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

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

Version: ~ [ linux-5.8-rc5 ] ~ [ linux-5.7.8 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.51 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.132 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.188 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.230 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.230 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ 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