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.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
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
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
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.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.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>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <linux/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.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>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
96 #include <net/compat.h>
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
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
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.
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
139 dev->hard_header != NULL
140 mac_header -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
156 struct packet_mreq_max {
158 unsigned short mr_type;
159 unsigned short mr_alen;
160 unsigned char mr_address[MAX_ADDR_LEN];
164 struct tpacket_hdr *h1;
165 struct tpacket2_hdr *h2;
166 struct tpacket3_hdr *h3;
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 int closing, int tx_ring);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define PGV_FROM_VMALLOC 1
182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
193 struct packet_type *pt, struct net_device *orig_dev);
195 static void *packet_previous_frame(struct packet_sock *po,
196 struct packet_ring_buffer *rb,
198 static void packet_increment_head(struct packet_ring_buffer *buff);
199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
200 struct tpacket_block_desc *);
201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
202 struct packet_sock *);
203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
204 struct packet_sock *, unsigned int status);
205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
206 static void prb_open_block(struct tpacket_kbdq_core *,
207 struct tpacket_block_desc *);
208 static void prb_retire_rx_blk_timer_expired(unsigned long);
209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
210 static void prb_init_blk_timer(struct packet_sock *,
211 struct tpacket_kbdq_core *,
212 void (*func) (unsigned long));
213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
214 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
217 struct tpacket3_hdr *);
218 static void packet_flush_mclist(struct sock *sk);
220 struct packet_skb_cb {
222 struct sockaddr_pkt pkt;
224 /* Trick: alias skb original length with
225 * ll.sll_family and ll.protocol in order
228 unsigned int origlen;
229 struct sockaddr_ll ll;
234 #define vio_le() virtio_legacy_is_little_endian()
236 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
238 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
239 #define GET_PBLOCK_DESC(x, bid) \
240 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
242 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
243 #define GET_NEXT_PRB_BLK_NUM(x) \
244 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
245 ((x)->kactive_blk_num+1) : 0)
247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
248 static void __fanout_link(struct sock *sk, struct packet_sock *po);
250 static int packet_direct_xmit(struct sk_buff *skb)
252 struct net_device *dev = skb->dev;
253 struct sk_buff *orig_skb = skb;
254 struct netdev_queue *txq;
255 int ret = NETDEV_TX_BUSY;
257 if (unlikely(!netif_running(dev) ||
258 !netif_carrier_ok(dev)))
261 skb = validate_xmit_skb_list(skb, dev);
265 txq = skb_get_tx_queue(dev, skb);
269 HARD_TX_LOCK(dev, txq, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq))
271 ret = netdev_start_xmit(skb, dev, txq, false);
272 HARD_TX_UNLOCK(dev, txq);
276 if (!dev_xmit_complete(ret))
281 atomic_long_inc(&dev->tx_dropped);
283 return NET_XMIT_DROP;
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
288 struct net_device *dev;
291 dev = rcu_dereference(po->cached_dev);
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
302 rcu_assign_pointer(po->cached_dev, dev);
305 static void packet_cached_dev_reset(struct packet_sock *po)
307 RCU_INIT_POINTER(po->cached_dev, NULL);
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
312 return po->xmit == packet_direct_xmit;
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
322 const struct net_device_ops *ops = dev->netdev_ops;
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
330 queue_index = __packet_pick_tx_queue(dev, skb);
333 skb_set_queue_mapping(skb, queue_index);
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
340 static void register_prot_hook(struct sock *sk)
342 struct packet_sock *po = pkt_sk(sk);
346 __fanout_link(sk, po);
348 dev_add_pack(&po->prot_hook);
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
364 struct packet_sock *po = pkt_sk(sk);
369 __fanout_unlink(sk, po);
371 __dev_remove_pack(&po->prot_hook);
376 spin_unlock(&po->bind_lock);
378 spin_lock(&po->bind_lock);
382 static void unregister_prot_hook(struct sock *sk, bool sync)
384 struct packet_sock *po = pkt_sk(sk);
387 __unregister_prot_hook(sk, sync);
390 static inline struct page * __pure pgv_to_page(void *addr)
392 if (is_vmalloc_addr(addr))
393 return vmalloc_to_page(addr);
394 return virt_to_page(addr);
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
399 union tpacket_uhdr h;
402 switch (po->tp_version) {
404 h.h1->tp_status = status;
405 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
408 h.h2->tp_status = status;
409 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
412 h.h3->tp_status = status;
413 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
416 WARN(1, "TPACKET version not supported.\n");
423 static int __packet_get_status(struct packet_sock *po, void *frame)
425 union tpacket_uhdr h;
430 switch (po->tp_version) {
432 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
433 return h.h1->tp_status;
435 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
436 return h.h2->tp_status;
438 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
439 return h.h3->tp_status;
441 WARN(1, "TPACKET version not supported.\n");
447 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
450 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
453 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
454 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
455 return TP_STATUS_TS_RAW_HARDWARE;
457 if (ktime_to_timespec_cond(skb->tstamp, ts))
458 return TP_STATUS_TS_SOFTWARE;
463 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
466 union tpacket_uhdr h;
470 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
474 switch (po->tp_version) {
476 h.h1->tp_sec = ts.tv_sec;
477 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
480 h.h2->tp_sec = ts.tv_sec;
481 h.h2->tp_nsec = ts.tv_nsec;
484 h.h3->tp_sec = ts.tv_sec;
485 h.h3->tp_nsec = ts.tv_nsec;
488 WARN(1, "TPACKET version not supported.\n");
492 /* one flush is safe, as both fields always lie on the same cacheline */
493 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
499 static void *packet_lookup_frame(struct packet_sock *po,
500 struct packet_ring_buffer *rb,
501 unsigned int position,
504 unsigned int pg_vec_pos, frame_offset;
505 union tpacket_uhdr h;
507 pg_vec_pos = position / rb->frames_per_block;
508 frame_offset = position % rb->frames_per_block;
510 h.raw = rb->pg_vec[pg_vec_pos].buffer +
511 (frame_offset * rb->frame_size);
513 if (status != __packet_get_status(po, h.raw))
519 static void *packet_current_frame(struct packet_sock *po,
520 struct packet_ring_buffer *rb,
523 return packet_lookup_frame(po, rb, rb->head, status);
526 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
528 del_timer_sync(&pkc->retire_blk_timer);
531 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
532 struct sk_buff_head *rb_queue)
534 struct tpacket_kbdq_core *pkc;
536 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
538 spin_lock_bh(&rb_queue->lock);
539 pkc->delete_blk_timer = 1;
540 spin_unlock_bh(&rb_queue->lock);
542 prb_del_retire_blk_timer(pkc);
545 static void prb_init_blk_timer(struct packet_sock *po,
546 struct tpacket_kbdq_core *pkc,
547 void (*func) (unsigned long))
549 init_timer(&pkc->retire_blk_timer);
550 pkc->retire_blk_timer.data = (long)po;
551 pkc->retire_blk_timer.function = func;
552 pkc->retire_blk_timer.expires = jiffies;
555 static void prb_setup_retire_blk_timer(struct packet_sock *po)
557 struct tpacket_kbdq_core *pkc;
559 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
560 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
563 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
564 int blk_size_in_bytes)
566 struct net_device *dev;
567 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
568 struct ethtool_link_ksettings ecmd;
572 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
573 if (unlikely(!dev)) {
575 return DEFAULT_PRB_RETIRE_TOV;
577 err = __ethtool_get_link_ksettings(dev, &ecmd);
581 * If the link speed is so slow you don't really
582 * need to worry about perf anyways
584 if (ecmd.base.speed < SPEED_1000 ||
585 ecmd.base.speed == SPEED_UNKNOWN) {
586 return DEFAULT_PRB_RETIRE_TOV;
589 div = ecmd.base.speed / 1000;
593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
606 union tpacket_req_u *req_u)
608 p1->feature_req_word = req_u->req3.tp_feature_req_word;
611 static void init_prb_bdqc(struct packet_sock *po,
612 struct packet_ring_buffer *rb,
614 union tpacket_req_u *req_u)
616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
617 struct tpacket_block_desc *pbd;
619 memset(p1, 0x0, sizeof(*p1));
621 p1->knxt_seq_num = 1;
623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
624 p1->pkblk_start = pg_vec[0].buffer;
625 p1->kblk_size = req_u->req3.tp_block_size;
626 p1->knum_blocks = req_u->req3.tp_block_nr;
627 p1->hdrlen = po->tp_hdrlen;
628 p1->version = po->tp_version;
629 p1->last_kactive_blk_num = 0;
630 po->stats.stats3.tp_freeze_q_cnt = 0;
631 if (req_u->req3.tp_retire_blk_tov)
632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 req_u->req3.tp_block_size);
636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
639 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
640 prb_init_ft_ops(p1, req_u);
641 prb_setup_retire_blk_timer(po);
642 prb_open_block(p1, pbd);
645 /* Do NOT update the last_blk_num first.
646 * Assumes sk_buff_head lock is held.
648 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
650 mod_timer(&pkc->retire_blk_timer,
651 jiffies + pkc->tov_in_jiffies);
652 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
657 * 1) We refresh the timer only when we open a block.
658 * By doing this we don't waste cycles refreshing the timer
659 * on packet-by-packet basis.
661 * With a 1MB block-size, on a 1Gbps line, it will take
662 * i) ~8 ms to fill a block + ii) memcpy etc.
663 * In this cut we are not accounting for the memcpy time.
665 * So, if the user sets the 'tmo' to 10ms then the timer
666 * will never fire while the block is still getting filled
667 * (which is what we want). However, the user could choose
668 * to close a block early and that's fine.
670 * But when the timer does fire, we check whether or not to refresh it.
671 * Since the tmo granularity is in msecs, it is not too expensive
672 * to refresh the timer, lets say every '8' msecs.
673 * Either the user can set the 'tmo' or we can derive it based on
674 * a) line-speed and b) block-size.
675 * prb_calc_retire_blk_tmo() calculates the tmo.
678 static void prb_retire_rx_blk_timer_expired(unsigned long data)
680 struct packet_sock *po = (struct packet_sock *)data;
681 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
683 struct tpacket_block_desc *pbd;
685 spin_lock(&po->sk.sk_receive_queue.lock);
687 frozen = prb_queue_frozen(pkc);
688 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
690 if (unlikely(pkc->delete_blk_timer))
693 /* We only need to plug the race when the block is partially filled.
695 * lock(); increment BLOCK_NUM_PKTS; unlock()
696 * copy_bits() is in progress ...
697 * timer fires on other cpu:
698 * we can't retire the current block because copy_bits
702 if (BLOCK_NUM_PKTS(pbd)) {
703 while (atomic_read(&pkc->blk_fill_in_prog)) {
704 /* Waiting for skb_copy_bits to finish... */
709 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
711 if (!BLOCK_NUM_PKTS(pbd)) {
712 /* An empty block. Just refresh the timer. */
715 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
716 if (!prb_dispatch_next_block(pkc, po))
721 /* Case 1. Queue was frozen because user-space was
724 if (prb_curr_blk_in_use(pkc, pbd)) {
726 * Ok, user-space is still behind.
727 * So just refresh the timer.
731 /* Case 2. queue was frozen,user-space caught up,
732 * now the link went idle && the timer fired.
733 * We don't have a block to close.So we open this
734 * block and restart the timer.
735 * opening a block thaws the queue,restarts timer
736 * Thawing/timer-refresh is a side effect.
738 prb_open_block(pkc, pbd);
745 _prb_refresh_rx_retire_blk_timer(pkc);
748 spin_unlock(&po->sk.sk_receive_queue.lock);
751 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
752 struct tpacket_block_desc *pbd1, __u32 status)
754 /* Flush everything minus the block header */
756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
761 /* Skip the block header(we know header WILL fit in 4K) */
764 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
765 for (; start < end; start += PAGE_SIZE)
766 flush_dcache_page(pgv_to_page(start));
771 /* Now update the block status. */
773 BLOCK_STATUS(pbd1) = status;
775 /* Flush the block header */
777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
779 flush_dcache_page(pgv_to_page(start));
789 * 2) Increment active_blk_num
791 * Note:We DONT refresh the timer on purpose.
792 * Because almost always the next block will be opened.
794 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
795 struct tpacket_block_desc *pbd1,
796 struct packet_sock *po, unsigned int stat)
798 __u32 status = TP_STATUS_USER | stat;
800 struct tpacket3_hdr *last_pkt;
801 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
802 struct sock *sk = &po->sk;
804 if (po->stats.stats3.tp_drops)
805 status |= TP_STATUS_LOSING;
807 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
808 last_pkt->tp_next_offset = 0;
810 /* Get the ts of the last pkt */
811 if (BLOCK_NUM_PKTS(pbd1)) {
812 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
813 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
815 /* Ok, we tmo'd - so get the current time.
817 * It shouldn't really happen as we don't close empty
818 * blocks. See prb_retire_rx_blk_timer_expired().
822 h1->ts_last_pkt.ts_sec = ts.tv_sec;
823 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
828 /* Flush the block */
829 prb_flush_block(pkc1, pbd1, status);
831 sk->sk_data_ready(sk);
833 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
836 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
838 pkc->reset_pending_on_curr_blk = 0;
842 * Side effect of opening a block:
844 * 1) prb_queue is thawed.
845 * 2) retire_blk_timer is refreshed.
848 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
849 struct tpacket_block_desc *pbd1)
852 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
856 /* We could have just memset this but we will lose the
857 * flexibility of making the priv area sticky
860 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
861 BLOCK_NUM_PKTS(pbd1) = 0;
862 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 h1->ts_first_pkt.ts_sec = ts.tv_sec;
867 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
869 pkc1->pkblk_start = (char *)pbd1;
870 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
872 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
873 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
875 pbd1->version = pkc1->version;
876 pkc1->prev = pkc1->nxt_offset;
877 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
879 prb_thaw_queue(pkc1);
880 _prb_refresh_rx_retire_blk_timer(pkc1);
886 * Queue freeze logic:
887 * 1) Assume tp_block_nr = 8 blocks.
888 * 2) At time 't0', user opens Rx ring.
889 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
890 * 4) user-space is either sleeping or processing block '0'.
891 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
892 * it will close block-7,loop around and try to fill block '0'.
894 * __packet_lookup_frame_in_block
895 * prb_retire_current_block()
896 * prb_dispatch_next_block()
897 * |->(BLOCK_STATUS == USER) evaluates to true
898 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
899 * 6) Now there are two cases:
900 * 6.1) Link goes idle right after the queue is frozen.
901 * But remember, the last open_block() refreshed the timer.
902 * When this timer expires,it will refresh itself so that we can
903 * re-open block-0 in near future.
904 * 6.2) Link is busy and keeps on receiving packets. This is a simple
905 * case and __packet_lookup_frame_in_block will check if block-0
906 * is free and can now be re-used.
908 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
909 struct packet_sock *po)
911 pkc->reset_pending_on_curr_blk = 1;
912 po->stats.stats3.tp_freeze_q_cnt++;
915 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
918 * If the next block is free then we will dispatch it
919 * and return a good offset.
920 * Else, we will freeze the queue.
921 * So, caller must check the return value.
923 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
924 struct packet_sock *po)
926 struct tpacket_block_desc *pbd;
930 /* 1. Get current block num */
931 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
933 /* 2. If this block is currently in_use then freeze the queue */
934 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
935 prb_freeze_queue(pkc, po);
941 * open this block and return the offset where the first packet
942 * needs to get stored.
944 prb_open_block(pkc, pbd);
945 return (void *)pkc->nxt_offset;
948 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
949 struct packet_sock *po, unsigned int status)
951 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
953 /* retire/close the current block */
954 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
956 * Plug the case where copy_bits() is in progress on
957 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
958 * have space to copy the pkt in the current block and
959 * called prb_retire_current_block()
961 * We don't need to worry about the TMO case because
962 * the timer-handler already handled this case.
964 if (!(status & TP_STATUS_BLK_TMO)) {
965 while (atomic_read(&pkc->blk_fill_in_prog)) {
966 /* Waiting for skb_copy_bits to finish... */
970 prb_close_block(pkc, pbd, po, status);
975 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
976 struct tpacket_block_desc *pbd)
978 return TP_STATUS_USER & BLOCK_STATUS(pbd);
981 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
983 return pkc->reset_pending_on_curr_blk;
986 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
988 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
989 atomic_dec(&pkc->blk_fill_in_prog);
992 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
993 struct tpacket3_hdr *ppd)
995 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
998 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
999 struct tpacket3_hdr *ppd)
1001 ppd->hv1.tp_rxhash = 0;
1004 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1005 struct tpacket3_hdr *ppd)
1007 if (skb_vlan_tag_present(pkc->skb)) {
1008 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1009 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1010 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1012 ppd->hv1.tp_vlan_tci = 0;
1013 ppd->hv1.tp_vlan_tpid = 0;
1014 ppd->tp_status = TP_STATUS_AVAILABLE;
1018 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1019 struct tpacket3_hdr *ppd)
1021 ppd->hv1.tp_padding = 0;
1022 prb_fill_vlan_info(pkc, ppd);
1024 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1025 prb_fill_rxhash(pkc, ppd);
1027 prb_clear_rxhash(pkc, ppd);
1030 static void prb_fill_curr_block(char *curr,
1031 struct tpacket_kbdq_core *pkc,
1032 struct tpacket_block_desc *pbd,
1035 struct tpacket3_hdr *ppd;
1037 ppd = (struct tpacket3_hdr *)curr;
1038 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1040 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1042 BLOCK_NUM_PKTS(pbd) += 1;
1043 atomic_inc(&pkc->blk_fill_in_prog);
1044 prb_run_all_ft_ops(pkc, ppd);
1047 /* Assumes caller has the sk->rx_queue.lock */
1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1049 struct sk_buff *skb,
1054 struct tpacket_kbdq_core *pkc;
1055 struct tpacket_block_desc *pbd;
1058 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1059 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1061 /* Queue is frozen when user space is lagging behind */
1062 if (prb_queue_frozen(pkc)) {
1064 * Check if that last block which caused the queue to freeze,
1065 * is still in_use by user-space.
1067 if (prb_curr_blk_in_use(pkc, pbd)) {
1068 /* Can't record this packet */
1072 * Ok, the block was released by user-space.
1073 * Now let's open that block.
1074 * opening a block also thaws the queue.
1075 * Thawing is a side effect.
1077 prb_open_block(pkc, pbd);
1082 curr = pkc->nxt_offset;
1084 end = (char *)pbd + pkc->kblk_size;
1086 /* first try the current block */
1087 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1088 prb_fill_curr_block(curr, pkc, pbd, len);
1089 return (void *)curr;
1092 /* Ok, close the current block */
1093 prb_retire_current_block(pkc, po, 0);
1095 /* Now, try to dispatch the next block */
1096 curr = (char *)prb_dispatch_next_block(pkc, po);
1098 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1099 prb_fill_curr_block(curr, pkc, pbd, len);
1100 return (void *)curr;
1104 * No free blocks are available.user_space hasn't caught up yet.
1105 * Queue was just frozen and now this packet will get dropped.
1110 static void *packet_current_rx_frame(struct packet_sock *po,
1111 struct sk_buff *skb,
1112 int status, unsigned int len)
1115 switch (po->tp_version) {
1118 curr = packet_lookup_frame(po, &po->rx_ring,
1119 po->rx_ring.head, status);
1122 return __packet_lookup_frame_in_block(po, skb, status, len);
1124 WARN(1, "TPACKET version not supported\n");
1130 static void *prb_lookup_block(struct packet_sock *po,
1131 struct packet_ring_buffer *rb,
1135 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1136 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1138 if (status != BLOCK_STATUS(pbd))
1143 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1146 if (rb->prb_bdqc.kactive_blk_num)
1147 prev = rb->prb_bdqc.kactive_blk_num-1;
1149 prev = rb->prb_bdqc.knum_blocks-1;
1153 /* Assumes caller has held the rx_queue.lock */
1154 static void *__prb_previous_block(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1158 unsigned int previous = prb_previous_blk_num(rb);
1159 return prb_lookup_block(po, rb, previous, status);
1162 static void *packet_previous_rx_frame(struct packet_sock *po,
1163 struct packet_ring_buffer *rb,
1166 if (po->tp_version <= TPACKET_V2)
1167 return packet_previous_frame(po, rb, status);
1169 return __prb_previous_block(po, rb, status);
1172 static void packet_increment_rx_head(struct packet_sock *po,
1173 struct packet_ring_buffer *rb)
1175 switch (po->tp_version) {
1178 return packet_increment_head(rb);
1181 WARN(1, "TPACKET version not supported.\n");
1187 static void *packet_previous_frame(struct packet_sock *po,
1188 struct packet_ring_buffer *rb,
1191 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1192 return packet_lookup_frame(po, rb, previous, status);
1195 static void packet_increment_head(struct packet_ring_buffer *buff)
1197 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1200 static void packet_inc_pending(struct packet_ring_buffer *rb)
1202 this_cpu_inc(*rb->pending_refcnt);
1205 static void packet_dec_pending(struct packet_ring_buffer *rb)
1207 this_cpu_dec(*rb->pending_refcnt);
1210 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1212 unsigned int refcnt = 0;
1215 /* We don't use pending refcount in rx_ring. */
1216 if (rb->pending_refcnt == NULL)
1219 for_each_possible_cpu(cpu)
1220 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1225 static int packet_alloc_pending(struct packet_sock *po)
1227 po->rx_ring.pending_refcnt = NULL;
1229 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1230 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1236 static void packet_free_pending(struct packet_sock *po)
1238 free_percpu(po->tx_ring.pending_refcnt);
1241 #define ROOM_POW_OFF 2
1242 #define ROOM_NONE 0x0
1243 #define ROOM_LOW 0x1
1244 #define ROOM_NORMAL 0x2
1246 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1250 len = po->rx_ring.frame_max + 1;
1251 idx = po->rx_ring.head;
1253 idx += len >> pow_off;
1256 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1259 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1263 len = po->rx_ring.prb_bdqc.knum_blocks;
1264 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1266 idx += len >> pow_off;
1269 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1272 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1274 struct sock *sk = &po->sk;
1275 int ret = ROOM_NONE;
1277 if (po->prot_hook.func != tpacket_rcv) {
1278 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1279 - (skb ? skb->truesize : 0);
1280 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1288 if (po->tp_version == TPACKET_V3) {
1289 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1291 else if (__tpacket_v3_has_room(po, 0))
1294 if (__tpacket_has_room(po, ROOM_POW_OFF))
1296 else if (__tpacket_has_room(po, 0))
1303 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1308 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1309 ret = __packet_rcv_has_room(po, skb);
1310 has_room = ret == ROOM_NORMAL;
1311 if (po->pressure == has_room)
1312 po->pressure = !has_room;
1313 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1318 static void packet_sock_destruct(struct sock *sk)
1320 skb_queue_purge(&sk->sk_error_queue);
1322 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1323 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1325 if (!sock_flag(sk, SOCK_DEAD)) {
1326 pr_err("Attempt to release alive packet socket: %p\n", sk);
1330 sk_refcnt_debug_dec(sk);
1333 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1338 rxhash = skb_get_hash(skb);
1339 for (i = 0; i < ROLLOVER_HLEN; i++)
1340 if (po->rollover->history[i] == rxhash)
1343 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1344 return count > (ROLLOVER_HLEN >> 1);
1347 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1348 struct sk_buff *skb,
1351 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1354 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1355 struct sk_buff *skb,
1358 unsigned int val = atomic_inc_return(&f->rr_cur);
1363 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1364 struct sk_buff *skb,
1367 return smp_processor_id() % num;
1370 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1371 struct sk_buff *skb,
1374 return prandom_u32_max(num);
1377 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1378 struct sk_buff *skb,
1379 unsigned int idx, bool try_self,
1382 struct packet_sock *po, *po_next, *po_skip = NULL;
1383 unsigned int i, j, room = ROOM_NONE;
1385 po = pkt_sk(f->arr[idx]);
1388 room = packet_rcv_has_room(po, skb);
1389 if (room == ROOM_NORMAL ||
1390 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1395 i = j = min_t(int, po->rollover->sock, num - 1);
1397 po_next = pkt_sk(f->arr[i]);
1398 if (po_next != po_skip && !po_next->pressure &&
1399 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1401 po->rollover->sock = i;
1402 atomic_long_inc(&po->rollover->num);
1403 if (room == ROOM_LOW)
1404 atomic_long_inc(&po->rollover->num_huge);
1412 atomic_long_inc(&po->rollover->num_failed);
1416 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1417 struct sk_buff *skb,
1420 return skb_get_queue_mapping(skb) % num;
1423 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1424 struct sk_buff *skb,
1427 struct bpf_prog *prog;
1428 unsigned int ret = 0;
1431 prog = rcu_dereference(f->bpf_prog);
1433 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1439 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1441 return f->flags & (flag >> 8);
1444 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1445 struct packet_type *pt, struct net_device *orig_dev)
1447 struct packet_fanout *f = pt->af_packet_priv;
1448 unsigned int num = READ_ONCE(f->num_members);
1449 struct net *net = read_pnet(&f->net);
1450 struct packet_sock *po;
1453 if (!net_eq(dev_net(dev), net) || !num) {
1458 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1459 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1464 case PACKET_FANOUT_HASH:
1466 idx = fanout_demux_hash(f, skb, num);
1468 case PACKET_FANOUT_LB:
1469 idx = fanout_demux_lb(f, skb, num);
1471 case PACKET_FANOUT_CPU:
1472 idx = fanout_demux_cpu(f, skb, num);
1474 case PACKET_FANOUT_RND:
1475 idx = fanout_demux_rnd(f, skb, num);
1477 case PACKET_FANOUT_QM:
1478 idx = fanout_demux_qm(f, skb, num);
1480 case PACKET_FANOUT_ROLLOVER:
1481 idx = fanout_demux_rollover(f, skb, 0, false, num);
1483 case PACKET_FANOUT_CBPF:
1484 case PACKET_FANOUT_EBPF:
1485 idx = fanout_demux_bpf(f, skb, num);
1489 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1490 idx = fanout_demux_rollover(f, skb, idx, true, num);
1492 po = pkt_sk(f->arr[idx]);
1493 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1496 DEFINE_MUTEX(fanout_mutex);
1497 EXPORT_SYMBOL_GPL(fanout_mutex);
1498 static LIST_HEAD(fanout_list);
1499 static u16 fanout_next_id;
1501 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1503 struct packet_fanout *f = po->fanout;
1505 spin_lock(&f->lock);
1506 f->arr[f->num_members] = sk;
1509 if (f->num_members == 1)
1510 dev_add_pack(&f->prot_hook);
1511 spin_unlock(&f->lock);
1514 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1516 struct packet_fanout *f = po->fanout;
1519 spin_lock(&f->lock);
1520 for (i = 0; i < f->num_members; i++) {
1521 if (f->arr[i] == sk)
1524 BUG_ON(i >= f->num_members);
1525 f->arr[i] = f->arr[f->num_members - 1];
1527 if (f->num_members == 0)
1528 __dev_remove_pack(&f->prot_hook);
1529 spin_unlock(&f->lock);
1532 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1534 if (sk->sk_family != PF_PACKET)
1537 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1540 static void fanout_init_data(struct packet_fanout *f)
1543 case PACKET_FANOUT_LB:
1544 atomic_set(&f->rr_cur, 0);
1546 case PACKET_FANOUT_CBPF:
1547 case PACKET_FANOUT_EBPF:
1548 RCU_INIT_POINTER(f->bpf_prog, NULL);
1553 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1555 struct bpf_prog *old;
1557 spin_lock(&f->lock);
1558 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1559 rcu_assign_pointer(f->bpf_prog, new);
1560 spin_unlock(&f->lock);
1564 bpf_prog_destroy(old);
1568 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1571 struct bpf_prog *new;
1572 struct sock_fprog fprog;
1575 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1577 if (len != sizeof(fprog))
1579 if (copy_from_user(&fprog, data, len))
1582 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1586 __fanout_set_data_bpf(po->fanout, new);
1590 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1593 struct bpf_prog *new;
1596 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1598 if (len != sizeof(fd))
1600 if (copy_from_user(&fd, data, len))
1603 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1605 return PTR_ERR(new);
1607 __fanout_set_data_bpf(po->fanout, new);
1611 static int fanout_set_data(struct packet_sock *po, char __user *data,
1614 switch (po->fanout->type) {
1615 case PACKET_FANOUT_CBPF:
1616 return fanout_set_data_cbpf(po, data, len);
1617 case PACKET_FANOUT_EBPF:
1618 return fanout_set_data_ebpf(po, data, len);
1624 static void fanout_release_data(struct packet_fanout *f)
1627 case PACKET_FANOUT_CBPF:
1628 case PACKET_FANOUT_EBPF:
1629 __fanout_set_data_bpf(f, NULL);
1633 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1635 struct packet_fanout *f;
1637 list_for_each_entry(f, &fanout_list, list) {
1638 if (f->id == candidate_id &&
1639 read_pnet(&f->net) == sock_net(sk)) {
1646 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1648 u16 id = fanout_next_id;
1651 if (__fanout_id_is_free(sk, id)) {
1653 fanout_next_id = id + 1;
1658 } while (id != fanout_next_id);
1663 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1665 struct packet_rollover *rollover = NULL;
1666 struct packet_sock *po = pkt_sk(sk);
1667 struct packet_fanout *f, *match;
1668 u8 type = type_flags & 0xff;
1669 u8 flags = type_flags >> 8;
1673 case PACKET_FANOUT_ROLLOVER:
1674 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1676 case PACKET_FANOUT_HASH:
1677 case PACKET_FANOUT_LB:
1678 case PACKET_FANOUT_CPU:
1679 case PACKET_FANOUT_RND:
1680 case PACKET_FANOUT_QM:
1681 case PACKET_FANOUT_CBPF:
1682 case PACKET_FANOUT_EBPF:
1688 mutex_lock(&fanout_mutex);
1698 if (type == PACKET_FANOUT_ROLLOVER ||
1699 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1701 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1704 atomic_long_set(&rollover->num, 0);
1705 atomic_long_set(&rollover->num_huge, 0);
1706 atomic_long_set(&rollover->num_failed, 0);
1707 po->rollover = rollover;
1710 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1715 if (!fanout_find_new_id(sk, &id)) {
1719 /* ephemeral flag for the first socket in the group: drop it */
1720 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1724 list_for_each_entry(f, &fanout_list, list) {
1726 read_pnet(&f->net) == sock_net(sk)) {
1732 if (match && match->flags != flags)
1736 match = kzalloc(sizeof(*match), GFP_KERNEL);
1739 write_pnet(&match->net, sock_net(sk));
1742 match->flags = flags;
1743 INIT_LIST_HEAD(&match->list);
1744 spin_lock_init(&match->lock);
1745 atomic_set(&match->sk_ref, 0);
1746 fanout_init_data(match);
1747 match->prot_hook.type = po->prot_hook.type;
1748 match->prot_hook.dev = po->prot_hook.dev;
1749 match->prot_hook.func = packet_rcv_fanout;
1750 match->prot_hook.af_packet_priv = match;
1751 match->prot_hook.id_match = match_fanout_group;
1752 list_add(&match->list, &fanout_list);
1755 if (match->type == type &&
1756 match->prot_hook.type == po->prot_hook.type &&
1757 match->prot_hook.dev == po->prot_hook.dev) {
1759 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1760 __dev_remove_pack(&po->prot_hook);
1762 atomic_inc(&match->sk_ref);
1763 __fanout_link(sk, po);
1768 if (err && rollover) {
1770 po->rollover = NULL;
1772 mutex_unlock(&fanout_mutex);
1776 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1777 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1778 * It is the responsibility of the caller to call fanout_release_data() and
1779 * free the returned packet_fanout (after synchronize_net())
1781 static struct packet_fanout *fanout_release(struct sock *sk)
1783 struct packet_sock *po = pkt_sk(sk);
1784 struct packet_fanout *f;
1786 mutex_lock(&fanout_mutex);
1791 if (atomic_dec_and_test(&f->sk_ref))
1797 kfree_rcu(po->rollover, rcu);
1799 mutex_unlock(&fanout_mutex);
1804 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1805 struct sk_buff *skb)
1807 /* Earlier code assumed this would be a VLAN pkt, double-check
1808 * this now that we have the actual packet in hand. We can only
1809 * do this check on Ethernet devices.
1811 if (unlikely(dev->type != ARPHRD_ETHER))
1814 skb_reset_mac_header(skb);
1815 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1818 static const struct proto_ops packet_ops;
1820 static const struct proto_ops packet_ops_spkt;
1822 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1823 struct packet_type *pt, struct net_device *orig_dev)
1826 struct sockaddr_pkt *spkt;
1829 * When we registered the protocol we saved the socket in the data
1830 * field for just this event.
1833 sk = pt->af_packet_priv;
1836 * Yank back the headers [hope the device set this
1837 * right or kerboom...]
1839 * Incoming packets have ll header pulled,
1842 * For outgoing ones skb->data == skb_mac_header(skb)
1843 * so that this procedure is noop.
1846 if (skb->pkt_type == PACKET_LOOPBACK)
1849 if (!net_eq(dev_net(dev), sock_net(sk)))
1852 skb = skb_share_check(skb, GFP_ATOMIC);
1856 /* drop any routing info */
1859 /* drop conntrack reference */
1862 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1864 skb_push(skb, skb->data - skb_mac_header(skb));
1867 * The SOCK_PACKET socket receives _all_ frames.
1870 spkt->spkt_family = dev->type;
1871 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1872 spkt->spkt_protocol = skb->protocol;
1875 * Charge the memory to the socket. This is done specifically
1876 * to prevent sockets using all the memory up.
1879 if (sock_queue_rcv_skb(sk, skb) == 0)
1890 * Output a raw packet to a device layer. This bypasses all the other
1891 * protocol layers and you must therefore supply it with a complete frame
1894 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1897 struct sock *sk = sock->sk;
1898 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1899 struct sk_buff *skb = NULL;
1900 struct net_device *dev;
1901 struct sockcm_cookie sockc;
1907 * Get and verify the address.
1911 if (msg->msg_namelen < sizeof(struct sockaddr))
1913 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1914 proto = saddr->spkt_protocol;
1916 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1919 * Find the device first to size check it
1922 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1925 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1931 if (!(dev->flags & IFF_UP))
1935 * You may not queue a frame bigger than the mtu. This is the lowest level
1936 * raw protocol and you must do your own fragmentation at this level.
1939 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1940 if (!netif_supports_nofcs(dev)) {
1941 err = -EPROTONOSUPPORT;
1944 extra_len = 4; /* We're doing our own CRC */
1948 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1952 size_t reserved = LL_RESERVED_SPACE(dev);
1953 int tlen = dev->needed_tailroom;
1954 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1957 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1960 /* FIXME: Save some space for broken drivers that write a hard
1961 * header at transmission time by themselves. PPP is the notable
1962 * one here. This should really be fixed at the driver level.
1964 skb_reserve(skb, reserved);
1965 skb_reset_network_header(skb);
1967 /* Try to align data part correctly */
1972 skb_reset_network_header(skb);
1974 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1980 if (!dev_validate_header(dev, skb->data, len)) {
1984 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1985 !packet_extra_vlan_len_allowed(dev, skb)) {
1990 sockc.tsflags = sk->sk_tsflags;
1991 if (msg->msg_controllen) {
1992 err = sock_cmsg_send(sk, msg, &sockc);
1997 skb->protocol = proto;
1999 skb->priority = sk->sk_priority;
2000 skb->mark = sk->sk_mark;
2002 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2004 if (unlikely(extra_len == 4))
2007 skb_probe_transport_header(skb, 0);
2009 dev_queue_xmit(skb);
2020 static unsigned int run_filter(struct sk_buff *skb,
2021 const struct sock *sk,
2024 struct sk_filter *filter;
2027 filter = rcu_dereference(sk->sk_filter);
2029 res = bpf_prog_run_clear_cb(filter->prog, skb);
2035 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2038 struct virtio_net_hdr vnet_hdr;
2040 if (*len < sizeof(vnet_hdr))
2042 *len -= sizeof(vnet_hdr);
2044 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
2047 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2051 * This function makes lazy skb cloning in hope that most of packets
2052 * are discarded by BPF.
2054 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2055 * and skb->cb are mangled. It works because (and until) packets
2056 * falling here are owned by current CPU. Output packets are cloned
2057 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2058 * sequencially, so that if we return skb to original state on exit,
2059 * we will not harm anyone.
2062 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2063 struct packet_type *pt, struct net_device *orig_dev)
2066 struct sockaddr_ll *sll;
2067 struct packet_sock *po;
2068 u8 *skb_head = skb->data;
2069 int skb_len = skb->len;
2070 unsigned int snaplen, res;
2071 bool is_drop_n_account = false;
2073 if (skb->pkt_type == PACKET_LOOPBACK)
2076 sk = pt->af_packet_priv;
2079 if (!net_eq(dev_net(dev), sock_net(sk)))
2084 if (dev->header_ops) {
2085 /* The device has an explicit notion of ll header,
2086 * exported to higher levels.
2088 * Otherwise, the device hides details of its frame
2089 * structure, so that corresponding packet head is
2090 * never delivered to user.
2092 if (sk->sk_type != SOCK_DGRAM)
2093 skb_push(skb, skb->data - skb_mac_header(skb));
2094 else if (skb->pkt_type == PACKET_OUTGOING) {
2095 /* Special case: outgoing packets have ll header at head */
2096 skb_pull(skb, skb_network_offset(skb));
2102 res = run_filter(skb, sk, snaplen);
2104 goto drop_n_restore;
2108 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2111 if (skb_shared(skb)) {
2112 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2116 if (skb_head != skb->data) {
2117 skb->data = skb_head;
2124 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2126 sll = &PACKET_SKB_CB(skb)->sa.ll;
2127 sll->sll_hatype = dev->type;
2128 sll->sll_pkttype = skb->pkt_type;
2129 if (unlikely(po->origdev))
2130 sll->sll_ifindex = orig_dev->ifindex;
2132 sll->sll_ifindex = dev->ifindex;
2134 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2136 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2137 * Use their space for storing the original skb length.
2139 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2141 if (pskb_trim(skb, snaplen))
2144 skb_set_owner_r(skb, sk);
2148 /* drop conntrack reference */
2151 spin_lock(&sk->sk_receive_queue.lock);
2152 po->stats.stats1.tp_packets++;
2153 sock_skb_set_dropcount(sk, skb);
2154 __skb_queue_tail(&sk->sk_receive_queue, skb);
2155 spin_unlock(&sk->sk_receive_queue.lock);
2156 sk->sk_data_ready(sk);
2160 is_drop_n_account = true;
2161 spin_lock(&sk->sk_receive_queue.lock);
2162 po->stats.stats1.tp_drops++;
2163 atomic_inc(&sk->sk_drops);
2164 spin_unlock(&sk->sk_receive_queue.lock);
2167 if (skb_head != skb->data && skb_shared(skb)) {
2168 skb->data = skb_head;
2172 if (!is_drop_n_account)
2179 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2180 struct packet_type *pt, struct net_device *orig_dev)
2183 struct packet_sock *po;
2184 struct sockaddr_ll *sll;
2185 union tpacket_uhdr h;
2186 u8 *skb_head = skb->data;
2187 int skb_len = skb->len;
2188 unsigned int snaplen, res;
2189 unsigned long status = TP_STATUS_USER;
2190 unsigned short macoff, netoff, hdrlen;
2191 struct sk_buff *copy_skb = NULL;
2194 bool is_drop_n_account = false;
2196 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2197 * We may add members to them until current aligned size without forcing
2198 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2200 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2201 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2203 if (skb->pkt_type == PACKET_LOOPBACK)
2206 sk = pt->af_packet_priv;
2209 if (!net_eq(dev_net(dev), sock_net(sk)))
2212 if (dev->header_ops) {
2213 if (sk->sk_type != SOCK_DGRAM)
2214 skb_push(skb, skb->data - skb_mac_header(skb));
2215 else if (skb->pkt_type == PACKET_OUTGOING) {
2216 /* Special case: outgoing packets have ll header at head */
2217 skb_pull(skb, skb_network_offset(skb));
2223 res = run_filter(skb, sk, snaplen);
2225 goto drop_n_restore;
2227 if (skb->ip_summed == CHECKSUM_PARTIAL)
2228 status |= TP_STATUS_CSUMNOTREADY;
2229 else if (skb->pkt_type != PACKET_OUTGOING &&
2230 (skb->ip_summed == CHECKSUM_COMPLETE ||
2231 skb_csum_unnecessary(skb)))
2232 status |= TP_STATUS_CSUM_VALID;
2237 if (sk->sk_type == SOCK_DGRAM) {
2238 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2241 unsigned int maclen = skb_network_offset(skb);
2242 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2243 (maclen < 16 ? 16 : maclen)) +
2245 if (po->has_vnet_hdr)
2246 netoff += sizeof(struct virtio_net_hdr);
2247 macoff = netoff - maclen;
2249 if (po->tp_version <= TPACKET_V2) {
2250 if (macoff + snaplen > po->rx_ring.frame_size) {
2251 if (po->copy_thresh &&
2252 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2253 if (skb_shared(skb)) {
2254 copy_skb = skb_clone(skb, GFP_ATOMIC);
2256 copy_skb = skb_get(skb);
2257 skb_head = skb->data;
2260 skb_set_owner_r(copy_skb, sk);
2262 snaplen = po->rx_ring.frame_size - macoff;
2263 if ((int)snaplen < 0)
2266 } else if (unlikely(macoff + snaplen >
2267 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2270 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2271 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2272 snaplen, nval, macoff);
2274 if (unlikely((int)snaplen < 0)) {
2276 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2279 spin_lock(&sk->sk_receive_queue.lock);
2280 h.raw = packet_current_rx_frame(po, skb,
2281 TP_STATUS_KERNEL, (macoff+snaplen));
2283 goto drop_n_account;
2284 if (po->tp_version <= TPACKET_V2) {
2285 packet_increment_rx_head(po, &po->rx_ring);
2287 * LOSING will be reported till you read the stats,
2288 * because it's COR - Clear On Read.
2289 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2292 if (po->stats.stats1.tp_drops)
2293 status |= TP_STATUS_LOSING;
2295 po->stats.stats1.tp_packets++;
2297 status |= TP_STATUS_COPY;
2298 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2300 spin_unlock(&sk->sk_receive_queue.lock);
2302 if (po->has_vnet_hdr) {
2303 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2304 sizeof(struct virtio_net_hdr),
2306 spin_lock(&sk->sk_receive_queue.lock);
2307 goto drop_n_account;
2311 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2313 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2314 getnstimeofday(&ts);
2316 status |= ts_status;
2318 switch (po->tp_version) {
2320 h.h1->tp_len = skb->len;
2321 h.h1->tp_snaplen = snaplen;
2322 h.h1->tp_mac = macoff;
2323 h.h1->tp_net = netoff;
2324 h.h1->tp_sec = ts.tv_sec;
2325 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2326 hdrlen = sizeof(*h.h1);
2329 h.h2->tp_len = skb->len;
2330 h.h2->tp_snaplen = snaplen;
2331 h.h2->tp_mac = macoff;
2332 h.h2->tp_net = netoff;
2333 h.h2->tp_sec = ts.tv_sec;
2334 h.h2->tp_nsec = ts.tv_nsec;
2335 if (skb_vlan_tag_present(skb)) {
2336 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2337 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2338 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2340 h.h2->tp_vlan_tci = 0;
2341 h.h2->tp_vlan_tpid = 0;
2343 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2344 hdrlen = sizeof(*h.h2);
2347 /* tp_nxt_offset,vlan are already populated above.
2348 * So DONT clear those fields here
2350 h.h3->tp_status |= status;
2351 h.h3->tp_len = skb->len;
2352 h.h3->tp_snaplen = snaplen;
2353 h.h3->tp_mac = macoff;
2354 h.h3->tp_net = netoff;
2355 h.h3->tp_sec = ts.tv_sec;
2356 h.h3->tp_nsec = ts.tv_nsec;
2357 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2358 hdrlen = sizeof(*h.h3);
2364 sll = h.raw + TPACKET_ALIGN(hdrlen);
2365 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2366 sll->sll_family = AF_PACKET;
2367 sll->sll_hatype = dev->type;
2368 sll->sll_protocol = skb->protocol;
2369 sll->sll_pkttype = skb->pkt_type;
2370 if (unlikely(po->origdev))
2371 sll->sll_ifindex = orig_dev->ifindex;
2373 sll->sll_ifindex = dev->ifindex;
2377 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2378 if (po->tp_version <= TPACKET_V2) {
2381 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2384 for (start = h.raw; start < end; start += PAGE_SIZE)
2385 flush_dcache_page(pgv_to_page(start));
2390 if (po->tp_version <= TPACKET_V2) {
2391 __packet_set_status(po, h.raw, status);
2392 sk->sk_data_ready(sk);
2394 prb_clear_blk_fill_status(&po->rx_ring);
2398 if (skb_head != skb->data && skb_shared(skb)) {
2399 skb->data = skb_head;
2403 if (!is_drop_n_account)
2410 is_drop_n_account = true;
2411 po->stats.stats1.tp_drops++;
2412 spin_unlock(&sk->sk_receive_queue.lock);
2414 sk->sk_data_ready(sk);
2415 kfree_skb(copy_skb);
2416 goto drop_n_restore;
2419 static void tpacket_destruct_skb(struct sk_buff *skb)
2421 struct packet_sock *po = pkt_sk(skb->sk);
2423 if (likely(po->tx_ring.pg_vec)) {
2427 ph = skb_shinfo(skb)->destructor_arg;
2428 packet_dec_pending(&po->tx_ring);
2430 ts = __packet_set_timestamp(po, ph, skb);
2431 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2437 static void tpacket_set_protocol(const struct net_device *dev,
2438 struct sk_buff *skb)
2440 if (dev->type == ARPHRD_ETHER) {
2441 skb_reset_mac_header(skb);
2442 skb->protocol = eth_hdr(skb)->h_proto;
2446 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2448 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2449 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2450 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2451 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2452 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2453 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2454 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2456 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2462 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2463 struct virtio_net_hdr *vnet_hdr)
2465 if (*len < sizeof(*vnet_hdr))
2467 *len -= sizeof(*vnet_hdr);
2469 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2472 return __packet_snd_vnet_parse(vnet_hdr, *len);
2475 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2476 void *frame, struct net_device *dev, void *data, int tp_len,
2477 __be16 proto, unsigned char *addr, int hlen, int copylen,
2478 const struct sockcm_cookie *sockc)
2480 union tpacket_uhdr ph;
2481 int to_write, offset, len, nr_frags, len_max;
2482 struct socket *sock = po->sk.sk_socket;
2488 skb->protocol = proto;
2490 skb->priority = po->sk.sk_priority;
2491 skb->mark = po->sk.sk_mark;
2492 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2493 skb_shinfo(skb)->destructor_arg = ph.raw;
2495 skb_reserve(skb, hlen);
2496 skb_reset_network_header(skb);
2500 if (sock->type == SOCK_DGRAM) {
2501 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2503 if (unlikely(err < 0))
2505 } else if (copylen) {
2506 int hdrlen = min_t(int, copylen, tp_len);
2508 skb_push(skb, dev->hard_header_len);
2509 skb_put(skb, copylen - dev->hard_header_len);
2510 err = skb_store_bits(skb, 0, data, hdrlen);
2513 if (!dev_validate_header(dev, skb->data, hdrlen))
2516 tpacket_set_protocol(dev, skb);
2522 offset = offset_in_page(data);
2523 len_max = PAGE_SIZE - offset;
2524 len = ((to_write > len_max) ? len_max : to_write);
2526 skb->data_len = to_write;
2527 skb->len += to_write;
2528 skb->truesize += to_write;
2529 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2531 while (likely(to_write)) {
2532 nr_frags = skb_shinfo(skb)->nr_frags;
2534 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2535 pr_err("Packet exceed the number of skb frags(%lu)\n",
2540 page = pgv_to_page(data);
2542 flush_dcache_page(page);
2544 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2547 len_max = PAGE_SIZE;
2548 len = ((to_write > len_max) ? len_max : to_write);
2551 skb_probe_transport_header(skb, 0);
2556 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2557 int size_max, void **data)
2559 union tpacket_uhdr ph;
2564 switch (po->tp_version) {
2566 if (ph.h3->tp_next_offset != 0) {
2567 pr_warn_once("variable sized slot not supported");
2570 tp_len = ph.h3->tp_len;
2573 tp_len = ph.h2->tp_len;
2576 tp_len = ph.h1->tp_len;
2579 if (unlikely(tp_len > size_max)) {
2580 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2584 if (unlikely(po->tp_tx_has_off)) {
2585 int off_min, off_max;
2587 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2588 off_max = po->tx_ring.frame_size - tp_len;
2589 if (po->sk.sk_type == SOCK_DGRAM) {
2590 switch (po->tp_version) {
2592 off = ph.h3->tp_net;
2595 off = ph.h2->tp_net;
2598 off = ph.h1->tp_net;
2602 switch (po->tp_version) {
2604 off = ph.h3->tp_mac;
2607 off = ph.h2->tp_mac;
2610 off = ph.h1->tp_mac;
2614 if (unlikely((off < off_min) || (off_max < off)))
2617 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2620 *data = frame + off;
2624 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2626 struct sk_buff *skb;
2627 struct net_device *dev;
2628 struct virtio_net_hdr *vnet_hdr = NULL;
2629 struct sockcm_cookie sockc;
2631 int err, reserve = 0;
2633 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2634 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2635 int tp_len, size_max;
2636 unsigned char *addr;
2639 int status = TP_STATUS_AVAILABLE;
2640 int hlen, tlen, copylen = 0;
2642 mutex_lock(&po->pg_vec_lock);
2644 if (likely(saddr == NULL)) {
2645 dev = packet_cached_dev_get(po);
2650 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2652 if (msg->msg_namelen < (saddr->sll_halen
2653 + offsetof(struct sockaddr_ll,
2656 proto = saddr->sll_protocol;
2657 addr = saddr->sll_addr;
2658 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2661 sockc.tsflags = po->sk.sk_tsflags;
2662 if (msg->msg_controllen) {
2663 err = sock_cmsg_send(&po->sk, msg, &sockc);
2669 if (unlikely(dev == NULL))
2672 if (unlikely(!(dev->flags & IFF_UP)))
2675 if (po->sk.sk_socket->type == SOCK_RAW)
2676 reserve = dev->hard_header_len;
2677 size_max = po->tx_ring.frame_size
2678 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2680 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2681 size_max = dev->mtu + reserve + VLAN_HLEN;
2684 ph = packet_current_frame(po, &po->tx_ring,
2685 TP_STATUS_SEND_REQUEST);
2686 if (unlikely(ph == NULL)) {
2687 if (need_wait && need_resched())
2693 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2697 status = TP_STATUS_SEND_REQUEST;
2698 hlen = LL_RESERVED_SPACE(dev);
2699 tlen = dev->needed_tailroom;
2700 if (po->has_vnet_hdr) {
2702 data += sizeof(*vnet_hdr);
2703 tp_len -= sizeof(*vnet_hdr);
2705 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2709 copylen = __virtio16_to_cpu(vio_le(),
2712 copylen = max_t(int, copylen, dev->hard_header_len);
2713 skb = sock_alloc_send_skb(&po->sk,
2714 hlen + tlen + sizeof(struct sockaddr_ll) +
2715 (copylen - dev->hard_header_len),
2718 if (unlikely(skb == NULL)) {
2719 /* we assume the socket was initially writeable ... */
2720 if (likely(len_sum > 0))
2724 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2725 addr, hlen, copylen, &sockc);
2726 if (likely(tp_len >= 0) &&
2727 tp_len > dev->mtu + reserve &&
2728 !po->has_vnet_hdr &&
2729 !packet_extra_vlan_len_allowed(dev, skb))
2732 if (unlikely(tp_len < 0)) {
2735 __packet_set_status(po, ph,
2736 TP_STATUS_AVAILABLE);
2737 packet_increment_head(&po->tx_ring);
2741 status = TP_STATUS_WRONG_FORMAT;
2747 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2753 packet_pick_tx_queue(dev, skb);
2755 skb->destructor = tpacket_destruct_skb;
2756 __packet_set_status(po, ph, TP_STATUS_SENDING);
2757 packet_inc_pending(&po->tx_ring);
2759 status = TP_STATUS_SEND_REQUEST;
2760 err = po->xmit(skb);
2761 if (unlikely(err > 0)) {
2762 err = net_xmit_errno(err);
2763 if (err && __packet_get_status(po, ph) ==
2764 TP_STATUS_AVAILABLE) {
2765 /* skb was destructed already */
2770 * skb was dropped but not destructed yet;
2771 * let's treat it like congestion or err < 0
2775 packet_increment_head(&po->tx_ring);
2777 } while (likely((ph != NULL) ||
2778 /* Note: packet_read_pending() might be slow if we have
2779 * to call it as it's per_cpu variable, but in fast-path
2780 * we already short-circuit the loop with the first
2781 * condition, and luckily don't have to go that path
2784 (need_wait && packet_read_pending(&po->tx_ring))));
2790 __packet_set_status(po, ph, status);
2795 mutex_unlock(&po->pg_vec_lock);
2799 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2800 size_t reserve, size_t len,
2801 size_t linear, int noblock,
2804 struct sk_buff *skb;
2806 /* Under a page? Don't bother with paged skb. */
2807 if (prepad + len < PAGE_SIZE || !linear)
2810 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2815 skb_reserve(skb, reserve);
2816 skb_put(skb, linear);
2817 skb->data_len = len - linear;
2818 skb->len += len - linear;
2823 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2825 struct sock *sk = sock->sk;
2826 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2827 struct sk_buff *skb;
2828 struct net_device *dev;
2830 unsigned char *addr;
2831 int err, reserve = 0;
2832 struct sockcm_cookie sockc;
2833 struct virtio_net_hdr vnet_hdr = { 0 };
2835 struct packet_sock *po = pkt_sk(sk);
2836 int hlen, tlen, linear;
2840 * Get and verify the address.
2843 if (likely(saddr == NULL)) {
2844 dev = packet_cached_dev_get(po);
2849 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2851 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2853 proto = saddr->sll_protocol;
2854 addr = saddr->sll_addr;
2855 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2859 if (unlikely(dev == NULL))
2862 if (unlikely(!(dev->flags & IFF_UP)))
2865 sockc.tsflags = sk->sk_tsflags;
2866 sockc.mark = sk->sk_mark;
2867 if (msg->msg_controllen) {
2868 err = sock_cmsg_send(sk, msg, &sockc);
2873 if (sock->type == SOCK_RAW)
2874 reserve = dev->hard_header_len;
2875 if (po->has_vnet_hdr) {
2876 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2881 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2882 if (!netif_supports_nofcs(dev)) {
2883 err = -EPROTONOSUPPORT;
2886 extra_len = 4; /* We're doing our own CRC */
2890 if (!vnet_hdr.gso_type &&
2891 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2895 hlen = LL_RESERVED_SPACE(dev);
2896 tlen = dev->needed_tailroom;
2897 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2898 linear = max(linear, min_t(int, len, dev->hard_header_len));
2899 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2900 msg->msg_flags & MSG_DONTWAIT, &err);
2904 skb_set_network_header(skb, reserve);
2907 if (sock->type == SOCK_DGRAM) {
2908 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2909 if (unlikely(offset < 0))
2913 /* Returns -EFAULT on error */
2914 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2918 if (sock->type == SOCK_RAW &&
2919 !dev_validate_header(dev, skb->data, len)) {
2924 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2926 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2927 !packet_extra_vlan_len_allowed(dev, skb)) {
2932 skb->protocol = proto;
2934 skb->priority = sk->sk_priority;
2935 skb->mark = sockc.mark;
2937 packet_pick_tx_queue(dev, skb);
2939 if (po->has_vnet_hdr) {
2940 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2943 len += sizeof(vnet_hdr);
2946 skb_probe_transport_header(skb, reserve);
2948 if (unlikely(extra_len == 4))
2951 err = po->xmit(skb);
2952 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2968 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2970 struct sock *sk = sock->sk;
2971 struct packet_sock *po = pkt_sk(sk);
2973 if (po->tx_ring.pg_vec)
2974 return tpacket_snd(po, msg);
2976 return packet_snd(sock, msg, len);
2980 * Close a PACKET socket. This is fairly simple. We immediately go
2981 * to 'closed' state and remove our protocol entry in the device list.
2984 static int packet_release(struct socket *sock)
2986 struct sock *sk = sock->sk;
2987 struct packet_sock *po;
2988 struct packet_fanout *f;
2990 union tpacket_req_u req_u;
2998 mutex_lock(&net->packet.sklist_lock);
2999 sk_del_node_init_rcu(sk);
3000 mutex_unlock(&net->packet.sklist_lock);
3003 sock_prot_inuse_add(net, sk->sk_prot, -1);
3006 spin_lock(&po->bind_lock);
3007 unregister_prot_hook(sk, false);
3008 packet_cached_dev_reset(po);
3010 if (po->prot_hook.dev) {
3011 dev_put(po->prot_hook.dev);
3012 po->prot_hook.dev = NULL;
3014 spin_unlock(&po->bind_lock);
3016 packet_flush_mclist(sk);
3018 if (po->rx_ring.pg_vec) {
3019 memset(&req_u, 0, sizeof(req_u));
3020 packet_set_ring(sk, &req_u, 1, 0);
3023 if (po->tx_ring.pg_vec) {
3024 memset(&req_u, 0, sizeof(req_u));
3025 packet_set_ring(sk, &req_u, 1, 1);
3028 f = fanout_release(sk);
3033 fanout_release_data(f);
3037 * Now the socket is dead. No more input will appear.
3044 skb_queue_purge(&sk->sk_receive_queue);
3045 packet_free_pending(po);
3046 sk_refcnt_debug_release(sk);
3053 * Attach a packet hook.
3056 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3059 struct packet_sock *po = pkt_sk(sk);
3060 struct net_device *dev_curr;
3063 struct net_device *dev = NULL;
3065 bool unlisted = false;
3071 spin_lock(&po->bind_lock);
3075 dev = dev_get_by_name_rcu(sock_net(sk), name);
3080 } else if (ifindex) {
3081 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3091 proto_curr = po->prot_hook.type;
3092 dev_curr = po->prot_hook.dev;
3094 need_rehook = proto_curr != proto || dev_curr != dev;
3099 __unregister_prot_hook(sk, true);
3101 dev_curr = po->prot_hook.dev;
3103 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3108 po->prot_hook.type = proto;
3110 if (unlikely(unlisted)) {
3112 po->prot_hook.dev = NULL;
3114 packet_cached_dev_reset(po);
3116 po->prot_hook.dev = dev;
3117 po->ifindex = dev ? dev->ifindex : 0;
3118 packet_cached_dev_assign(po, dev);
3124 if (proto == 0 || !need_rehook)
3127 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3128 register_prot_hook(sk);
3130 sk->sk_err = ENETDOWN;
3131 if (!sock_flag(sk, SOCK_DEAD))
3132 sk->sk_error_report(sk);
3137 spin_unlock(&po->bind_lock);
3143 * Bind a packet socket to a device
3146 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3149 struct sock *sk = sock->sk;
3150 char name[sizeof(uaddr->sa_data) + 1];
3156 if (addr_len != sizeof(struct sockaddr))
3158 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3161 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3162 name[sizeof(uaddr->sa_data)] = 0;
3164 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3167 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3169 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3170 struct sock *sk = sock->sk;
3176 if (addr_len < sizeof(struct sockaddr_ll))
3178 if (sll->sll_family != AF_PACKET)
3181 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3182 sll->sll_protocol ? : pkt_sk(sk)->num);
3185 static struct proto packet_proto = {
3187 .owner = THIS_MODULE,
3188 .obj_size = sizeof(struct packet_sock),
3192 * Create a packet of type SOCK_PACKET.
3195 static int packet_create(struct net *net, struct socket *sock, int protocol,
3199 struct packet_sock *po;
3200 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3203 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3205 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3206 sock->type != SOCK_PACKET)
3207 return -ESOCKTNOSUPPORT;
3209 sock->state = SS_UNCONNECTED;
3212 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3216 sock->ops = &packet_ops;
3217 if (sock->type == SOCK_PACKET)
3218 sock->ops = &packet_ops_spkt;
3220 sock_init_data(sock, sk);
3223 sk->sk_family = PF_PACKET;
3225 po->xmit = dev_queue_xmit;
3227 err = packet_alloc_pending(po);
3231 packet_cached_dev_reset(po);
3233 sk->sk_destruct = packet_sock_destruct;
3234 sk_refcnt_debug_inc(sk);
3237 * Attach a protocol block
3240 spin_lock_init(&po->bind_lock);
3241 mutex_init(&po->pg_vec_lock);
3242 po->rollover = NULL;
3243 po->prot_hook.func = packet_rcv;
3245 if (sock->type == SOCK_PACKET)
3246 po->prot_hook.func = packet_rcv_spkt;
3248 po->prot_hook.af_packet_priv = sk;
3251 po->prot_hook.type = proto;
3252 register_prot_hook(sk);
3255 mutex_lock(&net->packet.sklist_lock);
3256 sk_add_node_rcu(sk, &net->packet.sklist);
3257 mutex_unlock(&net->packet.sklist_lock);
3260 sock_prot_inuse_add(net, &packet_proto, 1);
3271 * Pull a packet from our receive queue and hand it to the user.
3272 * If necessary we block.
3275 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3278 struct sock *sk = sock->sk;
3279 struct sk_buff *skb;
3281 int vnet_hdr_len = 0;
3282 unsigned int origlen = 0;
3285 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3289 /* What error should we return now? EUNATTACH? */
3290 if (pkt_sk(sk)->ifindex < 0)
3294 if (flags & MSG_ERRQUEUE) {
3295 err = sock_recv_errqueue(sk, msg, len,
3296 SOL_PACKET, PACKET_TX_TIMESTAMP);
3301 * Call the generic datagram receiver. This handles all sorts
3302 * of horrible races and re-entrancy so we can forget about it
3303 * in the protocol layers.
3305 * Now it will return ENETDOWN, if device have just gone down,
3306 * but then it will block.
3309 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3312 * An error occurred so return it. Because skb_recv_datagram()
3313 * handles the blocking we don't see and worry about blocking
3320 if (pkt_sk(sk)->pressure)
3321 packet_rcv_has_room(pkt_sk(sk), NULL);
3323 if (pkt_sk(sk)->has_vnet_hdr) {
3324 err = packet_rcv_vnet(msg, skb, &len);
3327 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3330 /* You lose any data beyond the buffer you gave. If it worries
3331 * a user program they can ask the device for its MTU
3337 msg->msg_flags |= MSG_TRUNC;
3340 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3344 if (sock->type != SOCK_PACKET) {
3345 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3347 /* Original length was stored in sockaddr_ll fields */
3348 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3349 sll->sll_family = AF_PACKET;
3350 sll->sll_protocol = skb->protocol;
3353 sock_recv_ts_and_drops(msg, sk, skb);
3355 if (msg->msg_name) {
3356 /* If the address length field is there to be filled
3357 * in, we fill it in now.
3359 if (sock->type == SOCK_PACKET) {
3360 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3361 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3363 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3365 msg->msg_namelen = sll->sll_halen +
3366 offsetof(struct sockaddr_ll, sll_addr);
3368 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3372 if (pkt_sk(sk)->auxdata) {
3373 struct tpacket_auxdata aux;
3375 aux.tp_status = TP_STATUS_USER;
3376 if (skb->ip_summed == CHECKSUM_PARTIAL)
3377 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3378 else if (skb->pkt_type != PACKET_OUTGOING &&
3379 (skb->ip_summed == CHECKSUM_COMPLETE ||
3380 skb_csum_unnecessary(skb)))
3381 aux.tp_status |= TP_STATUS_CSUM_VALID;
3383 aux.tp_len = origlen;
3384 aux.tp_snaplen = skb->len;
3386 aux.tp_net = skb_network_offset(skb);
3387 if (skb_vlan_tag_present(skb)) {
3388 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3389 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3390 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3392 aux.tp_vlan_tci = 0;
3393 aux.tp_vlan_tpid = 0;
3395 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3399 * Free or return the buffer as appropriate. Again this
3400 * hides all the races and re-entrancy issues from us.
3402 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3405 skb_free_datagram(sk, skb);
3410 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3411 int *uaddr_len, int peer)
3413 struct net_device *dev;
3414 struct sock *sk = sock->sk;
3419 uaddr->sa_family = AF_PACKET;
3420 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3422 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3424 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3426 *uaddr_len = sizeof(*uaddr);
3431 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3432 int *uaddr_len, int peer)
3434 struct net_device *dev;
3435 struct sock *sk = sock->sk;
3436 struct packet_sock *po = pkt_sk(sk);
3437 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3442 sll->sll_family = AF_PACKET;
3443 sll->sll_ifindex = po->ifindex;
3444 sll->sll_protocol = po->num;
3445 sll->sll_pkttype = 0;
3447 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3449 sll->sll_hatype = dev->type;
3450 sll->sll_halen = dev->addr_len;
3451 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3453 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3457 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3462 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3466 case PACKET_MR_MULTICAST:
3467 if (i->alen != dev->addr_len)
3470 return dev_mc_add(dev, i->addr);
3472 return dev_mc_del(dev, i->addr);
3474 case PACKET_MR_PROMISC:
3475 return dev_set_promiscuity(dev, what);
3476 case PACKET_MR_ALLMULTI:
3477 return dev_set_allmulti(dev, what);
3478 case PACKET_MR_UNICAST:
3479 if (i->alen != dev->addr_len)
3482 return dev_uc_add(dev, i->addr);
3484 return dev_uc_del(dev, i->addr);
3492 static void packet_dev_mclist_delete(struct net_device *dev,
3493 struct packet_mclist **mlp)
3495 struct packet_mclist *ml;
3497 while ((ml = *mlp) != NULL) {
3498 if (ml->ifindex == dev->ifindex) {
3499 packet_dev_mc(dev, ml, -1);
3507 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3509 struct packet_sock *po = pkt_sk(sk);
3510 struct packet_mclist *ml, *i;
3511 struct net_device *dev;
3517 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3522 if (mreq->mr_alen > dev->addr_len)
3526 i = kmalloc(sizeof(*i), GFP_KERNEL);
3531 for (ml = po->mclist; ml; ml = ml->next) {
3532 if (ml->ifindex == mreq->mr_ifindex &&
3533 ml->type == mreq->mr_type &&
3534 ml->alen == mreq->mr_alen &&
3535 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3537 /* Free the new element ... */
3543 i->type = mreq->mr_type;
3544 i->ifindex = mreq->mr_ifindex;
3545 i->alen = mreq->mr_alen;
3546 memcpy(i->addr, mreq->mr_address, i->alen);
3547 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3549 i->next = po->mclist;
3551 err = packet_dev_mc(dev, i, 1);
3553 po->mclist = i->next;
3562 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3564 struct packet_mclist *ml, **mlp;
3568 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3569 if (ml->ifindex == mreq->mr_ifindex &&
3570 ml->type == mreq->mr_type &&
3571 ml->alen == mreq->mr_alen &&
3572 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3573 if (--ml->count == 0) {
3574 struct net_device *dev;
3576 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3578 packet_dev_mc(dev, ml, -1);
3588 static void packet_flush_mclist(struct sock *sk)
3590 struct packet_sock *po = pkt_sk(sk);
3591 struct packet_mclist *ml;
3597 while ((ml = po->mclist) != NULL) {
3598 struct net_device *dev;
3600 po->mclist = ml->next;
3601 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3603 packet_dev_mc(dev, ml, -1);
3610 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3612 struct sock *sk = sock->sk;
3613 struct packet_sock *po = pkt_sk(sk);
3616 if (level != SOL_PACKET)
3617 return -ENOPROTOOPT;
3620 case PACKET_ADD_MEMBERSHIP:
3621 case PACKET_DROP_MEMBERSHIP:
3623 struct packet_mreq_max mreq;
3625 memset(&mreq, 0, sizeof(mreq));
3626 if (len < sizeof(struct packet_mreq))
3628 if (len > sizeof(mreq))
3630 if (copy_from_user(&mreq, optval, len))
3632 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3634 if (optname == PACKET_ADD_MEMBERSHIP)
3635 ret = packet_mc_add(sk, &mreq);
3637 ret = packet_mc_drop(sk, &mreq);
3641 case PACKET_RX_RING:
3642 case PACKET_TX_RING:
3644 union tpacket_req_u req_u;
3647 switch (po->tp_version) {
3650 len = sizeof(req_u.req);
3654 len = sizeof(req_u.req3);
3659 if (copy_from_user(&req_u.req, optval, len))
3661 return packet_set_ring(sk, &req_u, 0,
3662 optname == PACKET_TX_RING);
3664 case PACKET_COPY_THRESH:
3668 if (optlen != sizeof(val))
3670 if (copy_from_user(&val, optval, sizeof(val)))
3673 pkt_sk(sk)->copy_thresh = val;
3676 case PACKET_VERSION:
3680 if (optlen != sizeof(val))
3682 if (copy_from_user(&val, optval, sizeof(val)))
3693 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3696 po->tp_version = val;
3702 case PACKET_RESERVE:
3706 if (optlen != sizeof(val))
3708 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3710 if (copy_from_user(&val, optval, sizeof(val)))
3714 po->tp_reserve = val;
3721 if (optlen != sizeof(val))
3723 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3725 if (copy_from_user(&val, optval, sizeof(val)))
3727 po->tp_loss = !!val;
3730 case PACKET_AUXDATA:
3734 if (optlen < sizeof(val))
3736 if (copy_from_user(&val, optval, sizeof(val)))
3739 po->auxdata = !!val;
3742 case PACKET_ORIGDEV:
3746 if (optlen < sizeof(val))
3748 if (copy_from_user(&val, optval, sizeof(val)))
3751 po->origdev = !!val;
3754 case PACKET_VNET_HDR:
3758 if (sock->type != SOCK_RAW)
3760 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3762 if (optlen < sizeof(val))
3764 if (copy_from_user(&val, optval, sizeof(val)))
3767 po->has_vnet_hdr = !!val;
3770 case PACKET_TIMESTAMP:
3774 if (optlen != sizeof(val))
3776 if (copy_from_user(&val, optval, sizeof(val)))
3779 po->tp_tstamp = val;
3786 if (optlen != sizeof(val))
3788 if (copy_from_user(&val, optval, sizeof(val)))
3791 return fanout_add(sk, val & 0xffff, val >> 16);
3793 case PACKET_FANOUT_DATA:
3798 return fanout_set_data(po, optval, optlen);
3800 case PACKET_TX_HAS_OFF:
3804 if (optlen != sizeof(val))
3806 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3808 if (copy_from_user(&val, optval, sizeof(val)))
3810 po->tp_tx_has_off = !!val;
3813 case PACKET_QDISC_BYPASS:
3817 if (optlen != sizeof(val))
3819 if (copy_from_user(&val, optval, sizeof(val)))
3822 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3826 return -ENOPROTOOPT;
3830 static int packet_getsockopt(struct socket *sock, int level, int optname,
3831 char __user *optval, int __user *optlen)
3834 int val, lv = sizeof(val);
3835 struct sock *sk = sock->sk;
3836 struct packet_sock *po = pkt_sk(sk);
3838 union tpacket_stats_u st;
3839 struct tpacket_rollover_stats rstats;
3841 if (level != SOL_PACKET)
3842 return -ENOPROTOOPT;
3844 if (get_user(len, optlen))
3851 case PACKET_STATISTICS:
3852 spin_lock_bh(&sk->sk_receive_queue.lock);
3853 memcpy(&st, &po->stats, sizeof(st));
3854 memset(&po->stats, 0, sizeof(po->stats));
3855 spin_unlock_bh(&sk->sk_receive_queue.lock);
3857 if (po->tp_version == TPACKET_V3) {
3858 lv = sizeof(struct tpacket_stats_v3);
3859 st.stats3.tp_packets += st.stats3.tp_drops;
3862 lv = sizeof(struct tpacket_stats);
3863 st.stats1.tp_packets += st.stats1.tp_drops;
3868 case PACKET_AUXDATA:
3871 case PACKET_ORIGDEV:
3874 case PACKET_VNET_HDR:
3875 val = po->has_vnet_hdr;
3877 case PACKET_VERSION:
3878 val = po->tp_version;
3881 if (len > sizeof(int))
3883 if (len < sizeof(int))
3885 if (copy_from_user(&val, optval, len))
3889 val = sizeof(struct tpacket_hdr);
3892 val = sizeof(struct tpacket2_hdr);
3895 val = sizeof(struct tpacket3_hdr);
3901 case PACKET_RESERVE:
3902 val = po->tp_reserve;
3907 case PACKET_TIMESTAMP:
3908 val = po->tp_tstamp;
3912 ((u32)po->fanout->id |
3913 ((u32)po->fanout->type << 16) |
3914 ((u32)po->fanout->flags << 24)) :
3917 case PACKET_ROLLOVER_STATS:
3920 rstats.tp_all = atomic_long_read(&po->rollover->num);
3921 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3922 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3924 lv = sizeof(rstats);
3926 case PACKET_TX_HAS_OFF:
3927 val = po->tp_tx_has_off;
3929 case PACKET_QDISC_BYPASS:
3930 val = packet_use_direct_xmit(po);
3933 return -ENOPROTOOPT;
3938 if (put_user(len, optlen))
3940 if (copy_to_user(optval, data, len))
3946 #ifdef CONFIG_COMPAT
3947 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3948 char __user *optval, unsigned int optlen)
3950 struct packet_sock *po = pkt_sk(sock->sk);
3952 if (level != SOL_PACKET)
3953 return -ENOPROTOOPT;
3955 if (optname == PACKET_FANOUT_DATA &&
3956 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3957 optval = (char __user *)get_compat_bpf_fprog(optval);
3960 optlen = sizeof(struct sock_fprog);
3963 return packet_setsockopt(sock, level, optname, optval, optlen);
3967 static int packet_notifier(struct notifier_block *this,
3968 unsigned long msg, void *ptr)
3971 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3972 struct net *net = dev_net(dev);
3975 sk_for_each_rcu(sk, &net->packet.sklist) {
3976 struct packet_sock *po = pkt_sk(sk);
3979 case NETDEV_UNREGISTER:
3981 packet_dev_mclist_delete(dev, &po->mclist);
3985 if (dev->ifindex == po->ifindex) {
3986 spin_lock(&po->bind_lock);
3988 __unregister_prot_hook(sk, false);
3989 sk->sk_err = ENETDOWN;
3990 if (!sock_flag(sk, SOCK_DEAD))
3991 sk->sk_error_report(sk);
3993 if (msg == NETDEV_UNREGISTER) {
3994 packet_cached_dev_reset(po);
3996 if (po->prot_hook.dev)
3997 dev_put(po->prot_hook.dev);
3998 po->prot_hook.dev = NULL;
4000 spin_unlock(&po->bind_lock);
4004 if (dev->ifindex == po->ifindex) {
4005 spin_lock(&po->bind_lock);
4007 register_prot_hook(sk);
4008 spin_unlock(&po->bind_lock);
4018 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4021 struct sock *sk = sock->sk;
4026 int amount = sk_wmem_alloc_get(sk);
4028 return put_user(amount, (int __user *)arg);
4032 struct sk_buff *skb;
4035 spin_lock_bh(&sk->sk_receive_queue.lock);
4036 skb = skb_peek(&sk->sk_receive_queue);
4039 spin_unlock_bh(&sk->sk_receive_queue.lock);
4040 return put_user(amount, (int __user *)arg);
4043 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4045 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4055 case SIOCGIFBRDADDR:
4056 case SIOCSIFBRDADDR:
4057 case SIOCGIFNETMASK:
4058 case SIOCSIFNETMASK:
4059 case SIOCGIFDSTADDR:
4060 case SIOCSIFDSTADDR:
4062 return inet_dgram_ops.ioctl(sock, cmd, arg);
4066 return -ENOIOCTLCMD;
4071 static unsigned int packet_poll(struct file *file, struct socket *sock,
4074 struct sock *sk = sock->sk;
4075 struct packet_sock *po = pkt_sk(sk);
4076 unsigned int mask = datagram_poll(file, sock, wait);
4078 spin_lock_bh(&sk->sk_receive_queue.lock);
4079 if (po->rx_ring.pg_vec) {
4080 if (!packet_previous_rx_frame(po, &po->rx_ring,
4082 mask |= POLLIN | POLLRDNORM;
4084 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4086 spin_unlock_bh(&sk->sk_receive_queue.lock);
4087 spin_lock_bh(&sk->sk_write_queue.lock);
4088 if (po->tx_ring.pg_vec) {
4089 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4090 mask |= POLLOUT | POLLWRNORM;
4092 spin_unlock_bh(&sk->sk_write_queue.lock);
4097 /* Dirty? Well, I still did not learn better way to account
4101 static void packet_mm_open(struct vm_area_struct *vma)
4103 struct file *file = vma->vm_file;
4104 struct socket *sock = file->private_data;
4105 struct sock *sk = sock->sk;
4108 atomic_inc(&pkt_sk(sk)->mapped);
4111 static void packet_mm_close(struct vm_area_struct *vma)
4113 struct file *file = vma->vm_file;
4114 struct socket *sock = file->private_data;
4115 struct sock *sk = sock->sk;
4118 atomic_dec(&pkt_sk(sk)->mapped);
4121 static const struct vm_operations_struct packet_mmap_ops = {
4122 .open = packet_mm_open,
4123 .close = packet_mm_close,
4126 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4131 for (i = 0; i < len; i++) {
4132 if (likely(pg_vec[i].buffer)) {
4133 if (is_vmalloc_addr(pg_vec[i].buffer))
4134 vfree(pg_vec[i].buffer);
4136 free_pages((unsigned long)pg_vec[i].buffer,
4138 pg_vec[i].buffer = NULL;
4144 static char *alloc_one_pg_vec_page(unsigned long order)
4147 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4148 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4150 buffer = (char *) __get_free_pages(gfp_flags, order);
4154 /* __get_free_pages failed, fall back to vmalloc */
4155 buffer = vzalloc((1 << order) * PAGE_SIZE);
4159 /* vmalloc failed, lets dig into swap here */
4160 gfp_flags &= ~__GFP_NORETRY;
4161 buffer = (char *) __get_free_pages(gfp_flags, order);
4165 /* complete and utter failure */
4169 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4171 unsigned int block_nr = req->tp_block_nr;
4175 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4176 if (unlikely(!pg_vec))
4179 for (i = 0; i < block_nr; i++) {
4180 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4181 if (unlikely(!pg_vec[i].buffer))
4182 goto out_free_pgvec;
4189 free_pg_vec(pg_vec, order, block_nr);
4194 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4195 int closing, int tx_ring)
4197 struct pgv *pg_vec = NULL;
4198 struct packet_sock *po = pkt_sk(sk);
4199 int was_running, order = 0;
4200 struct packet_ring_buffer *rb;
4201 struct sk_buff_head *rb_queue;
4204 /* Added to avoid minimal code churn */
4205 struct tpacket_req *req = &req_u->req;
4209 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4210 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4214 if (atomic_read(&po->mapped))
4216 if (packet_read_pending(rb))
4220 if (req->tp_block_nr) {
4221 /* Sanity tests and some calculations */
4223 if (unlikely(rb->pg_vec))
4226 switch (po->tp_version) {
4228 po->tp_hdrlen = TPACKET_HDRLEN;
4231 po->tp_hdrlen = TPACKET2_HDRLEN;
4234 po->tp_hdrlen = TPACKET3_HDRLEN;
4239 if (unlikely((int)req->tp_block_size <= 0))
4241 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4243 if (po->tp_version >= TPACKET_V3 &&
4244 req->tp_block_size <=
4245 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
4247 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4250 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4253 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4254 if (unlikely(rb->frames_per_block == 0))
4256 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4258 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4263 order = get_order(req->tp_block_size);
4264 pg_vec = alloc_pg_vec(req, order);
4265 if (unlikely(!pg_vec))
4267 switch (po->tp_version) {
4269 /* Block transmit is not supported yet */
4271 init_prb_bdqc(po, rb, pg_vec, req_u);
4273 struct tpacket_req3 *req3 = &req_u->req3;
4275 if (req3->tp_retire_blk_tov ||
4276 req3->tp_sizeof_priv ||
4277 req3->tp_feature_req_word) {
4290 if (unlikely(req->tp_frame_nr))
4295 /* Detach socket from network */
4296 spin_lock(&po->bind_lock);
4297 was_running = po->running;
4301 __unregister_prot_hook(sk, false);
4303 spin_unlock(&po->bind_lock);
4308 mutex_lock(&po->pg_vec_lock);
4309 if (closing || atomic_read(&po->mapped) == 0) {
4311 spin_lock_bh(&rb_queue->lock);
4312 swap(rb->pg_vec, pg_vec);
4313 rb->frame_max = (req->tp_frame_nr - 1);
4315 rb->frame_size = req->tp_frame_size;
4316 spin_unlock_bh(&rb_queue->lock);
4318 swap(rb->pg_vec_order, order);
4319 swap(rb->pg_vec_len, req->tp_block_nr);
4321 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4322 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4323 tpacket_rcv : packet_rcv;
4324 skb_queue_purge(rb_queue);
4325 if (atomic_read(&po->mapped))
4326 pr_err("packet_mmap: vma is busy: %d\n",
4327 atomic_read(&po->mapped));
4329 mutex_unlock(&po->pg_vec_lock);
4331 spin_lock(&po->bind_lock);
4334 register_prot_hook(sk);
4336 spin_unlock(&po->bind_lock);
4337 if (closing && (po->tp_version > TPACKET_V2)) {
4338 /* Because we don't support block-based V3 on tx-ring */
4340 prb_shutdown_retire_blk_timer(po, rb_queue);
4344 free_pg_vec(pg_vec, order, req->tp_block_nr);
4350 static int packet_mmap(struct file *file, struct socket *sock,
4351 struct vm_area_struct *vma)
4353 struct sock *sk = sock->sk;
4354 struct packet_sock *po = pkt_sk(sk);
4355 unsigned long size, expected_size;
4356 struct packet_ring_buffer *rb;
4357 unsigned long start;
4364 mutex_lock(&po->pg_vec_lock);
4367 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4369 expected_size += rb->pg_vec_len
4375 if (expected_size == 0)
4378 size = vma->vm_end - vma->vm_start;
4379 if (size != expected_size)
4382 start = vma->vm_start;
4383 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4384 if (rb->pg_vec == NULL)
4387 for (i = 0; i < rb->pg_vec_len; i++) {
4389 void *kaddr = rb->pg_vec[i].buffer;
4392 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4393 page = pgv_to_page(kaddr);
4394 err = vm_insert_page(vma, start, page);
4403 atomic_inc(&po->mapped);
4404 vma->vm_ops = &packet_mmap_ops;
4408 mutex_unlock(&po->pg_vec_lock);
4412 static const struct proto_ops packet_ops_spkt = {
4413 .family = PF_PACKET,
4414 .owner = THIS_MODULE,
4415 .release = packet_release,
4416 .bind = packet_bind_spkt,
4417 .connect = sock_no_connect,
4418 .socketpair = sock_no_socketpair,
4419 .accept = sock_no_accept,
4420 .getname = packet_getname_spkt,
4421 .poll = datagram_poll,
4422 .ioctl = packet_ioctl,
4423 .listen = sock_no_listen,
4424 .shutdown = sock_no_shutdown,
4425 .setsockopt = sock_no_setsockopt,
4426 .getsockopt = sock_no_getsockopt,
4427 .sendmsg = packet_sendmsg_spkt,
4428 .recvmsg = packet_recvmsg,
4429 .mmap = sock_no_mmap,
4430 .sendpage = sock_no_sendpage,
4433 static const struct proto_ops packet_ops = {
4434 .family = PF_PACKET,
4435 .owner = THIS_MODULE,
4436 .release = packet_release,
4437 .bind = packet_bind,
4438 .connect = sock_no_connect,
4439 .socketpair = sock_no_socketpair,
4440 .accept = sock_no_accept,
4441 .getname = packet_getname,
4442 .poll = packet_poll,
4443 .ioctl = packet_ioctl,
4444 .listen = sock_no_listen,
4445 .shutdown = sock_no_shutdown,
4446 .setsockopt = packet_setsockopt,
4447 .getsockopt = packet_getsockopt,
4448 #ifdef CONFIG_COMPAT
4449 .compat_setsockopt = compat_packet_setsockopt,
4451 .sendmsg = packet_sendmsg,
4452 .recvmsg = packet_recvmsg,
4453 .mmap = packet_mmap,
4454 .sendpage = sock_no_sendpage,
4457 static const struct net_proto_family packet_family_ops = {
4458 .family = PF_PACKET,
4459 .create = packet_create,
4460 .owner = THIS_MODULE,
4463 static struct notifier_block packet_netdev_notifier = {
4464 .notifier_call = packet_notifier,
4467 #ifdef CONFIG_PROC_FS
4469 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4472 struct net *net = seq_file_net(seq);
4475 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4478 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4480 struct net *net = seq_file_net(seq);
4481 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4484 static void packet_seq_stop(struct seq_file *seq, void *v)
4490 static int packet_seq_show(struct seq_file *seq, void *v)
4492 if (v == SEQ_START_TOKEN)
4493 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4495 struct sock *s = sk_entry(v);
4496 const struct packet_sock *po = pkt_sk(s);
4499 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4501 atomic_read(&s->sk_refcnt),
4506 atomic_read(&s->sk_rmem_alloc),
4507 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4514 static const struct seq_operations packet_seq_ops = {
4515 .start = packet_seq_start,
4516 .next = packet_seq_next,
4517 .stop = packet_seq_stop,
4518 .show = packet_seq_show,
4521 static int packet_seq_open(struct inode *inode, struct file *file)
4523 return seq_open_net(inode, file, &packet_seq_ops,
4524 sizeof(struct seq_net_private));
4527 static const struct file_operations packet_seq_fops = {
4528 .owner = THIS_MODULE,
4529 .open = packet_seq_open,
4531 .llseek = seq_lseek,
4532 .release = seq_release_net,
4537 static int __net_init packet_net_init(struct net *net)
4539 mutex_init(&net->packet.sklist_lock);
4540 INIT_HLIST_HEAD(&net->packet.sklist);
4542 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4548 static void __net_exit packet_net_exit(struct net *net)
4550 remove_proc_entry("packet", net->proc_net);
4553 static struct pernet_operations packet_net_ops = {
4554 .init = packet_net_init,
4555 .exit = packet_net_exit,
4559 static void __exit packet_exit(void)
4561 unregister_netdevice_notifier(&packet_netdev_notifier);
4562 unregister_pernet_subsys(&packet_net_ops);
4563 sock_unregister(PF_PACKET);
4564 proto_unregister(&packet_proto);
4567 static int __init packet_init(void)
4569 int rc = proto_register(&packet_proto, 0);
4574 sock_register(&packet_family_ops);
4575 register_pernet_subsys(&packet_net_ops);
4576 register_netdevice_notifier(&packet_netdev_notifier);
4581 module_init(packet_init);
4582 module_exit(packet_exit);
4583 MODULE_LICENSE("GPL");
4584 MODULE_ALIAS_NETPROTO(PF_PACKET);