Merge tag 'ieee802154-for-davem-2021-06-03' of git://git.kernel.org/pub/scm/linux...
[linux-2.6-microblaze.git] / net / packet / af_packet.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * INET         An implementation of the TCP/IP protocol suite for the LINUX
4  *              operating system.  INET is implemented using the  BSD Socket
5  *              interface as the means of communication with the user level.
6  *
7  *              PACKET - implements raw packet sockets.
8  *
9  * Authors:     Ross Biro
10  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
12  *
13  * Fixes:
14  *              Alan Cox        :       verify_area() now used correctly
15  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
16  *              Alan Cox        :       tidied skbuff lists.
17  *              Alan Cox        :       Now uses generic datagram routines I
18  *                                      added. Also fixed the peek/read crash
19  *                                      from all old Linux datagram code.
20  *              Alan Cox        :       Uses the improved datagram code.
21  *              Alan Cox        :       Added NULL's for socket options.
22  *              Alan Cox        :       Re-commented the code.
23  *              Alan Cox        :       Use new kernel side addressing
24  *              Rob Janssen     :       Correct MTU usage.
25  *              Dave Platt      :       Counter leaks caused by incorrect
26  *                                      interrupt locking and some slightly
27  *                                      dubious gcc output. Can you read
28  *                                      compiler: it said _VOLATILE_
29  *      Richard Kooijman        :       Timestamp fixes.
30  *              Alan Cox        :       New buffers. Use sk->mac.raw.
31  *              Alan Cox        :       sendmsg/recvmsg support.
32  *              Alan Cox        :       Protocol setting support
33  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
34  *      Cyrus Durgin            :       Fixed kerneld for kmod.
35  *      Michal Ostrowski        :       Module initialization cleanup.
36  *         Ulises Alonso        :       Frame number limit removal and
37  *                                      packet_set_ring memory leak.
38  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
39  *                                      The convention is that longer addresses
40  *                                      will simply extend the hardware address
41  *                                      byte arrays at the end of sockaddr_ll
42  *                                      and packet_mreq.
43  *              Johann Baudy    :       Added TX RING.
44  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
45  *                                      layer.
46  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
47  */
48
49 #include <linux/ethtool.h>
50 #include <linux/types.h>
51 #include <linux/mm.h>
52 #include <linux/capability.h>
53 #include <linux/fcntl.h>
54 #include <linux/socket.h>
55 #include <linux/in.h>
56 #include <linux/inet.h>
57 #include <linux/netdevice.h>
58 #include <linux/if_packet.h>
59 #include <linux/wireless.h>
60 #include <linux/kernel.h>
61 #include <linux/kmod.h>
62 #include <linux/slab.h>
63 #include <linux/vmalloc.h>
64 #include <net/net_namespace.h>
65 #include <net/ip.h>
66 #include <net/protocol.h>
67 #include <linux/skbuff.h>
68 #include <net/sock.h>
69 #include <linux/errno.h>
70 #include <linux/timer.h>
71 #include <linux/uaccess.h>
72 #include <asm/ioctls.h>
73 #include <asm/page.h>
74 #include <asm/cacheflush.h>
75 #include <asm/io.h>
76 #include <linux/proc_fs.h>
77 #include <linux/seq_file.h>
78 #include <linux/poll.h>
79 #include <linux/module.h>
80 #include <linux/init.h>
81 #include <linux/mutex.h>
82 #include <linux/if_vlan.h>
83 #include <linux/virtio_net.h>
84 #include <linux/errqueue.h>
85 #include <linux/net_tstamp.h>
86 #include <linux/percpu.h>
87 #ifdef CONFIG_INET
88 #include <net/inet_common.h>
89 #endif
90 #include <linux/bpf.h>
91 #include <net/compat.h>
92
93 #include "internal.h"
94
95 /*
96    Assumptions:
97    - If the device has no dev->header_ops->create, there is no LL header
98      visible above the device. In this case, its hard_header_len should be 0.
99      The device may prepend its own header internally. In this case, its
100      needed_headroom should be set to the space needed for it to add its
101      internal header.
102      For example, a WiFi driver pretending to be an Ethernet driver should
103      set its hard_header_len to be the Ethernet header length, and set its
104      needed_headroom to be (the real WiFi header length - the fake Ethernet
105      header length).
106    - packet socket receives packets with pulled ll header,
107      so that SOCK_RAW should push it back.
108
109 On receive:
110 -----------
111
112 Incoming, dev_has_header(dev) == true
113    mac_header -> ll header
114    data       -> data
115
116 Outgoing, dev_has_header(dev) == true
117    mac_header -> ll header
118    data       -> ll header
119
120 Incoming, dev_has_header(dev) == false
121    mac_header -> data
122      However drivers often make it point to the ll header.
123      This is incorrect because the ll header should be invisible to us.
124    data       -> data
125
126 Outgoing, dev_has_header(dev) == false
127    mac_header -> data. ll header is invisible to us.
128    data       -> data
129
130 Resume
131   If dev_has_header(dev) == false we are unable to restore the ll header,
132     because it is invisible to us.
133
134
135 On transmit:
136 ------------
137
138 dev_has_header(dev) == true
139    mac_header -> ll header
140    data       -> ll header
141
142 dev_has_header(dev) == false (ll header is invisible to us)
143    mac_header -> data
144    data       -> data
145
146    We should set network_header on output to the correct position,
147    packet classifier depends on it.
148  */
149
150 /* Private packet socket structures. */
151
152 /* identical to struct packet_mreq except it has
153  * a longer address field.
154  */
155 struct packet_mreq_max {
156         int             mr_ifindex;
157         unsigned short  mr_type;
158         unsigned short  mr_alen;
159         unsigned char   mr_address[MAX_ADDR_LEN];
160 };
161
162 union tpacket_uhdr {
163         struct tpacket_hdr  *h1;
164         struct tpacket2_hdr *h2;
165         struct tpacket3_hdr *h3;
166         void *raw;
167 };
168
169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170                 int closing, int tx_ring);
171
172 #define V3_ALIGNMENT    (8)
173
174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178
179 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
180 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
181 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
182 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
183 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
184 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
185
186 struct packet_sock;
187 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
188                        struct packet_type *pt, struct net_device *orig_dev);
189
190 static void *packet_previous_frame(struct packet_sock *po,
191                 struct packet_ring_buffer *rb,
192                 int status);
193 static void packet_increment_head(struct packet_ring_buffer *buff);
194 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
195 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
196                         struct packet_sock *);
197 static void prb_retire_current_block(struct tpacket_kbdq_core *,
198                 struct packet_sock *, unsigned int status);
199 static int prb_queue_frozen(struct tpacket_kbdq_core *);
200 static void prb_open_block(struct tpacket_kbdq_core *,
201                 struct tpacket_block_desc *);
202 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
203 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
204 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
205 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
206                 struct tpacket3_hdr *);
207 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
208                 struct tpacket3_hdr *);
209 static void packet_flush_mclist(struct sock *sk);
210 static u16 packet_pick_tx_queue(struct sk_buff *skb);
211
212 struct packet_skb_cb {
213         union {
214                 struct sockaddr_pkt pkt;
215                 union {
216                         /* Trick: alias skb original length with
217                          * ll.sll_family and ll.protocol in order
218                          * to save room.
219                          */
220                         unsigned int origlen;
221                         struct sockaddr_ll ll;
222                 };
223         } sa;
224 };
225
226 #define vio_le() virtio_legacy_is_little_endian()
227
228 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
229
230 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
231 #define GET_PBLOCK_DESC(x, bid) \
232         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
233 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
234         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
235 #define GET_NEXT_PRB_BLK_NUM(x) \
236         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
237         ((x)->kactive_blk_num+1) : 0)
238
239 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
240 static void __fanout_link(struct sock *sk, struct packet_sock *po);
241
242 static int packet_direct_xmit(struct sk_buff *skb)
243 {
244         return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
245 }
246
247 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
248 {
249         struct net_device *dev;
250
251         rcu_read_lock();
252         dev = rcu_dereference(po->cached_dev);
253         if (likely(dev))
254                 dev_hold(dev);
255         rcu_read_unlock();
256
257         return dev;
258 }
259
260 static void packet_cached_dev_assign(struct packet_sock *po,
261                                      struct net_device *dev)
262 {
263         rcu_assign_pointer(po->cached_dev, dev);
264 }
265
266 static void packet_cached_dev_reset(struct packet_sock *po)
267 {
268         RCU_INIT_POINTER(po->cached_dev, NULL);
269 }
270
271 static bool packet_use_direct_xmit(const struct packet_sock *po)
272 {
273         return po->xmit == packet_direct_xmit;
274 }
275
276 static u16 packet_pick_tx_queue(struct sk_buff *skb)
277 {
278         struct net_device *dev = skb->dev;
279         const struct net_device_ops *ops = dev->netdev_ops;
280         int cpu = raw_smp_processor_id();
281         u16 queue_index;
282
283 #ifdef CONFIG_XPS
284         skb->sender_cpu = cpu + 1;
285 #endif
286         skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
287         if (ops->ndo_select_queue) {
288                 queue_index = ops->ndo_select_queue(dev, skb, NULL);
289                 queue_index = netdev_cap_txqueue(dev, queue_index);
290         } else {
291                 queue_index = netdev_pick_tx(dev, skb, NULL);
292         }
293
294         return queue_index;
295 }
296
297 /* __register_prot_hook must be invoked through register_prot_hook
298  * or from a context in which asynchronous accesses to the packet
299  * socket is not possible (packet_create()).
300  */
301 static void __register_prot_hook(struct sock *sk)
302 {
303         struct packet_sock *po = pkt_sk(sk);
304
305         if (!po->running) {
306                 if (po->fanout)
307                         __fanout_link(sk, po);
308                 else
309                         dev_add_pack(&po->prot_hook);
310
311                 sock_hold(sk);
312                 po->running = 1;
313         }
314 }
315
316 static void register_prot_hook(struct sock *sk)
317 {
318         lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
319         __register_prot_hook(sk);
320 }
321
322 /* If the sync parameter is true, we will temporarily drop
323  * the po->bind_lock and do a synchronize_net to make sure no
324  * asynchronous packet processing paths still refer to the elements
325  * of po->prot_hook.  If the sync parameter is false, it is the
326  * callers responsibility to take care of this.
327  */
328 static void __unregister_prot_hook(struct sock *sk, bool sync)
329 {
330         struct packet_sock *po = pkt_sk(sk);
331
332         lockdep_assert_held_once(&po->bind_lock);
333
334         po->running = 0;
335
336         if (po->fanout)
337                 __fanout_unlink(sk, po);
338         else
339                 __dev_remove_pack(&po->prot_hook);
340
341         __sock_put(sk);
342
343         if (sync) {
344                 spin_unlock(&po->bind_lock);
345                 synchronize_net();
346                 spin_lock(&po->bind_lock);
347         }
348 }
349
350 static void unregister_prot_hook(struct sock *sk, bool sync)
351 {
352         struct packet_sock *po = pkt_sk(sk);
353
354         if (po->running)
355                 __unregister_prot_hook(sk, sync);
356 }
357
358 static inline struct page * __pure pgv_to_page(void *addr)
359 {
360         if (is_vmalloc_addr(addr))
361                 return vmalloc_to_page(addr);
362         return virt_to_page(addr);
363 }
364
365 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
366 {
367         union tpacket_uhdr h;
368
369         h.raw = frame;
370         switch (po->tp_version) {
371         case TPACKET_V1:
372                 h.h1->tp_status = status;
373                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
374                 break;
375         case TPACKET_V2:
376                 h.h2->tp_status = status;
377                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
378                 break;
379         case TPACKET_V3:
380                 h.h3->tp_status = status;
381                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
382                 break;
383         default:
384                 WARN(1, "TPACKET version not supported.\n");
385                 BUG();
386         }
387
388         smp_wmb();
389 }
390
391 static int __packet_get_status(const struct packet_sock *po, void *frame)
392 {
393         union tpacket_uhdr h;
394
395         smp_rmb();
396
397         h.raw = frame;
398         switch (po->tp_version) {
399         case TPACKET_V1:
400                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
401                 return h.h1->tp_status;
402         case TPACKET_V2:
403                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
404                 return h.h2->tp_status;
405         case TPACKET_V3:
406                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
407                 return h.h3->tp_status;
408         default:
409                 WARN(1, "TPACKET version not supported.\n");
410                 BUG();
411                 return 0;
412         }
413 }
414
415 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
416                                    unsigned int flags)
417 {
418         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
419
420         if (shhwtstamps &&
421             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
422             ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
423                 return TP_STATUS_TS_RAW_HARDWARE;
424
425         if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
426             ktime_to_timespec64_cond(skb->tstamp, ts))
427                 return TP_STATUS_TS_SOFTWARE;
428
429         return 0;
430 }
431
432 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
433                                     struct sk_buff *skb)
434 {
435         union tpacket_uhdr h;
436         struct timespec64 ts;
437         __u32 ts_status;
438
439         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
440                 return 0;
441
442         h.raw = frame;
443         /*
444          * versions 1 through 3 overflow the timestamps in y2106, since they
445          * all store the seconds in a 32-bit unsigned integer.
446          * If we create a version 4, that should have a 64-bit timestamp,
447          * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
448          * nanoseconds.
449          */
450         switch (po->tp_version) {
451         case TPACKET_V1:
452                 h.h1->tp_sec = ts.tv_sec;
453                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
454                 break;
455         case TPACKET_V2:
456                 h.h2->tp_sec = ts.tv_sec;
457                 h.h2->tp_nsec = ts.tv_nsec;
458                 break;
459         case TPACKET_V3:
460                 h.h3->tp_sec = ts.tv_sec;
461                 h.h3->tp_nsec = ts.tv_nsec;
462                 break;
463         default:
464                 WARN(1, "TPACKET version not supported.\n");
465                 BUG();
466         }
467
468         /* one flush is safe, as both fields always lie on the same cacheline */
469         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
470         smp_wmb();
471
472         return ts_status;
473 }
474
475 static void *packet_lookup_frame(const struct packet_sock *po,
476                                  const struct packet_ring_buffer *rb,
477                                  unsigned int position,
478                                  int status)
479 {
480         unsigned int pg_vec_pos, frame_offset;
481         union tpacket_uhdr h;
482
483         pg_vec_pos = position / rb->frames_per_block;
484         frame_offset = position % rb->frames_per_block;
485
486         h.raw = rb->pg_vec[pg_vec_pos].buffer +
487                 (frame_offset * rb->frame_size);
488
489         if (status != __packet_get_status(po, h.raw))
490                 return NULL;
491
492         return h.raw;
493 }
494
495 static void *packet_current_frame(struct packet_sock *po,
496                 struct packet_ring_buffer *rb,
497                 int status)
498 {
499         return packet_lookup_frame(po, rb, rb->head, status);
500 }
501
502 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
503 {
504         del_timer_sync(&pkc->retire_blk_timer);
505 }
506
507 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
508                 struct sk_buff_head *rb_queue)
509 {
510         struct tpacket_kbdq_core *pkc;
511
512         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
513
514         spin_lock_bh(&rb_queue->lock);
515         pkc->delete_blk_timer = 1;
516         spin_unlock_bh(&rb_queue->lock);
517
518         prb_del_retire_blk_timer(pkc);
519 }
520
521 static void prb_setup_retire_blk_timer(struct packet_sock *po)
522 {
523         struct tpacket_kbdq_core *pkc;
524
525         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
526         timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
527                     0);
528         pkc->retire_blk_timer.expires = jiffies;
529 }
530
531 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
532                                 int blk_size_in_bytes)
533 {
534         struct net_device *dev;
535         unsigned int mbits, div;
536         struct ethtool_link_ksettings ecmd;
537         int err;
538
539         rtnl_lock();
540         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
541         if (unlikely(!dev)) {
542                 rtnl_unlock();
543                 return DEFAULT_PRB_RETIRE_TOV;
544         }
545         err = __ethtool_get_link_ksettings(dev, &ecmd);
546         rtnl_unlock();
547         if (err)
548                 return DEFAULT_PRB_RETIRE_TOV;
549
550         /* If the link speed is so slow you don't really
551          * need to worry about perf anyways
552          */
553         if (ecmd.base.speed < SPEED_1000 ||
554             ecmd.base.speed == SPEED_UNKNOWN)
555                 return DEFAULT_PRB_RETIRE_TOV;
556
557         div = ecmd.base.speed / 1000;
558         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
559
560         if (div)
561                 mbits /= div;
562
563         if (div)
564                 return mbits + 1;
565         return mbits;
566 }
567
568 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
569                         union tpacket_req_u *req_u)
570 {
571         p1->feature_req_word = req_u->req3.tp_feature_req_word;
572 }
573
574 static void init_prb_bdqc(struct packet_sock *po,
575                         struct packet_ring_buffer *rb,
576                         struct pgv *pg_vec,
577                         union tpacket_req_u *req_u)
578 {
579         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
580         struct tpacket_block_desc *pbd;
581
582         memset(p1, 0x0, sizeof(*p1));
583
584         p1->knxt_seq_num = 1;
585         p1->pkbdq = pg_vec;
586         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
587         p1->pkblk_start = pg_vec[0].buffer;
588         p1->kblk_size = req_u->req3.tp_block_size;
589         p1->knum_blocks = req_u->req3.tp_block_nr;
590         p1->hdrlen = po->tp_hdrlen;
591         p1->version = po->tp_version;
592         p1->last_kactive_blk_num = 0;
593         po->stats.stats3.tp_freeze_q_cnt = 0;
594         if (req_u->req3.tp_retire_blk_tov)
595                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
596         else
597                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
598                                                 req_u->req3.tp_block_size);
599         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
600         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
601         rwlock_init(&p1->blk_fill_in_prog_lock);
602
603         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
604         prb_init_ft_ops(p1, req_u);
605         prb_setup_retire_blk_timer(po);
606         prb_open_block(p1, pbd);
607 }
608
609 /*  Do NOT update the last_blk_num first.
610  *  Assumes sk_buff_head lock is held.
611  */
612 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
613 {
614         mod_timer(&pkc->retire_blk_timer,
615                         jiffies + pkc->tov_in_jiffies);
616         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
617 }
618
619 /*
620  * Timer logic:
621  * 1) We refresh the timer only when we open a block.
622  *    By doing this we don't waste cycles refreshing the timer
623  *        on packet-by-packet basis.
624  *
625  * With a 1MB block-size, on a 1Gbps line, it will take
626  * i) ~8 ms to fill a block + ii) memcpy etc.
627  * In this cut we are not accounting for the memcpy time.
628  *
629  * So, if the user sets the 'tmo' to 10ms then the timer
630  * will never fire while the block is still getting filled
631  * (which is what we want). However, the user could choose
632  * to close a block early and that's fine.
633  *
634  * But when the timer does fire, we check whether or not to refresh it.
635  * Since the tmo granularity is in msecs, it is not too expensive
636  * to refresh the timer, lets say every '8' msecs.
637  * Either the user can set the 'tmo' or we can derive it based on
638  * a) line-speed and b) block-size.
639  * prb_calc_retire_blk_tmo() calculates the tmo.
640  *
641  */
642 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
643 {
644         struct packet_sock *po =
645                 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
646         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
647         unsigned int frozen;
648         struct tpacket_block_desc *pbd;
649
650         spin_lock(&po->sk.sk_receive_queue.lock);
651
652         frozen = prb_queue_frozen(pkc);
653         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
654
655         if (unlikely(pkc->delete_blk_timer))
656                 goto out;
657
658         /* We only need to plug the race when the block is partially filled.
659          * tpacket_rcv:
660          *              lock(); increment BLOCK_NUM_PKTS; unlock()
661          *              copy_bits() is in progress ...
662          *              timer fires on other cpu:
663          *              we can't retire the current block because copy_bits
664          *              is in progress.
665          *
666          */
667         if (BLOCK_NUM_PKTS(pbd)) {
668                 /* Waiting for skb_copy_bits to finish... */
669                 write_lock(&pkc->blk_fill_in_prog_lock);
670                 write_unlock(&pkc->blk_fill_in_prog_lock);
671         }
672
673         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
674                 if (!frozen) {
675                         if (!BLOCK_NUM_PKTS(pbd)) {
676                                 /* An empty block. Just refresh the timer. */
677                                 goto refresh_timer;
678                         }
679                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
680                         if (!prb_dispatch_next_block(pkc, po))
681                                 goto refresh_timer;
682                         else
683                                 goto out;
684                 } else {
685                         /* Case 1. Queue was frozen because user-space was
686                          *         lagging behind.
687                          */
688                         if (prb_curr_blk_in_use(pbd)) {
689                                 /*
690                                  * Ok, user-space is still behind.
691                                  * So just refresh the timer.
692                                  */
693                                 goto refresh_timer;
694                         } else {
695                                /* Case 2. queue was frozen,user-space caught up,
696                                 * now the link went idle && the timer fired.
697                                 * We don't have a block to close.So we open this
698                                 * block and restart the timer.
699                                 * opening a block thaws the queue,restarts timer
700                                 * Thawing/timer-refresh is a side effect.
701                                 */
702                                 prb_open_block(pkc, pbd);
703                                 goto out;
704                         }
705                 }
706         }
707
708 refresh_timer:
709         _prb_refresh_rx_retire_blk_timer(pkc);
710
711 out:
712         spin_unlock(&po->sk.sk_receive_queue.lock);
713 }
714
715 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
716                 struct tpacket_block_desc *pbd1, __u32 status)
717 {
718         /* Flush everything minus the block header */
719
720 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
721         u8 *start, *end;
722
723         start = (u8 *)pbd1;
724
725         /* Skip the block header(we know header WILL fit in 4K) */
726         start += PAGE_SIZE;
727
728         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
729         for (; start < end; start += PAGE_SIZE)
730                 flush_dcache_page(pgv_to_page(start));
731
732         smp_wmb();
733 #endif
734
735         /* Now update the block status. */
736
737         BLOCK_STATUS(pbd1) = status;
738
739         /* Flush the block header */
740
741 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
742         start = (u8 *)pbd1;
743         flush_dcache_page(pgv_to_page(start));
744
745         smp_wmb();
746 #endif
747 }
748
749 /*
750  * Side effect:
751  *
752  * 1) flush the block
753  * 2) Increment active_blk_num
754  *
755  * Note:We DONT refresh the timer on purpose.
756  *      Because almost always the next block will be opened.
757  */
758 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
759                 struct tpacket_block_desc *pbd1,
760                 struct packet_sock *po, unsigned int stat)
761 {
762         __u32 status = TP_STATUS_USER | stat;
763
764         struct tpacket3_hdr *last_pkt;
765         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
766         struct sock *sk = &po->sk;
767
768         if (atomic_read(&po->tp_drops))
769                 status |= TP_STATUS_LOSING;
770
771         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
772         last_pkt->tp_next_offset = 0;
773
774         /* Get the ts of the last pkt */
775         if (BLOCK_NUM_PKTS(pbd1)) {
776                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
777                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
778         } else {
779                 /* Ok, we tmo'd - so get the current time.
780                  *
781                  * It shouldn't really happen as we don't close empty
782                  * blocks. See prb_retire_rx_blk_timer_expired().
783                  */
784                 struct timespec64 ts;
785                 ktime_get_real_ts64(&ts);
786                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
787                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
788         }
789
790         smp_wmb();
791
792         /* Flush the block */
793         prb_flush_block(pkc1, pbd1, status);
794
795         sk->sk_data_ready(sk);
796
797         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
798 }
799
800 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
801 {
802         pkc->reset_pending_on_curr_blk = 0;
803 }
804
805 /*
806  * Side effect of opening a block:
807  *
808  * 1) prb_queue is thawed.
809  * 2) retire_blk_timer is refreshed.
810  *
811  */
812 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
813         struct tpacket_block_desc *pbd1)
814 {
815         struct timespec64 ts;
816         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
817
818         smp_rmb();
819
820         /* We could have just memset this but we will lose the
821          * flexibility of making the priv area sticky
822          */
823
824         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
825         BLOCK_NUM_PKTS(pbd1) = 0;
826         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
827
828         ktime_get_real_ts64(&ts);
829
830         h1->ts_first_pkt.ts_sec = ts.tv_sec;
831         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
832
833         pkc1->pkblk_start = (char *)pbd1;
834         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
835
836         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
838
839         pbd1->version = pkc1->version;
840         pkc1->prev = pkc1->nxt_offset;
841         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
842
843         prb_thaw_queue(pkc1);
844         _prb_refresh_rx_retire_blk_timer(pkc1);
845
846         smp_wmb();
847 }
848
849 /*
850  * Queue freeze logic:
851  * 1) Assume tp_block_nr = 8 blocks.
852  * 2) At time 't0', user opens Rx ring.
853  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854  * 4) user-space is either sleeping or processing block '0'.
855  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856  *    it will close block-7,loop around and try to fill block '0'.
857  *    call-flow:
858  *    __packet_lookup_frame_in_block
859  *      prb_retire_current_block()
860  *      prb_dispatch_next_block()
861  *        |->(BLOCK_STATUS == USER) evaluates to true
862  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
863  * 6) Now there are two cases:
864  *    6.1) Link goes idle right after the queue is frozen.
865  *         But remember, the last open_block() refreshed the timer.
866  *         When this timer expires,it will refresh itself so that we can
867  *         re-open block-0 in near future.
868  *    6.2) Link is busy and keeps on receiving packets. This is a simple
869  *         case and __packet_lookup_frame_in_block will check if block-0
870  *         is free and can now be re-used.
871  */
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873                                   struct packet_sock *po)
874 {
875         pkc->reset_pending_on_curr_blk = 1;
876         po->stats.stats3.tp_freeze_q_cnt++;
877 }
878
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880
881 /*
882  * If the next block is free then we will dispatch it
883  * and return a good offset.
884  * Else, we will freeze the queue.
885  * So, caller must check the return value.
886  */
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888                 struct packet_sock *po)
889 {
890         struct tpacket_block_desc *pbd;
891
892         smp_rmb();
893
894         /* 1. Get current block num */
895         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896
897         /* 2. If this block is currently in_use then freeze the queue */
898         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899                 prb_freeze_queue(pkc, po);
900                 return NULL;
901         }
902
903         /*
904          * 3.
905          * open this block and return the offset where the first packet
906          * needs to get stored.
907          */
908         prb_open_block(pkc, pbd);
909         return (void *)pkc->nxt_offset;
910 }
911
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913                 struct packet_sock *po, unsigned int status)
914 {
915         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916
917         /* retire/close the current block */
918         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919                 /*
920                  * Plug the case where copy_bits() is in progress on
921                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922                  * have space to copy the pkt in the current block and
923                  * called prb_retire_current_block()
924                  *
925                  * We don't need to worry about the TMO case because
926                  * the timer-handler already handled this case.
927                  */
928                 if (!(status & TP_STATUS_BLK_TMO)) {
929                         /* Waiting for skb_copy_bits to finish... */
930                         write_lock(&pkc->blk_fill_in_prog_lock);
931                         write_unlock(&pkc->blk_fill_in_prog_lock);
932                 }
933                 prb_close_block(pkc, pbd, po, status);
934                 return;
935         }
936 }
937
938 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
939 {
940         return TP_STATUS_USER & BLOCK_STATUS(pbd);
941 }
942
943 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
944 {
945         return pkc->reset_pending_on_curr_blk;
946 }
947
948 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
949         __releases(&pkc->blk_fill_in_prog_lock)
950 {
951         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
952
953         read_unlock(&pkc->blk_fill_in_prog_lock);
954 }
955
956 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
957                         struct tpacket3_hdr *ppd)
958 {
959         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
960 }
961
962 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
963                         struct tpacket3_hdr *ppd)
964 {
965         ppd->hv1.tp_rxhash = 0;
966 }
967
968 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
969                         struct tpacket3_hdr *ppd)
970 {
971         if (skb_vlan_tag_present(pkc->skb)) {
972                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
973                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
974                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
975         } else {
976                 ppd->hv1.tp_vlan_tci = 0;
977                 ppd->hv1.tp_vlan_tpid = 0;
978                 ppd->tp_status = TP_STATUS_AVAILABLE;
979         }
980 }
981
982 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
983                         struct tpacket3_hdr *ppd)
984 {
985         ppd->hv1.tp_padding = 0;
986         prb_fill_vlan_info(pkc, ppd);
987
988         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989                 prb_fill_rxhash(pkc, ppd);
990         else
991                 prb_clear_rxhash(pkc, ppd);
992 }
993
994 static void prb_fill_curr_block(char *curr,
995                                 struct tpacket_kbdq_core *pkc,
996                                 struct tpacket_block_desc *pbd,
997                                 unsigned int len)
998         __acquires(&pkc->blk_fill_in_prog_lock)
999 {
1000         struct tpacket3_hdr *ppd;
1001
1002         ppd  = (struct tpacket3_hdr *)curr;
1003         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1004         pkc->prev = curr;
1005         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1007         BLOCK_NUM_PKTS(pbd) += 1;
1008         read_lock(&pkc->blk_fill_in_prog_lock);
1009         prb_run_all_ft_ops(pkc, ppd);
1010 }
1011
1012 /* Assumes caller has the sk->rx_queue.lock */
1013 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1014                                             struct sk_buff *skb,
1015                                             unsigned int len
1016                                             )
1017 {
1018         struct tpacket_kbdq_core *pkc;
1019         struct tpacket_block_desc *pbd;
1020         char *curr, *end;
1021
1022         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1023         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024
1025         /* Queue is frozen when user space is lagging behind */
1026         if (prb_queue_frozen(pkc)) {
1027                 /*
1028                  * Check if that last block which caused the queue to freeze,
1029                  * is still in_use by user-space.
1030                  */
1031                 if (prb_curr_blk_in_use(pbd)) {
1032                         /* Can't record this packet */
1033                         return NULL;
1034                 } else {
1035                         /*
1036                          * Ok, the block was released by user-space.
1037                          * Now let's open that block.
1038                          * opening a block also thaws the queue.
1039                          * Thawing is a side effect.
1040                          */
1041                         prb_open_block(pkc, pbd);
1042                 }
1043         }
1044
1045         smp_mb();
1046         curr = pkc->nxt_offset;
1047         pkc->skb = skb;
1048         end = (char *)pbd + pkc->kblk_size;
1049
1050         /* first try the current block */
1051         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052                 prb_fill_curr_block(curr, pkc, pbd, len);
1053                 return (void *)curr;
1054         }
1055
1056         /* Ok, close the current block */
1057         prb_retire_current_block(pkc, po, 0);
1058
1059         /* Now, try to dispatch the next block */
1060         curr = (char *)prb_dispatch_next_block(pkc, po);
1061         if (curr) {
1062                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063                 prb_fill_curr_block(curr, pkc, pbd, len);
1064                 return (void *)curr;
1065         }
1066
1067         /*
1068          * No free blocks are available.user_space hasn't caught up yet.
1069          * Queue was just frozen and now this packet will get dropped.
1070          */
1071         return NULL;
1072 }
1073
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075                                             struct sk_buff *skb,
1076                                             int status, unsigned int len)
1077 {
1078         char *curr = NULL;
1079         switch (po->tp_version) {
1080         case TPACKET_V1:
1081         case TPACKET_V2:
1082                 curr = packet_lookup_frame(po, &po->rx_ring,
1083                                         po->rx_ring.head, status);
1084                 return curr;
1085         case TPACKET_V3:
1086                 return __packet_lookup_frame_in_block(po, skb, len);
1087         default:
1088                 WARN(1, "TPACKET version not supported\n");
1089                 BUG();
1090                 return NULL;
1091         }
1092 }
1093
1094 static void *prb_lookup_block(const struct packet_sock *po,
1095                               const struct packet_ring_buffer *rb,
1096                               unsigned int idx,
1097                               int status)
1098 {
1099         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1100         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1101
1102         if (status != BLOCK_STATUS(pbd))
1103                 return NULL;
1104         return pbd;
1105 }
1106
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1108 {
1109         unsigned int prev;
1110         if (rb->prb_bdqc.kactive_blk_num)
1111                 prev = rb->prb_bdqc.kactive_blk_num-1;
1112         else
1113                 prev = rb->prb_bdqc.knum_blocks-1;
1114         return prev;
1115 }
1116
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119                                          struct packet_ring_buffer *rb,
1120                                          int status)
1121 {
1122         unsigned int previous = prb_previous_blk_num(rb);
1123         return prb_lookup_block(po, rb, previous, status);
1124 }
1125
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127                                              struct packet_ring_buffer *rb,
1128                                              int status)
1129 {
1130         if (po->tp_version <= TPACKET_V2)
1131                 return packet_previous_frame(po, rb, status);
1132
1133         return __prb_previous_block(po, rb, status);
1134 }
1135
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137                                             struct packet_ring_buffer *rb)
1138 {
1139         switch (po->tp_version) {
1140         case TPACKET_V1:
1141         case TPACKET_V2:
1142                 return packet_increment_head(rb);
1143         case TPACKET_V3:
1144         default:
1145                 WARN(1, "TPACKET version not supported.\n");
1146                 BUG();
1147                 return;
1148         }
1149 }
1150
1151 static void *packet_previous_frame(struct packet_sock *po,
1152                 struct packet_ring_buffer *rb,
1153                 int status)
1154 {
1155         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156         return packet_lookup_frame(po, rb, previous, status);
1157 }
1158
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1160 {
1161         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1162 }
1163
1164 static void packet_inc_pending(struct packet_ring_buffer *rb)
1165 {
1166         this_cpu_inc(*rb->pending_refcnt);
1167 }
1168
1169 static void packet_dec_pending(struct packet_ring_buffer *rb)
1170 {
1171         this_cpu_dec(*rb->pending_refcnt);
1172 }
1173
1174 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1175 {
1176         unsigned int refcnt = 0;
1177         int cpu;
1178
1179         /* We don't use pending refcount in rx_ring. */
1180         if (rb->pending_refcnt == NULL)
1181                 return 0;
1182
1183         for_each_possible_cpu(cpu)
1184                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1185
1186         return refcnt;
1187 }
1188
1189 static int packet_alloc_pending(struct packet_sock *po)
1190 {
1191         po->rx_ring.pending_refcnt = NULL;
1192
1193         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1194         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1195                 return -ENOBUFS;
1196
1197         return 0;
1198 }
1199
1200 static void packet_free_pending(struct packet_sock *po)
1201 {
1202         free_percpu(po->tx_ring.pending_refcnt);
1203 }
1204
1205 #define ROOM_POW_OFF    2
1206 #define ROOM_NONE       0x0
1207 #define ROOM_LOW        0x1
1208 #define ROOM_NORMAL     0x2
1209
1210 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1211 {
1212         int idx, len;
1213
1214         len = READ_ONCE(po->rx_ring.frame_max) + 1;
1215         idx = READ_ONCE(po->rx_ring.head);
1216         if (pow_off)
1217                 idx += len >> pow_off;
1218         if (idx >= len)
1219                 idx -= len;
1220         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1221 }
1222
1223 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1224 {
1225         int idx, len;
1226
1227         len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1228         idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1229         if (pow_off)
1230                 idx += len >> pow_off;
1231         if (idx >= len)
1232                 idx -= len;
1233         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1234 }
1235
1236 static int __packet_rcv_has_room(const struct packet_sock *po,
1237                                  const struct sk_buff *skb)
1238 {
1239         const struct sock *sk = &po->sk;
1240         int ret = ROOM_NONE;
1241
1242         if (po->prot_hook.func != tpacket_rcv) {
1243                 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1244                 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1245                                    - (skb ? skb->truesize : 0);
1246
1247                 if (avail > (rcvbuf >> ROOM_POW_OFF))
1248                         return ROOM_NORMAL;
1249                 else if (avail > 0)
1250                         return ROOM_LOW;
1251                 else
1252                         return ROOM_NONE;
1253         }
1254
1255         if (po->tp_version == TPACKET_V3) {
1256                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1257                         ret = ROOM_NORMAL;
1258                 else if (__tpacket_v3_has_room(po, 0))
1259                         ret = ROOM_LOW;
1260         } else {
1261                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1262                         ret = ROOM_NORMAL;
1263                 else if (__tpacket_has_room(po, 0))
1264                         ret = ROOM_LOW;
1265         }
1266
1267         return ret;
1268 }
1269
1270 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1271 {
1272         int pressure, ret;
1273
1274         ret = __packet_rcv_has_room(po, skb);
1275         pressure = ret != ROOM_NORMAL;
1276
1277         if (READ_ONCE(po->pressure) != pressure)
1278                 WRITE_ONCE(po->pressure, pressure);
1279
1280         return ret;
1281 }
1282
1283 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1284 {
1285         if (READ_ONCE(po->pressure) &&
1286             __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1287                 WRITE_ONCE(po->pressure,  0);
1288 }
1289
1290 static void packet_sock_destruct(struct sock *sk)
1291 {
1292         skb_queue_purge(&sk->sk_error_queue);
1293
1294         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1295         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1296
1297         if (!sock_flag(sk, SOCK_DEAD)) {
1298                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1299                 return;
1300         }
1301
1302         sk_refcnt_debug_dec(sk);
1303 }
1304
1305 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1306 {
1307         u32 *history = po->rollover->history;
1308         u32 victim, rxhash;
1309         int i, count = 0;
1310
1311         rxhash = skb_get_hash(skb);
1312         for (i = 0; i < ROLLOVER_HLEN; i++)
1313                 if (READ_ONCE(history[i]) == rxhash)
1314                         count++;
1315
1316         victim = prandom_u32() % ROLLOVER_HLEN;
1317
1318         /* Avoid dirtying the cache line if possible */
1319         if (READ_ONCE(history[victim]) != rxhash)
1320                 WRITE_ONCE(history[victim], rxhash);
1321
1322         return count > (ROLLOVER_HLEN >> 1);
1323 }
1324
1325 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1326                                       struct sk_buff *skb,
1327                                       unsigned int num)
1328 {
1329         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1330 }
1331
1332 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1333                                     struct sk_buff *skb,
1334                                     unsigned int num)
1335 {
1336         unsigned int val = atomic_inc_return(&f->rr_cur);
1337
1338         return val % num;
1339 }
1340
1341 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1342                                      struct sk_buff *skb,
1343                                      unsigned int num)
1344 {
1345         return smp_processor_id() % num;
1346 }
1347
1348 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1349                                      struct sk_buff *skb,
1350                                      unsigned int num)
1351 {
1352         return prandom_u32_max(num);
1353 }
1354
1355 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1356                                           struct sk_buff *skb,
1357                                           unsigned int idx, bool try_self,
1358                                           unsigned int num)
1359 {
1360         struct packet_sock *po, *po_next, *po_skip = NULL;
1361         unsigned int i, j, room = ROOM_NONE;
1362
1363         po = pkt_sk(rcu_dereference(f->arr[idx]));
1364
1365         if (try_self) {
1366                 room = packet_rcv_has_room(po, skb);
1367                 if (room == ROOM_NORMAL ||
1368                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1369                         return idx;
1370                 po_skip = po;
1371         }
1372
1373         i = j = min_t(int, po->rollover->sock, num - 1);
1374         do {
1375                 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1376                 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1377                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1378                         if (i != j)
1379                                 po->rollover->sock = i;
1380                         atomic_long_inc(&po->rollover->num);
1381                         if (room == ROOM_LOW)
1382                                 atomic_long_inc(&po->rollover->num_huge);
1383                         return i;
1384                 }
1385
1386                 if (++i == num)
1387                         i = 0;
1388         } while (i != j);
1389
1390         atomic_long_inc(&po->rollover->num_failed);
1391         return idx;
1392 }
1393
1394 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1395                                     struct sk_buff *skb,
1396                                     unsigned int num)
1397 {
1398         return skb_get_queue_mapping(skb) % num;
1399 }
1400
1401 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1402                                      struct sk_buff *skb,
1403                                      unsigned int num)
1404 {
1405         struct bpf_prog *prog;
1406         unsigned int ret = 0;
1407
1408         rcu_read_lock();
1409         prog = rcu_dereference(f->bpf_prog);
1410         if (prog)
1411                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1412         rcu_read_unlock();
1413
1414         return ret;
1415 }
1416
1417 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1418 {
1419         return f->flags & (flag >> 8);
1420 }
1421
1422 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1423                              struct packet_type *pt, struct net_device *orig_dev)
1424 {
1425         struct packet_fanout *f = pt->af_packet_priv;
1426         unsigned int num = READ_ONCE(f->num_members);
1427         struct net *net = read_pnet(&f->net);
1428         struct packet_sock *po;
1429         unsigned int idx;
1430
1431         if (!net_eq(dev_net(dev), net) || !num) {
1432                 kfree_skb(skb);
1433                 return 0;
1434         }
1435
1436         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1437                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1438                 if (!skb)
1439                         return 0;
1440         }
1441         switch (f->type) {
1442         case PACKET_FANOUT_HASH:
1443         default:
1444                 idx = fanout_demux_hash(f, skb, num);
1445                 break;
1446         case PACKET_FANOUT_LB:
1447                 idx = fanout_demux_lb(f, skb, num);
1448                 break;
1449         case PACKET_FANOUT_CPU:
1450                 idx = fanout_demux_cpu(f, skb, num);
1451                 break;
1452         case PACKET_FANOUT_RND:
1453                 idx = fanout_demux_rnd(f, skb, num);
1454                 break;
1455         case PACKET_FANOUT_QM:
1456                 idx = fanout_demux_qm(f, skb, num);
1457                 break;
1458         case PACKET_FANOUT_ROLLOVER:
1459                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1460                 break;
1461         case PACKET_FANOUT_CBPF:
1462         case PACKET_FANOUT_EBPF:
1463                 idx = fanout_demux_bpf(f, skb, num);
1464                 break;
1465         }
1466
1467         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1468                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1469
1470         po = pkt_sk(rcu_dereference(f->arr[idx]));
1471         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1472 }
1473
1474 DEFINE_MUTEX(fanout_mutex);
1475 EXPORT_SYMBOL_GPL(fanout_mutex);
1476 static LIST_HEAD(fanout_list);
1477 static u16 fanout_next_id;
1478
1479 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1480 {
1481         struct packet_fanout *f = po->fanout;
1482
1483         spin_lock(&f->lock);
1484         rcu_assign_pointer(f->arr[f->num_members], sk);
1485         smp_wmb();
1486         f->num_members++;
1487         if (f->num_members == 1)
1488                 dev_add_pack(&f->prot_hook);
1489         spin_unlock(&f->lock);
1490 }
1491
1492 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1493 {
1494         struct packet_fanout *f = po->fanout;
1495         int i;
1496
1497         spin_lock(&f->lock);
1498         for (i = 0; i < f->num_members; i++) {
1499                 if (rcu_dereference_protected(f->arr[i],
1500                                               lockdep_is_held(&f->lock)) == sk)
1501                         break;
1502         }
1503         BUG_ON(i >= f->num_members);
1504         rcu_assign_pointer(f->arr[i],
1505                            rcu_dereference_protected(f->arr[f->num_members - 1],
1506                                                      lockdep_is_held(&f->lock)));
1507         f->num_members--;
1508         if (f->num_members == 0)
1509                 __dev_remove_pack(&f->prot_hook);
1510         spin_unlock(&f->lock);
1511 }
1512
1513 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1514 {
1515         if (sk->sk_family != PF_PACKET)
1516                 return false;
1517
1518         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1519 }
1520
1521 static void fanout_init_data(struct packet_fanout *f)
1522 {
1523         switch (f->type) {
1524         case PACKET_FANOUT_LB:
1525                 atomic_set(&f->rr_cur, 0);
1526                 break;
1527         case PACKET_FANOUT_CBPF:
1528         case PACKET_FANOUT_EBPF:
1529                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1530                 break;
1531         }
1532 }
1533
1534 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1535 {
1536         struct bpf_prog *old;
1537
1538         spin_lock(&f->lock);
1539         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1540         rcu_assign_pointer(f->bpf_prog, new);
1541         spin_unlock(&f->lock);
1542
1543         if (old) {
1544                 synchronize_net();
1545                 bpf_prog_destroy(old);
1546         }
1547 }
1548
1549 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1550                                 unsigned int len)
1551 {
1552         struct bpf_prog *new;
1553         struct sock_fprog fprog;
1554         int ret;
1555
1556         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1557                 return -EPERM;
1558
1559         ret = copy_bpf_fprog_from_user(&fprog, data, len);
1560         if (ret)
1561                 return ret;
1562
1563         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1564         if (ret)
1565                 return ret;
1566
1567         __fanout_set_data_bpf(po->fanout, new);
1568         return 0;
1569 }
1570
1571 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1572                                 unsigned int len)
1573 {
1574         struct bpf_prog *new;
1575         u32 fd;
1576
1577         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1578                 return -EPERM;
1579         if (len != sizeof(fd))
1580                 return -EINVAL;
1581         if (copy_from_sockptr(&fd, data, len))
1582                 return -EFAULT;
1583
1584         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1585         if (IS_ERR(new))
1586                 return PTR_ERR(new);
1587
1588         __fanout_set_data_bpf(po->fanout, new);
1589         return 0;
1590 }
1591
1592 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1593                            unsigned int len)
1594 {
1595         switch (po->fanout->type) {
1596         case PACKET_FANOUT_CBPF:
1597                 return fanout_set_data_cbpf(po, data, len);
1598         case PACKET_FANOUT_EBPF:
1599                 return fanout_set_data_ebpf(po, data, len);
1600         default:
1601                 return -EINVAL;
1602         }
1603 }
1604
1605 static void fanout_release_data(struct packet_fanout *f)
1606 {
1607         switch (f->type) {
1608         case PACKET_FANOUT_CBPF:
1609         case PACKET_FANOUT_EBPF:
1610                 __fanout_set_data_bpf(f, NULL);
1611         }
1612 }
1613
1614 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1615 {
1616         struct packet_fanout *f;
1617
1618         list_for_each_entry(f, &fanout_list, list) {
1619                 if (f->id == candidate_id &&
1620                     read_pnet(&f->net) == sock_net(sk)) {
1621                         return false;
1622                 }
1623         }
1624         return true;
1625 }
1626
1627 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1628 {
1629         u16 id = fanout_next_id;
1630
1631         do {
1632                 if (__fanout_id_is_free(sk, id)) {
1633                         *new_id = id;
1634                         fanout_next_id = id + 1;
1635                         return true;
1636                 }
1637
1638                 id++;
1639         } while (id != fanout_next_id);
1640
1641         return false;
1642 }
1643
1644 static int fanout_add(struct sock *sk, struct fanout_args *args)
1645 {
1646         struct packet_rollover *rollover = NULL;
1647         struct packet_sock *po = pkt_sk(sk);
1648         u16 type_flags = args->type_flags;
1649         struct packet_fanout *f, *match;
1650         u8 type = type_flags & 0xff;
1651         u8 flags = type_flags >> 8;
1652         u16 id = args->id;
1653         int err;
1654
1655         switch (type) {
1656         case PACKET_FANOUT_ROLLOVER:
1657                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1658                         return -EINVAL;
1659         case PACKET_FANOUT_HASH:
1660         case PACKET_FANOUT_LB:
1661         case PACKET_FANOUT_CPU:
1662         case PACKET_FANOUT_RND:
1663         case PACKET_FANOUT_QM:
1664         case PACKET_FANOUT_CBPF:
1665         case PACKET_FANOUT_EBPF:
1666                 break;
1667         default:
1668                 return -EINVAL;
1669         }
1670
1671         mutex_lock(&fanout_mutex);
1672
1673         err = -EALREADY;
1674         if (po->fanout)
1675                 goto out;
1676
1677         if (type == PACKET_FANOUT_ROLLOVER ||
1678             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1679                 err = -ENOMEM;
1680                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1681                 if (!rollover)
1682                         goto out;
1683                 atomic_long_set(&rollover->num, 0);
1684                 atomic_long_set(&rollover->num_huge, 0);
1685                 atomic_long_set(&rollover->num_failed, 0);
1686         }
1687
1688         if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1689                 if (id != 0) {
1690                         err = -EINVAL;
1691                         goto out;
1692                 }
1693                 if (!fanout_find_new_id(sk, &id)) {
1694                         err = -ENOMEM;
1695                         goto out;
1696                 }
1697                 /* ephemeral flag for the first socket in the group: drop it */
1698                 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1699         }
1700
1701         match = NULL;
1702         list_for_each_entry(f, &fanout_list, list) {
1703                 if (f->id == id &&
1704                     read_pnet(&f->net) == sock_net(sk)) {
1705                         match = f;
1706                         break;
1707                 }
1708         }
1709         err = -EINVAL;
1710         if (match) {
1711                 if (match->flags != flags)
1712                         goto out;
1713                 if (args->max_num_members &&
1714                     args->max_num_members != match->max_num_members)
1715                         goto out;
1716         } else {
1717                 if (args->max_num_members > PACKET_FANOUT_MAX)
1718                         goto out;
1719                 if (!args->max_num_members)
1720                         /* legacy PACKET_FANOUT_MAX */
1721                         args->max_num_members = 256;
1722                 err = -ENOMEM;
1723                 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1724                                  GFP_KERNEL);
1725                 if (!match)
1726                         goto out;
1727                 write_pnet(&match->net, sock_net(sk));
1728                 match->id = id;
1729                 match->type = type;
1730                 match->flags = flags;
1731                 INIT_LIST_HEAD(&match->list);
1732                 spin_lock_init(&match->lock);
1733                 refcount_set(&match->sk_ref, 0);
1734                 fanout_init_data(match);
1735                 match->prot_hook.type = po->prot_hook.type;
1736                 match->prot_hook.dev = po->prot_hook.dev;
1737                 match->prot_hook.func = packet_rcv_fanout;
1738                 match->prot_hook.af_packet_priv = match;
1739                 match->prot_hook.id_match = match_fanout_group;
1740                 match->max_num_members = args->max_num_members;
1741                 list_add(&match->list, &fanout_list);
1742         }
1743         err = -EINVAL;
1744
1745         spin_lock(&po->bind_lock);
1746         if (po->running &&
1747             match->type == type &&
1748             match->prot_hook.type == po->prot_hook.type &&
1749             match->prot_hook.dev == po->prot_hook.dev) {
1750                 err = -ENOSPC;
1751                 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1752                         __dev_remove_pack(&po->prot_hook);
1753                         po->fanout = match;
1754                         po->rollover = rollover;
1755                         rollover = NULL;
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                 kvfree(match);
1766         }
1767
1768 out:
1769         kfree(rollover);
1770         mutex_unlock(&fanout_mutex);
1771         return err;
1772 }
1773
1774 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1775  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1776  * It is the responsibility of the caller to call fanout_release_data() and
1777  * free the returned packet_fanout (after synchronize_net())
1778  */
1779 static struct packet_fanout *fanout_release(struct sock *sk)
1780 {
1781         struct packet_sock *po = pkt_sk(sk);
1782         struct packet_fanout *f;
1783
1784         mutex_lock(&fanout_mutex);
1785         f = po->fanout;
1786         if (f) {
1787                 po->fanout = NULL;
1788
1789                 if (refcount_dec_and_test(&f->sk_ref))
1790                         list_del(&f->list);
1791                 else
1792                         f = NULL;
1793         }
1794         mutex_unlock(&fanout_mutex);
1795
1796         return f;
1797 }
1798
1799 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1800                                           struct sk_buff *skb)
1801 {
1802         /* Earlier code assumed this would be a VLAN pkt, double-check
1803          * this now that we have the actual packet in hand. We can only
1804          * do this check on Ethernet devices.
1805          */
1806         if (unlikely(dev->type != ARPHRD_ETHER))
1807                 return false;
1808
1809         skb_reset_mac_header(skb);
1810         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1811 }
1812
1813 static const struct proto_ops packet_ops;
1814
1815 static const struct proto_ops packet_ops_spkt;
1816
1817 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1818                            struct packet_type *pt, struct net_device *orig_dev)
1819 {
1820         struct sock *sk;
1821         struct sockaddr_pkt *spkt;
1822
1823         /*
1824          *      When we registered the protocol we saved the socket in the data
1825          *      field for just this event.
1826          */
1827
1828         sk = pt->af_packet_priv;
1829
1830         /*
1831          *      Yank back the headers [hope the device set this
1832          *      right or kerboom...]
1833          *
1834          *      Incoming packets have ll header pulled,
1835          *      push it back.
1836          *
1837          *      For outgoing ones skb->data == skb_mac_header(skb)
1838          *      so that this procedure is noop.
1839          */
1840
1841         if (skb->pkt_type == PACKET_LOOPBACK)
1842                 goto out;
1843
1844         if (!net_eq(dev_net(dev), sock_net(sk)))
1845                 goto out;
1846
1847         skb = skb_share_check(skb, GFP_ATOMIC);
1848         if (skb == NULL)
1849                 goto oom;
1850
1851         /* drop any routing info */
1852         skb_dst_drop(skb);
1853
1854         /* drop conntrack reference */
1855         nf_reset_ct(skb);
1856
1857         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1858
1859         skb_push(skb, skb->data - skb_mac_header(skb));
1860
1861         /*
1862          *      The SOCK_PACKET socket receives _all_ frames.
1863          */
1864
1865         spkt->spkt_family = dev->type;
1866         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1867         spkt->spkt_protocol = skb->protocol;
1868
1869         /*
1870          *      Charge the memory to the socket. This is done specifically
1871          *      to prevent sockets using all the memory up.
1872          */
1873
1874         if (sock_queue_rcv_skb(sk, skb) == 0)
1875                 return 0;
1876
1877 out:
1878         kfree_skb(skb);
1879 oom:
1880         return 0;
1881 }
1882
1883 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1884 {
1885         if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1886             sock->type == SOCK_RAW) {
1887                 skb_reset_mac_header(skb);
1888                 skb->protocol = dev_parse_header_protocol(skb);
1889         }
1890
1891         skb_probe_transport_header(skb);
1892 }
1893
1894 /*
1895  *      Output a raw packet to a device layer. This bypasses all the other
1896  *      protocol layers and you must therefore supply it with a complete frame
1897  */
1898
1899 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1900                                size_t len)
1901 {
1902         struct sock *sk = sock->sk;
1903         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1904         struct sk_buff *skb = NULL;
1905         struct net_device *dev;
1906         struct sockcm_cookie sockc;
1907         __be16 proto = 0;
1908         int err;
1909         int extra_len = 0;
1910
1911         /*
1912          *      Get and verify the address.
1913          */
1914
1915         if (saddr) {
1916                 if (msg->msg_namelen < sizeof(struct sockaddr))
1917                         return -EINVAL;
1918                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1919                         proto = saddr->spkt_protocol;
1920         } else
1921                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1922
1923         /*
1924          *      Find the device first to size check it
1925          */
1926
1927         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1928 retry:
1929         rcu_read_lock();
1930         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1931         err = -ENODEV;
1932         if (dev == NULL)
1933                 goto out_unlock;
1934
1935         err = -ENETDOWN;
1936         if (!(dev->flags & IFF_UP))
1937                 goto out_unlock;
1938
1939         /*
1940          * You may not queue a frame bigger than the mtu. This is the lowest level
1941          * raw protocol and you must do your own fragmentation at this level.
1942          */
1943
1944         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1945                 if (!netif_supports_nofcs(dev)) {
1946                         err = -EPROTONOSUPPORT;
1947                         goto out_unlock;
1948                 }
1949                 extra_len = 4; /* We're doing our own CRC */
1950         }
1951
1952         err = -EMSGSIZE;
1953         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1954                 goto out_unlock;
1955
1956         if (!skb) {
1957                 size_t reserved = LL_RESERVED_SPACE(dev);
1958                 int tlen = dev->needed_tailroom;
1959                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1960
1961                 rcu_read_unlock();
1962                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1963                 if (skb == NULL)
1964                         return -ENOBUFS;
1965                 /* FIXME: Save some space for broken drivers that write a hard
1966                  * header at transmission time by themselves. PPP is the notable
1967                  * one here. This should really be fixed at the driver level.
1968                  */
1969                 skb_reserve(skb, reserved);
1970                 skb_reset_network_header(skb);
1971
1972                 /* Try to align data part correctly */
1973                 if (hhlen) {
1974                         skb->data -= hhlen;
1975                         skb->tail -= hhlen;
1976                         if (len < hhlen)
1977                                 skb_reset_network_header(skb);
1978                 }
1979                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1980                 if (err)
1981                         goto out_free;
1982                 goto retry;
1983         }
1984
1985         if (!dev_validate_header(dev, skb->data, len)) {
1986                 err = -EINVAL;
1987                 goto out_unlock;
1988         }
1989         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1990             !packet_extra_vlan_len_allowed(dev, skb)) {
1991                 err = -EMSGSIZE;
1992                 goto out_unlock;
1993         }
1994
1995         sockcm_init(&sockc, sk);
1996         if (msg->msg_controllen) {
1997                 err = sock_cmsg_send(sk, msg, &sockc);
1998                 if (unlikely(err))
1999                         goto out_unlock;
2000         }
2001
2002         skb->protocol = proto;
2003         skb->dev = dev;
2004         skb->priority = sk->sk_priority;
2005         skb->mark = sk->sk_mark;
2006         skb->tstamp = sockc.transmit_time;
2007
2008         skb_setup_tx_timestamp(skb, sockc.tsflags);
2009
2010         if (unlikely(extra_len == 4))
2011                 skb->no_fcs = 1;
2012
2013         packet_parse_headers(skb, sock);
2014
2015         dev_queue_xmit(skb);
2016         rcu_read_unlock();
2017         return len;
2018
2019 out_unlock:
2020         rcu_read_unlock();
2021 out_free:
2022         kfree_skb(skb);
2023         return err;
2024 }
2025
2026 static unsigned int run_filter(struct sk_buff *skb,
2027                                const struct sock *sk,
2028                                unsigned int res)
2029 {
2030         struct sk_filter *filter;
2031
2032         rcu_read_lock();
2033         filter = rcu_dereference(sk->sk_filter);
2034         if (filter != NULL)
2035                 res = bpf_prog_run_clear_cb(filter->prog, skb);
2036         rcu_read_unlock();
2037
2038         return res;
2039 }
2040
2041 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2042                            size_t *len)
2043 {
2044         struct virtio_net_hdr vnet_hdr;
2045
2046         if (*len < sizeof(vnet_hdr))
2047                 return -EINVAL;
2048         *len -= sizeof(vnet_hdr);
2049
2050         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2051                 return -EINVAL;
2052
2053         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2054 }
2055
2056 /*
2057  * This function makes lazy skb cloning in hope that most of packets
2058  * are discarded by BPF.
2059  *
2060  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2061  * and skb->cb are mangled. It works because (and until) packets
2062  * falling here are owned by current CPU. Output packets are cloned
2063  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2064  * sequentially, so that if we return skb to original state on exit,
2065  * we will not harm anyone.
2066  */
2067
2068 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2069                       struct packet_type *pt, struct net_device *orig_dev)
2070 {
2071         struct sock *sk;
2072         struct sockaddr_ll *sll;
2073         struct packet_sock *po;
2074         u8 *skb_head = skb->data;
2075         int skb_len = skb->len;
2076         unsigned int snaplen, res;
2077         bool is_drop_n_account = false;
2078
2079         if (skb->pkt_type == PACKET_LOOPBACK)
2080                 goto drop;
2081
2082         sk = pt->af_packet_priv;
2083         po = pkt_sk(sk);
2084
2085         if (!net_eq(dev_net(dev), sock_net(sk)))
2086                 goto drop;
2087
2088         skb->dev = dev;
2089
2090         if (dev_has_header(dev)) {
2091                 /* The device has an explicit notion of ll header,
2092                  * exported to higher levels.
2093                  *
2094                  * Otherwise, the device hides details of its frame
2095                  * structure, so that corresponding packet head is
2096                  * never delivered to user.
2097                  */
2098                 if (sk->sk_type != SOCK_DGRAM)
2099                         skb_push(skb, skb->data - skb_mac_header(skb));
2100                 else if (skb->pkt_type == PACKET_OUTGOING) {
2101                         /* Special case: outgoing packets have ll header at head */
2102                         skb_pull(skb, skb_network_offset(skb));
2103                 }
2104         }
2105
2106         snaplen = skb->len;
2107
2108         res = run_filter(skb, sk, snaplen);
2109         if (!res)
2110                 goto drop_n_restore;
2111         if (snaplen > res)
2112                 snaplen = res;
2113
2114         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2115                 goto drop_n_acct;
2116
2117         if (skb_shared(skb)) {
2118                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2119                 if (nskb == NULL)
2120                         goto drop_n_acct;
2121
2122                 if (skb_head != skb->data) {
2123                         skb->data = skb_head;
2124                         skb->len = skb_len;
2125                 }
2126                 consume_skb(skb);
2127                 skb = nskb;
2128         }
2129
2130         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2131
2132         sll = &PACKET_SKB_CB(skb)->sa.ll;
2133         sll->sll_hatype = dev->type;
2134         sll->sll_pkttype = skb->pkt_type;
2135         if (unlikely(po->origdev))
2136                 sll->sll_ifindex = orig_dev->ifindex;
2137         else
2138                 sll->sll_ifindex = dev->ifindex;
2139
2140         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2141
2142         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2143          * Use their space for storing the original skb length.
2144          */
2145         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2146
2147         if (pskb_trim(skb, snaplen))
2148                 goto drop_n_acct;
2149
2150         skb_set_owner_r(skb, sk);
2151         skb->dev = NULL;
2152         skb_dst_drop(skb);
2153
2154         /* drop conntrack reference */
2155         nf_reset_ct(skb);
2156
2157         spin_lock(&sk->sk_receive_queue.lock);
2158         po->stats.stats1.tp_packets++;
2159         sock_skb_set_dropcount(sk, skb);
2160         __skb_queue_tail(&sk->sk_receive_queue, skb);
2161         spin_unlock(&sk->sk_receive_queue.lock);
2162         sk->sk_data_ready(sk);
2163         return 0;
2164
2165 drop_n_acct:
2166         is_drop_n_account = true;
2167         atomic_inc(&po->tp_drops);
2168         atomic_inc(&sk->sk_drops);
2169
2170 drop_n_restore:
2171         if (skb_head != skb->data && skb_shared(skb)) {
2172                 skb->data = skb_head;
2173                 skb->len = skb_len;
2174         }
2175 drop:
2176         if (!is_drop_n_account)
2177                 consume_skb(skb);
2178         else
2179                 kfree_skb(skb);
2180         return 0;
2181 }
2182
2183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2184                        struct packet_type *pt, struct net_device *orig_dev)
2185 {
2186         struct sock *sk;
2187         struct packet_sock *po;
2188         struct sockaddr_ll *sll;
2189         union tpacket_uhdr h;
2190         u8 *skb_head = skb->data;
2191         int skb_len = skb->len;
2192         unsigned int snaplen, res;
2193         unsigned long status = TP_STATUS_USER;
2194         unsigned short macoff, hdrlen;
2195         unsigned int netoff;
2196         struct sk_buff *copy_skb = NULL;
2197         struct timespec64 ts;
2198         __u32 ts_status;
2199         bool is_drop_n_account = false;
2200         unsigned int slot_id = 0;
2201         bool do_vnet = false;
2202
2203         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2204          * We may add members to them until current aligned size without forcing
2205          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2206          */
2207         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2208         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2209
2210         if (skb->pkt_type == PACKET_LOOPBACK)
2211                 goto drop;
2212
2213         sk = pt->af_packet_priv;
2214         po = pkt_sk(sk);
2215
2216         if (!net_eq(dev_net(dev), sock_net(sk)))
2217                 goto drop;
2218
2219         if (dev_has_header(dev)) {
2220                 if (sk->sk_type != SOCK_DGRAM)
2221                         skb_push(skb, skb->data - skb_mac_header(skb));
2222                 else if (skb->pkt_type == PACKET_OUTGOING) {
2223                         /* Special case: outgoing packets have ll header at head */
2224                         skb_pull(skb, skb_network_offset(skb));
2225                 }
2226         }
2227
2228         snaplen = skb->len;
2229
2230         res = run_filter(skb, sk, snaplen);
2231         if (!res)
2232                 goto drop_n_restore;
2233
2234         /* If we are flooded, just give up */
2235         if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2236                 atomic_inc(&po->tp_drops);
2237                 goto drop_n_restore;
2238         }
2239
2240         if (skb->ip_summed == CHECKSUM_PARTIAL)
2241                 status |= TP_STATUS_CSUMNOTREADY;
2242         else if (skb->pkt_type != PACKET_OUTGOING &&
2243                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2244                   skb_csum_unnecessary(skb)))
2245                 status |= TP_STATUS_CSUM_VALID;
2246
2247         if (snaplen > res)
2248                 snaplen = res;
2249
2250         if (sk->sk_type == SOCK_DGRAM) {
2251                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2252                                   po->tp_reserve;
2253         } else {
2254                 unsigned int maclen = skb_network_offset(skb);
2255                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2256                                        (maclen < 16 ? 16 : maclen)) +
2257                                        po->tp_reserve;
2258                 if (po->has_vnet_hdr) {
2259                         netoff += sizeof(struct virtio_net_hdr);
2260                         do_vnet = true;
2261                 }
2262                 macoff = netoff - maclen;
2263         }
2264         if (netoff > USHRT_MAX) {
2265                 atomic_inc(&po->tp_drops);
2266                 goto drop_n_restore;
2267         }
2268         if (po->tp_version <= TPACKET_V2) {
2269                 if (macoff + snaplen > po->rx_ring.frame_size) {
2270                         if (po->copy_thresh &&
2271                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2272                                 if (skb_shared(skb)) {
2273                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2274                                 } else {
2275                                         copy_skb = skb_get(skb);
2276                                         skb_head = skb->data;
2277                                 }
2278                                 if (copy_skb)
2279                                         skb_set_owner_r(copy_skb, sk);
2280                         }
2281                         snaplen = po->rx_ring.frame_size - macoff;
2282                         if ((int)snaplen < 0) {
2283                                 snaplen = 0;
2284                                 do_vnet = false;
2285                         }
2286                 }
2287         } else if (unlikely(macoff + snaplen >
2288                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2289                 u32 nval;
2290
2291                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2292                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2293                             snaplen, nval, macoff);
2294                 snaplen = nval;
2295                 if (unlikely((int)snaplen < 0)) {
2296                         snaplen = 0;
2297                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2298                         do_vnet = false;
2299                 }
2300         }
2301         spin_lock(&sk->sk_receive_queue.lock);
2302         h.raw = packet_current_rx_frame(po, skb,
2303                                         TP_STATUS_KERNEL, (macoff+snaplen));
2304         if (!h.raw)
2305                 goto drop_n_account;
2306
2307         if (po->tp_version <= TPACKET_V2) {
2308                 slot_id = po->rx_ring.head;
2309                 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2310                         goto drop_n_account;
2311                 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2312         }
2313
2314         if (do_vnet &&
2315             virtio_net_hdr_from_skb(skb, h.raw + macoff -
2316                                     sizeof(struct virtio_net_hdr),
2317                                     vio_le(), true, 0)) {
2318                 if (po->tp_version == TPACKET_V3)
2319                         prb_clear_blk_fill_status(&po->rx_ring);
2320                 goto drop_n_account;
2321         }
2322
2323         if (po->tp_version <= TPACKET_V2) {
2324                 packet_increment_rx_head(po, &po->rx_ring);
2325         /*
2326          * LOSING will be reported till you read the stats,
2327          * because it's COR - Clear On Read.
2328          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2329          * at packet level.
2330          */
2331                 if (atomic_read(&po->tp_drops))
2332                         status |= TP_STATUS_LOSING;
2333         }
2334
2335         po->stats.stats1.tp_packets++;
2336         if (copy_skb) {
2337                 status |= TP_STATUS_COPY;
2338                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2339         }
2340         spin_unlock(&sk->sk_receive_queue.lock);
2341
2342         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2343
2344         /* Always timestamp; prefer an existing software timestamp taken
2345          * closer to the time of capture.
2346          */
2347         ts_status = tpacket_get_timestamp(skb, &ts,
2348                                           po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2349         if (!ts_status)
2350                 ktime_get_real_ts64(&ts);
2351
2352         status |= ts_status;
2353
2354         switch (po->tp_version) {
2355         case TPACKET_V1:
2356                 h.h1->tp_len = skb->len;
2357                 h.h1->tp_snaplen = snaplen;
2358                 h.h1->tp_mac = macoff;
2359                 h.h1->tp_net = netoff;
2360                 h.h1->tp_sec = ts.tv_sec;
2361                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2362                 hdrlen = sizeof(*h.h1);
2363                 break;
2364         case TPACKET_V2:
2365                 h.h2->tp_len = skb->len;
2366                 h.h2->tp_snaplen = snaplen;
2367                 h.h2->tp_mac = macoff;
2368                 h.h2->tp_net = netoff;
2369                 h.h2->tp_sec = ts.tv_sec;
2370                 h.h2->tp_nsec = ts.tv_nsec;
2371                 if (skb_vlan_tag_present(skb)) {
2372                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2373                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2374                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2375                 } else {
2376                         h.h2->tp_vlan_tci = 0;
2377                         h.h2->tp_vlan_tpid = 0;
2378                 }
2379                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2380                 hdrlen = sizeof(*h.h2);
2381                 break;
2382         case TPACKET_V3:
2383                 /* tp_nxt_offset,vlan are already populated above.
2384                  * So DONT clear those fields here
2385                  */
2386                 h.h3->tp_status |= status;
2387                 h.h3->tp_len = skb->len;
2388                 h.h3->tp_snaplen = snaplen;
2389                 h.h3->tp_mac = macoff;
2390                 h.h3->tp_net = netoff;
2391                 h.h3->tp_sec  = ts.tv_sec;
2392                 h.h3->tp_nsec = ts.tv_nsec;
2393                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2394                 hdrlen = sizeof(*h.h3);
2395                 break;
2396         default:
2397                 BUG();
2398         }
2399
2400         sll = h.raw + TPACKET_ALIGN(hdrlen);
2401         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2402         sll->sll_family = AF_PACKET;
2403         sll->sll_hatype = dev->type;
2404         sll->sll_protocol = skb->protocol;
2405         sll->sll_pkttype = skb->pkt_type;
2406         if (unlikely(po->origdev))
2407                 sll->sll_ifindex = orig_dev->ifindex;
2408         else
2409                 sll->sll_ifindex = dev->ifindex;
2410
2411         smp_mb();
2412
2413 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2414         if (po->tp_version <= TPACKET_V2) {
2415                 u8 *start, *end;
2416
2417                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2418                                         macoff + snaplen);
2419
2420                 for (start = h.raw; start < end; start += PAGE_SIZE)
2421                         flush_dcache_page(pgv_to_page(start));
2422         }
2423         smp_wmb();
2424 #endif
2425
2426         if (po->tp_version <= TPACKET_V2) {
2427                 spin_lock(&sk->sk_receive_queue.lock);
2428                 __packet_set_status(po, h.raw, status);
2429                 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2430                 spin_unlock(&sk->sk_receive_queue.lock);
2431                 sk->sk_data_ready(sk);
2432         } else if (po->tp_version == TPACKET_V3) {
2433                 prb_clear_blk_fill_status(&po->rx_ring);
2434         }
2435
2436 drop_n_restore:
2437         if (skb_head != skb->data && skb_shared(skb)) {
2438                 skb->data = skb_head;
2439                 skb->len = skb_len;
2440         }
2441 drop:
2442         if (!is_drop_n_account)
2443                 consume_skb(skb);
2444         else
2445                 kfree_skb(skb);
2446         return 0;
2447
2448 drop_n_account:
2449         spin_unlock(&sk->sk_receive_queue.lock);
2450         atomic_inc(&po->tp_drops);
2451         is_drop_n_account = true;
2452
2453         sk->sk_data_ready(sk);
2454         kfree_skb(copy_skb);
2455         goto drop_n_restore;
2456 }
2457
2458 static void tpacket_destruct_skb(struct sk_buff *skb)
2459 {
2460         struct packet_sock *po = pkt_sk(skb->sk);
2461
2462         if (likely(po->tx_ring.pg_vec)) {
2463                 void *ph;
2464                 __u32 ts;
2465
2466                 ph = skb_zcopy_get_nouarg(skb);
2467                 packet_dec_pending(&po->tx_ring);
2468
2469                 ts = __packet_set_timestamp(po, ph, skb);
2470                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2471
2472                 if (!packet_read_pending(&po->tx_ring))
2473                         complete(&po->skb_completion);
2474         }
2475
2476         sock_wfree(skb);
2477 }
2478
2479 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2480 {
2481         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2482             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2483              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2484               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2485                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2486                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2487                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2488
2489         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2490                 return -EINVAL;
2491
2492         return 0;
2493 }
2494
2495 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2496                                  struct virtio_net_hdr *vnet_hdr)
2497 {
2498         if (*len < sizeof(*vnet_hdr))
2499                 return -EINVAL;
2500         *len -= sizeof(*vnet_hdr);
2501
2502         if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2503                 return -EFAULT;
2504
2505         return __packet_snd_vnet_parse(vnet_hdr, *len);
2506 }
2507
2508 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2509                 void *frame, struct net_device *dev, void *data, int tp_len,
2510                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2511                 const struct sockcm_cookie *sockc)
2512 {
2513         union tpacket_uhdr ph;
2514         int to_write, offset, len, nr_frags, len_max;
2515         struct socket *sock = po->sk.sk_socket;
2516         struct page *page;
2517         int err;
2518
2519         ph.raw = frame;
2520
2521         skb->protocol = proto;
2522         skb->dev = dev;
2523         skb->priority = po->sk.sk_priority;
2524         skb->mark = po->sk.sk_mark;
2525         skb->tstamp = sockc->transmit_time;
2526         skb_setup_tx_timestamp(skb, sockc->tsflags);
2527         skb_zcopy_set_nouarg(skb, ph.raw);
2528
2529         skb_reserve(skb, hlen);
2530         skb_reset_network_header(skb);
2531
2532         to_write = tp_len;
2533
2534         if (sock->type == SOCK_DGRAM) {
2535                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2536                                 NULL, tp_len);
2537                 if (unlikely(err < 0))
2538                         return -EINVAL;
2539         } else if (copylen) {
2540                 int hdrlen = min_t(int, copylen, tp_len);
2541
2542                 skb_push(skb, dev->hard_header_len);
2543                 skb_put(skb, copylen - dev->hard_header_len);
2544                 err = skb_store_bits(skb, 0, data, hdrlen);
2545                 if (unlikely(err))
2546                         return err;
2547                 if (!dev_validate_header(dev, skb->data, hdrlen))
2548                         return -EINVAL;
2549
2550                 data += hdrlen;
2551                 to_write -= hdrlen;
2552         }
2553
2554         offset = offset_in_page(data);
2555         len_max = PAGE_SIZE - offset;
2556         len = ((to_write > len_max) ? len_max : to_write);
2557
2558         skb->data_len = to_write;
2559         skb->len += to_write;
2560         skb->truesize += to_write;
2561         refcount_add(to_write, &po->sk.sk_wmem_alloc);
2562
2563         while (likely(to_write)) {
2564                 nr_frags = skb_shinfo(skb)->nr_frags;
2565
2566                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2567                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2568                                MAX_SKB_FRAGS);
2569                         return -EFAULT;
2570                 }
2571
2572                 page = pgv_to_page(data);
2573                 data += len;
2574                 flush_dcache_page(page);
2575                 get_page(page);
2576                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2577                 to_write -= len;
2578                 offset = 0;
2579                 len_max = PAGE_SIZE;
2580                 len = ((to_write > len_max) ? len_max : to_write);
2581         }
2582
2583         packet_parse_headers(skb, sock);
2584
2585         return tp_len;
2586 }
2587
2588 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2589                                 int size_max, void **data)
2590 {
2591         union tpacket_uhdr ph;
2592         int tp_len, off;
2593
2594         ph.raw = frame;
2595
2596         switch (po->tp_version) {
2597         case TPACKET_V3:
2598                 if (ph.h3->tp_next_offset != 0) {
2599                         pr_warn_once("variable sized slot not supported");
2600                         return -EINVAL;
2601                 }
2602                 tp_len = ph.h3->tp_len;
2603                 break;
2604         case TPACKET_V2:
2605                 tp_len = ph.h2->tp_len;
2606                 break;
2607         default:
2608                 tp_len = ph.h1->tp_len;
2609                 break;
2610         }
2611         if (unlikely(tp_len > size_max)) {
2612                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2613                 return -EMSGSIZE;
2614         }
2615
2616         if (unlikely(po->tp_tx_has_off)) {
2617                 int off_min, off_max;
2618
2619                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2620                 off_max = po->tx_ring.frame_size - tp_len;
2621                 if (po->sk.sk_type == SOCK_DGRAM) {
2622                         switch (po->tp_version) {
2623                         case TPACKET_V3:
2624                                 off = ph.h3->tp_net;
2625                                 break;
2626                         case TPACKET_V2:
2627                                 off = ph.h2->tp_net;
2628                                 break;
2629                         default:
2630                                 off = ph.h1->tp_net;
2631                                 break;
2632                         }
2633                 } else {
2634                         switch (po->tp_version) {
2635                         case TPACKET_V3:
2636                                 off = ph.h3->tp_mac;
2637                                 break;
2638                         case TPACKET_V2:
2639                                 off = ph.h2->tp_mac;
2640                                 break;
2641                         default:
2642                                 off = ph.h1->tp_mac;
2643                                 break;
2644                         }
2645                 }
2646                 if (unlikely((off < off_min) || (off_max < off)))
2647                         return -EINVAL;
2648         } else {
2649                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2650         }
2651
2652         *data = frame + off;
2653         return tp_len;
2654 }
2655
2656 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2657 {
2658         struct sk_buff *skb = NULL;
2659         struct net_device *dev;
2660         struct virtio_net_hdr *vnet_hdr = NULL;
2661         struct sockcm_cookie sockc;
2662         __be16 proto;
2663         int err, reserve = 0;
2664         void *ph;
2665         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2666         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2667         unsigned char *addr = NULL;
2668         int tp_len, size_max;
2669         void *data;
2670         int len_sum = 0;
2671         int status = TP_STATUS_AVAILABLE;
2672         int hlen, tlen, copylen = 0;
2673         long timeo = 0;
2674
2675         mutex_lock(&po->pg_vec_lock);
2676
2677         /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2678          * we need to confirm it under protection of pg_vec_lock.
2679          */
2680         if (unlikely(!po->tx_ring.pg_vec)) {
2681                 err = -EBUSY;
2682                 goto out;
2683         }
2684         if (likely(saddr == NULL)) {
2685                 dev     = packet_cached_dev_get(po);
2686                 proto   = po->num;
2687         } else {
2688                 err = -EINVAL;
2689                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2690                         goto out;
2691                 if (msg->msg_namelen < (saddr->sll_halen
2692                                         + offsetof(struct sockaddr_ll,
2693                                                 sll_addr)))
2694                         goto out;
2695                 proto   = saddr->sll_protocol;
2696                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2697                 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2698                         if (dev && msg->msg_namelen < dev->addr_len +
2699                                    offsetof(struct sockaddr_ll, sll_addr))
2700                                 goto out_put;
2701                         addr = saddr->sll_addr;
2702                 }
2703         }
2704
2705         err = -ENXIO;
2706         if (unlikely(dev == NULL))
2707                 goto out;
2708         err = -ENETDOWN;
2709         if (unlikely(!(dev->flags & IFF_UP)))
2710                 goto out_put;
2711
2712         sockcm_init(&sockc, &po->sk);
2713         if (msg->msg_controllen) {
2714                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2715                 if (unlikely(err))
2716                         goto out_put;
2717         }
2718
2719         if (po->sk.sk_socket->type == SOCK_RAW)
2720                 reserve = dev->hard_header_len;
2721         size_max = po->tx_ring.frame_size
2722                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2723
2724         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2725                 size_max = dev->mtu + reserve + VLAN_HLEN;
2726
2727         reinit_completion(&po->skb_completion);
2728
2729         do {
2730                 ph = packet_current_frame(po, &po->tx_ring,
2731                                           TP_STATUS_SEND_REQUEST);
2732                 if (unlikely(ph == NULL)) {
2733                         if (need_wait && skb) {
2734                                 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2735                                 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2736                                 if (timeo <= 0) {
2737                                         err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2738                                         goto out_put;
2739                                 }
2740                         }
2741                         /* check for additional frames */
2742                         continue;
2743                 }
2744
2745                 skb = NULL;
2746                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2747                 if (tp_len < 0)
2748                         goto tpacket_error;
2749
2750                 status = TP_STATUS_SEND_REQUEST;
2751                 hlen = LL_RESERVED_SPACE(dev);
2752                 tlen = dev->needed_tailroom;
2753                 if (po->has_vnet_hdr) {
2754                         vnet_hdr = data;
2755                         data += sizeof(*vnet_hdr);
2756                         tp_len -= sizeof(*vnet_hdr);
2757                         if (tp_len < 0 ||
2758                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2759                                 tp_len = -EINVAL;
2760                                 goto tpacket_error;
2761                         }
2762                         copylen = __virtio16_to_cpu(vio_le(),
2763                                                     vnet_hdr->hdr_len);
2764                 }
2765                 copylen = max_t(int, copylen, dev->hard_header_len);
2766                 skb = sock_alloc_send_skb(&po->sk,
2767                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2768                                 (copylen - dev->hard_header_len),
2769                                 !need_wait, &err);
2770
2771                 if (unlikely(skb == NULL)) {
2772                         /* we assume the socket was initially writeable ... */
2773                         if (likely(len_sum > 0))
2774                                 err = len_sum;
2775                         goto out_status;
2776                 }
2777                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2778                                           addr, hlen, copylen, &sockc);
2779                 if (likely(tp_len >= 0) &&
2780                     tp_len > dev->mtu + reserve &&
2781                     !po->has_vnet_hdr &&
2782                     !packet_extra_vlan_len_allowed(dev, skb))
2783                         tp_len = -EMSGSIZE;
2784
2785                 if (unlikely(tp_len < 0)) {
2786 tpacket_error:
2787                         if (po->tp_loss) {
2788                                 __packet_set_status(po, ph,
2789                                                 TP_STATUS_AVAILABLE);
2790                                 packet_increment_head(&po->tx_ring);
2791                                 kfree_skb(skb);
2792                                 continue;
2793                         } else {
2794                                 status = TP_STATUS_WRONG_FORMAT;
2795                                 err = tp_len;
2796                                 goto out_status;
2797                         }
2798                 }
2799
2800                 if (po->has_vnet_hdr) {
2801                         if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2802                                 tp_len = -EINVAL;
2803                                 goto tpacket_error;
2804                         }
2805                         virtio_net_hdr_set_proto(skb, vnet_hdr);
2806                 }
2807
2808                 skb->destructor = tpacket_destruct_skb;
2809                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2810                 packet_inc_pending(&po->tx_ring);
2811
2812                 status = TP_STATUS_SEND_REQUEST;
2813                 err = po->xmit(skb);
2814                 if (unlikely(err > 0)) {
2815                         err = net_xmit_errno(err);
2816                         if (err && __packet_get_status(po, ph) ==
2817                                    TP_STATUS_AVAILABLE) {
2818                                 /* skb was destructed already */
2819                                 skb = NULL;
2820                                 goto out_status;
2821                         }
2822                         /*
2823                          * skb was dropped but not destructed yet;
2824                          * let's treat it like congestion or err < 0
2825                          */
2826                         err = 0;
2827                 }
2828                 packet_increment_head(&po->tx_ring);
2829                 len_sum += tp_len;
2830         } while (likely((ph != NULL) ||
2831                 /* Note: packet_read_pending() might be slow if we have
2832                  * to call it as it's per_cpu variable, but in fast-path
2833                  * we already short-circuit the loop with the first
2834                  * condition, and luckily don't have to go that path
2835                  * anyway.
2836                  */
2837                  (need_wait && packet_read_pending(&po->tx_ring))));
2838
2839         err = len_sum;
2840         goto out_put;
2841
2842 out_status:
2843         __packet_set_status(po, ph, status);
2844         kfree_skb(skb);
2845 out_put:
2846         dev_put(dev);
2847 out:
2848         mutex_unlock(&po->pg_vec_lock);
2849         return err;
2850 }
2851
2852 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2853                                         size_t reserve, size_t len,
2854                                         size_t linear, int noblock,
2855                                         int *err)
2856 {
2857         struct sk_buff *skb;
2858
2859         /* Under a page?  Don't bother with paged skb. */
2860         if (prepad + len < PAGE_SIZE || !linear)
2861                 linear = len;
2862
2863         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2864                                    err, 0);
2865         if (!skb)
2866                 return NULL;
2867
2868         skb_reserve(skb, reserve);
2869         skb_put(skb, linear);
2870         skb->data_len = len - linear;
2871         skb->len += len - linear;
2872
2873         return skb;
2874 }
2875
2876 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2877 {
2878         struct sock *sk = sock->sk;
2879         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2880         struct sk_buff *skb;
2881         struct net_device *dev;
2882         __be16 proto;
2883         unsigned char *addr = NULL;
2884         int err, reserve = 0;
2885         struct sockcm_cookie sockc;
2886         struct virtio_net_hdr vnet_hdr = { 0 };
2887         int offset = 0;
2888         struct packet_sock *po = pkt_sk(sk);
2889         bool has_vnet_hdr = false;
2890         int hlen, tlen, linear;
2891         int extra_len = 0;
2892
2893         /*
2894          *      Get and verify the address.
2895          */
2896
2897         if (likely(saddr == NULL)) {
2898                 dev     = packet_cached_dev_get(po);
2899                 proto   = po->num;
2900         } else {
2901                 err = -EINVAL;
2902                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2903                         goto out;
2904                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2905                         goto out;
2906                 proto   = saddr->sll_protocol;
2907                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2908                 if (sock->type == SOCK_DGRAM) {
2909                         if (dev && msg->msg_namelen < dev->addr_len +
2910                                    offsetof(struct sockaddr_ll, sll_addr))
2911                                 goto out_unlock;
2912                         addr = saddr->sll_addr;
2913                 }
2914         }
2915
2916         err = -ENXIO;
2917         if (unlikely(dev == NULL))
2918                 goto out_unlock;
2919         err = -ENETDOWN;
2920         if (unlikely(!(dev->flags & IFF_UP)))
2921                 goto out_unlock;
2922
2923         sockcm_init(&sockc, sk);
2924         sockc.mark = sk->sk_mark;
2925         if (msg->msg_controllen) {
2926                 err = sock_cmsg_send(sk, msg, &sockc);
2927                 if (unlikely(err))
2928                         goto out_unlock;
2929         }
2930
2931         if (sock->type == SOCK_RAW)
2932                 reserve = dev->hard_header_len;
2933         if (po->has_vnet_hdr) {
2934                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2935                 if (err)
2936                         goto out_unlock;
2937                 has_vnet_hdr = true;
2938         }
2939
2940         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2941                 if (!netif_supports_nofcs(dev)) {
2942                         err = -EPROTONOSUPPORT;
2943                         goto out_unlock;
2944                 }
2945                 extra_len = 4; /* We're doing our own CRC */
2946         }
2947
2948         err = -EMSGSIZE;
2949         if (!vnet_hdr.gso_type &&
2950             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2951                 goto out_unlock;
2952
2953         err = -ENOBUFS;
2954         hlen = LL_RESERVED_SPACE(dev);
2955         tlen = dev->needed_tailroom;
2956         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2957         linear = max(linear, min_t(int, len, dev->hard_header_len));
2958         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2959                                msg->msg_flags & MSG_DONTWAIT, &err);
2960         if (skb == NULL)
2961                 goto out_unlock;
2962
2963         skb_reset_network_header(skb);
2964
2965         err = -EINVAL;
2966         if (sock->type == SOCK_DGRAM) {
2967                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2968                 if (unlikely(offset < 0))
2969                         goto out_free;
2970         } else if (reserve) {
2971                 skb_reserve(skb, -reserve);
2972                 if (len < reserve + sizeof(struct ipv6hdr) &&
2973                     dev->min_header_len != dev->hard_header_len)
2974                         skb_reset_network_header(skb);
2975         }
2976
2977         /* Returns -EFAULT on error */
2978         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2979         if (err)
2980                 goto out_free;
2981
2982         if (sock->type == SOCK_RAW &&
2983             !dev_validate_header(dev, skb->data, len)) {
2984                 err = -EINVAL;
2985                 goto out_free;
2986         }
2987
2988         skb_setup_tx_timestamp(skb, sockc.tsflags);
2989
2990         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2991             !packet_extra_vlan_len_allowed(dev, skb)) {
2992                 err = -EMSGSIZE;
2993                 goto out_free;
2994         }
2995
2996         skb->protocol = proto;
2997         skb->dev = dev;
2998         skb->priority = sk->sk_priority;
2999         skb->mark = sockc.mark;
3000         skb->tstamp = sockc.transmit_time;
3001
3002         if (has_vnet_hdr) {
3003                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3004                 if (err)
3005                         goto out_free;
3006                 len += sizeof(vnet_hdr);
3007                 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3008         }
3009
3010         packet_parse_headers(skb, sock);
3011
3012         if (unlikely(extra_len == 4))
3013                 skb->no_fcs = 1;
3014
3015         err = po->xmit(skb);
3016         if (err > 0 && (err = net_xmit_errno(err)) != 0)
3017                 goto out_unlock;
3018
3019         dev_put(dev);
3020
3021         return len;
3022
3023 out_free:
3024         kfree_skb(skb);
3025 out_unlock:
3026         if (dev)
3027                 dev_put(dev);
3028 out:
3029         return err;
3030 }
3031
3032 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3033 {
3034         struct sock *sk = sock->sk;
3035         struct packet_sock *po = pkt_sk(sk);
3036
3037         if (po->tx_ring.pg_vec)
3038                 return tpacket_snd(po, msg);
3039         else
3040                 return packet_snd(sock, msg, len);
3041 }
3042
3043 /*
3044  *      Close a PACKET socket. This is fairly simple. We immediately go
3045  *      to 'closed' state and remove our protocol entry in the device list.
3046  */
3047
3048 static int packet_release(struct socket *sock)
3049 {
3050         struct sock *sk = sock->sk;
3051         struct packet_sock *po;
3052         struct packet_fanout *f;
3053         struct net *net;
3054         union tpacket_req_u req_u;
3055
3056         if (!sk)
3057                 return 0;
3058
3059         net = sock_net(sk);
3060         po = pkt_sk(sk);
3061
3062         mutex_lock(&net->packet.sklist_lock);
3063         sk_del_node_init_rcu(sk);
3064         mutex_unlock(&net->packet.sklist_lock);
3065
3066         preempt_disable();
3067         sock_prot_inuse_add(net, sk->sk_prot, -1);
3068         preempt_enable();
3069
3070         spin_lock(&po->bind_lock);
3071         unregister_prot_hook(sk, false);
3072         packet_cached_dev_reset(po);
3073
3074         if (po->prot_hook.dev) {
3075                 dev_put(po->prot_hook.dev);
3076                 po->prot_hook.dev = NULL;
3077         }
3078         spin_unlock(&po->bind_lock);
3079
3080         packet_flush_mclist(sk);
3081
3082         lock_sock(sk);
3083         if (po->rx_ring.pg_vec) {
3084                 memset(&req_u, 0, sizeof(req_u));
3085                 packet_set_ring(sk, &req_u, 1, 0);
3086         }
3087
3088         if (po->tx_ring.pg_vec) {
3089                 memset(&req_u, 0, sizeof(req_u));
3090                 packet_set_ring(sk, &req_u, 1, 1);
3091         }
3092         release_sock(sk);
3093
3094         f = fanout_release(sk);
3095
3096         synchronize_net();
3097
3098         kfree(po->rollover);
3099         if (f) {
3100                 fanout_release_data(f);
3101                 kvfree(f);
3102         }
3103         /*
3104          *      Now the socket is dead. No more input will appear.
3105          */
3106         sock_orphan(sk);
3107         sock->sk = NULL;
3108
3109         /* Purge queues */
3110
3111         skb_queue_purge(&sk->sk_receive_queue);
3112         packet_free_pending(po);
3113         sk_refcnt_debug_release(sk);
3114
3115         sock_put(sk);
3116         return 0;
3117 }
3118
3119 /*
3120  *      Attach a packet hook.
3121  */
3122
3123 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3124                           __be16 proto)
3125 {
3126         struct packet_sock *po = pkt_sk(sk);
3127         struct net_device *dev_curr;
3128         __be16 proto_curr;
3129         bool need_rehook;
3130         struct net_device *dev = NULL;
3131         int ret = 0;
3132         bool unlisted = false;
3133
3134         lock_sock(sk);
3135         spin_lock(&po->bind_lock);
3136         rcu_read_lock();
3137
3138         if (po->fanout) {
3139                 ret = -EINVAL;
3140                 goto out_unlock;
3141         }
3142
3143         if (name) {
3144                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3145                 if (!dev) {
3146                         ret = -ENODEV;
3147                         goto out_unlock;
3148                 }
3149         } else if (ifindex) {
3150                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3151                 if (!dev) {
3152                         ret = -ENODEV;
3153                         goto out_unlock;
3154                 }
3155         }
3156
3157         if (dev)
3158                 dev_hold(dev);
3159
3160         proto_curr = po->prot_hook.type;
3161         dev_curr = po->prot_hook.dev;
3162
3163         need_rehook = proto_curr != proto || dev_curr != dev;
3164
3165         if (need_rehook) {
3166                 if (po->running) {
3167                         rcu_read_unlock();
3168                         /* prevents packet_notifier() from calling
3169                          * register_prot_hook()
3170                          */
3171                         po->num = 0;
3172                         __unregister_prot_hook(sk, true);
3173                         rcu_read_lock();
3174                         dev_curr = po->prot_hook.dev;
3175                         if (dev)
3176                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3177                                                                  dev->ifindex);
3178                 }
3179
3180                 BUG_ON(po->running);
3181                 po->num = proto;
3182                 po->prot_hook.type = proto;
3183
3184                 if (unlikely(unlisted)) {
3185                         dev_put(dev);
3186                         po->prot_hook.dev = NULL;
3187                         po->ifindex = -1;
3188                         packet_cached_dev_reset(po);
3189                 } else {
3190                         po->prot_hook.dev = dev;
3191                         po->ifindex = dev ? dev->ifindex : 0;
3192                         packet_cached_dev_assign(po, dev);
3193                 }
3194         }
3195         if (dev_curr)
3196                 dev_put(dev_curr);
3197
3198         if (proto == 0 || !need_rehook)
3199                 goto out_unlock;
3200
3201         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3202                 register_prot_hook(sk);
3203         } else {
3204                 sk->sk_err = ENETDOWN;
3205                 if (!sock_flag(sk, SOCK_DEAD))
3206                         sk->sk_error_report(sk);
3207         }
3208
3209 out_unlock:
3210         rcu_read_unlock();
3211         spin_unlock(&po->bind_lock);
3212         release_sock(sk);
3213         return ret;
3214 }
3215
3216 /*
3217  *      Bind a packet socket to a device
3218  */
3219
3220 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3221                             int addr_len)
3222 {
3223         struct sock *sk = sock->sk;
3224         char name[sizeof(uaddr->sa_data) + 1];
3225
3226         /*
3227          *      Check legality
3228          */
3229
3230         if (addr_len != sizeof(struct sockaddr))
3231                 return -EINVAL;
3232         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3233          * zero-terminated.
3234          */
3235         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3236         name[sizeof(uaddr->sa_data)] = 0;
3237
3238         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3239 }
3240
3241 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3242 {
3243         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3244         struct sock *sk = sock->sk;
3245
3246         /*
3247          *      Check legality
3248          */
3249
3250         if (addr_len < sizeof(struct sockaddr_ll))
3251                 return -EINVAL;
3252         if (sll->sll_family != AF_PACKET)
3253                 return -EINVAL;
3254
3255         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3256                               sll->sll_protocol ? : pkt_sk(sk)->num);
3257 }
3258
3259 static struct proto packet_proto = {
3260         .name     = "PACKET",
3261         .owner    = THIS_MODULE,
3262         .obj_size = sizeof(struct packet_sock),
3263 };
3264
3265 /*
3266  *      Create a packet of type SOCK_PACKET.
3267  */
3268
3269 static int packet_create(struct net *net, struct socket *sock, int protocol,
3270                          int kern)
3271 {
3272         struct sock *sk;
3273         struct packet_sock *po;
3274         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3275         int err;
3276
3277         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3278                 return -EPERM;
3279         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3280             sock->type != SOCK_PACKET)
3281                 return -ESOCKTNOSUPPORT;
3282
3283         sock->state = SS_UNCONNECTED;
3284
3285         err = -ENOBUFS;
3286         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3287         if (sk == NULL)
3288                 goto out;
3289
3290         sock->ops = &packet_ops;
3291         if (sock->type == SOCK_PACKET)
3292                 sock->ops = &packet_ops_spkt;
3293
3294         sock_init_data(sock, sk);
3295
3296         po = pkt_sk(sk);
3297         init_completion(&po->skb_completion);
3298         sk->sk_family = PF_PACKET;
3299         po->num = proto;
3300         po->xmit = dev_queue_xmit;
3301
3302         err = packet_alloc_pending(po);
3303         if (err)
3304                 goto out2;
3305
3306         packet_cached_dev_reset(po);
3307
3308         sk->sk_destruct = packet_sock_destruct;
3309         sk_refcnt_debug_inc(sk);
3310
3311         /*
3312          *      Attach a protocol block
3313          */
3314
3315         spin_lock_init(&po->bind_lock);
3316         mutex_init(&po->pg_vec_lock);
3317         po->rollover = NULL;
3318         po->prot_hook.func = packet_rcv;
3319
3320         if (sock->type == SOCK_PACKET)
3321                 po->prot_hook.func = packet_rcv_spkt;
3322
3323         po->prot_hook.af_packet_priv = sk;
3324
3325         if (proto) {
3326                 po->prot_hook.type = proto;
3327                 __register_prot_hook(sk);
3328         }
3329
3330         mutex_lock(&net->packet.sklist_lock);
3331         sk_add_node_tail_rcu(sk, &net->packet.sklist);
3332         mutex_unlock(&net->packet.sklist_lock);
3333
3334         preempt_disable();
3335         sock_prot_inuse_add(net, &packet_proto, 1);
3336         preempt_enable();
3337
3338         return 0;
3339 out2:
3340         sk_free(sk);
3341 out:
3342         return err;
3343 }
3344
3345 /*
3346  *      Pull a packet from our receive queue and hand it to the user.
3347  *      If necessary we block.
3348  */
3349
3350 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3351                           int flags)
3352 {
3353         struct sock *sk = sock->sk;
3354         struct sk_buff *skb;
3355         int copied, err;
3356         int vnet_hdr_len = 0;
3357         unsigned int origlen = 0;
3358
3359         err = -EINVAL;
3360         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3361                 goto out;
3362
3363 #if 0
3364         /* What error should we return now? EUNATTACH? */
3365         if (pkt_sk(sk)->ifindex < 0)
3366                 return -ENODEV;
3367 #endif
3368
3369         if (flags & MSG_ERRQUEUE) {
3370                 err = sock_recv_errqueue(sk, msg, len,
3371                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3372                 goto out;
3373         }
3374
3375         /*
3376          *      Call the generic datagram receiver. This handles all sorts
3377          *      of horrible races and re-entrancy so we can forget about it
3378          *      in the protocol layers.
3379          *
3380          *      Now it will return ENETDOWN, if device have just gone down,
3381          *      but then it will block.
3382          */
3383
3384         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3385
3386         /*
3387          *      An error occurred so return it. Because skb_recv_datagram()
3388          *      handles the blocking we don't see and worry about blocking
3389          *      retries.
3390          */
3391
3392         if (skb == NULL)
3393                 goto out;
3394
3395         packet_rcv_try_clear_pressure(pkt_sk(sk));
3396
3397         if (pkt_sk(sk)->has_vnet_hdr) {
3398                 err = packet_rcv_vnet(msg, skb, &len);
3399                 if (err)
3400                         goto out_free;
3401                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3402         }
3403
3404         /* You lose any data beyond the buffer you gave. If it worries
3405          * a user program they can ask the device for its MTU
3406          * anyway.
3407          */
3408         copied = skb->len;
3409         if (copied > len) {
3410                 copied = len;
3411                 msg->msg_flags |= MSG_TRUNC;
3412         }
3413
3414         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3415         if (err)
3416                 goto out_free;
3417
3418         if (sock->type != SOCK_PACKET) {
3419                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3420
3421                 /* Original length was stored in sockaddr_ll fields */
3422                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3423                 sll->sll_family = AF_PACKET;
3424                 sll->sll_protocol = skb->protocol;
3425         }
3426
3427         sock_recv_ts_and_drops(msg, sk, skb);
3428
3429         if (msg->msg_name) {
3430                 int copy_len;
3431
3432                 /* If the address length field is there to be filled
3433                  * in, we fill it in now.
3434                  */
3435                 if (sock->type == SOCK_PACKET) {
3436                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3437                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3438                         copy_len = msg->msg_namelen;
3439                 } else {
3440                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3441
3442                         msg->msg_namelen = sll->sll_halen +
3443                                 offsetof(struct sockaddr_ll, sll_addr);
3444                         copy_len = msg->msg_namelen;
3445                         if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3446                                 memset(msg->msg_name +
3447                                        offsetof(struct sockaddr_ll, sll_addr),
3448                                        0, sizeof(sll->sll_addr));
3449                                 msg->msg_namelen = sizeof(struct sockaddr_ll);
3450                         }
3451                 }
3452                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3453         }
3454
3455         if (pkt_sk(sk)->auxdata) {
3456                 struct tpacket_auxdata aux;
3457
3458                 aux.tp_status = TP_STATUS_USER;
3459                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3460                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3461                 else if (skb->pkt_type != PACKET_OUTGOING &&
3462                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3463                           skb_csum_unnecessary(skb)))
3464                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3465
3466                 aux.tp_len = origlen;
3467                 aux.tp_snaplen = skb->len;
3468                 aux.tp_mac = 0;
3469                 aux.tp_net = skb_network_offset(skb);
3470                 if (skb_vlan_tag_present(skb)) {
3471                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3472                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3473                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3474                 } else {
3475                         aux.tp_vlan_tci = 0;
3476                         aux.tp_vlan_tpid = 0;
3477                 }
3478                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3479         }
3480
3481         /*
3482          *      Free or return the buffer as appropriate. Again this
3483          *      hides all the races and re-entrancy issues from us.
3484          */
3485         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3486
3487 out_free:
3488         skb_free_datagram(sk, skb);
3489 out:
3490         return err;
3491 }
3492
3493 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3494                                int peer)
3495 {
3496         struct net_device *dev;
3497         struct sock *sk = sock->sk;
3498
3499         if (peer)
3500                 return -EOPNOTSUPP;
3501
3502         uaddr->sa_family = AF_PACKET;
3503         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3504         rcu_read_lock();
3505         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3506         if (dev)
3507                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3508         rcu_read_unlock();
3509
3510         return sizeof(*uaddr);
3511 }
3512
3513 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3514                           int peer)
3515 {
3516         struct net_device *dev;
3517         struct sock *sk = sock->sk;
3518         struct packet_sock *po = pkt_sk(sk);
3519         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3520
3521         if (peer)
3522                 return -EOPNOTSUPP;
3523
3524         sll->sll_family = AF_PACKET;
3525         sll->sll_ifindex = po->ifindex;
3526         sll->sll_protocol = po->num;
3527         sll->sll_pkttype = 0;
3528         rcu_read_lock();
3529         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3530         if (dev) {
3531                 sll->sll_hatype = dev->type;
3532                 sll->sll_halen = dev->addr_len;
3533                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3534         } else {
3535                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3536                 sll->sll_halen = 0;
3537         }
3538         rcu_read_unlock();
3539
3540         return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3541 }
3542
3543 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3544                          int what)
3545 {
3546         switch (i->type) {
3547         case PACKET_MR_MULTICAST:
3548                 if (i->alen != dev->addr_len)
3549                         return -EINVAL;
3550                 if (what > 0)
3551                         return dev_mc_add(dev, i->addr);
3552                 else
3553                         return dev_mc_del(dev, i->addr);
3554                 break;
3555         case PACKET_MR_PROMISC:
3556                 return dev_set_promiscuity(dev, what);
3557         case PACKET_MR_ALLMULTI:
3558                 return dev_set_allmulti(dev, what);
3559         case PACKET_MR_UNICAST:
3560                 if (i->alen != dev->addr_len)
3561                         return -EINVAL;
3562                 if (what > 0)
3563                         return dev_uc_add(dev, i->addr);
3564                 else
3565                         return dev_uc_del(dev, i->addr);
3566                 break;
3567         default:
3568                 break;
3569         }
3570         return 0;
3571 }
3572
3573 static void packet_dev_mclist_delete(struct net_device *dev,
3574                                      struct packet_mclist **mlp)
3575 {
3576         struct packet_mclist *ml;
3577
3578         while ((ml = *mlp) != NULL) {
3579                 if (ml->ifindex == dev->ifindex) {
3580                         packet_dev_mc(dev, ml, -1);
3581                         *mlp = ml->next;
3582                         kfree(ml);
3583                 } else
3584                         mlp = &ml->next;
3585         }
3586 }
3587
3588 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3589 {
3590         struct packet_sock *po = pkt_sk(sk);
3591         struct packet_mclist *ml, *i;
3592         struct net_device *dev;
3593         int err;
3594
3595         rtnl_lock();
3596
3597         err = -ENODEV;
3598         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3599         if (!dev)
3600                 goto done;
3601
3602         err = -EINVAL;
3603         if (mreq->mr_alen > dev->addr_len)
3604                 goto done;
3605
3606         err = -ENOBUFS;
3607         i = kmalloc(sizeof(*i), GFP_KERNEL);
3608         if (i == NULL)
3609                 goto done;
3610
3611         err = 0;
3612         for (ml = po->mclist; ml; ml = ml->next) {
3613                 if (ml->ifindex == mreq->mr_ifindex &&
3614                     ml->type == mreq->mr_type &&
3615                     ml->alen == mreq->mr_alen &&
3616                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3617                         ml->count++;
3618                         /* Free the new element ... */
3619                         kfree(i);
3620                         goto done;
3621                 }
3622         }
3623
3624         i->type = mreq->mr_type;
3625         i->ifindex = mreq->mr_ifindex;
3626         i->alen = mreq->mr_alen;
3627         memcpy(i->addr, mreq->mr_address, i->alen);
3628         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3629         i->count = 1;
3630         i->next = po->mclist;
3631         po->mclist = i;
3632         err = packet_dev_mc(dev, i, 1);
3633         if (err) {
3634                 po->mclist = i->next;
3635                 kfree(i);
3636         }
3637
3638 done:
3639         rtnl_unlock();
3640         return err;
3641 }
3642
3643 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3644 {
3645         struct packet_mclist *ml, **mlp;
3646
3647         rtnl_lock();
3648
3649         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3650                 if (ml->ifindex == mreq->mr_ifindex &&
3651                     ml->type == mreq->mr_type &&
3652                     ml->alen == mreq->mr_alen &&
3653                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3654                         if (--ml->count == 0) {
3655                                 struct net_device *dev;
3656                                 *mlp = ml->next;
3657                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3658                                 if (dev)
3659                                         packet_dev_mc(dev, ml, -1);
3660                                 kfree(ml);
3661                         }
3662                         break;
3663                 }
3664         }
3665         rtnl_unlock();
3666         return 0;
3667 }
3668
3669 static void packet_flush_mclist(struct sock *sk)
3670 {
3671         struct packet_sock *po = pkt_sk(sk);
3672         struct packet_mclist *ml;
3673
3674         if (!po->mclist)
3675                 return;
3676
3677         rtnl_lock();
3678         while ((ml = po->mclist) != NULL) {
3679                 struct net_device *dev;
3680
3681                 po->mclist = ml->next;
3682                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3683                 if (dev != NULL)
3684                         packet_dev_mc(dev, ml, -1);
3685                 kfree(ml);
3686         }
3687         rtnl_unlock();
3688 }
3689
3690 static int
3691 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3692                   unsigned int optlen)
3693 {
3694         struct sock *sk = sock->sk;
3695         struct packet_sock *po = pkt_sk(sk);
3696         int ret;
3697
3698         if (level != SOL_PACKET)
3699                 return -ENOPROTOOPT;
3700
3701         switch (optname) {
3702         case PACKET_ADD_MEMBERSHIP:
3703         case PACKET_DROP_MEMBERSHIP:
3704         {
3705                 struct packet_mreq_max mreq;
3706                 int len = optlen;
3707                 memset(&mreq, 0, sizeof(mreq));
3708                 if (len < sizeof(struct packet_mreq))
3709                         return -EINVAL;
3710                 if (len > sizeof(mreq))
3711                         len = sizeof(mreq);
3712                 if (copy_from_sockptr(&mreq, optval, len))
3713                         return -EFAULT;
3714                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3715                         return -EINVAL;
3716                 if (optname == PACKET_ADD_MEMBERSHIP)
3717                         ret = packet_mc_add(sk, &mreq);
3718                 else
3719                         ret = packet_mc_drop(sk, &mreq);
3720                 return ret;
3721         }
3722
3723         case PACKET_RX_RING:
3724         case PACKET_TX_RING:
3725         {
3726                 union tpacket_req_u req_u;
3727                 int len;
3728
3729                 lock_sock(sk);
3730                 switch (po->tp_version) {
3731                 case TPACKET_V1:
3732                 case TPACKET_V2:
3733                         len = sizeof(req_u.req);
3734                         break;
3735                 case TPACKET_V3:
3736                 default:
3737                         len = sizeof(req_u.req3);
3738                         break;
3739                 }
3740                 if (optlen < len) {
3741                         ret = -EINVAL;
3742                 } else {
3743                         if (copy_from_sockptr(&req_u.req, optval, len))
3744                                 ret = -EFAULT;
3745                         else
3746                                 ret = packet_set_ring(sk, &req_u, 0,
3747                                                     optname == PACKET_TX_RING);
3748                 }
3749                 release_sock(sk);
3750                 return ret;
3751         }
3752         case PACKET_COPY_THRESH:
3753         {
3754                 int val;
3755
3756                 if (optlen != sizeof(val))
3757                         return -EINVAL;
3758                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3759                         return -EFAULT;
3760
3761                 pkt_sk(sk)->copy_thresh = val;
3762                 return 0;
3763         }
3764         case PACKET_VERSION:
3765         {
3766                 int val;
3767
3768                 if (optlen != sizeof(val))
3769                         return -EINVAL;
3770                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3771                         return -EFAULT;
3772                 switch (val) {
3773                 case TPACKET_V1:
3774                 case TPACKET_V2:
3775                 case TPACKET_V3:
3776                         break;
3777                 default:
3778                         return -EINVAL;
3779                 }
3780                 lock_sock(sk);
3781                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3782                         ret = -EBUSY;
3783                 } else {
3784                         po->tp_version = val;
3785                         ret = 0;
3786                 }
3787                 release_sock(sk);
3788                 return ret;
3789         }
3790         case PACKET_RESERVE:
3791         {
3792                 unsigned int val;
3793
3794                 if (optlen != sizeof(val))
3795                         return -EINVAL;
3796                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3797                         return -EFAULT;
3798                 if (val > INT_MAX)
3799                         return -EINVAL;
3800                 lock_sock(sk);
3801                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3802                         ret = -EBUSY;
3803                 } else {
3804                         po->tp_reserve = val;
3805                         ret = 0;
3806                 }
3807                 release_sock(sk);
3808                 return ret;
3809         }
3810         case PACKET_LOSS:
3811         {
3812                 unsigned int val;
3813
3814                 if (optlen != sizeof(val))
3815                         return -EINVAL;
3816                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3817                         return -EFAULT;
3818
3819                 lock_sock(sk);
3820                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3821                         ret = -EBUSY;
3822                 } else {
3823                         po->tp_loss = !!val;
3824                         ret = 0;
3825                 }
3826                 release_sock(sk);
3827                 return ret;
3828         }
3829         case PACKET_AUXDATA:
3830         {
3831                 int val;
3832
3833                 if (optlen < sizeof(val))
3834                         return -EINVAL;
3835                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3836                         return -EFAULT;
3837
3838                 lock_sock(sk);
3839                 po->auxdata = !!val;
3840                 release_sock(sk);
3841                 return 0;
3842         }
3843         case PACKET_ORIGDEV:
3844         {
3845                 int val;
3846
3847                 if (optlen < sizeof(val))
3848                         return -EINVAL;
3849                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3850                         return -EFAULT;
3851
3852                 lock_sock(sk);
3853                 po->origdev = !!val;
3854                 release_sock(sk);
3855                 return 0;
3856         }
3857         case PACKET_VNET_HDR:
3858         {
3859                 int val;
3860
3861                 if (sock->type != SOCK_RAW)
3862                         return -EINVAL;
3863                 if (optlen < sizeof(val))
3864                         return -EINVAL;
3865                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3866                         return -EFAULT;
3867
3868                 lock_sock(sk);
3869                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3870                         ret = -EBUSY;
3871                 } else {
3872                         po->has_vnet_hdr = !!val;
3873                         ret = 0;
3874                 }
3875                 release_sock(sk);
3876                 return ret;
3877         }
3878         case PACKET_TIMESTAMP:
3879         {
3880                 int val;
3881
3882                 if (optlen != sizeof(val))
3883                         return -EINVAL;
3884                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3885                         return -EFAULT;
3886
3887                 po->tp_tstamp = val;
3888                 return 0;
3889         }
3890         case PACKET_FANOUT:
3891         {
3892                 struct fanout_args args = { 0 };
3893
3894                 if (optlen != sizeof(int) && optlen != sizeof(args))
3895                         return -EINVAL;
3896                 if (copy_from_sockptr(&args, optval, optlen))
3897                         return -EFAULT;
3898
3899                 return fanout_add(sk, &args);
3900         }
3901         case PACKET_FANOUT_DATA:
3902         {
3903                 if (!po->fanout)
3904                         return -EINVAL;
3905
3906                 return fanout_set_data(po, optval, optlen);
3907         }
3908         case PACKET_IGNORE_OUTGOING:
3909         {
3910                 int val;
3911
3912                 if (optlen != sizeof(val))
3913                         return -EINVAL;
3914                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3915                         return -EFAULT;
3916                 if (val < 0 || val > 1)
3917                         return -EINVAL;
3918
3919                 po->prot_hook.ignore_outgoing = !!val;
3920                 return 0;
3921         }
3922         case PACKET_TX_HAS_OFF:
3923         {
3924                 unsigned int val;
3925
3926                 if (optlen != sizeof(val))
3927                         return -EINVAL;
3928                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3929                         return -EFAULT;
3930
3931                 lock_sock(sk);
3932                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3933                         ret = -EBUSY;
3934                 } else {
3935                         po->tp_tx_has_off = !!val;
3936                         ret = 0;
3937                 }
3938                 release_sock(sk);
3939                 return 0;
3940         }
3941         case PACKET_QDISC_BYPASS:
3942         {
3943                 int val;
3944
3945                 if (optlen != sizeof(val))
3946                         return -EINVAL;
3947                 if (copy_from_sockptr(&val, optval, sizeof(val)))
3948                         return -EFAULT;
3949
3950                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3951                 return 0;
3952         }
3953         default:
3954                 return -ENOPROTOOPT;
3955         }
3956 }
3957
3958 static int packet_getsockopt(struct socket *sock, int level, int optname,
3959                              char __user *optval, int __user *optlen)
3960 {
3961         int len;
3962         int val, lv = sizeof(val);
3963         struct sock *sk = sock->sk;
3964         struct packet_sock *po = pkt_sk(sk);
3965         void *data = &val;
3966         union tpacket_stats_u st;
3967         struct tpacket_rollover_stats rstats;
3968         int drops;
3969
3970         if (level != SOL_PACKET)
3971                 return -ENOPROTOOPT;
3972
3973         if (get_user(len, optlen))
3974                 return -EFAULT;
3975
3976         if (len < 0)
3977                 return -EINVAL;
3978
3979         switch (optname) {
3980         case PACKET_STATISTICS:
3981                 spin_lock_bh(&sk->sk_receive_queue.lock);
3982                 memcpy(&st, &po->stats, sizeof(st));
3983                 memset(&po->stats, 0, sizeof(po->stats));
3984                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3985                 drops = atomic_xchg(&po->tp_drops, 0);
3986
3987                 if (po->tp_version == TPACKET_V3) {
3988                         lv = sizeof(struct tpacket_stats_v3);
3989                         st.stats3.tp_drops = drops;
3990                         st.stats3.tp_packets += drops;
3991                         data = &st.stats3;
3992                 } else {
3993                         lv = sizeof(struct tpacket_stats);
3994                         st.stats1.tp_drops = drops;
3995                         st.stats1.tp_packets += drops;
3996                         data = &st.stats1;
3997                 }
3998
3999                 break;
4000         case PACKET_AUXDATA:
4001                 val = po->auxdata;
4002                 break;
4003         case PACKET_ORIGDEV:
4004                 val = po->origdev;
4005                 break;
4006         case PACKET_VNET_HDR:
4007                 val = po->has_vnet_hdr;
4008                 break;
4009         case PACKET_VERSION:
4010                 val = po->tp_version;
4011                 break;
4012         case PACKET_HDRLEN:
4013                 if (len > sizeof(int))
4014                         len = sizeof(int);
4015                 if (len < sizeof(int))
4016                         return -EINVAL;
4017                 if (copy_from_user(&val, optval, len))
4018                         return -EFAULT;
4019                 switch (val) {
4020                 case TPACKET_V1:
4021                         val = sizeof(struct tpacket_hdr);
4022                         break;
4023                 case TPACKET_V2:
4024                         val = sizeof(struct tpacket2_hdr);
4025                         break;
4026                 case TPACKET_V3:
4027                         val = sizeof(struct tpacket3_hdr);
4028                         break;
4029                 default:
4030                         return -EINVAL;
4031                 }
4032                 break;
4033         case PACKET_RESERVE:
4034                 val = po->tp_reserve;
4035                 break;
4036         case PACKET_LOSS:
4037                 val = po->tp_loss;
4038                 break;
4039         case PACKET_TIMESTAMP:
4040                 val = po->tp_tstamp;
4041                 break;
4042         case PACKET_FANOUT:
4043                 val = (po->fanout ?
4044                        ((u32)po->fanout->id |
4045                         ((u32)po->fanout->type << 16) |
4046                         ((u32)po->fanout->flags << 24)) :
4047                        0);
4048                 break;
4049         case PACKET_IGNORE_OUTGOING:
4050                 val = po->prot_hook.ignore_outgoing;
4051                 break;
4052         case PACKET_ROLLOVER_STATS:
4053                 if (!po->rollover)
4054                         return -EINVAL;
4055                 rstats.tp_all = atomic_long_read(&po->rollover->num);
4056                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4057                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4058                 data = &rstats;
4059                 lv = sizeof(rstats);
4060                 break;
4061         case PACKET_TX_HAS_OFF:
4062                 val = po->tp_tx_has_off;
4063                 break;
4064         case PACKET_QDISC_BYPASS:
4065                 val = packet_use_direct_xmit(po);
4066                 break;
4067         default:
4068                 return -ENOPROTOOPT;
4069         }
4070
4071         if (len > lv)
4072                 len = lv;
4073         if (put_user(len, optlen))
4074                 return -EFAULT;
4075         if (copy_to_user(optval, data, len))
4076                 return -EFAULT;
4077         return 0;
4078 }
4079
4080 static int packet_notifier(struct notifier_block *this,
4081                            unsigned long msg, void *ptr)
4082 {
4083         struct sock *sk;
4084         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4085         struct net *net = dev_net(dev);
4086
4087         rcu_read_lock();
4088         sk_for_each_rcu(sk, &net->packet.sklist) {
4089                 struct packet_sock *po = pkt_sk(sk);
4090
4091                 switch (msg) {
4092                 case NETDEV_UNREGISTER:
4093                         if (po->mclist)
4094                                 packet_dev_mclist_delete(dev, &po->mclist);
4095                         fallthrough;
4096
4097                 case NETDEV_DOWN:
4098                         if (dev->ifindex == po->ifindex) {
4099                                 spin_lock(&po->bind_lock);
4100                                 if (po->running) {
4101                                         __unregister_prot_hook(sk, false);
4102                                         sk->sk_err = ENETDOWN;
4103                                         if (!sock_flag(sk, SOCK_DEAD))
4104                                                 sk->sk_error_report(sk);
4105                                 }
4106                                 if (msg == NETDEV_UNREGISTER) {
4107                                         packet_cached_dev_reset(po);
4108                                         po->ifindex = -1;
4109                                         if (po->prot_hook.dev)
4110                                                 dev_put(po->prot_hook.dev);
4111                                         po->prot_hook.dev = NULL;
4112                                 }
4113                                 spin_unlock(&po->bind_lock);
4114                         }
4115                         break;
4116                 case NETDEV_UP:
4117                         if (dev->ifindex == po->ifindex) {
4118                                 spin_lock(&po->bind_lock);
4119                                 if (po->num)
4120                                         register_prot_hook(sk);
4121                                 spin_unlock(&po->bind_lock);
4122                         }
4123                         break;
4124                 }
4125         }
4126         rcu_read_unlock();
4127         return NOTIFY_DONE;
4128 }
4129
4130
4131 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4132                         unsigned long arg)
4133 {
4134         struct sock *sk = sock->sk;
4135
4136         switch (cmd) {
4137         case SIOCOUTQ:
4138         {
4139                 int amount = sk_wmem_alloc_get(sk);
4140
4141                 return put_user(amount, (int __user *)arg);
4142         }
4143         case SIOCINQ:
4144         {
4145                 struct sk_buff *skb;
4146                 int amount = 0;
4147
4148                 spin_lock_bh(&sk->sk_receive_queue.lock);
4149                 skb = skb_peek(&sk->sk_receive_queue);
4150                 if (skb)
4151                         amount = skb->len;
4152                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4153                 return put_user(amount, (int __user *)arg);
4154         }
4155 #ifdef CONFIG_INET
4156         case SIOCADDRT:
4157         case SIOCDELRT:
4158         case SIOCDARP:
4159         case SIOCGARP:
4160         case SIOCSARP:
4161         case SIOCGIFADDR:
4162         case SIOCSIFADDR:
4163         case SIOCGIFBRDADDR:
4164         case SIOCSIFBRDADDR:
4165         case SIOCGIFNETMASK:
4166         case SIOCSIFNETMASK:
4167         case SIOCGIFDSTADDR:
4168         case SIOCSIFDSTADDR:
4169         case SIOCSIFFLAGS:
4170                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4171 #endif
4172
4173         default:
4174                 return -ENOIOCTLCMD;
4175         }
4176         return 0;
4177 }
4178
4179 static __poll_t packet_poll(struct file *file, struct socket *sock,
4180                                 poll_table *wait)
4181 {
4182         struct sock *sk = sock->sk;
4183         struct packet_sock *po = pkt_sk(sk);
4184         __poll_t mask = datagram_poll(file, sock, wait);
4185
4186         spin_lock_bh(&sk->sk_receive_queue.lock);
4187         if (po->rx_ring.pg_vec) {
4188                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4189                         TP_STATUS_KERNEL))
4190                         mask |= EPOLLIN | EPOLLRDNORM;
4191         }
4192         packet_rcv_try_clear_pressure(po);
4193         spin_unlock_bh(&sk->sk_receive_queue.lock);
4194         spin_lock_bh(&sk->sk_write_queue.lock);
4195         if (po->tx_ring.pg_vec) {
4196                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4197                         mask |= EPOLLOUT | EPOLLWRNORM;
4198         }
4199         spin_unlock_bh(&sk->sk_write_queue.lock);
4200         return mask;
4201 }
4202
4203
4204 /* Dirty? Well, I still did not learn better way to account
4205  * for user mmaps.
4206  */
4207
4208 static void packet_mm_open(struct vm_area_struct *vma)
4209 {
4210         struct file *file = vma->vm_file;
4211         struct socket *sock = file->private_data;
4212         struct sock *sk = sock->sk;
4213
4214         if (sk)
4215                 atomic_inc(&pkt_sk(sk)->mapped);
4216 }
4217
4218 static void packet_mm_close(struct vm_area_struct *vma)
4219 {
4220         struct file *file = vma->vm_file;
4221         struct socket *sock = file->private_data;
4222         struct sock *sk = sock->sk;
4223
4224         if (sk)
4225                 atomic_dec(&pkt_sk(sk)->mapped);
4226 }
4227
4228 static const struct vm_operations_struct packet_mmap_ops = {
4229         .open   =       packet_mm_open,
4230         .close  =       packet_mm_close,
4231 };
4232
4233 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4234                         unsigned int len)
4235 {
4236         int i;
4237
4238         for (i = 0; i < len; i++) {
4239                 if (likely(pg_vec[i].buffer)) {
4240                         if (is_vmalloc_addr(pg_vec[i].buffer))
4241                                 vfree(pg_vec[i].buffer);
4242                         else
4243                                 free_pages((unsigned long)pg_vec[i].buffer,
4244                                            order);
4245                         pg_vec[i].buffer = NULL;
4246                 }
4247         }
4248         kfree(pg_vec);
4249 }
4250
4251 static char *alloc_one_pg_vec_page(unsigned long order)
4252 {
4253         char *buffer;
4254         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4255                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4256
4257         buffer = (char *) __get_free_pages(gfp_flags, order);
4258         if (buffer)
4259                 return buffer;
4260
4261         /* __get_free_pages failed, fall back to vmalloc */
4262         buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4263         if (buffer)
4264                 return buffer;
4265
4266         /* vmalloc failed, lets dig into swap here */
4267         gfp_flags &= ~__GFP_NORETRY;
4268         buffer = (char *) __get_free_pages(gfp_flags, order);
4269         if (buffer)
4270                 return buffer;
4271
4272         /* complete and utter failure */
4273         return NULL;
4274 }
4275
4276 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4277 {
4278         unsigned int block_nr = req->tp_block_nr;
4279         struct pgv *pg_vec;
4280         int i;
4281
4282         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4283         if (unlikely(!pg_vec))
4284                 goto out;
4285
4286         for (i = 0; i < block_nr; i++) {
4287                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4288                 if (unlikely(!pg_vec[i].buffer))
4289                         goto out_free_pgvec;
4290         }
4291
4292 out:
4293         return pg_vec;
4294
4295 out_free_pgvec:
4296         free_pg_vec(pg_vec, order, block_nr);
4297         pg_vec = NULL;
4298         goto out;
4299 }
4300
4301 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4302                 int closing, int tx_ring)
4303 {
4304         struct pgv *pg_vec = NULL;
4305         struct packet_sock *po = pkt_sk(sk);
4306         unsigned long *rx_owner_map = NULL;
4307         int was_running, order = 0;
4308         struct packet_ring_buffer *rb;
4309         struct sk_buff_head *rb_queue;
4310         __be16 num;
4311         int err;
4312         /* Added to avoid minimal code churn */
4313         struct tpacket_req *req = &req_u->req;
4314
4315         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4316         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4317
4318         err = -EBUSY;
4319         if (!closing) {
4320                 if (atomic_read(&po->mapped))
4321                         goto out;
4322                 if (packet_read_pending(rb))
4323                         goto out;
4324         }
4325
4326         if (req->tp_block_nr) {
4327                 unsigned int min_frame_size;
4328
4329                 /* Sanity tests and some calculations */
4330                 err = -EBUSY;
4331                 if (unlikely(rb->pg_vec))
4332                         goto out;
4333
4334                 switch (po->tp_version) {
4335                 case TPACKET_V1:
4336                         po->tp_hdrlen = TPACKET_HDRLEN;
4337                         break;
4338                 case TPACKET_V2:
4339                         po->tp_hdrlen = TPACKET2_HDRLEN;
4340                         break;
4341                 case TPACKET_V3:
4342                         po->tp_hdrlen = TPACKET3_HDRLEN;
4343                         break;
4344                 }
4345
4346                 err = -EINVAL;
4347                 if (unlikely((int)req->tp_block_size <= 0))
4348                         goto out;
4349                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4350                         goto out;
4351                 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4352                 if (po->tp_version >= TPACKET_V3 &&
4353                     req->tp_block_size <
4354                     BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4355                         goto out;
4356                 if (unlikely(req->tp_frame_size < min_frame_size))
4357                         goto out;
4358                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4359                         goto out;
4360
4361                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4362                 if (unlikely(rb->frames_per_block == 0))
4363                         goto out;
4364                 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4365                         goto out;
4366                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4367                                         req->tp_frame_nr))
4368                         goto out;
4369
4370                 err = -ENOMEM;
4371                 order = get_order(req->tp_block_size);
4372                 pg_vec = alloc_pg_vec(req, order);
4373                 if (unlikely(!pg_vec))
4374                         goto out;
4375                 switch (po->tp_version) {
4376                 case TPACKET_V3:
4377                         /* Block transmit is not supported yet */
4378                         if (!tx_ring) {
4379                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4380                         } else {
4381                                 struct tpacket_req3 *req3 = &req_u->req3;
4382
4383                                 if (req3->tp_retire_blk_tov ||
4384                                     req3->tp_sizeof_priv ||
4385                                     req3->tp_feature_req_word) {
4386                                         err = -EINVAL;
4387                                         goto out_free_pg_vec;
4388                                 }
4389                         }
4390                         break;
4391                 default:
4392                         if (!tx_ring) {
4393                                 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4394                                         GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4395                                 if (!rx_owner_map)
4396                                         goto out_free_pg_vec;
4397                         }
4398                         break;
4399                 }
4400         }
4401         /* Done */
4402         else {
4403                 err = -EINVAL;
4404                 if (unlikely(req->tp_frame_nr))
4405                         goto out;
4406         }
4407
4408
4409         /* Detach socket from network */
4410         spin_lock(&po->bind_lock);
4411         was_running = po->running;
4412         num = po->num;
4413         if (was_running) {
4414                 po->num = 0;
4415                 __unregister_prot_hook(sk, false);
4416         }
4417         spin_unlock(&po->bind_lock);
4418
4419         synchronize_net();
4420
4421         err = -EBUSY;
4422         mutex_lock(&po->pg_vec_lock);
4423         if (closing || atomic_read(&po->mapped) == 0) {
4424                 err = 0;
4425                 spin_lock_bh(&rb_queue->lock);
4426                 swap(rb->pg_vec, pg_vec);
4427                 if (po->tp_version <= TPACKET_V2)
4428                         swap(rb->rx_owner_map, rx_owner_map);
4429                 rb->frame_max = (req->tp_frame_nr - 1);
4430                 rb->head = 0;
4431                 rb->frame_size = req->tp_frame_size;
4432                 spin_unlock_bh(&rb_queue->lock);
4433
4434                 swap(rb->pg_vec_order, order);
4435                 swap(rb->pg_vec_len, req->tp_block_nr);
4436
4437                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4438                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4439                                                 tpacket_rcv : packet_rcv;
4440                 skb_queue_purge(rb_queue);
4441                 if (atomic_read(&po->mapped))
4442                         pr_err("packet_mmap: vma is busy: %d\n",
4443                                atomic_read(&po->mapped));
4444         }
4445         mutex_unlock(&po->pg_vec_lock);
4446
4447         spin_lock(&po->bind_lock);
4448         if (was_running) {
4449                 po->num = num;
4450                 register_prot_hook(sk);
4451         }
4452         spin_unlock(&po->bind_lock);
4453         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4454                 /* Because we don't support block-based V3 on tx-ring */
4455                 if (!tx_ring)
4456                         prb_shutdown_retire_blk_timer(po, rb_queue);
4457         }
4458
4459 out_free_pg_vec:
4460         bitmap_free(rx_owner_map);
4461         if (pg_vec)
4462                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4463 out:
4464         return err;
4465 }
4466
4467 static int packet_mmap(struct file *file, struct socket *sock,
4468                 struct vm_area_struct *vma)
4469 {
4470         struct sock *sk = sock->sk;
4471         struct packet_sock *po = pkt_sk(sk);
4472         unsigned long size, expected_size;
4473         struct packet_ring_buffer *rb;
4474         unsigned long start;
4475         int err = -EINVAL;
4476         int i;
4477
4478         if (vma->vm_pgoff)
4479                 return -EINVAL;
4480
4481         mutex_lock(&po->pg_vec_lock);
4482
4483         expected_size = 0;
4484         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4485                 if (rb->pg_vec) {
4486                         expected_size += rb->pg_vec_len
4487                                                 * rb->pg_vec_pages
4488                                                 * PAGE_SIZE;
4489                 }
4490         }
4491
4492         if (expected_size == 0)
4493                 goto out;
4494
4495         size = vma->vm_end - vma->vm_start;
4496         if (size != expected_size)
4497                 goto out;
4498
4499         start = vma->vm_start;
4500         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4501                 if (rb->pg_vec == NULL)
4502                         continue;
4503
4504                 for (i = 0; i < rb->pg_vec_len; i++) {
4505                         struct page *page;
4506                         void *kaddr = rb->pg_vec[i].buffer;
4507                         int pg_num;
4508
4509                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4510                                 page = pgv_to_page(kaddr);
4511                                 err = vm_insert_page(vma, start, page);
4512                                 if (unlikely(err))
4513                                         goto out;
4514                                 start += PAGE_SIZE;
4515                                 kaddr += PAGE_SIZE;
4516                         }
4517                 }
4518         }
4519
4520         atomic_inc(&po->mapped);
4521         vma->vm_ops = &packet_mmap_ops;
4522         err = 0;
4523
4524 out:
4525         mutex_unlock(&po->pg_vec_lock);
4526         return err;
4527 }
4528
4529 static const struct proto_ops packet_ops_spkt = {
4530         .family =       PF_PACKET,
4531         .owner =        THIS_MODULE,
4532         .release =      packet_release,
4533         .bind =         packet_bind_spkt,
4534         .connect =      sock_no_connect,
4535         .socketpair =   sock_no_socketpair,
4536         .accept =       sock_no_accept,
4537         .getname =      packet_getname_spkt,
4538         .poll =         datagram_poll,
4539         .ioctl =        packet_ioctl,
4540         .gettstamp =    sock_gettstamp,
4541         .listen =       sock_no_listen,
4542         .shutdown =     sock_no_shutdown,
4543         .sendmsg =      packet_sendmsg_spkt,
4544         .recvmsg =      packet_recvmsg,
4545         .mmap =         sock_no_mmap,
4546         .sendpage =     sock_no_sendpage,
4547 };
4548
4549 static const struct proto_ops packet_ops = {
4550         .family =       PF_PACKET,
4551         .owner =        THIS_MODULE,
4552         .release =      packet_release,
4553         .bind =         packet_bind,
4554         .connect =      sock_no_connect,
4555         .socketpair =   sock_no_socketpair,
4556         .accept =       sock_no_accept,
4557         .getname =      packet_getname,
4558         .poll =         packet_poll,
4559         .ioctl =        packet_ioctl,
4560         .gettstamp =    sock_gettstamp,
4561         .listen =       sock_no_listen,
4562         .shutdown =     sock_no_shutdown,
4563         .setsockopt =   packet_setsockopt,
4564         .getsockopt =   packet_getsockopt,
4565         .sendmsg =      packet_sendmsg,
4566         .recvmsg =      packet_recvmsg,
4567         .mmap =         packet_mmap,
4568         .sendpage =     sock_no_sendpage,
4569 };
4570
4571 static const struct net_proto_family packet_family_ops = {
4572         .family =       PF_PACKET,
4573         .create =       packet_create,
4574         .owner  =       THIS_MODULE,
4575 };
4576
4577 static struct notifier_block packet_netdev_notifier = {
4578         .notifier_call =        packet_notifier,
4579 };
4580
4581 #ifdef CONFIG_PROC_FS
4582
4583 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4584         __acquires(RCU)
4585 {
4586         struct net *net = seq_file_net(seq);
4587
4588         rcu_read_lock();
4589         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4590 }
4591
4592 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4593 {
4594         struct net *net = seq_file_net(seq);
4595         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4596 }
4597
4598 static void packet_seq_stop(struct seq_file *seq, void *v)
4599         __releases(RCU)
4600 {
4601         rcu_read_unlock();
4602 }
4603
4604 static int packet_seq_show(struct seq_file *seq, void *v)
4605 {
4606         if (v == SEQ_START_TOKEN)
4607                 seq_printf(seq,
4608                            "%*sRefCnt Type Proto  Iface R Rmem   User   Inode\n",
4609                            IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4610         else {
4611                 struct sock *s = sk_entry(v);
4612                 const struct packet_sock *po = pkt_sk(s);
4613
4614                 seq_printf(seq,
4615                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4616                            s,
4617                            refcount_read(&s->sk_refcnt),
4618                            s->sk_type,
4619                            ntohs(po->num),
4620                            po->ifindex,
4621                            po->running,
4622                            atomic_read(&s->sk_rmem_alloc),
4623                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4624                            sock_i_ino(s));
4625         }
4626
4627         return 0;
4628 }
4629
4630 static const struct seq_operations packet_seq_ops = {
4631         .start  = packet_seq_start,
4632         .next   = packet_seq_next,
4633         .stop   = packet_seq_stop,
4634         .show   = packet_seq_show,
4635 };
4636 #endif
4637
4638 static int __net_init packet_net_init(struct net *net)
4639 {
4640         mutex_init(&net->packet.sklist_lock);
4641         INIT_HLIST_HEAD(&net->packet.sklist);
4642
4643 #ifdef CONFIG_PROC_FS
4644         if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4645                         sizeof(struct seq_net_private)))
4646                 return -ENOMEM;
4647 #endif /* CONFIG_PROC_FS */
4648
4649         return 0;
4650 }
4651
4652 static void __net_exit packet_net_exit(struct net *net)
4653 {
4654         remove_proc_entry("packet", net->proc_net);
4655         WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4656 }
4657
4658 static struct pernet_operations packet_net_ops = {
4659         .init = packet_net_init,
4660         .exit = packet_net_exit,
4661 };
4662
4663
4664 static void __exit packet_exit(void)
4665 {
4666         unregister_netdevice_notifier(&packet_netdev_notifier);
4667         unregister_pernet_subsys(&packet_net_ops);
4668         sock_unregister(PF_PACKET);
4669         proto_unregister(&packet_proto);
4670 }
4671
4672 static int __init packet_init(void)
4673 {
4674         int rc;
4675
4676         rc = proto_register(&packet_proto, 0);
4677         if (rc)
4678                 goto out;
4679         rc = sock_register(&packet_family_ops);
4680         if (rc)
4681                 goto out_proto;
4682         rc = register_pernet_subsys(&packet_net_ops);
4683         if (rc)
4684                 goto out_sock;
4685         rc = register_netdevice_notifier(&packet_netdev_notifier);
4686         if (rc)
4687                 goto out_pernet;
4688
4689         return 0;
4690
4691 out_pernet:
4692         unregister_pernet_subsys(&packet_net_ops);
4693 out_sock:
4694         sock_unregister(PF_PACKET);
4695 out_proto:
4696         proto_unregister(&packet_proto);
4697 out:
4698         return rc;
4699 }
4700
4701 module_init(packet_init);
4702 module_exit(packet_exit);
4703 MODULE_LICENSE("GPL");
4704 MODULE_ALIAS_NETPROTO(PF_PACKET);