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 BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
190 struct packet_type *pt, struct net_device *orig_dev);
192 static void *packet_previous_frame(struct packet_sock *po,
193 struct packet_ring_buffer *rb,
195 static void packet_increment_head(struct packet_ring_buffer *buff);
196 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
198 struct packet_sock *);
199 static void prb_retire_current_block(struct tpacket_kbdq_core *,
200 struct packet_sock *, unsigned int status);
201 static int prb_queue_frozen(struct tpacket_kbdq_core *);
202 static void prb_open_block(struct tpacket_kbdq_core *,
203 struct tpacket_block_desc *);
204 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
206 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
207 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
208 struct tpacket3_hdr *);
209 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
210 struct tpacket3_hdr *);
211 static void packet_flush_mclist(struct sock *sk);
212 static u16 packet_pick_tx_queue(struct sk_buff *skb);
214 struct packet_skb_cb {
216 struct sockaddr_pkt pkt;
218 /* Trick: alias skb original length with
219 * ll.sll_family and ll.protocol in order
222 unsigned int origlen;
223 struct sockaddr_ll ll;
228 #define vio_le() virtio_legacy_is_little_endian()
230 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
232 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
233 #define GET_PBLOCK_DESC(x, bid) \
234 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
235 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
237 #define GET_NEXT_PRB_BLK_NUM(x) \
238 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
239 ((x)->kactive_blk_num+1) : 0)
241 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
242 static void __fanout_link(struct sock *sk, struct packet_sock *po);
244 static int packet_direct_xmit(struct sk_buff *skb)
246 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
249 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
251 struct net_device *dev;
254 dev = rcu_dereference(po->cached_dev);
262 static void packet_cached_dev_assign(struct packet_sock *po,
263 struct net_device *dev)
265 rcu_assign_pointer(po->cached_dev, dev);
268 static void packet_cached_dev_reset(struct packet_sock *po)
270 RCU_INIT_POINTER(po->cached_dev, NULL);
273 static bool packet_use_direct_xmit(const struct packet_sock *po)
275 return po->xmit == packet_direct_xmit;
278 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb,
279 struct net_device *sb_dev)
281 return dev_pick_tx_cpu_id(dev, skb, sb_dev, NULL);
284 static u16 packet_pick_tx_queue(struct sk_buff *skb)
286 struct net_device *dev = skb->dev;
287 const struct net_device_ops *ops = dev->netdev_ops;
290 if (ops->ndo_select_queue) {
291 queue_index = ops->ndo_select_queue(dev, skb, NULL,
292 __packet_pick_tx_queue);
293 queue_index = netdev_cap_txqueue(dev, queue_index);
295 queue_index = __packet_pick_tx_queue(dev, skb, NULL);
301 /* __register_prot_hook must be invoked through register_prot_hook
302 * or from a context in which asynchronous accesses to the packet
303 * socket is not possible (packet_create()).
305 static void __register_prot_hook(struct sock *sk)
307 struct packet_sock *po = pkt_sk(sk);
311 __fanout_link(sk, po);
313 dev_add_pack(&po->prot_hook);
320 static void register_prot_hook(struct sock *sk)
322 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
323 __register_prot_hook(sk);
326 /* If the sync parameter is true, we will temporarily drop
327 * the po->bind_lock and do a synchronize_net to make sure no
328 * asynchronous packet processing paths still refer to the elements
329 * of po->prot_hook. If the sync parameter is false, it is the
330 * callers responsibility to take care of this.
332 static void __unregister_prot_hook(struct sock *sk, bool sync)
334 struct packet_sock *po = pkt_sk(sk);
336 lockdep_assert_held_once(&po->bind_lock);
341 __fanout_unlink(sk, po);
343 __dev_remove_pack(&po->prot_hook);
348 spin_unlock(&po->bind_lock);
350 spin_lock(&po->bind_lock);
354 static void unregister_prot_hook(struct sock *sk, bool sync)
356 struct packet_sock *po = pkt_sk(sk);
359 __unregister_prot_hook(sk, sync);
362 static inline struct page * __pure pgv_to_page(void *addr)
364 if (is_vmalloc_addr(addr))
365 return vmalloc_to_page(addr);
366 return virt_to_page(addr);
369 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
371 union tpacket_uhdr h;
374 switch (po->tp_version) {
376 h.h1->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
380 h.h2->tp_status = status;
381 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
384 h.h3->tp_status = status;
385 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
388 WARN(1, "TPACKET version not supported.\n");
395 static int __packet_get_status(struct packet_sock *po, void *frame)
397 union tpacket_uhdr h;
402 switch (po->tp_version) {
404 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
405 return h.h1->tp_status;
407 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
408 return h.h2->tp_status;
410 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
411 return h.h3->tp_status;
413 WARN(1, "TPACKET version not supported.\n");
419 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
422 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
425 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
426 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
427 return TP_STATUS_TS_RAW_HARDWARE;
429 if (ktime_to_timespec_cond(skb->tstamp, ts))
430 return TP_STATUS_TS_SOFTWARE;
435 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
438 union tpacket_uhdr h;
442 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
446 switch (po->tp_version) {
448 h.h1->tp_sec = ts.tv_sec;
449 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
452 h.h2->tp_sec = ts.tv_sec;
453 h.h2->tp_nsec = ts.tv_nsec;
456 h.h3->tp_sec = ts.tv_sec;
457 h.h3->tp_nsec = ts.tv_nsec;
460 WARN(1, "TPACKET version not supported.\n");
464 /* one flush is safe, as both fields always lie on the same cacheline */
465 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
471 static void *packet_lookup_frame(struct packet_sock *po,
472 struct packet_ring_buffer *rb,
473 unsigned int position,
476 unsigned int pg_vec_pos, frame_offset;
477 union tpacket_uhdr h;
479 pg_vec_pos = position / rb->frames_per_block;
480 frame_offset = position % rb->frames_per_block;
482 h.raw = rb->pg_vec[pg_vec_pos].buffer +
483 (frame_offset * rb->frame_size);
485 if (status != __packet_get_status(po, h.raw))
491 static void *packet_current_frame(struct packet_sock *po,
492 struct packet_ring_buffer *rb,
495 return packet_lookup_frame(po, rb, rb->head, status);
498 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
500 del_timer_sync(&pkc->retire_blk_timer);
503 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
504 struct sk_buff_head *rb_queue)
506 struct tpacket_kbdq_core *pkc;
508 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
510 spin_lock_bh(&rb_queue->lock);
511 pkc->delete_blk_timer = 1;
512 spin_unlock_bh(&rb_queue->lock);
514 prb_del_retire_blk_timer(pkc);
517 static void prb_setup_retire_blk_timer(struct packet_sock *po)
519 struct tpacket_kbdq_core *pkc;
521 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
522 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
524 pkc->retire_blk_timer.expires = jiffies;
527 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
528 int blk_size_in_bytes)
530 struct net_device *dev;
531 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
532 struct ethtool_link_ksettings ecmd;
536 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
537 if (unlikely(!dev)) {
539 return DEFAULT_PRB_RETIRE_TOV;
541 err = __ethtool_get_link_ksettings(dev, &ecmd);
545 * If the link speed is so slow you don't really
546 * need to worry about perf anyways
548 if (ecmd.base.speed < SPEED_1000 ||
549 ecmd.base.speed == SPEED_UNKNOWN) {
550 return DEFAULT_PRB_RETIRE_TOV;
553 div = ecmd.base.speed / 1000;
557 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
569 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
570 union tpacket_req_u *req_u)
572 p1->feature_req_word = req_u->req3.tp_feature_req_word;
575 static void init_prb_bdqc(struct packet_sock *po,
576 struct packet_ring_buffer *rb,
578 union tpacket_req_u *req_u)
580 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
581 struct tpacket_block_desc *pbd;
583 memset(p1, 0x0, sizeof(*p1));
585 p1->knxt_seq_num = 1;
587 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
588 p1->pkblk_start = pg_vec[0].buffer;
589 p1->kblk_size = req_u->req3.tp_block_size;
590 p1->knum_blocks = req_u->req3.tp_block_nr;
591 p1->hdrlen = po->tp_hdrlen;
592 p1->version = po->tp_version;
593 p1->last_kactive_blk_num = 0;
594 po->stats.stats3.tp_freeze_q_cnt = 0;
595 if (req_u->req3.tp_retire_blk_tov)
596 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
598 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
599 req_u->req3.tp_block_size);
600 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
601 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
603 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
604 prb_init_ft_ops(p1, req_u);
605 prb_setup_retire_blk_timer(po);
606 prb_open_block(p1, pbd);
609 /* Do NOT update the last_blk_num first.
610 * Assumes sk_buff_head lock is held.
612 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
614 mod_timer(&pkc->retire_blk_timer,
615 jiffies + pkc->tov_in_jiffies);
616 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
621 * 1) We refresh the timer only when we open a block.
622 * By doing this we don't waste cycles refreshing the timer
623 * on packet-by-packet basis.
625 * With a 1MB block-size, on a 1Gbps line, it will take
626 * i) ~8 ms to fill a block + ii) memcpy etc.
627 * In this cut we are not accounting for the memcpy time.
629 * So, if the user sets the 'tmo' to 10ms then the timer
630 * will never fire while the block is still getting filled
631 * (which is what we want). However, the user could choose
632 * to close a block early and that's fine.
634 * But when the timer does fire, we check whether or not to refresh it.
635 * Since the tmo granularity is in msecs, it is not too expensive
636 * to refresh the timer, lets say every '8' msecs.
637 * Either the user can set the 'tmo' or we can derive it based on
638 * a) line-speed and b) block-size.
639 * prb_calc_retire_blk_tmo() calculates the tmo.
642 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
644 struct packet_sock *po =
645 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
646 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
648 struct tpacket_block_desc *pbd;
650 spin_lock(&po->sk.sk_receive_queue.lock);
652 frozen = prb_queue_frozen(pkc);
653 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
655 if (unlikely(pkc->delete_blk_timer))
658 /* We only need to plug the race when the block is partially filled.
660 * lock(); increment BLOCK_NUM_PKTS; unlock()
661 * copy_bits() is in progress ...
662 * timer fires on other cpu:
663 * we can't retire the current block because copy_bits
667 if (BLOCK_NUM_PKTS(pbd)) {
668 while (atomic_read(&pkc->blk_fill_in_prog)) {
669 /* Waiting for skb_copy_bits to finish... */
674 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
676 if (!BLOCK_NUM_PKTS(pbd)) {
677 /* An empty block. Just refresh the timer. */
680 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
681 if (!prb_dispatch_next_block(pkc, po))
686 /* Case 1. Queue was frozen because user-space was
689 if (prb_curr_blk_in_use(pbd)) {
691 * Ok, user-space is still behind.
692 * So just refresh the timer.
696 /* Case 2. queue was frozen,user-space caught up,
697 * now the link went idle && the timer fired.
698 * We don't have a block to close.So we open this
699 * block and restart the timer.
700 * opening a block thaws the queue,restarts timer
701 * Thawing/timer-refresh is a side effect.
703 prb_open_block(pkc, pbd);
710 _prb_refresh_rx_retire_blk_timer(pkc);
713 spin_unlock(&po->sk.sk_receive_queue.lock);
716 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
717 struct tpacket_block_desc *pbd1, __u32 status)
719 /* Flush everything minus the block header */
721 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
726 /* Skip the block header(we know header WILL fit in 4K) */
729 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
730 for (; start < end; start += PAGE_SIZE)
731 flush_dcache_page(pgv_to_page(start));
736 /* Now update the block status. */
738 BLOCK_STATUS(pbd1) = status;
740 /* Flush the block header */
742 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
744 flush_dcache_page(pgv_to_page(start));
754 * 2) Increment active_blk_num
756 * Note:We DONT refresh the timer on purpose.
757 * Because almost always the next block will be opened.
759 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
760 struct tpacket_block_desc *pbd1,
761 struct packet_sock *po, unsigned int stat)
763 __u32 status = TP_STATUS_USER | stat;
765 struct tpacket3_hdr *last_pkt;
766 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
767 struct sock *sk = &po->sk;
769 if (po->stats.stats3.tp_drops)
770 status |= TP_STATUS_LOSING;
772 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
773 last_pkt->tp_next_offset = 0;
775 /* Get the ts of the last pkt */
776 if (BLOCK_NUM_PKTS(pbd1)) {
777 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
778 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
780 /* Ok, we tmo'd - so get the current time.
782 * It shouldn't really happen as we don't close empty
783 * blocks. See prb_retire_rx_blk_timer_expired().
787 h1->ts_last_pkt.ts_sec = ts.tv_sec;
788 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
793 /* Flush the block */
794 prb_flush_block(pkc1, pbd1, status);
796 sk->sk_data_ready(sk);
798 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
801 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
803 pkc->reset_pending_on_curr_blk = 0;
807 * Side effect of opening a block:
809 * 1) prb_queue is thawed.
810 * 2) retire_blk_timer is refreshed.
813 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
814 struct tpacket_block_desc *pbd1)
817 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
821 /* We could have just memset this but we will lose the
822 * flexibility of making the priv area sticky
825 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
826 BLOCK_NUM_PKTS(pbd1) = 0;
827 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831 h1->ts_first_pkt.ts_sec = ts.tv_sec;
832 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
834 pkc1->pkblk_start = (char *)pbd1;
835 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
838 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
840 pbd1->version = pkc1->version;
841 pkc1->prev = pkc1->nxt_offset;
842 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
844 prb_thaw_queue(pkc1);
845 _prb_refresh_rx_retire_blk_timer(pkc1);
851 * Queue freeze logic:
852 * 1) Assume tp_block_nr = 8 blocks.
853 * 2) At time 't0', user opens Rx ring.
854 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
855 * 4) user-space is either sleeping or processing block '0'.
856 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
857 * it will close block-7,loop around and try to fill block '0'.
859 * __packet_lookup_frame_in_block
860 * prb_retire_current_block()
861 * prb_dispatch_next_block()
862 * |->(BLOCK_STATUS == USER) evaluates to true
863 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
864 * 6) Now there are two cases:
865 * 6.1) Link goes idle right after the queue is frozen.
866 * But remember, the last open_block() refreshed the timer.
867 * When this timer expires,it will refresh itself so that we can
868 * re-open block-0 in near future.
869 * 6.2) Link is busy and keeps on receiving packets. This is a simple
870 * case and __packet_lookup_frame_in_block will check if block-0
871 * is free and can now be re-used.
873 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
874 struct packet_sock *po)
876 pkc->reset_pending_on_curr_blk = 1;
877 po->stats.stats3.tp_freeze_q_cnt++;
880 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
883 * If the next block is free then we will dispatch it
884 * and return a good offset.
885 * Else, we will freeze the queue.
886 * So, caller must check the return value.
888 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
889 struct packet_sock *po)
891 struct tpacket_block_desc *pbd;
895 /* 1. Get current block num */
896 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
898 /* 2. If this block is currently in_use then freeze the queue */
899 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
900 prb_freeze_queue(pkc, po);
906 * open this block and return the offset where the first packet
907 * needs to get stored.
909 prb_open_block(pkc, pbd);
910 return (void *)pkc->nxt_offset;
913 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
914 struct packet_sock *po, unsigned int status)
916 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
918 /* retire/close the current block */
919 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
921 * Plug the case where copy_bits() is in progress on
922 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
923 * have space to copy the pkt in the current block and
924 * called prb_retire_current_block()
926 * We don't need to worry about the TMO case because
927 * the timer-handler already handled this case.
929 if (!(status & TP_STATUS_BLK_TMO)) {
930 while (atomic_read(&pkc->blk_fill_in_prog)) {
931 /* Waiting for skb_copy_bits to finish... */
935 prb_close_block(pkc, pbd, po, status);
940 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
942 return TP_STATUS_USER & BLOCK_STATUS(pbd);
945 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
947 return pkc->reset_pending_on_curr_blk;
950 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
952 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
953 atomic_dec(&pkc->blk_fill_in_prog);
956 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
957 struct tpacket3_hdr *ppd)
959 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
962 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
963 struct tpacket3_hdr *ppd)
965 ppd->hv1.tp_rxhash = 0;
968 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
969 struct tpacket3_hdr *ppd)
971 if (skb_vlan_tag_present(pkc->skb)) {
972 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
973 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
974 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
976 ppd->hv1.tp_vlan_tci = 0;
977 ppd->hv1.tp_vlan_tpid = 0;
978 ppd->tp_status = TP_STATUS_AVAILABLE;
982 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
983 struct tpacket3_hdr *ppd)
985 ppd->hv1.tp_padding = 0;
986 prb_fill_vlan_info(pkc, ppd);
988 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
989 prb_fill_rxhash(pkc, ppd);
991 prb_clear_rxhash(pkc, ppd);
994 static void prb_fill_curr_block(char *curr,
995 struct tpacket_kbdq_core *pkc,
996 struct tpacket_block_desc *pbd,
999 struct tpacket3_hdr *ppd;
1001 ppd = (struct tpacket3_hdr *)curr;
1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_NUM_PKTS(pbd) += 1;
1007 atomic_inc(&pkc->blk_fill_in_prog);
1008 prb_run_all_ft_ops(pkc, ppd);
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013 struct sk_buff *skb,
1018 struct tpacket_kbdq_core *pkc;
1019 struct tpacket_block_desc *pbd;
1022 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1023 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1025 /* Queue is frozen when user space is lagging behind */
1026 if (prb_queue_frozen(pkc)) {
1028 * Check if that last block which caused the queue to freeze,
1029 * is still in_use by user-space.
1031 if (prb_curr_blk_in_use(pbd)) {
1032 /* Can't record this packet */
1036 * Ok, the block was released by user-space.
1037 * Now let's open that block.
1038 * opening a block also thaws the queue.
1039 * Thawing is a side effect.
1041 prb_open_block(pkc, pbd);
1046 curr = pkc->nxt_offset;
1048 end = (char *)pbd + pkc->kblk_size;
1050 /* first try the current block */
1051 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1052 prb_fill_curr_block(curr, pkc, pbd, len);
1053 return (void *)curr;
1056 /* Ok, close the current block */
1057 prb_retire_current_block(pkc, po, 0);
1059 /* Now, try to dispatch the next block */
1060 curr = (char *)prb_dispatch_next_block(pkc, po);
1062 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 prb_fill_curr_block(curr, pkc, pbd, len);
1064 return (void *)curr;
1068 * No free blocks are available.user_space hasn't caught up yet.
1069 * Queue was just frozen and now this packet will get dropped.
1074 static void *packet_current_rx_frame(struct packet_sock *po,
1075 struct sk_buff *skb,
1076 int status, unsigned int len)
1079 switch (po->tp_version) {
1082 curr = packet_lookup_frame(po, &po->rx_ring,
1083 po->rx_ring.head, status);
1086 return __packet_lookup_frame_in_block(po, skb, status, len);
1088 WARN(1, "TPACKET version not supported\n");
1094 static void *prb_lookup_block(struct packet_sock *po,
1095 struct packet_ring_buffer *rb,
1099 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1100 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1102 if (status != BLOCK_STATUS(pbd))
1107 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1110 if (rb->prb_bdqc.kactive_blk_num)
1111 prev = rb->prb_bdqc.kactive_blk_num-1;
1113 prev = rb->prb_bdqc.knum_blocks-1;
1117 /* Assumes caller has held the rx_queue.lock */
1118 static void *__prb_previous_block(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1122 unsigned int previous = prb_previous_blk_num(rb);
1123 return prb_lookup_block(po, rb, previous, status);
1126 static void *packet_previous_rx_frame(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1130 if (po->tp_version <= TPACKET_V2)
1131 return packet_previous_frame(po, rb, status);
1133 return __prb_previous_block(po, rb, status);
1136 static void packet_increment_rx_head(struct packet_sock *po,
1137 struct packet_ring_buffer *rb)
1139 switch (po->tp_version) {
1142 return packet_increment_head(rb);
1145 WARN(1, "TPACKET version not supported.\n");
1151 static void *packet_previous_frame(struct packet_sock *po,
1152 struct packet_ring_buffer *rb,
1155 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1156 return packet_lookup_frame(po, rb, previous, status);
1159 static void packet_increment_head(struct packet_ring_buffer *buff)
1161 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1164 static void packet_inc_pending(struct packet_ring_buffer *rb)
1166 this_cpu_inc(*rb->pending_refcnt);
1169 static void packet_dec_pending(struct packet_ring_buffer *rb)
1171 this_cpu_dec(*rb->pending_refcnt);
1174 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1176 unsigned int refcnt = 0;
1179 /* We don't use pending refcount in rx_ring. */
1180 if (rb->pending_refcnt == NULL)
1183 for_each_possible_cpu(cpu)
1184 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1189 static int packet_alloc_pending(struct packet_sock *po)
1191 po->rx_ring.pending_refcnt = NULL;
1193 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1194 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1200 static void packet_free_pending(struct packet_sock *po)
1202 free_percpu(po->tx_ring.pending_refcnt);
1205 #define ROOM_POW_OFF 2
1206 #define ROOM_NONE 0x0
1207 #define ROOM_LOW 0x1
1208 #define ROOM_NORMAL 0x2
1210 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1214 len = po->rx_ring.frame_max + 1;
1215 idx = po->rx_ring.head;
1217 idx += len >> pow_off;
1220 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1223 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1227 len = po->rx_ring.prb_bdqc.knum_blocks;
1228 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1230 idx += len >> pow_off;
1233 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1236 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1238 struct sock *sk = &po->sk;
1239 int ret = ROOM_NONE;
1241 if (po->prot_hook.func != tpacket_rcv) {
1242 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1243 - (skb ? skb->truesize : 0);
1244 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1252 if (po->tp_version == TPACKET_V3) {
1253 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1255 else if (__tpacket_v3_has_room(po, 0))
1258 if (__tpacket_has_room(po, ROOM_POW_OFF))
1260 else if (__tpacket_has_room(po, 0))
1267 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1272 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1273 ret = __packet_rcv_has_room(po, skb);
1274 has_room = ret == ROOM_NORMAL;
1275 if (po->pressure == has_room)
1276 po->pressure = !has_room;
1277 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1282 static void packet_sock_destruct(struct sock *sk)
1284 skb_queue_purge(&sk->sk_error_queue);
1286 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1287 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1289 if (!sock_flag(sk, SOCK_DEAD)) {
1290 pr_err("Attempt to release alive packet socket: %p\n", sk);
1294 sk_refcnt_debug_dec(sk);
1297 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1302 rxhash = skb_get_hash(skb);
1303 for (i = 0; i < ROLLOVER_HLEN; i++)
1304 if (po->rollover->history[i] == rxhash)
1307 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1308 return count > (ROLLOVER_HLEN >> 1);
1311 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1312 struct sk_buff *skb,
1315 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1318 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1319 struct sk_buff *skb,
1322 unsigned int val = atomic_inc_return(&f->rr_cur);
1327 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1328 struct sk_buff *skb,
1331 return smp_processor_id() % num;
1334 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1335 struct sk_buff *skb,
1338 return prandom_u32_max(num);
1341 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1342 struct sk_buff *skb,
1343 unsigned int idx, bool try_self,
1346 struct packet_sock *po, *po_next, *po_skip = NULL;
1347 unsigned int i, j, room = ROOM_NONE;
1349 po = pkt_sk(f->arr[idx]);
1352 room = packet_rcv_has_room(po, skb);
1353 if (room == ROOM_NORMAL ||
1354 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1359 i = j = min_t(int, po->rollover->sock, num - 1);
1361 po_next = pkt_sk(f->arr[i]);
1362 if (po_next != po_skip && !po_next->pressure &&
1363 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1365 po->rollover->sock = i;
1366 atomic_long_inc(&po->rollover->num);
1367 if (room == ROOM_LOW)
1368 atomic_long_inc(&po->rollover->num_huge);
1376 atomic_long_inc(&po->rollover->num_failed);
1380 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1381 struct sk_buff *skb,
1384 return skb_get_queue_mapping(skb) % num;
1387 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1388 struct sk_buff *skb,
1391 struct bpf_prog *prog;
1392 unsigned int ret = 0;
1395 prog = rcu_dereference(f->bpf_prog);
1397 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1403 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1405 return f->flags & (flag >> 8);
1408 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1409 struct packet_type *pt, struct net_device *orig_dev)
1411 struct packet_fanout *f = pt->af_packet_priv;
1412 unsigned int num = READ_ONCE(f->num_members);
1413 struct net *net = read_pnet(&f->net);
1414 struct packet_sock *po;
1417 if (!net_eq(dev_net(dev), net) || !num) {
1422 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1423 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1428 case PACKET_FANOUT_HASH:
1430 idx = fanout_demux_hash(f, skb, num);
1432 case PACKET_FANOUT_LB:
1433 idx = fanout_demux_lb(f, skb, num);
1435 case PACKET_FANOUT_CPU:
1436 idx = fanout_demux_cpu(f, skb, num);
1438 case PACKET_FANOUT_RND:
1439 idx = fanout_demux_rnd(f, skb, num);
1441 case PACKET_FANOUT_QM:
1442 idx = fanout_demux_qm(f, skb, num);
1444 case PACKET_FANOUT_ROLLOVER:
1445 idx = fanout_demux_rollover(f, skb, 0, false, num);
1447 case PACKET_FANOUT_CBPF:
1448 case PACKET_FANOUT_EBPF:
1449 idx = fanout_demux_bpf(f, skb, num);
1453 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1454 idx = fanout_demux_rollover(f, skb, idx, true, num);
1456 po = pkt_sk(f->arr[idx]);
1457 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1460 DEFINE_MUTEX(fanout_mutex);
1461 EXPORT_SYMBOL_GPL(fanout_mutex);
1462 static LIST_HEAD(fanout_list);
1463 static u16 fanout_next_id;
1465 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1467 struct packet_fanout *f = po->fanout;
1469 spin_lock(&f->lock);
1470 f->arr[f->num_members] = sk;
1473 if (f->num_members == 1)
1474 dev_add_pack(&f->prot_hook);
1475 spin_unlock(&f->lock);
1478 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1480 struct packet_fanout *f = po->fanout;
1483 spin_lock(&f->lock);
1484 for (i = 0; i < f->num_members; i++) {
1485 if (f->arr[i] == sk)
1488 BUG_ON(i >= f->num_members);
1489 f->arr[i] = f->arr[f->num_members - 1];
1491 if (f->num_members == 0)
1492 __dev_remove_pack(&f->prot_hook);
1493 spin_unlock(&f->lock);
1496 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1498 if (sk->sk_family != PF_PACKET)
1501 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1504 static void fanout_init_data(struct packet_fanout *f)
1507 case PACKET_FANOUT_LB:
1508 atomic_set(&f->rr_cur, 0);
1510 case PACKET_FANOUT_CBPF:
1511 case PACKET_FANOUT_EBPF:
1512 RCU_INIT_POINTER(f->bpf_prog, NULL);
1517 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1519 struct bpf_prog *old;
1521 spin_lock(&f->lock);
1522 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1523 rcu_assign_pointer(f->bpf_prog, new);
1524 spin_unlock(&f->lock);
1528 bpf_prog_destroy(old);
1532 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1535 struct bpf_prog *new;
1536 struct sock_fprog fprog;
1539 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1541 if (len != sizeof(fprog))
1543 if (copy_from_user(&fprog, data, len))
1546 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1550 __fanout_set_data_bpf(po->fanout, new);
1554 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1557 struct bpf_prog *new;
1560 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1562 if (len != sizeof(fd))
1564 if (copy_from_user(&fd, data, len))
1567 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 return PTR_ERR(new);
1571 __fanout_set_data_bpf(po->fanout, new);
1575 static int fanout_set_data(struct packet_sock *po, char __user *data,
1578 switch (po->fanout->type) {
1579 case PACKET_FANOUT_CBPF:
1580 return fanout_set_data_cbpf(po, data, len);
1581 case PACKET_FANOUT_EBPF:
1582 return fanout_set_data_ebpf(po, data, len);
1588 static void fanout_release_data(struct packet_fanout *f)
1591 case PACKET_FANOUT_CBPF:
1592 case PACKET_FANOUT_EBPF:
1593 __fanout_set_data_bpf(f, NULL);
1597 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1599 struct packet_fanout *f;
1601 list_for_each_entry(f, &fanout_list, list) {
1602 if (f->id == candidate_id &&
1603 read_pnet(&f->net) == sock_net(sk)) {
1610 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1612 u16 id = fanout_next_id;
1615 if (__fanout_id_is_free(sk, id)) {
1617 fanout_next_id = id + 1;
1622 } while (id != fanout_next_id);
1627 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1629 struct packet_rollover *rollover = NULL;
1630 struct packet_sock *po = pkt_sk(sk);
1631 struct packet_fanout *f, *match;
1632 u8 type = type_flags & 0xff;
1633 u8 flags = type_flags >> 8;
1637 case PACKET_FANOUT_ROLLOVER:
1638 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1640 case PACKET_FANOUT_HASH:
1641 case PACKET_FANOUT_LB:
1642 case PACKET_FANOUT_CPU:
1643 case PACKET_FANOUT_RND:
1644 case PACKET_FANOUT_QM:
1645 case PACKET_FANOUT_CBPF:
1646 case PACKET_FANOUT_EBPF:
1652 mutex_lock(&fanout_mutex);
1658 if (type == PACKET_FANOUT_ROLLOVER ||
1659 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1661 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1664 atomic_long_set(&rollover->num, 0);
1665 atomic_long_set(&rollover->num_huge, 0);
1666 atomic_long_set(&rollover->num_failed, 0);
1669 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1674 if (!fanout_find_new_id(sk, &id)) {
1678 /* ephemeral flag for the first socket in the group: drop it */
1679 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1683 list_for_each_entry(f, &fanout_list, list) {
1685 read_pnet(&f->net) == sock_net(sk)) {
1691 if (match && match->flags != flags)
1695 match = kzalloc(sizeof(*match), GFP_KERNEL);
1698 write_pnet(&match->net, sock_net(sk));
1701 match->flags = flags;
1702 INIT_LIST_HEAD(&match->list);
1703 spin_lock_init(&match->lock);
1704 refcount_set(&match->sk_ref, 0);
1705 fanout_init_data(match);
1706 match->prot_hook.type = po->prot_hook.type;
1707 match->prot_hook.dev = po->prot_hook.dev;
1708 match->prot_hook.func = packet_rcv_fanout;
1709 match->prot_hook.af_packet_priv = match;
1710 match->prot_hook.id_match = match_fanout_group;
1711 list_add(&match->list, &fanout_list);
1715 spin_lock(&po->bind_lock);
1717 match->type == type &&
1718 match->prot_hook.type == po->prot_hook.type &&
1719 match->prot_hook.dev == po->prot_hook.dev) {
1721 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1722 __dev_remove_pack(&po->prot_hook);
1724 po->rollover = rollover;
1726 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1727 __fanout_link(sk, po);
1731 spin_unlock(&po->bind_lock);
1733 if (err && !refcount_read(&match->sk_ref)) {
1734 list_del(&match->list);
1740 mutex_unlock(&fanout_mutex);
1744 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1745 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1746 * It is the responsibility of the caller to call fanout_release_data() and
1747 * free the returned packet_fanout (after synchronize_net())
1749 static struct packet_fanout *fanout_release(struct sock *sk)
1751 struct packet_sock *po = pkt_sk(sk);
1752 struct packet_fanout *f;
1754 mutex_lock(&fanout_mutex);
1759 if (refcount_dec_and_test(&f->sk_ref))
1764 mutex_unlock(&fanout_mutex);
1769 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1770 struct sk_buff *skb)
1772 /* Earlier code assumed this would be a VLAN pkt, double-check
1773 * this now that we have the actual packet in hand. We can only
1774 * do this check on Ethernet devices.
1776 if (unlikely(dev->type != ARPHRD_ETHER))
1779 skb_reset_mac_header(skb);
1780 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1783 static const struct proto_ops packet_ops;
1785 static const struct proto_ops packet_ops_spkt;
1787 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1788 struct packet_type *pt, struct net_device *orig_dev)
1791 struct sockaddr_pkt *spkt;
1794 * When we registered the protocol we saved the socket in the data
1795 * field for just this event.
1798 sk = pt->af_packet_priv;
1801 * Yank back the headers [hope the device set this
1802 * right or kerboom...]
1804 * Incoming packets have ll header pulled,
1807 * For outgoing ones skb->data == skb_mac_header(skb)
1808 * so that this procedure is noop.
1811 if (skb->pkt_type == PACKET_LOOPBACK)
1814 if (!net_eq(dev_net(dev), sock_net(sk)))
1817 skb = skb_share_check(skb, GFP_ATOMIC);
1821 /* drop any routing info */
1824 /* drop conntrack reference */
1827 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1829 skb_push(skb, skb->data - skb_mac_header(skb));
1832 * The SOCK_PACKET socket receives _all_ frames.
1835 spkt->spkt_family = dev->type;
1836 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1837 spkt->spkt_protocol = skb->protocol;
1840 * Charge the memory to the socket. This is done specifically
1841 * to prevent sockets using all the memory up.
1844 if (sock_queue_rcv_skb(sk, skb) == 0)
1855 * Output a raw packet to a device layer. This bypasses all the other
1856 * protocol layers and you must therefore supply it with a complete frame
1859 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1862 struct sock *sk = sock->sk;
1863 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1864 struct sk_buff *skb = NULL;
1865 struct net_device *dev;
1866 struct sockcm_cookie sockc;
1872 * Get and verify the address.
1876 if (msg->msg_namelen < sizeof(struct sockaddr))
1878 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1879 proto = saddr->spkt_protocol;
1881 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1884 * Find the device first to size check it
1887 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1890 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1896 if (!(dev->flags & IFF_UP))
1900 * You may not queue a frame bigger than the mtu. This is the lowest level
1901 * raw protocol and you must do your own fragmentation at this level.
1904 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1905 if (!netif_supports_nofcs(dev)) {
1906 err = -EPROTONOSUPPORT;
1909 extra_len = 4; /* We're doing our own CRC */
1913 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1917 size_t reserved = LL_RESERVED_SPACE(dev);
1918 int tlen = dev->needed_tailroom;
1919 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1922 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1925 /* FIXME: Save some space for broken drivers that write a hard
1926 * header at transmission time by themselves. PPP is the notable
1927 * one here. This should really be fixed at the driver level.
1929 skb_reserve(skb, reserved);
1930 skb_reset_network_header(skb);
1932 /* Try to align data part correctly */
1937 skb_reset_network_header(skb);
1939 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1945 if (!dev_validate_header(dev, skb->data, len)) {
1949 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1950 !packet_extra_vlan_len_allowed(dev, skb)) {
1955 sockcm_init(&sockc, sk);
1956 if (msg->msg_controllen) {
1957 err = sock_cmsg_send(sk, msg, &sockc);
1962 skb->protocol = proto;
1964 skb->priority = sk->sk_priority;
1965 skb->mark = sk->sk_mark;
1966 skb->tstamp = sockc.transmit_time;
1968 skb_setup_tx_timestamp(skb, sockc.tsflags);
1970 if (unlikely(extra_len == 4))
1973 skb_probe_transport_header(skb, 0);
1975 dev_queue_xmit(skb);
1986 static unsigned int run_filter(struct sk_buff *skb,
1987 const struct sock *sk,
1990 struct sk_filter *filter;
1993 filter = rcu_dereference(sk->sk_filter);
1995 res = bpf_prog_run_clear_cb(filter->prog, skb);
2001 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2004 struct virtio_net_hdr vnet_hdr;
2006 if (*len < sizeof(vnet_hdr))
2008 *len -= sizeof(vnet_hdr);
2010 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2013 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2017 * This function makes lazy skb cloning in hope that most of packets
2018 * are discarded by BPF.
2020 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2021 * and skb->cb are mangled. It works because (and until) packets
2022 * falling here are owned by current CPU. Output packets are cloned
2023 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2024 * sequencially, so that if we return skb to original state on exit,
2025 * we will not harm anyone.
2028 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2029 struct packet_type *pt, struct net_device *orig_dev)
2032 struct sockaddr_ll *sll;
2033 struct packet_sock *po;
2034 u8 *skb_head = skb->data;
2035 int skb_len = skb->len;
2036 unsigned int snaplen, res;
2037 bool is_drop_n_account = false;
2039 if (skb->pkt_type == PACKET_LOOPBACK)
2042 sk = pt->af_packet_priv;
2045 if (!net_eq(dev_net(dev), sock_net(sk)))
2050 if (dev->header_ops) {
2051 /* The device has an explicit notion of ll header,
2052 * exported to higher levels.
2054 * Otherwise, the device hides details of its frame
2055 * structure, so that corresponding packet head is
2056 * never delivered to user.
2058 if (sk->sk_type != SOCK_DGRAM)
2059 skb_push(skb, skb->data - skb_mac_header(skb));
2060 else if (skb->pkt_type == PACKET_OUTGOING) {
2061 /* Special case: outgoing packets have ll header at head */
2062 skb_pull(skb, skb_network_offset(skb));
2068 res = run_filter(skb, sk, snaplen);
2070 goto drop_n_restore;
2074 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2077 if (skb_shared(skb)) {
2078 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2082 if (skb_head != skb->data) {
2083 skb->data = skb_head;
2090 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2092 sll = &PACKET_SKB_CB(skb)->sa.ll;
2093 sll->sll_hatype = dev->type;
2094 sll->sll_pkttype = skb->pkt_type;
2095 if (unlikely(po->origdev))
2096 sll->sll_ifindex = orig_dev->ifindex;
2098 sll->sll_ifindex = dev->ifindex;
2100 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2102 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2103 * Use their space for storing the original skb length.
2105 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2107 if (pskb_trim(skb, snaplen))
2110 skb_set_owner_r(skb, sk);
2114 /* drop conntrack reference */
2117 spin_lock(&sk->sk_receive_queue.lock);
2118 po->stats.stats1.tp_packets++;
2119 sock_skb_set_dropcount(sk, skb);
2120 __skb_queue_tail(&sk->sk_receive_queue, skb);
2121 spin_unlock(&sk->sk_receive_queue.lock);
2122 sk->sk_data_ready(sk);
2126 is_drop_n_account = true;
2127 spin_lock(&sk->sk_receive_queue.lock);
2128 po->stats.stats1.tp_drops++;
2129 atomic_inc(&sk->sk_drops);
2130 spin_unlock(&sk->sk_receive_queue.lock);
2133 if (skb_head != skb->data && skb_shared(skb)) {
2134 skb->data = skb_head;
2138 if (!is_drop_n_account)
2145 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2146 struct packet_type *pt, struct net_device *orig_dev)
2149 struct packet_sock *po;
2150 struct sockaddr_ll *sll;
2151 union tpacket_uhdr h;
2152 u8 *skb_head = skb->data;
2153 int skb_len = skb->len;
2154 unsigned int snaplen, res;
2155 unsigned long status = TP_STATUS_USER;
2156 unsigned short macoff, netoff, hdrlen;
2157 struct sk_buff *copy_skb = NULL;
2160 bool is_drop_n_account = false;
2161 bool do_vnet = false;
2163 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2164 * We may add members to them until current aligned size without forcing
2165 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2167 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2168 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2170 if (skb->pkt_type == PACKET_LOOPBACK)
2173 sk = pt->af_packet_priv;
2176 if (!net_eq(dev_net(dev), sock_net(sk)))
2179 if (dev->header_ops) {
2180 if (sk->sk_type != SOCK_DGRAM)
2181 skb_push(skb, skb->data - skb_mac_header(skb));
2182 else if (skb->pkt_type == PACKET_OUTGOING) {
2183 /* Special case: outgoing packets have ll header at head */
2184 skb_pull(skb, skb_network_offset(skb));
2190 res = run_filter(skb, sk, snaplen);
2192 goto drop_n_restore;
2194 if (skb->ip_summed == CHECKSUM_PARTIAL)
2195 status |= TP_STATUS_CSUMNOTREADY;
2196 else if (skb->pkt_type != PACKET_OUTGOING &&
2197 (skb->ip_summed == CHECKSUM_COMPLETE ||
2198 skb_csum_unnecessary(skb)))
2199 status |= TP_STATUS_CSUM_VALID;
2204 if (sk->sk_type == SOCK_DGRAM) {
2205 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2208 unsigned int maclen = skb_network_offset(skb);
2209 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2210 (maclen < 16 ? 16 : maclen)) +
2212 if (po->has_vnet_hdr) {
2213 netoff += sizeof(struct virtio_net_hdr);
2216 macoff = netoff - maclen;
2218 if (po->tp_version <= TPACKET_V2) {
2219 if (macoff + snaplen > po->rx_ring.frame_size) {
2220 if (po->copy_thresh &&
2221 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2222 if (skb_shared(skb)) {
2223 copy_skb = skb_clone(skb, GFP_ATOMIC);
2225 copy_skb = skb_get(skb);
2226 skb_head = skb->data;
2229 skb_set_owner_r(copy_skb, sk);
2231 snaplen = po->rx_ring.frame_size - macoff;
2232 if ((int)snaplen < 0) {
2237 } else if (unlikely(macoff + snaplen >
2238 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2241 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2242 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2243 snaplen, nval, macoff);
2245 if (unlikely((int)snaplen < 0)) {
2247 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2251 spin_lock(&sk->sk_receive_queue.lock);
2252 h.raw = packet_current_rx_frame(po, skb,
2253 TP_STATUS_KERNEL, (macoff+snaplen));
2255 goto drop_n_account;
2256 if (po->tp_version <= TPACKET_V2) {
2257 packet_increment_rx_head(po, &po->rx_ring);
2259 * LOSING will be reported till you read the stats,
2260 * because it's COR - Clear On Read.
2261 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2264 if (po->stats.stats1.tp_drops)
2265 status |= TP_STATUS_LOSING;
2269 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2270 sizeof(struct virtio_net_hdr),
2272 goto drop_n_account;
2274 po->stats.stats1.tp_packets++;
2276 status |= TP_STATUS_COPY;
2277 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2279 spin_unlock(&sk->sk_receive_queue.lock);
2281 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2283 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2284 getnstimeofday(&ts);
2286 status |= ts_status;
2288 switch (po->tp_version) {
2290 h.h1->tp_len = skb->len;
2291 h.h1->tp_snaplen = snaplen;
2292 h.h1->tp_mac = macoff;
2293 h.h1->tp_net = netoff;
2294 h.h1->tp_sec = ts.tv_sec;
2295 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2296 hdrlen = sizeof(*h.h1);
2299 h.h2->tp_len = skb->len;
2300 h.h2->tp_snaplen = snaplen;
2301 h.h2->tp_mac = macoff;
2302 h.h2->tp_net = netoff;
2303 h.h2->tp_sec = ts.tv_sec;
2304 h.h2->tp_nsec = ts.tv_nsec;
2305 if (skb_vlan_tag_present(skb)) {
2306 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2307 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2308 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2310 h.h2->tp_vlan_tci = 0;
2311 h.h2->tp_vlan_tpid = 0;
2313 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2314 hdrlen = sizeof(*h.h2);
2317 /* tp_nxt_offset,vlan are already populated above.
2318 * So DONT clear those fields here
2320 h.h3->tp_status |= status;
2321 h.h3->tp_len = skb->len;
2322 h.h3->tp_snaplen = snaplen;
2323 h.h3->tp_mac = macoff;
2324 h.h3->tp_net = netoff;
2325 h.h3->tp_sec = ts.tv_sec;
2326 h.h3->tp_nsec = ts.tv_nsec;
2327 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2328 hdrlen = sizeof(*h.h3);
2334 sll = h.raw + TPACKET_ALIGN(hdrlen);
2335 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2336 sll->sll_family = AF_PACKET;
2337 sll->sll_hatype = dev->type;
2338 sll->sll_protocol = skb->protocol;
2339 sll->sll_pkttype = skb->pkt_type;
2340 if (unlikely(po->origdev))
2341 sll->sll_ifindex = orig_dev->ifindex;
2343 sll->sll_ifindex = dev->ifindex;
2347 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2348 if (po->tp_version <= TPACKET_V2) {
2351 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2354 for (start = h.raw; start < end; start += PAGE_SIZE)
2355 flush_dcache_page(pgv_to_page(start));
2360 if (po->tp_version <= TPACKET_V2) {
2361 __packet_set_status(po, h.raw, status);
2362 sk->sk_data_ready(sk);
2364 prb_clear_blk_fill_status(&po->rx_ring);
2368 if (skb_head != skb->data && skb_shared(skb)) {
2369 skb->data = skb_head;
2373 if (!is_drop_n_account)
2380 is_drop_n_account = true;
2381 po->stats.stats1.tp_drops++;
2382 spin_unlock(&sk->sk_receive_queue.lock);
2384 sk->sk_data_ready(sk);
2385 kfree_skb(copy_skb);
2386 goto drop_n_restore;
2389 static void tpacket_destruct_skb(struct sk_buff *skb)
2391 struct packet_sock *po = pkt_sk(skb->sk);
2393 if (likely(po->tx_ring.pg_vec)) {
2397 ph = skb_zcopy_get_nouarg(skb);
2398 packet_dec_pending(&po->tx_ring);
2400 ts = __packet_set_timestamp(po, ph, skb);
2401 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2407 static void tpacket_set_protocol(const struct net_device *dev,
2408 struct sk_buff *skb)
2410 if (dev->type == ARPHRD_ETHER) {
2411 skb_reset_mac_header(skb);
2412 skb->protocol = eth_hdr(skb)->h_proto;
2416 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2418 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2419 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2420 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2421 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2422 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2423 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2424 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2426 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2432 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2433 struct virtio_net_hdr *vnet_hdr)
2435 if (*len < sizeof(*vnet_hdr))
2437 *len -= sizeof(*vnet_hdr);
2439 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2442 return __packet_snd_vnet_parse(vnet_hdr, *len);
2445 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2446 void *frame, struct net_device *dev, void *data, int tp_len,
2447 __be16 proto, unsigned char *addr, int hlen, int copylen,
2448 const struct sockcm_cookie *sockc)
2450 union tpacket_uhdr ph;
2451 int to_write, offset, len, nr_frags, len_max;
2452 struct socket *sock = po->sk.sk_socket;
2458 skb->protocol = proto;
2460 skb->priority = po->sk.sk_priority;
2461 skb->mark = po->sk.sk_mark;
2462 skb->tstamp = sockc->transmit_time;
2463 skb_setup_tx_timestamp(skb, sockc->tsflags);
2464 skb_zcopy_set_nouarg(skb, ph.raw);
2466 skb_reserve(skb, hlen);
2467 skb_reset_network_header(skb);
2471 if (sock->type == SOCK_DGRAM) {
2472 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2474 if (unlikely(err < 0))
2476 } else if (copylen) {
2477 int hdrlen = min_t(int, copylen, tp_len);
2479 skb_push(skb, dev->hard_header_len);
2480 skb_put(skb, copylen - dev->hard_header_len);
2481 err = skb_store_bits(skb, 0, data, hdrlen);
2484 if (!dev_validate_header(dev, skb->data, hdrlen))
2487 tpacket_set_protocol(dev, skb);
2493 offset = offset_in_page(data);
2494 len_max = PAGE_SIZE - offset;
2495 len = ((to_write > len_max) ? len_max : to_write);
2497 skb->data_len = to_write;
2498 skb->len += to_write;
2499 skb->truesize += to_write;
2500 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2502 while (likely(to_write)) {
2503 nr_frags = skb_shinfo(skb)->nr_frags;
2505 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2506 pr_err("Packet exceed the number of skb frags(%lu)\n",
2511 page = pgv_to_page(data);
2513 flush_dcache_page(page);
2515 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2518 len_max = PAGE_SIZE;
2519 len = ((to_write > len_max) ? len_max : to_write);
2522 skb_probe_transport_header(skb, 0);
2527 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2528 int size_max, void **data)
2530 union tpacket_uhdr ph;
2535 switch (po->tp_version) {
2537 if (ph.h3->tp_next_offset != 0) {
2538 pr_warn_once("variable sized slot not supported");
2541 tp_len = ph.h3->tp_len;
2544 tp_len = ph.h2->tp_len;
2547 tp_len = ph.h1->tp_len;
2550 if (unlikely(tp_len > size_max)) {
2551 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2555 if (unlikely(po->tp_tx_has_off)) {
2556 int off_min, off_max;
2558 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2559 off_max = po->tx_ring.frame_size - tp_len;
2560 if (po->sk.sk_type == SOCK_DGRAM) {
2561 switch (po->tp_version) {
2563 off = ph.h3->tp_net;
2566 off = ph.h2->tp_net;
2569 off = ph.h1->tp_net;
2573 switch (po->tp_version) {
2575 off = ph.h3->tp_mac;
2578 off = ph.h2->tp_mac;
2581 off = ph.h1->tp_mac;
2585 if (unlikely((off < off_min) || (off_max < off)))
2588 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2591 *data = frame + off;
2595 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2597 struct sk_buff *skb;
2598 struct net_device *dev;
2599 struct virtio_net_hdr *vnet_hdr = NULL;
2600 struct sockcm_cookie sockc;
2602 int err, reserve = 0;
2604 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2605 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2606 int tp_len, size_max;
2607 unsigned char *addr;
2610 int status = TP_STATUS_AVAILABLE;
2611 int hlen, tlen, copylen = 0;
2613 mutex_lock(&po->pg_vec_lock);
2615 if (likely(saddr == NULL)) {
2616 dev = packet_cached_dev_get(po);
2621 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2623 if (msg->msg_namelen < (saddr->sll_halen
2624 + offsetof(struct sockaddr_ll,
2627 proto = saddr->sll_protocol;
2628 addr = saddr->sll_halen ? saddr->sll_addr : NULL;
2629 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2630 if (addr && dev && saddr->sll_halen < dev->addr_len)
2635 if (unlikely(dev == NULL))
2638 if (unlikely(!(dev->flags & IFF_UP)))
2641 sockcm_init(&sockc, &po->sk);
2642 if (msg->msg_controllen) {
2643 err = sock_cmsg_send(&po->sk, msg, &sockc);
2648 if (po->sk.sk_socket->type == SOCK_RAW)
2649 reserve = dev->hard_header_len;
2650 size_max = po->tx_ring.frame_size
2651 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2653 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2654 size_max = dev->mtu + reserve + VLAN_HLEN;
2657 ph = packet_current_frame(po, &po->tx_ring,
2658 TP_STATUS_SEND_REQUEST);
2659 if (unlikely(ph == NULL)) {
2660 if (need_wait && need_resched())
2666 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2670 status = TP_STATUS_SEND_REQUEST;
2671 hlen = LL_RESERVED_SPACE(dev);
2672 tlen = dev->needed_tailroom;
2673 if (po->has_vnet_hdr) {
2675 data += sizeof(*vnet_hdr);
2676 tp_len -= sizeof(*vnet_hdr);
2678 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2682 copylen = __virtio16_to_cpu(vio_le(),
2685 copylen = max_t(int, copylen, dev->hard_header_len);
2686 skb = sock_alloc_send_skb(&po->sk,
2687 hlen + tlen + sizeof(struct sockaddr_ll) +
2688 (copylen - dev->hard_header_len),
2691 if (unlikely(skb == NULL)) {
2692 /* we assume the socket was initially writeable ... */
2693 if (likely(len_sum > 0))
2697 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2698 addr, hlen, copylen, &sockc);
2699 if (likely(tp_len >= 0) &&
2700 tp_len > dev->mtu + reserve &&
2701 !po->has_vnet_hdr &&
2702 !packet_extra_vlan_len_allowed(dev, skb))
2705 if (unlikely(tp_len < 0)) {
2708 __packet_set_status(po, ph,
2709 TP_STATUS_AVAILABLE);
2710 packet_increment_head(&po->tx_ring);
2714 status = TP_STATUS_WRONG_FORMAT;
2720 if (po->has_vnet_hdr) {
2721 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2725 virtio_net_hdr_set_proto(skb, vnet_hdr);
2728 skb->destructor = tpacket_destruct_skb;
2729 __packet_set_status(po, ph, TP_STATUS_SENDING);
2730 packet_inc_pending(&po->tx_ring);
2732 status = TP_STATUS_SEND_REQUEST;
2733 err = po->xmit(skb);
2734 if (unlikely(err > 0)) {
2735 err = net_xmit_errno(err);
2736 if (err && __packet_get_status(po, ph) ==
2737 TP_STATUS_AVAILABLE) {
2738 /* skb was destructed already */
2743 * skb was dropped but not destructed yet;
2744 * let's treat it like congestion or err < 0
2748 packet_increment_head(&po->tx_ring);
2750 } while (likely((ph != NULL) ||
2751 /* Note: packet_read_pending() might be slow if we have
2752 * to call it as it's per_cpu variable, but in fast-path
2753 * we already short-circuit the loop with the first
2754 * condition, and luckily don't have to go that path
2757 (need_wait && packet_read_pending(&po->tx_ring))));
2763 __packet_set_status(po, ph, status);
2768 mutex_unlock(&po->pg_vec_lock);
2772 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2773 size_t reserve, size_t len,
2774 size_t linear, int noblock,
2777 struct sk_buff *skb;
2779 /* Under a page? Don't bother with paged skb. */
2780 if (prepad + len < PAGE_SIZE || !linear)
2783 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2788 skb_reserve(skb, reserve);
2789 skb_put(skb, linear);
2790 skb->data_len = len - linear;
2791 skb->len += len - linear;
2796 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2798 struct sock *sk = sock->sk;
2799 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2800 struct sk_buff *skb;
2801 struct net_device *dev;
2803 unsigned char *addr;
2804 int err, reserve = 0;
2805 struct sockcm_cookie sockc;
2806 struct virtio_net_hdr vnet_hdr = { 0 };
2808 struct packet_sock *po = pkt_sk(sk);
2809 bool has_vnet_hdr = false;
2810 int hlen, tlen, linear;
2814 * Get and verify the address.
2817 if (likely(saddr == NULL)) {
2818 dev = packet_cached_dev_get(po);
2823 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2825 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2827 proto = saddr->sll_protocol;
2828 addr = saddr->sll_halen ? saddr->sll_addr : NULL;
2829 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2830 if (addr && dev && saddr->sll_halen < dev->addr_len)
2835 if (unlikely(dev == NULL))
2838 if (unlikely(!(dev->flags & IFF_UP)))
2841 sockcm_init(&sockc, sk);
2842 sockc.mark = sk->sk_mark;
2843 if (msg->msg_controllen) {
2844 err = sock_cmsg_send(sk, msg, &sockc);
2849 if (sock->type == SOCK_RAW)
2850 reserve = dev->hard_header_len;
2851 if (po->has_vnet_hdr) {
2852 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2855 has_vnet_hdr = true;
2858 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2859 if (!netif_supports_nofcs(dev)) {
2860 err = -EPROTONOSUPPORT;
2863 extra_len = 4; /* We're doing our own CRC */
2867 if (!vnet_hdr.gso_type &&
2868 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2872 hlen = LL_RESERVED_SPACE(dev);
2873 tlen = dev->needed_tailroom;
2874 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2875 linear = max(linear, min_t(int, len, dev->hard_header_len));
2876 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2877 msg->msg_flags & MSG_DONTWAIT, &err);
2881 skb_reset_network_header(skb);
2884 if (sock->type == SOCK_DGRAM) {
2885 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2886 if (unlikely(offset < 0))
2888 } else if (reserve) {
2889 skb_reserve(skb, -reserve);
2891 skb_reset_network_header(skb);
2894 /* Returns -EFAULT on error */
2895 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2899 if (sock->type == SOCK_RAW &&
2900 !dev_validate_header(dev, skb->data, len)) {
2905 skb_setup_tx_timestamp(skb, sockc.tsflags);
2907 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2908 !packet_extra_vlan_len_allowed(dev, skb)) {
2913 skb->protocol = proto;
2915 skb->priority = sk->sk_priority;
2916 skb->mark = sockc.mark;
2917 skb->tstamp = sockc.transmit_time;
2920 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2923 len += sizeof(vnet_hdr);
2924 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2927 skb_probe_transport_header(skb, reserve);
2929 if (unlikely(extra_len == 4))
2932 err = po->xmit(skb);
2933 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2949 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2951 struct sock *sk = sock->sk;
2952 struct packet_sock *po = pkt_sk(sk);
2954 if (po->tx_ring.pg_vec)
2955 return tpacket_snd(po, msg);
2957 return packet_snd(sock, msg, len);
2961 * Close a PACKET socket. This is fairly simple. We immediately go
2962 * to 'closed' state and remove our protocol entry in the device list.
2965 static int packet_release(struct socket *sock)
2967 struct sock *sk = sock->sk;
2968 struct packet_sock *po;
2969 struct packet_fanout *f;
2971 union tpacket_req_u req_u;
2979 mutex_lock(&net->packet.sklist_lock);
2980 sk_del_node_init_rcu(sk);
2981 mutex_unlock(&net->packet.sklist_lock);
2984 sock_prot_inuse_add(net, sk->sk_prot, -1);
2987 spin_lock(&po->bind_lock);
2988 unregister_prot_hook(sk, false);
2989 packet_cached_dev_reset(po);
2991 if (po->prot_hook.dev) {
2992 dev_put(po->prot_hook.dev);
2993 po->prot_hook.dev = NULL;
2995 spin_unlock(&po->bind_lock);
2997 packet_flush_mclist(sk);
3000 if (po->rx_ring.pg_vec) {
3001 memset(&req_u, 0, sizeof(req_u));
3002 packet_set_ring(sk, &req_u, 1, 0);
3005 if (po->tx_ring.pg_vec) {
3006 memset(&req_u, 0, sizeof(req_u));
3007 packet_set_ring(sk, &req_u, 1, 1);
3011 f = fanout_release(sk);
3016 kfree(po->rollover);
3017 fanout_release_data(f);
3021 * Now the socket is dead. No more input will appear.
3028 skb_queue_purge(&sk->sk_receive_queue);
3029 packet_free_pending(po);
3030 sk_refcnt_debug_release(sk);
3037 * Attach a packet hook.
3040 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3043 struct packet_sock *po = pkt_sk(sk);
3044 struct net_device *dev_curr;
3047 struct net_device *dev = NULL;
3049 bool unlisted = false;
3052 spin_lock(&po->bind_lock);
3061 dev = dev_get_by_name_rcu(sock_net(sk), name);
3066 } else if (ifindex) {
3067 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3077 proto_curr = po->prot_hook.type;
3078 dev_curr = po->prot_hook.dev;
3080 need_rehook = proto_curr != proto || dev_curr != dev;
3085 /* prevents packet_notifier() from calling
3086 * register_prot_hook()
3089 __unregister_prot_hook(sk, true);
3091 dev_curr = po->prot_hook.dev;
3093 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3097 BUG_ON(po->running);
3099 po->prot_hook.type = proto;
3101 if (unlikely(unlisted)) {
3103 po->prot_hook.dev = NULL;
3105 packet_cached_dev_reset(po);
3107 po->prot_hook.dev = dev;
3108 po->ifindex = dev ? dev->ifindex : 0;
3109 packet_cached_dev_assign(po, dev);
3115 if (proto == 0 || !need_rehook)
3118 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3119 register_prot_hook(sk);
3121 sk->sk_err = ENETDOWN;
3122 if (!sock_flag(sk, SOCK_DEAD))
3123 sk->sk_error_report(sk);
3128 spin_unlock(&po->bind_lock);
3134 * Bind a packet socket to a device
3137 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3140 struct sock *sk = sock->sk;
3141 char name[sizeof(uaddr->sa_data) + 1];
3147 if (addr_len != sizeof(struct sockaddr))
3149 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3152 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3153 name[sizeof(uaddr->sa_data)] = 0;
3155 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3158 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3160 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3161 struct sock *sk = sock->sk;
3167 if (addr_len < sizeof(struct sockaddr_ll))
3169 if (sll->sll_family != AF_PACKET)
3172 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3173 sll->sll_protocol ? : pkt_sk(sk)->num);
3176 static struct proto packet_proto = {
3178 .owner = THIS_MODULE,
3179 .obj_size = sizeof(struct packet_sock),
3183 * Create a packet of type SOCK_PACKET.
3186 static int packet_create(struct net *net, struct socket *sock, int protocol,
3190 struct packet_sock *po;
3191 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3194 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3196 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3197 sock->type != SOCK_PACKET)
3198 return -ESOCKTNOSUPPORT;
3200 sock->state = SS_UNCONNECTED;
3203 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3207 sock->ops = &packet_ops;
3208 if (sock->type == SOCK_PACKET)
3209 sock->ops = &packet_ops_spkt;
3211 sock_init_data(sock, sk);
3214 sk->sk_family = PF_PACKET;
3216 po->xmit = dev_queue_xmit;
3218 err = packet_alloc_pending(po);
3222 packet_cached_dev_reset(po);
3224 sk->sk_destruct = packet_sock_destruct;
3225 sk_refcnt_debug_inc(sk);
3228 * Attach a protocol block
3231 spin_lock_init(&po->bind_lock);
3232 mutex_init(&po->pg_vec_lock);
3233 po->rollover = NULL;
3234 po->prot_hook.func = packet_rcv;
3236 if (sock->type == SOCK_PACKET)
3237 po->prot_hook.func = packet_rcv_spkt;
3239 po->prot_hook.af_packet_priv = sk;
3242 po->prot_hook.type = proto;
3243 __register_prot_hook(sk);
3246 mutex_lock(&net->packet.sklist_lock);
3247 sk_add_node_rcu(sk, &net->packet.sklist);
3248 mutex_unlock(&net->packet.sklist_lock);
3251 sock_prot_inuse_add(net, &packet_proto, 1);
3262 * Pull a packet from our receive queue and hand it to the user.
3263 * If necessary we block.
3266 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3269 struct sock *sk = sock->sk;
3270 struct sk_buff *skb;
3272 int vnet_hdr_len = 0;
3273 unsigned int origlen = 0;
3276 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3280 /* What error should we return now? EUNATTACH? */
3281 if (pkt_sk(sk)->ifindex < 0)
3285 if (flags & MSG_ERRQUEUE) {
3286 err = sock_recv_errqueue(sk, msg, len,
3287 SOL_PACKET, PACKET_TX_TIMESTAMP);
3292 * Call the generic datagram receiver. This handles all sorts
3293 * of horrible races and re-entrancy so we can forget about it
3294 * in the protocol layers.
3296 * Now it will return ENETDOWN, if device have just gone down,
3297 * but then it will block.
3300 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3303 * An error occurred so return it. Because skb_recv_datagram()
3304 * handles the blocking we don't see and worry about blocking
3311 if (pkt_sk(sk)->pressure)
3312 packet_rcv_has_room(pkt_sk(sk), NULL);
3314 if (pkt_sk(sk)->has_vnet_hdr) {
3315 err = packet_rcv_vnet(msg, skb, &len);
3318 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3321 /* You lose any data beyond the buffer you gave. If it worries
3322 * a user program they can ask the device for its MTU
3328 msg->msg_flags |= MSG_TRUNC;
3331 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3335 if (sock->type != SOCK_PACKET) {
3336 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3338 /* Original length was stored in sockaddr_ll fields */
3339 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3340 sll->sll_family = AF_PACKET;
3341 sll->sll_protocol = skb->protocol;
3344 sock_recv_ts_and_drops(msg, sk, skb);
3346 if (msg->msg_name) {
3347 /* If the address length field is there to be filled
3348 * in, we fill it in now.
3350 if (sock->type == SOCK_PACKET) {
3351 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3352 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3354 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3356 msg->msg_namelen = sll->sll_halen +
3357 offsetof(struct sockaddr_ll, sll_addr);
3359 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3363 if (pkt_sk(sk)->auxdata) {
3364 struct tpacket_auxdata aux;
3366 aux.tp_status = TP_STATUS_USER;
3367 if (skb->ip_summed == CHECKSUM_PARTIAL)
3368 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3369 else if (skb->pkt_type != PACKET_OUTGOING &&
3370 (skb->ip_summed == CHECKSUM_COMPLETE ||
3371 skb_csum_unnecessary(skb)))
3372 aux.tp_status |= TP_STATUS_CSUM_VALID;
3374 aux.tp_len = origlen;
3375 aux.tp_snaplen = skb->len;
3377 aux.tp_net = skb_network_offset(skb);
3378 if (skb_vlan_tag_present(skb)) {
3379 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3380 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3381 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3383 aux.tp_vlan_tci = 0;
3384 aux.tp_vlan_tpid = 0;
3386 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3390 * Free or return the buffer as appropriate. Again this
3391 * hides all the races and re-entrancy issues from us.
3393 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3396 skb_free_datagram(sk, skb);
3401 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3404 struct net_device *dev;
3405 struct sock *sk = sock->sk;
3410 uaddr->sa_family = AF_PACKET;
3411 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3413 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3415 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3418 return sizeof(*uaddr);
3421 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3424 struct net_device *dev;
3425 struct sock *sk = sock->sk;
3426 struct packet_sock *po = pkt_sk(sk);
3427 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3432 sll->sll_family = AF_PACKET;
3433 sll->sll_ifindex = po->ifindex;
3434 sll->sll_protocol = po->num;
3435 sll->sll_pkttype = 0;
3437 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3439 sll->sll_hatype = dev->type;
3440 sll->sll_halen = dev->addr_len;
3441 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3443 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3448 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3451 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3455 case PACKET_MR_MULTICAST:
3456 if (i->alen != dev->addr_len)
3459 return dev_mc_add(dev, i->addr);
3461 return dev_mc_del(dev, i->addr);
3463 case PACKET_MR_PROMISC:
3464 return dev_set_promiscuity(dev, what);
3465 case PACKET_MR_ALLMULTI:
3466 return dev_set_allmulti(dev, what);
3467 case PACKET_MR_UNICAST:
3468 if (i->alen != dev->addr_len)
3471 return dev_uc_add(dev, i->addr);
3473 return dev_uc_del(dev, i->addr);
3481 static void packet_dev_mclist_delete(struct net_device *dev,
3482 struct packet_mclist **mlp)
3484 struct packet_mclist *ml;
3486 while ((ml = *mlp) != NULL) {
3487 if (ml->ifindex == dev->ifindex) {
3488 packet_dev_mc(dev, ml, -1);
3496 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3498 struct packet_sock *po = pkt_sk(sk);
3499 struct packet_mclist *ml, *i;
3500 struct net_device *dev;
3506 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3511 if (mreq->mr_alen > dev->addr_len)
3515 i = kmalloc(sizeof(*i), GFP_KERNEL);
3520 for (ml = po->mclist; ml; ml = ml->next) {
3521 if (ml->ifindex == mreq->mr_ifindex &&
3522 ml->type == mreq->mr_type &&
3523 ml->alen == mreq->mr_alen &&
3524 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3526 /* Free the new element ... */
3532 i->type = mreq->mr_type;
3533 i->ifindex = mreq->mr_ifindex;
3534 i->alen = mreq->mr_alen;
3535 memcpy(i->addr, mreq->mr_address, i->alen);
3536 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3538 i->next = po->mclist;
3540 err = packet_dev_mc(dev, i, 1);
3542 po->mclist = i->next;
3551 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3553 struct packet_mclist *ml, **mlp;
3557 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3558 if (ml->ifindex == mreq->mr_ifindex &&
3559 ml->type == mreq->mr_type &&
3560 ml->alen == mreq->mr_alen &&
3561 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3562 if (--ml->count == 0) {
3563 struct net_device *dev;
3565 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3567 packet_dev_mc(dev, ml, -1);
3577 static void packet_flush_mclist(struct sock *sk)
3579 struct packet_sock *po = pkt_sk(sk);
3580 struct packet_mclist *ml;
3586 while ((ml = po->mclist) != NULL) {
3587 struct net_device *dev;
3589 po->mclist = ml->next;
3590 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3592 packet_dev_mc(dev, ml, -1);
3599 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3601 struct sock *sk = sock->sk;
3602 struct packet_sock *po = pkt_sk(sk);
3605 if (level != SOL_PACKET)
3606 return -ENOPROTOOPT;
3609 case PACKET_ADD_MEMBERSHIP:
3610 case PACKET_DROP_MEMBERSHIP:
3612 struct packet_mreq_max mreq;
3614 memset(&mreq, 0, sizeof(mreq));
3615 if (len < sizeof(struct packet_mreq))
3617 if (len > sizeof(mreq))
3619 if (copy_from_user(&mreq, optval, len))
3621 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3623 if (optname == PACKET_ADD_MEMBERSHIP)
3624 ret = packet_mc_add(sk, &mreq);
3626 ret = packet_mc_drop(sk, &mreq);
3630 case PACKET_RX_RING:
3631 case PACKET_TX_RING:
3633 union tpacket_req_u req_u;
3637 switch (po->tp_version) {
3640 len = sizeof(req_u.req);
3644 len = sizeof(req_u.req3);
3650 if (copy_from_user(&req_u.req, optval, len))
3653 ret = packet_set_ring(sk, &req_u, 0,
3654 optname == PACKET_TX_RING);
3659 case PACKET_COPY_THRESH:
3663 if (optlen != sizeof(val))
3665 if (copy_from_user(&val, optval, sizeof(val)))
3668 pkt_sk(sk)->copy_thresh = val;
3671 case PACKET_VERSION:
3675 if (optlen != sizeof(val))
3677 if (copy_from_user(&val, optval, sizeof(val)))
3688 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3691 po->tp_version = val;
3697 case PACKET_RESERVE:
3701 if (optlen != sizeof(val))
3703 if (copy_from_user(&val, optval, sizeof(val)))
3708 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3711 po->tp_reserve = val;
3721 if (optlen != sizeof(val))
3723 if (copy_from_user(&val, optval, sizeof(val)))
3727 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3730 po->tp_loss = !!val;
3736 case PACKET_AUXDATA:
3740 if (optlen < sizeof(val))
3742 if (copy_from_user(&val, optval, sizeof(val)))
3746 po->auxdata = !!val;
3750 case PACKET_ORIGDEV:
3754 if (optlen < sizeof(val))
3756 if (copy_from_user(&val, optval, sizeof(val)))
3760 po->origdev = !!val;
3764 case PACKET_VNET_HDR:
3768 if (sock->type != SOCK_RAW)
3770 if (optlen < sizeof(val))
3772 if (copy_from_user(&val, optval, sizeof(val)))
3776 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3779 po->has_vnet_hdr = !!val;
3785 case PACKET_TIMESTAMP:
3789 if (optlen != sizeof(val))
3791 if (copy_from_user(&val, optval, sizeof(val)))
3794 po->tp_tstamp = val;
3801 if (optlen != sizeof(val))
3803 if (copy_from_user(&val, optval, sizeof(val)))
3806 return fanout_add(sk, val & 0xffff, val >> 16);
3808 case PACKET_FANOUT_DATA:
3813 return fanout_set_data(po, optval, optlen);
3815 case PACKET_IGNORE_OUTGOING:
3819 if (optlen != sizeof(val))
3821 if (copy_from_user(&val, optval, sizeof(val)))
3823 if (val < 0 || val > 1)
3826 po->prot_hook.ignore_outgoing = !!val;
3829 case PACKET_TX_HAS_OFF:
3833 if (optlen != sizeof(val))
3835 if (copy_from_user(&val, optval, sizeof(val)))
3839 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3842 po->tp_tx_has_off = !!val;
3848 case PACKET_QDISC_BYPASS:
3852 if (optlen != sizeof(val))
3854 if (copy_from_user(&val, optval, sizeof(val)))
3857 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3861 return -ENOPROTOOPT;
3865 static int packet_getsockopt(struct socket *sock, int level, int optname,
3866 char __user *optval, int __user *optlen)
3869 int val, lv = sizeof(val);
3870 struct sock *sk = sock->sk;
3871 struct packet_sock *po = pkt_sk(sk);
3873 union tpacket_stats_u st;
3874 struct tpacket_rollover_stats rstats;
3876 if (level != SOL_PACKET)
3877 return -ENOPROTOOPT;
3879 if (get_user(len, optlen))
3886 case PACKET_STATISTICS:
3887 spin_lock_bh(&sk->sk_receive_queue.lock);
3888 memcpy(&st, &po->stats, sizeof(st));
3889 memset(&po->stats, 0, sizeof(po->stats));
3890 spin_unlock_bh(&sk->sk_receive_queue.lock);
3892 if (po->tp_version == TPACKET_V3) {
3893 lv = sizeof(struct tpacket_stats_v3);
3894 st.stats3.tp_packets += st.stats3.tp_drops;
3897 lv = sizeof(struct tpacket_stats);
3898 st.stats1.tp_packets += st.stats1.tp_drops;
3903 case PACKET_AUXDATA:
3906 case PACKET_ORIGDEV:
3909 case PACKET_VNET_HDR:
3910 val = po->has_vnet_hdr;
3912 case PACKET_VERSION:
3913 val = po->tp_version;
3916 if (len > sizeof(int))
3918 if (len < sizeof(int))
3920 if (copy_from_user(&val, optval, len))
3924 val = sizeof(struct tpacket_hdr);
3927 val = sizeof(struct tpacket2_hdr);
3930 val = sizeof(struct tpacket3_hdr);
3936 case PACKET_RESERVE:
3937 val = po->tp_reserve;
3942 case PACKET_TIMESTAMP:
3943 val = po->tp_tstamp;
3947 ((u32)po->fanout->id |
3948 ((u32)po->fanout->type << 16) |
3949 ((u32)po->fanout->flags << 24)) :
3952 case PACKET_IGNORE_OUTGOING:
3953 val = po->prot_hook.ignore_outgoing;
3955 case PACKET_ROLLOVER_STATS:
3958 rstats.tp_all = atomic_long_read(&po->rollover->num);
3959 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3960 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3962 lv = sizeof(rstats);
3964 case PACKET_TX_HAS_OFF:
3965 val = po->tp_tx_has_off;
3967 case PACKET_QDISC_BYPASS:
3968 val = packet_use_direct_xmit(po);
3971 return -ENOPROTOOPT;
3976 if (put_user(len, optlen))
3978 if (copy_to_user(optval, data, len))
3984 #ifdef CONFIG_COMPAT
3985 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3986 char __user *optval, unsigned int optlen)
3988 struct packet_sock *po = pkt_sk(sock->sk);
3990 if (level != SOL_PACKET)
3991 return -ENOPROTOOPT;
3993 if (optname == PACKET_FANOUT_DATA &&
3994 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3995 optval = (char __user *)get_compat_bpf_fprog(optval);
3998 optlen = sizeof(struct sock_fprog);
4001 return packet_setsockopt(sock, level, optname, optval, optlen);
4005 static int packet_notifier(struct notifier_block *this,
4006 unsigned long msg, void *ptr)
4009 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4010 struct net *net = dev_net(dev);
4013 sk_for_each_rcu(sk, &net->packet.sklist) {
4014 struct packet_sock *po = pkt_sk(sk);
4017 case NETDEV_UNREGISTER:
4019 packet_dev_mclist_delete(dev, &po->mclist);
4023 if (dev->ifindex == po->ifindex) {
4024 spin_lock(&po->bind_lock);
4026 __unregister_prot_hook(sk, false);
4027 sk->sk_err = ENETDOWN;
4028 if (!sock_flag(sk, SOCK_DEAD))
4029 sk->sk_error_report(sk);
4031 if (msg == NETDEV_UNREGISTER) {
4032 packet_cached_dev_reset(po);
4034 if (po->prot_hook.dev)
4035 dev_put(po->prot_hook.dev);
4036 po->prot_hook.dev = NULL;
4038 spin_unlock(&po->bind_lock);
4042 if (dev->ifindex == po->ifindex) {
4043 spin_lock(&po->bind_lock);
4045 register_prot_hook(sk);
4046 spin_unlock(&po->bind_lock);
4056 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4059 struct sock *sk = sock->sk;
4064 int amount = sk_wmem_alloc_get(sk);
4066 return put_user(amount, (int __user *)arg);
4070 struct sk_buff *skb;
4073 spin_lock_bh(&sk->sk_receive_queue.lock);
4074 skb = skb_peek(&sk->sk_receive_queue);
4077 spin_unlock_bh(&sk->sk_receive_queue.lock);
4078 return put_user(amount, (int __user *)arg);
4081 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4083 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4093 case SIOCGIFBRDADDR:
4094 case SIOCSIFBRDADDR:
4095 case SIOCGIFNETMASK:
4096 case SIOCSIFNETMASK:
4097 case SIOCGIFDSTADDR:
4098 case SIOCSIFDSTADDR:
4100 return inet_dgram_ops.ioctl(sock, cmd, arg);
4104 return -ENOIOCTLCMD;
4109 static __poll_t packet_poll(struct file *file, struct socket *sock,
4112 struct sock *sk = sock->sk;
4113 struct packet_sock *po = pkt_sk(sk);
4114 __poll_t mask = datagram_poll(file, sock, wait);
4116 spin_lock_bh(&sk->sk_receive_queue.lock);
4117 if (po->rx_ring.pg_vec) {
4118 if (!packet_previous_rx_frame(po, &po->rx_ring,
4120 mask |= EPOLLIN | EPOLLRDNORM;
4122 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4124 spin_unlock_bh(&sk->sk_receive_queue.lock);
4125 spin_lock_bh(&sk->sk_write_queue.lock);
4126 if (po->tx_ring.pg_vec) {
4127 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4128 mask |= EPOLLOUT | EPOLLWRNORM;
4130 spin_unlock_bh(&sk->sk_write_queue.lock);
4135 /* Dirty? Well, I still did not learn better way to account
4139 static void packet_mm_open(struct vm_area_struct *vma)
4141 struct file *file = vma->vm_file;
4142 struct socket *sock = file->private_data;
4143 struct sock *sk = sock->sk;
4146 atomic_inc(&pkt_sk(sk)->mapped);
4149 static void packet_mm_close(struct vm_area_struct *vma)
4151 struct file *file = vma->vm_file;
4152 struct socket *sock = file->private_data;
4153 struct sock *sk = sock->sk;
4156 atomic_dec(&pkt_sk(sk)->mapped);
4159 static const struct vm_operations_struct packet_mmap_ops = {
4160 .open = packet_mm_open,
4161 .close = packet_mm_close,
4164 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4169 for (i = 0; i < len; i++) {
4170 if (likely(pg_vec[i].buffer)) {
4171 if (is_vmalloc_addr(pg_vec[i].buffer))
4172 vfree(pg_vec[i].buffer);
4174 free_pages((unsigned long)pg_vec[i].buffer,
4176 pg_vec[i].buffer = NULL;
4182 static char *alloc_one_pg_vec_page(unsigned long order)
4185 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4186 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4188 buffer = (char *) __get_free_pages(gfp_flags, order);
4192 /* __get_free_pages failed, fall back to vmalloc */
4193 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4197 /* vmalloc failed, lets dig into swap here */
4198 gfp_flags &= ~__GFP_NORETRY;
4199 buffer = (char *) __get_free_pages(gfp_flags, order);
4203 /* complete and utter failure */
4207 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4209 unsigned int block_nr = req->tp_block_nr;
4213 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4214 if (unlikely(!pg_vec))
4217 for (i = 0; i < block_nr; i++) {
4218 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4219 if (unlikely(!pg_vec[i].buffer))
4220 goto out_free_pgvec;
4227 free_pg_vec(pg_vec, order, block_nr);
4232 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4233 int closing, int tx_ring)
4235 struct pgv *pg_vec = NULL;
4236 struct packet_sock *po = pkt_sk(sk);
4237 int was_running, order = 0;
4238 struct packet_ring_buffer *rb;
4239 struct sk_buff_head *rb_queue;
4242 /* Added to avoid minimal code churn */
4243 struct tpacket_req *req = &req_u->req;
4245 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4246 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4250 if (atomic_read(&po->mapped))
4252 if (packet_read_pending(rb))
4256 if (req->tp_block_nr) {
4257 unsigned int min_frame_size;
4259 /* Sanity tests and some calculations */
4261 if (unlikely(rb->pg_vec))
4264 switch (po->tp_version) {
4266 po->tp_hdrlen = TPACKET_HDRLEN;
4269 po->tp_hdrlen = TPACKET2_HDRLEN;
4272 po->tp_hdrlen = TPACKET3_HDRLEN;
4277 if (unlikely((int)req->tp_block_size <= 0))
4279 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4281 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4282 if (po->tp_version >= TPACKET_V3 &&
4283 req->tp_block_size <
4284 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4286 if (unlikely(req->tp_frame_size < min_frame_size))
4288 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4291 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4292 if (unlikely(rb->frames_per_block == 0))
4294 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4296 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4301 order = get_order(req->tp_block_size);
4302 pg_vec = alloc_pg_vec(req, order);
4303 if (unlikely(!pg_vec))
4305 switch (po->tp_version) {
4307 /* Block transmit is not supported yet */
4309 init_prb_bdqc(po, rb, pg_vec, req_u);
4311 struct tpacket_req3 *req3 = &req_u->req3;
4313 if (req3->tp_retire_blk_tov ||
4314 req3->tp_sizeof_priv ||
4315 req3->tp_feature_req_word) {
4328 if (unlikely(req->tp_frame_nr))
4333 /* Detach socket from network */
4334 spin_lock(&po->bind_lock);
4335 was_running = po->running;
4339 __unregister_prot_hook(sk, false);
4341 spin_unlock(&po->bind_lock);
4346 mutex_lock(&po->pg_vec_lock);
4347 if (closing || atomic_read(&po->mapped) == 0) {
4349 spin_lock_bh(&rb_queue->lock);
4350 swap(rb->pg_vec, pg_vec);
4351 rb->frame_max = (req->tp_frame_nr - 1);
4353 rb->frame_size = req->tp_frame_size;
4354 spin_unlock_bh(&rb_queue->lock);
4356 swap(rb->pg_vec_order, order);
4357 swap(rb->pg_vec_len, req->tp_block_nr);
4359 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4360 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4361 tpacket_rcv : packet_rcv;
4362 skb_queue_purge(rb_queue);
4363 if (atomic_read(&po->mapped))
4364 pr_err("packet_mmap: vma is busy: %d\n",
4365 atomic_read(&po->mapped));
4367 mutex_unlock(&po->pg_vec_lock);
4369 spin_lock(&po->bind_lock);
4372 register_prot_hook(sk);
4374 spin_unlock(&po->bind_lock);
4375 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4376 /* Because we don't support block-based V3 on tx-ring */
4378 prb_shutdown_retire_blk_timer(po, rb_queue);
4382 free_pg_vec(pg_vec, order, req->tp_block_nr);
4387 static int packet_mmap(struct file *file, struct socket *sock,
4388 struct vm_area_struct *vma)
4390 struct sock *sk = sock->sk;
4391 struct packet_sock *po = pkt_sk(sk);
4392 unsigned long size, expected_size;
4393 struct packet_ring_buffer *rb;
4394 unsigned long start;
4401 mutex_lock(&po->pg_vec_lock);
4404 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4406 expected_size += rb->pg_vec_len
4412 if (expected_size == 0)
4415 size = vma->vm_end - vma->vm_start;
4416 if (size != expected_size)
4419 start = vma->vm_start;
4420 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4421 if (rb->pg_vec == NULL)
4424 for (i = 0; i < rb->pg_vec_len; i++) {
4426 void *kaddr = rb->pg_vec[i].buffer;
4429 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4430 page = pgv_to_page(kaddr);
4431 err = vm_insert_page(vma, start, page);
4440 atomic_inc(&po->mapped);
4441 vma->vm_ops = &packet_mmap_ops;
4445 mutex_unlock(&po->pg_vec_lock);
4449 static const struct proto_ops packet_ops_spkt = {
4450 .family = PF_PACKET,
4451 .owner = THIS_MODULE,
4452 .release = packet_release,
4453 .bind = packet_bind_spkt,
4454 .connect = sock_no_connect,
4455 .socketpair = sock_no_socketpair,
4456 .accept = sock_no_accept,
4457 .getname = packet_getname_spkt,
4458 .poll = datagram_poll,
4459 .ioctl = packet_ioctl,
4460 .listen = sock_no_listen,
4461 .shutdown = sock_no_shutdown,
4462 .setsockopt = sock_no_setsockopt,
4463 .getsockopt = sock_no_getsockopt,
4464 .sendmsg = packet_sendmsg_spkt,
4465 .recvmsg = packet_recvmsg,
4466 .mmap = sock_no_mmap,
4467 .sendpage = sock_no_sendpage,
4470 static const struct proto_ops packet_ops = {
4471 .family = PF_PACKET,
4472 .owner = THIS_MODULE,
4473 .release = packet_release,
4474 .bind = packet_bind,
4475 .connect = sock_no_connect,
4476 .socketpair = sock_no_socketpair,
4477 .accept = sock_no_accept,
4478 .getname = packet_getname,
4479 .poll = packet_poll,
4480 .ioctl = packet_ioctl,
4481 .listen = sock_no_listen,
4482 .shutdown = sock_no_shutdown,
4483 .setsockopt = packet_setsockopt,
4484 .getsockopt = packet_getsockopt,
4485 #ifdef CONFIG_COMPAT
4486 .compat_setsockopt = compat_packet_setsockopt,
4488 .sendmsg = packet_sendmsg,
4489 .recvmsg = packet_recvmsg,
4490 .mmap = packet_mmap,
4491 .sendpage = sock_no_sendpage,
4494 static const struct net_proto_family packet_family_ops = {
4495 .family = PF_PACKET,
4496 .create = packet_create,
4497 .owner = THIS_MODULE,
4500 static struct notifier_block packet_netdev_notifier = {
4501 .notifier_call = packet_notifier,
4504 #ifdef CONFIG_PROC_FS
4506 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4509 struct net *net = seq_file_net(seq);
4512 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4515 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4517 struct net *net = seq_file_net(seq);
4518 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4521 static void packet_seq_stop(struct seq_file *seq, void *v)
4527 static int packet_seq_show(struct seq_file *seq, void *v)
4529 if (v == SEQ_START_TOKEN)
4530 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4532 struct sock *s = sk_entry(v);
4533 const struct packet_sock *po = pkt_sk(s);
4536 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4538 refcount_read(&s->sk_refcnt),
4543 atomic_read(&s->sk_rmem_alloc),
4544 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4551 static const struct seq_operations packet_seq_ops = {
4552 .start = packet_seq_start,
4553 .next = packet_seq_next,
4554 .stop = packet_seq_stop,
4555 .show = packet_seq_show,
4559 static int __net_init packet_net_init(struct net *net)
4561 mutex_init(&net->packet.sklist_lock);
4562 INIT_HLIST_HEAD(&net->packet.sklist);
4564 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4565 sizeof(struct seq_net_private)))
4571 static void __net_exit packet_net_exit(struct net *net)
4573 remove_proc_entry("packet", net->proc_net);
4574 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4577 static struct pernet_operations packet_net_ops = {
4578 .init = packet_net_init,
4579 .exit = packet_net_exit,
4583 static void __exit packet_exit(void)
4585 unregister_netdevice_notifier(&packet_netdev_notifier);
4586 unregister_pernet_subsys(&packet_net_ops);
4587 sock_unregister(PF_PACKET);
4588 proto_unregister(&packet_proto);
4591 static int __init packet_init(void)
4593 int rc = proto_register(&packet_proto, 0);
4598 sock_register(&packet_family_ops);
4599 register_pernet_subsys(&packet_net_ops);
4600 register_netdevice_notifier(&packet_netdev_notifier);
4605 module_init(packet_init);
4606 module_exit(packet_exit);
4607 MODULE_LICENSE("GPL");
4608 MODULE_ALIAS_NETPROTO(PF_PACKET);