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

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

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

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

~ [ 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