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

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

Version: ~ [ linux-5.8-rc3 ] ~ [ linux-5.7.5 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.48 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.129 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.185 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.228 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.228 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
  3  *              operating system.  INET is implemented using the  BSD Socket
  4  *              interface as the means of communication with the user level.
  5  *
  6  *              PACKET - implements raw packet sockets.
  7  *
  8  * Authors:     Ross Biro
  9  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
 10  *              Alan Cox, <gw4pts@gw4pts.ampr.org>
 11  *
 12  * Fixes:
 13  *              Alan Cox        :       verify_area() now used correctly
 14  *              Alan Cox        :       new skbuff lists, look ma no backlogs!
 15  *              Alan Cox        :       tidied skbuff lists.
 16  *              Alan Cox        :       Now uses generic datagram routines I
 17  *                                      added. Also fixed the peek/read crash
 18  *                                      from all old Linux datagram code.
 19  *              Alan Cox        :       Uses the improved datagram code.
 20  *              Alan Cox        :       Added NULL's for socket options.
 21  *              Alan Cox        :       Re-commented the code.
 22  *              Alan Cox        :       Use new kernel side addressing
 23  *              Rob Janssen     :       Correct MTU usage.
 24  *              Dave Platt      :       Counter leaks caused by incorrect
 25  *                                      interrupt locking and some slightly
 26  *                                      dubious gcc output. Can you read
 27  *                                      compiler: it said _VOLATILE_
 28  *      Richard Kooijman        :       Timestamp fixes.
 29  *              Alan Cox        :       New buffers. Use sk->mac.raw.
 30  *              Alan Cox        :       sendmsg/recvmsg support.
 31  *              Alan Cox        :       Protocol setting support
 32  *      Alexey Kuznetsov        :       Untied from IPv4 stack.
 33  *      Cyrus Durgin            :       Fixed kerneld for kmod.
 34  *      Michal Ostrowski        :       Module initialization cleanup.
 35  *         Ulises Alonso        :       Frame number limit removal and
 36  *                                      packet_set_ring memory leak.
 37  *              Eric Biederman  :       Allow for > 8 byte hardware addresses.
 38  *                                      The convention is that longer addresses
 39  *                                      will simply extend the hardware address
 40  *                                      byte arrays at the end of sockaddr_ll
 41  *                                      and packet_mreq.
 42  *              Johann Baudy    :       Added TX RING.
 43  *              Chetan Loke     :       Implemented TPACKET_V3 block abstraction
 44  *                                      layer.
 45  *                                      Copyright (C) 2011, <lokec@ccs.neu.edu>
 46  *
 47  *
 48  *              This program is free software; you can redistribute it and/or
 49  *              modify it under the terms of the GNU General Public License
 50  *              as published by the Free Software Foundation; either version
 51  *              2 of the License, or (at your option) any later version.
 52  *
 53  */
 54 
 55 #include <linux/types.h>
 56 #include <linux/mm.h>
 57 #include <linux/capability.h>
 58 #include <linux/fcntl.h>
 59 #include <linux/socket.h>
 60 #include <linux/in.h>
 61 #include <linux/inet.h>
 62 #include <linux/netdevice.h>
 63 #include <linux/if_packet.h>
 64 #include <linux/wireless.h>
 65 #include <linux/kernel.h>
 66 #include <linux/kmod.h>
 67 #include <linux/slab.h>
 68 #include <linux/vmalloc.h>
 69 #include <net/net_namespace.h>
 70 #include <net/ip.h>
 71 #include <net/protocol.h>
 72 #include <linux/skbuff.h>
 73 #include <net/sock.h>
 74 #include <linux/errno.h>
 75 #include <linux/timer.h>
 76 #include <linux/uaccess.h>
 77 #include <asm/ioctls.h>
 78 #include <asm/page.h>
 79 #include <asm/cacheflush.h>
 80 #include <asm/io.h>
 81 #include <linux/proc_fs.h>
 82 #include <linux/seq_file.h>
 83 #include <linux/poll.h>
 84 #include <linux/module.h>
 85 #include <linux/init.h>
 86 #include <linux/mutex.h>
 87 #include <linux/if_vlan.h>
 88 #include <linux/virtio_net.h>
 89 #include <linux/errqueue.h>
 90 #include <linux/net_tstamp.h>
 91 #include <linux/percpu.h>
 92 #ifdef CONFIG_INET
 93 #include <net/inet_common.h>
 94 #endif
 95 #include <linux/bpf.h>
 96 #include <net/compat.h>
 97 
 98 #include "internal.h"
 99 
100 /*
101    Assumptions:
102    - if device has no dev->hard_header routine, it adds and removes ll header
103      inside itself. In this case ll header is invisible outside of device,
104      but higher levels still should reserve dev->hard_header_len.
105      Some devices are enough clever to reallocate skb, when header
106      will not fit to reserved space (tunnel), another ones are silly
107      (PPP).
108    - packet socket receives packets with pulled ll header,
109      so that SOCK_RAW should push it back.
110 
111 On receive:
112 -----------
113 
114 Incoming, dev->hard_header!=NULL
115    mac_header -> ll header
116    data       -> data
117 
118 Outgoing, dev->hard_header!=NULL
119    mac_header -> ll header
120    data       -> ll header
121 
122 Incoming, dev->hard_header==NULL
123    mac_header -> UNKNOWN position. It is very likely, that it points to ll
124                  header.  PPP makes it, that is wrong, because introduce
125                  assymetry between rx and tx paths.
126    data       -> data
127 
128 Outgoing, dev->hard_header==NULL
129    mac_header -> data. ll header is still not built!
130    data       -> data
131 
132 Resume
133   If dev->hard_header==NULL we are unlikely to restore sensible ll header.
134 
135 
136 On transmit:
137 ------------
138 
139 dev->hard_header != NULL
140    mac_header -> ll header
141    data       -> ll header
142 
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
144    mac_header -> data
145    data       -> data
146 
147    We should set nh.raw on output to correct posistion,
148    packet classifier depends on it.
149  */
150 
151 /* Private packet socket structures. */
152 
153 /* identical to struct packet_mreq except it has
154  * a longer address field.
155  */
156 struct packet_mreq_max {
157         int             mr_ifindex;
158         unsigned short  mr_type;
159         unsigned short  mr_alen;
160         unsigned char   mr_address[MAX_ADDR_LEN];
161 };
162 
163 union tpacket_uhdr {
164         struct tpacket_hdr  *h1;
165         struct tpacket2_hdr *h2;
166         struct tpacket3_hdr *h3;
167         void *raw;
168 };
169 
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171                 int closing, int tx_ring);
172 
173 #define V3_ALIGNMENT    (8)
174 
175 #define BLK_HDR_LEN     (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178         (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 
180 #define 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_rcv(struct sk_buff *skb, struct net_device *dev,
190                        struct packet_type *pt, struct net_device *orig_dev);
191 
192 static void *packet_previous_frame(struct packet_sock *po,
193                 struct packet_ring_buffer *rb,
194                 int status);
195 static void packet_increment_head(struct packet_ring_buffer *buff);
196 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
198                         struct packet_sock *);
199 static void prb_retire_current_block(struct tpacket_kbdq_core *,
200                 struct packet_sock *, unsigned int status);
201 static int prb_queue_frozen(struct tpacket_kbdq_core *);
202 static void prb_open_block(struct tpacket_kbdq_core *,
203                 struct tpacket_block_desc *);
204 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
206 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
207 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
208                 struct tpacket3_hdr *);
209 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
210                 struct tpacket3_hdr *);
211 static void packet_flush_mclist(struct sock *sk);
212 static u16 packet_pick_tx_queue(struct sk_buff *skb);
213 
214 struct packet_skb_cb {
215         union {
216                 struct sockaddr_pkt pkt;
217                 union {
218                         /* Trick: alias skb original length with
219                          * ll.sll_family and ll.protocol in order
220                          * to save room.
221                          */
222                         unsigned int origlen;
223                         struct sockaddr_ll ll;
224                 };
225         } sa;
226 };
227 
228 #define vio_le() virtio_legacy_is_little_endian()
229 
230 #define PACKET_SKB_CB(__skb)    ((struct packet_skb_cb *)((__skb)->cb))
231 
232 #define GET_PBDQC_FROM_RB(x)    ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
233 #define GET_PBLOCK_DESC(x, bid) \
234         ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
235 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x)       \
236         ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
237 #define GET_NEXT_PRB_BLK_NUM(x) \
238         (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
239         ((x)->kactive_blk_num+1) : 0)
240 
241 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
242 static void __fanout_link(struct sock *sk, struct packet_sock *po);
243 
244 static int packet_direct_xmit(struct sk_buff *skb)
245 {
246         return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
247 }
248 
249 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
250 {
251         struct net_device *dev;
252 
253         rcu_read_lock();
254         dev = rcu_dereference(po->cached_dev);
255         if (likely(dev))
256                 dev_hold(dev);
257         rcu_read_unlock();
258 
259         return dev;
260 }
261 
262 static void packet_cached_dev_assign(struct packet_sock *po,
263                                      struct net_device *dev)
264 {
265         rcu_assign_pointer(po->cached_dev, dev);
266 }
267 
268 static void packet_cached_dev_reset(struct packet_sock *po)
269 {
270         RCU_INIT_POINTER(po->cached_dev, NULL);
271 }
272 
273 static bool packet_use_direct_xmit(const struct packet_sock *po)
274 {
275         return po->xmit == packet_direct_xmit;
276 }
277 
278 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
279 {
280         return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
281 }
282 
283 static u16 packet_pick_tx_queue(struct sk_buff *skb)
284 {
285         struct net_device *dev = skb->dev;
286         const struct net_device_ops *ops = dev->netdev_ops;
287         u16 queue_index;
288 
289         if (ops->ndo_select_queue) {
290                 queue_index = ops->ndo_select_queue(dev, skb, NULL,
291                                                     __packet_pick_tx_queue);
292                 queue_index = netdev_cap_txqueue(dev, queue_index);
293         } else {
294                 queue_index = __packet_pick_tx_queue(dev, skb);
295         }
296 
297         return queue_index;
298 }
299 
300 /* __register_prot_hook must be invoked through register_prot_hook
301  * or from a context in which asynchronous accesses to the packet
302  * socket is not possible (packet_create()).
303  */
304 static void __register_prot_hook(struct sock *sk)
305 {
306         struct packet_sock *po = pkt_sk(sk);
307 
308         if (!po->running) {
309                 if (po->fanout)
310                         __fanout_link(sk, po);
311                 else
312                         dev_add_pack(&po->prot_hook);
313 
314                 sock_hold(sk);
315                 po->running = 1;
316         }
317 }
318 
319 static void register_prot_hook(struct sock *sk)
320 {
321         lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
322         __register_prot_hook(sk);
323 }
324 
325 /* If the sync parameter is true, we will temporarily drop
326  * the po->bind_lock and do a synchronize_net to make sure no
327  * asynchronous packet processing paths still refer to the elements
328  * of po->prot_hook.  If the sync parameter is false, it is the
329  * callers responsibility to take care of this.
330  */
331 static void __unregister_prot_hook(struct sock *sk, bool sync)
332 {
333         struct packet_sock *po = pkt_sk(sk);
334 
335         lockdep_assert_held_once(&po->bind_lock);
336 
337         po->running = 0;
338 
339         if (po->fanout)
340                 __fanout_unlink(sk, po);
341         else
342                 __dev_remove_pack(&po->prot_hook);
343 
344         __sock_put(sk);
345 
346         if (sync) {
347                 spin_unlock(&po->bind_lock);
348                 synchronize_net();
349                 spin_lock(&po->bind_lock);
350         }
351 }
352 
353 static void unregister_prot_hook(struct sock *sk, bool sync)
354 {
355         struct packet_sock *po = pkt_sk(sk);
356 
357         if (po->running)
358                 __unregister_prot_hook(sk, sync);
359 }
360 
361 static inline struct page * __pure pgv_to_page(void *addr)
362 {
363         if (is_vmalloc_addr(addr))
364                 return vmalloc_to_page(addr);
365         return virt_to_page(addr);
366 }
367 
368 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
369 {
370         union tpacket_uhdr h;
371 
372         h.raw = frame;
373         switch (po->tp_version) {
374         case TPACKET_V1:
375                 h.h1->tp_status = status;
376                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
377                 break;
378         case TPACKET_V2:
379                 h.h2->tp_status = status;
380                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
381                 break;
382         case TPACKET_V3:
383                 h.h3->tp_status = status;
384                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
385                 break;
386         default:
387                 WARN(1, "TPACKET version not supported.\n");
388                 BUG();
389         }
390 
391         smp_wmb();
392 }
393 
394 static int __packet_get_status(struct packet_sock *po, void *frame)
395 {
396         union tpacket_uhdr h;
397 
398         smp_rmb();
399 
400         h.raw = frame;
401         switch (po->tp_version) {
402         case TPACKET_V1:
403                 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
404                 return h.h1->tp_status;
405         case TPACKET_V2:
406                 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
407                 return h.h2->tp_status;
408         case TPACKET_V3:
409                 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
410                 return h.h3->tp_status;
411         default:
412                 WARN(1, "TPACKET version not supported.\n");
413                 BUG();
414                 return 0;
415         }
416 }
417 
418 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
419                                    unsigned int flags)
420 {
421         struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
422 
423         if (shhwtstamps &&
424             (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
425             ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
426                 return TP_STATUS_TS_RAW_HARDWARE;
427 
428         if (ktime_to_timespec_cond(skb->tstamp, ts))
429                 return TP_STATUS_TS_SOFTWARE;
430 
431         return 0;
432 }
433 
434 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
435                                     struct sk_buff *skb)
436 {
437         union tpacket_uhdr h;
438         struct timespec ts;
439         __u32 ts_status;
440 
441         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
442                 return 0;
443 
444         h.raw = frame;
445         switch (po->tp_version) {
446         case TPACKET_V1:
447                 h.h1->tp_sec = ts.tv_sec;
448                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
449                 break;
450         case TPACKET_V2:
451                 h.h2->tp_sec = ts.tv_sec;
452                 h.h2->tp_nsec = ts.tv_nsec;
453                 break;
454         case TPACKET_V3:
455                 h.h3->tp_sec = ts.tv_sec;
456                 h.h3->tp_nsec = ts.tv_nsec;
457                 break;
458         default:
459                 WARN(1, "TPACKET version not supported.\n");
460                 BUG();
461         }
462 
463         /* one flush is safe, as both fields always lie on the same cacheline */
464         flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
465         smp_wmb();
466 
467         return ts_status;
468 }
469 
470 static void *packet_lookup_frame(struct packet_sock *po,
471                 struct packet_ring_buffer *rb,
472                 unsigned int position,
473                 int status)
474 {
475         unsigned int pg_vec_pos, frame_offset;
476         union tpacket_uhdr h;
477 
478         pg_vec_pos = position / rb->frames_per_block;
479         frame_offset = position % rb->frames_per_block;
480 
481         h.raw = rb->pg_vec[pg_vec_pos].buffer +
482                 (frame_offset * rb->frame_size);
483 
484         if (status != __packet_get_status(po, h.raw))
485                 return NULL;
486 
487         return h.raw;
488 }
489 
490 static void *packet_current_frame(struct packet_sock *po,
491                 struct packet_ring_buffer *rb,
492                 int status)
493 {
494         return packet_lookup_frame(po, rb, rb->head, status);
495 }
496 
497 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
498 {
499         del_timer_sync(&pkc->retire_blk_timer);
500 }
501 
502 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
503                 struct sk_buff_head *rb_queue)
504 {
505         struct tpacket_kbdq_core *pkc;
506 
507         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
508 
509         spin_lock_bh(&rb_queue->lock);
510         pkc->delete_blk_timer = 1;
511         spin_unlock_bh(&rb_queue->lock);
512 
513         prb_del_retire_blk_timer(pkc);
514 }
515 
516 static void prb_setup_retire_blk_timer(struct packet_sock *po)
517 {
518         struct tpacket_kbdq_core *pkc;
519 
520         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
521         timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
522                     0);
523         pkc->retire_blk_timer.expires = jiffies;
524 }
525 
526 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
527                                 int blk_size_in_bytes)
528 {
529         struct net_device *dev;
530         unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
531         struct ethtool_link_ksettings ecmd;
532         int err;
533 
534         rtnl_lock();
535         dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
536         if (unlikely(!dev)) {
537                 rtnl_unlock();
538                 return DEFAULT_PRB_RETIRE_TOV;
539         }
540         err = __ethtool_get_link_ksettings(dev, &ecmd);
541         rtnl_unlock();
542         if (!err) {
543                 /*
544                  * If the link speed is so slow you don't really
545                  * need to worry about perf anyways
546                  */
547                 if (ecmd.base.speed < SPEED_1000 ||
548                     ecmd.base.speed == SPEED_UNKNOWN) {
549                         return DEFAULT_PRB_RETIRE_TOV;
550                 } else {
551                         msec = 1;
552                         div = ecmd.base.speed / 1000;
553                 }
554         }
555 
556         mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
557 
558         if (div)
559                 mbits /= div;
560 
561         tmo = mbits * msec;
562 
563         if (div)
564                 return tmo+1;
565         return tmo;
566 }
567 
568 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
569                         union tpacket_req_u *req_u)
570 {
571         p1->feature_req_word = req_u->req3.tp_feature_req_word;
572 }
573 
574 static void init_prb_bdqc(struct packet_sock *po,
575                         struct packet_ring_buffer *rb,
576                         struct pgv *pg_vec,
577                         union tpacket_req_u *req_u)
578 {
579         struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
580         struct tpacket_block_desc *pbd;
581 
582         memset(p1, 0x0, sizeof(*p1));
583 
584         p1->knxt_seq_num = 1;
585         p1->pkbdq = pg_vec;
586         pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
587         p1->pkblk_start = pg_vec[0].buffer;
588         p1->kblk_size = req_u->req3.tp_block_size;
589         p1->knum_blocks = req_u->req3.tp_block_nr;
590         p1->hdrlen = po->tp_hdrlen;
591         p1->version = po->tp_version;
592         p1->last_kactive_blk_num = 0;
593         po->stats.stats3.tp_freeze_q_cnt = 0;
594         if (req_u->req3.tp_retire_blk_tov)
595                 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
596         else
597                 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
598                                                 req_u->req3.tp_block_size);
599         p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
600         p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
601 
602         p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
603         prb_init_ft_ops(p1, req_u);
604         prb_setup_retire_blk_timer(po);
605         prb_open_block(p1, pbd);
606 }
607 
608 /*  Do NOT update the last_blk_num first.
609  *  Assumes sk_buff_head lock is held.
610  */
611 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
612 {
613         mod_timer(&pkc->retire_blk_timer,
614                         jiffies + pkc->tov_in_jiffies);
615         pkc->last_kactive_blk_num = pkc->kactive_blk_num;
616 }
617 
618 /*
619  * Timer logic:
620  * 1) We refresh the timer only when we open a block.
621  *    By doing this we don't waste cycles refreshing the timer
622  *        on packet-by-packet basis.
623  *
624  * With a 1MB block-size, on a 1Gbps line, it will take
625  * i) ~8 ms to fill a block + ii) memcpy etc.
626  * In this cut we are not accounting for the memcpy time.
627  *
628  * So, if the user sets the 'tmo' to 10ms then the timer
629  * will never fire while the block is still getting filled
630  * (which is what we want). However, the user could choose
631  * to close a block early and that's fine.
632  *
633  * But when the timer does fire, we check whether or not to refresh it.
634  * Since the tmo granularity is in msecs, it is not too expensive
635  * to refresh the timer, lets say every '8' msecs.
636  * Either the user can set the 'tmo' or we can derive it based on
637  * a) line-speed and b) block-size.
638  * prb_calc_retire_blk_tmo() calculates the tmo.
639  *
640  */
641 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
642 {
643         struct packet_sock *po =
644                 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
645         struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
646         unsigned int frozen;
647         struct tpacket_block_desc *pbd;
648 
649         spin_lock(&po->sk.sk_receive_queue.lock);
650 
651         frozen = prb_queue_frozen(pkc);
652         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
653 
654         if (unlikely(pkc->delete_blk_timer))
655                 goto out;
656 
657         /* We only need to plug the race when the block is partially filled.
658          * tpacket_rcv:
659          *              lock(); increment BLOCK_NUM_PKTS; unlock()
660          *              copy_bits() is in progress ...
661          *              timer fires on other cpu:
662          *              we can't retire the current block because copy_bits
663          *              is in progress.
664          *
665          */
666         if (BLOCK_NUM_PKTS(pbd)) {
667                 while (atomic_read(&pkc->blk_fill_in_prog)) {
668                         /* Waiting for skb_copy_bits to finish... */
669                         cpu_relax();
670                 }
671         }
672 
673         if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
674                 if (!frozen) {
675                         if (!BLOCK_NUM_PKTS(pbd)) {
676                                 /* An empty block. Just refresh the timer. */
677                                 goto refresh_timer;
678                         }
679                         prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
680                         if (!prb_dispatch_next_block(pkc, po))
681                                 goto refresh_timer;
682                         else
683                                 goto out;
684                 } else {
685                         /* Case 1. Queue was frozen because user-space was
686                          *         lagging behind.
687                          */
688                         if (prb_curr_blk_in_use(pbd)) {
689                                 /*
690                                  * Ok, user-space is still behind.
691                                  * So just refresh the timer.
692                                  */
693                                 goto refresh_timer;
694                         } else {
695                                /* Case 2. queue was frozen,user-space caught up,
696                                 * now the link went idle && the timer fired.
697                                 * We don't have a block to close.So we open this
698                                 * block and restart the timer.
699                                 * opening a block thaws the queue,restarts timer
700                                 * Thawing/timer-refresh is a side effect.
701                                 */
702                                 prb_open_block(pkc, pbd);
703                                 goto out;
704                         }
705                 }
706         }
707 
708 refresh_timer:
709         _prb_refresh_rx_retire_blk_timer(pkc);
710 
711 out:
712         spin_unlock(&po->sk.sk_receive_queue.lock);
713 }
714 
715 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
716                 struct tpacket_block_desc *pbd1, __u32 status)
717 {
718         /* Flush everything minus the block header */
719 
720 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
721         u8 *start, *end;
722 
723         start = (u8 *)pbd1;
724 
725         /* Skip the block header(we know header WILL fit in 4K) */
726         start += PAGE_SIZE;
727 
728         end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
729         for (; start < end; start += PAGE_SIZE)
730                 flush_dcache_page(pgv_to_page(start));
731 
732         smp_wmb();
733 #endif
734 
735         /* Now update the block status. */
736 
737         BLOCK_STATUS(pbd1) = status;
738 
739         /* Flush the block header */
740 
741 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
742         start = (u8 *)pbd1;
743         flush_dcache_page(pgv_to_page(start));
744 
745         smp_wmb();
746 #endif
747 }
748 
749 /*
750  * Side effect:
751  *
752  * 1) flush the block
753  * 2) Increment active_blk_num
754  *
755  * Note:We DONT refresh the timer on purpose.
756  *      Because almost always the next block will be opened.
757  */
758 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
759                 struct tpacket_block_desc *pbd1,
760                 struct packet_sock *po, unsigned int stat)
761 {
762         __u32 status = TP_STATUS_USER | stat;
763 
764         struct tpacket3_hdr *last_pkt;
765         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
766         struct sock *sk = &po->sk;
767 
768         if (po->stats.stats3.tp_drops)
769                 status |= TP_STATUS_LOSING;
770 
771         last_pkt = (struct tpacket3_hdr *)pkc1->prev;
772         last_pkt->tp_next_offset = 0;
773 
774         /* Get the ts of the last pkt */
775         if (BLOCK_NUM_PKTS(pbd1)) {
776                 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
777                 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
778         } else {
779                 /* Ok, we tmo'd - so get the current time.
780                  *
781                  * It shouldn't really happen as we don't close empty
782                  * blocks. See prb_retire_rx_blk_timer_expired().
783                  */
784                 struct timespec ts;
785                 getnstimeofday(&ts);
786                 h1->ts_last_pkt.ts_sec = ts.tv_sec;
787                 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
788         }
789 
790         smp_wmb();
791 
792         /* Flush the block */
793         prb_flush_block(pkc1, pbd1, status);
794 
795         sk->sk_data_ready(sk);
796 
797         pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
798 }
799 
800 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
801 {
802         pkc->reset_pending_on_curr_blk = 0;
803 }
804 
805 /*
806  * Side effect of opening a block:
807  *
808  * 1) prb_queue is thawed.
809  * 2) retire_blk_timer is refreshed.
810  *
811  */
812 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
813         struct tpacket_block_desc *pbd1)
814 {
815         struct timespec ts;
816         struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
817 
818         smp_rmb();
819 
820         /* We could have just memset this but we will lose the
821          * flexibility of making the priv area sticky
822          */
823 
824         BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
825         BLOCK_NUM_PKTS(pbd1) = 0;
826         BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
827 
828         getnstimeofday(&ts);
829 
830         h1->ts_first_pkt.ts_sec = ts.tv_sec;
831         h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
832 
833         pkc1->pkblk_start = (char *)pbd1;
834         pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
835 
836         BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837         BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
838 
839         pbd1->version = pkc1->version;
840         pkc1->prev = pkc1->nxt_offset;
841         pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
842 
843         prb_thaw_queue(pkc1);
844         _prb_refresh_rx_retire_blk_timer(pkc1);
845 
846         smp_wmb();
847 }
848 
849 /*
850  * Queue freeze logic:
851  * 1) Assume tp_block_nr = 8 blocks.
852  * 2) At time 't0', user opens Rx ring.
853  * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854  * 4) user-space is either sleeping or processing block ''.
855  * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856  *    it will close block-7,loop around and try to fill block ''.
857  *    call-flow:
858  *    __packet_lookup_frame_in_block
859  *      prb_retire_current_block()
860  *      prb_dispatch_next_block()
861  *        |->(BLOCK_STATUS == USER) evaluates to true
862  *    5.1) Since block-0 is currently in-use, we just freeze the queue.
863  * 6) Now there are two cases:
864  *    6.1) Link goes idle right after the queue is frozen.
865  *         But remember, the last open_block() refreshed the timer.
866  *         When this timer expires,it will refresh itself so that we can
867  *         re-open block-0 in near future.
868  *    6.2) Link is busy and keeps on receiving packets. This is a simple
869  *         case and __packet_lookup_frame_in_block will check if block-0
870  *         is free and can now be re-used.
871  */
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873                                   struct packet_sock *po)
874 {
875         pkc->reset_pending_on_curr_blk = 1;
876         po->stats.stats3.tp_freeze_q_cnt++;
877 }
878 
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
880 
881 /*
882  * If the next block is free then we will dispatch it
883  * and return a good offset.
884  * Else, we will freeze the queue.
885  * So, caller must check the return value.
886  */
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888                 struct packet_sock *po)
889 {
890         struct tpacket_block_desc *pbd;
891 
892         smp_rmb();
893 
894         /* 1. Get current block num */
895         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896 
897         /* 2. If this block is currently in_use then freeze the queue */
898         if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899                 prb_freeze_queue(pkc, po);
900                 return NULL;
901         }
902 
903         /*
904          * 3.
905          * open this block and return the offset where the first packet
906          * needs to get stored.
907          */
908         prb_open_block(pkc, pbd);
909         return (void *)pkc->nxt_offset;
910 }
911 
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913                 struct packet_sock *po, unsigned int status)
914 {
915         struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916 
917         /* retire/close the current block */
918         if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919                 /*
920                  * Plug the case where copy_bits() is in progress on
921                  * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922                  * have space to copy the pkt in the current block and
923                  * called prb_retire_current_block()
924                  *
925                  * We don't need to worry about the TMO case because
926                  * the timer-handler already handled this case.
927                  */
928                 if (!(status & TP_STATUS_BLK_TMO)) {
929                         while (atomic_read(&pkc->blk_fill_in_prog)) {
930                                 /* Waiting for skb_copy_bits to finish... */
931                                 cpu_relax();
932                         }
933                 }
934                 prb_close_block(pkc, pbd, po, status);
935                 return;
936         }
937 }
938 
939 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
940 {
941         return TP_STATUS_USER & BLOCK_STATUS(pbd);
942 }
943 
944 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
945 {
946         return pkc->reset_pending_on_curr_blk;
947 }
948 
949 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
950 {
951         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
952         atomic_dec(&pkc->blk_fill_in_prog);
953 }
954 
955 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
956                         struct tpacket3_hdr *ppd)
957 {
958         ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
959 }
960 
961 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
962                         struct tpacket3_hdr *ppd)
963 {
964         ppd->hv1.tp_rxhash = 0;
965 }
966 
967 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
968                         struct tpacket3_hdr *ppd)
969 {
970         if (skb_vlan_tag_present(pkc->skb)) {
971                 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
972                 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
973                 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
974         } else {
975                 ppd->hv1.tp_vlan_tci = 0;
976                 ppd->hv1.tp_vlan_tpid = 0;
977                 ppd->tp_status = TP_STATUS_AVAILABLE;
978         }
979 }
980 
981 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
982                         struct tpacket3_hdr *ppd)
983 {
984         ppd->hv1.tp_padding = 0;
985         prb_fill_vlan_info(pkc, ppd);
986 
987         if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
988                 prb_fill_rxhash(pkc, ppd);
989         else
990                 prb_clear_rxhash(pkc, ppd);
991 }
992 
993 static void prb_fill_curr_block(char *curr,
994                                 struct tpacket_kbdq_core *pkc,
995                                 struct tpacket_block_desc *pbd,
996                                 unsigned int len)
997 {
998         struct tpacket3_hdr *ppd;
999 
1000         ppd  = (struct tpacket3_hdr *)curr;
1001         ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1002         pkc->prev = curr;
1003         pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1004         BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005         BLOCK_NUM_PKTS(pbd) += 1;
1006         atomic_inc(&pkc->blk_fill_in_prog);
1007         prb_run_all_ft_ops(pkc, ppd);
1008 }
1009 
1010 /* Assumes caller has the sk->rx_queue.lock */
1011 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1012                                             struct sk_buff *skb,
1013                                                 int status,
1014                                             unsigned int len
1015                                             )
1016 {
1017         struct tpacket_kbdq_core *pkc;
1018         struct tpacket_block_desc *pbd;
1019         char *curr, *end;
1020 
1021         pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1022         pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1023 
1024         /* Queue is frozen when user space is lagging behind */
1025         if (prb_queue_frozen(pkc)) {
1026                 /*
1027                  * Check if that last block which caused the queue to freeze,
1028                  * is still in_use by user-space.
1029                  */
1030                 if (prb_curr_blk_in_use(pbd)) {
1031                         /* Can't record this packet */
1032                         return NULL;
1033                 } else {
1034                         /*
1035                          * Ok, the block was released by user-space.
1036                          * Now let's open that block.
1037                          * opening a block also thaws the queue.
1038                          * Thawing is a side effect.
1039                          */
1040                         prb_open_block(pkc, pbd);
1041                 }
1042         }
1043 
1044         smp_mb();
1045         curr = pkc->nxt_offset;
1046         pkc->skb = skb;
1047         end = (char *)pbd + pkc->kblk_size;
1048 
1049         /* first try the current block */
1050         if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1051                 prb_fill_curr_block(curr, pkc, pbd, len);
1052                 return (void *)curr;
1053         }
1054 
1055         /* Ok, close the current block */
1056         prb_retire_current_block(pkc, po, 0);
1057 
1058         /* Now, try to dispatch the next block */
1059         curr = (char *)prb_dispatch_next_block(pkc, po);
1060         if (curr) {
1061                 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1062                 prb_fill_curr_block(curr, pkc, pbd, len);
1063                 return (void *)curr;
1064         }
1065 
1066         /*
1067          * No free blocks are available.user_space hasn't caught up yet.
1068          * Queue was just frozen and now this packet will get dropped.
1069          */
1070         return NULL;
1071 }
1072 
1073 static void *packet_current_rx_frame(struct packet_sock *po,
1074                                             struct sk_buff *skb,
1075                                             int status, unsigned int len)
1076 {
1077         char *curr = NULL;
1078         switch (po->tp_version) {
1079         case TPACKET_V1:
1080         case TPACKET_V2:
1081                 curr = packet_lookup_frame(po, &po->rx_ring,
1082                                         po->rx_ring.head, status);
1083                 return curr;
1084         case TPACKET_V3:
1085                 return __packet_lookup_frame_in_block(po, skb, status, len);
1086         default:
1087                 WARN(1, "TPACKET version not supported\n");
1088                 BUG();
1089                 return NULL;
1090         }
1091 }
1092 
1093 static void *prb_lookup_block(struct packet_sock *po,
1094                                      struct packet_ring_buffer *rb,
1095                                      unsigned int idx,
1096                                      int status)
1097 {
1098         struct tpacket_kbdq_core *pkc  = GET_PBDQC_FROM_RB(rb);
1099         struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1100 
1101         if (status != BLOCK_STATUS(pbd))
1102                 return NULL;
1103         return pbd;
1104 }
1105 
1106 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1107 {
1108         unsigned int prev;
1109         if (rb->prb_bdqc.kactive_blk_num)
1110                 prev = rb->prb_bdqc.kactive_blk_num-1;
1111         else
1112                 prev = rb->prb_bdqc.knum_blocks-1;
1113         return prev;
1114 }
1115 
1116 /* Assumes caller has held the rx_queue.lock */
1117 static void *__prb_previous_block(struct packet_sock *po,
1118                                          struct packet_ring_buffer *rb,
1119                                          int status)
1120 {
1121         unsigned int previous = prb_previous_blk_num(rb);
1122         return prb_lookup_block(po, rb, previous, status);
1123 }
1124 
1125 static void *packet_previous_rx_frame(struct packet_sock *po,
1126                                              struct packet_ring_buffer *rb,
1127                                              int status)
1128 {
1129         if (po->tp_version <= TPACKET_V2)
1130                 return packet_previous_frame(po, rb, status);
1131 
1132         return __prb_previous_block(po, rb, status);
1133 }
1134 
1135 static void packet_increment_rx_head(struct packet_sock *po,
1136                                             struct packet_ring_buffer *rb)
1137 {
1138         switch (po->tp_version) {
1139         case TPACKET_V1:
1140         case TPACKET_V2:
1141                 return packet_increment_head(rb);
1142         case TPACKET_V3:
1143         default:
1144                 WARN(1, "TPACKET version not supported.\n");
1145                 BUG();
1146                 return;
1147         }
1148 }
1149 
1150 static void *packet_previous_frame(struct packet_sock *po,
1151                 struct packet_ring_buffer *rb,
1152                 int status)
1153 {
1154         unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1155         return packet_lookup_frame(po, rb, previous, status);
1156 }
1157 
1158 static void packet_increment_head(struct packet_ring_buffer *buff)
1159 {
1160         buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1161 }
1162 
1163 static void packet_inc_pending(struct packet_ring_buffer *rb)
1164 {
1165         this_cpu_inc(*rb->pending_refcnt);
1166 }
1167 
1168 static void packet_dec_pending(struct packet_ring_buffer *rb)
1169 {
1170         this_cpu_dec(*rb->pending_refcnt);
1171 }
1172 
1173 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1174 {
1175         unsigned int refcnt = 0;
1176         int cpu;
1177 
1178         /* We don't use pending refcount in rx_ring. */
1179         if (rb->pending_refcnt == NULL)
1180                 return 0;
1181 
1182         for_each_possible_cpu(cpu)
1183                 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1184 
1185         return refcnt;
1186 }
1187 
1188 static int packet_alloc_pending(struct packet_sock *po)
1189 {
1190         po->rx_ring.pending_refcnt = NULL;
1191 
1192         po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1193         if (unlikely(po->tx_ring.pending_refcnt == NULL))
1194                 return -ENOBUFS;
1195 
1196         return 0;
1197 }
1198 
1199 static void packet_free_pending(struct packet_sock *po)
1200 {
1201         free_percpu(po->tx_ring.pending_refcnt);
1202 }
1203 
1204 #define ROOM_POW_OFF    2
1205 #define ROOM_NONE       0x0
1206 #define ROOM_LOW        0x1
1207 #define ROOM_NORMAL     0x2
1208 
1209 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1210 {
1211         int idx, len;
1212 
1213         len = po->rx_ring.frame_max + 1;
1214         idx = po->rx_ring.head;
1215         if (pow_off)
1216                 idx += len >> pow_off;
1217         if (idx >= len)
1218                 idx -= len;
1219         return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1220 }
1221 
1222 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1223 {
1224         int idx, len;
1225 
1226         len = po->rx_ring.prb_bdqc.knum_blocks;
1227         idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1228         if (pow_off)
1229                 idx += len >> pow_off;
1230         if (idx >= len)
1231                 idx -= len;
1232         return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1233 }
1234 
1235 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1236 {
1237         struct sock *sk = &po->sk;
1238         int ret = ROOM_NONE;
1239 
1240         if (po->prot_hook.func != tpacket_rcv) {
1241                 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1242                                           - (skb ? skb->truesize : 0);
1243                 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1244                         return ROOM_NORMAL;
1245                 else if (avail > 0)
1246                         return ROOM_LOW;
1247                 else
1248                         return ROOM_NONE;
1249         }
1250 
1251         if (po->tp_version == TPACKET_V3) {
1252                 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1253                         ret = ROOM_NORMAL;
1254                 else if (__tpacket_v3_has_room(po, 0))
1255                         ret = ROOM_LOW;
1256         } else {
1257                 if (__tpacket_has_room(po, ROOM_POW_OFF))
1258                         ret = ROOM_NORMAL;
1259                 else if (__tpacket_has_room(po, 0))
1260                         ret = ROOM_LOW;
1261         }
1262 
1263         return ret;
1264 }
1265 
1266 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1267 {
1268         int ret;
1269         bool has_room;
1270 
1271         spin_lock_bh(&po->sk.sk_receive_queue.lock);
1272         ret = __packet_rcv_has_room(po, skb);
1273         has_room = ret == ROOM_NORMAL;
1274         if (po->pressure == has_room)
1275                 po->pressure = !has_room;
1276         spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1277 
1278         return ret;
1279 }
1280 
1281 static void packet_sock_destruct(struct sock *sk)
1282 {
1283         skb_queue_purge(&sk->sk_error_queue);
1284 
1285         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1286         WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1287 
1288         if (!sock_flag(sk, SOCK_DEAD)) {
1289                 pr_err("Attempt to release alive packet socket: %p\n", sk);
1290                 return;
1291         }
1292 
1293         sk_refcnt_debug_dec(sk);
1294 }
1295 
1296 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1297 {
1298         u32 rxhash;
1299         int i, count = 0;
1300 
1301         rxhash = skb_get_hash(skb);
1302         for (i = 0; i < ROLLOVER_HLEN; i++)
1303                 if (po->rollover->history[i] == rxhash)
1304                         count++;
1305 
1306         po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1307         return count > (ROLLOVER_HLEN >> 1);
1308 }
1309 
1310 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1311                                       struct sk_buff *skb,
1312                                       unsigned int num)
1313 {
1314         return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1315 }
1316 
1317 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1318                                     struct sk_buff *skb,
1319                                     unsigned int num)
1320 {
1321         unsigned int val = atomic_inc_return(&f->rr_cur);
1322 
1323         return val % num;
1324 }
1325 
1326 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1327                                      struct sk_buff *skb,
1328                                      unsigned int num)
1329 {
1330         return smp_processor_id() % num;
1331 }
1332 
1333 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1334                                      struct sk_buff *skb,
1335                                      unsigned int num)
1336 {
1337         return prandom_u32_max(num);
1338 }
1339 
1340 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1341                                           struct sk_buff *skb,
1342                                           unsigned int idx, bool try_self,
1343                                           unsigned int num)
1344 {
1345         struct packet_sock *po, *po_next, *po_skip = NULL;
1346         unsigned int i, j, room = ROOM_NONE;
1347 
1348         po = pkt_sk(f->arr[idx]);
1349 
1350         if (try_self) {
1351                 room = packet_rcv_has_room(po, skb);
1352                 if (room == ROOM_NORMAL ||
1353                     (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1354                         return idx;
1355                 po_skip = po;
1356         }
1357 
1358         i = j = min_t(int, po->rollover->sock, num - 1);
1359         do {
1360                 po_next = pkt_sk(f->arr[i]);
1361                 if (po_next != po_skip && !po_next->pressure &&
1362                     packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1363                         if (i != j)
1364                                 po->rollover->sock = i;
1365                         atomic_long_inc(&po->rollover->num);
1366                         if (room == ROOM_LOW)
1367                                 atomic_long_inc(&po->rollover->num_huge);
1368                         return i;
1369                 }
1370 
1371                 if (++i == num)
1372                         i = 0;
1373         } while (i != j);
1374 
1375         atomic_long_inc(&po->rollover->num_failed);
1376         return idx;
1377 }
1378 
1379 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1380                                     struct sk_buff *skb,
1381                                     unsigned int num)
1382 {
1383         return skb_get_queue_mapping(skb) % num;
1384 }
1385 
1386 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1387                                      struct sk_buff *skb,
1388                                      unsigned int num)
1389 {
1390         struct bpf_prog *prog;
1391         unsigned int ret = 0;
1392 
1393         rcu_read_lock();
1394         prog = rcu_dereference(f->bpf_prog);
1395         if (prog)
1396                 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1397         rcu_read_unlock();
1398 
1399         return ret;
1400 }
1401 
1402 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1403 {
1404         return f->flags & (flag >> 8);
1405 }
1406 
1407 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1408                              struct packet_type *pt, struct net_device *orig_dev)
1409 {
1410         struct packet_fanout *f = pt->af_packet_priv;
1411         unsigned int num = READ_ONCE(f->num_members);
1412         struct net *net = read_pnet(&f->net);
1413         struct packet_sock *po;
1414         unsigned int idx;
1415 
1416         if (!net_eq(dev_net(dev), net) || !num) {
1417                 kfree_skb(skb);
1418                 return 0;
1419         }
1420 
1421         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1422                 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1423                 if (!skb)
1424                         return 0;
1425         }
1426         switch (f->type) {
1427         case PACKET_FANOUT_HASH:
1428         default:
1429                 idx = fanout_demux_hash(f, skb, num);
1430                 break;
1431         case PACKET_FANOUT_LB:
1432                 idx = fanout_demux_lb(f, skb, num);
1433                 break;
1434         case PACKET_FANOUT_CPU:
1435                 idx = fanout_demux_cpu(f, skb, num);
1436                 break;
1437         case PACKET_FANOUT_RND:
1438                 idx = fanout_demux_rnd(f, skb, num);
1439                 break;
1440         case PACKET_FANOUT_QM:
1441                 idx = fanout_demux_qm(f, skb, num);
1442                 break;
1443         case PACKET_FANOUT_ROLLOVER:
1444                 idx = fanout_demux_rollover(f, skb, 0, false, num);
1445                 break;
1446         case PACKET_FANOUT_CBPF:
1447         case PACKET_FANOUT_EBPF:
1448                 idx = fanout_demux_bpf(f, skb, num);
1449                 break;
1450         }
1451 
1452         if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1453                 idx = fanout_demux_rollover(f, skb, idx, true, num);
1454 
1455         po = pkt_sk(f->arr[idx]);
1456         return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1457 }
1458 
1459 DEFINE_MUTEX(fanout_mutex);
1460 EXPORT_SYMBOL_GPL(fanout_mutex);
1461 static LIST_HEAD(fanout_list);
1462 static u16 fanout_next_id;
1463 
1464 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1465 {
1466         struct packet_fanout *f = po->fanout;
1467 
1468         spin_lock(&f->lock);
1469         f->arr[f->num_members] = sk;
1470         smp_wmb();
1471         f->num_members++;
1472         if (f->num_members == 1)
1473                 dev_add_pack(&f->prot_hook);
1474         spin_unlock(&f->lock);
1475 }
1476 
1477 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1478 {
1479         struct packet_fanout *f = po->fanout;
1480         int i;
1481 
1482         spin_lock(&f->lock);
1483         for (i = 0; i < f->num_members; i++) {
1484                 if (f->arr[i] == sk)
1485                         break;
1486         }
1487         BUG_ON(i >= f->num_members);
1488         f->arr[i] = f->arr[f->num_members - 1];
1489         f->num_members--;
1490         if (f->num_members == 0)
1491                 __dev_remove_pack(&f->prot_hook);
1492         spin_unlock(&f->lock);
1493 }
1494 
1495 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1496 {
1497         if (sk->sk_family != PF_PACKET)
1498                 return false;
1499 
1500         return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1501 }
1502 
1503 static void fanout_init_data(struct packet_fanout *f)
1504 {
1505         switch (f->type) {
1506         case PACKET_FANOUT_LB:
1507                 atomic_set(&f->rr_cur, 0);
1508                 break;
1509         case PACKET_FANOUT_CBPF:
1510         case PACKET_FANOUT_EBPF:
1511                 RCU_INIT_POINTER(f->bpf_prog, NULL);
1512                 break;
1513         }
1514 }
1515 
1516 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1517 {
1518         struct bpf_prog *old;
1519 
1520         spin_lock(&f->lock);
1521         old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1522         rcu_assign_pointer(f->bpf_prog, new);
1523         spin_unlock(&f->lock);
1524 
1525         if (old) {
1526                 synchronize_net();
1527                 bpf_prog_destroy(old);
1528         }
1529 }
1530 
1531 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1532                                 unsigned int len)
1533 {
1534         struct bpf_prog *new;
1535         struct sock_fprog fprog;
1536         int ret;
1537 
1538         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1539                 return -EPERM;
1540         if (len != sizeof(fprog))
1541                 return -EINVAL;
1542         if (copy_from_user(&fprog, data, len))
1543                 return -EFAULT;
1544 
1545         ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1546         if (ret)
1547                 return ret;
1548 
1549         __fanout_set_data_bpf(po->fanout, new);
1550         return 0;
1551 }
1552 
1553 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1554                                 unsigned int len)
1555 {
1556         struct bpf_prog *new;
1557         u32 fd;
1558 
1559         if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1560                 return -EPERM;
1561         if (len != sizeof(fd))
1562                 return -EINVAL;
1563         if (copy_from_user(&fd, data, len))
1564                 return -EFAULT;
1565 
1566         new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1567         if (IS_ERR(new))
1568                 return PTR_ERR(new);
1569 
1570         __fanout_set_data_bpf(po->fanout, new);
1571         return 0;
1572 }
1573 
1574 static int fanout_set_data(struct packet_sock *po, char __user *data,
1575                            unsigned int len)
1576 {
1577         switch (po->fanout->type) {
1578         case PACKET_FANOUT_CBPF:
1579                 return fanout_set_data_cbpf(po, data, len);
1580         case PACKET_FANOUT_EBPF:
1581                 return fanout_set_data_ebpf(po, data, len);
1582         default:
1583                 return -EINVAL;
1584         };
1585 }
1586 
1587 static void fanout_release_data(struct packet_fanout *f)
1588 {
1589         switch (f->type) {
1590         case PACKET_FANOUT_CBPF:
1591         case PACKET_FANOUT_EBPF:
1592                 __fanout_set_data_bpf(f, NULL);
1593         };
1594 }
1595 
1596 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1597 {
1598         struct packet_fanout *f;
1599 
1600         list_for_each_entry(f, &fanout_list, list) {
1601                 if (f->id == candidate_id &&
1602                     read_pnet(&f->net) == sock_net(sk)) {
1603                         return false;
1604                 }
1605         }
1606         return true;
1607 }
1608 
1609 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1610 {
1611         u16 id = fanout_next_id;
1612 
1613         do {
1614                 if (__fanout_id_is_free(sk, id)) {
1615                         *new_id = id;
1616                         fanout_next_id = id + 1;
1617                         return true;
1618                 }
1619 
1620                 id++;
1621         } while (id != fanout_next_id);
1622 
1623         return false;
1624 }
1625 
1626 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1627 {
1628         struct packet_rollover *rollover = NULL;
1629         struct packet_sock *po = pkt_sk(sk);
1630         struct packet_fanout *f, *match;
1631         u8 type = type_flags & 0xff;
1632         u8 flags = type_flags >> 8;
1633         int err;
1634 
1635         switch (type) {
1636         case PACKET_FANOUT_ROLLOVER:
1637                 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1638                         return -EINVAL;
1639         case PACKET_FANOUT_HASH:
1640         case PACKET_FANOUT_LB:
1641         case PACKET_FANOUT_CPU:
1642         case PACKET_FANOUT_RND:
1643         case PACKET_FANOUT_QM:
1644         case PACKET_FANOUT_CBPF:
1645         case PACKET_FANOUT_EBPF:
1646                 break;
1647         default:
1648                 return -EINVAL;
1649         }
1650 
1651         mutex_lock(&fanout_mutex);
1652 
1653         err = -EALREADY;
1654         if (po->fanout)
1655                 goto out;
1656 
1657         if (type == PACKET_FANOUT_ROLLOVER ||
1658             (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1659                 err = -ENOMEM;
1660                 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1661                 if (!rollover)
1662                         goto out;
1663                 atomic_long_set(&rollover->num, 0);
1664                 atomic_long_set(&rollover->num_huge, 0);
1665                 atomic_long_set(&rollover->num_failed, 0);
1666         }
1667 
1668         if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1669                 if (id != 0) {
1670                         err = -EINVAL;
1671                         goto out;
1672                 }
1673                 if (!fanout_find_new_id(sk, &id)) {
1674                         err = -ENOMEM;
1675                         goto out;
1676                 }
1677                 /* ephemeral flag for the first socket in the group: drop it */
1678                 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1679         }
1680 
1681         match = NULL;
1682         list_for_each_entry(f, &fanout_list, list) {
1683                 if (f->id == id &&
1684                     read_pnet(&f->net) == sock_net(sk)) {
1685                         match = f;
1686                         break;
1687                 }
1688         }
1689         err = -EINVAL;
1690         if (match && match->flags != flags)
1691                 goto out;
1692         if (!match) {
1693                 err = -ENOMEM;
1694                 match = kzalloc(sizeof(*match), GFP_KERNEL);
1695                 if (!match)
1696                         goto out;
1697                 write_pnet(&match->net, sock_net(sk));
1698                 match->id = id;
1699                 match->type = type;
1700                 match->flags = flags;
1701                 INIT_LIST_HEAD(&match->list);
1702                 spin_lock_init(&match->lock);
1703                 refcount_set(&match->sk_ref, 0);
1704                 fanout_init_data(match);
1705                 match->prot_hook.type = po->prot_hook.type;
1706                 match->prot_hook.dev = po->prot_hook.dev;
1707                 match->prot_hook.func = packet_rcv_fanout;
1708                 match->prot_hook.af_packet_priv = match;
1709                 match->prot_hook.id_match = match_fanout_group;
1710                 list_add(&match->list, &fanout_list);
1711         }
1712         err = -EINVAL;
1713 
1714         spin_lock(&po->bind_lock);
1715         if (po->running &&
1716             match->type == type &&
1717             match->prot_hook.type == po->prot_hook.type &&
1718             match->prot_hook.dev == po->prot_hook.dev) {
1719                 err = -ENOSPC;
1720                 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1721                         __dev_remove_pack(&po->prot_hook);
1722                         po->fanout = match;
1723                         po->rollover = rollover;
1724                         rollover = NULL;
1725                         refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1726                         __fanout_link(sk, po);
1727                         err = 0;
1728                 }
1729         }
1730         spin_unlock(&po->bind_lock);
1731 
1732         if (err && !refcount_read(&match->sk_ref)) {
1733                 list_del(&match->list);
1734                 kfree(match);
1735         }
1736 
1737 out:
1738         kfree(rollover);
1739         mutex_unlock(&fanout_mutex);
1740         return err;
1741 }
1742 
1743 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1744  * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1745  * It is the responsibility of the caller to call fanout_release_data() and
1746  * free the returned packet_fanout (after synchronize_net())
1747  */
1748 static struct packet_fanout *fanout_release(struct sock *sk)
1749 {
1750         struct packet_sock *po = pkt_sk(sk);
1751         struct packet_fanout *f;
1752 
1753         mutex_lock(&fanout_mutex);
1754         f = po->fanout;
1755         if (f) {
1756                 po->fanout = NULL;
1757 
1758                 if (refcount_dec_and_test(&f->sk_ref))
1759                         list_del(&f->list);
1760                 else
1761                         f = NULL;
1762         }
1763         mutex_unlock(&fanout_mutex);
1764 
1765         return f;
1766 }
1767 
1768 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1769                                           struct sk_buff *skb)
1770 {
1771         /* Earlier code assumed this would be a VLAN pkt, double-check
1772          * this now that we have the actual packet in hand. We can only
1773          * do this check on Ethernet devices.
1774          */
1775         if (unlikely(dev->type != ARPHRD_ETHER))
1776                 return false;
1777 
1778         skb_reset_mac_header(skb);
1779         return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1780 }
1781 
1782 static const struct proto_ops packet_ops;
1783 
1784 static const struct proto_ops packet_ops_spkt;
1785 
1786 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1787                            struct packet_type *pt, struct net_device *orig_dev)
1788 {
1789         struct sock *sk;
1790         struct sockaddr_pkt *spkt;
1791 
1792         /*
1793          *      When we registered the protocol we saved the socket in the data
1794          *      field for just this event.
1795          */
1796 
1797         sk = pt->af_packet_priv;
1798 
1799         /*
1800          *      Yank back the headers [hope the device set this
1801          *      right or kerboom...]
1802          *
1803          *      Incoming packets have ll header pulled,
1804          *      push it back.
1805          *
1806          *      For outgoing ones skb->data == skb_mac_header(skb)
1807          *      so that this procedure is noop.
1808          */
1809 
1810         if (skb->pkt_type == PACKET_LOOPBACK)
1811                 goto out;
1812 
1813         if (!net_eq(dev_net(dev), sock_net(sk)))
1814                 goto out;
1815 
1816         skb = skb_share_check(skb, GFP_ATOMIC);
1817         if (skb == NULL)
1818                 goto oom;
1819 
1820         /* drop any routing info */
1821         skb_dst_drop(skb);
1822 
1823         /* drop conntrack reference */
1824         nf_reset(skb);
1825 
1826         spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1827 
1828         skb_push(skb, skb->data - skb_mac_header(skb));
1829 
1830         /*
1831          *      The SOCK_PACKET socket receives _all_ frames.
1832          */
1833 
1834         spkt->spkt_family = dev->type;
1835         strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1836         spkt->spkt_protocol = skb->protocol;
1837 
1838         /*
1839          *      Charge the memory to the socket. This is done specifically
1840          *      to prevent sockets using all the memory up.
1841          */
1842 
1843         if (sock_queue_rcv_skb(sk, skb) == 0)
1844                 return 0;
1845 
1846 out:
1847         kfree_skb(skb);
1848 oom:
1849         return 0;
1850 }
1851 
1852 
1853 /*
1854  *      Output a raw packet to a device layer. This bypasses all the other
1855  *      protocol layers and you must therefore supply it with a complete frame
1856  */
1857 
1858 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1859                                size_t len)
1860 {
1861         struct sock *sk = sock->sk;
1862         DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1863         struct sk_buff *skb = NULL;
1864         struct net_device *dev;
1865         struct sockcm_cookie sockc;
1866         __be16 proto = 0;
1867         int err;
1868         int extra_len = 0;
1869 
1870         /*
1871          *      Get and verify the address.
1872          */
1873 
1874         if (saddr) {
1875                 if (msg->msg_namelen < sizeof(struct sockaddr))
1876                         return -EINVAL;
1877                 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1878                         proto = saddr->spkt_protocol;
1879         } else
1880                 return -ENOTCONN;       /* SOCK_PACKET must be sent giving an address */
1881 
1882         /*
1883          *      Find the device first to size check it
1884          */
1885 
1886         saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1887 retry:
1888         rcu_read_lock();
1889         dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1890         err = -ENODEV;
1891         if (dev == NULL)
1892                 goto out_unlock;
1893 
1894         err = -ENETDOWN;
1895         if (!(dev->flags & IFF_UP))
1896                 goto out_unlock;
1897 
1898         /*
1899          * You may not queue a frame bigger than the mtu. This is the lowest level
1900          * raw protocol and you must do your own fragmentation at this level.
1901          */
1902 
1903         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1904                 if (!netif_supports_nofcs(dev)) {
1905                         err = -EPROTONOSUPPORT;
1906                         goto out_unlock;
1907                 }
1908                 extra_len = 4; /* We're doing our own CRC */
1909         }
1910 
1911         err = -EMSGSIZE;
1912         if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1913                 goto out_unlock;
1914 
1915         if (!skb) {
1916                 size_t reserved = LL_RESERVED_SPACE(dev);
1917                 int tlen = dev->needed_tailroom;
1918                 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1919 
1920                 rcu_read_unlock();
1921                 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1922                 if (skb == NULL)
1923                         return -ENOBUFS;
1924                 /* FIXME: Save some space for broken drivers that write a hard
1925                  * header at transmission time by themselves. PPP is the notable
1926                  * one here. This should really be fixed at the driver level.
1927                  */
1928                 skb_reserve(skb, reserved);
1929                 skb_reset_network_header(skb);
1930 
1931                 /* Try to align data part correctly */
1932                 if (hhlen) {
1933                         skb->data -= hhlen;
1934                         skb->tail -= hhlen;
1935                         if (len < hhlen)
1936                                 skb_reset_network_header(skb);
1937                 }
1938                 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1939                 if (err)
1940                         goto out_free;
1941                 goto retry;
1942         }
1943 
1944         if (!dev_validate_header(dev, skb->data, len)) {
1945                 err = -EINVAL;
1946                 goto out_unlock;
1947         }
1948         if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1949             !packet_extra_vlan_len_allowed(dev, skb)) {
1950                 err = -EMSGSIZE;
1951                 goto out_unlock;
1952         }
1953 
1954         sockc.tsflags = sk->sk_tsflags;
1955         if (msg->msg_controllen) {
1956                 err = sock_cmsg_send(sk, msg, &sockc);
1957                 if (unlikely(err))
1958                         goto out_unlock;
1959         }
1960 
1961         skb->protocol = proto;
1962         skb->dev = dev;
1963         skb->priority = sk->sk_priority;
1964         skb->mark = sk->sk_mark;
1965 
1966         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1967 
1968         if (unlikely(extra_len == 4))
1969                 skb->no_fcs = 1;
1970 
1971         skb_probe_transport_header(skb, 0);
1972 
1973         dev_queue_xmit(skb);
1974         rcu_read_unlock();
1975         return len;
1976 
1977 out_unlock:
1978         rcu_read_unlock();
1979 out_free:
1980         kfree_skb(skb);
1981         return err;
1982 }
1983 
1984 static unsigned int run_filter(struct sk_buff *skb,
1985                                const struct sock *sk,
1986                                unsigned int res)
1987 {
1988         struct sk_filter *filter;
1989 
1990         rcu_read_lock();
1991         filter = rcu_dereference(sk->sk_filter);
1992         if (filter != NULL)
1993                 res = bpf_prog_run_clear_cb(filter->prog, skb);
1994         rcu_read_unlock();
1995 
1996         return res;
1997 }
1998 
1999 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2000                            size_t *len)
2001 {
2002         struct virtio_net_hdr vnet_hdr;
2003 
2004         if (*len < sizeof(vnet_hdr))
2005                 return -EINVAL;
2006         *len -= sizeof(vnet_hdr);
2007 
2008         if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2009                 return -EINVAL;
2010 
2011         return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2012 }
2013 
2014 /*
2015  * This function makes lazy skb cloning in hope that most of packets
2016  * are discarded by BPF.
2017  *
2018  * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2019  * and skb->cb are mangled. It works because (and until) packets
2020  * falling here are owned by current CPU. Output packets are cloned
2021  * by dev_queue_xmit_nit(), input packets are processed by net_bh
2022  * sequencially, so that if we return skb to original state on exit,
2023  * we will not harm anyone.
2024  */
2025 
2026 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2027                       struct packet_type *pt, struct net_device *orig_dev)
2028 {
2029         struct sock *sk;
2030         struct sockaddr_ll *sll;
2031         struct packet_sock *po;
2032         u8 *skb_head = skb->data;
2033         int skb_len = skb->len;
2034         unsigned int snaplen, res;
2035         bool is_drop_n_account = false;
2036 
2037         if (skb->pkt_type == PACKET_LOOPBACK)
2038                 goto drop;
2039 
2040         sk = pt->af_packet_priv;
2041         po = pkt_sk(sk);
2042 
2043         if (!net_eq(dev_net(dev), sock_net(sk)))
2044                 goto drop;
2045 
2046         skb->dev = dev;
2047 
2048         if (dev->header_ops) {
2049                 /* The device has an explicit notion of ll header,
2050                  * exported to higher levels.
2051                  *
2052                  * Otherwise, the device hides details of its frame
2053                  * structure, so that corresponding packet head is
2054                  * never delivered to user.
2055                  */
2056                 if (sk->sk_type != SOCK_DGRAM)
2057                         skb_push(skb, skb->data - skb_mac_header(skb));
2058                 else if (skb->pkt_type == PACKET_OUTGOING) {
2059                         /* Special case: outgoing packets have ll header at head */
2060                         skb_pull(skb, skb_network_offset(skb));
2061                 }
2062         }
2063 
2064         snaplen = skb->len;
2065 
2066         res = run_filter(skb, sk, snaplen);
2067         if (!res)
2068                 goto drop_n_restore;
2069         if (snaplen > res)
2070                 snaplen = res;
2071 
2072         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2073                 goto drop_n_acct;
2074 
2075         if (skb_shared(skb)) {
2076                 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2077                 if (nskb == NULL)
2078                         goto drop_n_acct;
2079 
2080                 if (skb_head != skb->data) {
2081                         skb->data = skb_head;
2082                         skb->len = skb_len;
2083                 }
2084                 consume_skb(skb);
2085                 skb = nskb;
2086         }
2087 
2088         sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2089 
2090         sll = &PACKET_SKB_CB(skb)->sa.ll;
2091         sll->sll_hatype = dev->type;
2092         sll->sll_pkttype = skb->pkt_type;
2093         if (unlikely(po->origdev))
2094                 sll->sll_ifindex = orig_dev->ifindex;
2095         else
2096                 sll->sll_ifindex = dev->ifindex;
2097 
2098         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2099 
2100         /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2101          * Use their space for storing the original skb length.
2102          */
2103         PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2104 
2105         if (pskb_trim(skb, snaplen))
2106                 goto drop_n_acct;
2107 
2108         skb_set_owner_r(skb, sk);
2109         skb->dev = NULL;
2110         skb_dst_drop(skb);
2111 
2112         /* drop conntrack reference */
2113         nf_reset(skb);
2114 
2115         spin_lock(&sk->sk_receive_queue.lock);
2116         po->stats.stats1.tp_packets++;
2117         sock_skb_set_dropcount(sk, skb);
2118         __skb_queue_tail(&sk->sk_receive_queue, skb);
2119         spin_unlock(&sk->sk_receive_queue.lock);
2120         sk->sk_data_ready(sk);
2121         return 0;
2122 
2123 drop_n_acct:
2124         is_drop_n_account = true;
2125         spin_lock(&sk->sk_receive_queue.lock);
2126         po->stats.stats1.tp_drops++;
2127         atomic_inc(&sk->sk_drops);
2128         spin_unlock(&sk->sk_receive_queue.lock);
2129 
2130 drop_n_restore:
2131         if (skb_head != skb->data && skb_shared(skb)) {
2132                 skb->data = skb_head;
2133                 skb->len = skb_len;
2134         }
2135 drop:
2136         if (!is_drop_n_account)
2137                 consume_skb(skb);
2138         else
2139                 kfree_skb(skb);
2140         return 0;
2141 }
2142 
2143 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2144                        struct packet_type *pt, struct net_device *orig_dev)
2145 {
2146         struct sock *sk;
2147         struct packet_sock *po;
2148         struct sockaddr_ll *sll;
2149         union tpacket_uhdr h;
2150         u8 *skb_head = skb->data;
2151         int skb_len = skb->len;
2152         unsigned int snaplen, res;
2153         unsigned long status = TP_STATUS_USER;
2154         unsigned short macoff, netoff, hdrlen;
2155         struct sk_buff *copy_skb = NULL;
2156         struct timespec ts;
2157         __u32 ts_status;
2158         bool is_drop_n_account = false;
2159         bool do_vnet = false;
2160 
2161         /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2162          * We may add members to them until current aligned size without forcing
2163          * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2164          */
2165         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2166         BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2167 
2168         if (skb->pkt_type == PACKET_LOOPBACK)
2169                 goto drop;
2170 
2171         sk = pt->af_packet_priv;
2172         po = pkt_sk(sk);
2173 
2174         if (!net_eq(dev_net(dev), sock_net(sk)))
2175                 goto drop;
2176 
2177         if (dev->header_ops) {
2178                 if (sk->sk_type != SOCK_DGRAM)
2179                         skb_push(skb, skb->data - skb_mac_header(skb));
2180                 else if (skb->pkt_type == PACKET_OUTGOING) {
2181                         /* Special case: outgoing packets have ll header at head */
2182                         skb_pull(skb, skb_network_offset(skb));
2183                 }
2184         }
2185 
2186         snaplen = skb->len;
2187 
2188         res = run_filter(skb, sk, snaplen);
2189         if (!res)
2190                 goto drop_n_restore;
2191 
2192         if (skb->ip_summed == CHECKSUM_PARTIAL)
2193                 status |= TP_STATUS_CSUMNOTREADY;
2194         else if (skb->pkt_type != PACKET_OUTGOING &&
2195                  (skb->ip_summed == CHECKSUM_COMPLETE ||
2196                   skb_csum_unnecessary(skb)))
2197                 status |= TP_STATUS_CSUM_VALID;
2198 
2199         if (snaplen > res)
2200                 snaplen = res;
2201 
2202         if (sk->sk_type == SOCK_DGRAM) {
2203                 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2204                                   po->tp_reserve;
2205         } else {
2206                 unsigned int maclen = skb_network_offset(skb);
2207                 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2208                                        (maclen < 16 ? 16 : maclen)) +
2209                                        po->tp_reserve;
2210                 if (po->has_vnet_hdr) {
2211                         netoff += sizeof(struct virtio_net_hdr);
2212                         do_vnet = true;
2213                 }
2214                 macoff = netoff - maclen;
2215         }
2216         if (po->tp_version <= TPACKET_V2) {
2217                 if (macoff + snaplen > po->rx_ring.frame_size) {
2218                         if (po->copy_thresh &&
2219                             atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2220                                 if (skb_shared(skb)) {
2221                                         copy_skb = skb_clone(skb, GFP_ATOMIC);
2222                                 } else {
2223                                         copy_skb = skb_get(skb);
2224                                         skb_head = skb->data;
2225                                 }
2226                                 if (copy_skb)
2227                                         skb_set_owner_r(copy_skb, sk);
2228                         }
2229                         snaplen = po->rx_ring.frame_size - macoff;
2230                         if ((int)snaplen < 0) {
2231                                 snaplen = 0;
2232                                 do_vnet = false;
2233                         }
2234                 }
2235         } else if (unlikely(macoff + snaplen >
2236                             GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2237                 u32 nval;
2238 
2239                 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2240                 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2241                             snaplen, nval, macoff);
2242                 snaplen = nval;
2243                 if (unlikely((int)snaplen < 0)) {
2244                         snaplen = 0;
2245                         macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2246                         do_vnet = false;
2247                 }
2248         }
2249         spin_lock(&sk->sk_receive_queue.lock);
2250         h.raw = packet_current_rx_frame(po, skb,
2251                                         TP_STATUS_KERNEL, (macoff+snaplen));
2252         if (!h.raw)
2253                 goto drop_n_account;
2254         if (po->tp_version <= TPACKET_V2) {
2255                 packet_increment_rx_head(po, &po->rx_ring);
2256         /*
2257          * LOSING will be reported till you read the stats,
2258          * because it's COR - Clear On Read.
2259          * Anyways, moving it for V1/V2 only as V3 doesn't need this
2260          * at packet level.
2261          */
2262                 if (po->stats.stats1.tp_drops)
2263                         status |= TP_STATUS_LOSING;
2264         }
2265 
2266         if (do_vnet &&
2267             virtio_net_hdr_from_skb(skb, h.raw + macoff -
2268                                     sizeof(struct virtio_net_hdr),
2269                                     vio_le(), true, 0))
2270                 goto drop_n_account;
2271 
2272         po->stats.stats1.tp_packets++;
2273         if (copy_skb) {
2274                 status |= TP_STATUS_COPY;
2275                 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2276         }
2277         spin_unlock(&sk->sk_receive_queue.lock);
2278 
2279         skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2280 
2281         if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2282                 getnstimeofday(&ts);
2283 
2284         status |= ts_status;
2285 
2286         switch (po->tp_version) {
2287         case TPACKET_V1:
2288                 h.h1->tp_len = skb->len;
2289                 h.h1->tp_snaplen = snaplen;
2290                 h.h1->tp_mac = macoff;
2291                 h.h1->tp_net = netoff;
2292                 h.h1->tp_sec = ts.tv_sec;
2293                 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2294                 hdrlen = sizeof(*h.h1);
2295                 break;
2296         case TPACKET_V2:
2297                 h.h2->tp_len = skb->len;
2298                 h.h2->tp_snaplen = snaplen;
2299                 h.h2->tp_mac = macoff;
2300                 h.h2->tp_net = netoff;
2301                 h.h2->tp_sec = ts.tv_sec;
2302                 h.h2->tp_nsec = ts.tv_nsec;
2303                 if (skb_vlan_tag_present(skb)) {
2304                         h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2305                         h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2306                         status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2307                 } else {
2308                         h.h2->tp_vlan_tci = 0;
2309                         h.h2->tp_vlan_tpid = 0;
2310                 }
2311                 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2312                 hdrlen = sizeof(*h.h2);
2313                 break;
2314         case TPACKET_V3:
2315                 /* tp_nxt_offset,vlan are already populated above.
2316                  * So DONT clear those fields here
2317                  */
2318                 h.h3->tp_status |= status;
2319                 h.h3->tp_len = skb->len;
2320                 h.h3->tp_snaplen = snaplen;
2321                 h.h3->tp_mac = macoff;
2322                 h.h3->tp_net = netoff;
2323                 h.h3->tp_sec  = ts.tv_sec;
2324                 h.h3->tp_nsec = ts.tv_nsec;
2325                 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2326                 hdrlen = sizeof(*h.h3);
2327                 break;
2328         default:
2329                 BUG();
2330         }
2331 
2332         sll = h.raw + TPACKET_ALIGN(hdrlen);
2333         sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2334         sll->sll_family = AF_PACKET;
2335         sll->sll_hatype = dev->type;
2336         sll->sll_protocol = skb->protocol;
2337         sll->sll_pkttype = skb->pkt_type;
2338         if (unlikely(po->origdev))
2339                 sll->sll_ifindex = orig_dev->ifindex;
2340         else
2341                 sll->sll_ifindex = dev->ifindex;
2342 
2343         smp_mb();
2344 
2345 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2346         if (po->tp_version <= TPACKET_V2) {
2347                 u8 *start, *end;
2348 
2349                 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2350                                         macoff + snaplen);
2351 
2352                 for (start = h.raw; start < end; start += PAGE_SIZE)
2353                         flush_dcache_page(pgv_to_page(start));
2354         }
2355         smp_wmb();
2356 #endif
2357 
2358         if (po->tp_version <= TPACKET_V2) {
2359                 __packet_set_status(po, h.raw, status);
2360                 sk->sk_data_ready(sk);
2361         } else {
2362                 prb_clear_blk_fill_status(&po->rx_ring);
2363         }
2364 
2365 drop_n_restore:
2366         if (skb_head != skb->data && skb_shared(skb)) {
2367                 skb->data = skb_head;
2368                 skb->len = skb_len;
2369         }
2370 drop:
2371         if (!is_drop_n_account)
2372                 consume_skb(skb);
2373         else
2374                 kfree_skb(skb);
2375         return 0;
2376 
2377 drop_n_account:
2378         is_drop_n_account = true;
2379         po->stats.stats1.tp_drops++;
2380         spin_unlock(&sk->sk_receive_queue.lock);
2381 
2382         sk->sk_data_ready(sk);
2383         kfree_skb(copy_skb);
2384         goto drop_n_restore;
2385 }
2386 
2387 static void tpacket_destruct_skb(struct sk_buff *skb)
2388 {
2389         struct packet_sock *po = pkt_sk(skb->sk);
2390 
2391         if (likely(po->tx_ring.pg_vec)) {
2392                 void *ph;
2393                 __u32 ts;
2394 
2395                 ph = skb_shinfo(skb)->destructor_arg;
2396                 packet_dec_pending(&po->tx_ring);
2397 
2398                 ts = __packet_set_timestamp(po, ph, skb);
2399                 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2400         }
2401 
2402         sock_wfree(skb);
2403 }
2404 
2405 static void tpacket_set_protocol(const struct net_device *dev,
2406                                  struct sk_buff *skb)
2407 {
2408         if (dev->type == ARPHRD_ETHER) {
2409                 skb_reset_mac_header(skb);
2410                 skb->protocol = eth_hdr(skb)->h_proto;
2411         }
2412 }
2413 
2414 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2415 {
2416         if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2417             (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2418              __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2419               __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2420                 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2421                          __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2422                         __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2423 
2424         if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2425                 return -EINVAL;
2426 
2427         return 0;
2428 }
2429 
2430 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2431                                  struct virtio_net_hdr *vnet_hdr)
2432 {
2433         if (*len < sizeof(*vnet_hdr))
2434                 return -EINVAL;
2435         *len -= sizeof(*vnet_hdr);
2436 
2437         if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2438                 return -EFAULT;
2439 
2440         return __packet_snd_vnet_parse(vnet_hdr, *len);
2441 }
2442 
2443 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2444                 void *frame, struct net_device *dev, void *data, int tp_len,
2445                 __be16 proto, unsigned char *addr, int hlen, int copylen,
2446                 const struct sockcm_cookie *sockc)
2447 {
2448         union tpacket_uhdr ph;
2449         int to_write, offset, len, nr_frags, len_max;
2450         struct socket *sock = po->sk.sk_socket;
2451         struct page *page;
2452         int err;
2453 
2454         ph.raw = frame;
2455 
2456         skb->protocol = proto;
2457         skb->dev = dev;
2458         skb->priority = po->sk.sk_priority;
2459         skb->mark = po->sk.sk_mark;
2460         sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2461         skb_shinfo(skb)->destructor_arg = ph.raw;
2462 
2463         skb_reserve(skb, hlen);
2464         skb_reset_network_header(skb);
2465 
2466         to_write = tp_len;
2467 
2468         if (sock->type == SOCK_DGRAM) {
2469                 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2470                                 NULL, tp_len);
2471                 if (unlikely(err < 0))
2472                         return -EINVAL;
2473         } else if (copylen) {
2474                 int hdrlen = min_t(int, copylen, tp_len);
2475 
2476                 skb_push(skb, dev->hard_header_len);
2477                 skb_put(skb, copylen - dev->hard_header_len);
2478                 err = skb_store_bits(skb, 0, data, hdrlen);
2479                 if (unlikely(err))
2480                         return err;
2481                 if (!dev_validate_header(dev, skb->data, hdrlen))
2482                         return -EINVAL;
2483                 if (!skb->protocol)
2484                         tpacket_set_protocol(dev, skb);
2485 
2486                 data += hdrlen;
2487                 to_write -= hdrlen;
2488         }
2489 
2490         offset = offset_in_page(data);
2491         len_max = PAGE_SIZE - offset;
2492         len = ((to_write > len_max) ? len_max : to_write);
2493 
2494         skb->data_len = to_write;
2495         skb->len += to_write;
2496         skb->truesize += to_write;
2497         refcount_add(to_write, &po->sk.sk_wmem_alloc);
2498 
2499         while (likely(to_write)) {
2500                 nr_frags = skb_shinfo(skb)->nr_frags;
2501 
2502                 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2503                         pr_err("Packet exceed the number of skb frags(%lu)\n",
2504                                MAX_SKB_FRAGS);
2505                         return -EFAULT;
2506                 }
2507 
2508                 page = pgv_to_page(data);
2509                 data += len;
2510                 flush_dcache_page(page);
2511                 get_page(page);
2512                 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2513                 to_write -= len;
2514                 offset = 0;
2515                 len_max = PAGE_SIZE;
2516                 len = ((to_write > len_max) ? len_max : to_write);
2517         }
2518 
2519         skb_probe_transport_header(skb, 0);
2520 
2521         return tp_len;
2522 }
2523 
2524 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2525                                 int size_max, void **data)
2526 {
2527         union tpacket_uhdr ph;
2528         int tp_len, off;
2529 
2530         ph.raw = frame;
2531 
2532         switch (po->tp_version) {
2533         case TPACKET_V3:
2534                 if (ph.h3->tp_next_offset != 0) {
2535                         pr_warn_once("variable sized slot not supported");
2536                         return -EINVAL;
2537                 }
2538                 tp_len = ph.h3->tp_len;
2539                 break;
2540         case TPACKET_V2:
2541                 tp_len = ph.h2->tp_len;
2542                 break;
2543         default:
2544                 tp_len = ph.h1->tp_len;
2545                 break;
2546         }
2547         if (unlikely(tp_len > size_max)) {
2548                 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2549                 return -EMSGSIZE;
2550         }
2551 
2552         if (unlikely(po->tp_tx_has_off)) {
2553                 int off_min, off_max;
2554 
2555                 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2556                 off_max = po->tx_ring.frame_size - tp_len;
2557                 if (po->sk.sk_type == SOCK_DGRAM) {
2558                         switch (po->tp_version) {
2559                         case TPACKET_V3:
2560                                 off = ph.h3->tp_net;
2561                                 break;
2562                         case TPACKET_V2:
2563                                 off = ph.h2->tp_net;
2564                                 break;
2565                         default:
2566                                 off = ph.h1->tp_net;
2567                                 break;
2568                         }
2569                 } else {
2570                         switch (po->tp_version) {
2571                         case TPACKET_V3:
2572                                 off = ph.h3->tp_mac;
2573                                 break;
2574                         case TPACKET_V2:
2575                                 off = ph.h2->tp_mac;
2576                                 break;
2577                         default:
2578                                 off = ph.h1->tp_mac;
2579                                 break;
2580                         }
2581                 }
2582                 if (unlikely((off < off_min) || (off_max < off)))
2583                         return -EINVAL;
2584         } else {
2585                 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2586         }
2587 
2588         *data = frame + off;
2589         return tp_len;
2590 }
2591 
2592 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2593 {
2594         struct sk_buff *skb;
2595         struct net_device *dev;
2596         struct virtio_net_hdr *vnet_hdr = NULL;
2597         struct sockcm_cookie sockc;
2598         __be16 proto;
2599         int err, reserve = 0;
2600         void *ph;
2601         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2602         bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2603         int tp_len, size_max;
2604         unsigned char *addr;
2605         void *data;
2606         int len_sum = 0;
2607         int status = TP_STATUS_AVAILABLE;
2608         int hlen, tlen, copylen = 0;
2609 
2610         mutex_lock(&po->pg_vec_lock);
2611 
2612         if (likely(saddr == NULL)) {
2613                 dev     = packet_cached_dev_get(po);
2614                 proto   = po->num;
2615                 addr    = NULL;
2616         } else {
2617                 err = -EINVAL;
2618                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2619                         goto out;
2620                 if (msg->msg_namelen < (saddr->sll_halen
2621                                         + offsetof(struct sockaddr_ll,
2622                                                 sll_addr)))
2623                         goto out;
2624                 proto   = saddr->sll_protocol;
2625                 addr    = saddr->sll_addr;
2626                 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2627         }
2628 
2629         err = -ENXIO;
2630         if (unlikely(dev == NULL))
2631                 goto out;
2632         err = -ENETDOWN;
2633         if (unlikely(!(dev->flags & IFF_UP)))
2634                 goto out_put;
2635 
2636         sockc.tsflags = po->sk.sk_tsflags;
2637         if (msg->msg_controllen) {
2638                 err = sock_cmsg_send(&po->sk, msg, &sockc);
2639                 if (unlikely(err))
2640                         goto out_put;
2641         }
2642 
2643         if (po->sk.sk_socket->type == SOCK_RAW)
2644                 reserve = dev->hard_header_len;
2645         size_max = po->tx_ring.frame_size
2646                 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2647 
2648         if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2649                 size_max = dev->mtu + reserve + VLAN_HLEN;
2650 
2651         do {
2652                 ph = packet_current_frame(po, &po->tx_ring,
2653                                           TP_STATUS_SEND_REQUEST);
2654                 if (unlikely(ph == NULL)) {
2655                         if (need_wait && need_resched())
2656                                 schedule();
2657                         continue;
2658                 }
2659 
2660                 skb = NULL;
2661                 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2662                 if (tp_len < 0)
2663                         goto tpacket_error;
2664 
2665                 status = TP_STATUS_SEND_REQUEST;
2666                 hlen = LL_RESERVED_SPACE(dev);
2667                 tlen = dev->needed_tailroom;
2668                 if (po->has_vnet_hdr) {
2669                         vnet_hdr = data;
2670                         data += sizeof(*vnet_hdr);
2671                         tp_len -= sizeof(*vnet_hdr);
2672                         if (tp_len < 0 ||
2673                             __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2674                                 tp_len = -EINVAL;
2675                                 goto tpacket_error;
2676                         }
2677                         copylen = __virtio16_to_cpu(vio_le(),
2678                                                     vnet_hdr->hdr_len);
2679                 }
2680                 copylen = max_t(int, copylen, dev->hard_header_len);
2681                 skb = sock_alloc_send_skb(&po->sk,
2682                                 hlen + tlen + sizeof(struct sockaddr_ll) +
2683                                 (copylen - dev->hard_header_len),
2684                                 !need_wait, &err);
2685 
2686                 if (unlikely(skb == NULL)) {
2687                         /* we assume the socket was initially writeable ... */
2688                         if (likely(len_sum > 0))
2689                                 err = len_sum;
2690                         goto out_status;
2691                 }
2692                 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2693                                           addr, hlen, copylen, &sockc);
2694                 if (likely(tp_len >= 0) &&
2695                     tp_len > dev->mtu + reserve &&
2696                     !po->has_vnet_hdr &&
2697                     !packet_extra_vlan_len_allowed(dev, skb))
2698                         tp_len = -EMSGSIZE;
2699 
2700                 if (unlikely(tp_len < 0)) {
2701 tpacket_error:
2702                         if (po->tp_loss) {
2703                                 __packet_set_status(po, ph,
2704                                                 TP_STATUS_AVAILABLE);
2705                                 packet_increment_head(&po->tx_ring);
2706                                 kfree_skb(skb);
2707                                 continue;
2708                         } else {
2709                                 status = TP_STATUS_WRONG_FORMAT;
2710                                 err = tp_len;
2711                                 goto out_status;
2712                         }
2713                 }
2714 
2715                 if (po->has_vnet_hdr) {
2716                         if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2717                                 tp_len = -EINVAL;
2718                                 goto tpacket_error;
2719                         }
2720                         virtio_net_hdr_set_proto(skb, vnet_hdr);
2721                 }
2722 
2723                 skb->destructor = tpacket_destruct_skb;
2724                 __packet_set_status(po, ph, TP_STATUS_SENDING);
2725                 packet_inc_pending(&po->tx_ring);
2726 
2727                 status = TP_STATUS_SEND_REQUEST;
2728                 err = po->xmit(skb);
2729                 if (unlikely(err > 0)) {
2730                         err = net_xmit_errno(err);
2731                         if (err && __packet_get_status(po, ph) ==
2732                                    TP_STATUS_AVAILABLE) {
2733                                 /* skb was destructed already */
2734                                 skb = NULL;
2735                                 goto out_status;
2736                         }
2737                         /*
2738                          * skb was dropped but not destructed yet;
2739                          * let's treat it like congestion or err < 0
2740                          */
2741                         err = 0;
2742                 }
2743                 packet_increment_head(&po->tx_ring);
2744                 len_sum += tp_len;
2745         } while (likely((ph != NULL) ||
2746                 /* Note: packet_read_pending() might be slow if we have
2747                  * to call it as it's per_cpu variable, but in fast-path
2748                  * we already short-circuit the loop with the first
2749                  * condition, and luckily don't have to go that path
2750                  * anyway.
2751                  */
2752                  (need_wait && packet_read_pending(&po->tx_ring))));
2753 
2754         err = len_sum;
2755         goto out_put;
2756 
2757 out_status:
2758         __packet_set_status(po, ph, status);
2759         kfree_skb(skb);
2760 out_put:
2761         dev_put(dev);
2762 out:
2763         mutex_unlock(&po->pg_vec_lock);
2764         return err;
2765 }
2766 
2767 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2768                                         size_t reserve, size_t len,
2769                                         size_t linear, int noblock,
2770                                         int *err)
2771 {
2772         struct sk_buff *skb;
2773 
2774         /* Under a page?  Don't bother with paged skb. */
2775         if (prepad + len < PAGE_SIZE || !linear)
2776                 linear = len;
2777 
2778         skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2779                                    err, 0);
2780         if (!skb)
2781                 return NULL;
2782 
2783         skb_reserve(skb, reserve);
2784         skb_put(skb, linear);
2785         skb->data_len = len - linear;
2786         skb->len += len - linear;
2787 
2788         return skb;
2789 }
2790 
2791 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2792 {
2793         struct sock *sk = sock->sk;
2794         DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2795         struct sk_buff *skb;
2796         struct net_device *dev;
2797         __be16 proto;
2798         unsigned char *addr;
2799         int err, reserve = 0;
2800         struct sockcm_cookie sockc;
2801         struct virtio_net_hdr vnet_hdr = { 0 };
2802         int offset = 0;
2803         struct packet_sock *po = pkt_sk(sk);
2804         bool has_vnet_hdr = false;
2805         int hlen, tlen, linear;
2806         int extra_len = 0;
2807 
2808         /*
2809          *      Get and verify the address.
2810          */
2811 
2812         if (likely(saddr == NULL)) {
2813                 dev     = packet_cached_dev_get(po);
2814                 proto   = po->num;
2815                 addr    = NULL;
2816         } else {
2817                 err = -EINVAL;
2818                 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2819                         goto out;
2820                 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2821                         goto out;
2822                 proto   = saddr->sll_protocol;
2823                 addr    = saddr->sll_addr;
2824                 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2825         }
2826 
2827         err = -ENXIO;
2828         if (unlikely(dev == NULL))
2829                 goto out_unlock;
2830         err = -ENETDOWN;
2831         if (unlikely(!(dev->flags & IFF_UP)))
2832                 goto out_unlock;
2833 
2834         sockc.tsflags = sk->sk_tsflags;
2835         sockc.mark = sk->sk_mark;
2836         if (msg->msg_controllen) {
2837                 err = sock_cmsg_send(sk, msg, &sockc);
2838                 if (unlikely(err))
2839                         goto out_unlock;
2840         }
2841 
2842         if (sock->type == SOCK_RAW)
2843                 reserve = dev->hard_header_len;
2844         if (po->has_vnet_hdr) {
2845                 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2846                 if (err)
2847                         goto out_unlock;
2848                 has_vnet_hdr = true;
2849         }
2850 
2851         if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2852                 if (!netif_supports_nofcs(dev)) {
2853                         err = -EPROTONOSUPPORT;
2854                         goto out_unlock;
2855                 }
2856                 extra_len = 4; /* We're doing our own CRC */
2857         }
2858 
2859         err = -EMSGSIZE;
2860         if (!vnet_hdr.gso_type &&
2861             (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2862                 goto out_unlock;
2863 
2864         err = -ENOBUFS;
2865         hlen = LL_RESERVED_SPACE(dev);
2866         tlen = dev->needed_tailroom;
2867         linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2868         linear = max(linear, min_t(int, len, dev->hard_header_len));
2869         skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2870                                msg->msg_flags & MSG_DONTWAIT, &err);
2871         if (skb == NULL)
2872                 goto out_unlock;
2873 
2874         skb_reset_network_header(skb);
2875 
2876         err = -EINVAL;
2877         if (sock->type == SOCK_DGRAM) {
2878                 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2879                 if (unlikely(offset < 0))
2880                         goto out_free;
2881         } else if (reserve) {
2882                 skb_reserve(skb, -reserve);
2883                 if (len < reserve)
2884                         skb_reset_network_header(skb);
2885         }
2886 
2887         /* Returns -EFAULT on error */
2888         err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2889         if (err)
2890                 goto out_free;
2891 
2892         if (sock->type == SOCK_RAW &&
2893             !dev_validate_header(dev, skb->data, len)) {
2894                 err = -EINVAL;
2895                 goto out_free;
2896         }
2897 
2898         sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2899 
2900         if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2901             !packet_extra_vlan_len_allowed(dev, skb)) {
2902                 err = -EMSGSIZE;
2903                 goto out_free;
2904         }
2905 
2906         skb->protocol = proto;
2907         skb->dev = dev;
2908         skb->priority = sk->sk_priority;
2909         skb->mark = sockc.mark;
2910 
2911         if (has_vnet_hdr) {
2912                 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2913                 if (err)
2914                         goto out_free;
2915                 len += sizeof(vnet_hdr);
2916                 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2917         }
2918 
2919         skb_probe_transport_header(skb, reserve);
2920 
2921         if (unlikely(extra_len == 4))
2922                 skb->no_fcs = 1;
2923 
2924         err = po->xmit(skb);
2925         if (err > 0 && (err = net_xmit_errno(err)) != 0)
2926                 goto out_unlock;
2927 
2928         dev_put(dev);
2929 
2930         return len;
2931 
2932 out_free:
2933         kfree_skb(skb);
2934 out_unlock:
2935         if (dev)
2936                 dev_put(dev);
2937 out:
2938         return err;
2939 }
2940 
2941 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2942 {
2943         struct sock *sk = sock->sk;
2944         struct packet_sock *po = pkt_sk(sk);
2945 
2946         if (po->tx_ring.pg_vec)
2947                 return tpacket_snd(po, msg);
2948         else
2949                 return packet_snd(sock, msg, len);
2950 }
2951 
2952 /*
2953  *      Close a PACKET socket. This is fairly simple. We immediately go
2954  *      to 'closed' state and remove our protocol entry in the device list.
2955  */
2956 
2957 static int packet_release(struct socket *sock)
2958 {
2959         struct sock *sk = sock->sk;
2960         struct packet_sock *po;
2961         struct packet_fanout *f;
2962         struct net *net;
2963         union tpacket_req_u req_u;
2964 
2965         if (!sk)
2966                 return 0;
2967 
2968         net = sock_net(sk);
2969         po = pkt_sk(sk);
2970 
2971         mutex_lock(&net->packet.sklist_lock);
2972         sk_del_node_init_rcu(sk);
2973         mutex_unlock(&net->packet.sklist_lock);
2974 
2975         preempt_disable();
2976         sock_prot_inuse_add(net, sk->sk_prot, -1);
2977         preempt_enable();
2978 
2979         spin_lock(&po->bind_lock);
2980         unregister_prot_hook(sk, false);
2981         packet_cached_dev_reset(po);
2982 
2983         if (po->prot_hook.dev) {
2984                 dev_put(po->prot_hook.dev);
2985                 po->prot_hook.dev = NULL;
2986         }
2987         spin_unlock(&po->bind_lock);
2988 
2989         packet_flush_mclist(sk);
2990 
2991         lock_sock(sk);
2992         if (po->rx_ring.pg_vec) {
2993                 memset(&req_u, 0, sizeof(req_u));
2994                 packet_set_ring(sk, &req_u, 1, 0);
2995         }
2996 
2997         if (po->tx_ring.pg_vec) {
2998                 memset(&req_u, 0, sizeof(req_u));
2999                 packet_set_ring(sk, &req_u, 1, 1);
3000         }
3001         release_sock(sk);
3002 
3003         f = fanout_release(sk);
3004 
3005         synchronize_net();
3006 
3007         if (f) {
3008                 kfree(po->rollover);
3009                 fanout_release_data(f);
3010                 kfree(f);
3011         }
3012         /*
3013          *      Now the socket is dead. No more input will appear.
3014          */
3015         sock_orphan(sk);
3016         sock->sk = NULL;
3017 
3018         /* Purge queues */
3019 
3020         skb_queue_purge(&sk->sk_receive_queue);
3021         packet_free_pending(po);
3022         sk_refcnt_debug_release(sk);
3023 
3024         sock_put(sk);
3025         return 0;
3026 }
3027 
3028 /*
3029  *      Attach a packet hook.
3030  */
3031 
3032 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3033                           __be16 proto)
3034 {
3035         struct packet_sock *po = pkt_sk(sk);
3036         struct net_device *dev_curr;
3037         __be16 proto_curr;
3038         bool need_rehook;
3039         struct net_device *dev = NULL;
3040         int ret = 0;
3041         bool unlisted = false;
3042 
3043         lock_sock(sk);
3044         spin_lock(&po->bind_lock);
3045         rcu_read_lock();
3046 
3047         if (po->fanout) {
3048                 ret = -EINVAL;
3049                 goto out_unlock;
3050         }
3051 
3052         if (name) {
3053                 dev = dev_get_by_name_rcu(sock_net(sk), name);
3054                 if (!dev) {
3055                         ret = -ENODEV;
3056                         goto out_unlock;
3057                 }
3058         } else if (ifindex) {
3059                 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3060                 if (!dev) {
3061                         ret = -ENODEV;
3062                         goto out_unlock;
3063                 }
3064         }
3065 
3066         if (dev)
3067                 dev_hold(dev);
3068 
3069         proto_curr = po->prot_hook.type;
3070         dev_curr = po->prot_hook.dev;
3071 
3072         need_rehook = proto_curr != proto || dev_curr != dev;
3073 
3074         if (need_rehook) {
3075                 if (po->running) {
3076                         rcu_read_unlock();
3077                         /* prevents packet_notifier() from calling
3078                          * register_prot_hook()
3079                          */
3080                         po->num = 0;
3081                         __unregister_prot_hook(sk, true);
3082                         rcu_read_lock();
3083                         dev_curr = po->prot_hook.dev;
3084                         if (dev)
3085                                 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3086                                                                  dev->ifindex);
3087                 }
3088 
3089                 BUG_ON(po->running);
3090                 po->num = proto;
3091                 po->prot_hook.type = proto;
3092 
3093                 if (unlikely(unlisted)) {
3094                         dev_put(dev);
3095                         po->prot_hook.dev = NULL;
3096                         po->ifindex = -1;
3097                         packet_cached_dev_reset(po);
3098                 } else {
3099                         po->prot_hook.dev = dev;
3100                         po->ifindex = dev ? dev->ifindex : 0;
3101                         packet_cached_dev_assign(po, dev);
3102                 }
3103         }
3104         if (dev_curr)
3105                 dev_put(dev_curr);
3106 
3107         if (proto == 0 || !need_rehook)
3108                 goto out_unlock;
3109 
3110         if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3111                 register_prot_hook(sk);
3112         } else {
3113                 sk->sk_err = ENETDOWN;
3114                 if (!sock_flag(sk, SOCK_DEAD))
3115                         sk->sk_error_report(sk);
3116         }
3117 
3118 out_unlock:
3119         rcu_read_unlock();
3120         spin_unlock(&po->bind_lock);
3121         release_sock(sk);
3122         return ret;
3123 }
3124 
3125 /*
3126  *      Bind a packet socket to a device
3127  */
3128 
3129 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3130                             int addr_len)
3131 {
3132         struct sock *sk = sock->sk;
3133         char name[sizeof(uaddr->sa_data) + 1];
3134 
3135         /*
3136          *      Check legality
3137          */
3138 
3139         if (addr_len != sizeof(struct sockaddr))
3140                 return -EINVAL;
3141         /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3142          * zero-terminated.
3143          */
3144         memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3145         name[sizeof(uaddr->sa_data)] = 0;
3146 
3147         return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3148 }
3149 
3150 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3151 {
3152         struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3153         struct sock *sk = sock->sk;
3154 
3155         /*
3156          *      Check legality
3157          */
3158 
3159         if (addr_len < sizeof(struct sockaddr_ll))
3160                 return -EINVAL;
3161         if (sll->sll_family != AF_PACKET)
3162                 return -EINVAL;
3163 
3164         return packet_do_bind(sk, NULL, sll->sll_ifindex,
3165                               sll->sll_protocol ? : pkt_sk(sk)->num);
3166 }
3167 
3168 static struct proto packet_proto = {
3169         .name     = "PACKET",
3170         .owner    = THIS_MODULE,
3171         .obj_size = sizeof(struct packet_sock),
3172 };
3173 
3174 /*
3175  *      Create a packet of type SOCK_PACKET.
3176  */
3177 
3178 static int packet_create(struct net *net, struct socket *sock, int protocol,
3179                          int kern)
3180 {
3181         struct sock *sk;
3182         struct packet_sock *po;
3183         __be16 proto = (__force __be16)protocol; /* weird, but documented */
3184         int err;
3185 
3186         if (!ns_capable(net->user_ns, CAP_NET_RAW))
3187                 return -EPERM;
3188         if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3189             sock->type != SOCK_PACKET)
3190                 return -ESOCKTNOSUPPORT;
3191 
3192         sock->state = SS_UNCONNECTED;
3193 
3194         err = -ENOBUFS;
3195         sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3196         if (sk == NULL)
3197                 goto out;
3198 
3199         sock->ops = &packet_ops;
3200         if (sock->type == SOCK_PACKET)
3201                 sock->ops = &packet_ops_spkt;
3202 
3203         sock_init_data(sock, sk);
3204 
3205         po = pkt_sk(sk);
3206         sk->sk_family = PF_PACKET;
3207         po->num = proto;
3208         po->xmit = dev_queue_xmit;
3209 
3210         err = packet_alloc_pending(po);
3211         if (err)
3212                 goto out2;
3213 
3214         packet_cached_dev_reset(po);
3215 
3216         sk->sk_destruct = packet_sock_destruct;
3217         sk_refcnt_debug_inc(sk);
3218 
3219         /*
3220          *      Attach a protocol block
3221          */
3222 
3223         spin_lock_init(&po->bind_lock);
3224         mutex_init(&po->pg_vec_lock);
3225         po->rollover = NULL;
3226         po->prot_hook.func = packet_rcv;
3227 
3228         if (sock->type == SOCK_PACKET)
3229                 po->prot_hook.func = packet_rcv_spkt;
3230 
3231         po->prot_hook.af_packet_priv = sk;
3232 
3233         if (proto) {
3234                 po->prot_hook.type = proto;
3235                 __register_prot_hook(sk);
3236         }
3237 
3238         mutex_lock(&net->packet.sklist_lock);
3239         sk_add_node_rcu(sk, &net->packet.sklist);
3240         mutex_unlock(&net->packet.sklist_lock);
3241 
3242         preempt_disable();
3243         sock_prot_inuse_add(net, &packet_proto, 1);
3244         preempt_enable();
3245 
3246         return 0;
3247 out2:
3248         sk_free(sk);
3249 out:
3250         return err;
3251 }
3252 
3253 /*
3254  *      Pull a packet from our receive queue and hand it to the user.
3255  *      If necessary we block.
3256  */
3257 
3258 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3259                           int flags)
3260 {
3261         struct sock *sk = sock->sk;
3262         struct sk_buff *skb;
3263         int copied, err;
3264         int vnet_hdr_len = 0;
3265         unsigned int origlen = 0;
3266 
3267         err = -EINVAL;
3268         if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3269                 goto out;
3270 
3271 #if 0
3272         /* What error should we return now? EUNATTACH? */
3273         if (pkt_sk(sk)->ifindex < 0)
3274                 return -ENODEV;
3275 #endif
3276 
3277         if (flags & MSG_ERRQUEUE) {
3278                 err = sock_recv_errqueue(sk, msg, len,
3279                                          SOL_PACKET, PACKET_TX_TIMESTAMP);
3280                 goto out;
3281         }
3282 
3283         /*
3284          *      Call the generic datagram receiver. This handles all sorts
3285          *      of horrible races and re-entrancy so we can forget about it
3286          *      in the protocol layers.
3287          *
3288          *      Now it will return ENETDOWN, if device have just gone down,
3289          *      but then it will block.
3290          */
3291 
3292         skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3293 
3294         /*
3295          *      An error occurred so return it. Because skb_recv_datagram()
3296          *      handles the blocking we don't see and worry about blocking
3297          *      retries.
3298          */
3299 
3300         if (skb == NULL)
3301                 goto out;
3302 
3303         if (pkt_sk(sk)->pressure)
3304                 packet_rcv_has_room(pkt_sk(sk), NULL);
3305 
3306         if (pkt_sk(sk)->has_vnet_hdr) {
3307                 err = packet_rcv_vnet(msg, skb, &len);
3308                 if (err)
3309                         goto out_free;
3310                 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3311         }
3312 
3313         /* You lose any data beyond the buffer you gave. If it worries
3314          * a user program they can ask the device for its MTU
3315          * anyway.
3316          */
3317         copied = skb->len;
3318         if (copied > len) {
3319                 copied = len;
3320                 msg->msg_flags |= MSG_TRUNC;
3321         }
3322 
3323         err = skb_copy_datagram_msg(skb, 0, msg, copied);
3324         if (err)
3325                 goto out_free;
3326 
3327         if (sock->type != SOCK_PACKET) {
3328                 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3329 
3330                 /* Original length was stored in sockaddr_ll fields */
3331                 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3332                 sll->sll_family = AF_PACKET;
3333                 sll->sll_protocol = skb->protocol;
3334         }
3335 
3336         sock_recv_ts_and_drops(msg, sk, skb);
3337 
3338         if (msg->msg_name) {
3339                 /* If the address length field is there to be filled
3340                  * in, we fill it in now.
3341                  */
3342                 if (sock->type == SOCK_PACKET) {
3343                         __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3344                         msg->msg_namelen = sizeof(struct sockaddr_pkt);
3345                 } else {
3346                         struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3347 
3348                         msg->msg_namelen = sll->sll_halen +
3349                                 offsetof(struct sockaddr_ll, sll_addr);
3350                 }
3351                 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3352                        msg->msg_namelen);
3353         }
3354 
3355         if (pkt_sk(sk)->auxdata) {
3356                 struct tpacket_auxdata aux;
3357 
3358                 aux.tp_status = TP_STATUS_USER;
3359                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3360                         aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3361                 else if (skb->pkt_type != PACKET_OUTGOING &&
3362                          (skb->ip_summed == CHECKSUM_COMPLETE ||
3363                           skb_csum_unnecessary(skb)))
3364                         aux.tp_status |= TP_STATUS_CSUM_VALID;
3365 
3366                 aux.tp_len = origlen;
3367                 aux.tp_snaplen = skb->len;
3368                 aux.tp_mac = 0;
3369                 aux.tp_net = skb_network_offset(skb);
3370                 if (skb_vlan_tag_present(skb)) {
3371                         aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3372                         aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3373                         aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3374                 } else {
3375                         aux.tp_vlan_tci = 0;
3376                         aux.tp_vlan_tpid = 0;
3377                 }
3378                 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3379         }
3380 
3381         /*
3382          *      Free or return the buffer as appropriate. Again this
3383          *      hides all the races and re-entrancy issues from us.
3384          */
3385         err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3386 
3387 out_free:
3388         skb_free_datagram(sk, skb);
3389 out:
3390         return err;
3391 }
3392 
3393 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3394                                int peer)
3395 {
3396         struct net_device *dev;
3397         struct sock *sk = sock->sk;
3398 
3399         if (peer)
3400                 return -EOPNOTSUPP;
3401 
3402         uaddr->sa_family = AF_PACKET;
3403         memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3404         rcu_read_lock();
3405         dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3406         if (dev)
3407                 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3408         rcu_read_unlock();
3409 
3410         return sizeof(*uaddr);
3411 }
3412 
3413 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3414                           int peer)
3415 {
3416         struct net_device *dev;
3417         struct sock *sk = sock->sk;
3418         struct packet_sock *po = pkt_sk(sk);
3419         DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3420 
3421         if (peer)
3422                 return -EOPNOTSUPP;
3423 
3424         sll->sll_family = AF_PACKET;
3425         sll->sll_ifindex = po->ifindex;
3426         sll->sll_protocol = po->num;
3427         sll->sll_pkttype = 0;
3428         rcu_read_lock();
3429         dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3430         if (dev) {
3431                 sll->sll_hatype = dev->type;
3432                 sll->sll_halen = dev->addr_len;
3433                 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3434         } else {
3435                 sll->sll_hatype = 0;    /* Bad: we have no ARPHRD_UNSPEC */
3436                 sll->sll_halen = 0;
3437         }
3438         rcu_read_unlock();
3439 
3440         return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3441 }
3442 
3443 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3444                          int what)
3445 {
3446         switch (i->type) {
3447         case PACKET_MR_MULTICAST:
3448                 if (i->alen != dev->addr_len)
3449                         return -EINVAL;
3450                 if (what > 0)
3451                         return dev_mc_add(dev, i->addr);
3452                 else
3453                         return dev_mc_del(dev, i->addr);
3454                 break;
3455         case PACKET_MR_PROMISC:
3456                 return dev_set_promiscuity(dev, what);
3457         case PACKET_MR_ALLMULTI:
3458                 return dev_set_allmulti(dev, what);
3459         case PACKET_MR_UNICAST:
3460                 if (i->alen != dev->addr_len)
3461                         return -EINVAL;
3462                 if (what > 0)
3463                         return dev_uc_add(dev, i->addr);
3464                 else
3465                         return dev_uc_del(dev, i->addr);
3466                 break;
3467         default:
3468                 break;
3469         }
3470         return 0;
3471 }
3472 
3473 static void packet_dev_mclist_delete(struct net_device *dev,
3474                                      struct packet_mclist **mlp)
3475 {
3476         struct packet_mclist *ml;
3477 
3478         while ((ml = *mlp) != NULL) {
3479                 if (ml->ifindex == dev->ifindex) {
3480                         packet_dev_mc(dev, ml, -1);
3481                         *mlp = ml->next;
3482                         kfree(ml);
3483                 } else
3484                         mlp = &ml->next;
3485         }
3486 }
3487 
3488 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3489 {
3490         struct packet_sock *po = pkt_sk(sk);
3491         struct packet_mclist *ml, *i;
3492         struct net_device *dev;
3493         int err;
3494 
3495         rtnl_lock();
3496 
3497         err = -ENODEV;
3498         dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3499         if (!dev)
3500                 goto done;
3501 
3502         err = -EINVAL;
3503         if (mreq->mr_alen > dev->addr_len)
3504                 goto done;
3505 
3506         err = -ENOBUFS;
3507         i = kmalloc(sizeof(*i), GFP_KERNEL);
3508         if (i == NULL)
3509                 goto done;
3510 
3511         err = 0;
3512         for (ml = po->mclist; ml; ml = ml->next) {
3513                 if (ml->ifindex == mreq->mr_ifindex &&
3514                     ml->type == mreq->mr_type &&
3515                     ml->alen == mreq->mr_alen &&
3516                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3517                         ml->count++;
3518                         /* Free the new element ... */
3519                         kfree(i);
3520                         goto done;
3521                 }
3522         }
3523 
3524         i->type = mreq->mr_type;
3525         i->ifindex = mreq->mr_ifindex;
3526         i->alen = mreq->mr_alen;
3527         memcpy(i->addr, mreq->mr_address, i->alen);
3528         memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3529         i->count = 1;
3530         i->next = po->mclist;
3531         po->mclist = i;
3532         err = packet_dev_mc(dev, i, 1);
3533         if (err) {
3534                 po->mclist = i->next;
3535                 kfree(i);
3536         }
3537 
3538 done:
3539         rtnl_unlock();
3540         return err;
3541 }
3542 
3543 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3544 {
3545         struct packet_mclist *ml, **mlp;
3546 
3547         rtnl_lock();
3548 
3549         for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3550                 if (ml->ifindex == mreq->mr_ifindex &&
3551                     ml->type == mreq->mr_type &&
3552                     ml->alen == mreq->mr_alen &&
3553                     memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3554                         if (--ml->count == 0) {
3555                                 struct net_device *dev;
3556                                 *mlp = ml->next;
3557                                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3558                                 if (dev)
3559                                         packet_dev_mc(dev, ml, -1);
3560                                 kfree(ml);
3561                         }
3562                         break;
3563                 }
3564         }
3565         rtnl_unlock();
3566         return 0;
3567 }
3568 
3569 static void packet_flush_mclist(struct sock *sk)
3570 {
3571         struct packet_sock *po = pkt_sk(sk);
3572         struct packet_mclist *ml;
3573 
3574         if (!po->mclist)
3575                 return;
3576 
3577         rtnl_lock();
3578         while ((ml = po->mclist) != NULL) {
3579                 struct net_device *dev;
3580 
3581                 po->mclist = ml->next;
3582                 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3583                 if (dev != NULL)
3584                         packet_dev_mc(dev, ml, -1);
3585                 kfree(ml);
3586         }
3587         rtnl_unlock();
3588 }
3589 
3590 static int
3591 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3592 {
3593         struct sock *sk = sock->sk;
3594         struct packet_sock *po = pkt_sk(sk);
3595         int ret;
3596 
3597         if (level != SOL_PACKET)
3598                 return -ENOPROTOOPT;
3599 
3600         switch (optname) {
3601         case PACKET_ADD_MEMBERSHIP:
3602         case PACKET_DROP_MEMBERSHIP:
3603         {
3604                 struct packet_mreq_max mreq;
3605                 int len = optlen;
3606                 memset(&mreq, 0, sizeof(mreq));
3607                 if (len < sizeof(struct packet_mreq))
3608                         return -EINVAL;
3609                 if (len > sizeof(mreq))
3610                         len = sizeof(mreq);
3611                 if (copy_from_user(&mreq, optval, len))
3612                         return -EFAULT;
3613                 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3614                         return -EINVAL;
3615                 if (optname == PACKET_ADD_MEMBERSHIP)
3616                         ret = packet_mc_add(sk, &mreq);
3617                 else
3618                         ret = packet_mc_drop(sk, &mreq);
3619                 return ret;
3620         }
3621 
3622         case PACKET_RX_RING:
3623         case PACKET_TX_RING:
3624         {
3625                 union tpacket_req_u req_u;
3626                 int len;
3627 
3628                 lock_sock(sk);
3629                 switch (po->tp_version) {
3630                 case TPACKET_V1:
3631                 case TPACKET_V2:
3632                         len = sizeof(req_u.req);
3633                         break;
3634                 case TPACKET_V3:
3635                 default:
3636                         len = sizeof(req_u.req3);
3637                         break;
3638                 }
3639                 if (optlen < len) {
3640                         ret = -EINVAL;
3641                 } else {
3642                         if (copy_from_user(&req_u.req, optval, len))
3643                                 ret = -EFAULT;
3644                         else
3645                                 ret = packet_set_ring(sk, &req_u, 0,
3646                                                     optname == PACKET_TX_RING);
3647                 }
3648                 release_sock(sk);
3649                 return ret;
3650         }
3651         case PACKET_COPY_THRESH:
3652         {
3653                 int val;
3654 
3655                 if (optlen != sizeof(val))
3656                         return -EINVAL;
3657                 if (copy_from_user(&val, optval, sizeof(val)))
3658                         return -EFAULT;
3659 
3660                 pkt_sk(sk)->copy_thresh = val;
3661                 return 0;
3662         }
3663         case PACKET_VERSION:
3664         {
3665                 int val;
3666 
3667                 if (optlen != sizeof(val))
3668                         return -EINVAL;
3669                 if (copy_from_user(&val, optval, sizeof(val)))
3670                         return -EFAULT;
3671                 switch (val) {
3672                 case TPACKET_V1:
3673                 case TPACKET_V2:
3674                 case TPACKET_V3:
3675                         break;
3676                 default:
3677                         return -EINVAL;
3678                 }
3679                 lock_sock(sk);
3680                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3681                         ret = -EBUSY;
3682                 } else {
3683                         po->tp_version = val;
3684                         ret = 0;
3685                 }
3686                 release_sock(sk);
3687                 return ret;
3688         }
3689         case PACKET_RESERVE:
3690         {
3691                 unsigned int val;
3692 
3693                 if (optlen != sizeof(val))
3694                         return -EINVAL;
3695                 if (copy_from_user(&val, optval, sizeof(val)))
3696                         return -EFAULT;
3697                 if (val > INT_MAX)
3698                         return -EINVAL;
3699                 lock_sock(sk);
3700                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3701                         ret = -EBUSY;
3702                 } else {
3703                         po->tp_reserve = val;
3704                         ret = 0;
3705                 }
3706                 release_sock(sk);
3707                 return ret;
3708         }
3709         case PACKET_LOSS:
3710         {
3711                 unsigned int val;
3712 
3713                 if (optlen != sizeof(val))
3714                         return -EINVAL;
3715                 if (copy_from_user(&val, optval, sizeof(val)))
3716                         return -EFAULT;
3717 
3718                 lock_sock(sk);
3719                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3720                         ret = -EBUSY;
3721                 } else {
3722                         po->tp_loss = !!val;
3723                         ret = 0;
3724                 }
3725                 release_sock(sk);
3726                 return ret;
3727         }
3728         case PACKET_AUXDATA:
3729         {
3730                 int val;
3731 
3732                 if (optlen < sizeof(val))
3733                         return -EINVAL;
3734                 if (copy_from_user(&val, optval, sizeof(val)))
3735                         return -EFAULT;
3736 
3737                 lock_sock(sk);
3738                 po->auxdata = !!val;
3739                 release_sock(sk);
3740                 return 0;
3741         }
3742         case PACKET_ORIGDEV:
3743         {
3744                 int val;
3745 
3746                 if (optlen < sizeof(val))
3747                         return -EINVAL;
3748                 if (copy_from_user(&val, optval, sizeof(val)))
3749                         return -EFAULT;
3750 
3751                 lock_sock(sk);
3752                 po->origdev = !!val;
3753                 release_sock(sk);
3754                 return 0;
3755         }
3756         case PACKET_VNET_HDR:
3757         {
3758                 int val;
3759 
3760                 if (sock->type != SOCK_RAW)
3761                         return -EINVAL;
3762                 if (optlen < sizeof(val))
3763                         return -EINVAL;
3764                 if (copy_from_user(&val, optval, sizeof(val)))
3765                         return -EFAULT;
3766 
3767                 lock_sock(sk);
3768                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3769                         ret = -EBUSY;
3770                 } else {
3771                         po->has_vnet_hdr = !!val;
3772                         ret = 0;
3773                 }
3774                 release_sock(sk);
3775                 return ret;
3776         }
3777         case PACKET_TIMESTAMP:
3778         {
3779                 int val;
3780 
3781                 if (optlen != sizeof(val))
3782                         return -EINVAL;
3783                 if (copy_from_user(&val, optval, sizeof(val)))
3784                         return -EFAULT;
3785 
3786                 po->tp_tstamp = val;
3787                 return 0;
3788         }
3789         case PACKET_FANOUT:
3790         {
3791                 int val;
3792 
3793                 if (optlen != sizeof(val))
3794                         return -EINVAL;
3795                 if (copy_from_user(&val, optval, sizeof(val)))
3796                         return -EFAULT;
3797 
3798                 return fanout_add(sk, val & 0xffff, val >> 16);
3799         }
3800         case PACKET_FANOUT_DATA:
3801         {
3802                 if (!po->fanout)
3803                         return -EINVAL;
3804 
3805                 return fanout_set_data(po, optval, optlen);
3806         }
3807         case PACKET_TX_HAS_OFF:
3808         {
3809                 unsigned int val;
3810 
3811                 if (optlen != sizeof(val))
3812                         return -EINVAL;
3813                 if (copy_from_user(&val, optval, sizeof(val)))
3814                         return -EFAULT;
3815 
3816                 lock_sock(sk);
3817                 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3818                         ret = -EBUSY;
3819                 } else {
3820                         po->tp_tx_has_off = !!val;
3821                         ret = 0;
3822                 }
3823                 release_sock(sk);
3824                 return 0;
3825         }
3826         case PACKET_QDISC_BYPASS:
3827         {
3828                 int val;
3829 
3830                 if (optlen != sizeof(val))
3831                         return -EINVAL;
3832                 if (copy_from_user(&val, optval, sizeof(val)))
3833                         return -EFAULT;
3834 
3835                 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3836                 return 0;
3837         }
3838         default:
3839                 return -ENOPROTOOPT;
3840         }
3841 }
3842 
3843 static int packet_getsockopt(struct socket *sock, int level, int optname,
3844                              char __user *optval, int __user *optlen)
3845 {
3846         int len;
3847         int val, lv = sizeof(val);
3848         struct sock *sk = sock->sk;
3849         struct packet_sock *po = pkt_sk(sk);
3850         void *data = &val;
3851         union tpacket_stats_u st;
3852         struct tpacket_rollover_stats rstats;
3853 
3854         if (level != SOL_PACKET)
3855                 return -ENOPROTOOPT;
3856 
3857         if (get_user(len, optlen))
3858                 return -EFAULT;
3859 
3860         if (len < 0)
3861                 return -EINVAL;
3862 
3863         switch (optname) {
3864         case PACKET_STATISTICS:
3865                 spin_lock_bh(&sk->sk_receive_queue.lock);
3866                 memcpy(&st, &po->stats, sizeof(st));
3867                 memset(&po->stats, 0, sizeof(po->stats));
3868                 spin_unlock_bh(&sk->sk_receive_queue.lock);
3869 
3870                 if (po->tp_version == TPACKET_V3) {
3871                         lv = sizeof(struct tpacket_stats_v3);
3872                         st.stats3.tp_packets += st.stats3.tp_drops;
3873                         data = &st.stats3;
3874                 } else {
3875                         lv = sizeof(struct tpacket_stats);
3876                         st.stats1.tp_packets += st.stats1.tp_drops;
3877                         data = &st.stats1;
3878                 }
3879 
3880                 break;
3881         case PACKET_AUXDATA:
3882                 val = po->auxdata;
3883                 break;
3884         case PACKET_ORIGDEV:
3885                 val = po->origdev;
3886                 break;
3887         case PACKET_VNET_HDR:
3888                 val = po->has_vnet_hdr;
3889                 break;
3890         case PACKET_VERSION:
3891                 val = po->tp_version;
3892                 break;
3893         case PACKET_HDRLEN:
3894                 if (len > sizeof(int))
3895                         len = sizeof(int);
3896                 if (len < sizeof(int))
3897                         return -EINVAL;
3898                 if (copy_from_user(&val, optval, len))
3899                         return -EFAULT;
3900                 switch (val) {
3901                 case TPACKET_V1:
3902                         val = sizeof(struct tpacket_hdr);
3903                         break;
3904                 case TPACKET_V2:
3905                         val = sizeof(struct tpacket2_hdr);
3906                         break;
3907                 case TPACKET_V3:
3908                         val = sizeof(struct tpacket3_hdr);
3909                         break;
3910                 default:
3911                         return -EINVAL;
3912                 }
3913                 break;
3914         case PACKET_RESERVE:
3915                 val = po->tp_reserve;
3916                 break;
3917         case PACKET_LOSS:
3918                 val = po->tp_loss;
3919                 break;
3920         case PACKET_TIMESTAMP:
3921                 val = po->tp_tstamp;
3922                 break;
3923         case PACKET_FANOUT:
3924                 val = (po->fanout ?
3925                        ((u32)po->fanout->id |
3926                         ((u32)po->fanout->type << 16) |
3927                         ((u32)po->fanout->flags << 24)) :
3928                        0);
3929                 break;
3930         case PACKET_ROLLOVER_STATS:
3931                 if (!po->rollover)
3932                         return -EINVAL;
3933                 rstats.tp_all = atomic_long_read(&po->rollover->num);
3934                 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3935                 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3936                 data = &rstats;
3937                 lv = sizeof(rstats);
3938                 break;
3939         case PACKET_TX_HAS_OFF:
3940                 val = po->tp_tx_has_off;
3941                 break;
3942         case PACKET_QDISC_BYPASS:
3943                 val = packet_use_direct_xmit(po);
3944                 break;
3945         default:
3946                 return -ENOPROTOOPT;
3947         }
3948 
3949         if (len > lv)
3950                 len = lv;
3951         if (put_user(len, optlen))
3952                 return -EFAULT;
3953         if (copy_to_user(optval, data, len))
3954                 return -EFAULT;
3955         return 0;
3956 }
3957 
3958 
3959 #ifdef CONFIG_COMPAT
3960 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3961                                     char __user *optval, unsigned int optlen)
3962 {
3963         struct packet_sock *po = pkt_sk(sock->sk);
3964 
3965         if (level != SOL_PACKET)
3966                 return -ENOPROTOOPT;
3967 
3968         if (optname == PACKET_FANOUT_DATA &&
3969             po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3970                 optval = (char __user *)get_compat_bpf_fprog(optval);
3971                 if (!optval)
3972                         return -EFAULT;
3973                 optlen = sizeof(struct sock_fprog);
3974         }
3975 
3976         return packet_setsockopt(sock, level, optname, optval, optlen);
3977 }
3978 #endif
3979 
3980 static int packet_notifier(struct notifier_block *this,
3981                            unsigned long msg, void *ptr)
3982 {
3983         struct sock *sk;
3984         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3985         struct net *net = dev_net(dev);
3986 
3987         rcu_read_lock();
3988         sk_for_each_rcu(sk, &net->packet.sklist) {
3989                 struct packet_sock *po = pkt_sk(sk);
3990 
3991                 switch (msg) {
3992                 case NETDEV_UNREGISTER:
3993                         if (po->mclist)
3994                                 packet_dev_mclist_delete(dev, &po->mclist);
3995                         /* fallthrough */
3996 
3997                 case NETDEV_DOWN:
3998                         if (dev->ifindex == po->ifindex) {
3999                                 spin_lock(&po->bind_lock);
4000                                 if (po->running) {
4001                                         __unregister_prot_hook(sk, false);
4002                                         sk->sk_err = ENETDOWN;
4003                                         if (!sock_flag(sk, SOCK_DEAD))
4004                                                 sk->sk_error_report(sk);
4005                                 }
4006                                 if (msg == NETDEV_UNREGISTER) {
4007                                         packet_cached_dev_reset(po);
4008                                         po->ifindex = -1;
4009                                         if (po->prot_hook.dev)
4010                                                 dev_put(po->prot_hook.dev);
4011                                         po->prot_hook.dev = NULL;
4012                                 }
4013                                 spin_unlock(&po->bind_lock);
4014                         }
4015                         break;
4016                 case NETDEV_UP:
4017                         if (dev->ifindex == po->ifindex) {
4018                                 spin_lock(&po->bind_lock);
4019                                 if (po->num)
4020                                         register_prot_hook(sk);
4021                                 spin_unlock(&po->bind_lock);
4022                         }
4023                         break;
4024                 }
4025         }
4026         rcu_read_unlock();
4027         return NOTIFY_DONE;
4028 }
4029 
4030 
4031 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4032                         unsigned long arg)
4033 {
4034         struct sock *sk = sock->sk;
4035 
4036         switch (cmd) {
4037         case SIOCOUTQ:
4038         {
4039                 int amount = sk_wmem_alloc_get(sk);
4040 
4041                 return put_user(amount, (int __user *)arg);
4042         }
4043         case SIOCINQ:
4044         {
4045                 struct sk_buff *skb;
4046                 int amount = 0;
4047 
4048                 spin_lock_bh(&sk->sk_receive_queue.lock);
4049                 skb = skb_peek(&sk->sk_receive_queue);
4050                 if (skb)
4051                         amount = skb->len;
4052                 spin_unlock_bh(&sk->sk_receive_queue.lock);
4053                 return put_user(amount, (int __user *)arg);
4054         }
4055         case SIOCGSTAMP:
4056                 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4057         case SIOCGSTAMPNS:
4058                 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4059 
4060 #ifdef CONFIG_INET
4061         case SIOCADDRT:
4062         case SIOCDELRT:
4063         case SIOCDARP:
4064         case SIOCGARP:
4065         case SIOCSARP:
4066         case SIOCGIFADDR:
4067         case SIOCSIFADDR:
4068         case SIOCGIFBRDADDR:
4069         case SIOCSIFBRDADDR:
4070         case SIOCGIFNETMASK:
4071         case SIOCSIFNETMASK:
4072         case SIOCGIFDSTADDR:
4073         case SIOCSIFDSTADDR:
4074         case SIOCSIFFLAGS:
4075                 return inet_dgram_ops.ioctl(sock, cmd, arg);
4076 #endif
4077 
4078         default:
4079                 return -ENOIOCTLCMD;
4080         }
4081         return 0;
4082 }
4083 
4084 static __poll_t packet_poll(struct file *file, struct socket *sock,
4085                                 poll_table *wait)
4086 {
4087         struct sock *sk = sock->sk;
4088         struct packet_sock *po = pkt_sk(sk);
4089         __poll_t mask = datagram_poll(file, sock, wait);
4090 
4091         spin_lock_bh(&sk->sk_receive_queue.lock);
4092         if (po->rx_ring.pg_vec) {
4093                 if (!packet_previous_rx_frame(po, &po->rx_ring,
4094                         TP_STATUS_KERNEL))
4095                         mask |= EPOLLIN | EPOLLRDNORM;
4096         }
4097         if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4098                 po->pressure = 0;
4099         spin_unlock_bh(&sk->sk_receive_queue.lock);
4100         spin_lock_bh(&sk->sk_write_queue.lock);
4101         if (po->tx_ring.pg_vec) {
4102                 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4103                         mask |= EPOLLOUT | EPOLLWRNORM;
4104         }
4105         spin_unlock_bh(&sk->sk_write_queue.lock);
4106         return mask;
4107 }
4108 
4109 
4110 /* Dirty? Well, I still did not learn better way to account
4111  * for user mmaps.
4112  */
4113 
4114 static void packet_mm_open(struct vm_area_struct *vma)
4115 {
4116         struct file *file = vma->vm_file;
4117         struct socket *sock = file->private_data;
4118         struct sock *sk = sock->sk;
4119 
4120         if (sk)
4121                 atomic_inc(&pkt_sk(sk)->mapped);
4122 }
4123 
4124 static void packet_mm_close(struct vm_area_struct *vma)
4125 {
4126         struct file *file = vma->vm_file;
4127         struct socket *sock = file->private_data;
4128         struct sock *sk = sock->sk;
4129 
4130         if (sk)
4131                 atomic_dec(&pkt_sk(sk)->mapped);
4132 }
4133 
4134 static const struct vm_operations_struct packet_mmap_ops = {
4135         .open   =       packet_mm_open,
4136         .close  =       packet_mm_close,
4137 };
4138 
4139 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4140                         unsigned int len)
4141 {
4142         int i;
4143 
4144         for (i = 0; i < len; i++) {
4145                 if (likely(pg_vec[i].buffer)) {
4146                         if (is_vmalloc_addr(pg_vec[i].buffer))
4147                                 vfree(pg_vec[i].buffer);
4148                         else
4149                                 free_pages((unsigned long)pg_vec[i].buffer,
4150                                            order);
4151                         pg_vec[i].buffer = NULL;
4152                 }
4153         }
4154         kfree(pg_vec);
4155 }
4156 
4157 static char *alloc_one_pg_vec_page(unsigned long order)
4158 {
4159         char *buffer;
4160         gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4161                           __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4162 
4163         buffer = (char *) __get_free_pages(gfp_flags, order);
4164         if (buffer)
4165                 return buffer;
4166 
4167         /* __get_free_pages failed, fall back to vmalloc */
4168         buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4169         if (buffer)
4170                 return buffer;
4171 
4172         /* vmalloc failed, lets dig into swap here */
4173         gfp_flags &= ~__GFP_NORETRY;
4174         buffer = (char *) __get_free_pages(gfp_flags, order);
4175         if (buffer)
4176                 return buffer;
4177 
4178         /* complete and utter failure */
4179         return NULL;
4180 }
4181 
4182 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4183 {
4184         unsigned int block_nr = req->tp_block_nr;
4185         struct pgv *pg_vec;
4186         int i;
4187 
4188         pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4189         if (unlikely(!pg_vec))
4190                 goto out;
4191 
4192         for (i = 0; i < block_nr; i++) {
4193                 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4194                 if (unlikely(!pg_vec[i].buffer))
4195                         goto out_free_pgvec;
4196         }
4197 
4198 out:
4199         return pg_vec;
4200 
4201 out_free_pgvec:
4202         free_pg_vec(pg_vec, order, block_nr);
4203         pg_vec = NULL;
4204         goto out;
4205 }
4206 
4207 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4208                 int closing, int tx_ring)
4209 {
4210         struct pgv *pg_vec = NULL;
4211         struct packet_sock *po = pkt_sk(sk);
4212         int was_running, order = 0;
4213         struct packet_ring_buffer *rb;
4214         struct sk_buff_head *rb_queue;
4215         __be16 num;
4216         int err = -EINVAL;
4217         /* Added to avoid minimal code churn */
4218         struct tpacket_req *req = &req_u->req;
4219 
4220         rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4221         rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4222 
4223         err = -EBUSY;
4224         if (!closing) {
4225                 if (atomic_read(&po->mapped))
4226                         goto out;
4227                 if (packet_read_pending(rb))
4228                         goto out;
4229         }
4230 
4231         if (req->tp_block_nr) {
4232                 unsigned int min_frame_size;
4233 
4234                 /* Sanity tests and some calculations */
4235                 err = -EBUSY;
4236                 if (unlikely(rb->pg_vec))
4237                         goto out;
4238 
4239                 switch (po->tp_version) {
4240                 case TPACKET_V1:
4241                         po->tp_hdrlen = TPACKET_HDRLEN;
4242                         break;
4243                 case TPACKET_V2:
4244                         po->tp_hdrlen = TPACKET2_HDRLEN;
4245                         break;
4246                 case TPACKET_V3:
4247                         po->tp_hdrlen = TPACKET3_HDRLEN;
4248                         break;
4249                 }
4250 
4251                 err = -EINVAL;
4252                 if (unlikely((int)req->tp_block_size <= 0))
4253                         goto out;
4254                 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4255                         goto out;
4256                 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4257                 if (po->tp_version >= TPACKET_V3 &&
4258                     req->tp_block_size <
4259                     BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4260                         goto out;
4261                 if (unlikely(req->tp_frame_size < min_frame_size))
4262                         goto out;
4263                 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4264                         goto out;
4265 
4266                 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4267                 if (unlikely(rb->frames_per_block == 0))
4268                         goto out;
4269                 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4270                         goto out;
4271                 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4272                                         req->tp_frame_nr))
4273                         goto out;
4274 
4275                 err = -ENOMEM;
4276                 order = get_order(req->tp_block_size);
4277                 pg_vec = alloc_pg_vec(req, order);
4278                 if (unlikely(!pg_vec))
4279                         goto out;
4280                 switch (po->tp_version) {
4281                 case TPACKET_V3:
4282                         /* Block transmit is not supported yet */
4283                         if (!tx_ring) {
4284                                 init_prb_bdqc(po, rb, pg_vec, req_u);
4285                         } else {
4286                                 struct tpacket_req3 *req3 = &req_u->req3;
4287 
4288                                 if (req3->tp_retire_blk_tov ||
4289                                     req3->tp_sizeof_priv ||
4290                                     req3->tp_feature_req_word) {
4291                                         err = -EINVAL;
4292                                         goto out;
4293                                 }
4294                         }
4295                         break;
4296                 default:
4297                         break;
4298                 }
4299         }
4300         /* Done */
4301         else {
4302                 err = -EINVAL;
4303                 if (unlikely(req->tp_frame_nr))
4304                         goto out;
4305         }
4306 
4307 
4308         /* Detach socket from network */
4309         spin_lock(&po->bind_lock);
4310         was_running = po->running;
4311         num = po->num;
4312         if (was_running) {
4313                 po->num = 0;
4314                 __unregister_prot_hook(sk, false);
4315         }
4316         spin_unlock(&po->bind_lock);
4317 
4318         synchronize_net();
4319 
4320         err = -EBUSY;
4321         mutex_lock(&po->pg_vec_lock);
4322         if (closing || atomic_read(&po->mapped) == 0) {
4323                 err = 0;
4324                 spin_lock_bh(&rb_queue->lock);
4325                 swap(rb->pg_vec, pg_vec);
4326                 rb->frame_max = (req->tp_frame_nr - 1);
4327                 rb->head = 0;
4328                 rb->frame_size = req->tp_frame_size;
4329                 spin_unlock_bh(&rb_queue->lock);
4330 
4331                 swap(rb->pg_vec_order, order);
4332                 swap(rb->pg_vec_len, req->tp_block_nr);
4333 
4334                 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4335                 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4336                                                 tpacket_rcv : packet_rcv;
4337                 skb_queue_purge(rb_queue);
4338                 if (atomic_read(&po->mapped))
4339                         pr_err("packet_mmap: vma is busy: %d\n",
4340                                atomic_read(&po->mapped));
4341         }
4342         mutex_unlock(&po->pg_vec_lock);
4343 
4344         spin_lock(&po->bind_lock);
4345         if (was_running) {
4346                 po->num = num;
4347                 register_prot_hook(sk);
4348         }
4349         spin_unlock(&po->bind_lock);
4350         if (pg_vec && (po->tp_version > TPACKET_V2)) {
4351                 /* Because we don't support block-based V3 on tx-ring */
4352                 if (!tx_ring)
4353                         prb_shutdown_retire_blk_timer(po, rb_queue);
4354         }
4355 
4356         if (pg_vec)
4357                 free_pg_vec(pg_vec, order, req->tp_block_nr);
4358 out:
4359         return err;
4360 }
4361 
4362 static int packet_mmap(struct file *file, struct socket *sock,
4363                 struct vm_area_struct *vma)
4364 {
4365         struct sock *sk = sock->sk;
4366         struct packet_sock *po = pkt_sk(sk);
4367         unsigned long size, expected_size;
4368         struct packet_ring_buffer *rb;
4369         unsigned long start;
4370         int err = -EINVAL;
4371         int i;
4372 
4373         if (vma->vm_pgoff)
4374                 return -EINVAL;
4375 
4376         mutex_lock(&po->pg_vec_lock);
4377 
4378         expected_size = 0;
4379         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4380                 if (rb->pg_vec) {
4381                         expected_size += rb->pg_vec_len
4382                                                 * rb->pg_vec_pages
4383                                                 * PAGE_SIZE;
4384                 }
4385         }
4386 
4387         if (expected_size == 0)
4388                 goto out;
4389 
4390         size = vma->vm_end - vma->vm_start;
4391         if (size != expected_size)
4392                 goto out;
4393 
4394         start = vma->vm_start;
4395         for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4396                 if (rb->pg_vec == NULL)
4397                         continue;
4398 
4399                 for (i = 0; i < rb->pg_vec_len; i++) {
4400                         struct page *page;
4401                         void *kaddr = rb->pg_vec[i].buffer;
4402                         int pg_num;
4403 
4404                         for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4405                                 page = pgv_to_page(kaddr);
4406                                 err = vm_insert_page(vma, start, page);
4407                                 if (unlikely(err))
4408                                         goto out;
4409                                 start += PAGE_SIZE;
4410                                 kaddr += PAGE_SIZE;
4411                         }
4412                 }
4413         }
4414 
4415         atomic_inc(&po->mapped);
4416         vma->vm_ops = &packet_mmap_ops;
4417         err = 0;
4418 
4419 out:
4420         mutex_unlock(&po->pg_vec_lock);
4421         return err;
4422 }
4423 
4424 static const struct proto_ops packet_ops_spkt = {
4425         .family =       PF_PACKET,
4426         .owner =        THIS_MODULE,
4427         .release =      packet_release,
4428         .bind =         packet_bind_spkt,
4429         .connect =      sock_no_connect,
4430         .socketpair =   sock_no_socketpair,
4431         .accept =       sock_no_accept,
4432         .getname =      packet_getname_spkt,
4433         .poll =         datagram_poll,
4434         .ioctl =        packet_ioctl,
4435         .listen =       sock_no_listen,
4436         .shutdown =     sock_no_shutdown,
4437         .setsockopt =   sock_no_setsockopt,
4438         .getsockopt =   sock_no_getsockopt,
4439         .sendmsg =      packet_sendmsg_spkt,
4440         .recvmsg =      packet_recvmsg,
4441         .mmap =         sock_no_mmap,
4442         .sendpage =     sock_no_sendpage,
4443 };
4444 
4445 static const struct proto_ops packet_ops = {
4446         .family =       PF_PACKET,
4447         .owner =        THIS_MODULE,
4448         .release =      packet_release,
4449         .bind =         packet_bind,
4450         .connect =      sock_no_connect,
4451         .socketpair =   sock_no_socketpair,
4452         .accept =       sock_no_accept,
4453         .getname =      packet_getname,
4454         .poll =         packet_poll,
4455         .ioctl =        packet_ioctl,
4456         .listen =       sock_no_listen,
4457         .shutdown =     sock_no_shutdown,
4458         .setsockopt =   packet_setsockopt,
4459         .getsockopt =   packet_getsockopt,
4460 #ifdef CONFIG_COMPAT
4461         .compat_setsockopt = compat_packet_setsockopt,
4462 #endif
4463         .sendmsg =      packet_sendmsg,
4464         .recvmsg =      packet_recvmsg,
4465         .mmap =         packet_mmap,
4466         .sendpage =     sock_no_sendpage,
4467 };
4468 
4469 static const struct net_proto_family packet_family_ops = {
4470         .family =       PF_PACKET,
4471         .create =       packet_create,
4472         .owner  =       THIS_MODULE,
4473 };
4474 
4475 static struct notifier_block packet_netdev_notifier = {
4476         .notifier_call =        packet_notifier,
4477 };
4478 
4479 #ifdef CONFIG_PROC_FS
4480 
4481 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4482         __acquires(RCU)
4483 {
4484         struct net *net = seq_file_net(seq);
4485 
4486         rcu_read_lock();
4487         return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4488 }
4489 
4490 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4491 {
4492         struct net *net = seq_file_net(seq);
4493         return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4494 }
4495 
4496 static void packet_seq_stop(struct seq_file *seq, void *v)
4497         __releases(RCU)
4498 {
4499         rcu_read_unlock();
4500 }
4501 
4502 static int packet_seq_show(struct seq_file *seq, void *v)
4503 {
4504         if (v == SEQ_START_TOKEN)
4505                 seq_puts(seq, "sk       RefCnt Type Proto  Iface R Rmem   User   Inode\n");
4506         else {
4507                 struct sock *s = sk_entry(v);
4508                 const struct packet_sock *po = pkt_sk(s);
4509 
4510                 seq_printf(seq,
4511                            "%pK %-6d %-4d %04x   %-5d %1d %-6u %-6u %-6lu\n",
4512                            s,
4513                            refcount_read(&s->sk_refcnt),
4514                            s->sk_type,
4515                            ntohs(po->num),
4516                            po->ifindex,
4517                            po->running,
4518                            atomic_read(&s->sk_rmem_alloc),
4519                            from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4520                            sock_i_ino(s));
4521         }
4522 
4523         return 0;
4524 }
4525 
4526 static const struct seq_operations packet_seq_ops = {
4527         .start  = packet_seq_start,
4528         .next   = packet_seq_next,
4529         .stop   = packet_seq_stop,
4530         .show   = packet_seq_show,
4531 };
4532 #endif
4533 
4534 static int __net_init packet_net_init(struct net *net)
4535 {
4536         mutex_init(&net->packet.sklist_lock);
4537         INIT_HLIST_HEAD(&net->packet.sklist);
4538 
4539         if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4540                         sizeof(struct seq_net_private)))
4541                 return -ENOMEM;
4542 
4543         return 0;
4544 }
4545 
4546 static void __net_exit packet_net_exit(struct net *net)
4547 {
4548         remove_proc_entry("packet", net->proc_net);
4549         WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4550 }
4551 
4552 static struct pernet_operations packet_net_ops = {
4553         .init = packet_net_init,
4554         .exit = packet_net_exit,
4555 };
4556 
4557 
4558 static void __exit packet_exit(void)
4559 {
4560         unregister_netdevice_notifier(&packet_netdev_notifier);
4561         unregister_pernet_subsys(&packet_net_ops);
4562         sock_unregister(PF_PACKET);
4563         proto_unregister(&packet_proto);
4564 }
4565 
4566 static int __init packet_init(void)
4567 {
4568         int rc = proto_register(&packet_proto, 0);
4569 
4570         if (rc != 0)
4571                 goto out;
4572 
4573         sock_register(&packet_family_ops);
4574         register_pernet_subsys(&packet_net_ops);
4575         register_netdevice_notifier(&packet_netdev_notifier);
4576 out:
4577         return rc;
4578 }
4579 
4580 module_init(packet_init);
4581 module_exit(packet_exit);
4582 MODULE_LICENSE("GPL");
4583 MODULE_ALIAS_NETPROTO(PF_PACKET);
4584 

~ [ 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