~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/packet/af_packet.c

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
  3  *              operating system.  INET is implemented using the  BSD Socket
  4  *              interface as the means of communication with the user level.
  5  *
  6  *              PACKET - implements raw packet sockets.
  7  *
  8  * Authors:     Ross Biro
  9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
 11  *
 12  * Fixes:
 13  *              Alan Cox        :       verify_area() now used correctly
 14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
 15  *              Alan Cox        :       tidied skbuff lists.
 16  *              Alan Cox        :       Now uses generic datagram routines I
 17  *                                      added. Also fixed the peek/read crash
 18  *                                      from all old Linux datagram code.
 19  *              Alan Cox        :       Uses the improved datagram code.
 20  *              Alan Cox        :       Added NULL's for socket options.
 21  *              Alan Cox        :       Re-commented the code.
 22  *              Alan Cox        :       Use new kernel side addressing
 23  *              Rob Janssen     :       Correct MTU usage.
 24  *              Dave Platt      :       Counter leaks caused by incorrect
 25  *                                      interrupt locking and some slightly
 26  *                                      dubious gcc output. Can you read
 27  *                                      compiler: it said _VOLATILE_
 28  *      Richard Kooijman        :       Timestamp fixes.
 29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
 30  *              Alan Cox        :       sendmsg/recvmsg support.
 31  *              Alan Cox        :       Protocol setting support
 32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
 33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
 34  *      Michal Ostrowski        :       Module initialization cleanup.
 35  *         Ulises Alonso        :       Frame number limit removal and
 36  *                                      packet_set_ring memory leak.
 37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
 38  *                                      The convention is that longer addresses
 39  *                                      will simply extend the hardware address
 40  *                                      byte arrays at the end of sockaddr_ll
 41  *                                      and packet_mreq.
 42  *              Johann Baudy    :       Added TX RING.
 43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
 44  *                                      layer.
 45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
 46  *
 47  *
 48  *              This program is free software; you can redistribute it and/or
 49  *              modify it under the terms of the GNU General Public License
 50  *              as published by the Free Software Foundation; either version
 51  *              2 of the License, or (at your option) any later version.
 52  *
 53  */
 54 
 55 #include <linux/types.h>
 56 #include <linux/mm.h>
 57 #include <linux/capability.h>
 58 #include <linux/fcntl.h>
 59 #include <linux/socket.h>
 60 #include <linux/in.h>
 61 #include <linux/inet.h>
 62 #include <linux/netdevice.h>
 63 #include <linux/if_packet.h>
 64 #include <linux/wireless.h>
 65 #include <linux/kernel.h>
 66 #include <linux/kmod.h>
 67 #include <linux/slab.h>
 68 #include <linux/vmalloc.h>
 69 #include <net/net_namespace.h>
 70 #include <net/ip.h>
 71 #include <net/protocol.h>
 72 #include <linux/skbuff.h>
 73 #include <net/sock.h>
 74 #include <linux/errno.h>
 75 #include <linux/timer.h>
 76 #include <asm/uaccess.h>
 77 #include <asm/ioctls.h>
 78 #include <asm/page.h>
 79 #include <asm/cacheflush.h>
 80 #include <asm/io.h>
 81 #include <linux/proc_fs.h>
 82 #include <linux/seq_file.h>
 83 #include <linux/poll.h>
 84 #include <linux/module.h>
 85 #include <linux/init.h>
 86 #include <linux/mutex.h>
 87 #include <linux/if_vlan.h>
 88 #include <linux/virtio_net.h>
 89 #include <linux/errqueue.h>
 90 #include <linux/net_tstamp.h>
 91 #include <linux/percpu.h>
 92 #ifdef CONFIG_INET
 93 #include <net/inet_common.h>
 94 #endif
 95 #include <linux/bpf.h>
 96 #include <net/compat.h>
 97 
 98 #include "internal.h"
 99 
100 /*
101    Assumptions:
102    - if device has no dev->hard_header routine, it adds and removes ll header
103      inside itself. In this case ll header is invisible outside of device,
104      but higher levels still should reserve dev->hard_header_len.
105      Some devices are enough clever to reallocate skb, when header
106      will not fit to reserved space (tunnel), another ones are silly
107      (PPP).
108    - packet socket receives packets with pulled ll header,
109      so that SOCK_RAW should push it back.
110 
111 On receive:
112 -----------
113 
114 Incoming, dev->hard_header!=NULL
115    mac_header -> ll header
116    data       -> data
117 
118 Outgoing, dev->hard_header!=NULL
119    mac_header -> ll header
120    data       -> ll header
121 
122 Incoming, dev->hard_header==NULL
123    mac_header -> UNKNOWN position. It is very likely, that it points to ll
124                  header.  PPP makes it, that is wrong, because introduce
125                  assymetry between rx and tx paths.
126    data       -> data
127 
128 Outgoing, dev->hard_header==NULL
129    mac_header -> data. ll header is still not built!
130    data       -> data
131 
132 Resume
133   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
134 
135 
136 On transmit:
137 ------------
138 
139 dev->hard_header != NULL
140    mac_header -> ll header
141    data       -> ll header
142 
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
144    mac_header -> data
145    data       -> data
146 
147    We should set nh.raw on output to correct posistion,
148    packet classifier depends on it.
149  */
150 
151 /* Private packet socket structures. */
152 
153 /* identical to struct packet_mreq except it has
154  * a longer address field.
155  */
156 struct packet_mreq_max {
157         int             mr_ifindex;
158         unsigned short  mr_type;
159         unsigned short  mr_alen;
160         unsigned char   mr_address[MAX_ADDR_LEN];
161 };
162 
163 union tpacket_uhdr {
164         struct tpacket_hdr  *h1;
165         struct tpacket2_hdr *h2;
166         struct tpacket3_hdr *h3;
167         void *raw;
168 };
169 
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171                 int closing, int tx_ring);
172 
173 #define V3_ALIGNMENT    (8)
174 
175 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 
180 #define PGV_FROM_VMALLOC 1
181 
182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
188 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 
190 struct packet_sock;
191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
193                        struct packet_type *pt, struct net_device *orig_dev);
194 
195 static void *packet_previous_frame(struct packet_sock *po,
196                 struct packet_ring_buffer *rb,
197                 int status);
198 static void packet_increment_head(struct packet_ring_buffer *buff);
199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
200                         struct tpacket_block_desc *);
201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
202                         struct packet_sock *);
203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
204                 struct packet_sock *, unsigned int status);
205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
206 static void prb_open_block(struct tpacket_kbdq_core *,
207                 struct tpacket_block_desc *);
208 static void prb_retire_rx_blk_timer_expired(unsigned long);
209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
210 static void prb_init_blk_timer(struct packet_sock *,
211                 struct tpacket_kbdq_core *,
212                 void (*func) (unsigned long));
213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
214 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
215                 struct tpacket3_hdr *);
216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
217                 struct tpacket3_hdr *);
218 static void packet_flush_mclist(struct sock *sk);
219 
220 struct packet_skb_cb {
221         union {
222                 struct sockaddr_pkt pkt;
223                 union {
224                         /* Trick: alias skb original length with
225                          * ll.sll_family and ll.protocol in order
226                          * to save room.
227                          */
228                         unsigned int origlen;
229                         struct sockaddr_ll ll;
230                 };
231         } sa;
232 };
233 
234 #define vio_le() virtio_legacy_is_little_endian()
235 
236 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
237 
238 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
239 #define GET_PBLOCK_DESC(x, bid) \
240         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
242         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
243 #define GET_NEXT_PRB_BLK_NUM(x) \
244         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
245         ((x)->kactive_blk_num+1) : 0)
246 
247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
248 static void __fanout_link(struct sock *sk, struct packet_sock *po);
249 
250 static int packet_direct_xmit(struct sk_buff *skb)
251 {
252         struct net_device *dev = skb->dev;
253         struct sk_buff *orig_skb = skb;
254         struct netdev_queue *txq;
255         int ret = NETDEV_TX_BUSY;
256 
257         if (unlikely(!netif_running(dev) ||
258                      !netif_carrier_ok(dev)))
259                 goto drop;
260 
261         skb = validate_xmit_skb_list(skb, dev);
262         if (skb != orig_skb)
263                 goto drop;
264 
265         txq = skb_get_tx_queue(dev, skb);
266 
267         local_bh_disable();
268 
269         HARD_TX_LOCK(dev, txq, smp_processor_id());
270         if (!netif_xmit_frozen_or_drv_stopped(txq))
271                 ret = netdev_start_xmit(skb, dev, txq, false);
272         HARD_TX_UNLOCK(dev, txq);
273 
274         local_bh_enable();
275 
276         if (!dev_xmit_complete(ret))
277                 kfree_skb(skb);
278 
279         return ret;
280 drop:
281         atomic_long_inc(&dev->tx_dropped);
282         kfree_skb_list(skb);
283         return NET_XMIT_DROP;
284 }
285 
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287 {
288         struct net_device *dev;
289 
290         rcu_read_lock();
291         dev = rcu_dereference(po->cached_dev);
292         if (likely(dev))
293                 dev_hold(dev);
294         rcu_read_unlock();
295 
296         return dev;
297 }
298 
299 static void packet_cached_dev_assign(struct packet_sock *po,
300                                      struct net_device *dev)
301 {
302         rcu_assign_pointer(po->cached_dev, dev);
303 }
304 
305 static void packet_cached_dev_reset(struct packet_sock *po)
306 {
307         RCU_INIT_POINTER(po->cached_dev, NULL);
308 }
309 
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
311 {
312         return po->xmit == packet_direct_xmit;
313 }
314 
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
316 {
317         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
318 }
319 
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
321 {
322         const struct net_device_ops *ops = dev->netdev_ops;
323         u16 queue_index;
324 
325         if (ops->ndo_select_queue) {
326                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327                                                     __packet_pick_tx_queue);
328                 queue_index = netdev_cap_txqueue(dev, queue_index);
329         } else {
330                 queue_index = __packet_pick_tx_queue(dev, skb);
331         }
332 
333         skb_set_queue_mapping(skb, queue_index);
334 }
335 
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337  * or from a context in which asynchronous accesses to the packet
338  * socket is not possible (packet_create()).
339  */
340 static void register_prot_hook(struct sock *sk)
341 {
342         struct packet_sock *po = pkt_sk(sk);
343 
344         if (!po->running) {
345                 if (po->fanout)
346                         __fanout_link(sk, po);
347                 else
348                         dev_add_pack(&po->prot_hook);
349 
350                 sock_hold(sk);
351                 po->running = 1;
352         }
353 }
354 
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356  * held.   If the sync parameter is true, we will temporarily drop
357  * the po->bind_lock and do a synchronize_net to make sure no
358  * asynchronous packet processing paths still refer to the elements
359  * of po->prot_hook.  If the sync parameter is false, it is the
360  * callers responsibility to take care of this.
361  */
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
363 {
364         struct packet_sock *po = pkt_sk(sk);
365 
366         po->running = 0;
367 
368         if (po->fanout)
369                 __fanout_unlink(sk, po);
370         else
371                 __dev_remove_pack(&po->prot_hook);
372 
373         __sock_put(sk);
374 
375         if (sync) {
376                 spin_unlock(&po->bind_lock);
377                 synchronize_net();
378                 spin_lock(&po->bind_lock);
379         }
380 }
381 
382 static void unregister_prot_hook(struct sock *sk, bool sync)
383 {
384         struct packet_sock *po = pkt_sk(sk);
385 
386         if (po->running)
387                 __unregister_prot_hook(sk, sync);
388 }
389 
390 static inline struct page * __pure pgv_to_page(void *addr)
391 {
392         if (is_vmalloc_addr(addr))
393                 return vmalloc_to_page(addr);
394         return virt_to_page(addr);
395 }
396 
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
398 {
399         union tpacket_uhdr h;
400 
401         h.raw = frame;
402         switch (po->tp_version) {
403         case TPACKET_V1:
404                 h.h1->tp_status = status;
405                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
406                 break;
407         case TPACKET_V2:
408                 h.h2->tp_status = status;
409                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
410                 break;
411         case TPACKET_V3:
412         default:
413                 WARN(1, "TPACKET version not supported.\n");
414                 BUG();
415         }
416 
417         smp_wmb();
418 }
419 
420 static int __packet_get_status(struct packet_sock *po, void *frame)
421 {
422         union tpacket_uhdr h;
423 
424         smp_rmb();
425 
426         h.raw = frame;
427         switch (po->tp_version) {
428         case TPACKET_V1:
429                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
430                 return h.h1->tp_status;
431         case TPACKET_V2:
432                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
433                 return h.h2->tp_status;
434         case TPACKET_V3:
435         default:
436                 WARN(1, "TPACKET version not supported.\n");
437                 BUG();
438                 return 0;
439         }
440 }
441 
442 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
443                                    unsigned int flags)
444 {
445         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
446 
447         if (shhwtstamps &&
448             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450                 return TP_STATUS_TS_RAW_HARDWARE;
451 
452         if (ktime_to_timespec_cond(skb->tstamp, ts))
453                 return TP_STATUS_TS_SOFTWARE;
454 
455         return 0;
456 }
457 
458 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
459                                     struct sk_buff *skb)
460 {
461         union tpacket_uhdr h;
462         struct timespec ts;
463         __u32 ts_status;
464 
465         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
466                 return 0;
467 
468         h.raw = frame;
469         switch (po->tp_version) {
470         case TPACKET_V1:
471                 h.h1->tp_sec = ts.tv_sec;
472                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
473                 break;
474         case TPACKET_V2:
475                 h.h2->tp_sec = ts.tv_sec;
476                 h.h2->tp_nsec = ts.tv_nsec;
477                 break;
478         case TPACKET_V3:
479         default:
480                 WARN(1, "TPACKET version not supported.\n");
481                 BUG();
482         }
483 
484         /* one flush is safe, as both fields always lie on the same cacheline */
485         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
486         smp_wmb();
487 
488         return ts_status;
489 }
490 
491 static void *packet_lookup_frame(struct packet_sock *po,
492                 struct packet_ring_buffer *rb,
493                 unsigned int position,
494                 int status)
495 {
496         unsigned int pg_vec_pos, frame_offset;
497         union tpacket_uhdr h;
498 
499         pg_vec_pos = position / rb->frames_per_block;
500         frame_offset = position % rb->frames_per_block;
501 
502         h.raw = rb->pg_vec[pg_vec_pos].buffer +
503                 (frame_offset * rb->frame_size);
504 
505         if (status != __packet_get_status(po, h.raw))
506                 return NULL;
507 
508         return h.raw;
509 }
510 
511 static void *packet_current_frame(struct packet_sock *po,
512                 struct packet_ring_buffer *rb,
513                 int status)
514 {
515         return packet_lookup_frame(po, rb, rb->head, status);
516 }
517 
518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
519 {
520         del_timer_sync(&pkc->retire_blk_timer);
521 }
522 
523 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
524                 struct sk_buff_head *rb_queue)
525 {
526         struct tpacket_kbdq_core *pkc;
527 
528         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
529 
530         spin_lock_bh(&rb_queue->lock);
531         pkc->delete_blk_timer = 1;
532         spin_unlock_bh(&rb_queue->lock);
533 
534         prb_del_retire_blk_timer(pkc);
535 }
536 
537 static void prb_init_blk_timer(struct packet_sock *po,
538                 struct tpacket_kbdq_core *pkc,
539                 void (*func) (unsigned long))
540 {
541         init_timer(&pkc->retire_blk_timer);
542         pkc->retire_blk_timer.data = (long)po;
543         pkc->retire_blk_timer.function = func;
544         pkc->retire_blk_timer.expires = jiffies;
545 }
546 
547 static void prb_setup_retire_blk_timer(struct packet_sock *po)
548 {
549         struct tpacket_kbdq_core *pkc;
550 
551         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552         prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
553 }
554 
555 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
556                                 int blk_size_in_bytes)
557 {
558         struct net_device *dev;
559         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
560         struct ethtool_link_ksettings ecmd;
561         int err;
562 
563         rtnl_lock();
564         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
565         if (unlikely(!dev)) {
566                 rtnl_unlock();
567                 return DEFAULT_PRB_RETIRE_TOV;
568         }
569         err = __ethtool_get_link_ksettings(dev, &ecmd);
570         rtnl_unlock();
571         if (!err) {
572                 /*
573                  * If the link speed is so slow you don't really
574                  * need to worry about perf anyways
575                  */
576                 if (ecmd.base.speed < SPEED_1000 ||
577                     ecmd.base.speed == SPEED_UNKNOWN) {
578                         return DEFAULT_PRB_RETIRE_TOV;
579                 } else {
580                         msec = 1;
581                         div = ecmd.base.speed / 1000;
582                 }
583         }
584 
585         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
586 
587         if (div)
588                 mbits /= div;
589 
590         tmo = mbits * msec;
591 
592         if (div)
593                 return tmo+1;
594         return tmo;
595 }
596 
597 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
598                         union tpacket_req_u *req_u)
599 {
600         p1->feature_req_word = req_u->req3.tp_feature_req_word;
601 }
602 
603 static void init_prb_bdqc(struct packet_sock *po,
604                         struct packet_ring_buffer *rb,
605                         struct pgv *pg_vec,
606                         union tpacket_req_u *req_u)
607 {
608         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
609         struct tpacket_block_desc *pbd;
610 
611         memset(p1, 0x0, sizeof(*p1));
612 
613         p1->knxt_seq_num = 1;
614         p1->pkbdq = pg_vec;
615         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
616         p1->pkblk_start = pg_vec[0].buffer;
617         p1->kblk_size = req_u->req3.tp_block_size;
618         p1->knum_blocks = req_u->req3.tp_block_nr;
619         p1->hdrlen = po->tp_hdrlen;
620         p1->version = po->tp_version;
621         p1->last_kactive_blk_num = 0;
622         po->stats.stats3.tp_freeze_q_cnt = 0;
623         if (req_u->req3.tp_retire_blk_tov)
624                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
625         else
626                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
627                                                 req_u->req3.tp_block_size);
628         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
629         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
630 
631         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
632         prb_init_ft_ops(p1, req_u);
633         prb_setup_retire_blk_timer(po);
634         prb_open_block(p1, pbd);
635 }
636 
637 /*  Do NOT update the last_blk_num first.
638  *  Assumes sk_buff_head lock is held.
639  */
640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
641 {
642         mod_timer(&pkc->retire_blk_timer,
643                         jiffies + pkc->tov_in_jiffies);
644         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
645 }
646 
647 /*
648  * Timer logic:
649  * 1) We refresh the timer only when we open a block.
650  *    By doing this we don't waste cycles refreshing the timer
651  *        on packet-by-packet basis.
652  *
653  * With a 1MB block-size, on a 1Gbps line, it will take
654  * i) ~8 ms to fill a block + ii) memcpy etc.
655  * In this cut we are not accounting for the memcpy time.
656  *
657  * So, if the user sets the 'tmo' to 10ms then the timer
658  * will never fire while the block is still getting filled
659  * (which is what we want). However, the user could choose
660  * to close a block early and that's fine.
661  *
662  * But when the timer does fire, we check whether or not to refresh it.
663  * Since the tmo granularity is in msecs, it is not too expensive
664  * to refresh the timer, lets say every '8' msecs.
665  * Either the user can set the 'tmo' or we can derive it based on
666  * a) line-speed and b) block-size.
667  * prb_calc_retire_blk_tmo() calculates the tmo.
668  *
669  */
670 static void prb_retire_rx_blk_timer_expired(unsigned long data)
671 {
672         struct packet_sock *po = (struct packet_sock *)data;
673         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
674         unsigned int frozen;
675         struct tpacket_block_desc *pbd;
676 
677         spin_lock(&po->sk.sk_receive_queue.lock);
678 
679         frozen = prb_queue_frozen(pkc);
680         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
681 
682         if (unlikely(pkc->delete_blk_timer))
683                 goto out;
684 
685         /* We only need to plug the race when the block is partially filled.
686          * tpacket_rcv:
687          *              lock(); increment BLOCK_NUM_PKTS; unlock()
688          *              copy_bits() is in progress ...
689          *              timer fires on other cpu:
690          *              we can't retire the current block because copy_bits
691          *              is in progress.
692          *
693          */
694         if (BLOCK_NUM_PKTS(pbd)) {
695                 while (atomic_read(&pkc->blk_fill_in_prog)) {
696                         /* Waiting for skb_copy_bits to finish... */
697                         cpu_relax();
698                 }
699         }
700 
701         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
702                 if (!frozen) {
703                         if (!BLOCK_NUM_PKTS(pbd)) {
704                                 /* An empty block. Just refresh the timer. */
705                                 goto refresh_timer;
706                         }
707                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
708                         if (!prb_dispatch_next_block(pkc, po))
709                                 goto refresh_timer;
710                         else
711                                 goto out;
712                 } else {
713                         /* Case 1. Queue was frozen because user-space was
714                          *         lagging behind.
715                          */
716                         if (prb_curr_blk_in_use(pkc, pbd)) {
717                                 /*
718                                  * Ok, user-space is still behind.
719                                  * So just refresh the timer.
720                                  */
721                                 goto refresh_timer;
722                         } else {
723                                /* Case 2. queue was frozen,user-space caught up,
724                                 * now the link went idle && the timer fired.
725                                 * We don't have a block to close.So we open this
726                                 * block and restart the timer.
727                                 * opening a block thaws the queue,restarts timer
728                                 * Thawing/timer-refresh is a side effect.
729                                 */
730                                 prb_open_block(pkc, pbd);
731                                 goto out;
732                         }
733                 }
734         }
735 
736 refresh_timer:
737         _prb_refresh_rx_retire_blk_timer(pkc);
738 
739 out:
740         spin_unlock(&po->sk.sk_receive_queue.lock);
741 }
742 
743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
744                 struct tpacket_block_desc *pbd1, __u32 status)
745 {
746         /* Flush everything minus the block header */
747 
748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
749         u8 *start, *end;
750 
751         start = (u8 *)pbd1;
752 
753         /* Skip the block header(we know header WILL fit in 4K) */
754         start += PAGE_SIZE;
755 
756         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
757         for (; start < end; start += PAGE_SIZE)
758                 flush_dcache_page(pgv_to_page(start));
759 
760         smp_wmb();
761 #endif
762 
763         /* Now update the block status. */
764 
765         BLOCK_STATUS(pbd1) = status;
766 
767         /* Flush the block header */
768 
769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
770         start = (u8 *)pbd1;
771         flush_dcache_page(pgv_to_page(start));
772 
773         smp_wmb();
774 #endif
775 }
776 
777 /*
778  * Side effect:
779  *
780  * 1) flush the block
781  * 2) Increment active_blk_num
782  *
783  * Note:We DONT refresh the timer on purpose.
784  *      Because almost always the next block will be opened.
785  */
786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
787                 struct tpacket_block_desc *pbd1,
788                 struct packet_sock *po, unsigned int stat)
789 {
790         __u32 status = TP_STATUS_USER | stat;
791 
792         struct tpacket3_hdr *last_pkt;
793         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
794         struct sock *sk = &po->sk;
795 
796         if (po->stats.stats3.tp_drops)
797                 status |= TP_STATUS_LOSING;
798 
799         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
800         last_pkt->tp_next_offset = 0;
801 
802         /* Get the ts of the last pkt */
803         if (BLOCK_NUM_PKTS(pbd1)) {
804                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
805                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
806         } else {
807                 /* Ok, we tmo'd - so get the current time.
808                  *
809                  * It shouldn't really happen as we don't close empty
810                  * blocks. See prb_retire_rx_blk_timer_expired().
811                  */
812                 struct timespec ts;
813                 getnstimeofday(&ts);
814                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
815                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
816         }
817 
818         smp_wmb();
819 
820         /* Flush the block */
821         prb_flush_block(pkc1, pbd1, status);
822 
823         sk->sk_data_ready(sk);
824 
825         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
826 }
827 
828 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
829 {
830         pkc->reset_pending_on_curr_blk = 0;
831 }
832 
833 /*
834  * Side effect of opening a block:
835  *
836  * 1) prb_queue is thawed.
837  * 2) retire_blk_timer is refreshed.
838  *
839  */
840 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
841         struct tpacket_block_desc *pbd1)
842 {
843         struct timespec ts;
844         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
845 
846         smp_rmb();
847 
848         /* We could have just memset this but we will lose the
849          * flexibility of making the priv area sticky
850          */
851 
852         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
853         BLOCK_NUM_PKTS(pbd1) = 0;
854         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
855 
856         getnstimeofday(&ts);
857 
858         h1->ts_first_pkt.ts_sec = ts.tv_sec;
859         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
860 
861         pkc1->pkblk_start = (char *)pbd1;
862         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
863 
864         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
866 
867         pbd1->version = pkc1->version;
868         pkc1->prev = pkc1->nxt_offset;
869         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
870 
871         prb_thaw_queue(pkc1);
872         _prb_refresh_rx_retire_blk_timer(pkc1);
873 
874         smp_wmb();
875 }
876 
877 /*
878  * Queue freeze logic:
879  * 1) Assume tp_block_nr = 8 blocks.
880  * 2) At time 't0', user opens Rx ring.
881  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882  * 4) user-space is either sleeping or processing block ''.
883  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884  *    it will close block-7,loop around and try to fill block ''.
885  *    call-flow:
886  *    __packet_lookup_frame_in_block
887  *      prb_retire_current_block()
888  *      prb_dispatch_next_block()
889  *        |->(BLOCK_STATUS == USER) evaluates to true
890  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
891  * 6) Now there are two cases:
892  *    6.1) Link goes idle right after the queue is frozen.
893  *         But remember, the last open_block() refreshed the timer.
894  *         When this timer expires,it will refresh itself so that we can
895  *         re-open block-0 in near future.
896  *    6.2) Link is busy and keeps on receiving packets. This is a simple
897  *         case and __packet_lookup_frame_in_block will check if block-0
898  *         is free and can now be re-used.
899  */
900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
901                                   struct packet_sock *po)
902 {
903         pkc->reset_pending_on_curr_blk = 1;
904         po->stats.stats3.tp_freeze_q_cnt++;
905 }
906 
907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
908 
909 /*
910  * If the next block is free then we will dispatch it
911  * and return a good offset.
912  * Else, we will freeze the queue.
913  * So, caller must check the return value.
914  */
915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
916                 struct packet_sock *po)
917 {
918         struct tpacket_block_desc *pbd;
919 
920         smp_rmb();
921 
922         /* 1. Get current block num */
923         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
924 
925         /* 2. If this block is currently in_use then freeze the queue */
926         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
927                 prb_freeze_queue(pkc, po);
928                 return NULL;
929         }
930 
931         /*
932          * 3.
933          * open this block and return the offset where the first packet
934          * needs to get stored.
935          */
936         prb_open_block(pkc, pbd);
937         return (void *)pkc->nxt_offset;
938 }
939 
940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
941                 struct packet_sock *po, unsigned int status)
942 {
943         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
944 
945         /* retire/close the current block */
946         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
947                 /*
948                  * Plug the case where copy_bits() is in progress on
949                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950                  * have space to copy the pkt in the current block and
951                  * called prb_retire_current_block()
952                  *
953                  * We don't need to worry about the TMO case because
954                  * the timer-handler already handled this case.
955                  */
956                 if (!(status & TP_STATUS_BLK_TMO)) {
957                         while (atomic_read(&pkc->blk_fill_in_prog)) {
958                                 /* Waiting for skb_copy_bits to finish... */
959                                 cpu_relax();
960                         }
961                 }
962                 prb_close_block(pkc, pbd, po, status);
963                 return;
964         }
965 }
966 
967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
968                                       struct tpacket_block_desc *pbd)
969 {
970         return TP_STATUS_USER & BLOCK_STATUS(pbd);
971 }
972 
973 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
974 {
975         return pkc->reset_pending_on_curr_blk;
976 }
977 
978 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
979 {
980         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
981         atomic_dec(&pkc->blk_fill_in_prog);
982 }
983 
984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
985                         struct tpacket3_hdr *ppd)
986 {
987         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
988 }
989 
990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
991                         struct tpacket3_hdr *ppd)
992 {
993         ppd->hv1.tp_rxhash = 0;
994 }
995 
996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
997                         struct tpacket3_hdr *ppd)
998 {
999         if (skb_vlan_tag_present(pkc->skb)) {
1000                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1003         } else {
1004                 ppd->hv1.tp_vlan_tci = 0;
1005                 ppd->hv1.tp_vlan_tpid = 0;
1006                 ppd->tp_status = TP_STATUS_AVAILABLE;
1007         }
1008 }
1009 
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011                         struct tpacket3_hdr *ppd)
1012 {
1013         ppd->hv1.tp_padding = 0;
1014         prb_fill_vlan_info(pkc, ppd);
1015 
1016         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017                 prb_fill_rxhash(pkc, ppd);
1018         else
1019                 prb_clear_rxhash(pkc, ppd);
1020 }
1021 
1022 static void prb_fill_curr_block(char *curr,
1023                                 struct tpacket_kbdq_core *pkc,
1024                                 struct tpacket_block_desc *pbd,
1025                                 unsigned int len)
1026 {
1027         struct tpacket3_hdr *ppd;
1028 
1029         ppd  = (struct tpacket3_hdr *)curr;
1030         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1031         pkc->prev = curr;
1032         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034         BLOCK_NUM_PKTS(pbd) += 1;
1035         atomic_inc(&pkc->blk_fill_in_prog);
1036         prb_run_all_ft_ops(pkc, ppd);
1037 }
1038 
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041                                             struct sk_buff *skb,
1042                                                 int status,
1043                                             unsigned int len
1044                                             )
1045 {
1046         struct tpacket_kbdq_core *pkc;
1047         struct tpacket_block_desc *pbd;
1048         char *curr, *end;
1049 
1050         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1052 
1053         /* Queue is frozen when user space is lagging behind */
1054         if (prb_queue_frozen(pkc)) {
1055                 /*
1056                  * Check if that last block which caused the queue to freeze,
1057                  * is still in_use by user-space.
1058                  */
1059                 if (prb_curr_blk_in_use(pkc, pbd)) {
1060                         /* Can't record this packet */
1061                         return NULL;
1062                 } else {
1063                         /*
1064                          * Ok, the block was released by user-space.
1065                          * Now let's open that block.
1066                          * opening a block also thaws the queue.
1067                          * Thawing is a side effect.
1068                          */
1069                         prb_open_block(pkc, pbd);
1070                 }
1071         }
1072 
1073         smp_mb();
1074         curr = pkc->nxt_offset;
1075         pkc->skb = skb;
1076         end = (char *)pbd + pkc->kblk_size;
1077 
1078         /* first try the current block */
1079         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080                 prb_fill_curr_block(curr, pkc, pbd, len);
1081                 return (void *)curr;
1082         }
1083 
1084         /* Ok, close the current block */
1085         prb_retire_current_block(pkc, po, 0);
1086 
1087         /* Now, try to dispatch the next block */
1088         curr = (char *)prb_dispatch_next_block(pkc, po);
1089         if (curr) {
1090                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091                 prb_fill_curr_block(curr, pkc, pbd, len);
1092                 return (void *)curr;
1093         }
1094 
1095         /*
1096          * No free blocks are available.user_space hasn't caught up yet.
1097          * Queue was just frozen and now this packet will get dropped.
1098          */
1099         return NULL;
1100 }
1101 
1102 static void *packet_current_rx_frame(struct packet_sock *po,
1103                                             struct sk_buff *skb,
1104                                             int status, unsigned int len)
1105 {
1106         char *curr = NULL;
1107         switch (po->tp_version) {
1108         case TPACKET_V1:
1109         case TPACKET_V2:
1110                 curr = packet_lookup_frame(po, &po->rx_ring,
1111                                         po->rx_ring.head, status);
1112                 return curr;
1113         case TPACKET_V3:
1114                 return __packet_lookup_frame_in_block(po, skb, status, len);
1115         default:
1116                 WARN(1, "TPACKET version not supported\n");
1117                 BUG();
1118                 return NULL;
1119         }
1120 }
1121 
1122 static void *prb_lookup_block(struct packet_sock *po,
1123                                      struct packet_ring_buffer *rb,
1124                                      unsigned int idx,
1125                                      int status)
1126 {
1127         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1128         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1129 
1130         if (status != BLOCK_STATUS(pbd))
1131                 return NULL;
1132         return pbd;
1133 }
1134 
1135 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1136 {
1137         unsigned int prev;
1138         if (rb->prb_bdqc.kactive_blk_num)
1139                 prev = rb->prb_bdqc.kactive_blk_num-1;
1140         else
1141                 prev = rb->prb_bdqc.knum_blocks-1;
1142         return prev;
1143 }
1144 
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock *po,
1147                                          struct packet_ring_buffer *rb,
1148                                          int status)
1149 {
1150         unsigned int previous = prb_previous_blk_num(rb);
1151         return prb_lookup_block(po, rb, previous, status);
1152 }
1153 
1154 static void *packet_previous_rx_frame(struct packet_sock *po,
1155                                              struct packet_ring_buffer *rb,
1156                                              int status)
1157 {
1158         if (po->tp_version <= TPACKET_V2)
1159                 return packet_previous_frame(po, rb, status);
1160 
1161         return __prb_previous_block(po, rb, status);
1162 }
1163 
1164 static void packet_increment_rx_head(struct packet_sock *po,
1165                                             struct packet_ring_buffer *rb)
1166 {
1167         switch (po->tp_version) {
1168         case TPACKET_V1:
1169         case TPACKET_V2:
1170                 return packet_increment_head(rb);
1171         case TPACKET_V3:
1172         default:
1173                 WARN(1, "TPACKET version not supported.\n");
1174                 BUG();
1175                 return;
1176         }
1177 }
1178 
1179 static void *packet_previous_frame(struct packet_sock *po,
1180                 struct packet_ring_buffer *rb,
1181                 int status)
1182 {
1183         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184         return packet_lookup_frame(po, rb, previous, status);
1185 }
1186 
1187 static void packet_increment_head(struct packet_ring_buffer *buff)
1188 {
1189         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1190 }
1191 
1192 static void packet_inc_pending(struct packet_ring_buffer *rb)
1193 {
1194         this_cpu_inc(*rb->pending_refcnt);
1195 }
1196 
1197 static void packet_dec_pending(struct packet_ring_buffer *rb)
1198 {
1199         this_cpu_dec(*rb->pending_refcnt);
1200 }
1201 
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1203 {
1204         unsigned int refcnt = 0;
1205         int cpu;
1206 
1207         /* We don't use pending refcount in rx_ring. */
1208         if (rb->pending_refcnt == NULL)
1209                 return 0;
1210 
1211         for_each_possible_cpu(cpu)
1212                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1213 
1214         return refcnt;
1215 }
1216 
1217 static int packet_alloc_pending(struct packet_sock *po)
1218 {
1219         po->rx_ring.pending_refcnt = NULL;
1220 
1221         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1223                 return -ENOBUFS;
1224 
1225         return 0;
1226 }
1227 
1228 static void packet_free_pending(struct packet_sock *po)
1229 {
1230         free_percpu(po->tx_ring.pending_refcnt);
1231 }
1232 
1233 #define ROOM_POW_OFF    2
1234 #define ROOM_NONE       0x0
1235 #define ROOM_LOW        0x1
1236 #define ROOM_NORMAL     0x2
1237 
1238 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1239 {
1240         int idx, len;
1241 
1242         len = po->rx_ring.frame_max + 1;
1243         idx = po->rx_ring.head;
1244         if (pow_off)
1245                 idx += len >> pow_off;
1246         if (idx >= len)
1247                 idx -= len;
1248         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1249 }
1250 
1251 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1252 {
1253         int idx, len;
1254 
1255         len = po->rx_ring.prb_bdqc.knum_blocks;
1256         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1257         if (pow_off)
1258                 idx += len >> pow_off;
1259         if (idx >= len)
1260                 idx -= len;
1261         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1262 }
1263 
1264 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1265 {
1266         struct sock *sk = &po->sk;
1267         int ret = ROOM_NONE;
1268 
1269         if (po->prot_hook.func != tpacket_rcv) {
1270                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271                                           - (skb ? skb->truesize : 0);
1272                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1273                         return ROOM_NORMAL;
1274                 else if (avail > 0)
1275                         return ROOM_LOW;
1276                 else
1277                         return ROOM_NONE;
1278         }
1279 
1280         if (po->tp_version == TPACKET_V3) {
1281                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1282                         ret = ROOM_NORMAL;
1283                 else if (__tpacket_v3_has_room(po, 0))
1284                         ret = ROOM_LOW;
1285         } else {
1286                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1287                         ret = ROOM_NORMAL;
1288                 else if (__tpacket_has_room(po, 0))
1289                         ret = ROOM_LOW;
1290         }
1291 
1292         return ret;
1293 }
1294 
1295 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1296 {
1297         int ret;
1298         bool has_room;
1299 
1300         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301         ret = __packet_rcv_has_room(po, skb);
1302         has_room = ret == ROOM_NORMAL;
1303         if (po->pressure == has_room)
1304                 po->pressure = !has_room;
1305         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1306 
1307         return ret;
1308 }
1309 
1310 static void packet_sock_destruct(struct sock *sk)
1311 {
1312         skb_queue_purge(&sk->sk_error_queue);
1313 
1314         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1316 
1317         if (!sock_flag(sk, SOCK_DEAD)) {
1318                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1319                 return;
1320         }
1321 
1322         sk_refcnt_debug_dec(sk);
1323 }
1324 
1325 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1326 {
1327         u32 rxhash;
1328         int i, count = 0;
1329 
1330         rxhash = skb_get_hash(skb);
1331         for (i = 0; i < ROLLOVER_HLEN; i++)
1332                 if (po->rollover->history[i] == rxhash)
1333                         count++;
1334 
1335         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336         return count > (ROLLOVER_HLEN >> 1);
1337 }
1338 
1339 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340                                       struct sk_buff *skb,
1341                                       unsigned int num)
1342 {
1343         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1344 }
1345 
1346 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347                                     struct sk_buff *skb,
1348                                     unsigned int num)
1349 {
1350         unsigned int val = atomic_inc_return(&f->rr_cur);
1351 
1352         return val % num;
1353 }
1354 
1355 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356                                      struct sk_buff *skb,
1357                                      unsigned int num)
1358 {
1359         return smp_processor_id() % num;
1360 }
1361 
1362 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363                                      struct sk_buff *skb,
1364                                      unsigned int num)
1365 {
1366         return prandom_u32_max(num);
1367 }
1368 
1369 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370                                           struct sk_buff *skb,
1371                                           unsigned int idx, bool try_self,
1372                                           unsigned int num)
1373 {
1374         struct packet_sock *po, *po_next, *po_skip = NULL;
1375         unsigned int i, j, room = ROOM_NONE;
1376 
1377         po = pkt_sk(f->arr[idx]);
1378 
1379         if (try_self) {
1380                 room = packet_rcv_has_room(po, skb);
1381                 if (room == ROOM_NORMAL ||
1382                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1383                         return idx;
1384                 po_skip = po;
1385         }
1386 
1387         i = j = min_t(int, po->rollover->sock, num - 1);
1388         do {
1389                 po_next = pkt_sk(f->arr[i]);
1390                 if (po_next != po_skip && !po_next->pressure &&
1391                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1392                         if (i != j)
1393                                 po->rollover->sock = i;
1394                         atomic_long_inc(&po->rollover->num);
1395                         if (room == ROOM_LOW)
1396                                 atomic_long_inc(&po->rollover->num_huge);
1397                         return i;
1398                 }
1399 
1400                 if (++i == num)
1401                         i = 0;
1402         } while (i != j);
1403 
1404         atomic_long_inc(&po->rollover->num_failed);
1405         return idx;
1406 }
1407 
1408 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409                                     struct sk_buff *skb,
1410                                     unsigned int num)
1411 {
1412         return skb_get_queue_mapping(skb) % num;
1413 }
1414 
1415 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416                                      struct sk_buff *skb,
1417                                      unsigned int num)
1418 {
1419         struct bpf_prog *prog;
1420         unsigned int ret = 0;
1421 
1422         rcu_read_lock();
1423         prog = rcu_dereference(f->bpf_prog);
1424         if (prog)
1425                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1426         rcu_read_unlock();
1427 
1428         return ret;
1429 }
1430 
1431 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1432 {
1433         return f->flags & (flag >> 8);
1434 }
1435 
1436 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437                              struct packet_type *pt, struct net_device *orig_dev)
1438 {
1439         struct packet_fanout *f = pt->af_packet_priv;
1440         unsigned int num = READ_ONCE(f->num_members);
1441         struct net *net = read_pnet(&f->net);
1442         struct packet_sock *po;
1443         unsigned int idx;
1444 
1445         if (!net_eq(dev_net(dev), net) || !num) {
1446                 kfree_skb(skb);
1447                 return 0;
1448         }
1449 
1450         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1452                 if (!skb)
1453                         return 0;
1454         }
1455         switch (f->type) {
1456         case PACKET_FANOUT_HASH:
1457         default:
1458                 idx = fanout_demux_hash(f, skb, num);
1459                 break;
1460         case PACKET_FANOUT_LB:
1461                 idx = fanout_demux_lb(f, skb, num);
1462                 break;
1463         case PACKET_FANOUT_CPU:
1464                 idx = fanout_demux_cpu(f, skb, num);
1465                 break;
1466         case PACKET_FANOUT_RND:
1467                 idx = fanout_demux_rnd(f, skb, num);
1468                 break;
1469         case PACKET_FANOUT_QM:
1470                 idx = fanout_demux_qm(f, skb, num);
1471                 break;
1472         case PACKET_FANOUT_ROLLOVER:
1473                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1474                 break;
1475         case PACKET_FANOUT_CBPF:
1476         case PACKET_FANOUT_EBPF:
1477                 idx = fanout_demux_bpf(f, skb, num);
1478                 break;
1479         }
1480 
1481         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1483 
1484         po = pkt_sk(f->arr[idx]);
1485         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1486 }
1487 
1488 DEFINE_MUTEX(fanout_mutex);
1489 EXPORT_SYMBOL_GPL(fanout_mutex);
1490 static LIST_HEAD(fanout_list);
1491 
1492 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1493 {
1494         struct packet_fanout *f = po->fanout;
1495 
1496         spin_lock(&f->lock);
1497         f->arr[f->num_members] = sk;
1498         smp_wmb();
1499         f->num_members++;
1500         spin_unlock(&f->lock);
1501 }
1502 
1503 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1504 {
1505         struct packet_fanout *f = po->fanout;
1506         int i;
1507 
1508         spin_lock(&f->lock);
1509         for (i = 0; i < f->num_members; i++) {
1510                 if (f->arr[i] == sk)
1511                         break;
1512         }
1513         BUG_ON(i >= f->num_members);
1514         f->arr[i] = f->arr[f->num_members - 1];
1515         f->num_members--;
1516         spin_unlock(&f->lock);
1517 }
1518 
1519 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1520 {
1521         if (sk->sk_family != PF_PACKET)
1522                 return false;
1523 
1524         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1525 }
1526 
1527 static void fanout_init_data(struct packet_fanout *f)
1528 {
1529         switch (f->type) {
1530         case PACKET_FANOUT_LB:
1531                 atomic_set(&f->rr_cur, 0);
1532                 break;
1533         case PACKET_FANOUT_CBPF:
1534         case PACKET_FANOUT_EBPF:
1535                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1536                 break;
1537         }
1538 }
1539 
1540 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1541 {
1542         struct bpf_prog *old;
1543 
1544         spin_lock(&f->lock);
1545         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1546         rcu_assign_pointer(f->bpf_prog, new);
1547         spin_unlock(&f->lock);
1548 
1549         if (old) {
1550                 synchronize_net();
1551                 bpf_prog_destroy(old);
1552         }
1553 }
1554 
1555 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1556                                 unsigned int len)
1557 {
1558         struct bpf_prog *new;
1559         struct sock_fprog fprog;
1560         int ret;
1561 
1562         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1563                 return -EPERM;
1564         if (len != sizeof(fprog))
1565                 return -EINVAL;
1566         if (copy_from_user(&fprog, data, len))
1567                 return -EFAULT;
1568 
1569         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1570         if (ret)
1571                 return ret;
1572 
1573         __fanout_set_data_bpf(po->fanout, new);
1574         return 0;
1575 }
1576 
1577 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1578                                 unsigned int len)
1579 {
1580         struct bpf_prog *new;
1581         u32 fd;
1582 
1583         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1584                 return -EPERM;
1585         if (len != sizeof(fd))
1586                 return -EINVAL;
1587         if (copy_from_user(&fd, data, len))
1588                 return -EFAULT;
1589 
1590         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1591         if (IS_ERR(new))
1592                 return PTR_ERR(new);
1593 
1594         __fanout_set_data_bpf(po->fanout, new);
1595         return 0;
1596 }
1597 
1598 static int fanout_set_data(struct packet_sock *po, char __user *data,
1599                            unsigned int len)
1600 {
1601         switch (po->fanout->type) {
1602         case PACKET_FANOUT_CBPF:
1603                 return fanout_set_data_cbpf(po, data, len);
1604         case PACKET_FANOUT_EBPF:
1605                 return fanout_set_data_ebpf(po, data, len);
1606         default:
1607                 return -EINVAL;
1608         };
1609 }
1610 
1611 static void fanout_release_data(struct packet_fanout *f)
1612 {
1613         switch (f->type) {
1614         case PACKET_FANOUT_CBPF:
1615         case PACKET_FANOUT_EBPF:
1616                 __fanout_set_data_bpf(f, NULL);
1617         };
1618 }
1619 
1620 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1621 {
1622         struct packet_sock *po = pkt_sk(sk);
1623         struct packet_fanout *f, *match;
1624         u8 type = type_flags & 0xff;
1625         u8 flags = type_flags >> 8;
1626         int err;
1627 
1628         switch (type) {
1629         case PACKET_FANOUT_ROLLOVER:
1630                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1631                         return -EINVAL;
1632         case PACKET_FANOUT_HASH:
1633         case PACKET_FANOUT_LB:
1634         case PACKET_FANOUT_CPU:
1635         case PACKET_FANOUT_RND:
1636         case PACKET_FANOUT_QM:
1637         case PACKET_FANOUT_CBPF:
1638         case PACKET_FANOUT_EBPF:
1639                 break;
1640         default:
1641                 return -EINVAL;
1642         }
1643 
1644         if (!po->running)
1645                 return -EINVAL;
1646 
1647         if (po->fanout)
1648                 return -EALREADY;
1649 
1650         if (type == PACKET_FANOUT_ROLLOVER ||
1651             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1652                 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1653                 if (!po->rollover)
1654                         return -ENOMEM;
1655                 atomic_long_set(&po->rollover->num, 0);
1656                 atomic_long_set(&po->rollover->num_huge, 0);
1657                 atomic_long_set(&po->rollover->num_failed, 0);
1658         }
1659 
1660         mutex_lock(&fanout_mutex);
1661         match = NULL;
1662         list_for_each_entry(f, &fanout_list, list) {
1663                 if (f->id == id &&
1664                     read_pnet(&f->net) == sock_net(sk)) {
1665                         match = f;
1666                         break;
1667                 }
1668         }
1669         err = -EINVAL;
1670         if (match && match->flags != flags)
1671                 goto out;
1672         if (!match) {
1673                 err = -ENOMEM;
1674                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1675                 if (!match)
1676                         goto out;
1677                 write_pnet(&match->net, sock_net(sk));
1678                 match->id = id;
1679                 match->type = type;
1680                 match->flags = flags;
1681                 INIT_LIST_HEAD(&match->list);
1682                 spin_lock_init(&match->lock);
1683                 atomic_set(&match->sk_ref, 0);
1684                 fanout_init_data(match);
1685                 match->prot_hook.type = po->prot_hook.type;
1686                 match->prot_hook.dev = po->prot_hook.dev;
1687                 match->prot_hook.func = packet_rcv_fanout;
1688                 match->prot_hook.af_packet_priv = match;
1689                 match->prot_hook.id_match = match_fanout_group;
1690                 dev_add_pack(&match->prot_hook);
1691                 list_add(&match->list, &fanout_list);
1692         }
1693         err = -EINVAL;
1694         if (match->type == type &&
1695             match->prot_hook.type == po->prot_hook.type &&
1696             match->prot_hook.dev == po->prot_hook.dev) {
1697                 err = -ENOSPC;
1698                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1699                         __dev_remove_pack(&po->prot_hook);
1700                         po->fanout = match;
1701                         atomic_inc(&match->sk_ref);
1702                         __fanout_link(sk, po);
1703                         err = 0;
1704                 }
1705         }
1706 out:
1707         mutex_unlock(&fanout_mutex);
1708         if (err) {
1709                 kfree(po->rollover);
1710                 po->rollover = NULL;
1711         }
1712         return err;
1713 }
1714 
1715 static void fanout_release(struct sock *sk)
1716 {
1717         struct packet_sock *po = pkt_sk(sk);
1718         struct packet_fanout *f;
1719 
1720         f = po->fanout;
1721         if (!f)
1722                 return;
1723 
1724         mutex_lock(&fanout_mutex);
1725         po->fanout = NULL;
1726 
1727         if (atomic_dec_and_test(&f->sk_ref)) {
1728                 list_del(&f->list);
1729                 dev_remove_pack(&f->prot_hook);
1730                 fanout_release_data(f);
1731                 kfree(f);
1732         }
1733         mutex_unlock(&fanout_mutex);
1734 
1735         if (po->rollover)
1736                 kfree_rcu(po->rollover, rcu);
1737 }
1738 
1739 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1740                                           struct sk_buff *skb)
1741 {
1742         /* Earlier code assumed this would be a VLAN pkt, double-check
1743          * this now that we have the actual packet in hand. We can only
1744          * do this check on Ethernet devices.
1745          */
1746         if (unlikely(dev->type != ARPHRD_ETHER))
1747                 return false;
1748 
1749         skb_reset_mac_header(skb);
1750         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1751 }
1752 
1753 static const struct proto_ops packet_ops;
1754 
1755 static const struct proto_ops packet_ops_spkt;
1756 
1757 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1758                            struct packet_type *pt, struct net_device *orig_dev)
1759 {
1760         struct sock *sk;
1761         struct sockaddr_pkt *spkt;
1762 
1763         /*
1764          *      When we registered the protocol we saved the socket in the data
1765          *      field for just this event.
1766          */
1767 
1768         sk = pt->af_packet_priv;
1769 
1770         /*
1771          *      Yank back the headers [hope the device set this
1772          *      right or kerboom...]
1773          *
1774          *      Incoming packets have ll header pulled,
1775          *      push it back.
1776          *
1777          *      For outgoing ones skb->data == skb_mac_header(skb)
1778          *      so that this procedure is noop.
1779          */
1780 
1781         if (skb->pkt_type == PACKET_LOOPBACK)
1782                 goto out;
1783 
1784         if (!net_eq(dev_net(dev), sock_net(sk)))
1785                 goto out;
1786 
1787         skb = skb_share_check(skb, GFP_ATOMIC);
1788         if (skb == NULL)
1789                 goto oom;
1790 
1791         /* drop any routing info */
1792         skb_dst_drop(skb);
1793 
1794         /* drop conntrack reference */
1795         nf_reset(skb);
1796 
1797         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1798 
1799         skb_push(skb, skb->data - skb_mac_header(skb));
1800 
1801         /*
1802          *      The SOCK_PACKET socket receives _all_ frames.
1803          */
1804 
1805         spkt->spkt_family = dev->type;
1806         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1807         spkt->spkt_protocol = skb->protocol;
1808 
1809         /*
1810          *      Charge the memory to the socket. This is done specifically
1811          *      to prevent sockets using all the memory up.
1812          */
1813 
1814         if (sock_queue_rcv_skb(sk, skb) == 0)
1815                 return 0;
1816 
1817 out:
1818         kfree_skb(skb);
1819 oom:
1820         return 0;
1821 }
1822 
1823 
1824 /*
1825  *      Output a raw packet to a device layer. This bypasses all the other
1826  *      protocol layers and you must therefore supply it with a complete frame
1827  */
1828 
1829 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1830                                size_t len)
1831 {
1832         struct sock *sk = sock->sk;
1833         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1834         struct sk_buff *skb = NULL;
1835         struct net_device *dev;
1836         struct sockcm_cookie sockc;
1837         __be16 proto = 0;
1838         int err;
1839         int extra_len = 0;
1840 
1841         /*
1842          *      Get and verify the address.
1843          */
1844 
1845         if (saddr) {
1846                 if (msg->msg_namelen < sizeof(struct sockaddr))
1847                         return -EINVAL;
1848                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1849                         proto = saddr->spkt_protocol;
1850         } else
1851                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1852 
1853         /*
1854          *      Find the device first to size check it
1855          */
1856 
1857         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1858 retry:
1859         rcu_read_lock();
1860         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1861         err = -ENODEV;
1862         if (dev == NULL)
1863                 goto out_unlock;
1864 
1865         err = -ENETDOWN;
1866         if (!(dev->flags & IFF_UP))
1867                 goto out_unlock;
1868 
1869         /*
1870          * You may not queue a frame bigger than the mtu. This is the lowest level
1871          * raw protocol and you must do your own fragmentation at this level.
1872          */
1873 
1874         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1875                 if (!netif_supports_nofcs(dev)) {
1876                         err = -EPROTONOSUPPORT;
1877                         goto out_unlock;
1878                 }
1879                 extra_len = 4; /* We're doing our own CRC */
1880         }
1881 
1882         err = -EMSGSIZE;
1883         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1884                 goto out_unlock;
1885 
1886         if (!skb) {
1887                 size_t reserved = LL_RESERVED_SPACE(dev);
1888                 int tlen = dev->needed_tailroom;
1889                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1890 
1891                 rcu_read_unlock();
1892                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1893                 if (skb == NULL)
1894                         return -ENOBUFS;
1895                 /* FIXME: Save some space for broken drivers that write a hard
1896                  * header at transmission time by themselves. PPP is the notable
1897                  * one here. This should really be fixed at the driver level.
1898                  */
1899                 skb_reserve(skb, reserved);
1900                 skb_reset_network_header(skb);
1901 
1902                 /* Try to align data part correctly */
1903                 if (hhlen) {
1904                         skb->data -= hhlen;
1905                         skb->tail -= hhlen;
1906                         if (len < hhlen)
1907                                 skb_reset_network_header(skb);
1908                 }
1909                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1910                 if (err)
1911                         goto out_free;
1912                 goto retry;
1913         }
1914 
1915         if (!dev_validate_header(dev, skb->data, len)) {
1916                 err = -EINVAL;
1917                 goto out_unlock;
1918         }
1919         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1920             !packet_extra_vlan_len_allowed(dev, skb)) {
1921                 err = -EMSGSIZE;
1922                 goto out_unlock;
1923         }
1924 
1925         sockc.tsflags = sk->sk_tsflags;
1926         if (msg->msg_controllen) {
1927                 err = sock_cmsg_send(sk, msg, &sockc);
1928                 if (unlikely(err))
1929                         goto out_unlock;
1930         }
1931 
1932         skb->protocol = proto;
1933         skb->dev = dev;
1934         skb->priority = sk->sk_priority;
1935         skb->mark = sk->sk_mark;
1936 
1937         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1938 
1939         if (unlikely(extra_len == 4))
1940                 skb->no_fcs = 1;
1941 
1942         skb_probe_transport_header(skb, 0);
1943 
1944         dev_queue_xmit(skb);
1945         rcu_read_unlock();
1946         return len;
1947 
1948 out_unlock:
1949         rcu_read_unlock();
1950 out_free:
1951         kfree_skb(skb);
1952         return err;
1953 }
1954 
1955 static unsigned int run_filter(struct sk_buff *skb,
1956                                const struct sock *sk,
1957                                unsigned int res)
1958 {
1959         struct sk_filter *filter;
1960 
1961         rcu_read_lock();
1962         filter = rcu_dereference(sk->sk_filter);
1963         if (filter != NULL)
1964                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1965         rcu_read_unlock();
1966 
1967         return res;
1968 }
1969 
1970 static int __packet_rcv_vnet(const struct sk_buff *skb,
1971                              struct virtio_net_hdr *vnet_hdr)
1972 {
1973         *vnet_hdr = (const struct virtio_net_hdr) { 0 };
1974 
1975         if (virtio_net_hdr_from_skb(skb, vnet_hdr, vio_le()))
1976                 BUG();
1977 
1978         return 0;
1979 }
1980 
1981 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
1982                            size_t *len)
1983 {
1984         struct virtio_net_hdr vnet_hdr;
1985 
1986         if (*len < sizeof(vnet_hdr))
1987                 return -EINVAL;
1988         *len -= sizeof(vnet_hdr);
1989 
1990         if (__packet_rcv_vnet(skb, &vnet_hdr))
1991                 return -EINVAL;
1992 
1993         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
1994 }
1995 
1996 /*
1997  * This function makes lazy skb cloning in hope that most of packets
1998  * are discarded by BPF.
1999  *
2000  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2001  * and skb->cb are mangled. It works because (and until) packets
2002  * falling here are owned by current CPU. Output packets are cloned
2003  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2004  * sequencially, so that if we return skb to original state on exit,
2005  * we will not harm anyone.
2006  */
2007 
2008 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2009                       struct packet_type *pt, struct net_device *orig_dev)
2010 {
2011         struct sock *sk;
2012         struct sockaddr_ll *sll;
2013         struct packet_sock *po;
2014         u8 *skb_head = skb->data;
2015         int skb_len = skb->len;
2016         unsigned int snaplen, res;
2017         bool is_drop_n_account = false;
2018 
2019         if (skb->pkt_type == PACKET_LOOPBACK)
2020                 goto drop;
2021 
2022         sk = pt->af_packet_priv;
2023         po = pkt_sk(sk);
2024 
2025         if (!net_eq(dev_net(dev), sock_net(sk)))
2026                 goto drop;
2027 
2028         skb->dev = dev;
2029 
2030         if (dev->header_ops) {
2031                 /* The device has an explicit notion of ll header,
2032                  * exported to higher levels.
2033                  *
2034                  * Otherwise, the device hides details of its frame
2035                  * structure, so that corresponding packet head is
2036                  * never delivered to user.
2037                  */
2038                 if (sk->sk_type != SOCK_DGRAM)
2039                         skb_push(skb, skb->data - skb_mac_header(skb));
2040                 else if (skb->pkt_type == PACKET_OUTGOING) {
2041                         /* Special case: outgoing packets have ll header at head */
2042                         skb_pull(skb, skb_network_offset(skb));
2043                 }
2044         }
2045 
2046         snaplen = skb->len;
2047 
2048         res = run_filter(skb, sk, snaplen);
2049         if (!res)
2050                 goto drop_n_restore;
2051         if (snaplen > res)
2052                 snaplen = res;
2053 
2054         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2055                 goto drop_n_acct;
2056 
2057         if (skb_shared(skb)) {
2058                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2059                 if (nskb == NULL)
2060                         goto drop_n_acct;
2061 
2062                 if (skb_head != skb->data) {
2063                         skb->data = skb_head;
2064                         skb->len = skb_len;
2065                 }
2066                 consume_skb(skb);
2067                 skb = nskb;
2068         }
2069 
2070         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2071 
2072         sll = &PACKET_SKB_CB(skb)->sa.ll;
2073         sll->sll_hatype = dev->type;
2074         sll->sll_pkttype = skb->pkt_type;
2075         if (unlikely(po->origdev))
2076                 sll->sll_ifindex = orig_dev->ifindex;
2077         else
2078                 sll->sll_ifindex = dev->ifindex;
2079 
2080         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2081 
2082         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2083          * Use their space for storing the original skb length.
2084          */
2085         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2086 
2087         if (pskb_trim(skb, snaplen))
2088                 goto drop_n_acct;
2089 
2090         skb_set_owner_r(skb, sk);
2091         skb->dev = NULL;
2092         skb_dst_drop(skb);
2093 
2094         /* drop conntrack reference */
2095         nf_reset(skb);
2096 
2097         spin_lock(&sk->sk_receive_queue.lock);
2098         po->stats.stats1.tp_packets++;
2099         sock_skb_set_dropcount(sk, skb);
2100         __skb_queue_tail(&sk->sk_receive_queue, skb);
2101         spin_unlock(&sk->sk_receive_queue.lock);
2102         sk->sk_data_ready(sk);
2103         return 0;
2104 
2105 drop_n_acct:
2106         is_drop_n_account = true;
2107         spin_lock(&sk->sk_receive_queue.lock);
2108         po->stats.stats1.tp_drops++;
2109         atomic_inc(&sk->sk_drops);
2110         spin_unlock(&sk->sk_receive_queue.lock);
2111 
2112 drop_n_restore:
2113         if (skb_head != skb->data && skb_shared(skb)) {
2114                 skb->data = skb_head;
2115                 skb->len = skb_len;
2116         }
2117 drop:
2118         if (!is_drop_n_account)
2119                 consume_skb(skb);
2120         else
2121                 kfree_skb(skb);
2122         return 0;
2123 }
2124 
2125 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2126                        struct packet_type *pt, struct net_device *orig_dev)
2127 {
2128         struct sock *sk;
2129         struct packet_sock *po;
2130         struct sockaddr_ll *sll;
2131         union tpacket_uhdr h;
2132         u8 *skb_head = skb->data;
2133         int skb_len = skb->len;
2134         unsigned int snaplen, res;
2135         unsigned long status = TP_STATUS_USER;
2136         unsigned short macoff, netoff, hdrlen;
2137         struct sk_buff *copy_skb = NULL;
2138         struct timespec ts;
2139         __u32 ts_status;
2140         bool is_drop_n_account = false;
2141 
2142         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2143          * We may add members to them until current aligned size without forcing
2144          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2145          */
2146         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2147         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2148 
2149         if (skb->pkt_type == PACKET_LOOPBACK)
2150                 goto drop;
2151 
2152         sk = pt->af_packet_priv;
2153         po = pkt_sk(sk);
2154 
2155         if (!net_eq(dev_net(dev), sock_net(sk)))
2156                 goto drop;
2157 
2158         if (dev->header_ops) {
2159                 if (sk->sk_type != SOCK_DGRAM)
2160                         skb_push(skb, skb->data - skb_mac_header(skb));
2161                 else if (skb->pkt_type == PACKET_OUTGOING) {
2162                         /* Special case: outgoing packets have ll header at head */
2163                         skb_pull(skb, skb_network_offset(skb));
2164                 }
2165         }
2166 
2167         snaplen = skb->len;
2168 
2169         res = run_filter(skb, sk, snaplen);
2170         if (!res)
2171                 goto drop_n_restore;
2172 
2173         if (skb->ip_summed == CHECKSUM_PARTIAL)
2174                 status |= TP_STATUS_CSUMNOTREADY;
2175         else if (skb->pkt_type != PACKET_OUTGOING &&
2176                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2177                   skb_csum_unnecessary(skb)))
2178                 status |= TP_STATUS_CSUM_VALID;
2179 
2180         if (snaplen > res)
2181                 snaplen = res;
2182 
2183         if (sk->sk_type == SOCK_DGRAM) {
2184                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2185                                   po->tp_reserve;
2186         } else {
2187                 unsigned int maclen = skb_network_offset(skb);
2188                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2189                                        (maclen < 16 ? 16 : maclen)) +
2190                                        po->tp_reserve;
2191                 if (po->has_vnet_hdr)
2192                         netoff += sizeof(struct virtio_net_hdr);
2193                 macoff = netoff - maclen;
2194         }
2195         if (po->tp_version <= TPACKET_V2) {
2196                 if (macoff + snaplen > po->rx_ring.frame_size) {
2197                         if (po->copy_thresh &&
2198                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2199                                 if (skb_shared(skb)) {
2200                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2201                                 } else {
2202                                         copy_skb = skb_get(skb);
2203                                         skb_head = skb->data;
2204                                 }
2205                                 if (copy_skb)
2206                                         skb_set_owner_r(copy_skb, sk);
2207                         }
2208                         snaplen = po->rx_ring.frame_size - macoff;
2209                         if ((int)snaplen < 0)
2210                                 snaplen = 0;
2211                 }
2212         } else if (unlikely(macoff + snaplen >
2213                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2214                 u32 nval;
2215 
2216                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2217                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2218                             snaplen, nval, macoff);
2219                 snaplen = nval;
2220                 if (unlikely((int)snaplen < 0)) {
2221                         snaplen = 0;
2222                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2223                 }
2224         }
2225         spin_lock(&sk->sk_receive_queue.lock);
2226         h.raw = packet_current_rx_frame(po, skb,
2227                                         TP_STATUS_KERNEL, (macoff+snaplen));
2228         if (!h.raw)
2229                 goto drop_n_account;
2230         if (po->tp_version <= TPACKET_V2) {
2231                 packet_increment_rx_head(po, &po->rx_ring);
2232         /*
2233          * LOSING will be reported till you read the stats,
2234          * because it's COR - Clear On Read.
2235          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2236          * at packet level.
2237          */
2238                 if (po->stats.stats1.tp_drops)
2239                         status |= TP_STATUS_LOSING;
2240         }
2241         po->stats.stats1.tp_packets++;
2242         if (copy_skb) {
2243                 status |= TP_STATUS_COPY;
2244                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2245         }
2246         spin_unlock(&sk->sk_receive_queue.lock);
2247 
2248         if (po->has_vnet_hdr) {
2249                 if (__packet_rcv_vnet(skb, h.raw + macoff -
2250                                            sizeof(struct virtio_net_hdr))) {
2251                         spin_lock(&sk->sk_receive_queue.lock);
2252                         goto drop_n_account;
2253                 }
2254         }
2255 
2256         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2257 
2258         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2259                 getnstimeofday(&ts);
2260 
2261         status |= ts_status;
2262 
2263         switch (po->tp_version) {
2264         case TPACKET_V1:
2265                 h.h1->tp_len = skb->len;
2266                 h.h1->tp_snaplen = snaplen;
2267                 h.h1->tp_mac = macoff;
2268                 h.h1->tp_net = netoff;
2269                 h.h1->tp_sec = ts.tv_sec;
2270                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2271                 hdrlen = sizeof(*h.h1);
2272                 break;
2273         case TPACKET_V2:
2274                 h.h2->tp_len = skb->len;
2275                 h.h2->tp_snaplen = snaplen;
2276                 h.h2->tp_mac = macoff;
2277                 h.h2->tp_net = netoff;
2278                 h.h2->tp_sec = ts.tv_sec;
2279                 h.h2->tp_nsec = ts.tv_nsec;
2280                 if (skb_vlan_tag_present(skb)) {
2281                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2282                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2283                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2284                 } else {
2285                         h.h2->tp_vlan_tci = 0;
2286                         h.h2->tp_vlan_tpid = 0;
2287                 }
2288                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2289                 hdrlen = sizeof(*h.h2);
2290                 break;
2291         case TPACKET_V3:
2292                 /* tp_nxt_offset,vlan are already populated above.
2293                  * So DONT clear those fields here
2294                  */
2295                 h.h3->tp_status |= status;
2296                 h.h3->tp_len = skb->len;
2297                 h.h3->tp_snaplen = snaplen;
2298                 h.h3->tp_mac = macoff;
2299                 h.h3->tp_net = netoff;
2300                 h.h3->tp_sec  = ts.tv_sec;
2301                 h.h3->tp_nsec = ts.tv_nsec;
2302                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2303                 hdrlen = sizeof(*h.h3);
2304                 break;
2305         default:
2306                 BUG();
2307         }
2308 
2309         sll = h.raw + TPACKET_ALIGN(hdrlen);
2310         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2311         sll->sll_family = AF_PACKET;
2312         sll->sll_hatype = dev->type;
2313         sll->sll_protocol = skb->protocol;
2314         sll->sll_pkttype = skb->pkt_type;
2315         if (unlikely(po->origdev))
2316                 sll->sll_ifindex = orig_dev->ifindex;
2317         else
2318                 sll->sll_ifindex = dev->ifindex;
2319 
2320         smp_mb();
2321 
2322 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2323         if (po->tp_version <= TPACKET_V2) {
2324                 u8 *start, *end;
2325 
2326                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2327                                         macoff + snaplen);
2328 
2329                 for (start = h.raw; start < end; start += PAGE_SIZE)
2330                         flush_dcache_page(pgv_to_page(start));
2331         }
2332         smp_wmb();
2333 #endif
2334 
2335         if (po->tp_version <= TPACKET_V2) {
2336                 __packet_set_status(po, h.raw, status);
2337                 sk->sk_data_ready(sk);
2338         } else {
2339                 prb_clear_blk_fill_status(&po->rx_ring);
2340         }
2341 
2342 drop_n_restore:
2343         if (skb_head != skb->data && skb_shared(skb)) {
2344                 skb->data = skb_head;
2345                 skb->len = skb_len;
2346         }
2347 drop:
2348         if (!is_drop_n_account)
2349                 consume_skb(skb);
2350         else
2351                 kfree_skb(skb);
2352         return 0;
2353 
2354 drop_n_account:
2355         is_drop_n_account = true;
2356         po->stats.stats1.tp_drops++;
2357         spin_unlock(&sk->sk_receive_queue.lock);
2358 
2359         sk->sk_data_ready(sk);
2360         kfree_skb(copy_skb);
2361         goto drop_n_restore;
2362 }
2363 
2364 static void tpacket_destruct_skb(struct sk_buff *skb)
2365 {
2366         struct packet_sock *po = pkt_sk(skb->sk);
2367 
2368         if (likely(po->tx_ring.pg_vec)) {
2369                 void *ph;
2370                 __u32 ts;
2371 
2372                 ph = skb_shinfo(skb)->destructor_arg;
2373                 packet_dec_pending(&po->tx_ring);
2374 
2375                 ts = __packet_set_timestamp(po, ph, skb);
2376                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2377         }
2378 
2379         sock_wfree(skb);
2380 }
2381 
2382 static void tpacket_set_protocol(const struct net_device *dev,
2383                                  struct sk_buff *skb)
2384 {
2385         if (dev->type == ARPHRD_ETHER) {
2386                 skb_reset_mac_header(skb);
2387                 skb->protocol = eth_hdr(skb)->h_proto;
2388         }
2389 }
2390 
2391 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2392 {
2393         unsigned short gso_type = 0;
2394 
2395         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2396             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2397              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2398               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2399                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2400                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2401                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2402 
2403         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2404                 return -EINVAL;
2405 
2406         if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2407                 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2408                 case VIRTIO_NET_HDR_GSO_TCPV4:
2409                         gso_type = SKB_GSO_TCPV4;
2410                         break;
2411                 case VIRTIO_NET_HDR_GSO_TCPV6:
2412                         gso_type = SKB_GSO_TCPV6;
2413                         break;
2414                 case VIRTIO_NET_HDR_GSO_UDP:
2415                         gso_type = SKB_GSO_UDP;
2416                         break;
2417                 default:
2418                         return -EINVAL;
2419                 }
2420 
2421                 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
2422                         gso_type |= SKB_GSO_TCP_ECN;
2423 
2424                 if (vnet_hdr->gso_size == 0)
2425                         return -EINVAL;
2426         }
2427 
2428         vnet_hdr->gso_type = gso_type;  /* changes type, temporary storage */
2429         return 0;
2430 }
2431 
2432 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2433                                  struct virtio_net_hdr *vnet_hdr)
2434 {
2435         int n;
2436 
2437         if (*len < sizeof(*vnet_hdr))
2438                 return -EINVAL;
2439         *len -= sizeof(*vnet_hdr);
2440 
2441         n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter);
2442         if (n != sizeof(*vnet_hdr))
2443                 return -EFAULT;
2444 
2445         return __packet_snd_vnet_parse(vnet_hdr, *len);
2446 }
2447 
2448 static int packet_snd_vnet_gso(struct sk_buff *skb,
2449                                struct virtio_net_hdr *vnet_hdr)
2450 {
2451         if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2452                 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start);
2453                 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset);
2454 
2455                 if (!skb_partial_csum_set(skb, s, o))
2456                         return -EINVAL;
2457         }
2458 
2459         skb_shinfo(skb)->gso_size =
2460                 __virtio16_to_cpu(vio_le(), vnet_hdr->gso_size);
2461         skb_shinfo(skb)->gso_type = vnet_hdr->gso_type;
2462 
2463         /* Header must be checked, and gso_segs computed. */
2464         skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2465         skb_shinfo(skb)->gso_segs = 0;
2466         return 0;
2467 }
2468 
2469 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2470                 void *frame, struct net_device *dev, void *data, int tp_len,
2471                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2472                 const struct sockcm_cookie *sockc)
2473 {
2474         union tpacket_uhdr ph;
2475         int to_write, offset, len, nr_frags, len_max;
2476         struct socket *sock = po->sk.sk_socket;
2477         struct page *page;
2478         int err;
2479 
2480         ph.raw = frame;
2481 
2482         skb->protocol = proto;
2483         skb->dev = dev;
2484         skb->priority = po->sk.sk_priority;
2485         skb->mark = po->sk.sk_mark;
2486         sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2487         skb_shinfo(skb)->destructor_arg = ph.raw;
2488 
2489         skb_reserve(skb, hlen);
2490         skb_reset_network_header(skb);
2491 
2492         to_write = tp_len;
2493 
2494         if (sock->type == SOCK_DGRAM) {
2495                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2496                                 NULL, tp_len);
2497                 if (unlikely(err < 0))
2498                         return -EINVAL;
2499         } else if (copylen) {
2500                 int hdrlen = min_t(int, copylen, tp_len);
2501 
2502                 skb_push(skb, dev->hard_header_len);
2503                 skb_put(skb, copylen - dev->hard_header_len);
2504                 err = skb_store_bits(skb, 0, data, hdrlen);
2505                 if (unlikely(err))
2506                         return err;
2507                 if (!dev_validate_header(dev, skb->data, hdrlen))
2508                         return -EINVAL;
2509                 if (!skb->protocol)
2510                         tpacket_set_protocol(dev, skb);
2511 
2512                 data += hdrlen;
2513                 to_write -= hdrlen;
2514         }
2515 
2516         offset = offset_in_page(data);
2517         len_max = PAGE_SIZE - offset;
2518         len = ((to_write > len_max) ? len_max : to_write);
2519 
2520         skb->data_len = to_write;
2521         skb->len += to_write;
2522         skb->truesize += to_write;
2523         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2524 
2525         while (likely(to_write)) {
2526                 nr_frags = skb_shinfo(skb)->nr_frags;
2527 
2528                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2529                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2530                                MAX_SKB_FRAGS);
2531                         return -EFAULT;
2532                 }
2533 
2534                 page = pgv_to_page(data);
2535                 data += len;
2536                 flush_dcache_page(page);
2537                 get_page(page);
2538                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2539                 to_write -= len;
2540                 offset = 0;
2541                 len_max = PAGE_SIZE;
2542                 len = ((to_write > len_max) ? len_max : to_write);
2543         }
2544 
2545         skb_probe_transport_header(skb, 0);
2546 
2547         return tp_len;
2548 }
2549 
2550 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2551                                 int size_max, void **data)
2552 {
2553         union tpacket_uhdr ph;
2554         int tp_len, off;
2555 
2556         ph.raw = frame;
2557 
2558         switch (po->tp_version) {
2559         case TPACKET_V2:
2560                 tp_len = ph.h2->tp_len;
2561                 break;
2562         default:
2563                 tp_len = ph.h1->tp_len;
2564                 break;
2565         }
2566         if (unlikely(tp_len > size_max)) {
2567                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2568                 return -EMSGSIZE;
2569         }
2570 
2571         if (unlikely(po->tp_tx_has_off)) {
2572                 int off_min, off_max;
2573 
2574                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2575                 off_max = po->tx_ring.frame_size - tp_len;
2576                 if (po->sk.sk_type == SOCK_DGRAM) {
2577                         switch (po->tp_version) {
2578                         case TPACKET_V2:
2579                                 off = ph.h2->tp_net;
2580                                 break;
2581                         default:
2582                                 off = ph.h1->tp_net;
2583                                 break;
2584                         }
2585                 } else {
2586                         switch (po->tp_version) {
2587                         case TPACKET_V2:
2588                                 off = ph.h2->tp_mac;
2589                                 break;
2590                         default:
2591                                 off = ph.h1->tp_mac;
2592                                 break;
2593                         }
2594                 }
2595                 if (unlikely((off < off_min) || (off_max < off)))
2596                         return -EINVAL;
2597         } else {
2598                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2599         }
2600 
2601         *data = frame + off;
2602         return tp_len;
2603 }
2604 
2605 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2606 {
2607         struct sk_buff *skb;
2608         struct net_device *dev;
2609         struct virtio_net_hdr *vnet_hdr = NULL;
2610         struct sockcm_cookie sockc;
2611         __be16 proto;
2612         int err, reserve = 0;
2613         void *ph;
2614         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2615         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2616         int tp_len, size_max;
2617         unsigned char *addr;
2618         void *data;
2619         int len_sum = 0;
2620         int status = TP_STATUS_AVAILABLE;
2621         int hlen, tlen, copylen = 0;
2622 
2623         mutex_lock(&po->pg_vec_lock);
2624 
2625         if (likely(saddr == NULL)) {
2626                 dev     = packet_cached_dev_get(po);
2627                 proto   = po->num;
2628                 addr    = NULL;
2629         } else {
2630                 err = -EINVAL;
2631                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2632                         goto out;
2633                 if (msg->msg_namelen < (saddr->sll_halen
2634                                         + offsetof(struct sockaddr_ll,
2635                                                 sll_addr)))
2636                         goto out;
2637                 proto   = saddr->sll_protocol;
2638                 addr    = saddr->sll_addr;
2639                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2640         }
2641 
2642         sockc.tsflags = po->sk.sk_tsflags;
2643         if (msg->msg_controllen) {
2644                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2645                 if (unlikely(err))
2646                         goto out;
2647         }
2648 
2649         err = -ENXIO;
2650         if (unlikely(dev == NULL))
2651                 goto out;
2652         err = -ENETDOWN;
2653         if (unlikely(!(dev->flags & IFF_UP)))
2654                 goto out_put;
2655 
2656         if (po->sk.sk_socket->type == SOCK_RAW)
2657                 reserve = dev->hard_header_len;
2658         size_max = po->tx_ring.frame_size
2659                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2660 
2661         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2662                 size_max = dev->mtu + reserve + VLAN_HLEN;
2663 
2664         do {
2665                 ph = packet_current_frame(po, &po->tx_ring,
2666                                           TP_STATUS_SEND_REQUEST);
2667                 if (unlikely(ph == NULL)) {
2668                         if (need_wait && need_resched())
2669                                 schedule();
2670                         continue;
2671                 }
2672 
2673                 skb = NULL;
2674                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2675                 if (tp_len < 0)
2676                         goto tpacket_error;
2677 
2678                 status = TP_STATUS_SEND_REQUEST;
2679                 hlen = LL_RESERVED_SPACE(dev);
2680                 tlen = dev->needed_tailroom;
2681                 if (po->has_vnet_hdr) {
2682                         vnet_hdr = data;
2683                         data += sizeof(*vnet_hdr);
2684                         tp_len -= sizeof(*vnet_hdr);
2685                         if (tp_len < 0 ||
2686                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2687                                 tp_len = -EINVAL;
2688                                 goto tpacket_error;
2689                         }
2690                         copylen = __virtio16_to_cpu(vio_le(),
2691                                                     vnet_hdr->hdr_len);
2692                 }
2693                 copylen = max_t(int, copylen, dev->hard_header_len);
2694                 skb = sock_alloc_send_skb(&po->sk,
2695                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2696                                 (copylen - dev->hard_header_len),
2697                                 !need_wait, &err);
2698 
2699                 if (unlikely(skb == NULL)) {
2700                         /* we assume the socket was initially writeable ... */
2701                         if (likely(len_sum > 0))
2702                                 err = len_sum;
2703                         goto out_status;
2704                 }
2705                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2706                                           addr, hlen, copylen, &sockc);
2707                 if (likely(tp_len >= 0) &&
2708                     tp_len > dev->mtu + reserve &&
2709                     !po->has_vnet_hdr &&
2710                     !packet_extra_vlan_len_allowed(dev, skb))
2711                         tp_len = -EMSGSIZE;
2712 
2713                 if (unlikely(tp_len < 0)) {
2714 tpacket_error:
2715                         if (po->tp_loss) {
2716                                 __packet_set_status(po, ph,
2717                                                 TP_STATUS_AVAILABLE);
2718                                 packet_increment_head(&po->tx_ring);
2719                                 kfree_skb(skb);
2720                                 continue;
2721                         } else {
2722                                 status = TP_STATUS_WRONG_FORMAT;
2723                                 err = tp_len;
2724                                 goto out_status;
2725                         }
2726                 }
2727 
2728                 if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) {
2729                         tp_len = -EINVAL;
2730                         goto tpacket_error;
2731                 }
2732 
2733                 packet_pick_tx_queue(dev, skb);
2734 
2735                 skb->destructor = tpacket_destruct_skb;
2736                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2737                 packet_inc_pending(&po->tx_ring);
2738 
2739                 status = TP_STATUS_SEND_REQUEST;
2740                 err = po->xmit(skb);
2741                 if (unlikely(err > 0)) {
2742                         err = net_xmit_errno(err);
2743                         if (err && __packet_get_status(po, ph) ==
2744                                    TP_STATUS_AVAILABLE) {
2745                                 /* skb was destructed already */
2746                                 skb = NULL;
2747                                 goto out_status;
2748                         }
2749                         /*
2750                          * skb was dropped but not destructed yet;
2751                          * let's treat it like congestion or err < 0
2752                          */
2753                         err = 0;
2754                 }
2755                 packet_increment_head(&po->tx_ring);
2756                 len_sum += tp_len;
2757         } while (likely((ph != NULL) ||
2758                 /* Note: packet_read_pending() might be slow if we have
2759                  * to call it as it's per_cpu variable, but in fast-path
2760                  * we already short-circuit the loop with the first
2761                  * condition, and luckily don't have to go that path
2762                  * anyway.
2763                  */
2764                  (need_wait && packet_read_pending(&po->tx_ring))));
2765 
2766         err = len_sum;
2767         goto out_put;
2768 
2769 out_status:
2770         __packet_set_status(po, ph, status);
2771         kfree_skb(skb);
2772 out_put:
2773         dev_put(dev);
2774 out:
2775         mutex_unlock(&po->pg_vec_lock);
2776         return err;
2777 }
2778 
2779 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2780                                         size_t reserve, size_t len,
2781                                         size_t linear, int noblock,
2782                                         int *err)
2783 {
2784         struct sk_buff *skb;
2785 
2786         /* Under a page?  Don't bother with paged skb. */
2787         if (prepad + len < PAGE_SIZE || !linear)
2788                 linear = len;
2789 
2790         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2791                                    err, 0);
2792         if (!skb)
2793                 return NULL;
2794 
2795         skb_reserve(skb, reserve);
2796         skb_put(skb, linear);
2797         skb->data_len = len - linear;
2798         skb->len += len - linear;
2799 
2800         return skb;
2801 }
2802 
2803 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2804 {
2805         struct sock *sk = sock->sk;
2806         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2807         struct sk_buff *skb;
2808         struct net_device *dev;
2809         __be16 proto;
2810         unsigned char *addr;
2811         int err, reserve = 0;
2812         struct sockcm_cookie sockc;
2813         struct virtio_net_hdr vnet_hdr = { 0 };
2814         int offset = 0;
2815         struct packet_sock *po = pkt_sk(sk);
2816         int hlen, tlen;
2817         int extra_len = 0;
2818 
2819         /*
2820          *      Get and verify the address.
2821          */
2822 
2823         if (likely(saddr == NULL)) {
2824                 dev     = packet_cached_dev_get(po);
2825                 proto   = po->num;
2826                 addr    = NULL;
2827         } else {
2828                 err = -EINVAL;
2829                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2830                         goto out;
2831                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2832                         goto out;
2833                 proto   = saddr->sll_protocol;
2834                 addr    = saddr->sll_addr;
2835                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2836         }
2837 
2838         err = -ENXIO;
2839         if (unlikely(dev == NULL))
2840                 goto out_unlock;
2841         err = -ENETDOWN;
2842         if (unlikely(!(dev->flags & IFF_UP)))
2843                 goto out_unlock;
2844 
2845         sockc.tsflags = sk->sk_tsflags;
2846         sockc.mark = sk->sk_mark;
2847         if (msg->msg_controllen) {
2848                 err = sock_cmsg_send(sk, msg, &sockc);
2849                 if (unlikely(err))
2850                         goto out_unlock;
2851         }
2852 
2853         if (sock->type == SOCK_RAW)
2854                 reserve = dev->hard_header_len;
2855         if (po->has_vnet_hdr) {
2856                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2857                 if (err)
2858                         goto out_unlock;
2859         }
2860 
2861         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2862                 if (!netif_supports_nofcs(dev)) {
2863                         err = -EPROTONOSUPPORT;
2864                         goto out_unlock;
2865                 }
2866                 extra_len = 4; /* We're doing our own CRC */
2867         }
2868 
2869         err = -EMSGSIZE;
2870         if (!vnet_hdr.gso_type &&
2871             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2872                 goto out_unlock;
2873 
2874         err = -ENOBUFS;
2875         hlen = LL_RESERVED_SPACE(dev);
2876         tlen = dev->needed_tailroom;
2877         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2878                                __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2879                                msg->msg_flags & MSG_DONTWAIT, &err);
2880         if (skb == NULL)
2881                 goto out_unlock;
2882 
2883         skb_set_network_header(skb, reserve);
2884 
2885         err = -EINVAL;
2886         if (sock->type == SOCK_DGRAM) {
2887                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2888                 if (unlikely(offset < 0))
2889                         goto out_free;
2890         }
2891 
2892         /* Returns -EFAULT on error */
2893         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2894         if (err)
2895                 goto out_free;
2896 
2897         if (sock->type == SOCK_RAW &&
2898             !dev_validate_header(dev, skb->data, len)) {
2899                 err = -EINVAL;
2900                 goto out_free;
2901         }
2902 
2903         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2904 
2905         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2906             !packet_extra_vlan_len_allowed(dev, skb)) {
2907                 err = -EMSGSIZE;
2908                 goto out_free;
2909         }
2910 
2911         skb->protocol = proto;
2912         skb->dev = dev;
2913         skb->priority = sk->sk_priority;
2914         skb->mark = sockc.mark;
2915 
2916         packet_pick_tx_queue(dev, skb);
2917 
2918         if (po->has_vnet_hdr) {
2919                 err = packet_snd_vnet_gso(skb, &vnet_hdr);
2920                 if (err)
2921                         goto out_free;
2922                 len += sizeof(vnet_hdr);
2923         }
2924 
2925         skb_probe_transport_header(skb, reserve);
2926 
2927         if (unlikely(extra_len == 4))
2928                 skb->no_fcs = 1;
2929 
2930         err = po->xmit(skb);
2931         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2932                 goto out_unlock;
2933 
2934         dev_put(dev);
2935 
2936         return len;
2937 
2938 out_free:
2939         kfree_skb(skb);
2940 out_unlock:
2941         if (dev)
2942                 dev_put(dev);
2943 out:
2944         return err;
2945 }
2946 
2947 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2948 {
2949         struct sock *sk = sock->sk;
2950         struct packet_sock *po = pkt_sk(sk);
2951 
2952         if (po->tx_ring.pg_vec)
2953                 return tpacket_snd(po, msg);
2954         else
2955                 return packet_snd(sock, msg, len);
2956 }
2957 
2958 /*
2959  *      Close a PACKET socket. This is fairly simple. We immediately go
2960  *      to 'closed' state and remove our protocol entry in the device list.
2961  */
2962 
2963 static int packet_release(struct socket *sock)
2964 {
2965         struct sock *sk = sock->sk;
2966         struct packet_sock *po;
2967         struct net *net;
2968         union tpacket_req_u req_u;
2969 
2970         if (!sk)
2971                 return 0;
2972 
2973         net = sock_net(sk);
2974         po = pkt_sk(sk);
2975 
2976         mutex_lock(&net->packet.sklist_lock);
2977         sk_del_node_init_rcu(sk);
2978         mutex_unlock(&net->packet.sklist_lock);
2979 
2980         preempt_disable();
2981         sock_prot_inuse_add(net, sk->sk_prot, -1);
2982         preempt_enable();
2983 
2984         spin_lock(&po->bind_lock);
2985         unregister_prot_hook(sk, false);
2986         packet_cached_dev_reset(po);
2987 
2988         if (po->prot_hook.dev) {
2989                 dev_put(po->prot_hook.dev);
2990                 po->prot_hook.dev = NULL;
2991         }
2992         spin_unlock(&po->bind_lock);
2993 
2994         packet_flush_mclist(sk);
2995 
2996         if (po->rx_ring.pg_vec) {
2997                 memset(&req_u, 0, sizeof(req_u));
2998                 packet_set_ring(sk, &req_u, 1, 0);
2999         }
3000 
3001         if (po->tx_ring.pg_vec) {
3002                 memset(&req_u, 0, sizeof(req_u));
3003                 packet_set_ring(sk, &req_u, 1, 1);
3004         }
3005 
3006         fanout_release(sk);
3007 
3008         synchronize_net();
3009         /*
3010          *      Now the socket is dead. No more input will appear.
3011          */
3012         sock_orphan(sk);
3013         sock->sk = NULL;
3014 
3015         /* Purge queues */
3016 
3017         skb_queue_purge(&sk->sk_receive_queue);
3018         packet_free_pending(po);
3019         sk_refcnt_debug_release(sk);
3020 
3021         sock_put(sk);
3022         return 0;
3023 }
3024 
3025 /*
3026  *      Attach a packet hook.
3027  */
3028 
3029 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3030                           __be16 proto)
3031 {
3032         struct packet_sock *po = pkt_sk(sk);
3033         struct net_device *dev_curr;
3034         __be16 proto_curr;
3035         bool need_rehook;
3036         struct net_device *dev = NULL;
3037         int ret = 0;
3038         bool unlisted = false;
3039 
3040         if (po->fanout)
3041                 return -EINVAL;
3042 
3043         lock_sock(sk);
3044         spin_lock(&po->bind_lock);
3045         rcu_read_lock();
3046 
3047         if (name) {
3048                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3049                 if (!dev) {
3050                         ret = -ENODEV;
3051                         goto out_unlock;
3052                 }
3053         } else if (ifindex) {
3054                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3055                 if (!dev) {
3056                         ret = -ENODEV;
3057                         goto out_unlock;
3058                 }
3059         }
3060 
3061         if (dev)
3062                 dev_hold(dev);
3063 
3064         proto_curr = po->prot_hook.type;
3065         dev_curr = po->prot_hook.dev;
3066 
3067         need_rehook = proto_curr != proto || dev_curr != dev;
3068 
3069         if (need_rehook) {
3070                 if (po->running) {
3071                         rcu_read_unlock();
3072                         __unregister_prot_hook(sk, true);
3073                         rcu_read_lock();
3074                         dev_curr = po->prot_hook.dev;
3075                         if (dev)
3076                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3077                                                                  dev->ifindex);
3078                 }
3079 
3080                 po->num = proto;
3081                 po->prot_hook.type = proto;
3082 
3083                 if (unlikely(unlisted)) {
3084                         dev_put(dev);
3085                         po->prot_hook.dev = NULL;
3086                         po->ifindex = -1;
3087                         packet_cached_dev_reset(po);
3088                 } else {
3089                         po->prot_hook.dev = dev;
3090                         po->ifindex = dev ? dev->ifindex : 0;
3091                         packet_cached_dev_assign(po, dev);
3092                 }
3093         }
3094         if (dev_curr)
3095                 dev_put(dev_curr);
3096 
3097         if (proto == 0 || !need_rehook)
3098                 goto out_unlock;
3099 
3100         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3101                 register_prot_hook(sk);
3102         } else {
3103                 sk->sk_err = ENETDOWN;
3104                 if (!sock_flag(sk, SOCK_DEAD))
3105                         sk->sk_error_report(sk);
3106         }
3107 
3108 out_unlock:
3109         rcu_read_unlock();
3110         spin_unlock(&po->bind_lock);
3111         release_sock(sk);
3112         return ret;
3113 }
3114 
3115 /*
3116  *      Bind a packet socket to a device
3117  */
3118 
3119 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3120                             int addr_len)
3121 {
3122         struct sock *sk = sock->sk;
3123         char name[15];
3124 
3125         /*
3126          *      Check legality
3127          */
3128 
3129         if (addr_len != sizeof(struct sockaddr))
3130                 return -EINVAL;
3131         strlcpy(name, uaddr->sa_data, sizeof(name));
3132 
3133         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3134 }
3135 
3136 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3137 {
3138         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3139         struct sock *sk = sock->sk;
3140 
3141         /*
3142          *      Check legality
3143          */
3144 
3145         if (addr_len < sizeof(struct sockaddr_ll))
3146                 return -EINVAL;
3147         if (sll->sll_family != AF_PACKET)
3148                 return -EINVAL;
3149 
3150         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3151                               sll->sll_protocol ? : pkt_sk(sk)->num);
3152 }
3153 
3154 static struct proto packet_proto = {
3155         .name     = "PACKET",
3156         .owner    = THIS_MODULE,
3157         .obj_size = sizeof(struct packet_sock),
3158 };
3159 
3160 /*
3161  *      Create a packet of type SOCK_PACKET.
3162  */
3163 
3164 static int packet_create(struct net *net, struct socket *sock, int protocol,
3165                          int kern)
3166 {
3167         struct sock *sk;
3168         struct packet_sock *po;
3169         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3170         int err;
3171 
3172         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3173                 return -EPERM;
3174         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3175             sock->type != SOCK_PACKET)
3176                 return -ESOCKTNOSUPPORT;
3177 
3178         sock->state = SS_UNCONNECTED;
3179 
3180         err = -ENOBUFS;
3181         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3182         if (sk == NULL)
3183                 goto out;
3184 
3185         sock->ops = &packet_ops;
3186         if (sock->type == SOCK_PACKET)
3187                 sock->ops = &packet_ops_spkt;
3188 
3189         sock_init_data(sock, sk);
3190 
3191         po = pkt_sk(sk);
3192         sk->sk_family = PF_PACKET;
3193         po->num = proto;
3194         po->xmit = dev_queue_xmit;
3195 
3196         err = packet_alloc_pending(po);
3197         if (err)
3198                 goto out2;
3199 
3200         packet_cached_dev_reset(po);
3201 
3202         sk->sk_destruct = packet_sock_destruct;
3203         sk_refcnt_debug_inc(sk);
3204 
3205         /*
3206          *      Attach a protocol block
3207          */
3208 
3209         spin_lock_init(&po->bind_lock);
3210         mutex_init(&po->pg_vec_lock);
3211         po->rollover = NULL;
3212         po->prot_hook.func = packet_rcv;
3213 
3214         if (sock->type == SOCK_PACKET)
3215                 po->prot_hook.func = packet_rcv_spkt;
3216 
3217         po->prot_hook.af_packet_priv = sk;
3218 
3219         if (proto) {
3220                 po->prot_hook.type = proto;
3221                 register_prot_hook(sk);
3222         }
3223 
3224         mutex_lock(&net->packet.sklist_lock);
3225         sk_add_node_rcu(sk, &net->packet.sklist);
3226         mutex_unlock(&net->packet.sklist_lock);
3227 
3228         preempt_disable();
3229         sock_prot_inuse_add(net, &packet_proto, 1);
3230         preempt_enable();
3231 
3232         return 0;
3233 out2:
3234         sk_free(sk);
3235 out:
3236         return err;
3237 }
3238 
3239 /*
3240  *      Pull a packet from our receive queue and hand it to the user.
3241  *      If necessary we block.
3242  */
3243 
3244 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3245                           int flags)
3246 {
3247         struct sock *sk = sock->sk;
3248         struct sk_buff *skb;
3249         int copied, err;
3250         int vnet_hdr_len = 0;
3251         unsigned int origlen = 0;
3252 
3253         err = -EINVAL;
3254         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3255                 goto out;
3256 
3257 #if 0
3258         /* What error should we return now? EUNATTACH? */
3259         if (pkt_sk(sk)->ifindex < 0)
3260                 return -ENODEV;
3261 #endif
3262 
3263         if (flags & MSG_ERRQUEUE) {
3264                 err = sock_recv_errqueue(sk, msg, len,
3265                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3266                 goto out;
3267         }
3268 
3269         /*
3270          *      Call the generic datagram receiver. This handles all sorts
3271          *      of horrible races and re-entrancy so we can forget about it
3272          *      in the protocol layers.
3273          *
3274          *      Now it will return ENETDOWN, if device have just gone down,
3275          *      but then it will block.
3276          */
3277 
3278         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3279 
3280         /*
3281          *      An error occurred so return it. Because skb_recv_datagram()
3282          *      handles the blocking we don't see and worry about blocking
3283          *      retries.
3284          */
3285 
3286         if (skb == NULL)
3287                 goto out;
3288 
3289         if (pkt_sk(sk)->pressure)
3290                 packet_rcv_has_room(pkt_sk(sk), NULL);
3291 
3292         if (pkt_sk(sk)->has_vnet_hdr) {
3293                 err = packet_rcv_vnet(msg, skb, &len);
3294                 if (err)
3295                         goto out_free;
3296                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3297         }
3298 
3299         /* You lose any data beyond the buffer you gave. If it worries
3300          * a user program they can ask the device for its MTU
3301          * anyway.
3302          */
3303         copied = skb->len;
3304         if (copied > len) {
3305                 copied = len;
3306                 msg->msg_flags |= MSG_TRUNC;
3307         }
3308 
3309         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3310         if (err)
3311                 goto out_free;
3312 
3313         if (sock->type != SOCK_PACKET) {
3314                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3315 
3316                 /* Original length was stored in sockaddr_ll fields */
3317                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3318                 sll->sll_family = AF_PACKET;
3319                 sll->sll_protocol = skb->protocol;
3320         }
3321 
3322         sock_recv_ts_and_drops(msg, sk, skb);
3323 
3324         if (msg->msg_name) {
3325                 /* If the address length field is there to be filled
3326                  * in, we fill it in now.
3327                  */
3328                 if (sock->type == SOCK_PACKET) {
3329                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3330                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3331                 } else {
3332                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3333 
3334                         msg->msg_namelen = sll->sll_halen +
3335                                 offsetof(struct sockaddr_ll, sll_addr);
3336                 }
3337                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3338                        msg->msg_namelen);
3339         }
3340 
3341         if (pkt_sk(sk)->auxdata) {
3342                 struct tpacket_auxdata aux;
3343 
3344                 aux.tp_status = TP_STATUS_USER;
3345                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3346                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3347                 else if (skb->pkt_type != PACKET_OUTGOING &&
3348                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3349                           skb_csum_unnecessary(skb)))
3350                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3351 
3352                 aux.tp_len = origlen;
3353                 aux.tp_snaplen = skb->len;
3354                 aux.tp_mac = 0;
3355                 aux.tp_net = skb_network_offset(skb);
3356                 if (skb_vlan_tag_present(skb)) {
3357                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3358                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3359                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3360                 } else {
3361                         aux.tp_vlan_tci = 0;
3362                         aux.tp_vlan_tpid = 0;
3363                 }
3364                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3365         }
3366 
3367         /*
3368          *      Free or return the buffer as appropriate. Again this
3369          *      hides all the races and re-entrancy issues from us.
3370          */
3371         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3372 
3373 out_free:
3374         skb_free_datagram(sk, skb);
3375 out:
3376         return err;
3377 }
3378 
3379 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3380                                int *uaddr_len, int peer)
3381 {
3382         struct net_device *dev;
3383         struct sock *sk = sock->sk;
3384 
3385         if (peer)
3386                 return -EOPNOTSUPP;
3387 
3388         uaddr->sa_family = AF_PACKET;
3389         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3390         rcu_read_lock();
3391         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3392         if (dev)
3393                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3394         rcu_read_unlock();
3395         *uaddr_len = sizeof(*uaddr);
3396 
3397         return 0;
3398 }
3399 
3400 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3401                           int *uaddr_len, int peer)
3402 {
3403         struct net_device *dev;
3404         struct sock *sk = sock->sk;
3405         struct packet_sock *po = pkt_sk(sk);
3406         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3407 
3408         if (peer)
3409                 return -EOPNOTSUPP;
3410 
3411         sll->sll_family = AF_PACKET;
3412         sll->sll_ifindex = po->ifindex;
3413         sll->sll_protocol = po->num;
3414         sll->sll_pkttype = 0;
3415         rcu_read_lock();
3416         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3417         if (dev) {
3418                 sll->sll_hatype = dev->type;
3419                 sll->sll_halen = dev->addr_len;
3420                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3421         } else {
3422                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3423                 sll->sll_halen = 0;
3424         }
3425         rcu_read_unlock();
3426         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3427 
3428         return 0;
3429 }
3430 
3431 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3432                          int what)
3433 {
3434         switch (i->type) {
3435         case PACKET_MR_MULTICAST:
3436                 if (i->alen != dev->addr_len)
3437                         return -EINVAL;
3438                 if (what > 0)
3439                         return dev_mc_add(dev, i->addr);
3440                 else
3441                         return dev_mc_del(dev, i->addr);
3442                 break;
3443         case PACKET_MR_PROMISC:
3444                 return dev_set_promiscuity(dev, what);
3445         case PACKET_MR_ALLMULTI:
3446                 return dev_set_allmulti(dev, what);
3447         case PACKET_MR_UNICAST:
3448                 if (i->alen != dev->addr_len)
3449                         return -EINVAL;
3450                 if (what > 0)
3451                         return dev_uc_add(dev, i->addr);
3452                 else
3453                         return dev_uc_del(dev, i->addr);
3454                 break;
3455         default:
3456                 break;
3457         }
3458         return 0;
3459 }
3460 
3461 static void packet_dev_mclist_delete(struct net_device *dev,
3462                                      struct packet_mclist **mlp)
3463 {
3464         struct packet_mclist *ml;
3465 
3466         while ((ml = *mlp) != NULL) {
3467                 if (ml->ifindex == dev->ifindex) {
3468                         packet_dev_mc(dev, ml, -1);
3469                         *mlp = ml->next;
3470                         kfree(ml);
3471                 } else
3472                         mlp = &ml->next;
3473         }
3474 }
3475 
3476 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3477 {
3478         struct packet_sock *po = pkt_sk(sk);
3479         struct packet_mclist *ml, *i;
3480         struct net_device *dev;
3481         int err;
3482 
3483         rtnl_lock();
3484 
3485         err = -ENODEV;
3486         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3487         if (!dev)
3488                 goto done;
3489 
3490         err = -EINVAL;
3491         if (mreq->mr_alen > dev->addr_len)
3492                 goto done;
3493 
3494         err = -ENOBUFS;
3495         i = kmalloc(sizeof(*i), GFP_KERNEL);
3496         if (i == NULL)
3497                 goto done;
3498 
3499         err = 0;
3500         for (ml = po->mclist; ml; ml = ml->next) {
3501                 if (ml->ifindex == mreq->mr_ifindex &&
3502                     ml->type == mreq->mr_type &&
3503                     ml->alen == mreq->mr_alen &&
3504                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3505                         ml->count++;
3506                         /* Free the new element ... */
3507                         kfree(i);
3508                         goto done;
3509                 }
3510         }
3511 
3512         i->type = mreq->mr_type;
3513         i->ifindex = mreq->mr_ifindex;
3514         i->alen = mreq->mr_alen;
3515         memcpy(i->addr, mreq->mr_address, i->alen);
3516         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3517         i->count = 1;
3518         i->next = po->mclist;
3519         po->mclist = i;
3520         err = packet_dev_mc(dev, i, 1);
3521         if (err) {
3522                 po->mclist = i->next;
3523                 kfree(i);
3524         }
3525 
3526 done:
3527         rtnl_unlock();
3528         return err;
3529 }
3530 
3531 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3532 {
3533         struct packet_mclist *ml, **mlp;
3534 
3535         rtnl_lock();
3536 
3537         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3538                 if (ml->ifindex == mreq->mr_ifindex &&
3539                     ml->type == mreq->mr_type &&
3540                     ml->alen == mreq->mr_alen &&
3541                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3542                         if (--ml->count == 0) {
3543                                 struct net_device *dev;
3544                                 *mlp = ml->next;
3545                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3546                                 if (dev)
3547                                         packet_dev_mc(dev, ml, -1);
3548                                 kfree(ml);
3549                         }
3550                         break;
3551                 }
3552         }
3553         rtnl_unlock();
3554         return 0;
3555 }
3556 
3557 static void packet_flush_mclist(struct sock *sk)
3558 {
3559         struct packet_sock *po = pkt_sk(sk);
3560         struct packet_mclist *ml;
3561 
3562         if (!po->mclist)
3563                 return;
3564 
3565         rtnl_lock();
3566         while ((ml = po->mclist) != NULL) {
3567                 struct net_device *dev;
3568 
3569                 po->mclist = ml->next;
3570                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3571                 if (dev != NULL)
3572                         packet_dev_mc(dev, ml, -1);
3573                 kfree(ml);
3574         }
3575         rtnl_unlock();
3576 }
3577 
3578 static int
3579 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3580 {
3581         struct sock *sk = sock->sk;
3582         struct packet_sock *po = pkt_sk(sk);
3583         int ret;
3584 
3585         if (level != SOL_PACKET)
3586                 return -ENOPROTOOPT;
3587 
3588         switch (optname) {
3589         case PACKET_ADD_MEMBERSHIP:
3590         case PACKET_DROP_MEMBERSHIP:
3591         {
3592                 struct packet_mreq_max mreq;
3593                 int len = optlen;
3594                 memset(&mreq, 0, sizeof(mreq));
3595                 if (len < sizeof(struct packet_mreq))
3596                         return -EINVAL;
3597                 if (len > sizeof(mreq))
3598                         len = sizeof(mreq);
3599                 if (copy_from_user(&mreq, optval, len))
3600                         return -EFAULT;
3601                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3602                         return -EINVAL;
3603                 if (optname == PACKET_ADD_MEMBERSHIP)
3604                         ret = packet_mc_add(sk, &mreq);
3605                 else
3606                         ret = packet_mc_drop(sk, &mreq);
3607                 return ret;
3608         }
3609 
3610         case PACKET_RX_RING:
3611         case PACKET_TX_RING:
3612         {
3613                 union tpacket_req_u req_u;
3614                 int len;
3615 
3616                 switch (po->tp_version) {
3617                 case TPACKET_V1:
3618                 case TPACKET_V2:
3619                         len = sizeof(req_u.req);
3620                         break;
3621                 case TPACKET_V3:
3622                 default:
3623                         len = sizeof(req_u.req3);
3624                         break;
3625                 }
3626                 if (optlen < len)
3627                         return -EINVAL;
3628                 if (copy_from_user(&req_u.req, optval, len))
3629                         return -EFAULT;
3630                 return packet_set_ring(sk, &req_u, 0,
3631                         optname == PACKET_TX_RING);
3632         }
3633         case PACKET_COPY_THRESH:
3634         {
3635                 int val;
3636 
3637                 if (optlen != sizeof(val))
3638                         return -EINVAL;
3639                 if (copy_from_user(&val, optval, sizeof(val)))
3640                         return -EFAULT;
3641 
3642                 pkt_sk(sk)->copy_thresh = val;
3643                 return 0;
3644         }
3645         case PACKET_VERSION:
3646         {
3647                 int val;
3648 
3649                 if (optlen != sizeof(val))
3650                         return -EINVAL;
3651                 if (copy_from_user(&val, optval, sizeof(val)))
3652                         return -EFAULT;
3653                 switch (val) {
3654                 case TPACKET_V1:
3655                 case TPACKET_V2:
3656                 case TPACKET_V3:
3657                         break;
3658                 default:
3659                         return -EINVAL;
3660                 }
3661                 lock_sock(sk);
3662                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3663                         ret = -EBUSY;
3664                 } else {
3665                         po->tp_version = val;
3666                         ret = 0;
3667                 }
3668                 release_sock(sk);
3669                 return ret;
3670         }
3671         case PACKET_RESERVE:
3672         {
3673                 unsigned int val;
3674 
3675                 if (optlen != sizeof(val))
3676                         return -EINVAL;
3677                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3678                         return -EBUSY;
3679                 if (copy_from_user(&val, optval, sizeof(val)))
3680                         return -EFAULT;
3681                 po->tp_reserve = val;
3682                 return 0;
3683         }
3684         case PACKET_LOSS:
3685         {
3686                 unsigned int val;
3687 
3688                 if (optlen != sizeof(val))
3689                         return -EINVAL;
3690                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3691                         return -EBUSY;
3692                 if (copy_from_user(&val, optval, sizeof(val)))
3693                         return -EFAULT;
3694                 po->tp_loss = !!val;
3695                 return 0;
3696         }
3697         case PACKET_AUXDATA:
3698         {
3699                 int val;
3700 
3701                 if (optlen < sizeof(val))
3702                         return -EINVAL;
3703                 if (copy_from_user(&val, optval, sizeof(val)))
3704                         return -EFAULT;
3705 
3706                 po->auxdata = !!val;
3707                 return 0;
3708         }
3709         case PACKET_ORIGDEV:
3710         {
3711                 int val;
3712 
3713                 if (optlen < sizeof(val))
3714                         return -EINVAL;
3715                 if (copy_from_user(&val, optval, sizeof(val)))
3716                         return -EFAULT;
3717 
3718                 po->origdev = !!val;
3719                 return 0;
3720         }
3721         case PACKET_VNET_HDR:
3722         {
3723                 int val;
3724 
3725                 if (sock->type != SOCK_RAW)
3726                         return -EINVAL;
3727                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3728                         return -EBUSY;
3729                 if (optlen < sizeof(val))
3730                         return -EINVAL;
3731                 if (copy_from_user(&val, optval, sizeof(val)))
3732                         return -EFAULT;
3733 
3734                 po->has_vnet_hdr = !!val;
3735                 return 0;
3736         }
3737         case PACKET_TIMESTAMP:
3738         {
3739                 int val;
3740 
3741                 if (optlen != sizeof(val))
3742                         return -EINVAL;
3743                 if (copy_from_user(&val, optval, sizeof(val)))
3744                         return -EFAULT;
3745 
3746                 po->tp_tstamp = val;
3747                 return 0;
3748         }
3749         case PACKET_FANOUT:
3750         {
3751                 int val;
3752 
3753                 if (optlen != sizeof(val))
3754                         return -EINVAL;
3755                 if (copy_from_user(&val, optval, sizeof(val)))
3756                         return -EFAULT;
3757 
3758                 return fanout_add(sk, val & 0xffff, val >> 16);
3759         }
3760         case PACKET_FANOUT_DATA:
3761         {
3762                 if (!po->fanout)
3763                         return -EINVAL;
3764 
3765                 return fanout_set_data(po, optval, optlen);
3766         }
3767         case PACKET_TX_HAS_OFF:
3768         {
3769                 unsigned int val;
3770 
3771                 if (optlen != sizeof(val))
3772                         return -EINVAL;
3773                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3774                         return -EBUSY;
3775                 if (copy_from_user(&val, optval, sizeof(val)))
3776                         return -EFAULT;
3777                 po->tp_tx_has_off = !!val;
3778                 return 0;
3779         }
3780         case PACKET_QDISC_BYPASS:
3781         {
3782                 int val;
3783 
3784                 if (optlen != sizeof(val))
3785                         return -EINVAL;
3786                 if (copy_from_user(&val, optval, sizeof(val)))
3787                         return -EFAULT;
3788 
3789                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3790                 return 0;
3791         }
3792         default:
3793                 return -ENOPROTOOPT;
3794         }
3795 }
3796 
3797 static int packet_getsockopt(struct socket *sock, int level, int optname,
3798                              char __user *optval, int __user *optlen)
3799 {
3800         int len;
3801         int val, lv = sizeof(val);
3802         struct sock *sk = sock->sk;
3803         struct packet_sock *po = pkt_sk(sk);
3804         void *data = &val;
3805         union tpacket_stats_u st;
3806         struct tpacket_rollover_stats rstats;
3807 
3808         if (level != SOL_PACKET)
3809                 return -ENOPROTOOPT;
3810 
3811         if (get_user(len, optlen))
3812                 return -EFAULT;
3813 
3814         if (len < 0)
3815                 return -EINVAL;
3816 
3817         switch (optname) {
3818         case PACKET_STATISTICS:
3819                 spin_lock_bh(&sk->sk_receive_queue.lock);
3820                 memcpy(&st, &po->stats, sizeof(st));
3821                 memset(&po->stats, 0, sizeof(po->stats));
3822                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3823 
3824                 if (po->tp_version == TPACKET_V3) {
3825                         lv = sizeof(struct tpacket_stats_v3);
3826                         st.stats3.tp_packets += st.stats3.tp_drops;
3827                         data = &st.stats3;
3828                 } else {
3829                         lv = sizeof(struct tpacket_stats);
3830                         st.stats1.tp_packets += st.stats1.tp_drops;
3831                         data = &st.stats1;
3832                 }
3833 
3834                 break;
3835         case PACKET_AUXDATA:
3836                 val = po->auxdata;
3837                 break;
3838         case PACKET_ORIGDEV:
3839                 val = po->origdev;
3840                 break;
3841         case PACKET_VNET_HDR:
3842                 val = po->has_vnet_hdr;
3843                 break;
3844         case PACKET_VERSION:
3845                 val = po->tp_version;
3846                 break;
3847         case PACKET_HDRLEN:
3848                 if (len > sizeof(int))
3849                         len = sizeof(int);
3850                 if (copy_from_user(&val, optval, len))
3851                         return -EFAULT;
3852                 switch (val) {
3853                 case TPACKET_V1:
3854                         val = sizeof(struct tpacket_hdr);
3855                         break;
3856                 case TPACKET_V2:
3857                         val = sizeof(struct tpacket2_hdr);
3858                         break;
3859                 case TPACKET_V3:
3860                         val = sizeof(struct tpacket3_hdr);
3861                         break;
3862                 default:
3863                         return -EINVAL;
3864                 }
3865                 break;
3866         case PACKET_RESERVE:
3867                 val = po->tp_reserve;
3868                 break;
3869         case PACKET_LOSS:
3870                 val = po->tp_loss;
3871                 break;
3872         case PACKET_TIMESTAMP:
3873                 val = po->tp_tstamp;
3874                 break;
3875         case PACKET_FANOUT:
3876                 val = (po->fanout ?
3877                        ((u32)po->fanout->id |
3878                         ((u32)po->fanout->type << 16) |
3879                         ((u32)po->fanout->flags << 24)) :
3880                        0);
3881                 break;
3882         case PACKET_ROLLOVER_STATS:
3883                 if (!po->rollover)
3884                         return -EINVAL;
3885                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3886                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3887                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3888                 data = &rstats;
3889                 lv = sizeof(rstats);
3890                 break;
3891         case PACKET_TX_HAS_OFF:
3892                 val = po->tp_tx_has_off;
3893                 break;
3894         case PACKET_QDISC_BYPASS:
3895                 val = packet_use_direct_xmit(po);
3896                 break;
3897         default:
3898                 return -ENOPROTOOPT;
3899         }
3900 
3901         if (len > lv)
3902                 len = lv;
3903         if (put_user(len, optlen))
3904                 return -EFAULT;
3905         if (copy_to_user(optval, data, len))
3906                 return -EFAULT;
3907         return 0;
3908 }
3909 
3910 
3911 #ifdef CONFIG_COMPAT
3912 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3913                                     char __user *optval, unsigned int optlen)
3914 {
3915         struct packet_sock *po = pkt_sk(sock->sk);
3916 
3917         if (level != SOL_PACKET)
3918                 return -ENOPROTOOPT;
3919 
3920         if (optname == PACKET_FANOUT_DATA &&
3921             po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3922                 optval = (char __user *)get_compat_bpf_fprog(optval);
3923                 if (!optval)
3924                         return -EFAULT;
3925                 optlen = sizeof(struct sock_fprog);
3926         }
3927 
3928         return packet_setsockopt(sock, level, optname, optval, optlen);
3929 }
3930 #endif
3931 
3932 static int packet_notifier(struct notifier_block *this,
3933                            unsigned long msg, void *ptr)
3934 {
3935         struct sock *sk;
3936         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3937         struct net *net = dev_net(dev);
3938 
3939         rcu_read_lock();
3940         sk_for_each_rcu(sk, &net->packet.sklist) {
3941                 struct packet_sock *po = pkt_sk(sk);
3942 
3943                 switch (msg) {
3944                 case NETDEV_UNREGISTER:
3945                         if (po->mclist)
3946                                 packet_dev_mclist_delete(dev, &po->mclist);
3947                         /* fallthrough */
3948 
3949                 case NETDEV_DOWN:
3950                         if (dev->ifindex == po->ifindex) {
3951                                 spin_lock(&po->bind_lock);
3952                                 if (po->running) {
3953                                         __unregister_prot_hook(sk, false);
3954                                         sk->sk_err = ENETDOWN;
3955                                         if (!sock_flag(sk, SOCK_DEAD))
3956                                                 sk->sk_error_report(sk);
3957                                 }
3958                                 if (msg == NETDEV_UNREGISTER) {
3959                                         packet_cached_dev_reset(po);
3960                                         fanout_release(sk);
3961                                         po->ifindex = -1;
3962                                         if (po->prot_hook.dev)
3963                                                 dev_put(po->prot_hook.dev);
3964                                         po->prot_hook.dev = NULL;
3965                                 }
3966                                 spin_unlock(&po->bind_lock);
3967                         }
3968                         break;
3969                 case NETDEV_UP:
3970                         if (dev->ifindex == po->ifindex) {
3971                                 spin_lock(&po->bind_lock);
3972                                 if (po->num)
3973                                         register_prot_hook(sk);
3974                                 spin_unlock(&po->bind_lock);
3975                         }
3976                         break;
3977                 }
3978         }
3979         rcu_read_unlock();
3980         return NOTIFY_DONE;
3981 }
3982 
3983 
3984 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3985                         unsigned long arg)
3986 {
3987         struct sock *sk = sock->sk;
3988 
3989         switch (cmd) {
3990         case SIOCOUTQ:
3991         {
3992                 int amount = sk_wmem_alloc_get(sk);
3993 
3994                 return put_user(amount, (int __user *)arg);
3995         }
3996         case SIOCINQ:
3997         {
3998                 struct sk_buff *skb;
3999                 int amount = 0;
4000 
4001                 spin_lock_bh(&sk->sk_receive_queue.lock);
4002                 skb = skb_peek(&sk->sk_receive_queue);
4003                 if (skb)
4004                         amount = skb->len;
4005                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4006                 return put_user(amount, (int __user *)arg);
4007         }
4008         case SIOCGSTAMP:
4009                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4010         case SIOCGSTAMPNS:
4011                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4012 
4013 #ifdef CONFIG_INET
4014         case SIOCADDRT:
4015         case SIOCDELRT:
4016         case SIOCDARP:
4017         case SIOCGARP:
4018         case SIOCSARP:
4019         case SIOCGIFADDR:
4020         case SIOCSIFADDR:
4021         case SIOCGIFBRDADDR:
4022         case SIOCSIFBRDADDR:
4023         case SIOCGIFNETMASK:
4024         case SIOCSIFNETMASK:
4025         case SIOCGIFDSTADDR:
4026         case SIOCSIFDSTADDR:
4027         case SIOCSIFFLAGS:
4028                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4029 #endif
4030 
4031         default:
4032                 return -ENOIOCTLCMD;
4033         }
4034         return 0;
4035 }
4036 
4037 static unsigned int packet_poll(struct file *file, struct socket *sock,
4038                                 poll_table *wait)
4039 {
4040         struct sock *sk = sock->sk;
4041         struct packet_sock *po = pkt_sk(sk);
4042         unsigned int mask = datagram_poll(file, sock, wait);
4043 
4044         spin_lock_bh(&sk->sk_receive_queue.lock);
4045         if (po->rx_ring.pg_vec) {
4046                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4047                         TP_STATUS_KERNEL))
4048                         mask |= POLLIN | POLLRDNORM;
4049         }
4050         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4051                 po->pressure = 0;
4052         spin_unlock_bh(&sk->sk_receive_queue.lock);
4053         spin_lock_bh(&sk->sk_write_queue.lock);
4054         if (po->tx_ring.pg_vec) {
4055                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4056                         mask |= POLLOUT | POLLWRNORM;
4057         }
4058         spin_unlock_bh(&sk->sk_write_queue.lock);
4059         return mask;
4060 }
4061 
4062 
4063 /* Dirty? Well, I still did not learn better way to account
4064  * for user mmaps.
4065  */
4066 
4067 static void packet_mm_open(struct vm_area_struct *vma)
4068 {
4069         struct file *file = vma->vm_file;
4070         struct socket *sock = file->private_data;
4071         struct sock *sk = sock->sk;
4072 
4073         if (sk)
4074                 atomic_inc(&pkt_sk(sk)->mapped);
4075 }
4076 
4077 static void packet_mm_close(struct vm_area_struct *vma)
4078 {
4079         struct file *file = vma->vm_file;
4080         struct socket *sock = file->private_data;
4081         struct sock *sk = sock->sk;
4082 
4083         if (sk)
4084                 atomic_dec(&pkt_sk(sk)->mapped);
4085 }
4086 
4087 static const struct vm_operations_struct packet_mmap_ops = {
4088         .open   =       packet_mm_open,
4089         .close  =       packet_mm_close,
4090 };
4091 
4092 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4093                         unsigned int len)
4094 {
4095         int i;
4096 
4097         for (i = 0; i < len; i++) {
4098                 if (likely(pg_vec[i].buffer)) {
4099                         if (is_vmalloc_addr(pg_vec[i].buffer))
4100                                 vfree(pg_vec[i].buffer);
4101                         else
4102                                 free_pages((unsigned long)pg_vec[i].buffer,
4103                                            order);
4104                         pg_vec[i].buffer = NULL;
4105                 }
4106         }
4107         kfree(pg_vec);
4108 }
4109 
4110 static char *alloc_one_pg_vec_page(unsigned long order)
4111 {
4112         char *buffer;
4113         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4114                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4115 
4116         buffer = (char *) __get_free_pages(gfp_flags, order);
4117         if (buffer)
4118                 return buffer;
4119 
4120         /* __get_free_pages failed, fall back to vmalloc */
4121         buffer = vzalloc((1 << order) * PAGE_SIZE);
4122         if (buffer)
4123                 return buffer;
4124 
4125         /* vmalloc failed, lets dig into swap here */
4126         gfp_flags &= ~__GFP_NORETRY;
4127         buffer = (char *) __get_free_pages(gfp_flags, order);
4128         if (buffer)
4129                 return buffer;
4130 
4131         /* complete and utter failure */
4132         return NULL;
4133 }
4134 
4135 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4136 {
4137         unsigned int block_nr = req->tp_block_nr;
4138         struct pgv *pg_vec;
4139         int i;
4140 
4141         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4142         if (unlikely(!pg_vec))
4143                 goto out;
4144 
4145         for (i = 0; i < block_nr; i++) {
4146                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4147                 if (unlikely(!pg_vec[i].buffer))
4148                         goto out_free_pgvec;
4149         }
4150 
4151 out:
4152         return pg_vec;
4153 
4154 out_free_pgvec:
4155         free_pg_vec(pg_vec, order, block_nr);
4156         pg_vec = NULL;
4157         goto out;
4158 }
4159 
4160 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4161                 int closing, int tx_ring)
4162 {
4163         struct pgv *pg_vec = NULL;
4164         struct packet_sock *po = pkt_sk(sk);
4165         int was_running, order = 0;
4166         struct packet_ring_buffer *rb;
4167         struct sk_buff_head *rb_queue;
4168         __be16 num;
4169         int err = -EINVAL;
4170         /* Added to avoid minimal code churn */
4171         struct tpacket_req *req = &req_u->req;
4172 
4173         lock_sock(sk);
4174         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4175         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4176                 net_warn_ratelimited("Tx-ring is not supported.\n");
4177                 goto out;
4178         }
4179 
4180         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4181         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4182 
4183         err = -EBUSY;
4184         if (!closing) {
4185                 if (atomic_read(&po->mapped))
4186                         goto out;
4187                 if (packet_read_pending(rb))
4188                         goto out;
4189         }
4190 
4191         if (req->tp_block_nr) {
4192                 /* Sanity tests and some calculations */
4193                 err = -EBUSY;
4194                 if (unlikely(rb->pg_vec))
4195                         goto out;
4196 
4197                 switch (po->tp_version) {
4198                 case TPACKET_V1:
4199                         po->tp_hdrlen = TPACKET_HDRLEN;
4200                         break;
4201                 case TPACKET_V2:
4202                         po->tp_hdrlen = TPACKET2_HDRLEN;
4203                         break;
4204                 case TPACKET_V3:
4205                         po->tp_hdrlen = TPACKET3_HDRLEN;
4206                         break;
4207                 }
4208 
4209                 err = -EINVAL;
4210                 if (unlikely((int)req->tp_block_size <= 0))
4211                         goto out;
4212                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4213                         goto out;
4214                 if (po->tp_version >= TPACKET_V3 &&
4215                     (int)(req->tp_block_size -
4216                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4217                         goto out;
4218                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4219                                         po->tp_reserve))
4220                         goto out;
4221                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4222                         goto out;
4223 
4224                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4225                 if (unlikely(rb->frames_per_block == 0))
4226                         goto out;
4227                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4228                                         req->tp_frame_nr))
4229                         goto out;
4230 
4231                 err = -ENOMEM;
4232                 order = get_order(req->tp_block_size);
4233                 pg_vec = alloc_pg_vec(req, order);
4234                 if (unlikely(!pg_vec))
4235                         goto out;
4236                 switch (po->tp_version) {
4237                 case TPACKET_V3:
4238                 /* Transmit path is not supported. We checked
4239                  * it above but just being paranoid
4240                  */
4241                         if (!tx_ring)
4242                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4243                         break;
4244                 default:
4245                         break;
4246                 }
4247         }
4248         /* Done */
4249         else {
4250                 err = -EINVAL;
4251                 if (unlikely(req->tp_frame_nr))
4252                         goto out;
4253         }
4254 
4255 
4256         /* Detach socket from network */
4257         spin_lock(&po->bind_lock);
4258         was_running = po->running;
4259         num = po->num;
4260         if (was_running) {
4261                 po->num = 0;
4262                 __unregister_prot_hook(sk, false);
4263         }
4264         spin_unlock(&po->bind_lock);
4265 
4266         synchronize_net();
4267 
4268         err = -EBUSY;
4269         mutex_lock(&po->pg_vec_lock);
4270         if (closing || atomic_read(&po->mapped) == 0) {
4271                 err = 0;
4272                 spin_lock_bh(&rb_queue->lock);
4273                 swap(rb->pg_vec, pg_vec);
4274                 rb->frame_max = (req->tp_frame_nr - 1);
4275                 rb->head = 0;
4276                 rb->frame_size = req->tp_frame_size;
4277                 spin_unlock_bh(&rb_queue->lock);
4278 
4279                 swap(rb->pg_vec_order, order);
4280                 swap(rb->pg_vec_len, req->tp_block_nr);
4281 
4282                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4283                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4284                                                 tpacket_rcv : packet_rcv;
4285                 skb_queue_purge(rb_queue);
4286                 if (atomic_read(&po->mapped))
4287                         pr_err("packet_mmap: vma is busy: %d\n",
4288                                atomic_read(&po->mapped));
4289         }
4290         mutex_unlock(&po->pg_vec_lock);
4291 
4292         spin_lock(&po->bind_lock);
4293         if (was_running) {
4294                 po->num = num;
4295                 register_prot_hook(sk);
4296         }
4297         spin_unlock(&po->bind_lock);
4298         if (closing && (po->tp_version > TPACKET_V2)) {
4299                 /* Because we don't support block-based V3 on tx-ring */
4300                 if (!tx_ring)
4301                         prb_shutdown_retire_blk_timer(po, rb_queue);
4302         }
4303 
4304         if (pg_vec)
4305                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4306 out:
4307         release_sock(sk);
4308         return err;
4309 }
4310 
4311 static int packet_mmap(struct file *file, struct socket *sock,
4312                 struct vm_area_struct *vma)
4313 {
4314         struct sock *sk = sock->sk;
4315         struct packet_sock *po = pkt_sk(sk);
4316         unsigned long size, expected_size;
4317         struct packet_ring_buffer *rb;
4318         unsigned long start;
4319         int err = -EINVAL;
4320         int i;
4321 
4322         if (vma->vm_pgoff)
4323                 return -EINVAL;
4324 
4325         mutex_lock(&po->pg_vec_lock);
4326 
4327         expected_size = 0;
4328         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4329                 if (rb->pg_vec) {
4330                         expected_size += rb->pg_vec_len
4331                                                 * rb->pg_vec_pages
4332                                                 * PAGE_SIZE;
4333                 }
4334         }
4335 
4336         if (expected_size == 0)
4337                 goto out;
4338 
4339         size = vma->vm_end - vma->vm_start;
4340         if (size != expected_size)
4341                 goto out;
4342 
4343         start = vma->vm_start;
4344         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4345                 if (rb->pg_vec == NULL)
4346                         continue;
4347 
4348                 for (i = 0; i < rb->pg_vec_len; i++) {
4349                         struct page *page;
4350                         void *kaddr = rb->pg_vec[i].buffer;
4351                         int pg_num;
4352 
4353                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4354                                 page = pgv_to_page(kaddr);
4355                                 err = vm_insert_page(vma, start, page);
4356                                 if (unlikely(err))
4357                                         goto out;
4358                                 start += PAGE_SIZE;
4359                                 kaddr += PAGE_SIZE;
4360                         }
4361                 }
4362         }
4363 
4364         atomic_inc(&po->mapped);
4365         vma->vm_ops = &packet_mmap_ops;
4366         err = 0;
4367 
4368 out:
4369         mutex_unlock(&po->pg_vec_lock);
4370         return err;
4371 }
4372 
4373 static const struct proto_ops packet_ops_spkt = {
4374         .family =       PF_PACKET,
4375         .owner =        THIS_MODULE,
4376         .release =      packet_release,
4377         .bind =         packet_bind_spkt,
4378         .connect =      sock_no_connect,
4379         .socketpair =   sock_no_socketpair,
4380         .accept =       sock_no_accept,
4381         .getname =      packet_getname_spkt,
4382         .poll =         datagram_poll,
4383         .ioctl =        packet_ioctl,
4384         .listen =       sock_no_listen,
4385         .shutdown =     sock_no_shutdown,
4386         .setsockopt =   sock_no_setsockopt,
4387         .getsockopt =   sock_no_getsockopt,
4388         .sendmsg =      packet_sendmsg_spkt,
4389         .recvmsg =      packet_recvmsg,
4390         .mmap =         sock_no_mmap,
4391         .sendpage =     sock_no_sendpage,
4392 };
4393 
4394 static const struct proto_ops packet_ops = {
4395         .family =       PF_PACKET,
4396         .owner =        THIS_MODULE,
4397         .release =      packet_release,
4398         .bind =         packet_bind,
4399         .connect =      sock_no_connect,
4400         .socketpair =   sock_no_socketpair,
4401         .accept =       sock_no_accept,
4402         .getname =      packet_getname,
4403         .poll =         packet_poll,
4404         .ioctl =        packet_ioctl,
4405         .listen =       sock_no_listen,
4406         .shutdown =     sock_no_shutdown,
4407         .setsockopt =   packet_setsockopt,
4408         .getsockopt =   packet_getsockopt,
4409 #ifdef CONFIG_COMPAT
4410         .compat_setsockopt = compat_packet_setsockopt,
4411 #endif
4412         .sendmsg =      packet_sendmsg,
4413         .recvmsg =      packet_recvmsg,
4414         .mmap =         packet_mmap,
4415         .sendpage =     sock_no_sendpage,
4416 };
4417 
4418 static const struct net_proto_family packet_family_ops = {
4419         .family =       PF_PACKET,
4420         .create =       packet_create,
4421         .owner  =       THIS_MODULE,
4422 };
4423 
4424 static struct notifier_block packet_netdev_notifier = {
4425         .notifier_call =        packet_notifier,
4426 };
4427 
4428 #ifdef CONFIG_PROC_FS
4429 
4430 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4431         __acquires(RCU)
4432 {
4433         struct net *net = seq_file_net(seq);
4434 
4435         rcu_read_lock();
4436         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4437 }
4438 
4439 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4440 {
4441         struct net *net = seq_file_net(seq);
4442         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4443 }
4444 
4445 static void packet_seq_stop(struct seq_file *seq, void *v)
4446         __releases(RCU)
4447 {
4448         rcu_read_unlock();
4449 }
4450 
4451 static int packet_seq_show(struct seq_file *seq, void *v)
4452 {
4453         if (v == SEQ_START_TOKEN)
4454                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4455         else {
4456                 struct sock *s = sk_entry(v);
4457                 const struct packet_sock *po = pkt_sk(s);
4458 
4459                 seq_printf(seq,
4460                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4461                            s,
4462                            atomic_read(&s->sk_refcnt),
4463                            s->sk_type,
4464                            ntohs(po->num),
4465                            po->ifindex,
4466                            po->running,
4467                            atomic_read(&s->sk_rmem_alloc),
4468                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4469                            sock_i_ino(s));
4470         }
4471 
4472         return 0;
4473 }
4474 
4475 static const struct seq_operations packet_seq_ops = {
4476         .start  = packet_seq_start,
4477         .next   = packet_seq_next,
4478         .stop   = packet_seq_stop,
4479         .show   = packet_seq_show,
4480 };
4481 
4482 static int packet_seq_open(struct inode *inode, struct file *file)
4483 {
4484         return seq_open_net(inode, file, &packet_seq_ops,
4485                             sizeof(struct seq_net_private));
4486 }
4487 
4488 static const struct file_operations packet_seq_fops = {
4489         .owner          = THIS_MODULE,
4490         .open           = packet_seq_open,
4491         .read           = seq_read,
4492         .llseek         = seq_lseek,
4493         .release        = seq_release_net,
4494 };
4495 
4496 #endif
4497 
4498 static int __net_init packet_net_init(struct net *net)
4499 {
4500         mutex_init(&net->packet.sklist_lock);
4501         INIT_HLIST_HEAD(&net->packet.sklist);
4502 
4503         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4504                 return -ENOMEM;
4505 
4506         return 0;
4507 }
4508 
4509 static void __net_exit packet_net_exit(struct net *net)
4510 {
4511         remove_proc_entry("packet", net->proc_net);
4512 }
4513 
4514 static struct pernet_operations packet_net_ops = {
4515         .init = packet_net_init,
4516         .exit = packet_net_exit,
4517 };
4518 
4519 
4520 static void __exit packet_exit(void)
4521 {
4522         unregister_netdevice_notifier(&packet_netdev_notifier);
4523         unregister_pernet_subsys(&packet_net_ops);
4524         sock_unregister(PF_PACKET);
4525         proto_unregister(&packet_proto);
4526 }
4527 
4528 static int __init packet_init(void)
4529 {
4530         int rc = proto_register(&packet_proto, 0);
4531 
4532         if (rc != 0)
4533                 goto out;
4534 
4535         sock_register(&packet_family_ops);
4536         register_pernet_subsys(&packet_net_ops);
4537         register_netdevice_notifier(&packet_netdev_notifier);
4538 out:
4539         return rc;
4540 }
4541 
4542 module_init(packet_init);
4543 module_exit(packet_exit);
4544 MODULE_LICENSE("GPL");
4545 MODULE_ALIAS_NETPROTO(PF_PACKET);
4546 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp