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

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

Version: ~ [ linux-5.17-rc1 ] ~ [ linux-5.16.2 ] ~ [ linux-5.15.16 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.93 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.173 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.225 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.262 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.297 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.299 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ 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 
 97 #include "internal.h"
 98 
 99 /*
100    Assumptions:
101    - if device has no dev->hard_header routine, it adds and removes ll header
102      inside itself. In this case ll header is invisible outside of device,
103      but higher levels still should reserve dev->hard_header_len.
104      Some devices are enough clever to reallocate skb, when header
105      will not fit to reserved space (tunnel), another ones are silly
106      (PPP).
107    - packet socket receives packets with pulled ll header,
108      so that SOCK_RAW should push it back.
109 
110 On receive:
111 -----------
112 
113 Incoming, dev->hard_header!=NULL
114    mac_header -> ll header
115    data       -> data
116 
117 Outgoing, dev->hard_header!=NULL
118    mac_header -> ll header
119    data       -> ll header
120 
121 Incoming, dev->hard_header==NULL
122    mac_header -> UNKNOWN position. It is very likely, that it points to ll
123                  header.  PPP makes it, that is wrong, because introduce
124                  assymetry between rx and tx paths.
125    data       -> data
126 
127 Outgoing, dev->hard_header==NULL
128    mac_header -> data. ll header is still not built!
129    data       -> data
130 
131 Resume
132   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
133 
134 
135 On transmit:
136 ------------
137 
138 dev->hard_header != NULL
139    mac_header -> ll header
140    data       -> ll header
141 
142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
143    mac_header -> data
144    data       -> data
145 
146    We should set nh.raw on output to correct posistion,
147    packet classifier depends on it.
148  */
149 
150 /* Private packet socket structures. */
151 
152 /* identical to struct packet_mreq except it has
153  * a longer address field.
154  */
155 struct packet_mreq_max {
156         int             mr_ifindex;
157         unsigned short  mr_type;
158         unsigned short  mr_alen;
159         unsigned char   mr_address[MAX_ADDR_LEN];
160 };
161 
162 union tpacket_uhdr {
163         struct tpacket_hdr  *h1;
164         struct tpacket2_hdr *h2;
165         struct tpacket3_hdr *h3;
166         void *raw;
167 };
168 
169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170                 int closing, int tx_ring);
171 
172 #define V3_ALIGNMENT    (8)
173 
174 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 
179 #define PGV_FROM_VMALLOC 1
180 
181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182 #define BLOCK_NUM_PKTS(x)       ((x)->hdr.bh1.num_pkts)
183 #define BLOCK_O2FP(x)           ((x)->hdr.bh1.offset_to_first_pkt)
184 #define BLOCK_LEN(x)            ((x)->hdr.bh1.blk_len)
185 #define BLOCK_SNUM(x)           ((x)->hdr.bh1.seq_num)
186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187 #define BLOCK_PRIV(x)           ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
188 
189 struct packet_sock;
190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192                        struct packet_type *pt, struct net_device *orig_dev);
193 
194 static void *packet_previous_frame(struct packet_sock *po,
195                 struct packet_ring_buffer *rb,
196                 int status);
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
199                         struct tpacket_block_desc *);
200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
201                         struct packet_sock *);
202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
203                 struct packet_sock *, unsigned int status);
204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
205 static void prb_open_block(struct tpacket_kbdq_core *,
206                 struct tpacket_block_desc *);
207 static void prb_retire_rx_blk_timer_expired(unsigned long);
208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
209 static void prb_init_blk_timer(struct packet_sock *,
210                 struct tpacket_kbdq_core *,
211                 void (*func) (unsigned long));
212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
214                 struct tpacket3_hdr *);
215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
216                 struct tpacket3_hdr *);
217 static void packet_flush_mclist(struct sock *sk);
218 
219 struct packet_skb_cb {
220         union {
221                 struct sockaddr_pkt pkt;
222                 union {
223                         /* Trick: alias skb original length with
224                          * ll.sll_family and ll.protocol in order
225                          * to save room.
226                          */
227                         unsigned int origlen;
228                         struct sockaddr_ll ll;
229                 };
230         } sa;
231 };
232 
233 #define vio_le() virtio_legacy_is_little_endian()
234 
235 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
236 
237 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
238 #define GET_PBLOCK_DESC(x, bid) \
239         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
241         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
242 #define GET_NEXT_PRB_BLK_NUM(x) \
243         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244         ((x)->kactive_blk_num+1) : 0)
245 
246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
248 
249 static int packet_direct_xmit(struct sk_buff *skb)
250 {
251         struct net_device *dev = skb->dev;
252         netdev_features_t features;
253         struct netdev_queue *txq;
254         int ret = NETDEV_TX_BUSY;
255 
256         if (unlikely(!netif_running(dev) ||
257                      !netif_carrier_ok(dev)))
258                 goto drop;
259 
260         features = netif_skb_features(skb);
261         if (skb_needs_linearize(skb, features) &&
262             __skb_linearize(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(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(fd);
1591         if (IS_ERR(new))
1592                 return PTR_ERR(new);
1593         if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1594                 bpf_prog_put(new);
1595                 return -EINVAL;
1596         }
1597 
1598         __fanout_set_data_bpf(po->fanout, new);
1599         return 0;
1600 }
1601 
1602 static int fanout_set_data(struct packet_sock *po, char __user *data,
1603                            unsigned int len)
1604 {
1605         switch (po->fanout->type) {
1606         case PACKET_FANOUT_CBPF:
1607                 return fanout_set_data_cbpf(po, data, len);
1608         case PACKET_FANOUT_EBPF:
1609                 return fanout_set_data_ebpf(po, data, len);
1610         default:
1611                 return -EINVAL;
1612         };
1613 }
1614 
1615 static void fanout_release_data(struct packet_fanout *f)
1616 {
1617         switch (f->type) {
1618         case PACKET_FANOUT_CBPF:
1619         case PACKET_FANOUT_EBPF:
1620                 __fanout_set_data_bpf(f, NULL);
1621         };
1622 }
1623 
1624 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1625 {
1626         struct packet_sock *po = pkt_sk(sk);
1627         struct packet_fanout *f, *match;
1628         u8 type = type_flags & 0xff;
1629         u8 flags = type_flags >> 8;
1630         int err;
1631 
1632         switch (type) {
1633         case PACKET_FANOUT_ROLLOVER:
1634                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1635                         return -EINVAL;
1636         case PACKET_FANOUT_HASH:
1637         case PACKET_FANOUT_LB:
1638         case PACKET_FANOUT_CPU:
1639         case PACKET_FANOUT_RND:
1640         case PACKET_FANOUT_QM:
1641         case PACKET_FANOUT_CBPF:
1642         case PACKET_FANOUT_EBPF:
1643                 break;
1644         default:
1645                 return -EINVAL;
1646         }
1647 
1648         if (!po->running)
1649                 return -EINVAL;
1650 
1651         if (po->fanout)
1652                 return -EALREADY;
1653 
1654         if (type == PACKET_FANOUT_ROLLOVER ||
1655             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1656                 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1657                 if (!po->rollover)
1658                         return -ENOMEM;
1659                 atomic_long_set(&po->rollover->num, 0);
1660                 atomic_long_set(&po->rollover->num_huge, 0);
1661                 atomic_long_set(&po->rollover->num_failed, 0);
1662         }
1663 
1664         mutex_lock(&fanout_mutex);
1665         match = NULL;
1666         list_for_each_entry(f, &fanout_list, list) {
1667                 if (f->id == id &&
1668                     read_pnet(&f->net) == sock_net(sk)) {
1669                         match = f;
1670                         break;
1671                 }
1672         }
1673         err = -EINVAL;
1674         if (match && match->flags != flags)
1675                 goto out;
1676         if (!match) {
1677                 err = -ENOMEM;
1678                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1679                 if (!match)
1680                         goto out;
1681                 write_pnet(&match->net, sock_net(sk));
1682                 match->id = id;
1683                 match->type = type;
1684                 match->flags = flags;
1685                 INIT_LIST_HEAD(&match->list);
1686                 spin_lock_init(&match->lock);
1687                 atomic_set(&match->sk_ref, 0);
1688                 fanout_init_data(match);
1689                 match->prot_hook.type = po->prot_hook.type;
1690                 match->prot_hook.dev = po->prot_hook.dev;
1691                 match->prot_hook.func = packet_rcv_fanout;
1692                 match->prot_hook.af_packet_priv = match;
1693                 match->prot_hook.id_match = match_fanout_group;
1694                 dev_add_pack(&match->prot_hook);
1695                 list_add(&match->list, &fanout_list);
1696         }
1697         err = -EINVAL;
1698         if (match->type == type &&
1699             match->prot_hook.type == po->prot_hook.type &&
1700             match->prot_hook.dev == po->prot_hook.dev) {
1701                 err = -ENOSPC;
1702                 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1703                         __dev_remove_pack(&po->prot_hook);
1704                         po->fanout = match;
1705                         atomic_inc(&match->sk_ref);
1706                         __fanout_link(sk, po);
1707                         err = 0;
1708                 }
1709         }
1710 out:
1711         mutex_unlock(&fanout_mutex);
1712         if (err) {
1713                 kfree(po->rollover);
1714                 po->rollover = NULL;
1715         }
1716         return err;
1717 }
1718 
1719 static void fanout_release(struct sock *sk)
1720 {
1721         struct packet_sock *po = pkt_sk(sk);
1722         struct packet_fanout *f;
1723 
1724         f = po->fanout;
1725         if (!f)
1726                 return;
1727 
1728         mutex_lock(&fanout_mutex);
1729         po->fanout = NULL;
1730 
1731         if (atomic_dec_and_test(&f->sk_ref)) {
1732                 list_del(&f->list);
1733                 dev_remove_pack(&f->prot_hook);
1734                 fanout_release_data(f);
1735                 kfree(f);
1736         }
1737         mutex_unlock(&fanout_mutex);
1738 
1739         if (po->rollover)
1740                 kfree_rcu(po->rollover, rcu);
1741 }
1742 
1743 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1744                                           struct sk_buff *skb)
1745 {
1746         /* Earlier code assumed this would be a VLAN pkt, double-check
1747          * this now that we have the actual packet in hand. We can only
1748          * do this check on Ethernet devices.
1749          */
1750         if (unlikely(dev->type != ARPHRD_ETHER))
1751                 return false;
1752 
1753         skb_reset_mac_header(skb);
1754         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1755 }
1756 
1757 static const struct proto_ops packet_ops;
1758 
1759 static const struct proto_ops packet_ops_spkt;
1760 
1761 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1762                            struct packet_type *pt, struct net_device *orig_dev)
1763 {
1764         struct sock *sk;
1765         struct sockaddr_pkt *spkt;
1766 
1767         /*
1768          *      When we registered the protocol we saved the socket in the data
1769          *      field for just this event.
1770          */
1771 
1772         sk = pt->af_packet_priv;
1773 
1774         /*
1775          *      Yank back the headers [hope the device set this
1776          *      right or kerboom...]
1777          *
1778          *      Incoming packets have ll header pulled,
1779          *      push it back.
1780          *
1781          *      For outgoing ones skb->data == skb_mac_header(skb)
1782          *      so that this procedure is noop.
1783          */
1784 
1785         if (skb->pkt_type == PACKET_LOOPBACK)
1786                 goto out;
1787 
1788         if (!net_eq(dev_net(dev), sock_net(sk)))
1789                 goto out;
1790 
1791         skb = skb_share_check(skb, GFP_ATOMIC);
1792         if (skb == NULL)
1793                 goto oom;
1794 
1795         /* drop any routing info */
1796         skb_dst_drop(skb);
1797 
1798         /* drop conntrack reference */
1799         nf_reset(skb);
1800 
1801         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1802 
1803         skb_push(skb, skb->data - skb_mac_header(skb));
1804 
1805         /*
1806          *      The SOCK_PACKET socket receives _all_ frames.
1807          */
1808 
1809         spkt->spkt_family = dev->type;
1810         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1811         spkt->spkt_protocol = skb->protocol;
1812 
1813         /*
1814          *      Charge the memory to the socket. This is done specifically
1815          *      to prevent sockets using all the memory up.
1816          */
1817 
1818         if (sock_queue_rcv_skb(sk, skb) == 0)
1819                 return 0;
1820 
1821 out:
1822         kfree_skb(skb);
1823 oom:
1824         return 0;
1825 }
1826 
1827 
1828 /*
1829  *      Output a raw packet to a device layer. This bypasses all the other
1830  *      protocol layers and you must therefore supply it with a complete frame
1831  */
1832 
1833 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1834                                size_t len)
1835 {
1836         struct sock *sk = sock->sk;
1837         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1838         struct sk_buff *skb = NULL;
1839         struct net_device *dev;
1840         __be16 proto = 0;
1841         int err;
1842         int extra_len = 0;
1843 
1844         /*
1845          *      Get and verify the address.
1846          */
1847 
1848         if (saddr) {
1849                 if (msg->msg_namelen < sizeof(struct sockaddr))
1850                         return -EINVAL;
1851                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1852                         proto = saddr->spkt_protocol;
1853         } else
1854                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1855 
1856         /*
1857          *      Find the device first to size check it
1858          */
1859 
1860         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1861 retry:
1862         rcu_read_lock();
1863         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1864         err = -ENODEV;
1865         if (dev == NULL)
1866                 goto out_unlock;
1867 
1868         err = -ENETDOWN;
1869         if (!(dev->flags & IFF_UP))
1870                 goto out_unlock;
1871 
1872         /*
1873          * You may not queue a frame bigger than the mtu. This is the lowest level
1874          * raw protocol and you must do your own fragmentation at this level.
1875          */
1876 
1877         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1878                 if (!netif_supports_nofcs(dev)) {
1879                         err = -EPROTONOSUPPORT;
1880                         goto out_unlock;
1881                 }
1882                 extra_len = 4; /* We're doing our own CRC */
1883         }
1884 
1885         err = -EMSGSIZE;
1886         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1887                 goto out_unlock;
1888 
1889         if (!skb) {
1890                 size_t reserved = LL_RESERVED_SPACE(dev);
1891                 int tlen = dev->needed_tailroom;
1892                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1893 
1894                 rcu_read_unlock();
1895                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1896                 if (skb == NULL)
1897                         return -ENOBUFS;
1898                 /* FIXME: Save some space for broken drivers that write a hard
1899                  * header at transmission time by themselves. PPP is the notable
1900                  * one here. This should really be fixed at the driver level.
1901                  */
1902                 skb_reserve(skb, reserved);
1903                 skb_reset_network_header(skb);
1904 
1905                 /* Try to align data part correctly */
1906                 if (hhlen) {
1907                         skb->data -= hhlen;
1908                         skb->tail -= hhlen;
1909                         if (len < hhlen)
1910                                 skb_reset_network_header(skb);
1911                 }
1912                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1913                 if (err)
1914                         goto out_free;
1915                 goto retry;
1916         }
1917 
1918         if (!dev_validate_header(dev, skb->data, len)) {
1919                 err = -EINVAL;
1920                 goto out_unlock;
1921         }
1922         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1923             !packet_extra_vlan_len_allowed(dev, skb)) {
1924                 err = -EMSGSIZE;
1925                 goto out_unlock;
1926         }
1927 
1928         skb->protocol = proto;
1929         skb->dev = dev;
1930         skb->priority = sk->sk_priority;
1931         skb->mark = sk->sk_mark;
1932 
1933         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1934 
1935         if (unlikely(extra_len == 4))
1936                 skb->no_fcs = 1;
1937 
1938         skb_probe_transport_header(skb, 0);
1939 
1940         dev_queue_xmit(skb);
1941         rcu_read_unlock();
1942         return len;
1943 
1944 out_unlock:
1945         rcu_read_unlock();
1946 out_free:
1947         kfree_skb(skb);
1948         return err;
1949 }
1950 
1951 static unsigned int run_filter(struct sk_buff *skb,
1952                                const struct sock *sk,
1953                                unsigned int res)
1954 {
1955         struct sk_filter *filter;
1956 
1957         rcu_read_lock();
1958         filter = rcu_dereference(sk->sk_filter);
1959         if (filter != NULL)
1960                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1961         rcu_read_unlock();
1962 
1963         return res;
1964 }
1965 
1966 static int __packet_rcv_vnet(const struct sk_buff *skb,
1967                              struct virtio_net_hdr *vnet_hdr)
1968 {
1969         *vnet_hdr = (const struct virtio_net_hdr) { 0 };
1970 
1971         if (skb_is_gso(skb)) {
1972                 struct skb_shared_info *sinfo = skb_shinfo(skb);
1973 
1974                 /* This is a hint as to how much should be linear. */
1975                 vnet_hdr->hdr_len =
1976                         __cpu_to_virtio16(vio_le(), skb_headlen(skb));
1977                 vnet_hdr->gso_size =
1978                         __cpu_to_virtio16(vio_le(), sinfo->gso_size);
1979 
1980                 if (sinfo->gso_type & SKB_GSO_TCPV4)
1981                         vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1982                 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1983                         vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1984                 else if (sinfo->gso_type & SKB_GSO_UDP)
1985                         vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
1986                 else if (sinfo->gso_type & SKB_GSO_FCOE)
1987                         return -EINVAL;
1988                 else
1989                         BUG();
1990 
1991                 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
1992                         vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1993         } else
1994                 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
1995 
1996         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1997                 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
1998                 vnet_hdr->csum_start = __cpu_to_virtio16(vio_le(),
1999                                   skb_checksum_start_offset(skb));
2000                 vnet_hdr->csum_offset = __cpu_to_virtio16(vio_le(),
2001                                                  skb->csum_offset);
2002         } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2003                 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
2004         } /* else everything is zero */
2005 
2006         return 0;
2007 }
2008 
2009 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2010                            size_t *len)
2011 {
2012         struct virtio_net_hdr vnet_hdr;
2013 
2014         if (*len < sizeof(vnet_hdr))
2015                 return -EINVAL;
2016         *len -= sizeof(vnet_hdr);
2017 
2018         if (__packet_rcv_vnet(skb, &vnet_hdr))
2019                 return -EINVAL;
2020 
2021         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2022 }
2023 
2024 /*
2025  * This function makes lazy skb cloning in hope that most of packets
2026  * are discarded by BPF.
2027  *
2028  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2029  * and skb->cb are mangled. It works because (and until) packets
2030  * falling here are owned by current CPU. Output packets are cloned
2031  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2032  * sequencially, so that if we return skb to original state on exit,
2033  * we will not harm anyone.
2034  */
2035 
2036 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2037                       struct packet_type *pt, struct net_device *orig_dev)
2038 {
2039         struct sock *sk;
2040         struct sockaddr_ll *sll;
2041         struct packet_sock *po;
2042         u8 *skb_head = skb->data;
2043         int skb_len = skb->len;
2044         unsigned int snaplen, res;
2045 
2046         if (skb->pkt_type == PACKET_LOOPBACK)
2047                 goto drop;
2048 
2049         sk = pt->af_packet_priv;
2050         po = pkt_sk(sk);
2051 
2052         if (!net_eq(dev_net(dev), sock_net(sk)))
2053                 goto drop;
2054 
2055         skb->dev = dev;
2056 
2057         if (dev->header_ops) {
2058                 /* The device has an explicit notion of ll header,
2059                  * exported to higher levels.
2060                  *
2061                  * Otherwise, the device hides details of its frame
2062                  * structure, so that corresponding packet head is
2063                  * never delivered to user.
2064                  */
2065                 if (sk->sk_type != SOCK_DGRAM)
2066                         skb_push(skb, skb->data - skb_mac_header(skb));
2067                 else if (skb->pkt_type == PACKET_OUTGOING) {
2068                         /* Special case: outgoing packets have ll header at head */
2069                         skb_pull(skb, skb_network_offset(skb));
2070                 }
2071         }
2072 
2073         snaplen = skb->len;
2074 
2075         res = run_filter(skb, sk, snaplen);
2076         if (!res)
2077                 goto drop_n_restore;
2078         if (snaplen > res)
2079                 snaplen = res;
2080 
2081         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2082                 goto drop_n_acct;
2083 
2084         if (skb_shared(skb)) {
2085                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2086                 if (nskb == NULL)
2087                         goto drop_n_acct;
2088 
2089                 if (skb_head != skb->data) {
2090                         skb->data = skb_head;
2091                         skb->len = skb_len;
2092                 }
2093                 consume_skb(skb);
2094                 skb = nskb;
2095         }
2096 
2097         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2098 
2099         sll = &PACKET_SKB_CB(skb)->sa.ll;
2100         sll->sll_hatype = dev->type;
2101         sll->sll_pkttype = skb->pkt_type;
2102         if (unlikely(po->origdev))
2103                 sll->sll_ifindex = orig_dev->ifindex;
2104         else
2105                 sll->sll_ifindex = dev->ifindex;
2106 
2107         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2108 
2109         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2110          * Use their space for storing the original skb length.
2111          */
2112         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2113 
2114         if (pskb_trim(skb, snaplen))
2115                 goto drop_n_acct;
2116 
2117         skb_set_owner_r(skb, sk);
2118         skb->dev = NULL;
2119         skb_dst_drop(skb);
2120 
2121         /* drop conntrack reference */
2122         nf_reset(skb);
2123 
2124         spin_lock(&sk->sk_receive_queue.lock);
2125         po->stats.stats1.tp_packets++;
2126         sock_skb_set_dropcount(sk, skb);
2127         __skb_queue_tail(&sk->sk_receive_queue, skb);
2128         spin_unlock(&sk->sk_receive_queue.lock);
2129         sk->sk_data_ready(sk);
2130         return 0;
2131 
2132 drop_n_acct:
2133         spin_lock(&sk->sk_receive_queue.lock);
2134         po->stats.stats1.tp_drops++;
2135         atomic_inc(&sk->sk_drops);
2136         spin_unlock(&sk->sk_receive_queue.lock);
2137 
2138 drop_n_restore:
2139         if (skb_head != skb->data && skb_shared(skb)) {
2140                 skb->data = skb_head;
2141                 skb->len = skb_len;
2142         }
2143 drop:
2144         consume_skb(skb);
2145         return 0;
2146 }
2147 
2148 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2149                        struct packet_type *pt, struct net_device *orig_dev)
2150 {
2151         struct sock *sk;
2152         struct packet_sock *po;
2153         struct sockaddr_ll *sll;
2154         union tpacket_uhdr h;
2155         u8 *skb_head = skb->data;
2156         int skb_len = skb->len;
2157         unsigned int snaplen, res;
2158         unsigned long status = TP_STATUS_USER;
2159         unsigned short macoff, netoff, hdrlen;
2160         struct sk_buff *copy_skb = NULL;
2161         struct timespec ts;
2162         __u32 ts_status;
2163 
2164         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2165          * We may add members to them until current aligned size without forcing
2166          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2167          */
2168         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2169         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2170 
2171         if (skb->pkt_type == PACKET_LOOPBACK)
2172                 goto drop;
2173 
2174         sk = pt->af_packet_priv;
2175         po = pkt_sk(sk);
2176 
2177         if (!net_eq(dev_net(dev), sock_net(sk)))
2178                 goto drop;
2179 
2180         if (dev->header_ops) {
2181                 if (sk->sk_type != SOCK_DGRAM)
2182                         skb_push(skb, skb->data - skb_mac_header(skb));
2183                 else if (skb->pkt_type == PACKET_OUTGOING) {
2184                         /* Special case: outgoing packets have ll header at head */
2185                         skb_pull(skb, skb_network_offset(skb));
2186                 }
2187         }
2188 
2189         snaplen = skb->len;
2190 
2191         res = run_filter(skb, sk, snaplen);
2192         if (!res)
2193                 goto drop_n_restore;
2194 
2195         if (skb->ip_summed == CHECKSUM_PARTIAL)
2196                 status |= TP_STATUS_CSUMNOTREADY;
2197         else if (skb->pkt_type != PACKET_OUTGOING &&
2198                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2199                   skb_csum_unnecessary(skb)))
2200                 status |= TP_STATUS_CSUM_VALID;
2201 
2202         if (snaplen > res)
2203                 snaplen = res;
2204 
2205         if (sk->sk_type == SOCK_DGRAM) {
2206                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2207                                   po->tp_reserve;
2208         } else {
2209                 unsigned int maclen = skb_network_offset(skb);
2210                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2211                                        (maclen < 16 ? 16 : maclen)) +
2212                                        po->tp_reserve;
2213                 if (po->has_vnet_hdr)
2214                         netoff += sizeof(struct virtio_net_hdr);
2215                 macoff = netoff - maclen;
2216         }
2217         if (po->tp_version <= TPACKET_V2) {
2218                 if (macoff + snaplen > po->rx_ring.frame_size) {
2219                         if (po->copy_thresh &&
2220                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2221                                 if (skb_shared(skb)) {
2222                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2223                                 } else {
2224                                         copy_skb = skb_get(skb);
2225                                         skb_head = skb->data;
2226                                 }
2227                                 if (copy_skb)
2228                                         skb_set_owner_r(copy_skb, sk);
2229                         }
2230                         snaplen = po->rx_ring.frame_size - macoff;
2231                         if ((int)snaplen < 0)
2232                                 snaplen = 0;
2233                 }
2234         } else if (unlikely(macoff + snaplen >
2235                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2236                 u32 nval;
2237 
2238                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2239                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2240                             snaplen, nval, macoff);
2241                 snaplen = nval;
2242                 if (unlikely((int)snaplen < 0)) {
2243                         snaplen = 0;
2244                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2245                 }
2246         }
2247         spin_lock(&sk->sk_receive_queue.lock);
2248         h.raw = packet_current_rx_frame(po, skb,
2249                                         TP_STATUS_KERNEL, (macoff+snaplen));
2250         if (!h.raw)
2251                 goto drop_n_account;
2252         if (po->tp_version <= TPACKET_V2) {
2253                 packet_increment_rx_head(po, &po->rx_ring);
2254         /*
2255          * LOSING will be reported till you read the stats,
2256          * because it's COR - Clear On Read.
2257          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2258          * at packet level.
2259          */
2260                 if (po->stats.stats1.tp_drops)
2261                         status |= TP_STATUS_LOSING;
2262         }
2263         po->stats.stats1.tp_packets++;
2264         if (copy_skb) {
2265                 status |= TP_STATUS_COPY;
2266                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2267         }
2268         spin_unlock(&sk->sk_receive_queue.lock);
2269 
2270         if (po->has_vnet_hdr) {
2271                 if (__packet_rcv_vnet(skb, h.raw + macoff -
2272                                            sizeof(struct virtio_net_hdr))) {
2273                         spin_lock(&sk->sk_receive_queue.lock);
2274                         goto drop_n_account;
2275                 }
2276         }
2277 
2278         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2279 
2280         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2281                 getnstimeofday(&ts);
2282 
2283         status |= ts_status;
2284 
2285         switch (po->tp_version) {
2286         case TPACKET_V1:
2287                 h.h1->tp_len = skb->len;
2288                 h.h1->tp_snaplen = snaplen;
2289                 h.h1->tp_mac = macoff;
2290                 h.h1->tp_net = netoff;
2291                 h.h1->tp_sec = ts.tv_sec;
2292                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2293                 hdrlen = sizeof(*h.h1);
2294                 break;
2295         case TPACKET_V2:
2296                 h.h2->tp_len = skb->len;
2297                 h.h2->tp_snaplen = snaplen;
2298                 h.h2->tp_mac = macoff;
2299                 h.h2->tp_net = netoff;
2300                 h.h2->tp_sec = ts.tv_sec;
2301                 h.h2->tp_nsec = ts.tv_nsec;
2302                 if (skb_vlan_tag_present(skb)) {
2303                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2304                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2305                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2306                 } else {
2307                         h.h2->tp_vlan_tci = 0;
2308                         h.h2->tp_vlan_tpid = 0;
2309                 }
2310                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2311                 hdrlen = sizeof(*h.h2);
2312                 break;
2313         case TPACKET_V3:
2314                 /* tp_nxt_offset,vlan are already populated above.
2315                  * So DONT clear those fields here
2316                  */
2317                 h.h3->tp_status |= status;
2318                 h.h3->tp_len = skb->len;
2319                 h.h3->tp_snaplen = snaplen;
2320                 h.h3->tp_mac = macoff;
2321                 h.h3->tp_net = netoff;
2322                 h.h3->tp_sec  = ts.tv_sec;
2323                 h.h3->tp_nsec = ts.tv_nsec;
2324                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2325                 hdrlen = sizeof(*h.h3);
2326                 break;
2327         default:
2328                 BUG();
2329         }
2330 
2331         sll = h.raw + TPACKET_ALIGN(hdrlen);
2332         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2333         sll->sll_family = AF_PACKET;
2334         sll->sll_hatype = dev->type;
2335         sll->sll_protocol = skb->protocol;
2336         sll->sll_pkttype = skb->pkt_type;
2337         if (unlikely(po->origdev))
2338                 sll->sll_ifindex = orig_dev->ifindex;
2339         else
2340                 sll->sll_ifindex = dev->ifindex;
2341 
2342         smp_mb();
2343 
2344 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2345         if (po->tp_version <= TPACKET_V2) {
2346                 u8 *start, *end;
2347 
2348                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2349                                         macoff + snaplen);
2350 
2351                 for (start = h.raw; start < end; start += PAGE_SIZE)
2352                         flush_dcache_page(pgv_to_page(start));
2353         }
2354         smp_wmb();
2355 #endif
2356 
2357         if (po->tp_version <= TPACKET_V2) {
2358                 __packet_set_status(po, h.raw, status);
2359                 sk->sk_data_ready(sk);
2360         } else {
2361                 prb_clear_blk_fill_status(&po->rx_ring);
2362         }
2363 
2364 drop_n_restore:
2365         if (skb_head != skb->data && skb_shared(skb)) {
2366                 skb->data = skb_head;
2367                 skb->len = skb_len;
2368         }
2369 drop:
2370         kfree_skb(skb);
2371         return 0;
2372 
2373 drop_n_account:
2374         po->stats.stats1.tp_drops++;
2375         spin_unlock(&sk->sk_receive_queue.lock);
2376 
2377         sk->sk_data_ready(sk);
2378         kfree_skb(copy_skb);
2379         goto drop_n_restore;
2380 }
2381 
2382 static void tpacket_destruct_skb(struct sk_buff *skb)
2383 {
2384         struct packet_sock *po = pkt_sk(skb->sk);
2385 
2386         if (likely(po->tx_ring.pg_vec)) {
2387                 void *ph;
2388                 __u32 ts;
2389 
2390                 ph = skb_shinfo(skb)->destructor_arg;
2391                 packet_dec_pending(&po->tx_ring);
2392 
2393                 ts = __packet_set_timestamp(po, ph, skb);
2394                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2395         }
2396 
2397         sock_wfree(skb);
2398 }
2399 
2400 static void tpacket_set_protocol(const struct net_device *dev,
2401                                  struct sk_buff *skb)
2402 {
2403         if (dev->type == ARPHRD_ETHER) {
2404                 skb_reset_mac_header(skb);
2405                 skb->protocol = eth_hdr(skb)->h_proto;
2406         }
2407 }
2408 
2409 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2410 {
2411         unsigned short gso_type = 0;
2412 
2413         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2414             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2415              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2416               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2417                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2418                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2419                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2420 
2421         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2422                 return -EINVAL;
2423 
2424         if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2425                 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2426                 case VIRTIO_NET_HDR_GSO_TCPV4:
2427                         gso_type = SKB_GSO_TCPV4;
2428                         break;
2429                 case VIRTIO_NET_HDR_GSO_TCPV6:
2430                         gso_type = SKB_GSO_TCPV6;
2431                         break;
2432                 case VIRTIO_NET_HDR_GSO_UDP:
2433                         gso_type = SKB_GSO_UDP;
2434                         break;
2435                 default:
2436                         return -EINVAL;
2437                 }
2438 
2439                 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
2440                         gso_type |= SKB_GSO_TCP_ECN;
2441 
2442                 if (vnet_hdr->gso_size == 0)
2443                         return -EINVAL;
2444         }
2445 
2446         vnet_hdr->gso_type = gso_type;  /* changes type, temporary storage */
2447         return 0;
2448 }
2449 
2450 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2451                                  struct virtio_net_hdr *vnet_hdr)
2452 {
2453         int n;
2454 
2455         if (*len < sizeof(*vnet_hdr))
2456                 return -EINVAL;
2457         *len -= sizeof(*vnet_hdr);
2458 
2459         n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter);
2460         if (n != sizeof(*vnet_hdr))
2461                 return -EFAULT;
2462 
2463         return __packet_snd_vnet_parse(vnet_hdr, *len);
2464 }
2465 
2466 static int packet_snd_vnet_gso(struct sk_buff *skb,
2467                                struct virtio_net_hdr *vnet_hdr)
2468 {
2469         if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2470                 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start);
2471                 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset);
2472 
2473                 if (!skb_partial_csum_set(skb, s, o))
2474                         return -EINVAL;
2475         }
2476 
2477         skb_shinfo(skb)->gso_size =
2478                 __virtio16_to_cpu(vio_le(), vnet_hdr->gso_size);
2479         skb_shinfo(skb)->gso_type = vnet_hdr->gso_type;
2480 
2481         /* Header must be checked, and gso_segs computed. */
2482         skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2483         skb_shinfo(skb)->gso_segs = 0;
2484         return 0;
2485 }
2486 
2487 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2488                 void *frame, struct net_device *dev, void *data, int tp_len,
2489                 __be16 proto, unsigned char *addr, int hlen, int copylen)
2490 {
2491         union tpacket_uhdr ph;
2492         int to_write, offset, len, nr_frags, len_max;
2493         struct socket *sock = po->sk.sk_socket;
2494         struct page *page;
2495         int err;
2496 
2497         ph.raw = frame;
2498 
2499         skb->protocol = proto;
2500         skb->dev = dev;
2501         skb->priority = po->sk.sk_priority;
2502         skb->mark = po->sk.sk_mark;
2503         sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2504         skb_shinfo(skb)->destructor_arg = ph.raw;
2505 
2506         skb_reserve(skb, hlen);
2507         skb_reset_network_header(skb);
2508 
2509         to_write = tp_len;
2510 
2511         if (sock->type == SOCK_DGRAM) {
2512                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2513                                 NULL, tp_len);
2514                 if (unlikely(err < 0))
2515                         return -EINVAL;
2516         } else if (copylen) {
2517                 int hdrlen = min_t(int, copylen, tp_len);
2518 
2519                 skb_push(skb, dev->hard_header_len);
2520                 skb_put(skb, copylen - dev->hard_header_len);
2521                 err = skb_store_bits(skb, 0, data, hdrlen);
2522                 if (unlikely(err))
2523                         return err;
2524                 if (!dev_validate_header(dev, skb->data, hdrlen))
2525                         return -EINVAL;
2526                 if (!skb->protocol)
2527                         tpacket_set_protocol(dev, skb);
2528 
2529                 data += hdrlen;
2530                 to_write -= hdrlen;
2531         }
2532 
2533         offset = offset_in_page(data);
2534         len_max = PAGE_SIZE - offset;
2535         len = ((to_write > len_max) ? len_max : to_write);
2536 
2537         skb->data_len = to_write;
2538         skb->len += to_write;
2539         skb->truesize += to_write;
2540         atomic_add(to_write, &po->sk.sk_wmem_alloc);
2541 
2542         while (likely(to_write)) {
2543                 nr_frags = skb_shinfo(skb)->nr_frags;
2544 
2545                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2546                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2547                                MAX_SKB_FRAGS);
2548                         return -EFAULT;
2549                 }
2550 
2551                 page = pgv_to_page(data);
2552                 data += len;
2553                 flush_dcache_page(page);
2554                 get_page(page);
2555                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2556                 to_write -= len;
2557                 offset = 0;
2558                 len_max = PAGE_SIZE;
2559                 len = ((to_write > len_max) ? len_max : to_write);
2560         }
2561 
2562         skb_probe_transport_header(skb, 0);
2563 
2564         return tp_len;
2565 }
2566 
2567 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2568                                 int size_max, void **data)
2569 {
2570         union tpacket_uhdr ph;
2571         int tp_len, off;
2572 
2573         ph.raw = frame;
2574 
2575         switch (po->tp_version) {
2576         case TPACKET_V2:
2577                 tp_len = ph.h2->tp_len;
2578                 break;
2579         default:
2580                 tp_len = ph.h1->tp_len;
2581                 break;
2582         }
2583         if (unlikely(tp_len > size_max)) {
2584                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2585                 return -EMSGSIZE;
2586         }
2587 
2588         if (unlikely(po->tp_tx_has_off)) {
2589                 int off_min, off_max;
2590 
2591                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2592                 off_max = po->tx_ring.frame_size - tp_len;
2593                 if (po->sk.sk_type == SOCK_DGRAM) {
2594                         switch (po->tp_version) {
2595                         case TPACKET_V2:
2596                                 off = ph.h2->tp_net;
2597                                 break;
2598                         default:
2599                                 off = ph.h1->tp_net;
2600                                 break;
2601                         }
2602                 } else {
2603                         switch (po->tp_version) {
2604                         case TPACKET_V2:
2605                                 off = ph.h2->tp_mac;
2606                                 break;
2607                         default:
2608                                 off = ph.h1->tp_mac;
2609                                 break;
2610                         }
2611                 }
2612                 if (unlikely((off < off_min) || (off_max < off)))
2613                         return -EINVAL;
2614         } else {
2615                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2616         }
2617 
2618         *data = frame + off;
2619         return tp_len;
2620 }
2621 
2622 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2623 {
2624         struct sk_buff *skb;
2625         struct net_device *dev;
2626         struct virtio_net_hdr *vnet_hdr = NULL;
2627         __be16 proto;
2628         int err, reserve = 0;
2629         void *ph;
2630         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2631         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2632         int tp_len, size_max;
2633         unsigned char *addr;
2634         void *data;
2635         int len_sum = 0;
2636         int status = TP_STATUS_AVAILABLE;
2637         int hlen, tlen, copylen = 0;
2638 
2639         mutex_lock(&po->pg_vec_lock);
2640 
2641         if (likely(saddr == NULL)) {
2642                 dev     = packet_cached_dev_get(po);
2643                 proto   = po->num;
2644                 addr    = NULL;
2645         } else {
2646                 err = -EINVAL;
2647                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2648                         goto out;
2649                 if (msg->msg_namelen < (saddr->sll_halen
2650                                         + offsetof(struct sockaddr_ll,
2651                                                 sll_addr)))
2652                         goto out;
2653                 proto   = saddr->sll_protocol;
2654                 addr    = saddr->sll_addr;
2655                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2656         }
2657 
2658         err = -ENXIO;
2659         if (unlikely(dev == NULL))
2660                 goto out;
2661         err = -ENETDOWN;
2662         if (unlikely(!(dev->flags & IFF_UP)))
2663                 goto out_put;
2664 
2665         if (po->sk.sk_socket->type == SOCK_RAW)
2666                 reserve = dev->hard_header_len;
2667         size_max = po->tx_ring.frame_size
2668                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2669 
2670         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2671                 size_max = dev->mtu + reserve + VLAN_HLEN;
2672 
2673         do {
2674                 ph = packet_current_frame(po, &po->tx_ring,
2675                                           TP_STATUS_SEND_REQUEST);
2676                 if (unlikely(ph == NULL)) {
2677                         if (need_wait && need_resched())
2678                                 schedule();
2679                         continue;
2680                 }
2681 
2682                 skb = NULL;
2683                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2684                 if (tp_len < 0)
2685                         goto tpacket_error;
2686 
2687                 status = TP_STATUS_SEND_REQUEST;
2688                 hlen = LL_RESERVED_SPACE(dev);
2689                 tlen = dev->needed_tailroom;
2690                 if (po->has_vnet_hdr) {
2691                         vnet_hdr = data;
2692                         data += sizeof(*vnet_hdr);
2693                         tp_len -= sizeof(*vnet_hdr);
2694                         if (tp_len < 0 ||
2695                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2696                                 tp_len = -EINVAL;
2697                                 goto tpacket_error;
2698                         }
2699                         copylen = __virtio16_to_cpu(vio_le(),
2700                                                     vnet_hdr->hdr_len);
2701                 }
2702                 copylen = max_t(int, copylen, dev->hard_header_len);
2703                 skb = sock_alloc_send_skb(&po->sk,
2704                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2705                                 (copylen - dev->hard_header_len),
2706                                 !need_wait, &err);
2707 
2708                 if (unlikely(skb == NULL)) {
2709                         /* we assume the socket was initially writeable ... */
2710                         if (likely(len_sum > 0))
2711                                 err = len_sum;
2712                         goto out_status;
2713                 }
2714                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2715                                           addr, hlen, copylen);
2716                 if (likely(tp_len >= 0) &&
2717                     tp_len > dev->mtu + reserve &&
2718                     !po->has_vnet_hdr &&
2719                     !packet_extra_vlan_len_allowed(dev, skb))
2720                         tp_len = -EMSGSIZE;
2721 
2722                 if (unlikely(tp_len < 0)) {
2723 tpacket_error:
2724                         if (po->tp_loss) {
2725                                 __packet_set_status(po, ph,
2726                                                 TP_STATUS_AVAILABLE);
2727                                 packet_increment_head(&po->tx_ring);
2728                                 kfree_skb(skb);
2729                                 continue;
2730                         } else {
2731                                 status = TP_STATUS_WRONG_FORMAT;
2732                                 err = tp_len;
2733                                 goto out_status;
2734                         }
2735                 }
2736 
2737                 if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) {
2738                         tp_len = -EINVAL;
2739                         goto tpacket_error;
2740                 }
2741 
2742                 packet_pick_tx_queue(dev, skb);
2743 
2744                 skb->destructor = tpacket_destruct_skb;
2745                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2746                 packet_inc_pending(&po->tx_ring);
2747 
2748                 status = TP_STATUS_SEND_REQUEST;
2749                 err = po->xmit(skb);
2750                 if (unlikely(err > 0)) {
2751                         err = net_xmit_errno(err);
2752                         if (err && __packet_get_status(po, ph) ==
2753                                    TP_STATUS_AVAILABLE) {
2754                                 /* skb was destructed already */
2755                                 skb = NULL;
2756                                 goto out_status;
2757                         }
2758                         /*
2759                          * skb was dropped but not destructed yet;
2760                          * let's treat it like congestion or err < 0
2761                          */
2762                         err = 0;
2763                 }
2764                 packet_increment_head(&po->tx_ring);
2765                 len_sum += tp_len;
2766         } while (likely((ph != NULL) ||
2767                 /* Note: packet_read_pending() might be slow if we have
2768                  * to call it as it's per_cpu variable, but in fast-path
2769                  * we already short-circuit the loop with the first
2770                  * condition, and luckily don't have to go that path
2771                  * anyway.
2772                  */
2773                  (need_wait && packet_read_pending(&po->tx_ring))));
2774 
2775         err = len_sum;
2776         goto out_put;
2777 
2778 out_status:
2779         __packet_set_status(po, ph, status);
2780         kfree_skb(skb);
2781 out_put:
2782         dev_put(dev);
2783 out:
2784         mutex_unlock(&po->pg_vec_lock);
2785         return err;
2786 }
2787 
2788 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2789                                         size_t reserve, size_t len,
2790                                         size_t linear, int noblock,
2791                                         int *err)
2792 {
2793         struct sk_buff *skb;
2794 
2795         /* Under a page?  Don't bother with paged skb. */
2796         if (prepad + len < PAGE_SIZE || !linear)
2797                 linear = len;
2798 
2799         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2800                                    err, 0);
2801         if (!skb)
2802                 return NULL;
2803 
2804         skb_reserve(skb, reserve);
2805         skb_put(skb, linear);
2806         skb->data_len = len - linear;
2807         skb->len += len - linear;
2808 
2809         return skb;
2810 }
2811 
2812 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2813 {
2814         struct sock *sk = sock->sk;
2815         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2816         struct sk_buff *skb;
2817         struct net_device *dev;
2818         __be16 proto;
2819         unsigned char *addr;
2820         int err, reserve = 0;
2821         struct sockcm_cookie sockc;
2822         struct virtio_net_hdr vnet_hdr = { 0 };
2823         int offset = 0;
2824         struct packet_sock *po = pkt_sk(sk);
2825         int hlen, tlen;
2826         int extra_len = 0;
2827 
2828         /*
2829          *      Get and verify the address.
2830          */
2831 
2832         if (likely(saddr == NULL)) {
2833                 dev     = packet_cached_dev_get(po);
2834                 proto   = po->num;
2835                 addr    = NULL;
2836         } else {
2837                 err = -EINVAL;
2838                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2839                         goto out;
2840                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2841                         goto out;
2842                 proto   = saddr->sll_protocol;
2843                 addr    = saddr->sll_addr;
2844                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2845         }
2846 
2847         err = -ENXIO;
2848         if (unlikely(dev == NULL))
2849                 goto out_unlock;
2850         err = -ENETDOWN;
2851         if (unlikely(!(dev->flags & IFF_UP)))
2852                 goto out_unlock;
2853 
2854         sockc.mark = sk->sk_mark;
2855         if (msg->msg_controllen) {
2856                 err = sock_cmsg_send(sk, msg, &sockc);
2857                 if (unlikely(err))
2858                         goto out_unlock;
2859         }
2860 
2861         if (sock->type == SOCK_RAW)
2862                 reserve = dev->hard_header_len;
2863         if (po->has_vnet_hdr) {
2864                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2865                 if (err)
2866                         goto out_unlock;
2867         }
2868 
2869         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2870                 if (!netif_supports_nofcs(dev)) {
2871                         err = -EPROTONOSUPPORT;
2872                         goto out_unlock;
2873                 }
2874                 extra_len = 4; /* We're doing our own CRC */
2875         }
2876 
2877         err = -EMSGSIZE;
2878         if (!vnet_hdr.gso_type &&
2879             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2880                 goto out_unlock;
2881 
2882         err = -ENOBUFS;
2883         hlen = LL_RESERVED_SPACE(dev);
2884         tlen = dev->needed_tailroom;
2885         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2886                                __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2887                                msg->msg_flags & MSG_DONTWAIT, &err);
2888         if (skb == NULL)
2889                 goto out_unlock;
2890 
2891         skb_set_network_header(skb, reserve);
2892 
2893         err = -EINVAL;
2894         if (sock->type == SOCK_DGRAM) {
2895                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2896                 if (unlikely(offset < 0))
2897                         goto out_free;
2898         }
2899 
2900         /* Returns -EFAULT on error */
2901         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2902         if (err)
2903                 goto out_free;
2904 
2905         if (sock->type == SOCK_RAW &&
2906             !dev_validate_header(dev, skb->data, len)) {
2907                 err = -EINVAL;
2908                 goto out_free;
2909         }
2910 
2911         sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2912 
2913         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2914             !packet_extra_vlan_len_allowed(dev, skb)) {
2915                 err = -EMSGSIZE;
2916                 goto out_free;
2917         }
2918 
2919         skb->protocol = proto;
2920         skb->dev = dev;
2921         skb->priority = sk->sk_priority;
2922         skb->mark = sockc.mark;
2923 
2924         packet_pick_tx_queue(dev, skb);
2925 
2926         if (po->has_vnet_hdr) {
2927                 err = packet_snd_vnet_gso(skb, &vnet_hdr);
2928                 if (err)
2929                         goto out_free;
2930                 len += sizeof(vnet_hdr);
2931         }
2932 
2933         skb_probe_transport_header(skb, reserve);
2934 
2935         if (unlikely(extra_len == 4))
2936                 skb->no_fcs = 1;
2937 
2938         err = po->xmit(skb);
2939         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2940                 goto out_unlock;
2941 
2942         dev_put(dev);
2943 
2944         return len;
2945 
2946 out_free:
2947         kfree_skb(skb);
2948 out_unlock:
2949         if (dev)
2950                 dev_put(dev);
2951 out:
2952         return err;
2953 }
2954 
2955 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2956 {
2957         struct sock *sk = sock->sk;
2958         struct packet_sock *po = pkt_sk(sk);
2959 
2960         if (po->tx_ring.pg_vec)
2961                 return tpacket_snd(po, msg);
2962         else
2963                 return packet_snd(sock, msg, len);
2964 }
2965 
2966 /*
2967  *      Close a PACKET socket. This is fairly simple. We immediately go
2968  *      to 'closed' state and remove our protocol entry in the device list.
2969  */
2970 
2971 static int packet_release(struct socket *sock)
2972 {
2973         struct sock *sk = sock->sk;
2974         struct packet_sock *po;
2975         struct net *net;
2976         union tpacket_req_u req_u;
2977 
2978         if (!sk)
2979                 return 0;
2980 
2981         net = sock_net(sk);
2982         po = pkt_sk(sk);
2983 
2984         mutex_lock(&net->packet.sklist_lock);
2985         sk_del_node_init_rcu(sk);
2986         mutex_unlock(&net->packet.sklist_lock);
2987 
2988         preempt_disable();
2989         sock_prot_inuse_add(net, sk->sk_prot, -1);
2990         preempt_enable();
2991 
2992         spin_lock(&po->bind_lock);
2993         unregister_prot_hook(sk, false);
2994         packet_cached_dev_reset(po);
2995 
2996         if (po->prot_hook.dev) {
2997                 dev_put(po->prot_hook.dev);
2998                 po->prot_hook.dev = NULL;
2999         }
3000         spin_unlock(&po->bind_lock);
3001 
3002         packet_flush_mclist(sk);
3003 
3004         if (po->rx_ring.pg_vec) {
3005                 memset(&req_u, 0, sizeof(req_u));
3006                 packet_set_ring(sk, &req_u, 1, 0);
3007         }
3008 
3009         if (po->tx_ring.pg_vec) {
3010                 memset(&req_u, 0, sizeof(req_u));
3011                 packet_set_ring(sk, &req_u, 1, 1);
3012         }
3013 
3014         fanout_release(sk);
3015 
3016         synchronize_net();
3017         /*
3018          *      Now the socket is dead. No more input will appear.
3019          */
3020         sock_orphan(sk);
3021         sock->sk = NULL;
3022 
3023         /* Purge queues */
3024 
3025         skb_queue_purge(&sk->sk_receive_queue);
3026         packet_free_pending(po);
3027         sk_refcnt_debug_release(sk);
3028 
3029         sock_put(sk);
3030         return 0;
3031 }
3032 
3033 /*
3034  *      Attach a packet hook.
3035  */
3036 
3037 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3038                           __be16 proto)
3039 {
3040         struct packet_sock *po = pkt_sk(sk);
3041         struct net_device *dev_curr;
3042         __be16 proto_curr;
3043         bool need_rehook;
3044         struct net_device *dev = NULL;
3045         int ret = 0;
3046         bool unlisted = false;
3047 
3048         if (po->fanout)
3049                 return -EINVAL;
3050 
3051         lock_sock(sk);
3052         spin_lock(&po->bind_lock);
3053         rcu_read_lock();
3054 
3055         if (name) {
3056                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3057                 if (!dev) {
3058                         ret = -ENODEV;
3059                         goto out_unlock;
3060                 }
3061         } else if (ifindex) {
3062                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3063                 if (!dev) {
3064                         ret = -ENODEV;
3065                         goto out_unlock;
3066                 }
3067         }
3068 
3069         if (dev)
3070                 dev_hold(dev);
3071 
3072         proto_curr = po->prot_hook.type;
3073         dev_curr = po->prot_hook.dev;
3074 
3075         need_rehook = proto_curr != proto || dev_curr != dev;
3076 
3077         if (need_rehook) {
3078                 if (po->running) {
3079                         rcu_read_unlock();
3080                         __unregister_prot_hook(sk, true);
3081                         rcu_read_lock();
3082                         dev_curr = po->prot_hook.dev;
3083                         if (dev)
3084                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3085                                                                  dev->ifindex);
3086                 }
3087 
3088                 po->num = proto;
3089                 po->prot_hook.type = proto;
3090 
3091                 if (unlikely(unlisted)) {
3092                         dev_put(dev);
3093                         po->prot_hook.dev = NULL;
3094                         po->ifindex = -1;
3095                         packet_cached_dev_reset(po);
3096                 } else {
3097                         po->prot_hook.dev = dev;
3098                         po->ifindex = dev ? dev->ifindex : 0;
3099                         packet_cached_dev_assign(po, dev);
3100                 }
3101         }
3102         if (dev_curr)
3103                 dev_put(dev_curr);
3104 
3105         if (proto == 0 || !need_rehook)
3106                 goto out_unlock;
3107 
3108         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3109                 register_prot_hook(sk);
3110         } else {
3111                 sk->sk_err = ENETDOWN;
3112                 if (!sock_flag(sk, SOCK_DEAD))
3113                         sk->sk_error_report(sk);
3114         }
3115 
3116 out_unlock:
3117         rcu_read_unlock();
3118         spin_unlock(&po->bind_lock);
3119         release_sock(sk);
3120         return ret;
3121 }
3122 
3123 /*
3124  *      Bind a packet socket to a device
3125  */
3126 
3127 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3128                             int addr_len)
3129 {
3130         struct sock *sk = sock->sk;
3131         char name[15];
3132 
3133         /*
3134          *      Check legality
3135          */
3136 
3137         if (addr_len != sizeof(struct sockaddr))
3138                 return -EINVAL;
3139         strlcpy(name, uaddr->sa_data, sizeof(name));
3140 
3141         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3142 }
3143 
3144 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3145 {
3146         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3147         struct sock *sk = sock->sk;
3148 
3149         /*
3150          *      Check legality
3151          */
3152 
3153         if (addr_len < sizeof(struct sockaddr_ll))
3154                 return -EINVAL;
3155         if (sll->sll_family != AF_PACKET)
3156                 return -EINVAL;
3157 
3158         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3159                               sll->sll_protocol ? : pkt_sk(sk)->num);
3160 }
3161 
3162 static struct proto packet_proto = {
3163         .name     = "PACKET",
3164         .owner    = THIS_MODULE,
3165         .obj_size = sizeof(struct packet_sock),
3166 };
3167 
3168 /*
3169  *      Create a packet of type SOCK_PACKET.
3170  */
3171 
3172 static int packet_create(struct net *net, struct socket *sock, int protocol,
3173                          int kern)
3174 {
3175         struct sock *sk;
3176         struct packet_sock *po;
3177         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3178         int err;
3179 
3180         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3181                 return -EPERM;
3182         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3183             sock->type != SOCK_PACKET)
3184                 return -ESOCKTNOSUPPORT;
3185 
3186         sock->state = SS_UNCONNECTED;
3187 
3188         err = -ENOBUFS;
3189         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3190         if (sk == NULL)
3191                 goto out;
3192 
3193         sock->ops = &packet_ops;
3194         if (sock->type == SOCK_PACKET)
3195                 sock->ops = &packet_ops_spkt;
3196 
3197         sock_init_data(sock, sk);
3198 
3199         po = pkt_sk(sk);
3200         sk->sk_family = PF_PACKET;
3201         po->num = proto;
3202         po->xmit = dev_queue_xmit;
3203 
3204         err = packet_alloc_pending(po);
3205         if (err)
3206                 goto out2;
3207 
3208         packet_cached_dev_reset(po);
3209 
3210         sk->sk_destruct = packet_sock_destruct;
3211         sk_refcnt_debug_inc(sk);
3212 
3213         /*
3214          *      Attach a protocol block
3215          */
3216 
3217         spin_lock_init(&po->bind_lock);
3218         mutex_init(&po->pg_vec_lock);
3219         po->rollover = NULL;
3220         po->prot_hook.func = packet_rcv;
3221 
3222         if (sock->type == SOCK_PACKET)
3223                 po->prot_hook.func = packet_rcv_spkt;
3224 
3225         po->prot_hook.af_packet_priv = sk;
3226 
3227         if (proto) {
3228                 po->prot_hook.type = proto;
3229                 register_prot_hook(sk);
3230         }
3231 
3232         mutex_lock(&net->packet.sklist_lock);
3233         sk_add_node_rcu(sk, &net->packet.sklist);
3234         mutex_unlock(&net->packet.sklist_lock);
3235 
3236         preempt_disable();
3237         sock_prot_inuse_add(net, &packet_proto, 1);
3238         preempt_enable();
3239 
3240         return 0;
3241 out2:
3242         sk_free(sk);
3243 out:
3244         return err;
3245 }
3246 
3247 /*
3248  *      Pull a packet from our receive queue and hand it to the user.
3249  *      If necessary we block.
3250  */
3251 
3252 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3253                           int flags)
3254 {
3255         struct sock *sk = sock->sk;
3256         struct sk_buff *skb;
3257         int copied, err;
3258         int vnet_hdr_len = 0;
3259         unsigned int origlen = 0;
3260 
3261         err = -EINVAL;
3262         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3263                 goto out;
3264 
3265 #if 0
3266         /* What error should we return now? EUNATTACH? */
3267         if (pkt_sk(sk)->ifindex < 0)
3268                 return -ENODEV;
3269 #endif
3270 
3271         if (flags & MSG_ERRQUEUE) {
3272                 err = sock_recv_errqueue(sk, msg, len,
3273                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3274                 goto out;
3275         }
3276 
3277         /*
3278          *      Call the generic datagram receiver. This handles all sorts
3279          *      of horrible races and re-entrancy so we can forget about it
3280          *      in the protocol layers.
3281          *
3282          *      Now it will return ENETDOWN, if device have just gone down,
3283          *      but then it will block.
3284          */
3285 
3286         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3287 
3288         /*
3289          *      An error occurred so return it. Because skb_recv_datagram()
3290          *      handles the blocking we don't see and worry about blocking
3291          *      retries.
3292          */
3293 
3294         if (skb == NULL)
3295                 goto out;
3296 
3297         if (pkt_sk(sk)->pressure)
3298                 packet_rcv_has_room(pkt_sk(sk), NULL);
3299 
3300         if (pkt_sk(sk)->has_vnet_hdr) {
3301                 err = packet_rcv_vnet(msg, skb, &len);
3302                 if (err)
3303                         goto out_free;
3304                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3305         }
3306 
3307         /* You lose any data beyond the buffer you gave. If it worries
3308          * a user program they can ask the device for its MTU
3309          * anyway.
3310          */
3311         copied = skb->len;
3312         if (copied > len) {
3313                 copied = len;
3314                 msg->msg_flags |= MSG_TRUNC;
3315         }
3316 
3317         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3318         if (err)
3319                 goto out_free;
3320 
3321         if (sock->type != SOCK_PACKET) {
3322                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3323 
3324                 /* Original length was stored in sockaddr_ll fields */
3325                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3326                 sll->sll_family = AF_PACKET;
3327                 sll->sll_protocol = skb->protocol;
3328         }
3329 
3330         sock_recv_ts_and_drops(msg, sk, skb);
3331 
3332         if (msg->msg_name) {
3333                 /* If the address length field is there to be filled
3334                  * in, we fill it in now.
3335                  */
3336                 if (sock->type == SOCK_PACKET) {
3337                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3338                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3339                 } else {
3340                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3341 
3342                         msg->msg_namelen = sll->sll_halen +
3343                                 offsetof(struct sockaddr_ll, sll_addr);
3344                 }
3345                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3346                        msg->msg_namelen);
3347         }
3348 
3349         if (pkt_sk(sk)->auxdata) {
3350                 struct tpacket_auxdata aux;
3351 
3352                 aux.tp_status = TP_STATUS_USER;
3353                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3354                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3355                 else if (skb->pkt_type != PACKET_OUTGOING &&
3356                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3357                           skb_csum_unnecessary(skb)))
3358                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3359 
3360                 aux.tp_len = origlen;
3361                 aux.tp_snaplen = skb->len;
3362                 aux.tp_mac = 0;
3363                 aux.tp_net = skb_network_offset(skb);
3364                 if (skb_vlan_tag_present(skb)) {
3365                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3366                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3367                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3368                 } else {
3369                         aux.tp_vlan_tci = 0;
3370                         aux.tp_vlan_tpid = 0;
3371                 }
3372                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3373         }
3374 
3375         /*
3376          *      Free or return the buffer as appropriate. Again this
3377          *      hides all the races and re-entrancy issues from us.
3378          */
3379         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3380 
3381 out_free:
3382         skb_free_datagram(sk, skb);
3383 out:
3384         return err;
3385 }
3386 
3387 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3388                                int *uaddr_len, int peer)
3389 {
3390         struct net_device *dev;
3391         struct sock *sk = sock->sk;
3392 
3393         if (peer)
3394                 return -EOPNOTSUPP;
3395 
3396         uaddr->sa_family = AF_PACKET;
3397         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3398         rcu_read_lock();
3399         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3400         if (dev)
3401                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3402         rcu_read_unlock();
3403         *uaddr_len = sizeof(*uaddr);
3404 
3405         return 0;
3406 }
3407 
3408 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3409                           int *uaddr_len, int peer)
3410 {
3411         struct net_device *dev;
3412         struct sock *sk = sock->sk;
3413         struct packet_sock *po = pkt_sk(sk);
3414         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3415 
3416         if (peer)
3417                 return -EOPNOTSUPP;
3418 
3419         sll->sll_family = AF_PACKET;
3420         sll->sll_ifindex = po->ifindex;
3421         sll->sll_protocol = po->num;
3422         sll->sll_pkttype = 0;
3423         rcu_read_lock();
3424         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3425         if (dev) {
3426                 sll->sll_hatype = dev->type;
3427                 sll->sll_halen = dev->addr_len;
3428                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3429         } else {
3430                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3431                 sll->sll_halen = 0;
3432         }
3433         rcu_read_unlock();
3434         *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3435 
3436         return 0;
3437 }
3438 
3439 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3440                          int what)
3441 {
3442         switch (i->type) {
3443         case PACKET_MR_MULTICAST:
3444                 if (i->alen != dev->addr_len)
3445                         return -EINVAL;
3446                 if (what > 0)
3447                         return dev_mc_add(dev, i->addr);
3448                 else
3449                         return dev_mc_del(dev, i->addr);
3450                 break;
3451         case PACKET_MR_PROMISC:
3452                 return dev_set_promiscuity(dev, what);
3453         case PACKET_MR_ALLMULTI:
3454                 return dev_set_allmulti(dev, what);
3455         case PACKET_MR_UNICAST:
3456                 if (i->alen != dev->addr_len)
3457                         return -EINVAL;
3458                 if (what > 0)
3459                         return dev_uc_add(dev, i->addr);
3460                 else
3461                         return dev_uc_del(dev, i->addr);
3462                 break;
3463         default:
3464                 break;
3465         }
3466         return 0;
3467 }
3468 
3469 static void packet_dev_mclist_delete(struct net_device *dev,
3470                                      struct packet_mclist **mlp)
3471 {
3472         struct packet_mclist *ml;
3473 
3474         while ((ml = *mlp) != NULL) {
3475                 if (ml->ifindex == dev->ifindex) {
3476                         packet_dev_mc(dev, ml, -1);
3477                         *mlp = ml->next;
3478                         kfree(ml);
3479                 } else
3480                         mlp = &ml->next;
3481         }
3482 }
3483 
3484 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3485 {
3486         struct packet_sock *po = pkt_sk(sk);
3487         struct packet_mclist *ml, *i;
3488         struct net_device *dev;
3489         int err;
3490 
3491         rtnl_lock();
3492 
3493         err = -ENODEV;
3494         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3495         if (!dev)
3496                 goto done;
3497 
3498         err = -EINVAL;
3499         if (mreq->mr_alen > dev->addr_len)
3500                 goto done;
3501 
3502         err = -ENOBUFS;
3503         i = kmalloc(sizeof(*i), GFP_KERNEL);
3504         if (i == NULL)
3505                 goto done;
3506 
3507         err = 0;
3508         for (ml = po->mclist; ml; ml = ml->next) {
3509                 if (ml->ifindex == mreq->mr_ifindex &&
3510                     ml->type == mreq->mr_type &&
3511                     ml->alen == mreq->mr_alen &&
3512                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3513                         ml->count++;
3514                         /* Free the new element ... */
3515                         kfree(i);
3516                         goto done;
3517                 }
3518         }
3519 
3520         i->type = mreq->mr_type;
3521         i->ifindex = mreq->mr_ifindex;
3522         i->alen = mreq->mr_alen;
3523         memcpy(i->addr, mreq->mr_address, i->alen);
3524         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3525         i->count = 1;
3526         i->next = po->mclist;
3527         po->mclist = i;
3528         err = packet_dev_mc(dev, i, 1);
3529         if (err) {
3530                 po->mclist = i->next;
3531                 kfree(i);
3532         }
3533 
3534 done:
3535         rtnl_unlock();
3536         return err;
3537 }
3538 
3539 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3540 {
3541         struct packet_mclist *ml, **mlp;
3542 
3543         rtnl_lock();
3544 
3545         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3546                 if (ml->ifindex == mreq->mr_ifindex &&
3547                     ml->type == mreq->mr_type &&
3548                     ml->alen == mreq->mr_alen &&
3549                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3550                         if (--ml->count == 0) {
3551                                 struct net_device *dev;
3552                                 *mlp = ml->next;
3553                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3554                                 if (dev)
3555                                         packet_dev_mc(dev, ml, -1);
3556                                 kfree(ml);
3557                         }
3558                         break;
3559                 }
3560         }
3561         rtnl_unlock();
3562         return 0;
3563 }
3564 
3565 static void packet_flush_mclist(struct sock *sk)
3566 {
3567         struct packet_sock *po = pkt_sk(sk);
3568         struct packet_mclist *ml;
3569 
3570         if (!po->mclist)
3571                 return;
3572 
3573         rtnl_lock();
3574         while ((ml = po->mclist) != NULL) {
3575                 struct net_device *dev;
3576 
3577                 po->mclist = ml->next;
3578                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3579                 if (dev != NULL)
3580                         packet_dev_mc(dev, ml, -1);
3581                 kfree(ml);
3582         }
3583         rtnl_unlock();
3584 }
3585 
3586 static int
3587 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3588 {
3589         struct sock *sk = sock->sk;
3590         struct packet_sock *po = pkt_sk(sk);
3591         int ret;
3592 
3593         if (level != SOL_PACKET)
3594                 return -ENOPROTOOPT;
3595 
3596         switch (optname) {
3597         case PACKET_ADD_MEMBERSHIP:
3598         case PACKET_DROP_MEMBERSHIP:
3599         {
3600                 struct packet_mreq_max mreq;
3601                 int len = optlen;
3602                 memset(&mreq, 0, sizeof(mreq));
3603                 if (len < sizeof(struct packet_mreq))
3604                         return -EINVAL;
3605                 if (len > sizeof(mreq))
3606                         len = sizeof(mreq);
3607                 if (copy_from_user(&mreq, optval, len))
3608                         return -EFAULT;
3609                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3610                         return -EINVAL;
3611                 if (optname == PACKET_ADD_MEMBERSHIP)
3612                         ret = packet_mc_add(sk, &mreq);
3613                 else
3614                         ret = packet_mc_drop(sk, &mreq);
3615                 return ret;
3616         }
3617 
3618         case PACKET_RX_RING:
3619         case PACKET_TX_RING:
3620         {
3621                 union tpacket_req_u req_u;
3622                 int len;
3623 
3624                 switch (po->tp_version) {
3625                 case TPACKET_V1:
3626                 case TPACKET_V2:
3627                         len = sizeof(req_u.req);
3628                         break;
3629                 case TPACKET_V3:
3630                 default:
3631                         len = sizeof(req_u.req3);
3632                         break;
3633                 }
3634                 if (optlen < len)
3635                         return -EINVAL;
3636                 if (copy_from_user(&req_u.req, optval, len))
3637                         return -EFAULT;
3638                 return packet_set_ring(sk, &req_u, 0,
3639                         optname == PACKET_TX_RING);
3640         }
3641         case PACKET_COPY_THRESH:
3642         {
3643                 int val;
3644 
3645                 if (optlen != sizeof(val))
3646                         return -EINVAL;
3647                 if (copy_from_user(&val, optval, sizeof(val)))
3648                         return -EFAULT;
3649 
3650                 pkt_sk(sk)->copy_thresh = val;
3651                 return 0;
3652         }
3653         case PACKET_VERSION:
3654         {
3655                 int val;
3656 
3657                 if (optlen != sizeof(val))
3658                         return -EINVAL;
3659                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3660                         return -EBUSY;
3661                 if (copy_from_user(&val, optval, sizeof(val)))
3662                         return -EFAULT;
3663                 switch (val) {
3664                 case TPACKET_V1:
3665                 case TPACKET_V2:
3666                 case TPACKET_V3:
3667                         po->tp_version = val;
3668                         return 0;
3669                 default:
3670                         return -EINVAL;
3671                 }
3672         }
3673         case PACKET_RESERVE:
3674         {
3675                 unsigned int val;
3676 
3677                 if (optlen != sizeof(val))
3678                         return -EINVAL;
3679                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3680                         return -EBUSY;
3681                 if (copy_from_user(&val, optval, sizeof(val)))
3682                         return -EFAULT;
3683                 po->tp_reserve = val;
3684                 return 0;
3685         }
3686         case PACKET_LOSS:
3687         {
3688                 unsigned int val;
3689 
3690                 if (optlen != sizeof(val))
3691                         return -EINVAL;
3692                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3693                         return -EBUSY;
3694                 if (copy_from_user(&val, optval, sizeof(val)))
3695                         return -EFAULT;
3696                 po->tp_loss = !!val;
3697                 return 0;
3698         }
3699         case PACKET_AUXDATA:
3700         {
3701                 int val;
3702 
3703                 if (optlen < sizeof(val))
3704                         return -EINVAL;
3705                 if (copy_from_user(&val, optval, sizeof(val)))
3706                         return -EFAULT;
3707 
3708                 po->auxdata = !!val;
3709                 return 0;
3710         }
3711         case PACKET_ORIGDEV:
3712         {
3713                 int val;
3714 
3715                 if (optlen < sizeof(val))
3716                         return -EINVAL;
3717                 if (copy_from_user(&val, optval, sizeof(val)))
3718                         return -EFAULT;
3719 
3720                 po->origdev = !!val;
3721                 return 0;
3722         }
3723         case PACKET_VNET_HDR:
3724         {
3725                 int val;
3726 
3727                 if (sock->type != SOCK_RAW)
3728                         return -EINVAL;
3729                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3730                         return -EBUSY;
3731                 if (optlen < sizeof(val))
3732                         return -EINVAL;
3733                 if (copy_from_user(&val, optval, sizeof(val)))
3734                         return -EFAULT;
3735 
3736                 po->has_vnet_hdr = !!val;
3737                 return 0;
3738         }
3739         case PACKET_TIMESTAMP:
3740         {
3741                 int val;
3742 
3743                 if (optlen != sizeof(val))
3744                         return -EINVAL;
3745                 if (copy_from_user(&val, optval, sizeof(val)))
3746                         return -EFAULT;
3747 
3748                 po->tp_tstamp = val;
3749                 return 0;
3750         }
3751         case PACKET_FANOUT:
3752         {
3753                 int val;
3754 
3755                 if (optlen != sizeof(val))
3756                         return -EINVAL;
3757                 if (copy_from_user(&val, optval, sizeof(val)))
3758                         return -EFAULT;
3759 
3760                 return fanout_add(sk, val & 0xffff, val >> 16);
3761         }
3762         case PACKET_FANOUT_DATA:
3763         {
3764                 if (!po->fanout)
3765                         return -EINVAL;
3766 
3767                 return fanout_set_data(po, optval, optlen);
3768         }
3769         case PACKET_TX_HAS_OFF:
3770         {
3771                 unsigned int val;
3772 
3773                 if (optlen != sizeof(val))
3774                         return -EINVAL;
3775                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3776                         return -EBUSY;
3777                 if (copy_from_user(&val, optval, sizeof(val)))
3778                         return -EFAULT;
3779                 po->tp_tx_has_off = !!val;
3780                 return 0;
3781         }
3782         case PACKET_QDISC_BYPASS:
3783         {
3784                 int val;
3785 
3786                 if (optlen != sizeof(val))
3787                         return -EINVAL;
3788                 if (copy_from_user(&val, optval, sizeof(val)))
3789                         return -EFAULT;
3790 
3791                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3792                 return 0;
3793         }
3794         default:
3795                 return -ENOPROTOOPT;
3796         }
3797 }
3798 
3799 static int packet_getsockopt(struct socket *sock, int level, int optname,
3800                              char __user *optval, int __user *optlen)
3801 {
3802         int len;
3803         int val, lv = sizeof(val);
3804         struct sock *sk = sock->sk;
3805         struct packet_sock *po = pkt_sk(sk);
3806         void *data = &val;
3807         union tpacket_stats_u st;
3808         struct tpacket_rollover_stats rstats;
3809 
3810         if (level != SOL_PACKET)
3811                 return -ENOPROTOOPT;
3812 
3813         if (get_user(len, optlen))
3814                 return -EFAULT;
3815 
3816         if (len < 0)
3817                 return -EINVAL;
3818 
3819         switch (optname) {
3820         case PACKET_STATISTICS:
3821                 spin_lock_bh(&sk->sk_receive_queue.lock);
3822                 memcpy(&st, &po->stats, sizeof(st));
3823                 memset(&po->stats, 0, sizeof(po->stats));
3824                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3825 
3826                 if (po->tp_version == TPACKET_V3) {
3827                         lv = sizeof(struct tpacket_stats_v3);
3828                         st.stats3.tp_packets += st.stats3.tp_drops;
3829                         data = &st.stats3;
3830                 } else {
3831                         lv = sizeof(struct tpacket_stats);
3832                         st.stats1.tp_packets += st.stats1.tp_drops;
3833                         data = &st.stats1;
3834                 }
3835 
3836                 break;
3837         case PACKET_AUXDATA:
3838                 val = po->auxdata;
3839                 break;
3840         case PACKET_ORIGDEV:
3841                 val = po->origdev;
3842                 break;
3843         case PACKET_VNET_HDR:
3844                 val = po->has_vnet_hdr;
3845                 break;
3846         case PACKET_VERSION:
3847                 val = po->tp_version;
3848                 break;
3849         case PACKET_HDRLEN:
3850                 if (len > sizeof(int))
3851                         len = sizeof(int);
3852                 if (copy_from_user(&val, optval, len))
3853                         return -EFAULT;
3854                 switch (val) {
3855                 case TPACKET_V1:
3856                         val = sizeof(struct tpacket_hdr);
3857                         break;
3858                 case TPACKET_V2:
3859                         val = sizeof(struct tpacket2_hdr);
3860                         break;
3861                 case TPACKET_V3:
3862                         val = sizeof(struct tpacket3_hdr);
3863                         break;
3864                 default:
3865                         return -EINVAL;
3866                 }
3867                 break;
3868         case PACKET_RESERVE:
3869                 val = po->tp_reserve;
3870                 break;
3871         case PACKET_LOSS:
3872                 val = po->tp_loss;
3873                 break;
3874         case PACKET_TIMESTAMP:
3875                 val = po->tp_tstamp;
3876                 break;
3877         case PACKET_FANOUT:
3878                 val = (po->fanout ?
3879                        ((u32)po->fanout->id |
3880                         ((u32)po->fanout->type << 16) |
3881                         ((u32)po->fanout->flags << 24)) :
3882                        0);
3883                 break;
3884         case PACKET_ROLLOVER_STATS:
3885                 if (!po->rollover)
3886                         return -EINVAL;
3887                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3888                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3889                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3890                 data = &rstats;
3891                 lv = sizeof(rstats);
3892                 break;
3893         case PACKET_TX_HAS_OFF:
3894                 val = po->tp_tx_has_off;
3895                 break;
3896         case PACKET_QDISC_BYPASS:
3897                 val = packet_use_direct_xmit(po);
3898                 break;
3899         default:
3900                 return -ENOPROTOOPT;
3901         }
3902 
3903         if (len > lv)
3904                 len = lv;
3905         if (put_user(len, optlen))
3906                 return -EFAULT;
3907         if (copy_to_user(optval, data, len))
3908                 return -EFAULT;
3909         return 0;
3910 }
3911 
3912 
3913 static int packet_notifier(struct notifier_block *this,
3914                            unsigned long msg, void *ptr)
3915 {
3916         struct sock *sk;
3917         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3918         struct net *net = dev_net(dev);
3919 
3920         rcu_read_lock();
3921         sk_for_each_rcu(sk, &net->packet.sklist) {
3922                 struct packet_sock *po = pkt_sk(sk);
3923 
3924                 switch (msg) {
3925                 case NETDEV_UNREGISTER:
3926                         if (po->mclist)
3927                                 packet_dev_mclist_delete(dev, &po->mclist);
3928                         /* fallthrough */
3929 
3930                 case NETDEV_DOWN:
3931                         if (dev->ifindex == po->ifindex) {
3932                                 spin_lock(&po->bind_lock);
3933                                 if (po->running) {
3934                                         __unregister_prot_hook(sk, false);
3935                                         sk->sk_err = ENETDOWN;
3936                                         if (!sock_flag(sk, SOCK_DEAD))
3937                                                 sk->sk_error_report(sk);
3938                                 }
3939                                 if (msg == NETDEV_UNREGISTER) {
3940                                         packet_cached_dev_reset(po);
3941                                         po->ifindex = -1;
3942                                         if (po->prot_hook.dev)
3943                                                 dev_put(po->prot_hook.dev);
3944                                         po->prot_hook.dev = NULL;
3945                                 }
3946                                 spin_unlock(&po->bind_lock);
3947                         }
3948                         break;
3949                 case NETDEV_UP:
3950                         if (dev->ifindex == po->ifindex) {
3951                                 spin_lock(&po->bind_lock);
3952                                 if (po->num)
3953                                         register_prot_hook(sk);
3954                                 spin_unlock(&po->bind_lock);
3955                         }
3956                         break;
3957                 }
3958         }
3959         rcu_read_unlock();
3960         return NOTIFY_DONE;
3961 }
3962 
3963 
3964 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3965                         unsigned long arg)
3966 {
3967         struct sock *sk = sock->sk;
3968 
3969         switch (cmd) {
3970         case SIOCOUTQ:
3971         {
3972                 int amount = sk_wmem_alloc_get(sk);
3973 
3974                 return put_user(amount, (int __user *)arg);
3975         }
3976         case SIOCINQ:
3977         {
3978                 struct sk_buff *skb;
3979                 int amount = 0;
3980 
3981                 spin_lock_bh(&sk->sk_receive_queue.lock);
3982                 skb = skb_peek(&sk->sk_receive_queue);
3983                 if (skb)
3984                         amount = skb->len;
3985                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3986                 return put_user(amount, (int __user *)arg);
3987         }
3988         case SIOCGSTAMP:
3989                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3990         case SIOCGSTAMPNS:
3991                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3992 
3993 #ifdef CONFIG_INET
3994         case SIOCADDRT:
3995         case SIOCDELRT:
3996         case SIOCDARP:
3997         case SIOCGARP:
3998         case SIOCSARP:
3999         case SIOCGIFADDR:
4000         case SIOCSIFADDR:
4001         case SIOCGIFBRDADDR:
4002         case SIOCSIFBRDADDR:
4003         case SIOCGIFNETMASK:
4004         case SIOCSIFNETMASK:
4005         case SIOCGIFDSTADDR:
4006         case SIOCSIFDSTADDR:
4007         case SIOCSIFFLAGS:
4008                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4009 #endif
4010 
4011         default:
4012                 return -ENOIOCTLCMD;
4013         }
4014         return 0;
4015 }
4016 
4017 static unsigned int packet_poll(struct file *file, struct socket *sock,
4018                                 poll_table *wait)
4019 {
4020         struct sock *sk = sock->sk;
4021         struct packet_sock *po = pkt_sk(sk);
4022         unsigned int mask = datagram_poll(file, sock, wait);
4023 
4024         spin_lock_bh(&sk->sk_receive_queue.lock);
4025         if (po->rx_ring.pg_vec) {
4026                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4027                         TP_STATUS_KERNEL))
4028                         mask |= POLLIN | POLLRDNORM;
4029         }
4030         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4031                 po->pressure = 0;
4032         spin_unlock_bh(&sk->sk_receive_queue.lock);
4033         spin_lock_bh(&sk->sk_write_queue.lock);
4034         if (po->tx_ring.pg_vec) {
4035                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4036                         mask |= POLLOUT | POLLWRNORM;
4037         }
4038         spin_unlock_bh(&sk->sk_write_queue.lock);
4039         return mask;
4040 }
4041 
4042 
4043 /* Dirty? Well, I still did not learn better way to account
4044  * for user mmaps.
4045  */
4046 
4047 static void packet_mm_open(struct vm_area_struct *vma)
4048 {
4049         struct file *file = vma->vm_file;
4050         struct socket *sock = file->private_data;
4051         struct sock *sk = sock->sk;
4052 
4053         if (sk)
4054                 atomic_inc(&pkt_sk(sk)->mapped);
4055 }
4056 
4057 static void packet_mm_close(struct vm_area_struct *vma)
4058 {
4059         struct file *file = vma->vm_file;
4060         struct socket *sock = file->private_data;
4061         struct sock *sk = sock->sk;
4062 
4063         if (sk)
4064                 atomic_dec(&pkt_sk(sk)->mapped);
4065 }
4066 
4067 static const struct vm_operations_struct packet_mmap_ops = {
4068         .open   =       packet_mm_open,
4069         .close  =       packet_mm_close,
4070 };
4071 
4072 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4073                         unsigned int len)
4074 {
4075         int i;
4076 
4077         for (i = 0; i < len; i++) {
4078                 if (likely(pg_vec[i].buffer)) {
4079                         if (is_vmalloc_addr(pg_vec[i].buffer))
4080                                 vfree(pg_vec[i].buffer);
4081                         else
4082                                 free_pages((unsigned long)pg_vec[i].buffer,
4083                                            order);
4084                         pg_vec[i].buffer = NULL;
4085                 }
4086         }
4087         kfree(pg_vec);
4088 }
4089 
4090 static char *alloc_one_pg_vec_page(unsigned long order)
4091 {
4092         char *buffer;
4093         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4094                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4095 
4096         buffer = (char *) __get_free_pages(gfp_flags, order);
4097         if (buffer)
4098                 return buffer;
4099 
4100         /* __get_free_pages failed, fall back to vmalloc */
4101         buffer = vzalloc((1 << order) * PAGE_SIZE);
4102         if (buffer)
4103                 return buffer;
4104 
4105         /* vmalloc failed, lets dig into swap here */
4106         gfp_flags &= ~__GFP_NORETRY;
4107         buffer = (char *) __get_free_pages(gfp_flags, order);
4108         if (buffer)
4109                 return buffer;
4110 
4111         /* complete and utter failure */
4112         return NULL;
4113 }
4114 
4115 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4116 {
4117         unsigned int block_nr = req->tp_block_nr;
4118         struct pgv *pg_vec;
4119         int i;
4120 
4121         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4122         if (unlikely(!pg_vec))
4123                 goto out;
4124 
4125         for (i = 0; i < block_nr; i++) {
4126                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4127                 if (unlikely(!pg_vec[i].buffer))
4128                         goto out_free_pgvec;
4129         }
4130 
4131 out:
4132         return pg_vec;
4133 
4134 out_free_pgvec:
4135         free_pg_vec(pg_vec, order, block_nr);
4136         pg_vec = NULL;
4137         goto out;
4138 }
4139 
4140 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4141                 int closing, int tx_ring)
4142 {
4143         struct pgv *pg_vec = NULL;
4144         struct packet_sock *po = pkt_sk(sk);
4145         int was_running, order = 0;
4146         struct packet_ring_buffer *rb;
4147         struct sk_buff_head *rb_queue;
4148         __be16 num;
4149         int err = -EINVAL;
4150         /* Added to avoid minimal code churn */
4151         struct tpacket_req *req = &req_u->req;
4152 
4153         /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4154         if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4155                 net_warn_ratelimited("Tx-ring is not supported.\n");
4156                 goto out;
4157         }
4158 
4159         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4160         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4161 
4162         err = -EBUSY;
4163         if (!closing) {
4164                 if (atomic_read(&po->mapped))
4165                         goto out;
4166                 if (packet_read_pending(rb))
4167                         goto out;
4168         }
4169 
4170         if (req->tp_block_nr) {
4171                 /* Sanity tests and some calculations */
4172                 err = -EBUSY;
4173                 if (unlikely(rb->pg_vec))
4174                         goto out;
4175 
4176                 switch (po->tp_version) {
4177                 case TPACKET_V1:
4178                         po->tp_hdrlen = TPACKET_HDRLEN;
4179                         break;
4180                 case TPACKET_V2:
4181                         po->tp_hdrlen = TPACKET2_HDRLEN;
4182                         break;
4183                 case TPACKET_V3:
4184                         po->tp_hdrlen = TPACKET3_HDRLEN;
4185                         break;
4186                 }
4187 
4188                 err = -EINVAL;
4189                 if (unlikely((int)req->tp_block_size <= 0))
4190                         goto out;
4191                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4192                         goto out;
4193                 if (po->tp_version >= TPACKET_V3 &&
4194                     (int)(req->tp_block_size -
4195                           BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4196                         goto out;
4197                 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4198                                         po->tp_reserve))
4199                         goto out;
4200                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4201                         goto out;
4202 
4203                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4204                 if (unlikely(rb->frames_per_block == 0))
4205                         goto out;
4206                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4207                                         req->tp_frame_nr))
4208                         goto out;
4209 
4210                 err = -ENOMEM;
4211                 order = get_order(req->tp_block_size);
4212                 pg_vec = alloc_pg_vec(req, order);
4213                 if (unlikely(!pg_vec))
4214                         goto out;
4215                 switch (po->tp_version) {
4216                 case TPACKET_V3:
4217                 /* Transmit path is not supported. We checked
4218                  * it above but just being paranoid
4219                  */
4220                         if (!tx_ring)
4221                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4222                         break;
4223                 default:
4224                         break;
4225                 }
4226         }
4227         /* Done */
4228         else {
4229                 err = -EINVAL;
4230                 if (unlikely(req->tp_frame_nr))
4231                         goto out;
4232         }
4233 
4234         lock_sock(sk);
4235 
4236         /* Detach socket from network */
4237         spin_lock(&po->bind_lock);
4238         was_running = po->running;
4239         num = po->num;
4240         if (was_running) {
4241                 po->num = 0;
4242                 __unregister_prot_hook(sk, false);
4243         }
4244         spin_unlock(&po->bind_lock);
4245 
4246         synchronize_net();
4247 
4248         err = -EBUSY;
4249         mutex_lock(&po->pg_vec_lock);
4250         if (closing || atomic_read(&po->mapped) == 0) {
4251                 err = 0;
4252                 spin_lock_bh(&rb_queue->lock);
4253                 swap(rb->pg_vec, pg_vec);
4254                 rb->frame_max = (req->tp_frame_nr - 1);
4255                 rb->head = 0;
4256                 rb->frame_size = req->tp_frame_size;
4257                 spin_unlock_bh(&rb_queue->lock);
4258 
4259                 swap(rb->pg_vec_order, order);
4260                 swap(rb->pg_vec_len, req->tp_block_nr);
4261 
4262                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4263                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4264                                                 tpacket_rcv : packet_rcv;
4265                 skb_queue_purge(rb_queue);
4266                 if (atomic_read(&po->mapped))
4267                         pr_err("packet_mmap: vma is busy: %d\n",
4268                                atomic_read(&po->mapped));
4269         }
4270         mutex_unlock(&po->pg_vec_lock);
4271 
4272         spin_lock(&po->bind_lock);
4273         if (was_running) {
4274                 po->num = num;
4275                 register_prot_hook(sk);
4276         }
4277         spin_unlock(&po->bind_lock);
4278         if (closing && (po->tp_version > TPACKET_V2)) {
4279                 /* Because we don't support block-based V3 on tx-ring */
4280                 if (!tx_ring)
4281                         prb_shutdown_retire_blk_timer(po, rb_queue);
4282         }
4283         release_sock(sk);
4284 
4285         if (pg_vec)
4286                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4287 out:
4288         return err;
4289 }
4290 
4291 static int packet_mmap(struct file *file, struct socket *sock,
4292                 struct vm_area_struct *vma)
4293 {
4294         struct sock *sk = sock->sk;
4295         struct packet_sock *po = pkt_sk(sk);
4296         unsigned long size, expected_size;
4297         struct packet_ring_buffer *rb;
4298         unsigned long start;
4299         int err = -EINVAL;
4300         int i;
4301 
4302         if (vma->vm_pgoff)
4303                 return -EINVAL;
4304 
4305         mutex_lock(&po->pg_vec_lock);
4306 
4307         expected_size = 0;
4308         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4309                 if (rb->pg_vec) {
4310                         expected_size += rb->pg_vec_len
4311                                                 * rb->pg_vec_pages
4312                                                 * PAGE_SIZE;
4313                 }
4314         }
4315 
4316         if (expected_size == 0)
4317                 goto out;
4318 
4319         size = vma->vm_end - vma->vm_start;
4320         if (size != expected_size)
4321                 goto out;
4322 
4323         start = vma->vm_start;
4324         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4325                 if (rb->pg_vec == NULL)
4326                         continue;
4327 
4328                 for (i = 0; i < rb->pg_vec_len; i++) {
4329                         struct page *page;
4330                         void *kaddr = rb->pg_vec[i].buffer;
4331                         int pg_num;
4332 
4333                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4334                                 page = pgv_to_page(kaddr);
4335                                 err = vm_insert_page(vma, start, page);
4336                                 if (unlikely(err))
4337                                         goto out;
4338                                 start += PAGE_SIZE;
4339                                 kaddr += PAGE_SIZE;
4340                         }
4341                 }
4342         }
4343 
4344         atomic_inc(&po->mapped);
4345         vma->vm_ops = &packet_mmap_ops;
4346         err = 0;
4347 
4348 out:
4349         mutex_unlock(&po->pg_vec_lock);
4350         return err;
4351 }
4352 
4353 static const struct proto_ops packet_ops_spkt = {
4354         .family =       PF_PACKET,
4355         .owner =        THIS_MODULE,
4356         .release =      packet_release,
4357         .bind =         packet_bind_spkt,
4358         .connect =      sock_no_connect,
4359         .socketpair =   sock_no_socketpair,
4360         .accept =       sock_no_accept,
4361         .getname =      packet_getname_spkt,
4362         .poll =         datagram_poll,
4363         .ioctl =        packet_ioctl,
4364         .listen =       sock_no_listen,
4365         .shutdown =     sock_no_shutdown,
4366         .setsockopt =   sock_no_setsockopt,
4367         .getsockopt =   sock_no_getsockopt,
4368         .sendmsg =      packet_sendmsg_spkt,
4369         .recvmsg =      packet_recvmsg,
4370         .mmap =         sock_no_mmap,
4371         .sendpage =     sock_no_sendpage,
4372 };
4373 
4374 static const struct proto_ops packet_ops = {
4375         .family =       PF_PACKET,
4376         .owner =        THIS_MODULE,
4377         .release =      packet_release,
4378         .bind =         packet_bind,
4379         .connect =      sock_no_connect,
4380         .socketpair =   sock_no_socketpair,
4381         .accept =       sock_no_accept,
4382         .getname =      packet_getname,
4383         .poll =         packet_poll,
4384         .ioctl =        packet_ioctl,
4385         .listen =       sock_no_listen,
4386         .shutdown =     sock_no_shutdown,
4387         .setsockopt =   packet_setsockopt,
4388         .getsockopt =   packet_getsockopt,
4389         .sendmsg =      packet_sendmsg,
4390         .recvmsg =      packet_recvmsg,
4391         .mmap =         packet_mmap,
4392         .sendpage =     sock_no_sendpage,
4393 };
4394 
4395 static const struct net_proto_family packet_family_ops = {
4396         .family =       PF_PACKET,
4397         .create =       packet_create,
4398         .owner  =       THIS_MODULE,
4399 };
4400 
4401 static struct notifier_block packet_netdev_notifier = {
4402         .notifier_call =        packet_notifier,
4403 };
4404 
4405 #ifdef CONFIG_PROC_FS
4406 
4407 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4408         __acquires(RCU)
4409 {
4410         struct net *net = seq_file_net(seq);
4411 
4412         rcu_read_lock();
4413         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4414 }
4415 
4416 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4417 {
4418         struct net *net = seq_file_net(seq);
4419         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4420 }
4421 
4422 static void packet_seq_stop(struct seq_file *seq, void *v)
4423         __releases(RCU)
4424 {
4425         rcu_read_unlock();
4426 }
4427 
4428 static int packet_seq_show(struct seq_file *seq, void *v)
4429 {
4430         if (v == SEQ_START_TOKEN)
4431                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4432         else {
4433                 struct sock *s = sk_entry(v);
4434                 const struct packet_sock *po = pkt_sk(s);
4435 
4436                 seq_printf(seq,
4437                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4438                            s,
4439                            atomic_read(&s->sk_refcnt),
4440                            s->sk_type,
4441                            ntohs(po->num),
4442                            po->ifindex,
4443                            po->running,
4444                            atomic_read(&s->sk_rmem_alloc),
4445                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4446                            sock_i_ino(s));
4447         }
4448 
4449         return 0;
4450 }
4451 
4452 static const struct seq_operations packet_seq_ops = {
4453         .start  = packet_seq_start,
4454         .next   = packet_seq_next,
4455         .stop   = packet_seq_stop,
4456         .show   = packet_seq_show,
4457 };
4458 
4459 static int packet_seq_open(struct inode *inode, struct file *file)
4460 {
4461         return seq_open_net(inode, file, &packet_seq_ops,
4462                             sizeof(struct seq_net_private));
4463 }
4464 
4465 static const struct file_operations packet_seq_fops = {
4466         .owner          = THIS_MODULE,
4467         .open           = packet_seq_open,
4468         .read           = seq_read,
4469         .llseek         = seq_lseek,
4470         .release        = seq_release_net,
4471 };
4472 
4473 #endif
4474 
4475 static int __net_init packet_net_init(struct net *net)
4476 {
4477         mutex_init(&net->packet.sklist_lock);
4478         INIT_HLIST_HEAD(&net->packet.sklist);
4479 
4480         if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4481                 return -ENOMEM;
4482 
4483         return 0;
4484 }
4485 
4486 static void __net_exit packet_net_exit(struct net *net)
4487 {
4488         remove_proc_entry("packet", net->proc_net);
4489 }
4490 
4491 static struct pernet_operations packet_net_ops = {
4492         .init = packet_net_init,
4493         .exit = packet_net_exit,
4494 };
4495 
4496 
4497 static void __exit packet_exit(void)
4498 {
4499         unregister_netdevice_notifier(&packet_netdev_notifier);
4500         unregister_pernet_subsys(&packet_net_ops);
4501         sock_unregister(PF_PACKET);
4502         proto_unregister(&packet_proto);
4503 }
4504 
4505 static int __init packet_init(void)
4506 {
4507         int rc = proto_register(&packet_proto, 0);
4508 
4509         if (rc != 0)
4510                 goto out;
4511 
4512         sock_register(&packet_family_ops);
4513         register_pernet_subsys(&packet_net_ops);
4514         register_netdevice_notifier(&packet_netdev_notifier);
4515 out:
4516         return rc;
4517 }
4518 
4519 module_init(packet_init);
4520 module_exit(packet_exit);
4521 MODULE_LICENSE("GPL");
4522 MODULE_ALIAS_NETPROTO(PF_PACKET);
4523 

~ [ 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