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 void packet_pick_tx_queue(struct net_device *dev, 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 struct net_device *dev = skb->dev;
247 struct sk_buff *orig_skb = skb;
248 struct netdev_queue *txq;
249 int ret = NETDEV_TX_BUSY;
252 if (unlikely(!netif_running(dev) ||
253 !netif_carrier_ok(dev)))
256 skb = validate_xmit_skb_list(skb, dev, &again);
260 packet_pick_tx_queue(dev, skb);
261 txq = skb_get_tx_queue(dev, skb);
265 HARD_TX_LOCK(dev, txq, smp_processor_id());
266 if (!netif_xmit_frozen_or_drv_stopped(txq))
267 ret = netdev_start_xmit(skb, dev, txq, false);
268 HARD_TX_UNLOCK(dev, txq);
272 if (!dev_xmit_complete(ret))
277 atomic_long_inc(&dev->tx_dropped);
279 return NET_XMIT_DROP;
282 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
284 struct net_device *dev;
287 dev = rcu_dereference(po->cached_dev);
295 static void packet_cached_dev_assign(struct packet_sock *po,
296 struct net_device *dev)
298 rcu_assign_pointer(po->cached_dev, dev);
301 static void packet_cached_dev_reset(struct packet_sock *po)
303 RCU_INIT_POINTER(po->cached_dev, NULL);
306 static bool packet_use_direct_xmit(const struct packet_sock *po)
308 return po->xmit == packet_direct_xmit;
311 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
313 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
316 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
318 const struct net_device_ops *ops = dev->netdev_ops;
321 if (ops->ndo_select_queue) {
322 queue_index = ops->ndo_select_queue(dev, skb, NULL,
323 __packet_pick_tx_queue);
324 queue_index = netdev_cap_txqueue(dev, queue_index);
326 queue_index = __packet_pick_tx_queue(dev, skb);
329 skb_set_queue_mapping(skb, queue_index);
332 /* __register_prot_hook must be invoked through register_prot_hook
333 * or from a context in which asynchronous accesses to the packet
334 * socket is not possible (packet_create()).
336 static void __register_prot_hook(struct sock *sk)
338 struct packet_sock *po = pkt_sk(sk);
342 __fanout_link(sk, po);
344 dev_add_pack(&po->prot_hook);
351 static void register_prot_hook(struct sock *sk)
353 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
354 __register_prot_hook(sk);
357 /* If the sync parameter is true, we will temporarily drop
358 * the po->bind_lock and do a synchronize_net to make sure no
359 * asynchronous packet processing paths still refer to the elements
360 * of po->prot_hook. If the sync parameter is false, it is the
361 * callers responsibility to take care of this.
363 static void __unregister_prot_hook(struct sock *sk, bool sync)
365 struct packet_sock *po = pkt_sk(sk);
367 lockdep_assert_held_once(&po->bind_lock);
372 __fanout_unlink(sk, po);
374 __dev_remove_pack(&po->prot_hook);
379 spin_unlock(&po->bind_lock);
381 spin_lock(&po->bind_lock);
385 static void unregister_prot_hook(struct sock *sk, bool sync)
387 struct packet_sock *po = pkt_sk(sk);
390 __unregister_prot_hook(sk, sync);
393 static inline struct page * __pure pgv_to_page(void *addr)
395 if (is_vmalloc_addr(addr))
396 return vmalloc_to_page(addr);
397 return virt_to_page(addr);
400 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
402 union tpacket_uhdr h;
405 switch (po->tp_version) {
407 h.h1->tp_status = status;
408 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
411 h.h2->tp_status = status;
412 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
415 h.h3->tp_status = status;
416 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
419 WARN(1, "TPACKET version not supported.\n");
426 static int __packet_get_status(struct packet_sock *po, void *frame)
428 union tpacket_uhdr h;
433 switch (po->tp_version) {
435 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
436 return h.h1->tp_status;
438 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
439 return h.h2->tp_status;
441 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
442 return h.h3->tp_status;
444 WARN(1, "TPACKET version not supported.\n");
450 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
453 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
456 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
457 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
458 return TP_STATUS_TS_RAW_HARDWARE;
460 if (ktime_to_timespec_cond(skb->tstamp, ts))
461 return TP_STATUS_TS_SOFTWARE;
466 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
469 union tpacket_uhdr h;
473 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
477 switch (po->tp_version) {
479 h.h1->tp_sec = ts.tv_sec;
480 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
483 h.h2->tp_sec = ts.tv_sec;
484 h.h2->tp_nsec = ts.tv_nsec;
487 h.h3->tp_sec = ts.tv_sec;
488 h.h3->tp_nsec = ts.tv_nsec;
491 WARN(1, "TPACKET version not supported.\n");
495 /* one flush is safe, as both fields always lie on the same cacheline */
496 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
502 static void *packet_lookup_frame(struct packet_sock *po,
503 struct packet_ring_buffer *rb,
504 unsigned int position,
507 unsigned int pg_vec_pos, frame_offset;
508 union tpacket_uhdr h;
510 pg_vec_pos = position / rb->frames_per_block;
511 frame_offset = position % rb->frames_per_block;
513 h.raw = rb->pg_vec[pg_vec_pos].buffer +
514 (frame_offset * rb->frame_size);
516 if (status != __packet_get_status(po, h.raw))
522 static void *packet_current_frame(struct packet_sock *po,
523 struct packet_ring_buffer *rb,
526 return packet_lookup_frame(po, rb, rb->head, status);
529 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
531 del_timer_sync(&pkc->retire_blk_timer);
534 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
535 struct sk_buff_head *rb_queue)
537 struct tpacket_kbdq_core *pkc;
539 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
541 spin_lock_bh(&rb_queue->lock);
542 pkc->delete_blk_timer = 1;
543 spin_unlock_bh(&rb_queue->lock);
545 prb_del_retire_blk_timer(pkc);
548 static void prb_setup_retire_blk_timer(struct packet_sock *po)
550 struct tpacket_kbdq_core *pkc;
552 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
553 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
555 pkc->retire_blk_timer.expires = jiffies;
558 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
559 int blk_size_in_bytes)
561 struct net_device *dev;
562 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
563 struct ethtool_link_ksettings ecmd;
567 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
568 if (unlikely(!dev)) {
570 return DEFAULT_PRB_RETIRE_TOV;
572 err = __ethtool_get_link_ksettings(dev, &ecmd);
576 * If the link speed is so slow you don't really
577 * need to worry about perf anyways
579 if (ecmd.base.speed < SPEED_1000 ||
580 ecmd.base.speed == SPEED_UNKNOWN) {
581 return DEFAULT_PRB_RETIRE_TOV;
584 div = ecmd.base.speed / 1000;
588 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
600 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
601 union tpacket_req_u *req_u)
603 p1->feature_req_word = req_u->req3.tp_feature_req_word;
606 static void init_prb_bdqc(struct packet_sock *po,
607 struct packet_ring_buffer *rb,
609 union tpacket_req_u *req_u)
611 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
612 struct tpacket_block_desc *pbd;
614 memset(p1, 0x0, sizeof(*p1));
616 p1->knxt_seq_num = 1;
618 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
619 p1->pkblk_start = pg_vec[0].buffer;
620 p1->kblk_size = req_u->req3.tp_block_size;
621 p1->knum_blocks = req_u->req3.tp_block_nr;
622 p1->hdrlen = po->tp_hdrlen;
623 p1->version = po->tp_version;
624 p1->last_kactive_blk_num = 0;
625 po->stats.stats3.tp_freeze_q_cnt = 0;
626 if (req_u->req3.tp_retire_blk_tov)
627 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
629 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
630 req_u->req3.tp_block_size);
631 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
632 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
634 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
635 prb_init_ft_ops(p1, req_u);
636 prb_setup_retire_blk_timer(po);
637 prb_open_block(p1, pbd);
640 /* Do NOT update the last_blk_num first.
641 * Assumes sk_buff_head lock is held.
643 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
645 mod_timer(&pkc->retire_blk_timer,
646 jiffies + pkc->tov_in_jiffies);
647 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
652 * 1) We refresh the timer only when we open a block.
653 * By doing this we don't waste cycles refreshing the timer
654 * on packet-by-packet basis.
656 * With a 1MB block-size, on a 1Gbps line, it will take
657 * i) ~8 ms to fill a block + ii) memcpy etc.
658 * In this cut we are not accounting for the memcpy time.
660 * So, if the user sets the 'tmo' to 10ms then the timer
661 * will never fire while the block is still getting filled
662 * (which is what we want). However, the user could choose
663 * to close a block early and that's fine.
665 * But when the timer does fire, we check whether or not to refresh it.
666 * Since the tmo granularity is in msecs, it is not too expensive
667 * to refresh the timer, lets say every '8' msecs.
668 * Either the user can set the 'tmo' or we can derive it based on
669 * a) line-speed and b) block-size.
670 * prb_calc_retire_blk_tmo() calculates the tmo.
673 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
675 struct packet_sock *po =
676 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
677 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
679 struct tpacket_block_desc *pbd;
681 spin_lock(&po->sk.sk_receive_queue.lock);
683 frozen = prb_queue_frozen(pkc);
684 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
686 if (unlikely(pkc->delete_blk_timer))
689 /* We only need to plug the race when the block is partially filled.
691 * lock(); increment BLOCK_NUM_PKTS; unlock()
692 * copy_bits() is in progress ...
693 * timer fires on other cpu:
694 * we can't retire the current block because copy_bits
698 if (BLOCK_NUM_PKTS(pbd)) {
699 while (atomic_read(&pkc->blk_fill_in_prog)) {
700 /* Waiting for skb_copy_bits to finish... */
705 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
707 if (!BLOCK_NUM_PKTS(pbd)) {
708 /* An empty block. Just refresh the timer. */
711 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
712 if (!prb_dispatch_next_block(pkc, po))
717 /* Case 1. Queue was frozen because user-space was
720 if (prb_curr_blk_in_use(pbd)) {
722 * Ok, user-space is still behind.
723 * So just refresh the timer.
727 /* Case 2. queue was frozen,user-space caught up,
728 * now the link went idle && the timer fired.
729 * We don't have a block to close.So we open this
730 * block and restart the timer.
731 * opening a block thaws the queue,restarts timer
732 * Thawing/timer-refresh is a side effect.
734 prb_open_block(pkc, pbd);
741 _prb_refresh_rx_retire_blk_timer(pkc);
744 spin_unlock(&po->sk.sk_receive_queue.lock);
747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
748 struct tpacket_block_desc *pbd1, __u32 status)
750 /* Flush everything minus the block header */
752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
757 /* Skip the block header(we know header WILL fit in 4K) */
760 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
761 for (; start < end; start += PAGE_SIZE)
762 flush_dcache_page(pgv_to_page(start));
767 /* Now update the block status. */
769 BLOCK_STATUS(pbd1) = status;
771 /* Flush the block header */
773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
775 flush_dcache_page(pgv_to_page(start));
785 * 2) Increment active_blk_num
787 * Note:We DONT refresh the timer on purpose.
788 * Because almost always the next block will be opened.
790 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
791 struct tpacket_block_desc *pbd1,
792 struct packet_sock *po, unsigned int stat)
794 __u32 status = TP_STATUS_USER | stat;
796 struct tpacket3_hdr *last_pkt;
797 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
798 struct sock *sk = &po->sk;
800 if (po->stats.stats3.tp_drops)
801 status |= TP_STATUS_LOSING;
803 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
804 last_pkt->tp_next_offset = 0;
806 /* Get the ts of the last pkt */
807 if (BLOCK_NUM_PKTS(pbd1)) {
808 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
809 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
811 /* Ok, we tmo'd - so get the current time.
813 * It shouldn't really happen as we don't close empty
814 * blocks. See prb_retire_rx_blk_timer_expired().
818 h1->ts_last_pkt.ts_sec = ts.tv_sec;
819 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
824 /* Flush the block */
825 prb_flush_block(pkc1, pbd1, status);
827 sk->sk_data_ready(sk);
829 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
834 pkc->reset_pending_on_curr_blk = 0;
838 * Side effect of opening a block:
840 * 1) prb_queue is thawed.
841 * 2) retire_blk_timer is refreshed.
844 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
845 struct tpacket_block_desc *pbd1)
848 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
852 /* We could have just memset this but we will lose the
853 * flexibility of making the priv area sticky
856 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
857 BLOCK_NUM_PKTS(pbd1) = 0;
858 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
862 h1->ts_first_pkt.ts_sec = ts.tv_sec;
863 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
865 pkc1->pkblk_start = (char *)pbd1;
866 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
868 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
869 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
871 pbd1->version = pkc1->version;
872 pkc1->prev = pkc1->nxt_offset;
873 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
875 prb_thaw_queue(pkc1);
876 _prb_refresh_rx_retire_blk_timer(pkc1);
882 * Queue freeze logic:
883 * 1) Assume tp_block_nr = 8 blocks.
884 * 2) At time 't0', user opens Rx ring.
885 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
886 * 4) user-space is either sleeping or processing block '0'.
887 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
888 * it will close block-7,loop around and try to fill block '0'.
890 * __packet_lookup_frame_in_block
891 * prb_retire_current_block()
892 * prb_dispatch_next_block()
893 * |->(BLOCK_STATUS == USER) evaluates to true
894 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
895 * 6) Now there are two cases:
896 * 6.1) Link goes idle right after the queue is frozen.
897 * But remember, the last open_block() refreshed the timer.
898 * When this timer expires,it will refresh itself so that we can
899 * re-open block-0 in near future.
900 * 6.2) Link is busy and keeps on receiving packets. This is a simple
901 * case and __packet_lookup_frame_in_block will check if block-0
902 * is free and can now be re-used.
904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
905 struct packet_sock *po)
907 pkc->reset_pending_on_curr_blk = 1;
908 po->stats.stats3.tp_freeze_q_cnt++;
911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
914 * If the next block is free then we will dispatch it
915 * and return a good offset.
916 * Else, we will freeze the queue.
917 * So, caller must check the return value.
919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
920 struct packet_sock *po)
922 struct tpacket_block_desc *pbd;
926 /* 1. Get current block num */
927 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
929 /* 2. If this block is currently in_use then freeze the queue */
930 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
931 prb_freeze_queue(pkc, po);
937 * open this block and return the offset where the first packet
938 * needs to get stored.
940 prb_open_block(pkc, pbd);
941 return (void *)pkc->nxt_offset;
944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
945 struct packet_sock *po, unsigned int status)
947 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
949 /* retire/close the current block */
950 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
952 * Plug the case where copy_bits() is in progress on
953 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
954 * have space to copy the pkt in the current block and
955 * called prb_retire_current_block()
957 * We don't need to worry about the TMO case because
958 * the timer-handler already handled this case.
960 if (!(status & TP_STATUS_BLK_TMO)) {
961 while (atomic_read(&pkc->blk_fill_in_prog)) {
962 /* Waiting for skb_copy_bits to finish... */
966 prb_close_block(pkc, pbd, po, status);
971 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
973 return TP_STATUS_USER & BLOCK_STATUS(pbd);
976 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
978 return pkc->reset_pending_on_curr_blk;
981 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
983 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
984 atomic_dec(&pkc->blk_fill_in_prog);
987 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
988 struct tpacket3_hdr *ppd)
990 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
993 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
994 struct tpacket3_hdr *ppd)
996 ppd->hv1.tp_rxhash = 0;
999 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1000 struct tpacket3_hdr *ppd)
1002 if (skb_vlan_tag_present(pkc->skb)) {
1003 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1004 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1005 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1007 ppd->hv1.tp_vlan_tci = 0;
1008 ppd->hv1.tp_vlan_tpid = 0;
1009 ppd->tp_status = TP_STATUS_AVAILABLE;
1013 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1014 struct tpacket3_hdr *ppd)
1016 ppd->hv1.tp_padding = 0;
1017 prb_fill_vlan_info(pkc, ppd);
1019 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1020 prb_fill_rxhash(pkc, ppd);
1022 prb_clear_rxhash(pkc, ppd);
1025 static void prb_fill_curr_block(char *curr,
1026 struct tpacket_kbdq_core *pkc,
1027 struct tpacket_block_desc *pbd,
1030 struct tpacket3_hdr *ppd;
1032 ppd = (struct tpacket3_hdr *)curr;
1033 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1035 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1036 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1037 BLOCK_NUM_PKTS(pbd) += 1;
1038 atomic_inc(&pkc->blk_fill_in_prog);
1039 prb_run_all_ft_ops(pkc, ppd);
1042 /* Assumes caller has the sk->rx_queue.lock */
1043 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1044 struct sk_buff *skb,
1049 struct tpacket_kbdq_core *pkc;
1050 struct tpacket_block_desc *pbd;
1053 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1054 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1056 /* Queue is frozen when user space is lagging behind */
1057 if (prb_queue_frozen(pkc)) {
1059 * Check if that last block which caused the queue to freeze,
1060 * is still in_use by user-space.
1062 if (prb_curr_blk_in_use(pbd)) {
1063 /* Can't record this packet */
1067 * Ok, the block was released by user-space.
1068 * Now let's open that block.
1069 * opening a block also thaws the queue.
1070 * Thawing is a side effect.
1072 prb_open_block(pkc, pbd);
1077 curr = pkc->nxt_offset;
1079 end = (char *)pbd + pkc->kblk_size;
1081 /* first try the current block */
1082 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1083 prb_fill_curr_block(curr, pkc, pbd, len);
1084 return (void *)curr;
1087 /* Ok, close the current block */
1088 prb_retire_current_block(pkc, po, 0);
1090 /* Now, try to dispatch the next block */
1091 curr = (char *)prb_dispatch_next_block(pkc, po);
1093 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1094 prb_fill_curr_block(curr, pkc, pbd, len);
1095 return (void *)curr;
1099 * No free blocks are available.user_space hasn't caught up yet.
1100 * Queue was just frozen and now this packet will get dropped.
1105 static void *packet_current_rx_frame(struct packet_sock *po,
1106 struct sk_buff *skb,
1107 int status, unsigned int len)
1110 switch (po->tp_version) {
1113 curr = packet_lookup_frame(po, &po->rx_ring,
1114 po->rx_ring.head, status);
1117 return __packet_lookup_frame_in_block(po, skb, status, len);
1119 WARN(1, "TPACKET version not supported\n");
1125 static void *prb_lookup_block(struct packet_sock *po,
1126 struct packet_ring_buffer *rb,
1130 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1131 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1133 if (status != BLOCK_STATUS(pbd))
1138 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1141 if (rb->prb_bdqc.kactive_blk_num)
1142 prev = rb->prb_bdqc.kactive_blk_num-1;
1144 prev = rb->prb_bdqc.knum_blocks-1;
1148 /* Assumes caller has held the rx_queue.lock */
1149 static void *__prb_previous_block(struct packet_sock *po,
1150 struct packet_ring_buffer *rb,
1153 unsigned int previous = prb_previous_blk_num(rb);
1154 return prb_lookup_block(po, rb, previous, status);
1157 static void *packet_previous_rx_frame(struct packet_sock *po,
1158 struct packet_ring_buffer *rb,
1161 if (po->tp_version <= TPACKET_V2)
1162 return packet_previous_frame(po, rb, status);
1164 return __prb_previous_block(po, rb, status);
1167 static void packet_increment_rx_head(struct packet_sock *po,
1168 struct packet_ring_buffer *rb)
1170 switch (po->tp_version) {
1173 return packet_increment_head(rb);
1176 WARN(1, "TPACKET version not supported.\n");
1182 static void *packet_previous_frame(struct packet_sock *po,
1183 struct packet_ring_buffer *rb,
1186 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1187 return packet_lookup_frame(po, rb, previous, status);
1190 static void packet_increment_head(struct packet_ring_buffer *buff)
1192 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1195 static void packet_inc_pending(struct packet_ring_buffer *rb)
1197 this_cpu_inc(*rb->pending_refcnt);
1200 static void packet_dec_pending(struct packet_ring_buffer *rb)
1202 this_cpu_dec(*rb->pending_refcnt);
1205 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1207 unsigned int refcnt = 0;
1210 /* We don't use pending refcount in rx_ring. */
1211 if (rb->pending_refcnt == NULL)
1214 for_each_possible_cpu(cpu)
1215 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1220 static int packet_alloc_pending(struct packet_sock *po)
1222 po->rx_ring.pending_refcnt = NULL;
1224 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1225 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1231 static void packet_free_pending(struct packet_sock *po)
1233 free_percpu(po->tx_ring.pending_refcnt);
1236 #define ROOM_POW_OFF 2
1237 #define ROOM_NONE 0x0
1238 #define ROOM_LOW 0x1
1239 #define ROOM_NORMAL 0x2
1241 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1245 len = po->rx_ring.frame_max + 1;
1246 idx = po->rx_ring.head;
1248 idx += len >> pow_off;
1251 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1254 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1258 len = po->rx_ring.prb_bdqc.knum_blocks;
1259 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1261 idx += len >> pow_off;
1264 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1267 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1269 struct sock *sk = &po->sk;
1270 int ret = ROOM_NONE;
1272 if (po->prot_hook.func != tpacket_rcv) {
1273 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1274 - (skb ? skb->truesize : 0);
1275 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1283 if (po->tp_version == TPACKET_V3) {
1284 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1286 else if (__tpacket_v3_has_room(po, 0))
1289 if (__tpacket_has_room(po, ROOM_POW_OFF))
1291 else if (__tpacket_has_room(po, 0))
1298 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1303 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1304 ret = __packet_rcv_has_room(po, skb);
1305 has_room = ret == ROOM_NORMAL;
1306 if (po->pressure == has_room)
1307 po->pressure = !has_room;
1308 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1313 static void packet_sock_destruct(struct sock *sk)
1315 skb_queue_purge(&sk->sk_error_queue);
1317 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1318 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1320 if (!sock_flag(sk, SOCK_DEAD)) {
1321 pr_err("Attempt to release alive packet socket: %p\n", sk);
1325 sk_refcnt_debug_dec(sk);
1328 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1333 rxhash = skb_get_hash(skb);
1334 for (i = 0; i < ROLLOVER_HLEN; i++)
1335 if (po->rollover->history[i] == rxhash)
1338 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1339 return count > (ROLLOVER_HLEN >> 1);
1342 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1343 struct sk_buff *skb,
1346 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1349 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1350 struct sk_buff *skb,
1353 unsigned int val = atomic_inc_return(&f->rr_cur);
1358 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1359 struct sk_buff *skb,
1362 return smp_processor_id() % num;
1365 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1366 struct sk_buff *skb,
1369 return prandom_u32_max(num);
1372 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1373 struct sk_buff *skb,
1374 unsigned int idx, bool try_self,
1377 struct packet_sock *po, *po_next, *po_skip = NULL;
1378 unsigned int i, j, room = ROOM_NONE;
1380 po = pkt_sk(f->arr[idx]);
1383 room = packet_rcv_has_room(po, skb);
1384 if (room == ROOM_NORMAL ||
1385 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1390 i = j = min_t(int, po->rollover->sock, num - 1);
1392 po_next = pkt_sk(f->arr[i]);
1393 if (po_next != po_skip && !po_next->pressure &&
1394 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1396 po->rollover->sock = i;
1397 atomic_long_inc(&po->rollover->num);
1398 if (room == ROOM_LOW)
1399 atomic_long_inc(&po->rollover->num_huge);
1407 atomic_long_inc(&po->rollover->num_failed);
1411 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1412 struct sk_buff *skb,
1415 return skb_get_queue_mapping(skb) % num;
1418 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1419 struct sk_buff *skb,
1422 struct bpf_prog *prog;
1423 unsigned int ret = 0;
1426 prog = rcu_dereference(f->bpf_prog);
1428 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1434 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1436 return f->flags & (flag >> 8);
1439 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1440 struct packet_type *pt, struct net_device *orig_dev)
1442 struct packet_fanout *f = pt->af_packet_priv;
1443 unsigned int num = READ_ONCE(f->num_members);
1444 struct net *net = read_pnet(&f->net);
1445 struct packet_sock *po;
1448 if (!net_eq(dev_net(dev), net) || !num) {
1453 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1454 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1459 case PACKET_FANOUT_HASH:
1461 idx = fanout_demux_hash(f, skb, num);
1463 case PACKET_FANOUT_LB:
1464 idx = fanout_demux_lb(f, skb, num);
1466 case PACKET_FANOUT_CPU:
1467 idx = fanout_demux_cpu(f, skb, num);
1469 case PACKET_FANOUT_RND:
1470 idx = fanout_demux_rnd(f, skb, num);
1472 case PACKET_FANOUT_QM:
1473 idx = fanout_demux_qm(f, skb, num);
1475 case PACKET_FANOUT_ROLLOVER:
1476 idx = fanout_demux_rollover(f, skb, 0, false, num);
1478 case PACKET_FANOUT_CBPF:
1479 case PACKET_FANOUT_EBPF:
1480 idx = fanout_demux_bpf(f, skb, num);
1484 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1485 idx = fanout_demux_rollover(f, skb, idx, true, num);
1487 po = pkt_sk(f->arr[idx]);
1488 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1491 DEFINE_MUTEX(fanout_mutex);
1492 EXPORT_SYMBOL_GPL(fanout_mutex);
1493 static LIST_HEAD(fanout_list);
1494 static u16 fanout_next_id;
1496 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1498 struct packet_fanout *f = po->fanout;
1500 spin_lock(&f->lock);
1501 f->arr[f->num_members] = sk;
1504 if (f->num_members == 1)
1505 dev_add_pack(&f->prot_hook);
1506 spin_unlock(&f->lock);
1509 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1511 struct packet_fanout *f = po->fanout;
1514 spin_lock(&f->lock);
1515 for (i = 0; i < f->num_members; i++) {
1516 if (f->arr[i] == sk)
1519 BUG_ON(i >= f->num_members);
1520 f->arr[i] = f->arr[f->num_members - 1];
1522 if (f->num_members == 0)
1523 __dev_remove_pack(&f->prot_hook);
1524 spin_unlock(&f->lock);
1527 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1529 if (sk->sk_family != PF_PACKET)
1532 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1535 static void fanout_init_data(struct packet_fanout *f)
1538 case PACKET_FANOUT_LB:
1539 atomic_set(&f->rr_cur, 0);
1541 case PACKET_FANOUT_CBPF:
1542 case PACKET_FANOUT_EBPF:
1543 RCU_INIT_POINTER(f->bpf_prog, NULL);
1548 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1550 struct bpf_prog *old;
1552 spin_lock(&f->lock);
1553 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1554 rcu_assign_pointer(f->bpf_prog, new);
1555 spin_unlock(&f->lock);
1559 bpf_prog_destroy(old);
1563 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1566 struct bpf_prog *new;
1567 struct sock_fprog fprog;
1570 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1572 if (len != sizeof(fprog))
1574 if (copy_from_user(&fprog, data, len))
1577 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1581 __fanout_set_data_bpf(po->fanout, new);
1585 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1588 struct bpf_prog *new;
1591 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1593 if (len != sizeof(fd))
1595 if (copy_from_user(&fd, data, len))
1598 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1600 return PTR_ERR(new);
1602 __fanout_set_data_bpf(po->fanout, new);
1606 static int fanout_set_data(struct packet_sock *po, char __user *data,
1609 switch (po->fanout->type) {
1610 case PACKET_FANOUT_CBPF:
1611 return fanout_set_data_cbpf(po, data, len);
1612 case PACKET_FANOUT_EBPF:
1613 return fanout_set_data_ebpf(po, data, len);
1619 static void fanout_release_data(struct packet_fanout *f)
1622 case PACKET_FANOUT_CBPF:
1623 case PACKET_FANOUT_EBPF:
1624 __fanout_set_data_bpf(f, NULL);
1628 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1630 struct packet_fanout *f;
1632 list_for_each_entry(f, &fanout_list, list) {
1633 if (f->id == candidate_id &&
1634 read_pnet(&f->net) == sock_net(sk)) {
1641 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1643 u16 id = fanout_next_id;
1646 if (__fanout_id_is_free(sk, id)) {
1648 fanout_next_id = id + 1;
1653 } while (id != fanout_next_id);
1658 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1660 struct packet_rollover *rollover = NULL;
1661 struct packet_sock *po = pkt_sk(sk);
1662 struct packet_fanout *f, *match;
1663 u8 type = type_flags & 0xff;
1664 u8 flags = type_flags >> 8;
1668 case PACKET_FANOUT_ROLLOVER:
1669 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1671 case PACKET_FANOUT_HASH:
1672 case PACKET_FANOUT_LB:
1673 case PACKET_FANOUT_CPU:
1674 case PACKET_FANOUT_RND:
1675 case PACKET_FANOUT_QM:
1676 case PACKET_FANOUT_CBPF:
1677 case PACKET_FANOUT_EBPF:
1683 mutex_lock(&fanout_mutex);
1689 if (type == PACKET_FANOUT_ROLLOVER ||
1690 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1692 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1695 atomic_long_set(&rollover->num, 0);
1696 atomic_long_set(&rollover->num_huge, 0);
1697 atomic_long_set(&rollover->num_failed, 0);
1700 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1705 if (!fanout_find_new_id(sk, &id)) {
1709 /* ephemeral flag for the first socket in the group: drop it */
1710 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1714 list_for_each_entry(f, &fanout_list, list) {
1716 read_pnet(&f->net) == sock_net(sk)) {
1722 if (match && match->flags != flags)
1726 match = kzalloc(sizeof(*match), GFP_KERNEL);
1729 write_pnet(&match->net, sock_net(sk));
1732 match->flags = flags;
1733 INIT_LIST_HEAD(&match->list);
1734 spin_lock_init(&match->lock);
1735 refcount_set(&match->sk_ref, 0);
1736 fanout_init_data(match);
1737 match->prot_hook.type = po->prot_hook.type;
1738 match->prot_hook.dev = po->prot_hook.dev;
1739 match->prot_hook.func = packet_rcv_fanout;
1740 match->prot_hook.af_packet_priv = match;
1741 match->prot_hook.id_match = match_fanout_group;
1742 list_add(&match->list, &fanout_list);
1746 spin_lock(&po->bind_lock);
1748 match->type == type &&
1749 match->prot_hook.type == po->prot_hook.type &&
1750 match->prot_hook.dev == po->prot_hook.dev) {
1752 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1753 __dev_remove_pack(&po->prot_hook);
1755 po->rollover = rollover;
1757 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1758 __fanout_link(sk, po);
1762 spin_unlock(&po->bind_lock);
1764 if (err && !refcount_read(&match->sk_ref)) {
1765 list_del(&match->list);
1771 mutex_unlock(&fanout_mutex);
1775 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1776 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1777 * It is the responsibility of the caller to call fanout_release_data() and
1778 * free the returned packet_fanout (after synchronize_net())
1780 static struct packet_fanout *fanout_release(struct sock *sk)
1782 struct packet_sock *po = pkt_sk(sk);
1783 struct packet_fanout *f;
1785 mutex_lock(&fanout_mutex);
1790 if (refcount_dec_and_test(&f->sk_ref))
1795 mutex_unlock(&fanout_mutex);
1800 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1801 struct sk_buff *skb)
1803 /* Earlier code assumed this would be a VLAN pkt, double-check
1804 * this now that we have the actual packet in hand. We can only
1805 * do this check on Ethernet devices.
1807 if (unlikely(dev->type != ARPHRD_ETHER))
1810 skb_reset_mac_header(skb);
1811 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1814 static const struct proto_ops packet_ops;
1816 static const struct proto_ops packet_ops_spkt;
1818 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1819 struct packet_type *pt, struct net_device *orig_dev)
1822 struct sockaddr_pkt *spkt;
1825 * When we registered the protocol we saved the socket in the data
1826 * field for just this event.
1829 sk = pt->af_packet_priv;
1832 * Yank back the headers [hope the device set this
1833 * right or kerboom...]
1835 * Incoming packets have ll header pulled,
1838 * For outgoing ones skb->data == skb_mac_header(skb)
1839 * so that this procedure is noop.
1842 if (skb->pkt_type == PACKET_LOOPBACK)
1845 if (!net_eq(dev_net(dev), sock_net(sk)))
1848 skb = skb_share_check(skb, GFP_ATOMIC);
1852 /* drop any routing info */
1855 /* drop conntrack reference */
1858 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1860 skb_push(skb, skb->data - skb_mac_header(skb));
1863 * The SOCK_PACKET socket receives _all_ frames.
1866 spkt->spkt_family = dev->type;
1867 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1868 spkt->spkt_protocol = skb->protocol;
1871 * Charge the memory to the socket. This is done specifically
1872 * to prevent sockets using all the memory up.
1875 if (sock_queue_rcv_skb(sk, skb) == 0)
1886 * Output a raw packet to a device layer. This bypasses all the other
1887 * protocol layers and you must therefore supply it with a complete frame
1890 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1893 struct sock *sk = sock->sk;
1894 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1895 struct sk_buff *skb = NULL;
1896 struct net_device *dev;
1897 struct sockcm_cookie sockc;
1903 * Get and verify the address.
1907 if (msg->msg_namelen < sizeof(struct sockaddr))
1909 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1910 proto = saddr->spkt_protocol;
1912 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1915 * Find the device first to size check it
1918 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1921 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1927 if (!(dev->flags & IFF_UP))
1931 * You may not queue a frame bigger than the mtu. This is the lowest level
1932 * raw protocol and you must do your own fragmentation at this level.
1935 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1936 if (!netif_supports_nofcs(dev)) {
1937 err = -EPROTONOSUPPORT;
1940 extra_len = 4; /* We're doing our own CRC */
1944 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1948 size_t reserved = LL_RESERVED_SPACE(dev);
1949 int tlen = dev->needed_tailroom;
1950 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1953 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1956 /* FIXME: Save some space for broken drivers that write a hard
1957 * header at transmission time by themselves. PPP is the notable
1958 * one here. This should really be fixed at the driver level.
1960 skb_reserve(skb, reserved);
1961 skb_reset_network_header(skb);
1963 /* Try to align data part correctly */
1968 skb_reset_network_header(skb);
1970 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1976 if (!dev_validate_header(dev, skb->data, len)) {
1980 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1981 !packet_extra_vlan_len_allowed(dev, skb)) {
1986 sockc.tsflags = sk->sk_tsflags;
1987 if (msg->msg_controllen) {
1988 err = sock_cmsg_send(sk, msg, &sockc);
1993 skb->protocol = proto;
1995 skb->priority = sk->sk_priority;
1996 skb->mark = sk->sk_mark;
1998 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2000 if (unlikely(extra_len == 4))
2003 skb_probe_transport_header(skb, 0);
2005 dev_queue_xmit(skb);
2016 static unsigned int run_filter(struct sk_buff *skb,
2017 const struct sock *sk,
2020 struct sk_filter *filter;
2023 filter = rcu_dereference(sk->sk_filter);
2025 res = bpf_prog_run_clear_cb(filter->prog, skb);
2031 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2034 struct virtio_net_hdr vnet_hdr;
2036 if (*len < sizeof(vnet_hdr))
2038 *len -= sizeof(vnet_hdr);
2040 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
2043 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2047 * This function makes lazy skb cloning in hope that most of packets
2048 * are discarded by BPF.
2050 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2051 * and skb->cb are mangled. It works because (and until) packets
2052 * falling here are owned by current CPU. Output packets are cloned
2053 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2054 * sequencially, so that if we return skb to original state on exit,
2055 * we will not harm anyone.
2058 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2059 struct packet_type *pt, struct net_device *orig_dev)
2062 struct sockaddr_ll *sll;
2063 struct packet_sock *po;
2064 u8 *skb_head = skb->data;
2065 int skb_len = skb->len;
2066 unsigned int snaplen, res;
2067 bool is_drop_n_account = false;
2069 if (skb->pkt_type == PACKET_LOOPBACK)
2072 sk = pt->af_packet_priv;
2075 if (!net_eq(dev_net(dev), sock_net(sk)))
2080 if (dev->header_ops) {
2081 /* The device has an explicit notion of ll header,
2082 * exported to higher levels.
2084 * Otherwise, the device hides details of its frame
2085 * structure, so that corresponding packet head is
2086 * never delivered to user.
2088 if (sk->sk_type != SOCK_DGRAM)
2089 skb_push(skb, skb->data - skb_mac_header(skb));
2090 else if (skb->pkt_type == PACKET_OUTGOING) {
2091 /* Special case: outgoing packets have ll header at head */
2092 skb_pull(skb, skb_network_offset(skb));
2098 res = run_filter(skb, sk, snaplen);
2100 goto drop_n_restore;
2104 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2107 if (skb_shared(skb)) {
2108 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2112 if (skb_head != skb->data) {
2113 skb->data = skb_head;
2120 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2122 sll = &PACKET_SKB_CB(skb)->sa.ll;
2123 sll->sll_hatype = dev->type;
2124 sll->sll_pkttype = skb->pkt_type;
2125 if (unlikely(po->origdev))
2126 sll->sll_ifindex = orig_dev->ifindex;
2128 sll->sll_ifindex = dev->ifindex;
2130 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2132 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2133 * Use their space for storing the original skb length.
2135 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2137 if (pskb_trim(skb, snaplen))
2140 skb_set_owner_r(skb, sk);
2144 /* drop conntrack reference */
2147 spin_lock(&sk->sk_receive_queue.lock);
2148 po->stats.stats1.tp_packets++;
2149 sock_skb_set_dropcount(sk, skb);
2150 __skb_queue_tail(&sk->sk_receive_queue, skb);
2151 spin_unlock(&sk->sk_receive_queue.lock);
2152 sk->sk_data_ready(sk);
2156 is_drop_n_account = true;
2157 spin_lock(&sk->sk_receive_queue.lock);
2158 po->stats.stats1.tp_drops++;
2159 atomic_inc(&sk->sk_drops);
2160 spin_unlock(&sk->sk_receive_queue.lock);
2163 if (skb_head != skb->data && skb_shared(skb)) {
2164 skb->data = skb_head;
2168 if (!is_drop_n_account)
2175 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2176 struct packet_type *pt, struct net_device *orig_dev)
2179 struct packet_sock *po;
2180 struct sockaddr_ll *sll;
2181 union tpacket_uhdr h;
2182 u8 *skb_head = skb->data;
2183 int skb_len = skb->len;
2184 unsigned int snaplen, res;
2185 unsigned long status = TP_STATUS_USER;
2186 unsigned short macoff, netoff, hdrlen;
2187 struct sk_buff *copy_skb = NULL;
2190 bool is_drop_n_account = false;
2191 bool do_vnet = false;
2193 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2194 * We may add members to them until current aligned size without forcing
2195 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2197 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2198 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2200 if (skb->pkt_type == PACKET_LOOPBACK)
2203 sk = pt->af_packet_priv;
2206 if (!net_eq(dev_net(dev), sock_net(sk)))
2209 if (dev->header_ops) {
2210 if (sk->sk_type != SOCK_DGRAM)
2211 skb_push(skb, skb->data - skb_mac_header(skb));
2212 else if (skb->pkt_type == PACKET_OUTGOING) {
2213 /* Special case: outgoing packets have ll header at head */
2214 skb_pull(skb, skb_network_offset(skb));
2220 res = run_filter(skb, sk, snaplen);
2222 goto drop_n_restore;
2224 if (skb->ip_summed == CHECKSUM_PARTIAL)
2225 status |= TP_STATUS_CSUMNOTREADY;
2226 else if (skb->pkt_type != PACKET_OUTGOING &&
2227 (skb->ip_summed == CHECKSUM_COMPLETE ||
2228 skb_csum_unnecessary(skb)))
2229 status |= TP_STATUS_CSUM_VALID;
2234 if (sk->sk_type == SOCK_DGRAM) {
2235 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2238 unsigned int maclen = skb_network_offset(skb);
2239 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2240 (maclen < 16 ? 16 : maclen)) +
2242 if (po->has_vnet_hdr) {
2243 netoff += sizeof(struct virtio_net_hdr);
2246 macoff = netoff - maclen;
2248 if (po->tp_version <= TPACKET_V2) {
2249 if (macoff + snaplen > po->rx_ring.frame_size) {
2250 if (po->copy_thresh &&
2251 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2252 if (skb_shared(skb)) {
2253 copy_skb = skb_clone(skb, GFP_ATOMIC);
2255 copy_skb = skb_get(skb);
2256 skb_head = skb->data;
2259 skb_set_owner_r(copy_skb, sk);
2261 snaplen = po->rx_ring.frame_size - macoff;
2262 if ((int)snaplen < 0) {
2267 } else if (unlikely(macoff + snaplen >
2268 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2271 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2272 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2273 snaplen, nval, macoff);
2275 if (unlikely((int)snaplen < 0)) {
2277 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2281 spin_lock(&sk->sk_receive_queue.lock);
2282 h.raw = packet_current_rx_frame(po, skb,
2283 TP_STATUS_KERNEL, (macoff+snaplen));
2285 goto drop_n_account;
2286 if (po->tp_version <= TPACKET_V2) {
2287 packet_increment_rx_head(po, &po->rx_ring);
2289 * LOSING will be reported till you read the stats,
2290 * because it's COR - Clear On Read.
2291 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2294 if (po->stats.stats1.tp_drops)
2295 status |= TP_STATUS_LOSING;
2297 po->stats.stats1.tp_packets++;
2299 status |= TP_STATUS_COPY;
2300 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2302 spin_unlock(&sk->sk_receive_queue.lock);
2305 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2306 sizeof(struct virtio_net_hdr),
2308 spin_lock(&sk->sk_receive_queue.lock);
2309 goto drop_n_account;
2313 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2315 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2316 getnstimeofday(&ts);
2318 status |= ts_status;
2320 switch (po->tp_version) {
2322 h.h1->tp_len = skb->len;
2323 h.h1->tp_snaplen = snaplen;
2324 h.h1->tp_mac = macoff;
2325 h.h1->tp_net = netoff;
2326 h.h1->tp_sec = ts.tv_sec;
2327 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2328 hdrlen = sizeof(*h.h1);
2331 h.h2->tp_len = skb->len;
2332 h.h2->tp_snaplen = snaplen;
2333 h.h2->tp_mac = macoff;
2334 h.h2->tp_net = netoff;
2335 h.h2->tp_sec = ts.tv_sec;
2336 h.h2->tp_nsec = ts.tv_nsec;
2337 if (skb_vlan_tag_present(skb)) {
2338 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2339 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2340 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2342 h.h2->tp_vlan_tci = 0;
2343 h.h2->tp_vlan_tpid = 0;
2345 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2346 hdrlen = sizeof(*h.h2);
2349 /* tp_nxt_offset,vlan are already populated above.
2350 * So DONT clear those fields here
2352 h.h3->tp_status |= status;
2353 h.h3->tp_len = skb->len;
2354 h.h3->tp_snaplen = snaplen;
2355 h.h3->tp_mac = macoff;
2356 h.h3->tp_net = netoff;
2357 h.h3->tp_sec = ts.tv_sec;
2358 h.h3->tp_nsec = ts.tv_nsec;
2359 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2360 hdrlen = sizeof(*h.h3);
2366 sll = h.raw + TPACKET_ALIGN(hdrlen);
2367 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2368 sll->sll_family = AF_PACKET;
2369 sll->sll_hatype = dev->type;
2370 sll->sll_protocol = skb->protocol;
2371 sll->sll_pkttype = skb->pkt_type;
2372 if (unlikely(po->origdev))
2373 sll->sll_ifindex = orig_dev->ifindex;
2375 sll->sll_ifindex = dev->ifindex;
2379 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2380 if (po->tp_version <= TPACKET_V2) {
2383 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2386 for (start = h.raw; start < end; start += PAGE_SIZE)
2387 flush_dcache_page(pgv_to_page(start));
2392 if (po->tp_version <= TPACKET_V2) {
2393 __packet_set_status(po, h.raw, status);
2394 sk->sk_data_ready(sk);
2396 prb_clear_blk_fill_status(&po->rx_ring);
2400 if (skb_head != skb->data && skb_shared(skb)) {
2401 skb->data = skb_head;
2405 if (!is_drop_n_account)
2412 is_drop_n_account = true;
2413 po->stats.stats1.tp_drops++;
2414 spin_unlock(&sk->sk_receive_queue.lock);
2416 sk->sk_data_ready(sk);
2417 kfree_skb(copy_skb);
2418 goto drop_n_restore;
2421 static void tpacket_destruct_skb(struct sk_buff *skb)
2423 struct packet_sock *po = pkt_sk(skb->sk);
2425 if (likely(po->tx_ring.pg_vec)) {
2429 ph = skb_shinfo(skb)->destructor_arg;
2430 packet_dec_pending(&po->tx_ring);
2432 ts = __packet_set_timestamp(po, ph, skb);
2433 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2439 static void tpacket_set_protocol(const struct net_device *dev,
2440 struct sk_buff *skb)
2442 if (dev->type == ARPHRD_ETHER) {
2443 skb_reset_mac_header(skb);
2444 skb->protocol = eth_hdr(skb)->h_proto;
2448 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2450 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2451 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2452 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2453 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2454 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2455 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2456 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2458 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2464 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2465 struct virtio_net_hdr *vnet_hdr)
2467 if (*len < sizeof(*vnet_hdr))
2469 *len -= sizeof(*vnet_hdr);
2471 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2474 return __packet_snd_vnet_parse(vnet_hdr, *len);
2477 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2478 void *frame, struct net_device *dev, void *data, int tp_len,
2479 __be16 proto, unsigned char *addr, int hlen, int copylen,
2480 const struct sockcm_cookie *sockc)
2482 union tpacket_uhdr ph;
2483 int to_write, offset, len, nr_frags, len_max;
2484 struct socket *sock = po->sk.sk_socket;
2490 skb->protocol = proto;
2492 skb->priority = po->sk.sk_priority;
2493 skb->mark = po->sk.sk_mark;
2494 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2495 skb_shinfo(skb)->destructor_arg = ph.raw;
2497 skb_reserve(skb, hlen);
2498 skb_reset_network_header(skb);
2502 if (sock->type == SOCK_DGRAM) {
2503 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2505 if (unlikely(err < 0))
2507 } else if (copylen) {
2508 int hdrlen = min_t(int, copylen, tp_len);
2510 skb_push(skb, dev->hard_header_len);
2511 skb_put(skb, copylen - dev->hard_header_len);
2512 err = skb_store_bits(skb, 0, data, hdrlen);
2515 if (!dev_validate_header(dev, skb->data, hdrlen))
2518 tpacket_set_protocol(dev, skb);
2524 offset = offset_in_page(data);
2525 len_max = PAGE_SIZE - offset;
2526 len = ((to_write > len_max) ? len_max : to_write);
2528 skb->data_len = to_write;
2529 skb->len += to_write;
2530 skb->truesize += to_write;
2531 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2533 while (likely(to_write)) {
2534 nr_frags = skb_shinfo(skb)->nr_frags;
2536 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2537 pr_err("Packet exceed the number of skb frags(%lu)\n",
2542 page = pgv_to_page(data);
2544 flush_dcache_page(page);
2546 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2549 len_max = PAGE_SIZE;
2550 len = ((to_write > len_max) ? len_max : to_write);
2553 skb_probe_transport_header(skb, 0);
2558 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2559 int size_max, void **data)
2561 union tpacket_uhdr ph;
2566 switch (po->tp_version) {
2568 if (ph.h3->tp_next_offset != 0) {
2569 pr_warn_once("variable sized slot not supported");
2572 tp_len = ph.h3->tp_len;
2575 tp_len = ph.h2->tp_len;
2578 tp_len = ph.h1->tp_len;
2581 if (unlikely(tp_len > size_max)) {
2582 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2586 if (unlikely(po->tp_tx_has_off)) {
2587 int off_min, off_max;
2589 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2590 off_max = po->tx_ring.frame_size - tp_len;
2591 if (po->sk.sk_type == SOCK_DGRAM) {
2592 switch (po->tp_version) {
2594 off = ph.h3->tp_net;
2597 off = ph.h2->tp_net;
2600 off = ph.h1->tp_net;
2604 switch (po->tp_version) {
2606 off = ph.h3->tp_mac;
2609 off = ph.h2->tp_mac;
2612 off = ph.h1->tp_mac;
2616 if (unlikely((off < off_min) || (off_max < off)))
2619 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2622 *data = frame + off;
2626 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2628 struct sk_buff *skb;
2629 struct net_device *dev;
2630 struct virtio_net_hdr *vnet_hdr = NULL;
2631 struct sockcm_cookie sockc;
2633 int err, reserve = 0;
2635 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2636 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2637 int tp_len, size_max;
2638 unsigned char *addr;
2641 int status = TP_STATUS_AVAILABLE;
2642 int hlen, tlen, copylen = 0;
2644 mutex_lock(&po->pg_vec_lock);
2646 if (likely(saddr == NULL)) {
2647 dev = packet_cached_dev_get(po);
2652 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2654 if (msg->msg_namelen < (saddr->sll_halen
2655 + offsetof(struct sockaddr_ll,
2658 proto = saddr->sll_protocol;
2659 addr = saddr->sll_addr;
2660 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2664 if (unlikely(dev == NULL))
2667 if (unlikely(!(dev->flags & IFF_UP)))
2670 sockc.tsflags = po->sk.sk_tsflags;
2671 if (msg->msg_controllen) {
2672 err = sock_cmsg_send(&po->sk, msg, &sockc);
2677 if (po->sk.sk_socket->type == SOCK_RAW)
2678 reserve = dev->hard_header_len;
2679 size_max = po->tx_ring.frame_size
2680 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2682 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2683 size_max = dev->mtu + reserve + VLAN_HLEN;
2686 ph = packet_current_frame(po, &po->tx_ring,
2687 TP_STATUS_SEND_REQUEST);
2688 if (unlikely(ph == NULL)) {
2689 if (need_wait && need_resched())
2695 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2699 status = TP_STATUS_SEND_REQUEST;
2700 hlen = LL_RESERVED_SPACE(dev);
2701 tlen = dev->needed_tailroom;
2702 if (po->has_vnet_hdr) {
2704 data += sizeof(*vnet_hdr);
2705 tp_len -= sizeof(*vnet_hdr);
2707 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2711 copylen = __virtio16_to_cpu(vio_le(),
2714 copylen = max_t(int, copylen, dev->hard_header_len);
2715 skb = sock_alloc_send_skb(&po->sk,
2716 hlen + tlen + sizeof(struct sockaddr_ll) +
2717 (copylen - dev->hard_header_len),
2720 if (unlikely(skb == NULL)) {
2721 /* we assume the socket was initially writeable ... */
2722 if (likely(len_sum > 0))
2726 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2727 addr, hlen, copylen, &sockc);
2728 if (likely(tp_len >= 0) &&
2729 tp_len > dev->mtu + reserve &&
2730 !po->has_vnet_hdr &&
2731 !packet_extra_vlan_len_allowed(dev, skb))
2734 if (unlikely(tp_len < 0)) {
2737 __packet_set_status(po, ph,
2738 TP_STATUS_AVAILABLE);
2739 packet_increment_head(&po->tx_ring);
2743 status = TP_STATUS_WRONG_FORMAT;
2749 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2755 skb->destructor = tpacket_destruct_skb;
2756 __packet_set_status(po, ph, TP_STATUS_SENDING);
2757 packet_inc_pending(&po->tx_ring);
2759 status = TP_STATUS_SEND_REQUEST;
2760 err = po->xmit(skb);
2761 if (unlikely(err > 0)) {
2762 err = net_xmit_errno(err);
2763 if (err && __packet_get_status(po, ph) ==
2764 TP_STATUS_AVAILABLE) {
2765 /* skb was destructed already */
2770 * skb was dropped but not destructed yet;
2771 * let's treat it like congestion or err < 0
2775 packet_increment_head(&po->tx_ring);
2777 } while (likely((ph != NULL) ||
2778 /* Note: packet_read_pending() might be slow if we have
2779 * to call it as it's per_cpu variable, but in fast-path
2780 * we already short-circuit the loop with the first
2781 * condition, and luckily don't have to go that path
2784 (need_wait && packet_read_pending(&po->tx_ring))));
2790 __packet_set_status(po, ph, status);
2795 mutex_unlock(&po->pg_vec_lock);
2799 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2800 size_t reserve, size_t len,
2801 size_t linear, int noblock,
2804 struct sk_buff *skb;
2806 /* Under a page? Don't bother with paged skb. */
2807 if (prepad + len < PAGE_SIZE || !linear)
2810 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2815 skb_reserve(skb, reserve);
2816 skb_put(skb, linear);
2817 skb->data_len = len - linear;
2818 skb->len += len - linear;
2823 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2825 struct sock *sk = sock->sk;
2826 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2827 struct sk_buff *skb;
2828 struct net_device *dev;
2830 unsigned char *addr;
2831 int err, reserve = 0;
2832 struct sockcm_cookie sockc;
2833 struct virtio_net_hdr vnet_hdr = { 0 };
2835 struct packet_sock *po = pkt_sk(sk);
2836 bool has_vnet_hdr = false;
2837 int hlen, tlen, linear;
2841 * Get and verify the address.
2844 if (likely(saddr == NULL)) {
2845 dev = packet_cached_dev_get(po);
2850 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2852 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2854 proto = saddr->sll_protocol;
2855 addr = saddr->sll_addr;
2856 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2860 if (unlikely(dev == NULL))
2863 if (unlikely(!(dev->flags & IFF_UP)))
2866 sockc.tsflags = sk->sk_tsflags;
2867 sockc.mark = sk->sk_mark;
2868 if (msg->msg_controllen) {
2869 err = sock_cmsg_send(sk, msg, &sockc);
2874 if (sock->type == SOCK_RAW)
2875 reserve = dev->hard_header_len;
2876 if (po->has_vnet_hdr) {
2877 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2880 has_vnet_hdr = true;
2883 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2884 if (!netif_supports_nofcs(dev)) {
2885 err = -EPROTONOSUPPORT;
2888 extra_len = 4; /* We're doing our own CRC */
2892 if (!vnet_hdr.gso_type &&
2893 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2897 hlen = LL_RESERVED_SPACE(dev);
2898 tlen = dev->needed_tailroom;
2899 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2900 linear = max(linear, min_t(int, len, dev->hard_header_len));
2901 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2902 msg->msg_flags & MSG_DONTWAIT, &err);
2906 skb_reset_network_header(skb);
2909 if (sock->type == SOCK_DGRAM) {
2910 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2911 if (unlikely(offset < 0))
2913 } else if (reserve) {
2914 skb_reserve(skb, -reserve);
2917 /* Returns -EFAULT on error */
2918 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2922 if (sock->type == SOCK_RAW &&
2923 !dev_validate_header(dev, skb->data, len)) {
2928 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2930 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2931 !packet_extra_vlan_len_allowed(dev, skb)) {
2936 skb->protocol = proto;
2938 skb->priority = sk->sk_priority;
2939 skb->mark = sockc.mark;
2942 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2945 len += sizeof(vnet_hdr);
2948 skb_probe_transport_header(skb, reserve);
2950 if (unlikely(extra_len == 4))
2953 err = po->xmit(skb);
2954 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2970 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2972 struct sock *sk = sock->sk;
2973 struct packet_sock *po = pkt_sk(sk);
2975 if (po->tx_ring.pg_vec)
2976 return tpacket_snd(po, msg);
2978 return packet_snd(sock, msg, len);
2982 * Close a PACKET socket. This is fairly simple. We immediately go
2983 * to 'closed' state and remove our protocol entry in the device list.
2986 static int packet_release(struct socket *sock)
2988 struct sock *sk = sock->sk;
2989 struct packet_sock *po;
2990 struct packet_fanout *f;
2992 union tpacket_req_u req_u;
3000 mutex_lock(&net->packet.sklist_lock);
3001 sk_del_node_init_rcu(sk);
3002 mutex_unlock(&net->packet.sklist_lock);
3005 sock_prot_inuse_add(net, sk->sk_prot, -1);
3008 spin_lock(&po->bind_lock);
3009 unregister_prot_hook(sk, false);
3010 packet_cached_dev_reset(po);
3012 if (po->prot_hook.dev) {
3013 dev_put(po->prot_hook.dev);
3014 po->prot_hook.dev = NULL;
3016 spin_unlock(&po->bind_lock);
3018 packet_flush_mclist(sk);
3021 if (po->rx_ring.pg_vec) {
3022 memset(&req_u, 0, sizeof(req_u));
3023 packet_set_ring(sk, &req_u, 1, 0);
3026 if (po->tx_ring.pg_vec) {
3027 memset(&req_u, 0, sizeof(req_u));
3028 packet_set_ring(sk, &req_u, 1, 1);
3032 f = fanout_release(sk);
3037 kfree(po->rollover);
3038 fanout_release_data(f);
3042 * Now the socket is dead. No more input will appear.
3049 skb_queue_purge(&sk->sk_receive_queue);
3050 packet_free_pending(po);
3051 sk_refcnt_debug_release(sk);
3058 * Attach a packet hook.
3061 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3064 struct packet_sock *po = pkt_sk(sk);
3065 struct net_device *dev_curr;
3068 struct net_device *dev = NULL;
3070 bool unlisted = false;
3073 spin_lock(&po->bind_lock);
3082 dev = dev_get_by_name_rcu(sock_net(sk), name);
3087 } else if (ifindex) {
3088 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3098 proto_curr = po->prot_hook.type;
3099 dev_curr = po->prot_hook.dev;
3101 need_rehook = proto_curr != proto || dev_curr != dev;
3106 /* prevents packet_notifier() from calling
3107 * register_prot_hook()
3110 __unregister_prot_hook(sk, true);
3112 dev_curr = po->prot_hook.dev;
3114 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3118 BUG_ON(po->running);
3120 po->prot_hook.type = proto;
3122 if (unlikely(unlisted)) {
3124 po->prot_hook.dev = NULL;
3126 packet_cached_dev_reset(po);
3128 po->prot_hook.dev = dev;
3129 po->ifindex = dev ? dev->ifindex : 0;
3130 packet_cached_dev_assign(po, dev);
3136 if (proto == 0 || !need_rehook)
3139 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3140 register_prot_hook(sk);
3142 sk->sk_err = ENETDOWN;
3143 if (!sock_flag(sk, SOCK_DEAD))
3144 sk->sk_error_report(sk);
3149 spin_unlock(&po->bind_lock);
3155 * Bind a packet socket to a device
3158 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3161 struct sock *sk = sock->sk;
3162 char name[sizeof(uaddr->sa_data) + 1];
3168 if (addr_len != sizeof(struct sockaddr))
3170 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3173 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3174 name[sizeof(uaddr->sa_data)] = 0;
3176 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3179 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3181 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3182 struct sock *sk = sock->sk;
3188 if (addr_len < sizeof(struct sockaddr_ll))
3190 if (sll->sll_family != AF_PACKET)
3193 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3194 sll->sll_protocol ? : pkt_sk(sk)->num);
3197 static struct proto packet_proto = {
3199 .owner = THIS_MODULE,
3200 .obj_size = sizeof(struct packet_sock),
3204 * Create a packet of type SOCK_PACKET.
3207 static int packet_create(struct net *net, struct socket *sock, int protocol,
3211 struct packet_sock *po;
3212 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3215 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3217 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3218 sock->type != SOCK_PACKET)
3219 return -ESOCKTNOSUPPORT;
3221 sock->state = SS_UNCONNECTED;
3224 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3228 sock->ops = &packet_ops;
3229 if (sock->type == SOCK_PACKET)
3230 sock->ops = &packet_ops_spkt;
3232 sock_init_data(sock, sk);
3235 sk->sk_family = PF_PACKET;
3237 po->xmit = dev_queue_xmit;
3239 err = packet_alloc_pending(po);
3243 packet_cached_dev_reset(po);
3245 sk->sk_destruct = packet_sock_destruct;
3246 sk_refcnt_debug_inc(sk);
3249 * Attach a protocol block
3252 spin_lock_init(&po->bind_lock);
3253 mutex_init(&po->pg_vec_lock);
3254 po->rollover = NULL;
3255 po->prot_hook.func = packet_rcv;
3257 if (sock->type == SOCK_PACKET)
3258 po->prot_hook.func = packet_rcv_spkt;
3260 po->prot_hook.af_packet_priv = sk;
3263 po->prot_hook.type = proto;
3264 __register_prot_hook(sk);
3267 mutex_lock(&net->packet.sklist_lock);
3268 sk_add_node_rcu(sk, &net->packet.sklist);
3269 mutex_unlock(&net->packet.sklist_lock);
3272 sock_prot_inuse_add(net, &packet_proto, 1);
3283 * Pull a packet from our receive queue and hand it to the user.
3284 * If necessary we block.
3287 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3290 struct sock *sk = sock->sk;
3291 struct sk_buff *skb;
3293 int vnet_hdr_len = 0;
3294 unsigned int origlen = 0;
3297 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3301 /* What error should we return now? EUNATTACH? */
3302 if (pkt_sk(sk)->ifindex < 0)
3306 if (flags & MSG_ERRQUEUE) {
3307 err = sock_recv_errqueue(sk, msg, len,
3308 SOL_PACKET, PACKET_TX_TIMESTAMP);
3313 * Call the generic datagram receiver. This handles all sorts
3314 * of horrible races and re-entrancy so we can forget about it
3315 * in the protocol layers.
3317 * Now it will return ENETDOWN, if device have just gone down,
3318 * but then it will block.
3321 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3324 * An error occurred so return it. Because skb_recv_datagram()
3325 * handles the blocking we don't see and worry about blocking
3332 if (pkt_sk(sk)->pressure)
3333 packet_rcv_has_room(pkt_sk(sk), NULL);
3335 if (pkt_sk(sk)->has_vnet_hdr) {
3336 err = packet_rcv_vnet(msg, skb, &len);
3339 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3342 /* You lose any data beyond the buffer you gave. If it worries
3343 * a user program they can ask the device for its MTU
3349 msg->msg_flags |= MSG_TRUNC;
3352 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3356 if (sock->type != SOCK_PACKET) {
3357 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3359 /* Original length was stored in sockaddr_ll fields */
3360 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3361 sll->sll_family = AF_PACKET;
3362 sll->sll_protocol = skb->protocol;
3365 sock_recv_ts_and_drops(msg, sk, skb);
3367 if (msg->msg_name) {
3368 /* If the address length field is there to be filled
3369 * in, we fill it in now.
3371 if (sock->type == SOCK_PACKET) {
3372 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3373 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3375 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3377 msg->msg_namelen = sll->sll_halen +
3378 offsetof(struct sockaddr_ll, sll_addr);
3380 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3384 if (pkt_sk(sk)->auxdata) {
3385 struct tpacket_auxdata aux;
3387 aux.tp_status = TP_STATUS_USER;
3388 if (skb->ip_summed == CHECKSUM_PARTIAL)
3389 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3390 else if (skb->pkt_type != PACKET_OUTGOING &&
3391 (skb->ip_summed == CHECKSUM_COMPLETE ||
3392 skb_csum_unnecessary(skb)))
3393 aux.tp_status |= TP_STATUS_CSUM_VALID;
3395 aux.tp_len = origlen;
3396 aux.tp_snaplen = skb->len;
3398 aux.tp_net = skb_network_offset(skb);
3399 if (skb_vlan_tag_present(skb)) {
3400 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3401 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3402 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3404 aux.tp_vlan_tci = 0;
3405 aux.tp_vlan_tpid = 0;
3407 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3411 * Free or return the buffer as appropriate. Again this
3412 * hides all the races and re-entrancy issues from us.
3414 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3417 skb_free_datagram(sk, skb);
3422 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3425 struct net_device *dev;
3426 struct sock *sk = sock->sk;
3431 uaddr->sa_family = AF_PACKET;
3432 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3434 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3436 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3439 return sizeof(*uaddr);
3442 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3445 struct net_device *dev;
3446 struct sock *sk = sock->sk;
3447 struct packet_sock *po = pkt_sk(sk);
3448 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3453 sll->sll_family = AF_PACKET;
3454 sll->sll_ifindex = po->ifindex;
3455 sll->sll_protocol = po->num;
3456 sll->sll_pkttype = 0;
3458 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3460 sll->sll_hatype = dev->type;
3461 sll->sll_halen = dev->addr_len;
3462 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3464 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3469 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3472 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3476 case PACKET_MR_MULTICAST:
3477 if (i->alen != dev->addr_len)
3480 return dev_mc_add(dev, i->addr);
3482 return dev_mc_del(dev, i->addr);
3484 case PACKET_MR_PROMISC:
3485 return dev_set_promiscuity(dev, what);
3486 case PACKET_MR_ALLMULTI:
3487 return dev_set_allmulti(dev, what);
3488 case PACKET_MR_UNICAST:
3489 if (i->alen != dev->addr_len)
3492 return dev_uc_add(dev, i->addr);
3494 return dev_uc_del(dev, i->addr);
3502 static void packet_dev_mclist_delete(struct net_device *dev,
3503 struct packet_mclist **mlp)
3505 struct packet_mclist *ml;
3507 while ((ml = *mlp) != NULL) {
3508 if (ml->ifindex == dev->ifindex) {
3509 packet_dev_mc(dev, ml, -1);
3517 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3519 struct packet_sock *po = pkt_sk(sk);
3520 struct packet_mclist *ml, *i;
3521 struct net_device *dev;
3527 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3532 if (mreq->mr_alen > dev->addr_len)
3536 i = kmalloc(sizeof(*i), GFP_KERNEL);
3541 for (ml = po->mclist; ml; ml = ml->next) {
3542 if (ml->ifindex == mreq->mr_ifindex &&
3543 ml->type == mreq->mr_type &&
3544 ml->alen == mreq->mr_alen &&
3545 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3547 /* Free the new element ... */
3553 i->type = mreq->mr_type;
3554 i->ifindex = mreq->mr_ifindex;
3555 i->alen = mreq->mr_alen;
3556 memcpy(i->addr, mreq->mr_address, i->alen);
3557 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3559 i->next = po->mclist;
3561 err = packet_dev_mc(dev, i, 1);
3563 po->mclist = i->next;
3572 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3574 struct packet_mclist *ml, **mlp;
3578 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3579 if (ml->ifindex == mreq->mr_ifindex &&
3580 ml->type == mreq->mr_type &&
3581 ml->alen == mreq->mr_alen &&
3582 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3583 if (--ml->count == 0) {
3584 struct net_device *dev;
3586 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3588 packet_dev_mc(dev, ml, -1);
3598 static void packet_flush_mclist(struct sock *sk)
3600 struct packet_sock *po = pkt_sk(sk);
3601 struct packet_mclist *ml;
3607 while ((ml = po->mclist) != NULL) {
3608 struct net_device *dev;
3610 po->mclist = ml->next;
3611 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3613 packet_dev_mc(dev, ml, -1);
3620 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3622 struct sock *sk = sock->sk;
3623 struct packet_sock *po = pkt_sk(sk);
3626 if (level != SOL_PACKET)
3627 return -ENOPROTOOPT;
3630 case PACKET_ADD_MEMBERSHIP:
3631 case PACKET_DROP_MEMBERSHIP:
3633 struct packet_mreq_max mreq;
3635 memset(&mreq, 0, sizeof(mreq));
3636 if (len < sizeof(struct packet_mreq))
3638 if (len > sizeof(mreq))
3640 if (copy_from_user(&mreq, optval, len))
3642 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3644 if (optname == PACKET_ADD_MEMBERSHIP)
3645 ret = packet_mc_add(sk, &mreq);
3647 ret = packet_mc_drop(sk, &mreq);
3651 case PACKET_RX_RING:
3652 case PACKET_TX_RING:
3654 union tpacket_req_u req_u;
3658 switch (po->tp_version) {
3661 len = sizeof(req_u.req);
3665 len = sizeof(req_u.req3);
3671 if (copy_from_user(&req_u.req, optval, len))
3674 ret = packet_set_ring(sk, &req_u, 0,
3675 optname == PACKET_TX_RING);
3680 case PACKET_COPY_THRESH:
3684 if (optlen != sizeof(val))
3686 if (copy_from_user(&val, optval, sizeof(val)))
3689 pkt_sk(sk)->copy_thresh = val;
3692 case PACKET_VERSION:
3696 if (optlen != sizeof(val))
3698 if (copy_from_user(&val, optval, sizeof(val)))
3709 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3712 po->tp_version = val;
3718 case PACKET_RESERVE:
3722 if (optlen != sizeof(val))
3724 if (copy_from_user(&val, optval, sizeof(val)))
3729 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3732 po->tp_reserve = val;
3742 if (optlen != sizeof(val))
3744 if (copy_from_user(&val, optval, sizeof(val)))
3748 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3751 po->tp_loss = !!val;
3757 case PACKET_AUXDATA:
3761 if (optlen < sizeof(val))
3763 if (copy_from_user(&val, optval, sizeof(val)))
3767 po->auxdata = !!val;
3771 case PACKET_ORIGDEV:
3775 if (optlen < sizeof(val))
3777 if (copy_from_user(&val, optval, sizeof(val)))
3781 po->origdev = !!val;
3785 case PACKET_VNET_HDR:
3789 if (sock->type != SOCK_RAW)
3791 if (optlen < sizeof(val))
3793 if (copy_from_user(&val, optval, sizeof(val)))
3797 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3800 po->has_vnet_hdr = !!val;
3806 case PACKET_TIMESTAMP:
3810 if (optlen != sizeof(val))
3812 if (copy_from_user(&val, optval, sizeof(val)))
3815 po->tp_tstamp = val;
3822 if (optlen != sizeof(val))
3824 if (copy_from_user(&val, optval, sizeof(val)))
3827 return fanout_add(sk, val & 0xffff, val >> 16);
3829 case PACKET_FANOUT_DATA:
3834 return fanout_set_data(po, optval, optlen);
3836 case PACKET_TX_HAS_OFF:
3840 if (optlen != sizeof(val))
3842 if (copy_from_user(&val, optval, sizeof(val)))
3846 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3849 po->tp_tx_has_off = !!val;
3855 case PACKET_QDISC_BYPASS:
3859 if (optlen != sizeof(val))
3861 if (copy_from_user(&val, optval, sizeof(val)))
3864 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3868 return -ENOPROTOOPT;
3872 static int packet_getsockopt(struct socket *sock, int level, int optname,
3873 char __user *optval, int __user *optlen)
3876 int val, lv = sizeof(val);
3877 struct sock *sk = sock->sk;
3878 struct packet_sock *po = pkt_sk(sk);
3880 union tpacket_stats_u st;
3881 struct tpacket_rollover_stats rstats;
3883 if (level != SOL_PACKET)
3884 return -ENOPROTOOPT;
3886 if (get_user(len, optlen))
3893 case PACKET_STATISTICS:
3894 spin_lock_bh(&sk->sk_receive_queue.lock);
3895 memcpy(&st, &po->stats, sizeof(st));
3896 memset(&po->stats, 0, sizeof(po->stats));
3897 spin_unlock_bh(&sk->sk_receive_queue.lock);
3899 if (po->tp_version == TPACKET_V3) {
3900 lv = sizeof(struct tpacket_stats_v3);
3901 st.stats3.tp_packets += st.stats3.tp_drops;
3904 lv = sizeof(struct tpacket_stats);
3905 st.stats1.tp_packets += st.stats1.tp_drops;
3910 case PACKET_AUXDATA:
3913 case PACKET_ORIGDEV:
3916 case PACKET_VNET_HDR:
3917 val = po->has_vnet_hdr;
3919 case PACKET_VERSION:
3920 val = po->tp_version;
3923 if (len > sizeof(int))
3925 if (len < sizeof(int))
3927 if (copy_from_user(&val, optval, len))
3931 val = sizeof(struct tpacket_hdr);
3934 val = sizeof(struct tpacket2_hdr);
3937 val = sizeof(struct tpacket3_hdr);
3943 case PACKET_RESERVE:
3944 val = po->tp_reserve;
3949 case PACKET_TIMESTAMP:
3950 val = po->tp_tstamp;
3954 ((u32)po->fanout->id |
3955 ((u32)po->fanout->type << 16) |
3956 ((u32)po->fanout->flags << 24)) :
3959 case PACKET_ROLLOVER_STATS:
3962 rstats.tp_all = atomic_long_read(&po->rollover->num);
3963 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3964 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3966 lv = sizeof(rstats);
3968 case PACKET_TX_HAS_OFF:
3969 val = po->tp_tx_has_off;
3971 case PACKET_QDISC_BYPASS:
3972 val = packet_use_direct_xmit(po);
3975 return -ENOPROTOOPT;
3980 if (put_user(len, optlen))
3982 if (copy_to_user(optval, data, len))
3988 #ifdef CONFIG_COMPAT
3989 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3990 char __user *optval, unsigned int optlen)
3992 struct packet_sock *po = pkt_sk(sock->sk);
3994 if (level != SOL_PACKET)
3995 return -ENOPROTOOPT;
3997 if (optname == PACKET_FANOUT_DATA &&
3998 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3999 optval = (char __user *)get_compat_bpf_fprog(optval);
4002 optlen = sizeof(struct sock_fprog);
4005 return packet_setsockopt(sock, level, optname, optval, optlen);
4009 static int packet_notifier(struct notifier_block *this,
4010 unsigned long msg, void *ptr)
4013 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4014 struct net *net = dev_net(dev);
4017 sk_for_each_rcu(sk, &net->packet.sklist) {
4018 struct packet_sock *po = pkt_sk(sk);
4021 case NETDEV_UNREGISTER:
4023 packet_dev_mclist_delete(dev, &po->mclist);
4027 if (dev->ifindex == po->ifindex) {
4028 spin_lock(&po->bind_lock);
4030 __unregister_prot_hook(sk, false);
4031 sk->sk_err = ENETDOWN;
4032 if (!sock_flag(sk, SOCK_DEAD))
4033 sk->sk_error_report(sk);
4035 if (msg == NETDEV_UNREGISTER) {
4036 packet_cached_dev_reset(po);
4038 if (po->prot_hook.dev)
4039 dev_put(po->prot_hook.dev);
4040 po->prot_hook.dev = NULL;
4042 spin_unlock(&po->bind_lock);
4046 if (dev->ifindex == po->ifindex) {
4047 spin_lock(&po->bind_lock);
4049 register_prot_hook(sk);
4050 spin_unlock(&po->bind_lock);
4060 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4063 struct sock *sk = sock->sk;
4068 int amount = sk_wmem_alloc_get(sk);
4070 return put_user(amount, (int __user *)arg);
4074 struct sk_buff *skb;
4077 spin_lock_bh(&sk->sk_receive_queue.lock);
4078 skb = skb_peek(&sk->sk_receive_queue);
4081 spin_unlock_bh(&sk->sk_receive_queue.lock);
4082 return put_user(amount, (int __user *)arg);
4085 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4087 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4097 case SIOCGIFBRDADDR:
4098 case SIOCSIFBRDADDR:
4099 case SIOCGIFNETMASK:
4100 case SIOCSIFNETMASK:
4101 case SIOCGIFDSTADDR:
4102 case SIOCSIFDSTADDR:
4104 return inet_dgram_ops.ioctl(sock, cmd, arg);
4108 return -ENOIOCTLCMD;
4113 static __poll_t packet_poll(struct file *file, struct socket *sock,
4116 struct sock *sk = sock->sk;
4117 struct packet_sock *po = pkt_sk(sk);
4118 __poll_t mask = datagram_poll(file, sock, wait);
4120 spin_lock_bh(&sk->sk_receive_queue.lock);
4121 if (po->rx_ring.pg_vec) {
4122 if (!packet_previous_rx_frame(po, &po->rx_ring,
4124 mask |= EPOLLIN | EPOLLRDNORM;
4126 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4128 spin_unlock_bh(&sk->sk_receive_queue.lock);
4129 spin_lock_bh(&sk->sk_write_queue.lock);
4130 if (po->tx_ring.pg_vec) {
4131 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4132 mask |= EPOLLOUT | EPOLLWRNORM;
4134 spin_unlock_bh(&sk->sk_write_queue.lock);
4139 /* Dirty? Well, I still did not learn better way to account
4143 static void packet_mm_open(struct vm_area_struct *vma)
4145 struct file *file = vma->vm_file;
4146 struct socket *sock = file->private_data;
4147 struct sock *sk = sock->sk;
4150 atomic_inc(&pkt_sk(sk)->mapped);
4153 static void packet_mm_close(struct vm_area_struct *vma)
4155 struct file *file = vma->vm_file;
4156 struct socket *sock = file->private_data;
4157 struct sock *sk = sock->sk;
4160 atomic_dec(&pkt_sk(sk)->mapped);
4163 static const struct vm_operations_struct packet_mmap_ops = {
4164 .open = packet_mm_open,
4165 .close = packet_mm_close,
4168 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4173 for (i = 0; i < len; i++) {
4174 if (likely(pg_vec[i].buffer)) {
4175 if (is_vmalloc_addr(pg_vec[i].buffer))
4176 vfree(pg_vec[i].buffer);
4178 free_pages((unsigned long)pg_vec[i].buffer,
4180 pg_vec[i].buffer = NULL;
4186 static char *alloc_one_pg_vec_page(unsigned long order)
4189 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4190 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4192 buffer = (char *) __get_free_pages(gfp_flags, order);
4196 /* __get_free_pages failed, fall back to vmalloc */
4197 buffer = vzalloc((1 << order) * PAGE_SIZE);
4201 /* vmalloc failed, lets dig into swap here */
4202 gfp_flags &= ~__GFP_NORETRY;
4203 buffer = (char *) __get_free_pages(gfp_flags, order);
4207 /* complete and utter failure */
4211 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4213 unsigned int block_nr = req->tp_block_nr;
4217 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4218 if (unlikely(!pg_vec))
4221 for (i = 0; i < block_nr; i++) {
4222 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4223 if (unlikely(!pg_vec[i].buffer))
4224 goto out_free_pgvec;
4231 free_pg_vec(pg_vec, order, block_nr);
4236 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4237 int closing, int tx_ring)
4239 struct pgv *pg_vec = NULL;
4240 struct packet_sock *po = pkt_sk(sk);
4241 int was_running, order = 0;
4242 struct packet_ring_buffer *rb;
4243 struct sk_buff_head *rb_queue;
4246 /* Added to avoid minimal code churn */
4247 struct tpacket_req *req = &req_u->req;
4249 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4250 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4254 if (atomic_read(&po->mapped))
4256 if (packet_read_pending(rb))
4260 if (req->tp_block_nr) {
4261 /* Sanity tests and some calculations */
4263 if (unlikely(rb->pg_vec))
4266 switch (po->tp_version) {
4268 po->tp_hdrlen = TPACKET_HDRLEN;
4271 po->tp_hdrlen = TPACKET2_HDRLEN;
4274 po->tp_hdrlen = TPACKET3_HDRLEN;
4279 if (unlikely((int)req->tp_block_size <= 0))
4281 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4283 if (po->tp_version >= TPACKET_V3 &&
4284 req->tp_block_size <=
4285 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
4287 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4290 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4293 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4294 if (unlikely(rb->frames_per_block == 0))
4296 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4298 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4303 order = get_order(req->tp_block_size);
4304 pg_vec = alloc_pg_vec(req, order);
4305 if (unlikely(!pg_vec))
4307 switch (po->tp_version) {
4309 /* Block transmit is not supported yet */
4311 init_prb_bdqc(po, rb, pg_vec, req_u);
4313 struct tpacket_req3 *req3 = &req_u->req3;
4315 if (req3->tp_retire_blk_tov ||
4316 req3->tp_sizeof_priv ||
4317 req3->tp_feature_req_word) {
4330 if (unlikely(req->tp_frame_nr))
4335 /* Detach socket from network */
4336 spin_lock(&po->bind_lock);
4337 was_running = po->running;
4341 __unregister_prot_hook(sk, false);
4343 spin_unlock(&po->bind_lock);
4348 mutex_lock(&po->pg_vec_lock);
4349 if (closing || atomic_read(&po->mapped) == 0) {
4351 spin_lock_bh(&rb_queue->lock);
4352 swap(rb->pg_vec, pg_vec);
4353 rb->frame_max = (req->tp_frame_nr - 1);
4355 rb->frame_size = req->tp_frame_size;
4356 spin_unlock_bh(&rb_queue->lock);
4358 swap(rb->pg_vec_order, order);
4359 swap(rb->pg_vec_len, req->tp_block_nr);
4361 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4362 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4363 tpacket_rcv : packet_rcv;
4364 skb_queue_purge(rb_queue);
4365 if (atomic_read(&po->mapped))
4366 pr_err("packet_mmap: vma is busy: %d\n",
4367 atomic_read(&po->mapped));
4369 mutex_unlock(&po->pg_vec_lock);
4371 spin_lock(&po->bind_lock);
4374 register_prot_hook(sk);
4376 spin_unlock(&po->bind_lock);
4377 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4378 /* Because we don't support block-based V3 on tx-ring */
4380 prb_shutdown_retire_blk_timer(po, rb_queue);
4384 free_pg_vec(pg_vec, order, req->tp_block_nr);
4389 static int packet_mmap(struct file *file, struct socket *sock,
4390 struct vm_area_struct *vma)
4392 struct sock *sk = sock->sk;
4393 struct packet_sock *po = pkt_sk(sk);
4394 unsigned long size, expected_size;
4395 struct packet_ring_buffer *rb;
4396 unsigned long start;
4403 mutex_lock(&po->pg_vec_lock);
4406 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4408 expected_size += rb->pg_vec_len
4414 if (expected_size == 0)
4417 size = vma->vm_end - vma->vm_start;
4418 if (size != expected_size)
4421 start = vma->vm_start;
4422 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4423 if (rb->pg_vec == NULL)
4426 for (i = 0; i < rb->pg_vec_len; i++) {
4428 void *kaddr = rb->pg_vec[i].buffer;
4431 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4432 page = pgv_to_page(kaddr);
4433 err = vm_insert_page(vma, start, page);
4442 atomic_inc(&po->mapped);
4443 vma->vm_ops = &packet_mmap_ops;
4447 mutex_unlock(&po->pg_vec_lock);
4451 static const struct proto_ops packet_ops_spkt = {
4452 .family = PF_PACKET,
4453 .owner = THIS_MODULE,
4454 .release = packet_release,
4455 .bind = packet_bind_spkt,
4456 .connect = sock_no_connect,
4457 .socketpair = sock_no_socketpair,
4458 .accept = sock_no_accept,
4459 .getname = packet_getname_spkt,
4460 .poll = datagram_poll,
4461 .ioctl = packet_ioctl,
4462 .listen = sock_no_listen,
4463 .shutdown = sock_no_shutdown,
4464 .setsockopt = sock_no_setsockopt,
4465 .getsockopt = sock_no_getsockopt,
4466 .sendmsg = packet_sendmsg_spkt,
4467 .recvmsg = packet_recvmsg,
4468 .mmap = sock_no_mmap,
4469 .sendpage = sock_no_sendpage,
4472 static const struct proto_ops packet_ops = {
4473 .family = PF_PACKET,
4474 .owner = THIS_MODULE,
4475 .release = packet_release,
4476 .bind = packet_bind,
4477 .connect = sock_no_connect,
4478 .socketpair = sock_no_socketpair,
4479 .accept = sock_no_accept,
4480 .getname = packet_getname,
4481 .poll = packet_poll,
4482 .ioctl = packet_ioctl,
4483 .listen = sock_no_listen,
4484 .shutdown = sock_no_shutdown,
4485 .setsockopt = packet_setsockopt,
4486 .getsockopt = packet_getsockopt,
4487 #ifdef CONFIG_COMPAT
4488 .compat_setsockopt = compat_packet_setsockopt,
4490 .sendmsg = packet_sendmsg,
4491 .recvmsg = packet_recvmsg,
4492 .mmap = packet_mmap,
4493 .sendpage = sock_no_sendpage,
4496 static const struct net_proto_family packet_family_ops = {
4497 .family = PF_PACKET,
4498 .create = packet_create,
4499 .owner = THIS_MODULE,
4502 static struct notifier_block packet_netdev_notifier = {
4503 .notifier_call = packet_notifier,
4506 #ifdef CONFIG_PROC_FS
4508 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4511 struct net *net = seq_file_net(seq);
4514 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4517 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4519 struct net *net = seq_file_net(seq);
4520 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4523 static void packet_seq_stop(struct seq_file *seq, void *v)
4529 static int packet_seq_show(struct seq_file *seq, void *v)
4531 if (v == SEQ_START_TOKEN)
4532 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4534 struct sock *s = sk_entry(v);
4535 const struct packet_sock *po = pkt_sk(s);
4538 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4540 refcount_read(&s->sk_refcnt),
4545 atomic_read(&s->sk_rmem_alloc),
4546 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4553 static const struct seq_operations packet_seq_ops = {
4554 .start = packet_seq_start,
4555 .next = packet_seq_next,
4556 .stop = packet_seq_stop,
4557 .show = packet_seq_show,
4560 static int packet_seq_open(struct inode *inode, struct file *file)
4562 return seq_open_net(inode, file, &packet_seq_ops,
4563 sizeof(struct seq_net_private));
4566 static const struct file_operations packet_seq_fops = {
4567 .open = packet_seq_open,
4569 .llseek = seq_lseek,
4570 .release = seq_release_net,
4575 static int __net_init packet_net_init(struct net *net)
4577 mutex_init(&net->packet.sklist_lock);
4578 INIT_HLIST_HEAD(&net->packet.sklist);
4580 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4586 static void __net_exit packet_net_exit(struct net *net)
4588 remove_proc_entry("packet", net->proc_net);
4589 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4592 static struct pernet_operations packet_net_ops = {
4593 .init = packet_net_init,
4594 .exit = packet_net_exit,
4598 static void __exit packet_exit(void)
4600 unregister_netdevice_notifier(&packet_netdev_notifier);
4601 unregister_pernet_subsys(&packet_net_ops);
4602 sock_unregister(PF_PACKET);
4603 proto_unregister(&packet_proto);
4606 static int __init packet_init(void)
4608 int rc = proto_register(&packet_proto, 0);
4613 sock_register(&packet_family_ops);
4614 register_pernet_subsys(&packet_net_ops);
4615 register_netdevice_notifier(&packet_netdev_notifier);
4620 module_init(packet_init);
4621 module_exit(packet_exit);
4622 MODULE_LICENSE("GPL");
4623 MODULE_ALIAS_NETPROTO(PF_PACKET);