclist;
3559         po->mclist = i;
3560         err = packet_dev_mc(dev, i, 1);
3561         if (err) {
3562                 po->mclist = i->next;
3563                 kfree(i);
3564         }
3565
3566 done:
3567         rtnl_unlock();
3568         return err;
3569 }
3570
3571 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3572 {
3573         struct packet_mclist *ml, **mlp;
3574
3575         rtnl_lock();
3576
3577         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3578                 if (ml->ifindex == mreq->mr_ifindex &&
3579                     ml->type == mreq->mr_type &&
3580                     ml->alen == mreq->mr_alen &&
3581                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3582                         if (--ml->count == 0) {
3583                                 struct net_device *dev;
3584                                 *mlp = ml->next;
3585                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3586                                 if (dev)
3587                                         packet_dev_mc(dev, ml, -1);
3588                                 kfree(ml);
3589                         }
3590                         break;
3591                 }
3592         }
3593         rtnl_unlock();
3594         return 0;
3595 }
3596
3597 static void packet_flush_mclist(struct sock *sk)
3598 {
3599         struct packet_sock *po = pkt_sk(sk);
3600         struct packet_mclist *ml;
3601
3602         if (!po->mclist)
3603                 return;
3604
3605         rtnl_lock();
3606         while ((ml = po->mclist) != NULL) {
3607                 struct net_device *dev;
3608
3609                 po->mclist = ml->next;
3610                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3611                 if (dev != NULL)
3612                         packet_dev_mc(dev, ml, -1);
3613                 kfree(ml);
3614         }
3615         rtnl_unlock();
3616 }
3617
3618 static int
3619 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3620 {
3621         struct sock *sk = sock->sk;
3622         struct packet_sock *po = pkt_sk(sk);
3623         int ret;
3624
3625         if (level != SOL_PACKET)
3626                 return -ENOPROTOOPT;
3627
3628         switch (optname) {
3629         case PACKET_ADD_MEMBERSHIP:
3630         case PACKET_DROP_MEMBERSHIP:
3631         {
3632                 struct packet_mreq_max mreq;
3633                 int len = optlen;
3634                 memset(&mreq, 0, sizeof(mreq));
3635                 if (len < sizeof(struct packet_mreq))
3636                         return -EINVAL;
3637                 if (len > sizeof(mreq))
3638                         len = sizeof(mreq);
3639                 if (copy_from_user(&mreq, optval, len))
3640                         return -EFAULT;
3641                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3642                         return -EINVAL;
3643                 if (optname == PACKET_ADD_MEMBERSHIP)
3644                         ret = packet_mc_add(sk, &mreq);
3645                 else
3646                         ret = packet_mc_drop(sk, &mreq);
3647                 return ret;
3648         }
3649
3650         case PACKET_RX_RING:
3651         case PACKET_TX_RING:
3652         {
3653                 union tpacket_req_u req_u;
3654                 int len;
3655
3656                 switch (po->tp_version) {
3657                 case TPACKET_V1:
3658                 case TPACKET_V2:
3659                         len = sizeof(req_u.req);
3660                         break;
3661                 case TPACKET_V3:
3662                 default:
3663                         len = sizeof(req_u.req3);
3664                         break;
3665                 }
3666                 if (optlen < len)
3667                         return -EINVAL;
3668                 if (copy_from_user(&req_u.req, optval, len))
3669                         return -EFAULT;
3670                 return packet_set_ring(sk, &req_u, 0,
3671                         optname == PACKET_TX_RING);
3672         }
3673         case PACKET_COPY_THRESH:
3674         {
3675                 int val;
3676
3677                 if (optlen != sizeof(val))
3678                         return -EINVAL;
3679                 if (copy_from_user(&val, optval, sizeof(val)))
3680                         return -EFAULT;
3681
3682                 pkt_sk(sk)->copy_thresh = val;
3683                 return 0;
3684         }
3685         case PACKET_VERSION:
3686         {
3687                 int val;
3688
3689                 if (optlen != sizeof(val))
3690                         return -EINVAL;
3691                 if (copy_from_user(&val, optval, sizeof(val)))
3692                         return -EFAULT;
3693                 switch (val) {
3694                 case TPACKET_V1:
3695                 case TPACKET_V2:
3696                 case TPACKET_V3:
3697                         break;
3698                 default:
3699                         return -EINVAL;
3700                 }
3701                 lock_sock(sk);
3702                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3703                         ret = -EBUSY;
3704                 } else {
3705                         po->tp_version = val;
3706                         ret = 0;
3707                 }
3708                 release_sock(sk);
3709                 return ret;
3710         }
3711         case PACKET_RESERVE:
3712         {
3713                 unsigned int val;
3714
3715                 if (optlen != sizeof(val))
3716                         return -EINVAL;
3717                 if (copy_from_user(&val, optval, sizeof(val)))
3718                         return -EFAULT;
3719                 if (val > INT_MAX)
3720                         return -EINVAL;
3721                 lock_sock(sk);
3722                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3723                         ret = -EBUSY;
3724                 } else {
3725                         po->tp_reserve = val;
3726                         ret = 0;
3727                 }
3728                 release_sock(sk);
3729                 return ret;
3730         }
3731         case PACKET_LOSS:
3732         {
3733                 unsigned int val;
3734
3735                 if (optlen != sizeof(val))
3736                         return -EINVAL;
3737                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3738                         return -EBUSY;
3739                 if (copy_from_user(&val, optval, sizeof(val)))
3740                         return -EFAULT;
3741                 po->tp_loss = !!val;
3742                 return 0;
3743         }
3744         case PACKET_AUXDATA:
3745         {
3746                 int val;
3747
3748                 if (optlen < sizeof(val))
3749                         return -EINVAL;
3750                 if (copy_from_user(&val, optval, sizeof(val)))
3751                         return -EFAULT;
3752
3753                 po->auxdata = !!val;
3754                 return 0;
3755         }
3756         case PACKET_ORIGDEV:
3757         {
3758                 int val;
3759
3760                 if (optlen < sizeof(val))
3761                         return -EINVAL;
3762                 if (copy_from_user(&val, optval, sizeof(val)))
3763                         return -EFAULT;
3764
3765                 po->origdev = !!val;
3766                 return 0;
3767         }
3768         case PACKET_VNET_HDR:
3769         {
3770                 int val;
3771
3772                 if (sock->type != SOCK_RAW)
3773                         return -EINVAL;
3774                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3775                         return -EBUSY;
3776                 if (optlen < sizeof(val))
3777                         return -EINVAL;
3778                 if (copy_from_user(&val, optval, sizeof(val)))
3779                         return -EFAULT;
3780
3781                 po->has_vnet_hdr = !!val;
3782                 return 0;
3783         }
3784         case PACKET_TIMESTAMP:
3785         {
3786                 int val;
3787
3788                 if (optlen != sizeof(val))
3789                         return -EINVAL;
3790                 if (copy_from_user(&val, optval, sizeof(val)))
3791                         return -EFAULT;
3792
3793                 po->tp_tstamp = val;
3794                 return 0;
3795         }
3796         case PACKET_FANOUT:
3797         {
3798                 int val;
3799
3800                 if (optlen != sizeof(val))
3801                         return -EINVAL;
3802                 if (copy_from_user(&val, optval, sizeof(val)))
3803                         return -EFAULT;
3804
3805                 return fanout_add(sk, val & 0xffff, val >> 16);
3806         }
3807         case PACKET_FANOUT_DATA:
3808         {
3809                 if (!po->fanout)
3810                         return -EINVAL;
3811
3812                 return fanout_set_data(po, optval, optlen);
3813         }
3814         case PACKET_TX_HAS_OFF:
3815         {
3816                 unsigned int val;
3817
3818                 if (optlen != sizeof(val))
3819                         return -EINVAL;
3820                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3821                         return -EBUSY;
3822                 if (copy_from_user(&val, optval, sizeof(val)))
3823                         return -EFAULT;
3824                 po->tp_tx_has_off = !!val;
3825                 return 0;
3826         }
3827         case PACKET_QDISC_BYPASS:
3828         {
3829                 int val;
3830
3831                 if (optlen != sizeof(val))
3832                         return -EINVAL;
3833                 if (copy_from_user(&val, optval, sizeof(val)))
3834                         return -EFAULT;
3835
3836                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3837                 return 0;
3838         }
3839         default:
3840                 return -ENOPROTOOPT;
3841         }
3842 }
3843
3844 static int packet_getsockopt(struct socket *sock, int level, int optname,
3845                              char __user *optval, int __user *optlen)
3846 {
3847         int len;
3848         int val, lv = sizeof(val);
3849         struct sock *sk = sock->sk;
3850         struct packet_sock *po = pkt_sk(sk);
3851         void *data = &val;
3852         union tpacket_stats_u st;
3853         struct tpacket_rollover_stats rstats;
3854         struct packet_rollover *rollover;
3855
3856         if (level != SOL_PACKET)
3857                 return -ENOPROTOOPT;
3858
3859         if (get_user(len, optlen))
3860                 return -EFAULT;
3861
3862         if (len < 0)
3863                 return -EINVAL;
3864
3865         switch (optname) {
3866         case PACKET_STATISTICS:
3867                 spin_lock_bh(&sk->sk_receive_queue.lock);
3868                 memcpy(&st, &po->stats, sizeof(st));
3869                 memset(&po->stats, 0, sizeof(po->stats));
3870                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3871
3872                 if (po->tp_version == TPACKET_V3) {
3873                         lv = sizeof(struct tpacket_stats_v3);
3874                         st.stats3.tp_packets += st.stats3.tp_drops;
3875                         data = &st.stats3;
3876                 } else {
3877                         lv = sizeof(struct tpacket_stats);
3878                         st.stats1.tp_packets += st.stats1.tp_drops;
3879                         data = &st.stats1;
3880                 }
3881
3882                 break;
3883         case PACKET_AUXDATA:
3884                 val = po->auxdata;
3885                 break;
3886         case PACKET_ORIGDEV:
3887                 val = po->origdev;
3888                 break;
3889         case PACKET_VNET_HDR:
3890                 val = po->has_vnet_hdr;
3891                 break;
3892         case PACKET_VERSION:
3893                 val = po->tp_version;
3894                 break;
3895         case PACKET_HDRLEN:
3896                 if (len > sizeof(int))
3897                         len = sizeof(int);
3898                 if (len < sizeof(int))
3899                         return -EINVAL;
3900                 if (copy_from_user(&val, optval, len))
3901                         return -EFAULT;
3902                 switch (val) {
3903                 case TPACKET_V1:
3904                         val = sizeof(struct tpacket_hdr);
3905                         break;
3906                 case TPACKET_V2:
3907                         val = sizeof(struct tpacket2_hdr);
3908                         break;
3909                 case TPACKET_V3:
3910                         val = sizeof(struct tpacket3_hdr);
3911                         break;
3912                 default:
3913                         return -EINVAL;
3914                 }
3915                 break;
3916         case PACKET_RESERVE:
3917                 val = po->tp_reserve;
3918                 break;
3919         case PACKET_LOSS:
3920                 val = po->tp_loss;
3921                 break;
3922         case PACKET_TIMESTAMP:
3923                 val = po->tp_tstamp;
3924                 break;
3925         case PACKET_FANOUT:
3926                 val = (po->fanout ?
3927                        ((u32)po->fanout->id |
3928                         ((u32)po->fanout->type << 16) |
3929                         ((u32)po->fanout->flags << 24)) :
3930                        0);
3931                 break;
3932         case PACKET_ROLLOVER_STATS:
3933                 rcu_read_lock();
3934                 rollover = rcu_dereference(po->rollover);
3935                 if (rollover) {
3936                         rstats.tp_all = atomic_long_read(&rollover->num);
3937                         rstats.tp_huge = atomic_long_read(&rollover->num_huge);
3938                         rstats.tp_failed = atomic_long_read(&rollover->num_failed);
3939                         data = &rstats;
3940                         lv = sizeof(rstats);
3941                 }
3942                 rcu_read_unlock();
3943                 if (!rollover)
3944                         return -EINVAL;
3945                 break;
3946         case PACKET_TX_HAS_OFF:
3947                 val = po->tp_tx_has_off;
3948                 break;
3949         case PACKET_QDISC_BYPASS:
3950                 val = packet_use_direct_xmit(po);
3951                 break;
3952         default:
3953                 return -ENOPROTOOPT;
3954         }
3955
3956         if (len > lv)
3957                 len = lv;
3958         if (put_user(len, optlen))
3959                 return -EFAULT;
3960         if (copy_to_user(optval, data, len))
3961                 return -EFAULT;
3962         return 0;
3963 }
3964
3965
3966 #ifdef CONFIG_COMPAT
3967 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3968                                     char __user *optval, unsigned int optlen)
3969 {
3970         struct packet_sock *po = pkt_sk(sock->sk);
3971
3972         if (level != SOL_PACKET)
3973                 return -ENOPROTOOPT;
3974
3975         if (optname == PACKET_FANOUT_DATA &&
3976             po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3977                 optval = (char __user *)get_compat_bpf_fprog(optval);
3978                 if (!optval)
3979                         return -EFAULT;
3980                 optlen = sizeof(struct sock_fprog);
3981         }
3982
3983         return packet_setsockopt(sock, level, optname, optval, optlen);
3984 }
3985 #endif
3986
3987 static int packet_notifier(struct notifier_block *this,
3988                            unsigned long msg, void *ptr)
3989 {
3990         struct sock *sk;
3991         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3992         struct net *net = dev_net(dev);
3993
3994         rcu_read_lock();
3995         sk_for_each_rcu(sk, &net->packet.sklist) {
3996                 struct packet_sock *po = pkt_sk(sk);
3997
3998                 switch (msg) {
3999                 case NETDEV_UNREGISTER:
4000                         if (po->mclist)
4001                                 packet_dev_mclist_delete(dev, &po->mclist);
4002                         /* fallthrough */
4003
4004                 case NETDEV_DOWN:
4005                         if (dev->ifindex == po->ifindex) {
4006                                 spin_lock(&po->bind_lock);
4007                                 if (po->running) {
4008                                         __unregister_prot_hook(sk, false);
4009                                         sk->sk_err = ENETDOWN;
4010                                         if (!sock_flag(sk, SOCK_DEAD))
4011                                                 sk->sk_error_report(sk);
4012                                 }
4013                                 if (msg == NETDEV_UNREGISTER) {
4014                                         packet_cached_dev_reset(po);
4015                                         po->ifindex = -1;
4016                                         if (po->prot_hook.dev)
4017                                                 dev_put(po->prot_hook.dev);
4018                                         po->prot_hook.dev = NULL;
4019                                 }
4020                                 spin_unlock(&po->bind_lock);
4021                         }
4022                         break;
4023                 case NETDEV_UP:
4024                         if (dev->ifindex == po->ifindex) {
4025                                 spin_lock(&po->bind_lock);
4026                                 if (po->num)
4027                                         register_prot_hook(sk);
4028                                 spin_unlock(&po->bind_lock);
4029                         }
4030                         break;
4031                 }
4032         }
4033         rcu_read_unlock();
4034         return NOTIFY_DONE;
4035 }
4036
4037
4038 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4039                         unsigned long arg)
4040 {
4041         struct sock *sk = sock->sk;
4042
4043         switch (cmd) {
4044         case SIOCOUTQ:
4045         {
4046                 int amount = sk_wmem_alloc_get(sk);
4047
4048                 return put_user(amount, (int __user *)arg);
4049         }
4050         case SIOCINQ:
4051         {
4052                 struct sk_buff *skb;
4053                 int amount = 0;
4054
4055                 spin_lock_bh(&sk->sk_receive_queue.lock);
4056                 skb = skb_peek(&sk->sk_receive_queue);
4057                 if (skb)
4058                         amount = skb->len;
4059                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4060                 return put_user(amount, (int __user *)arg);
4061         }
4062         case SIOCGSTAMP:
4063                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4064         case SIOCGSTAMPNS:
4065                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4066
4067 #ifdef CONFIG_INET
4068         case SIOCADDRT:
4069         case SIOCDELRT:
4070         case SIOCDARP:
4071         case SIOCGARP:
4072         case SIOCSARP:
4073         case SIOCGIFADDR:
4074         case SIOCSIFADDR:
4075         case SIOCGIFBRDADDR:
4076         case SIOCSIFBRDADDR:
4077         case SIOCGIFNETMASK:
4078         case SIOCSIFNETMASK:
4079         case SIOCGIFDSTADDR:
4080         case SIOCSIFDSTADDR:
4081         case SIOCSIFFLAGS:
4082                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4083 #endif
4084
4085         default:
4086                 return -ENOIOCTLCMD;
4087         }
4088         return 0;
4089 }
4090
4091 static unsigned int packet_poll(struct file *file, struct socket *sock,
4092                                 poll_table *wait)
4093 {
4094         struct sock *sk = sock->sk;
4095         struct packet_sock *po = pkt_sk(sk);
4096         unsigned int mask = datagram_poll(file, sock, wait);
4097
4098         spin_lock_bh(&sk->sk_receive_queue.lock);
4099         if (po->rx_ring.pg_vec) {
4100                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4101                         TP_STATUS_KERNEL))
4102                         mask |= POLLIN | POLLRDNORM;
4103         }
4104         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4105                 po->pressure = 0;
4106         spin_unlock_bh(&sk->sk_receive_queue.lock);
4107         spin_lock_bh(&sk->sk_write_queue.lock);
4108         if (po->tx_ring.pg_vec) {
4109                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4110                         mask |= POLLOUT | POLLWRNORM;
4111         }
4112         spin_unlock_bh(&sk->sk_write_queue.lock);
4113         return mask;
4114 }
4115
4116
4117 /* Dirty? Well, I still did not learn better way to account
4118  * for user mmaps.
4119  */
4120
4121 static void packet_mm_open(struct vm_area_struct *vma)
4122 {
4123         struct file *file = vma->vm_file;
4124         struct socket *sock = file->private_data;
4125         struct sock *sk = sock->sk;
4126
4127         if (sk)
4128                 atomic_inc(&pkt_sk(sk)->mapped);
4129 }
4130
4131 static void packet_mm_close(struct vm_area_struct *vma)
4132 {
4133         struct file *file = vma->vm_file;
4134         struct socket *sock = file->private_data;
4135         struct sock *sk = sock->sk;
4136
4137         if (sk)
4138                 atomic_dec(&pkt_sk(sk)->mapped);
4139 }
4140
4141 static const struct vm_operations_struct packet_mmap_ops = {
4142         .open   =       packet_mm_open,
4143         .close  =       packet_mm_close,
4144 };
4145
4146 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4147                         unsigned int len)
4148 {
4149         int i;
4150
4151         for (i = 0; i < len; i++) {
4152                 if (likely(pg_vec[i].buffer)) {
4153                         if (is_vmalloc_addr(pg_vec[i].buffer))
4154                                 vfree(pg_vec[i].buffer);
4155                         else
4156                                 free_pages((unsigned long)pg_vec[i].buffer,
4157                                            order);
4158                         pg_vec[i].buffer = NULL;
4159                 }
4160         }
4161         kfree(pg_vec);
4162 }
4163
4164 static char *alloc_one_pg_vec_page(unsigned long order)
4165 {
4166         char *buffer;
4167         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4168                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4169
4170         buffer = (char *) __get_free_pages(gfp_flags, order);
4171         if (buffer)
4172                 return buffer;
4173
4174         /* __get_free_pages failed, fall back to vmalloc */
4175         buffer = vzalloc((1 << order) * PAGE_SIZE);
4176         if (buffer)
4177                 return buffer;
4178
4179         /* vmalloc failed, lets dig into swap here */
4180         gfp_flags &= ~__GFP_NORETRY;
4181         buffer = (char *) __get_free_pages(gfp_flags, order);
4182         if (buffer)
4183                 return buffer;
4184
4185         /* complete and utter failure */
4186         return NULL;
4187 }
4188
4189 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4190 {
4191         unsigned int block_nr = req->tp_block_nr;
4192         struct pgv *pg_vec;
4193         int i;
4194
4195         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4196         if (unlikely(!pg_vec))
4197                 goto out;
4198
4199         for (i = 0; i < block_nr; i++) {
4200                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4201                 if (unlikely(!pg_vec[i].buffer))
4202                         goto out_free_pgvec;
4203         }
4204
4205 out:
4206         return pg_vec;
4207
4208 out_free_pgvec:
4209         free_pg_vec(pg_vec, order, block_nr);
4210         pg_vec = NULL;
4211         goto out;
4212 }
4213
4214 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4215                 int closing, int tx_ring)
4216 {
4217         struct pgv *pg_vec = NULL;
4218         struct packet_sock *po = pkt_sk(sk);
4219         int was_running, order = 0;
4220         struct packet_ring_buffer *rb;
4221         struct sk_buff_head *rb_queue;
4222         __be16 num;
4223         int err = -EINVAL;
4224         /* Added to avoid minimal code churn */
4225         struct tpacket_req *req = &req_u->req;
4226
4227         lock_sock(sk);
4228
4229         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4230         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4231
4232         err = -EBUSY;
4233         if (!closing) {
4234                 if (atomic_read(&po->mapped))
4235                         goto out;
4236                 if (packet_read_pending(rb))
4237                         goto out;
4238         }
4239
4240         if (req->tp_block_nr) {
4241                 /* Sanity tests and some calculations */
4242                 err = -EBUSY;
4243                 if (unlikely(rb->pg_vec))
4244                         goto out;
4245
4246                 switch (po->tp_version) {
4247                 case TPACKET_V1:
4248                         po->tp_hdrlen = TPACKET_HDRLEN;
4249                         break;
4250                 case TPACKET_V2:
4251                         po->tp_hdrlen = TPACKET2_HDRLEN;
4252                         break;
4253                 case TPACKET_V3:
4254                         po->tp_hdrlen = TPACKET3_HDRLEN;
4255                         break;
4256                 }
4257
4258                 err = -EINVAL;
4259                 if (unlikely((int)req->tp_block_size <= 0))
4260                         goto out;
4261                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4262                         goto out;
4263                 if (po->tp_version >= TPACKET_V3 &&
4264                     req->tp_block_size <=
4265                           BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
4266                         goto out;
4267                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4268                                         po->tp_reserve))
4269                         goto out;
4270                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4271                         goto out;
4272
4273                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4274                 if (unlikely(rb->frames_per_block == 0))
4275                         goto out;
4276                 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4277                         goto out;
4278                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4279                                         req->tp_frame_nr))
4280                         goto out;
4281
4282                 err = -ENOMEM;
4283                 order = get_order(req->tp_block_size);
4284                 pg_vec = alloc_pg_vec(req, order);
4285                 if (unlikely(!pg_vec))
4286                         goto out;
4287                 switch (po->tp_version) {
4288                 case TPACKET_V3:
4289                         /* Block transmit is not supported yet */
4290                         if (!tx_ring) {
4291                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4292                         } else {
4293                                 struct tpacket_req3 *req3 = &req_u->req3;
4294
4295                                 if (req3->tp_retire_blk_tov ||
4296                                     req3->tp_sizeof_priv ||
4297                                     req3->tp_feature_req_word) {
4298                                         err = -EINVAL;
4299                                         goto out;
4300                                 }
4301                         }
4302                         break;
4303                 default:
4304                         break;
4305                 }
4306         }
4307         /* Done */
4308         else {
4309                 err = -EINVAL;
4310                 if (unlikely(req->tp_frame_nr))
4311                         goto out;
4312         }
4313
4314
4315         /* Detach socket from network */
4316         spin_lock(&po->bind_lock);
4317         was_running = po->running;
4318         num = po->num;
4319         if (was_running) {
4320                 po->num = 0;
4321                 __unregister_prot_hook(sk, false);
4322         }
4323         spin_unlock(&po->bind_lock);
4324
4325         synchronize_net();
4326
4327         err = -EBUSY;
4328         mutex_lock(&po->pg_vec_lock);
4329         if (closing || atomic_read(&po->mapped) == 0) {
4330                 err = 0;
4331                 spin_lock_bh(&rb_queue->lock);
4332                 swap(rb->pg_vec, pg_vec);
4333                 rb->frame_max = (req->tp_frame_nr - 1);
4334                 rb->head = 0;
4335                 rb->frame_size = req->tp_frame_size;
4336                 spin_unlock_bh(&rb_queue->lock);
4337
4338                 swap(rb->pg_vec_order, order);
4339                 swap(rb->pg_vec_len, req->tp_block_nr);
4340
4341                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4342                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4343                                                 tpacket_rcv : packet_rcv;
4344                 skb_queue_purge(rb_queue);
4345                 if (atomic_read(&po->mapped))
4346                         pr_err("packet_mmap: vma is busy: %d\n",
4347                                atomic_read(&po->mapped));
4348         }
4349         mutex_unlock(&po->pg_vec_lock);
4350
4351         spin_lock(&po->bind_lock);
4352         if (was_running) {
4353                 po->num = num;
4354                 register_prot_hook(sk);
4355         }
4356         spin_unlock(&po->bind_lock);
4357         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4358                 /* Because we don't support block-based V3 on tx-ring */
4359                 if (!tx_ring)
4360                         prb_shutdown_retire_blk_timer(po, rb_queue);
4361         }
4362
4363         if (pg_vec)
4364                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4365 out:
4366         release_sock(sk);
4367         return err;
4368 }
4369
4370 static int packet_mmap(struct file *file, struct socket *sock,
4371                 struct vm_area_struct *vma)
4372 {
4373         struct sock *sk = sock->sk;
4374         struct packet_sock *po = pkt_sk(sk);
4375         unsigned long size, expected_size;
4376         struct packet_ring_buffer *rb;
4377         unsigned long start;
4378         int err = -EINVAL;
4379         int i;
4380
4381         if (vma->vm_pgoff)
4382                 return -EINVAL;
4383
4384         mutex_lock(&po->pg_vec_lock);
4385
4386         expected_size = 0;
4387         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4388                 if (rb->pg_vec) {
4389                         expected_size += rb->pg_vec_len
4390                                                 * rb->pg_vec_pages
4391                                                 * PAGE_SIZE;
4392                 }
4393         }
4394
4395         if (expected_size == 0)
4396                 goto out;
4397
4398         size = vma->vm_end - vma->vm_start;
4399         if (size != expected_size)
4400                 goto out;
4401
4402         start = vma->vm_start;
4403         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4404                 if (rb->pg_vec == NULL)
4405                         continue;
4406
4407                 for (i = 0; i < rb->pg_vec_len; i++) {
4408                         struct page *page;
4409                         void *kaddr = rb->pg_vec[i].buffer;
4410                         int pg_num;
4411
4412                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4413                                 page = pgv_to_page(kaddr);
4414                                 err = vm_insert_page(vma, start, page);
4415                                 if (unlikely(err))
4416                                         goto out;
4417                                 start += PAGE_SIZE;
4418                                 kaddr += PAGE_SIZE;
4419                         }
4420                 }
4421         }
4422
4423         atomic_inc(&po->mapped);
4424         vma->vm_ops = &packet_mmap_ops;
4425         err = 0;
4426
4427 out:
4428         mutex_unlock(&po->pg_vec_lock);
4429         return err;
4430 }
4431
4432 static const struct proto_ops packet_ops_spkt = {
4433         .family =       PF_PACKET,
4434         .owner =        THIS_MODULE,
4435         .release =      packet_release,
4436         .bind =         packet_bind_spkt,
4437         .connect =      sock_no_connect,
4438         .socketpair =   sock_no_socketpair,
4439         .accept =       sock_no_accept,
4440         .getname =      packet_getname_spkt,
4441         .poll =         datagram_poll,
4442         .ioctl =        packet_ioctl,
4443         .listen =       sock_no_listen,
4444         .shutdown =     sock_no_shutdown,
4445         .setsockopt =   sock_no_setsockopt,
4446         .getsockopt =   sock_no_getsockopt,
4447         .sendmsg =      packet_sendmsg_spkt,
4448         .recvmsg =      packet_recvmsg,
4449         .mmap =         sock_no_mmap,
4450         .sendpage =     sock_no_sendpage,
4451 };
4452
4453 static const struct proto_ops packet_ops = {
4454         .family =       PF_PACKET,
4455         .owner =        THIS_MODULE,
4456         .release =      packet_release,
4457         .bind =         packet_bind,
4458         .connect =      sock_no_connect,
4459         .socketpair =   sock_no_socketpair,
4460         .accept =       sock_no_accept,
4461         .getname =      packet_getname,
4462         .poll =         packet_poll,
4463         .ioctl =        packet_ioctl,
4464         .listen =       sock_no_listen,
4465         .shutdown =     sock_no_shutdown,
4466         .setsockopt =   packet_setsockopt,
4467         .getsockopt =   packet_getsockopt,
4468 #ifdef CONFIG_COMPAT
4469         .compat_setsockopt = compat_packet_setsockopt,
4470 #endif
4471         .sendmsg =      packet_sendmsg,
4472         .recvmsg =      packet_recvmsg,
4473         .mmap =         packet_mmap,
4474         .sendpage =     sock_no_sendpage,
4475 };
4476
4477 static const struct net_proto_family packet_family_ops = {
4478         .family =       PF_PACKET,
4479         .create =       packet_create,
4480         .owner  =       THIS_MODULE,
4481 };
4482
4483 static struct notifier_block packet_netdev_notifier = {
4484         .notifier_call =        packet_notifier,
4485 };
4486
4487 #ifdef CONFIG_PROC_FS
4488
4489 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4490         __acquires(RCU)
4491 {
4492         struct net *net = seq_file_net(seq);
4493
4494         rcu_read_lock();
4495         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4496 }
4497
4498 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4499 {
4500         struct net *net = seq_file_net(seq);
4501         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4502 }
4503
4504 static void packet_seq_stop(struct seq_file *seq, void *v)
4505         __releases(RCU)
4506 {
4507         rcu_read_unlock();
4508 }
4509
4510 static int packet_seq_show(struct seq_file *seq, void *v)
4511 {
4512         if (v == SEQ_START_TOKEN)
4513                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4514         else {
4515                 struct sock *s = sk_entry(v);
4516                 const struct packet_sock *po = pkt_sk(s);
4517
4518                 seq_printf(seq,
4519                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4520                            s,
4521                            refcount_read(&s->sk_refcnt),
4522                            s->sk_type,
4523                            ntohs(po->num),
4524                            po->ifindex,
4525                            po->running,
4526                            atomic_read(&s->sk_rmem_alloc),
4527                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4528                            sock_i_ino(s));
4529         }
4530
4531         return 0;
4532 }
4533
4534 static const struct seq_operations packet_seq_ops = {
4535         .start  = packet_seq_start,
4536         .next   = packet_seq_next,
4537         .stop   = packet_seq_stop,
4538         .show   = packet_seq_show,
4539 };
4540
4541 static int packet_seq_open(struct inode *inode, struct file *file)
4542 {
4543         return seq_open_net(inode, file, &packet_seq_ops,
4544                             sizeof(struct seq_net_private));
4545 }
4546
4547 static const struct file_operations packet_seq_fops = {
4548         .owner          = THIS_MODULE,
4549         .open           = packet_seq_open,
4550         .read           = seq_read,
4551         .llseek         = seq_lseek,
4552         .release        = seq_release_net,
4553 };
4554
4555 #endif
4556
4557 static int __net_init packet_net_init(struct net *net)
4558 {
4559         mutex_init(&net->packet.sklist_lock);
4560         INIT_HLIST_HEAD(&net->packet.sklist);
4561
4562         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4563                 return -ENOMEM;
4564
4565         return 0;
4566 }
4567
4568 static void __net_exit packet_net_exit(struct net *net)
4569 {
4570         remove_proc_entry("packet", net->proc_net);
4571 }
4572
4573 static struct pernet_operations packet_net_ops = {
4574         .init = packet_net_init,
4575         .exit = packet_net_exit,
4576 };
4577
4578
4579 static void __exit packet_exit(void)
4580 {
4581         unregister_netdevice_notifier(&packet_netdev_notifier);
4582         unregister_pernet_subsys(&packet_net_ops);
4583         sock_unregister(PF_PACKET);
4584         proto_unregister(&packet_proto);
4585 }
4586
4587 static int __init packet_init(void)
4588 {
4589         int rc = proto_register(&packet_proto, 0);
4590
4591         if (rc != 0)
4592                 goto out;
4593
4594         sock_register(&packet_family_ops);
4595         register_pernet_subsys(&packet_net_ops);
4596         register_netdevice_notifier(&packet_netdev_notifier);
4597 out:
4598         return rc;
4599 }
4600
4601 module_init(packet_init);
4602 module_exit(packet_exit);
4603 MODULE_LICENSE("GPL");
4604 MODULE_ALIAS_NETPROTO(PF_PACKET);