2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <linux/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
96 #include <net/compat.h>
102 - if device has no dev->hard_header routine, it adds and removes ll header
103 inside itself. In this case ll header is invisible outside of device,
104 but higher levels still should reserve dev->hard_header_len.
105 Some devices are enough clever to reallocate skb, when header
106 will not fit to reserved space (tunnel), another ones are silly
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
122 Incoming, dev->hard_header==NULL
123 mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 header. PPP makes it, that is wrong, because introduce
125 assymetry between rx and tx paths.
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
139 dev->hard_header != NULL
140 mac_header -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
156 struct packet_mreq_max {
158 unsigned short mr_type;
159 unsigned short mr_alen;
160 unsigned char mr_address[MAX_ADDR_LEN];
164 struct tpacket_hdr *h1;
165 struct tpacket2_hdr *h2;
166 struct tpacket3_hdr *h3;
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 int closing, int tx_ring);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define PGV_FROM_VMALLOC 1
182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
194 static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
217 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb);
219 struct packet_skb_cb {
221 struct sockaddr_pkt pkt;
223 /* Trick: alias skb original length with
224 * ll.sll_family and ll.protocol in order
227 unsigned int origlen;
228 struct sockaddr_ll ll;
233 #define vio_le() virtio_legacy_is_little_endian()
235 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
237 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
238 #define GET_PBLOCK_DESC(x, bid) \
239 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
240 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
241 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
242 #define GET_NEXT_PRB_BLK_NUM(x) \
243 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
244 ((x)->kactive_blk_num+1) : 0)
246 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
247 static void __fanout_link(struct sock *sk, struct packet_sock *po);
249 static int packet_direct_xmit(struct sk_buff *skb)
251 struct net_device *dev = skb->dev;
252 struct sk_buff *orig_skb = skb;
253 struct netdev_queue *txq;
254 int ret = NETDEV_TX_BUSY;
256 if (unlikely(!netif_running(dev) ||
257 !netif_carrier_ok(dev)))
260 skb = validate_xmit_skb_list(skb, dev);
264 packet_pick_tx_queue(dev, skb);
265 txq = skb_get_tx_queue(dev, skb);
269 HARD_TX_LOCK(dev, txq, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq))
271 ret = netdev_start_xmit(skb, dev, txq, false);
272 HARD_TX_UNLOCK(dev, txq);
276 if (!dev_xmit_complete(ret))
281 atomic_long_inc(&dev->tx_dropped);
283 return NET_XMIT_DROP;
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
288 struct net_device *dev;
291 dev = rcu_dereference(po->cached_dev);
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
302 rcu_assign_pointer(po->cached_dev, dev);
305 static void packet_cached_dev_reset(struct packet_sock *po)
307 RCU_INIT_POINTER(po->cached_dev, NULL);
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
312 return po->xmit == packet_direct_xmit;
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
322 const struct net_device_ops *ops = dev->netdev_ops;
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
330 queue_index = __packet_pick_tx_queue(dev, skb);
333 skb_set_queue_mapping(skb, queue_index);
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
340 static void register_prot_hook(struct sock *sk)
342 struct packet_sock *po = pkt_sk(sk);
346 __fanout_link(sk, po);
348 dev_add_pack(&po->prot_hook);
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
364 struct packet_sock *po = pkt_sk(sk);
369 __fanout_unlink(sk, po);
371 __dev_remove_pack(&po->prot_hook);
376 spin_unlock(&po->bind_lock);
378 spin_lock(&po->bind_lock);
382 static void unregister_prot_hook(struct sock *sk, bool sync)
384 struct packet_sock *po = pkt_sk(sk);
387 __unregister_prot_hook(sk, sync);
390 static inline struct page * __pure pgv_to_page(void *addr)
392 if (is_vmalloc_addr(addr))
393 return vmalloc_to_page(addr);
394 return virt_to_page(addr);
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
399 union tpacket_uhdr h;
402 switch (po->tp_version) {
404 h.h1->tp_status = status;
405 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
408 h.h2->tp_status = status;
409 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
412 h.h3->tp_status = status;
413 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
416 WARN(1, "TPACKET version not supported.\n");
423 static int __packet_get_status(struct packet_sock *po, void *frame)
425 union tpacket_uhdr h;
430 switch (po->tp_version) {
432 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
433 return h.h1->tp_status;
435 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
436 return h.h2->tp_status;
438 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
439 return h.h3->tp_status;
441 WARN(1, "TPACKET version not supported.\n");
447 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
450 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
453 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
454 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
455 return TP_STATUS_TS_RAW_HARDWARE;
457 if (ktime_to_timespec_cond(skb->tstamp, ts))
458 return TP_STATUS_TS_SOFTWARE;
463 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
466 union tpacket_uhdr h;
470 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
474 switch (po->tp_version) {
476 h.h1->tp_sec = ts.tv_sec;
477 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
480 h.h2->tp_sec = ts.tv_sec;
481 h.h2->tp_nsec = ts.tv_nsec;
484 h.h3->tp_sec = ts.tv_sec;
485 h.h3->tp_nsec = ts.tv_nsec;
488 WARN(1, "TPACKET version not supported.\n");
492 /* one flush is safe, as both fields always lie on the same cacheline */
493 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
499 static void *packet_lookup_frame(struct packet_sock *po,
500 struct packet_ring_buffer *rb,
501 unsigned int position,
504 unsigned int pg_vec_pos, frame_offset;
505 union tpacket_uhdr h;
507 pg_vec_pos = position / rb->frames_per_block;
508 frame_offset = position % rb->frames_per_block;
510 h.raw = rb->pg_vec[pg_vec_pos].buffer +
511 (frame_offset * rb->frame_size);
513 if (status != __packet_get_status(po, h.raw))
519 static void *packet_current_frame(struct packet_sock *po,
520 struct packet_ring_buffer *rb,
523 return packet_lookup_frame(po, rb, rb->head, status);
526 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
528 del_timer_sync(&pkc->retire_blk_timer);
531 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
532 struct sk_buff_head *rb_queue)
534 struct tpacket_kbdq_core *pkc;
536 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
538 spin_lock_bh(&rb_queue->lock);
539 pkc->delete_blk_timer = 1;
540 spin_unlock_bh(&rb_queue->lock);
542 prb_del_retire_blk_timer(pkc);
545 static void prb_init_blk_timer(struct packet_sock *po,
546 struct tpacket_kbdq_core *pkc,
547 void (*func) (unsigned long))
549 init_timer(&pkc->retire_blk_timer);
550 pkc->retire_blk_timer.data = (long)po;
551 pkc->retire_blk_timer.function = func;
552 pkc->retire_blk_timer.expires = jiffies;
555 static void prb_setup_retire_blk_timer(struct packet_sock *po)
557 struct tpacket_kbdq_core *pkc;
559 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
560 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
563 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
564 int blk_size_in_bytes)
566 struct net_device *dev;
567 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
568 struct ethtool_link_ksettings ecmd;
572 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
573 if (unlikely(!dev)) {
575 return DEFAULT_PRB_RETIRE_TOV;
577 err = __ethtool_get_link_ksettings(dev, &ecmd);
581 * If the link speed is so slow you don't really
582 * need to worry about perf anyways
584 if (ecmd.base.speed < SPEED_1000 ||
585 ecmd.base.speed == SPEED_UNKNOWN) {
586 return DEFAULT_PRB_RETIRE_TOV;
589 div = ecmd.base.speed / 1000;
593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
606 union tpacket_req_u *req_u)
608 p1->feature_req_word = req_u->req3.tp_feature_req_word;
611 static void init_prb_bdqc(struct packet_sock *po,
612 struct packet_ring_buffer *rb,
614 union tpacket_req_u *req_u)
616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
617 struct tpacket_block_desc *pbd;
619 memset(p1, 0x0, sizeof(*p1));
621 p1->knxt_seq_num = 1;
623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
624 p1->pkblk_start = pg_vec[0].buffer;
625 p1->kblk_size = req_u->req3.tp_block_size;
626 p1->knum_blocks = req_u->req3.tp_block_nr;
627 p1->hdrlen = po->tp_hdrlen;
628 p1->version = po->tp_version;
629 p1->last_kactive_blk_num = 0;
630 po->stats.stats3.tp_freeze_q_cnt = 0;
631 if (req_u->req3.tp_retire_blk_tov)
632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 req_u->req3.tp_block_size);
636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
639 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
640 prb_init_ft_ops(p1, req_u);
641 prb_setup_retire_blk_timer(po);
642 prb_open_block(p1, pbd);
645 /* Do NOT update the last_blk_num first.
646 * Assumes sk_buff_head lock is held.
648 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
650 mod_timer(&pkc->retire_blk_timer,
651 jiffies + pkc->tov_in_jiffies);
652 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
657 * 1) We refresh the timer only when we open a block.
658 * By doing this we don't waste cycles refreshing the timer
659 * on packet-by-packet basis.
661 * With a 1MB block-size, on a 1Gbps line, it will take
662 * i) ~8 ms to fill a block + ii) memcpy etc.
663 * In this cut we are not accounting for the memcpy time.
665 * So, if the user sets the 'tmo' to 10ms then the timer
666 * will never fire while the block is still getting filled
667 * (which is what we want). However, the user could choose
668 * to close a block early and that's fine.
670 * But when the timer does fire, we check whether or not to refresh it.
671 * Since the tmo granularity is in msecs, it is not too expensive
672 * to refresh the timer, lets say every '8' msecs.
673 * Either the user can set the 'tmo' or we can derive it based on
674 * a) line-speed and b) block-size.
675 * prb_calc_retire_blk_tmo() calculates the tmo.
678 static void prb_retire_rx_blk_timer_expired(unsigned long data)
680 struct packet_sock *po = (struct packet_sock *)data;
681 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
683 struct tpacket_block_desc *pbd;
685 spin_lock(&po->sk.sk_receive_queue.lock);
687 frozen = prb_queue_frozen(pkc);
688 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
690 if (unlikely(pkc->delete_blk_timer))
693 /* We only need to plug the race when the block is partially filled.
695 * lock(); increment BLOCK_NUM_PKTS; unlock()
696 * copy_bits() is in progress ...
697 * timer fires on other cpu:
698 * we can't retire the current block because copy_bits
702 if (BLOCK_NUM_PKTS(pbd)) {
703 while (atomic_read(&pkc->blk_fill_in_prog)) {
704 /* Waiting for skb_copy_bits to finish... */
709 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
711 if (!BLOCK_NUM_PKTS(pbd)) {
712 /* An empty block. Just refresh the timer. */
715 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
716 if (!prb_dispatch_next_block(pkc, po))
721 /* Case 1. Queue was frozen because user-space was
724 if (prb_curr_blk_in_use(pbd)) {
726 * Ok, user-space is still behind.
727 * So just refresh the timer.
731 /* Case 2. queue was frozen,user-space caught up,
732 * now the link went idle && the timer fired.
733 * We don't have a block to close.So we open this
734 * block and restart the timer.
735 * opening a block thaws the queue,restarts timer
736 * Thawing/timer-refresh is a side effect.
738 prb_open_block(pkc, pbd);
745 _prb_refresh_rx_retire_blk_timer(pkc);
748 spin_unlock(&po->sk.sk_receive_queue.lock);
751 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
752 struct tpacket_block_desc *pbd1, __u32 status)
754 /* Flush everything minus the block header */
756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
761 /* Skip the block header(we know header WILL fit in 4K) */
764 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
765 for (; start < end; start += PAGE_SIZE)
766 flush_dcache_page(pgv_to_page(start));
771 /* Now update the block status. */
773 BLOCK_STATUS(pbd1) = status;
775 /* Flush the block header */
777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
779 flush_dcache_page(pgv_to_page(start));
789 * 2) Increment active_blk_num
791 * Note:We DONT refresh the timer on purpose.
792 * Because almost always the next block will be opened.
794 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
795 struct tpacket_block_desc *pbd1,
796 struct packet_sock *po, unsigned int stat)
798 __u32 status = TP_STATUS_USER | stat;
800 struct tpacket3_hdr *last_pkt;
801 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
802 struct sock *sk = &po->sk;
804 if (po->stats.stats3.tp_drops)
805 status |= TP_STATUS_LOSING;
807 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
808 last_pkt->tp_next_offset = 0;
810 /* Get the ts of the last pkt */
811 if (BLOCK_NUM_PKTS(pbd1)) {
812 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
813 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
815 /* Ok, we tmo'd - so get the current time.
817 * It shouldn't really happen as we don't close empty
818 * blocks. See prb_retire_rx_blk_timer_expired().
822 h1->ts_last_pkt.ts_sec = ts.tv_sec;
823 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
828 /* Flush the block */
829 prb_flush_block(pkc1, pbd1, status);
831 sk->sk_data_ready(sk);
833 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
836 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
838 pkc->reset_pending_on_curr_blk = 0;
842 * Side effect of opening a block:
844 * 1) prb_queue is thawed.
845 * 2) retire_blk_timer is refreshed.
848 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
849 struct tpacket_block_desc *pbd1)
852 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
856 /* We could have just memset this but we will lose the
857 * flexibility of making the priv area sticky
860 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
861 BLOCK_NUM_PKTS(pbd1) = 0;
862 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 h1->ts_first_pkt.ts_sec = ts.tv_sec;
867 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
869 pkc1->pkblk_start = (char *)pbd1;
870 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
872 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
873 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
875 pbd1->version = pkc1->version;
876 pkc1->prev = pkc1->nxt_offset;
877 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
879 prb_thaw_queue(pkc1);
880 _prb_refresh_rx_retire_blk_timer(pkc1);
886 * Queue freeze logic:
887 * 1) Assume tp_block_nr = 8 blocks.
888 * 2) At time 't0', user opens Rx ring.
889 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
890 * 4) user-space is either sleeping or processing block '0'.
891 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
892 * it will close block-7,loop around and try to fill block '0'.
894 * __packet_lookup_frame_in_block
895 * prb_retire_current_block()
896 * prb_dispatch_next_block()
897 * |->(BLOCK_STATUS == USER) evaluates to true
898 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
899 * 6) Now there are two cases:
900 * 6.1) Link goes idle right after the queue is frozen.
901 * But remember, the last open_block() refreshed the timer.
902 * When this timer expires,it will refresh itself so that we can
903 * re-open block-0 in near future.
904 * 6.2) Link is busy and keeps on receiving packets. This is a simple
905 * case and __packet_lookup_frame_in_block will check if block-0
906 * is free and can now be re-used.
908 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
909 struct packet_sock *po)
911 pkc->reset_pending_on_curr_blk = 1;
912 po->stats.stats3.tp_freeze_q_cnt++;
915 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
918 * If the next block is free then we will dispatch it
919 * and return a good offset.
920 * Else, we will freeze the queue.
921 * So, caller must check the return value.
923 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
924 struct packet_sock *po)
926 struct tpacket_block_desc *pbd;
930 /* 1. Get current block num */
931 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
933 /* 2. If this block is currently in_use then freeze the queue */
934 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
935 prb_freeze_queue(pkc, po);
941 * open this block and return the offset where the first packet
942 * needs to get stored.
944 prb_open_block(pkc, pbd);
945 return (void *)pkc->nxt_offset;
948 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
949 struct packet_sock *po, unsigned int status)
951 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
953 /* retire/close the current block */
954 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
956 * Plug the case where copy_bits() is in progress on
957 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
958 * have space to copy the pkt in the current block and
959 * called prb_retire_current_block()
961 * We don't need to worry about the TMO case because
962 * the timer-handler already handled this case.
964 if (!(status & TP_STATUS_BLK_TMO)) {
965 while (atomic_read(&pkc->blk_fill_in_prog)) {
966 /* Waiting for skb_copy_bits to finish... */
970 prb_close_block(pkc, pbd, po, status);
975 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
977 return TP_STATUS_USER & BLOCK_STATUS(pbd);
980 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
982 return pkc->reset_pending_on_curr_blk;
985 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
987 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
988 atomic_dec(&pkc->blk_fill_in_prog);
991 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
992 struct tpacket3_hdr *ppd)
994 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
997 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
998 struct tpacket3_hdr *ppd)
1000 ppd->hv1.tp_rxhash = 0;
1003 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1004 struct tpacket3_hdr *ppd)
1006 if (skb_vlan_tag_present(pkc->skb)) {
1007 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1008 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1009 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1011 ppd->hv1.tp_vlan_tci = 0;
1012 ppd->hv1.tp_vlan_tpid = 0;
1013 ppd->tp_status = TP_STATUS_AVAILABLE;
1017 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1018 struct tpacket3_hdr *ppd)
1020 ppd->hv1.tp_padding = 0;
1021 prb_fill_vlan_info(pkc, ppd);
1023 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1024 prb_fill_rxhash(pkc, ppd);
1026 prb_clear_rxhash(pkc, ppd);
1029 static void prb_fill_curr_block(char *curr,
1030 struct tpacket_kbdq_core *pkc,
1031 struct tpacket_block_desc *pbd,
1034 struct tpacket3_hdr *ppd;
1036 ppd = (struct tpacket3_hdr *)curr;
1037 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1039 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1040 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 BLOCK_NUM_PKTS(pbd) += 1;
1042 atomic_inc(&pkc->blk_fill_in_prog);
1043 prb_run_all_ft_ops(pkc, ppd);
1046 /* Assumes caller has the sk->rx_queue.lock */
1047 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1048 struct sk_buff *skb,
1053 struct tpacket_kbdq_core *pkc;
1054 struct tpacket_block_desc *pbd;
1057 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1058 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1060 /* Queue is frozen when user space is lagging behind */
1061 if (prb_queue_frozen(pkc)) {
1063 * Check if that last block which caused the queue to freeze,
1064 * is still in_use by user-space.
1066 if (prb_curr_blk_in_use(pbd)) {
1067 /* Can't record this packet */
1071 * Ok, the block was released by user-space.
1072 * Now let's open that block.
1073 * opening a block also thaws the queue.
1074 * Thawing is a side effect.
1076 prb_open_block(pkc, pbd);
1081 curr = pkc->nxt_offset;
1083 end = (char *)pbd + pkc->kblk_size;
1085 /* first try the current block */
1086 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1087 prb_fill_curr_block(curr, pkc, pbd, len);
1088 return (void *)curr;
1091 /* Ok, close the current block */
1092 prb_retire_current_block(pkc, po, 0);
1094 /* Now, try to dispatch the next block */
1095 curr = (char *)prb_dispatch_next_block(pkc, po);
1097 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1098 prb_fill_curr_block(curr, pkc, pbd, len);
1099 return (void *)curr;
1103 * No free blocks are available.user_space hasn't caught up yet.
1104 * Queue was just frozen and now this packet will get dropped.
1109 static void *packet_current_rx_frame(struct packet_sock *po,
1110 struct sk_buff *skb,
1111 int status, unsigned int len)
1114 switch (po->tp_version) {
1117 curr = packet_lookup_frame(po, &po->rx_ring,
1118 po->rx_ring.head, status);
1121 return __packet_lookup_frame_in_block(po, skb, status, len);
1123 WARN(1, "TPACKET version not supported\n");
1129 static void *prb_lookup_block(struct packet_sock *po,
1130 struct packet_ring_buffer *rb,
1134 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1135 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1137 if (status != BLOCK_STATUS(pbd))
1142 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1145 if (rb->prb_bdqc.kactive_blk_num)
1146 prev = rb->prb_bdqc.kactive_blk_num-1;
1148 prev = rb->prb_bdqc.knum_blocks-1;
1152 /* Assumes caller has held the rx_queue.lock */
1153 static void *__prb_previous_block(struct packet_sock *po,
1154 struct packet_ring_buffer *rb,
1157 unsigned int previous = prb_previous_blk_num(rb);
1158 return prb_lookup_block(po, rb, previous, status);
1161 static void *packet_previous_rx_frame(struct packet_sock *po,
1162 struct packet_ring_buffer *rb,
1165 if (po->tp_version <= TPACKET_V2)
1166 return packet_previous_frame(po, rb, status);
1168 return __prb_previous_block(po, rb, status);
1171 static void packet_increment_rx_head(struct packet_sock *po,
1172 struct packet_ring_buffer *rb)
1174 switch (po->tp_version) {
1177 return packet_increment_head(rb);
1180 WARN(1, "TPACKET version not supported.\n");
1186 static void *packet_previous_frame(struct packet_sock *po,
1187 struct packet_ring_buffer *rb,
1190 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1191 return packet_lookup_frame(po, rb, previous, status);
1194 static void packet_increment_head(struct packet_ring_buffer *buff)
1196 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1199 static void packet_inc_pending(struct packet_ring_buffer *rb)
1201 this_cpu_inc(*rb->pending_refcnt);
1204 static void packet_dec_pending(struct packet_ring_buffer *rb)
1206 this_cpu_dec(*rb->pending_refcnt);
1209 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1211 unsigned int refcnt = 0;
1214 /* We don't use pending refcount in rx_ring. */
1215 if (rb->pending_refcnt == NULL)
1218 for_each_possible_cpu(cpu)
1219 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1224 static int packet_alloc_pending(struct packet_sock *po)
1226 po->rx_ring.pending_refcnt = NULL;
1228 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1229 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1235 static void packet_free_pending(struct packet_sock *po)
1237 free_percpu(po->tx_ring.pending_refcnt);
1240 #define ROOM_POW_OFF 2
1241 #define ROOM_NONE 0x0
1242 #define ROOM_LOW 0x1
1243 #define ROOM_NORMAL 0x2
1245 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1249 len = po->rx_ring.frame_max + 1;
1250 idx = po->rx_ring.head;
1252 idx += len >> pow_off;
1255 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1258 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1262 len = po->rx_ring.prb_bdqc.knum_blocks;
1263 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1265 idx += len >> pow_off;
1268 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1271 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1273 struct sock *sk = &po->sk;
1274 int ret = ROOM_NONE;
1276 if (po->prot_hook.func != tpacket_rcv) {
1277 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1278 - (skb ? skb->truesize : 0);
1279 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1287 if (po->tp_version == TPACKET_V3) {
1288 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1290 else if (__tpacket_v3_has_room(po, 0))
1293 if (__tpacket_has_room(po, ROOM_POW_OFF))
1295 else if (__tpacket_has_room(po, 0))
1302 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1307 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1308 ret = __packet_rcv_has_room(po, skb);
1309 has_room = ret == ROOM_NORMAL;
1310 if (po->pressure == has_room)
1311 po->pressure = !has_room;
1312 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1317 static void packet_sock_destruct(struct sock *sk)
1319 skb_queue_purge(&sk->sk_error_queue);
1321 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1322 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1324 if (!sock_flag(sk, SOCK_DEAD)) {
1325 pr_err("Attempt to release alive packet socket: %p\n", sk);
1329 sk_refcnt_debug_dec(sk);
1332 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1337 rxhash = skb_get_hash(skb);
1338 for (i = 0; i < ROLLOVER_HLEN; i++)
1339 if (po->rollover->history[i] == rxhash)
1342 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1343 return count > (ROLLOVER_HLEN >> 1);
1346 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1347 struct sk_buff *skb,
1350 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1353 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1354 struct sk_buff *skb,
1357 unsigned int val = atomic_inc_return(&f->rr_cur);
1362 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1363 struct sk_buff *skb,
1366 return smp_processor_id() % num;
1369 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1370 struct sk_buff *skb,
1373 return prandom_u32_max(num);
1376 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1377 struct sk_buff *skb,
1378 unsigned int idx, bool try_self,
1381 struct packet_sock *po, *po_next, *po_skip = NULL;
1382 unsigned int i, j, room = ROOM_NONE;
1384 po = pkt_sk(f->arr[idx]);
1387 room = packet_rcv_has_room(po, skb);
1388 if (room == ROOM_NORMAL ||
1389 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1394 i = j = min_t(int, po->rollover->sock, num - 1);
1396 po_next = pkt_sk(f->arr[i]);
1397 if (po_next != po_skip && !po_next->pressure &&
1398 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1400 po->rollover->sock = i;
1401 atomic_long_inc(&po->rollover->num);
1402 if (room == ROOM_LOW)
1403 atomic_long_inc(&po->rollover->num_huge);
1411 atomic_long_inc(&po->rollover->num_failed);
1415 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1416 struct sk_buff *skb,
1419 return skb_get_queue_mapping(skb) % num;
1422 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1423 struct sk_buff *skb,
1426 struct bpf_prog *prog;
1427 unsigned int ret = 0;
1430 prog = rcu_dereference(f->bpf_prog);
1432 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1438 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1440 return f->flags & (flag >> 8);
1443 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1444 struct packet_type *pt, struct net_device *orig_dev)
1446 struct packet_fanout *f = pt->af_packet_priv;
1447 unsigned int num = READ_ONCE(f->num_members);
1448 struct net *net = read_pnet(&f->net);
1449 struct packet_sock *po;
1452 if (!net_eq(dev_net(dev), net) || !num) {
1457 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1458 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1463 case PACKET_FANOUT_HASH:
1465 idx = fanout_demux_hash(f, skb, num);
1467 case PACKET_FANOUT_LB:
1468 idx = fanout_demux_lb(f, skb, num);
1470 case PACKET_FANOUT_CPU:
1471 idx = fanout_demux_cpu(f, skb, num);
1473 case PACKET_FANOUT_RND:
1474 idx = fanout_demux_rnd(f, skb, num);
1476 case PACKET_FANOUT_QM:
1477 idx = fanout_demux_qm(f, skb, num);
1479 case PACKET_FANOUT_ROLLOVER:
1480 idx = fanout_demux_rollover(f, skb, 0, false, num);
1482 case PACKET_FANOUT_CBPF:
1483 case PACKET_FANOUT_EBPF:
1484 idx = fanout_demux_bpf(f, skb, num);
1488 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1489 idx = fanout_demux_rollover(f, skb, idx, true, num);
1491 po = pkt_sk(f->arr[idx]);
1492 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1495 DEFINE_MUTEX(fanout_mutex);
1496 EXPORT_SYMBOL_GPL(fanout_mutex);
1497 static LIST_HEAD(fanout_list);
1498 static u16 fanout_next_id;
1500 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1502 struct packet_fanout *f = po->fanout;
1504 spin_lock(&f->lock);
1505 f->arr[f->num_members] = sk;
1508 if (f->num_members == 1)
1509 dev_add_pack(&f->prot_hook);
1510 spin_unlock(&f->lock);
1513 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1515 struct packet_fanout *f = po->fanout;
1518 spin_lock(&f->lock);
1519 for (i = 0; i < f->num_members; i++) {
1520 if (f->arr[i] == sk)
1523 BUG_ON(i >= f->num_members);
1524 f->arr[i] = f->arr[f->num_members - 1];
1526 if (f->num_members == 0)
1527 __dev_remove_pack(&f->prot_hook);
1528 spin_unlock(&f->lock);
1531 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1533 if (sk->sk_family != PF_PACKET)
1536 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1539 static void fanout_init_data(struct packet_fanout *f)
1542 case PACKET_FANOUT_LB:
1543 atomic_set(&f->rr_cur, 0);
1545 case PACKET_FANOUT_CBPF:
1546 case PACKET_FANOUT_EBPF:
1547 RCU_INIT_POINTER(f->bpf_prog, NULL);
1552 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1554 struct bpf_prog *old;
1556 spin_lock(&f->lock);
1557 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1558 rcu_assign_pointer(f->bpf_prog, new);
1559 spin_unlock(&f->lock);
1563 bpf_prog_destroy(old);
1567 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1570 struct bpf_prog *new;
1571 struct sock_fprog fprog;
1574 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1576 if (len != sizeof(fprog))
1578 if (copy_from_user(&fprog, data, len))
1581 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1585 __fanout_set_data_bpf(po->fanout, new);
1589 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1592 struct bpf_prog *new;
1595 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1597 if (len != sizeof(fd))
1599 if (copy_from_user(&fd, data, len))
1602 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1604 return PTR_ERR(new);
1606 __fanout_set_data_bpf(po->fanout, new);
1610 static int fanout_set_data(struct packet_sock *po, char __user *data,
1613 switch (po->fanout->type) {
1614 case PACKET_FANOUT_CBPF:
1615 return fanout_set_data_cbpf(po, data, len);
1616 case PACKET_FANOUT_EBPF:
1617 return fanout_set_data_ebpf(po, data, len);
1623 static void fanout_release_data(struct packet_fanout *f)
1626 case PACKET_FANOUT_CBPF:
1627 case PACKET_FANOUT_EBPF:
1628 __fanout_set_data_bpf(f, NULL);
1632 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1634 struct packet_fanout *f;
1636 list_for_each_entry(f, &fanout_list, list) {
1637 if (f->id == candidate_id &&
1638 read_pnet(&f->net) == sock_net(sk)) {
1645 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1647 u16 id = fanout_next_id;
1650 if (__fanout_id_is_free(sk, id)) {
1652 fanout_next_id = id + 1;
1657 } while (id != fanout_next_id);
1662 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1664 struct packet_rollover *rollover = NULL;
1665 struct packet_sock *po = pkt_sk(sk);
1666 struct packet_fanout *f, *match;
1667 u8 type = type_flags & 0xff;
1668 u8 flags = type_flags >> 8;
1672 case PACKET_FANOUT_ROLLOVER:
1673 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1675 case PACKET_FANOUT_HASH:
1676 case PACKET_FANOUT_LB:
1677 case PACKET_FANOUT_CPU:
1678 case PACKET_FANOUT_RND:
1679 case PACKET_FANOUT_QM:
1680 case PACKET_FANOUT_CBPF:
1681 case PACKET_FANOUT_EBPF:
1687 mutex_lock(&fanout_mutex);
1697 if (type == PACKET_FANOUT_ROLLOVER ||
1698 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1700 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1703 atomic_long_set(&rollover->num, 0);
1704 atomic_long_set(&rollover->num_huge, 0);
1705 atomic_long_set(&rollover->num_failed, 0);
1706 po->rollover = rollover;
1709 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1714 if (!fanout_find_new_id(sk, &id)) {
1718 /* ephemeral flag for the first socket in the group: drop it */
1719 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1723 list_for_each_entry(f, &fanout_list, list) {
1725 read_pnet(&f->net) == sock_net(sk)) {
1731 if (match && match->flags != flags)
1735 match = kzalloc(sizeof(*match), GFP_KERNEL);
1738 write_pnet(&match->net, sock_net(sk));
1741 match->flags = flags;
1742 INIT_LIST_HEAD(&match->list);
1743 spin_lock_init(&match->lock);
1744 refcount_set(&match->sk_ref, 0);
1745 fanout_init_data(match);
1746 match->prot_hook.type = po->prot_hook.type;
1747 match->prot_hook.dev = po->prot_hook.dev;
1748 match->prot_hook.func = packet_rcv_fanout;
1749 match->prot_hook.af_packet_priv = match;
1750 match->prot_hook.id_match = match_fanout_group;
1751 list_add(&match->list, &fanout_list);
1754 if (match->type == type &&
1755 match->prot_hook.type == po->prot_hook.type &&
1756 match->prot_hook.dev == po->prot_hook.dev) {
1758 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1759 __dev_remove_pack(&po->prot_hook);
1761 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1762 __fanout_link(sk, po);
1767 if (err && rollover) {
1769 po->rollover = NULL;
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))
1796 kfree_rcu(po->rollover, rcu);
1798 mutex_unlock(&fanout_mutex);
1803 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1804 struct sk_buff *skb)
1806 /* Earlier code assumed this would be a VLAN pkt, double-check
1807 * this now that we have the actual packet in hand. We can only
1808 * do this check on Ethernet devices.
1810 if (unlikely(dev->type != ARPHRD_ETHER))
1813 skb_reset_mac_header(skb);
1814 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1817 static const struct proto_ops packet_ops;
1819 static const struct proto_ops packet_ops_spkt;
1821 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1822 struct packet_type *pt, struct net_device *orig_dev)
1825 struct sockaddr_pkt *spkt;
1828 * When we registered the protocol we saved the socket in the data
1829 * field for just this event.
1832 sk = pt->af_packet_priv;
1835 * Yank back the headers [hope the device set this
1836 * right or kerboom...]
1838 * Incoming packets have ll header pulled,
1841 * For outgoing ones skb->data == skb_mac_header(skb)
1842 * so that this procedure is noop.
1845 if (skb->pkt_type == PACKET_LOOPBACK)
1848 if (!net_eq(dev_net(dev), sock_net(sk)))
1851 skb = skb_share_check(skb, GFP_ATOMIC);
1855 /* drop any routing info */
1858 /* drop conntrack reference */
1861 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1863 skb_push(skb, skb->data - skb_mac_header(skb));
1866 * The SOCK_PACKET socket receives _all_ frames.
1869 spkt->spkt_family = dev->type;
1870 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1871 spkt->spkt_protocol = skb->protocol;
1874 * Charge the memory to the socket. This is done specifically
1875 * to prevent sockets using all the memory up.
1878 if (sock_queue_rcv_skb(sk, skb) == 0)
1889 * Output a raw packet to a device layer. This bypasses all the other
1890 * protocol layers and you must therefore supply it with a complete frame
1893 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1896 struct sock *sk = sock->sk;
1897 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1898 struct sk_buff *skb = NULL;
1899 struct net_device *dev;
1900 struct sockcm_cookie sockc;
1906 * Get and verify the address.
1910 if (msg->msg_namelen < sizeof(struct sockaddr))
1912 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1913 proto = saddr->spkt_protocol;
1915 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1918 * Find the device first to size check it
1921 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1924 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1930 if (!(dev->flags & IFF_UP))
1934 * You may not queue a frame bigger than the mtu. This is the lowest level
1935 * raw protocol and you must do your own fragmentation at this level.
1938 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1939 if (!netif_supports_nofcs(dev)) {
1940 err = -EPROTONOSUPPORT;
1943 extra_len = 4; /* We're doing our own CRC */
1947 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1951 size_t reserved = LL_RESERVED_SPACE(dev);
1952 int tlen = dev->needed_tailroom;
1953 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1956 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1959 /* FIXME: Save some space for broken drivers that write a hard
1960 * header at transmission time by themselves. PPP is the notable
1961 * one here. This should really be fixed at the driver level.
1963 skb_reserve(skb, reserved);
1964 skb_reset_network_header(skb);
1966 /* Try to align data part correctly */
1971 skb_reset_network_header(skb);
1973 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1979 if (!dev_validate_header(dev, skb->data, len)) {
1983 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1984 !packet_extra_vlan_len_allowed(dev, skb)) {
1989 sockc.tsflags = sk->sk_tsflags;
1990 if (msg->msg_controllen) {
1991 err = sock_cmsg_send(sk, msg, &sockc);
1996 skb->protocol = proto;
1998 skb->priority = sk->sk_priority;
1999 skb->mark = sk->sk_mark;
2001 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2003 if (unlikely(extra_len == 4))
2006 skb_probe_transport_header(skb, 0);
2008 dev_queue_xmit(skb);
2019 static unsigned int run_filter(struct sk_buff *skb,
2020 const struct sock *sk,
2023 struct sk_filter *filter;
2026 filter = rcu_dereference(sk->sk_filter);
2028 res = bpf_prog_run_clear_cb(filter->prog, skb);
2034 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2037 struct virtio_net_hdr vnet_hdr;
2039 if (*len < sizeof(vnet_hdr))
2041 *len -= sizeof(vnet_hdr);
2043 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
2046 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2050 * This function makes lazy skb cloning in hope that most of packets
2051 * are discarded by BPF.
2053 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2054 * and skb->cb are mangled. It works because (and until) packets
2055 * falling here are owned by current CPU. Output packets are cloned
2056 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2057 * sequencially, so that if we return skb to original state on exit,
2058 * we will not harm anyone.
2061 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2062 struct packet_type *pt, struct net_device *orig_dev)
2065 struct sockaddr_ll *sll;
2066 struct packet_sock *po;
2067 u8 *skb_head = skb->data;
2068 int skb_len = skb->len;
2069 unsigned int snaplen, res;
2070 bool is_drop_n_account = false;
2072 if (skb->pkt_type == PACKET_LOOPBACK)
2075 sk = pt->af_packet_priv;
2078 if (!net_eq(dev_net(dev), sock_net(sk)))
2083 if (dev->header_ops) {
2084 /* The device has an explicit notion of ll header,
2085 * exported to higher levels.
2087 * Otherwise, the device hides details of its frame
2088 * structure, so that corresponding packet head is
2089 * never delivered to user.
2091 if (sk->sk_type != SOCK_DGRAM)
2092 skb_push(skb, skb->data - skb_mac_header(skb));
2093 else if (skb->pkt_type == PACKET_OUTGOING) {
2094 /* Special case: outgoing packets have ll header at head */
2095 skb_pull(skb, skb_network_offset(skb));
2101 res = run_filter(skb, sk, snaplen);
2103 goto drop_n_restore;
2107 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2110 if (skb_shared(skb)) {
2111 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2115 if (skb_head != skb->data) {
2116 skb->data = skb_head;
2123 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2125 sll = &PACKET_SKB_CB(skb)->sa.ll;
2126 sll->sll_hatype = dev->type;
2127 sll->sll_pkttype = skb->pkt_type;
2128 if (unlikely(po->origdev))
2129 sll->sll_ifindex = orig_dev->ifindex;
2131 sll->sll_ifindex = dev->ifindex;
2133 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2135 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2136 * Use their space for storing the original skb length.
2138 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2140 if (pskb_trim(skb, snaplen))
2143 skb_set_owner_r(skb, sk);
2147 /* drop conntrack reference */
2150 spin_lock(&sk->sk_receive_queue.lock);
2151 po->stats.stats1.tp_packets++;
2152 sock_skb_set_dropcount(sk, skb);
2153 __skb_queue_tail(&sk->sk_receive_queue, skb);
2154 spin_unlock(&sk->sk_receive_queue.lock);
2155 sk->sk_data_ready(sk);
2159 is_drop_n_account = true;
2160 spin_lock(&sk->sk_receive_queue.lock);
2161 po->stats.stats1.tp_drops++;
2162 atomic_inc(&sk->sk_drops);
2163 spin_unlock(&sk->sk_receive_queue.lock);
2166 if (skb_head != skb->data && skb_shared(skb)) {
2167 skb->data = skb_head;
2171 if (!is_drop_n_account)
2178 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2179 struct packet_type *pt, struct net_device *orig_dev)
2182 struct packet_sock *po;
2183 struct sockaddr_ll *sll;
2184 union tpacket_uhdr h;
2185 u8 *skb_head = skb->data;
2186 int skb_len = skb->len;
2187 unsigned int snaplen, res;
2188 unsigned long status = TP_STATUS_USER;
2189 unsigned short macoff, netoff, hdrlen;
2190 struct sk_buff *copy_skb = NULL;
2193 bool is_drop_n_account = false;
2195 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2196 * We may add members to them until current aligned size without forcing
2197 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2199 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2200 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2202 if (skb->pkt_type == PACKET_LOOPBACK)
2205 sk = pt->af_packet_priv;
2208 if (!net_eq(dev_net(dev), sock_net(sk)))
2211 if (dev->header_ops) {
2212 if (sk->sk_type != SOCK_DGRAM)
2213 skb_push(skb, skb->data - skb_mac_header(skb));
2214 else if (skb->pkt_type == PACKET_OUTGOING) {
2215 /* Special case: outgoing packets have ll header at head */
2216 skb_pull(skb, skb_network_offset(skb));
2222 res = run_filter(skb, sk, snaplen);
2224 goto drop_n_restore;
2226 if (skb->ip_summed == CHECKSUM_PARTIAL)
2227 status |= TP_STATUS_CSUMNOTREADY;
2228 else if (skb->pkt_type != PACKET_OUTGOING &&
2229 (skb->ip_summed == CHECKSUM_COMPLETE ||
2230 skb_csum_unnecessary(skb)))
2231 status |= TP_STATUS_CSUM_VALID;
2236 if (sk->sk_type == SOCK_DGRAM) {
2237 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2240 unsigned int maclen = skb_network_offset(skb);
2241 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2242 (maclen < 16 ? 16 : maclen)) +
2244 if (po->has_vnet_hdr)
2245 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)
2265 } else if (unlikely(macoff + snaplen >
2266 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2269 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2270 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2271 snaplen, nval, macoff);
2273 if (unlikely((int)snaplen < 0)) {
2275 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2278 spin_lock(&sk->sk_receive_queue.lock);
2279 h.raw = packet_current_rx_frame(po, skb,
2280 TP_STATUS_KERNEL, (macoff+snaplen));
2282 goto drop_n_account;
2283 if (po->tp_version <= TPACKET_V2) {
2284 packet_increment_rx_head(po, &po->rx_ring);
2286 * LOSING will be reported till you read the stats,
2287 * because it's COR - Clear On Read.
2288 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2291 if (po->stats.stats1.tp_drops)
2292 status |= TP_STATUS_LOSING;
2294 po->stats.stats1.tp_packets++;
2296 status |= TP_STATUS_COPY;
2297 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2299 spin_unlock(&sk->sk_receive_queue.lock);
2301 if (po->has_vnet_hdr) {
2302 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2303 sizeof(struct virtio_net_hdr),
2305 spin_lock(&sk->sk_receive_queue.lock);
2306 goto drop_n_account;
2310 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2312 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2313 getnstimeofday(&ts);
2315 status |= ts_status;
2317 switch (po->tp_version) {
2319 h.h1->tp_len = skb->len;
2320 h.h1->tp_snaplen = snaplen;
2321 h.h1->tp_mac = macoff;
2322 h.h1->tp_net = netoff;
2323 h.h1->tp_sec = ts.tv_sec;
2324 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2325 hdrlen = sizeof(*h.h1);
2328 h.h2->tp_len = skb->len;
2329 h.h2->tp_snaplen = snaplen;
2330 h.h2->tp_mac = macoff;
2331 h.h2->tp_net = netoff;
2332 h.h2->tp_sec = ts.tv_sec;
2333 h.h2->tp_nsec = ts.tv_nsec;
2334 if (skb_vlan_tag_present(skb)) {
2335 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2336 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2337 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2339 h.h2->tp_vlan_tci = 0;
2340 h.h2->tp_vlan_tpid = 0;
2342 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2343 hdrlen = sizeof(*h.h2);
2346 /* tp_nxt_offset,vlan are already populated above.
2347 * So DONT clear those fields here
2349 h.h3->tp_status |= status;
2350 h.h3->tp_len = skb->len;
2351 h.h3->tp_snaplen = snaplen;
2352 h.h3->tp_mac = macoff;
2353 h.h3->tp_net = netoff;
2354 h.h3->tp_sec = ts.tv_sec;
2355 h.h3->tp_nsec = ts.tv_nsec;
2356 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2357 hdrlen = sizeof(*h.h3);
2363 sll = h.raw + TPACKET_ALIGN(hdrlen);
2364 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2365 sll->sll_family = AF_PACKET;
2366 sll->sll_hatype = dev->type;
2367 sll->sll_protocol = skb->protocol;
2368 sll->sll_pkttype = skb->pkt_type;
2369 if (unlikely(po->origdev))
2370 sll->sll_ifindex = orig_dev->ifindex;
2372 sll->sll_ifindex = dev->ifindex;
2376 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2377 if (po->tp_version <= TPACKET_V2) {
2380 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2383 for (start = h.raw; start < end; start += PAGE_SIZE)
2384 flush_dcache_page(pgv_to_page(start));
2389 if (po->tp_version <= TPACKET_V2) {
2390 __packet_set_status(po, h.raw, status);
2391 sk->sk_data_ready(sk);
2393 prb_clear_blk_fill_status(&po->rx_ring);
2397 if (skb_head != skb->data && skb_shared(skb)) {
2398 skb->data = skb_head;
2402 if (!is_drop_n_account)
2409 is_drop_n_account = true;
2410 po->stats.stats1.tp_drops++;
2411 spin_unlock(&sk->sk_receive_queue.lock);
2413 sk->sk_data_ready(sk);
2414 kfree_skb(copy_skb);
2415 goto drop_n_restore;
2418 static void tpacket_destruct_skb(struct sk_buff *skb)
2420 struct packet_sock *po = pkt_sk(skb->sk);
2422 if (likely(po->tx_ring.pg_vec)) {
2426 ph = skb_shinfo(skb)->destructor_arg;
2427 packet_dec_pending(&po->tx_ring);
2429 ts = __packet_set_timestamp(po, ph, skb);
2430 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2436 static void tpacket_set_protocol(const struct net_device *dev,
2437 struct sk_buff *skb)
2439 if (dev->type == ARPHRD_ETHER) {
2440 skb_reset_mac_header(skb);
2441 skb->protocol = eth_hdr(skb)->h_proto;
2445 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2447 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2448 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2449 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2450 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2451 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2452 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2453 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2455 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2461 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2462 struct virtio_net_hdr *vnet_hdr)
2464 if (*len < sizeof(*vnet_hdr))
2466 *len -= sizeof(*vnet_hdr);
2468 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2471 return __packet_snd_vnet_parse(vnet_hdr, *len);
2474 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2475 void *frame, struct net_device *dev, void *data, int tp_len,
2476 __be16 proto, unsigned char *addr, int hlen, int copylen,
2477 const struct sockcm_cookie *sockc)
2479 union tpacket_uhdr ph;
2480 int to_write, offset, len, nr_frags, len_max;
2481 struct socket *sock = po->sk.sk_socket;
2487 skb->protocol = proto;
2489 skb->priority = po->sk.sk_priority;
2490 skb->mark = po->sk.sk_mark;
2491 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2492 skb_shinfo(skb)->destructor_arg = ph.raw;
2494 skb_reserve(skb, hlen);
2495 skb_reset_network_header(skb);
2499 if (sock->type == SOCK_DGRAM) {
2500 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2502 if (unlikely(err < 0))
2504 } else if (copylen) {
2505 int hdrlen = min_t(int, copylen, tp_len);
2507 skb_push(skb, dev->hard_header_len);
2508 skb_put(skb, copylen - dev->hard_header_len);
2509 err = skb_store_bits(skb, 0, data, hdrlen);
2512 if (!dev_validate_header(dev, skb->data, hdrlen))
2515 tpacket_set_protocol(dev, skb);
2521 offset = offset_in_page(data);
2522 len_max = PAGE_SIZE - offset;
2523 len = ((to_write > len_max) ? len_max : to_write);
2525 skb->data_len = to_write;
2526 skb->len += to_write;
2527 skb->truesize += to_write;
2528 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2530 while (likely(to_write)) {
2531 nr_frags = skb_shinfo(skb)->nr_frags;
2533 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2534 pr_err("Packet exceed the number of skb frags(%lu)\n",
2539 page = pgv_to_page(data);
2541 flush_dcache_page(page);
2543 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2546 len_max = PAGE_SIZE;
2547 len = ((to_write > len_max) ? len_max : to_write);
2550 skb_probe_transport_header(skb, 0);
2555 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2556 int size_max, void **data)
2558 union tpacket_uhdr ph;
2563 switch (po->tp_version) {
2565 if (ph.h3->tp_next_offset != 0) {
2566 pr_warn_once("variable sized slot not supported");
2569 tp_len = ph.h3->tp_len;
2572 tp_len = ph.h2->tp_len;
2575 tp_len = ph.h1->tp_len;
2578 if (unlikely(tp_len > size_max)) {
2579 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2583 if (unlikely(po->tp_tx_has_off)) {
2584 int off_min, off_max;
2586 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2587 off_max = po->tx_ring.frame_size - tp_len;
2588 if (po->sk.sk_type == SOCK_DGRAM) {
2589 switch (po->tp_version) {
2591 off = ph.h3->tp_net;
2594 off = ph.h2->tp_net;
2597 off = ph.h1->tp_net;
2601 switch (po->tp_version) {
2603 off = ph.h3->tp_mac;
2606 off = ph.h2->tp_mac;
2609 off = ph.h1->tp_mac;
2613 if (unlikely((off < off_min) || (off_max < off)))
2616 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2619 *data = frame + off;
2623 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2625 struct sk_buff *skb;
2626 struct net_device *dev;
2627 struct virtio_net_hdr *vnet_hdr = NULL;
2628 struct sockcm_cookie sockc;
2630 int err, reserve = 0;
2632 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2633 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2634 int tp_len, size_max;
2635 unsigned char *addr;
2638 int status = TP_STATUS_AVAILABLE;
2639 int hlen, tlen, copylen = 0;
2641 mutex_lock(&po->pg_vec_lock);
2643 if (likely(saddr == NULL)) {
2644 dev = packet_cached_dev_get(po);
2649 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2651 if (msg->msg_namelen < (saddr->sll_halen
2652 + offsetof(struct sockaddr_ll,
2655 proto = saddr->sll_protocol;
2656 addr = saddr->sll_addr;
2657 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2661 if (unlikely(dev == NULL))
2664 if (unlikely(!(dev->flags & IFF_UP)))
2667 sockc.tsflags = po->sk.sk_tsflags;
2668 if (msg->msg_controllen) {
2669 err = sock_cmsg_send(&po->sk, msg, &sockc);
2674 if (po->sk.sk_socket->type == SOCK_RAW)
2675 reserve = dev->hard_header_len;
2676 size_max = po->tx_ring.frame_size
2677 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2679 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2680 size_max = dev->mtu + reserve + VLAN_HLEN;
2683 ph = packet_current_frame(po, &po->tx_ring,
2684 TP_STATUS_SEND_REQUEST);
2685 if (unlikely(ph == NULL)) {
2686 if (need_wait && need_resched())
2692 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2696 status = TP_STATUS_SEND_REQUEST;
2697 hlen = LL_RESERVED_SPACE(dev);
2698 tlen = dev->needed_tailroom;
2699 if (po->has_vnet_hdr) {
2701 data += sizeof(*vnet_hdr);
2702 tp_len -= sizeof(*vnet_hdr);
2704 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2708 copylen = __virtio16_to_cpu(vio_le(),
2711 copylen = max_t(int, copylen, dev->hard_header_len);
2712 skb = sock_alloc_send_skb(&po->sk,
2713 hlen + tlen + sizeof(struct sockaddr_ll) +
2714 (copylen - dev->hard_header_len),
2717 if (unlikely(skb == NULL)) {
2718 /* we assume the socket was initially writeable ... */
2719 if (likely(len_sum > 0))
2723 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2724 addr, hlen, copylen, &sockc);
2725 if (likely(tp_len >= 0) &&
2726 tp_len > dev->mtu + reserve &&
2727 !po->has_vnet_hdr &&
2728 !packet_extra_vlan_len_allowed(dev, skb))
2731 if (unlikely(tp_len < 0)) {
2734 __packet_set_status(po, ph,
2735 TP_STATUS_AVAILABLE);
2736 packet_increment_head(&po->tx_ring);
2740 status = TP_STATUS_WRONG_FORMAT;
2746 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2752 skb->destructor = tpacket_destruct_skb;
2753 __packet_set_status(po, ph, TP_STATUS_SENDING);
2754 packet_inc_pending(&po->tx_ring);
2756 status = TP_STATUS_SEND_REQUEST;
2757 err = po->xmit(skb);
2758 if (unlikely(err > 0)) {
2759 err = net_xmit_errno(err);
2760 if (err && __packet_get_status(po, ph) ==
2761 TP_STATUS_AVAILABLE) {
2762 /* skb was destructed already */
2767 * skb was dropped but not destructed yet;
2768 * let's treat it like congestion or err < 0
2772 packet_increment_head(&po->tx_ring);
2774 } while (likely((ph != NULL) ||
2775 /* Note: packet_read_pending() might be slow if we have
2776 * to call it as it's per_cpu variable, but in fast-path
2777 * we already short-circuit the loop with the first
2778 * condition, and luckily don't have to go that path
2781 (need_wait && packet_read_pending(&po->tx_ring))));
2787 __packet_set_status(po, ph, status);
2792 mutex_unlock(&po->pg_vec_lock);
2796 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2797 size_t reserve, size_t len,
2798 size_t linear, int noblock,
2801 struct sk_buff *skb;
2803 /* Under a page? Don't bother with paged skb. */
2804 if (prepad + len < PAGE_SIZE || !linear)
2807 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2812 skb_reserve(skb, reserve);
2813 skb_put(skb, linear);
2814 skb->data_len = len - linear;
2815 skb->len += len - linear;
2820 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2822 struct sock *sk = sock->sk;
2823 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2824 struct sk_buff *skb;
2825 struct net_device *dev;
2827 unsigned char *addr;
2828 int err, reserve = 0;
2829 struct sockcm_cookie sockc;
2830 struct virtio_net_hdr vnet_hdr = { 0 };
2832 struct packet_sock *po = pkt_sk(sk);
2833 int hlen, tlen, linear;
2837 * Get and verify the address.
2840 if (likely(saddr == NULL)) {
2841 dev = packet_cached_dev_get(po);
2846 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2848 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2850 proto = saddr->sll_protocol;
2851 addr = saddr->sll_addr;
2852 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2856 if (unlikely(dev == NULL))
2859 if (unlikely(!(dev->flags & IFF_UP)))
2862 sockc.tsflags = sk->sk_tsflags;
2863 sockc.mark = sk->sk_mark;
2864 if (msg->msg_controllen) {
2865 err = sock_cmsg_send(sk, msg, &sockc);
2870 if (sock->type == SOCK_RAW)
2871 reserve = dev->hard_header_len;
2872 if (po->has_vnet_hdr) {
2873 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2878 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2879 if (!netif_supports_nofcs(dev)) {
2880 err = -EPROTONOSUPPORT;
2883 extra_len = 4; /* We're doing our own CRC */
2887 if (!vnet_hdr.gso_type &&
2888 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2892 hlen = LL_RESERVED_SPACE(dev);
2893 tlen = dev->needed_tailroom;
2894 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2895 linear = max(linear, min_t(int, len, dev->hard_header_len));
2896 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2897 msg->msg_flags & MSG_DONTWAIT, &err);
2901 skb_set_network_header(skb, reserve);
2904 if (sock->type == SOCK_DGRAM) {
2905 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2906 if (unlikely(offset < 0))
2910 /* Returns -EFAULT on error */
2911 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2915 if (sock->type == SOCK_RAW &&
2916 !dev_validate_header(dev, skb->data, len)) {
2921 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2923 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2924 !packet_extra_vlan_len_allowed(dev, skb)) {
2929 skb->protocol = proto;
2931 skb->priority = sk->sk_priority;
2932 skb->mark = sockc.mark;
2934 if (po->has_vnet_hdr) {
2935 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2938 len += sizeof(vnet_hdr);
2941 skb_probe_transport_header(skb, reserve);
2943 if (unlikely(extra_len == 4))
2946 err = po->xmit(skb);
2947 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2963 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2965 struct sock *sk = sock->sk;
2966 struct packet_sock *po = pkt_sk(sk);
2968 if (po->tx_ring.pg_vec)
2969 return tpacket_snd(po, msg);
2971 return packet_snd(sock, msg, len);
2975 * Close a PACKET socket. This is fairly simple. We immediately go
2976 * to 'closed' state and remove our protocol entry in the device list.
2979 static int packet_release(struct socket *sock)
2981 struct sock *sk = sock->sk;
2982 struct packet_sock *po;
2983 struct packet_fanout *f;
2985 union tpacket_req_u req_u;
2993 mutex_lock(&net->packet.sklist_lock);
2994 sk_del_node_init_rcu(sk);
2995 mutex_unlock(&net->packet.sklist_lock);
2998 sock_prot_inuse_add(net, sk->sk_prot, -1);
3001 spin_lock(&po->bind_lock);
3002 unregister_prot_hook(sk, false);
3003 packet_cached_dev_reset(po);
3005 if (po->prot_hook.dev) {
3006 dev_put(po->prot_hook.dev);
3007 po->prot_hook.dev = NULL;
3009 spin_unlock(&po->bind_lock);
3011 packet_flush_mclist(sk);
3013 if (po->rx_ring.pg_vec) {
3014 memset(&req_u, 0, sizeof(req_u));
3015 packet_set_ring(sk, &req_u, 1, 0);
3018 if (po->tx_ring.pg_vec) {
3019 memset(&req_u, 0, sizeof(req_u));
3020 packet_set_ring(sk, &req_u, 1, 1);
3023 f = fanout_release(sk);
3028 fanout_release_data(f);
3032 * Now the socket is dead. No more input will appear.
3039 skb_queue_purge(&sk->sk_receive_queue);
3040 packet_free_pending(po);
3041 sk_refcnt_debug_release(sk);
3048 * Attach a packet hook.
3051 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3054 struct packet_sock *po = pkt_sk(sk);
3055 struct net_device *dev_curr;
3058 struct net_device *dev = NULL;
3060 bool unlisted = false;
3066 spin_lock(&po->bind_lock);
3070 dev = dev_get_by_name_rcu(sock_net(sk), name);
3075 } else if (ifindex) {
3076 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3086 proto_curr = po->prot_hook.type;
3087 dev_curr = po->prot_hook.dev;
3089 need_rehook = proto_curr != proto || dev_curr != dev;
3094 __unregister_prot_hook(sk, true);
3096 dev_curr = po->prot_hook.dev;
3098 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3103 po->prot_hook.type = proto;
3105 if (unlikely(unlisted)) {
3107 po->prot_hook.dev = NULL;
3109 packet_cached_dev_reset(po);
3111 po->prot_hook.dev = dev;
3112 po->ifindex = dev ? dev->ifindex : 0;
3113 packet_cached_dev_assign(po, dev);
3119 if (proto == 0 || !need_rehook)
3122 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3123 register_prot_hook(sk);
3125 sk->sk_err = ENETDOWN;
3126 if (!sock_flag(sk, SOCK_DEAD))
3127 sk->sk_error_report(sk);
3132 spin_unlock(&po->bind_lock);
3138 * Bind a packet socket to a device
3141 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3144 struct sock *sk = sock->sk;
3145 char name[sizeof(uaddr->sa_data) + 1];
3151 if (addr_len != sizeof(struct sockaddr))
3153 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3156 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3157 name[sizeof(uaddr->sa_data)] = 0;
3159 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3162 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3164 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3165 struct sock *sk = sock->sk;
3171 if (addr_len < sizeof(struct sockaddr_ll))
3173 if (sll->sll_family != AF_PACKET)
3176 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3177 sll->sll_protocol ? : pkt_sk(sk)->num);
3180 static struct proto packet_proto = {
3182 .owner = THIS_MODULE,
3183 .obj_size = sizeof(struct packet_sock),
3187 * Create a packet of type SOCK_PACKET.
3190 static int packet_create(struct net *net, struct socket *sock, int protocol,
3194 struct packet_sock *po;
3195 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3198 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3200 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3201 sock->type != SOCK_PACKET)
3202 return -ESOCKTNOSUPPORT;
3204 sock->state = SS_UNCONNECTED;
3207 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3211 sock->ops = &packet_ops;
3212 if (sock->type == SOCK_PACKET)
3213 sock->ops = &packet_ops_spkt;
3215 sock_init_data(sock, sk);
3218 sk->sk_family = PF_PACKET;
3220 po->xmit = dev_queue_xmit;
3222 err = packet_alloc_pending(po);
3226 packet_cached_dev_reset(po);
3228 sk->sk_destruct = packet_sock_destruct;
3229 sk_refcnt_debug_inc(sk);
3232 * Attach a protocol block
3235 spin_lock_init(&po->bind_lock);
3236 mutex_init(&po->pg_vec_lock);
3237 po->rollover = NULL;
3238 po->prot_hook.func = packet_rcv;
3240 if (sock->type == SOCK_PACKET)
3241 po->prot_hook.func = packet_rcv_spkt;
3243 po->prot_hook.af_packet_priv = sk;
3246 po->prot_hook.type = proto;
3247 register_prot_hook(sk);
3250 mutex_lock(&net->packet.sklist_lock);
3251 sk_add_node_rcu(sk, &net->packet.sklist);
3252 mutex_unlock(&net->packet.sklist_lock);
3255 sock_prot_inuse_add(net, &packet_proto, 1);
3266 * Pull a packet from our receive queue and hand it to the user.
3267 * If necessary we block.
3270 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3273 struct sock *sk = sock->sk;
3274 struct sk_buff *skb;
3276 int vnet_hdr_len = 0;
3277 unsigned int origlen = 0;
3280 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3284 /* What error should we return now? EUNATTACH? */
3285 if (pkt_sk(sk)->ifindex < 0)
3289 if (flags & MSG_ERRQUEUE) {
3290 err = sock_recv_errqueue(sk, msg, len,
3291 SOL_PACKET, PACKET_TX_TIMESTAMP);
3296 * Call the generic datagram receiver. This handles all sorts
3297 * of horrible races and re-entrancy so we can forget about it
3298 * in the protocol layers.
3300 * Now it will return ENETDOWN, if device have just gone down,
3301 * but then it will block.
3304 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3307 * An error occurred so return it. Because skb_recv_datagram()
3308 * handles the blocking we don't see and worry about blocking
3315 if (pkt_sk(sk)->pressure)
3316 packet_rcv_has_room(pkt_sk(sk), NULL);
3318 if (pkt_sk(sk)->has_vnet_hdr) {
3319 err = packet_rcv_vnet(msg, skb, &len);
3322 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3325 /* You lose any data beyond the buffer you gave. If it worries
3326 * a user program they can ask the device for its MTU
3332 msg->msg_flags |= MSG_TRUNC;
3335 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3339 if (sock->type != SOCK_PACKET) {
3340 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3342 /* Original length was stored in sockaddr_ll fields */
3343 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3344 sll->sll_family = AF_PACKET;
3345 sll->sll_protocol = skb->protocol;
3348 sock_recv_ts_and_drops(msg, sk, skb);
3350 if (msg->msg_name) {
3351 /* If the address length field is there to be filled
3352 * in, we fill it in now.
3354 if (sock->type == SOCK_PACKET) {
3355 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3356 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3358 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3360 msg->msg_namelen = sll->sll_halen +
3361 offsetof(struct sockaddr_ll, sll_addr);
3363 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3367 if (pkt_sk(sk)->auxdata) {
3368 struct tpacket_auxdata aux;
3370 aux.tp_status = TP_STATUS_USER;
3371 if (skb->ip_summed == CHECKSUM_PARTIAL)
3372 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3373 else if (skb->pkt_type != PACKET_OUTGOING &&
3374 (skb->ip_summed == CHECKSUM_COMPLETE ||
3375 skb_csum_unnecessary(skb)))
3376 aux.tp_status |= TP_STATUS_CSUM_VALID;
3378 aux.tp_len = origlen;
3379 aux.tp_snaplen = skb->len;
3381 aux.tp_net = skb_network_offset(skb);
3382 if (skb_vlan_tag_present(skb)) {
3383 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3384 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3385 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3387 aux.tp_vlan_tci = 0;
3388 aux.tp_vlan_tpid = 0;
3390 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3394 * Free or return the buffer as appropriate. Again this
3395 * hides all the races and re-entrancy issues from us.
3397 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3400 skb_free_datagram(sk, skb);
3405 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3406 int *uaddr_len, int peer)
3408 struct net_device *dev;
3409 struct sock *sk = sock->sk;
3414 uaddr->sa_family = AF_PACKET;
3415 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3417 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3419 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3421 *uaddr_len = sizeof(*uaddr);
3426 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3427 int *uaddr_len, int peer)
3429 struct net_device *dev;
3430 struct sock *sk = sock->sk;
3431 struct packet_sock *po = pkt_sk(sk);
3432 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3437 sll->sll_family = AF_PACKET;
3438 sll->sll_ifindex = po->ifindex;
3439 sll->sll_protocol = po->num;
3440 sll->sll_pkttype = 0;
3442 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3444 sll->sll_hatype = dev->type;
3445 sll->sll_halen = dev->addr_len;
3446 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3448 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3452 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3457 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3461 case PACKET_MR_MULTICAST:
3462 if (i->alen != dev->addr_len)
3465 return dev_mc_add(dev, i->addr);
3467 return dev_mc_del(dev, i->addr);
3469 case PACKET_MR_PROMISC:
3470 return dev_set_promiscuity(dev, what);
3471 case PACKET_MR_ALLMULTI:
3472 return dev_set_allmulti(dev, what);
3473 case PACKET_MR_UNICAST:
3474 if (i->alen != dev->addr_len)
3477 return dev_uc_add(dev, i->addr);
3479 return dev_uc_del(dev, i->addr);
3487 static void packet_dev_mclist_delete(struct net_device *dev,
3488 struct packet_mclist **mlp)
3490 struct packet_mclist *ml;
3492 while ((ml = *mlp) != NULL) {
3493 if (ml->ifindex == dev->ifindex) {
3494 packet_dev_mc(dev, ml, -1);
3502 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3504 struct packet_sock *po = pkt_sk(sk);
3505 struct packet_mclist *ml, *i;
3506 struct net_device *dev;
3512 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3517 if (mreq->mr_alen > dev->addr_len)
3521 i = kmalloc(sizeof(*i), GFP_KERNEL);
3526 for (ml = po->mclist; ml; ml = ml->next) {
3527 if (ml->ifindex == mreq->mr_ifindex &&
3528 ml->type == mreq->mr_type &&
3529 ml->alen == mreq->mr_alen &&
3530 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3532 /* Free the new element ... */
3538 i->type = mreq->mr_type;
3539 i->ifindex = mreq->mr_ifindex;
3540 i->alen = mreq->mr_alen;
3541 memcpy(i->addr, mreq->mr_address, i->alen);
3542 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3544 i->next = po->mclist;
3546 err = packet_dev_mc(dev, i, 1);
3548 po->mclist = i->next;
3557 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3559 struct packet_mclist *ml, **mlp;
3563 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3564 if (ml->ifindex == mreq->mr_ifindex &&
3565 ml->type == mreq->mr_type &&
3566 ml->alen == mreq->mr_alen &&
3567 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3568 if (--ml->count == 0) {
3569 struct net_device *dev;
3571 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3573 packet_dev_mc(dev, ml, -1);
3583 static void packet_flush_mclist(struct sock *sk)
3585 struct packet_sock *po = pkt_sk(sk);
3586 struct packet_mclist *ml;
3592 while ((ml = po->mclist) != NULL) {
3593 struct net_device *dev;
3595 po->mclist = ml->next;
3596 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3598 packet_dev_mc(dev, ml, -1);
3605 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3607 struct sock *sk = sock->sk;
3608 struct packet_sock *po = pkt_sk(sk);
3611 if (level != SOL_PACKET)
3612 return -ENOPROTOOPT;
3615 case PACKET_ADD_MEMBERSHIP:
3616 case PACKET_DROP_MEMBERSHIP:
3618 struct packet_mreq_max mreq;
3620 memset(&mreq, 0, sizeof(mreq));
3621 if (len < sizeof(struct packet_mreq))
3623 if (len > sizeof(mreq))
3625 if (copy_from_user(&mreq, optval, len))
3627 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3629 if (optname == PACKET_ADD_MEMBERSHIP)
3630 ret = packet_mc_add(sk, &mreq);
3632 ret = packet_mc_drop(sk, &mreq);
3636 case PACKET_RX_RING:
3637 case PACKET_TX_RING:
3639 union tpacket_req_u req_u;
3642 switch (po->tp_version) {
3645 len = sizeof(req_u.req);
3649 len = sizeof(req_u.req3);
3654 if (copy_from_user(&req_u.req, optval, len))
3656 return packet_set_ring(sk, &req_u, 0,
3657 optname == PACKET_TX_RING);
3659 case PACKET_COPY_THRESH:
3663 if (optlen != sizeof(val))
3665 if (copy_from_user(&val, optval, sizeof(val)))
3668 pkt_sk(sk)->copy_thresh = val;
3671 case PACKET_VERSION:
3675 if (optlen != sizeof(val))
3677 if (copy_from_user(&val, optval, sizeof(val)))
3688 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3691 po->tp_version = val;
3697 case PACKET_RESERVE:
3701 if (optlen != sizeof(val))
3703 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3705 if (copy_from_user(&val, optval, sizeof(val)))
3709 po->tp_reserve = val;
3716 if (optlen != sizeof(val))
3718 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3720 if (copy_from_user(&val, optval, sizeof(val)))
3722 po->tp_loss = !!val;
3725 case PACKET_AUXDATA:
3729 if (optlen < sizeof(val))
3731 if (copy_from_user(&val, optval, sizeof(val)))
3734 po->auxdata = !!val;
3737 case PACKET_ORIGDEV:
3741 if (optlen < sizeof(val))
3743 if (copy_from_user(&val, optval, sizeof(val)))
3746 po->origdev = !!val;
3749 case PACKET_VNET_HDR:
3753 if (sock->type != SOCK_RAW)
3755 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3757 if (optlen < sizeof(val))
3759 if (copy_from_user(&val, optval, sizeof(val)))
3762 po->has_vnet_hdr = !!val;
3765 case PACKET_TIMESTAMP:
3769 if (optlen != sizeof(val))
3771 if (copy_from_user(&val, optval, sizeof(val)))
3774 po->tp_tstamp = val;
3781 if (optlen != sizeof(val))
3783 if (copy_from_user(&val, optval, sizeof(val)))
3786 return fanout_add(sk, val & 0xffff, val >> 16);
3788 case PACKET_FANOUT_DATA:
3793 return fanout_set_data(po, optval, optlen);
3795 case PACKET_TX_HAS_OFF:
3799 if (optlen != sizeof(val))
3801 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3803 if (copy_from_user(&val, optval, sizeof(val)))
3805 po->tp_tx_has_off = !!val;
3808 case PACKET_QDISC_BYPASS:
3812 if (optlen != sizeof(val))
3814 if (copy_from_user(&val, optval, sizeof(val)))
3817 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3821 return -ENOPROTOOPT;
3825 static int packet_getsockopt(struct socket *sock, int level, int optname,
3826 char __user *optval, int __user *optlen)
3829 int val, lv = sizeof(val);
3830 struct sock *sk = sock->sk;
3831 struct packet_sock *po = pkt_sk(sk);
3833 union tpacket_stats_u st;
3834 struct tpacket_rollover_stats rstats;
3836 if (level != SOL_PACKET)
3837 return -ENOPROTOOPT;
3839 if (get_user(len, optlen))
3846 case PACKET_STATISTICS:
3847 spin_lock_bh(&sk->sk_receive_queue.lock);
3848 memcpy(&st, &po->stats, sizeof(st));
3849 memset(&po->stats, 0, sizeof(po->stats));
3850 spin_unlock_bh(&sk->sk_receive_queue.lock);
3852 if (po->tp_version == TPACKET_V3) {
3853 lv = sizeof(struct tpacket_stats_v3);
3854 st.stats3.tp_packets += st.stats3.tp_drops;
3857 lv = sizeof(struct tpacket_stats);
3858 st.stats1.tp_packets += st.stats1.tp_drops;
3863 case PACKET_AUXDATA:
3866 case PACKET_ORIGDEV:
3869 case PACKET_VNET_HDR:
3870 val = po->has_vnet_hdr;
3872 case PACKET_VERSION:
3873 val = po->tp_version;
3876 if (len > sizeof(int))
3878 if (len < sizeof(int))
3880 if (copy_from_user(&val, optval, len))
3884 val = sizeof(struct tpacket_hdr);
3887 val = sizeof(struct tpacket2_hdr);
3890 val = sizeof(struct tpacket3_hdr);
3896 case PACKET_RESERVE:
3897 val = po->tp_reserve;
3902 case PACKET_TIMESTAMP:
3903 val = po->tp_tstamp;
3907 ((u32)po->fanout->id |
3908 ((u32)po->fanout->type << 16) |
3909 ((u32)po->fanout->flags << 24)) :
3912 case PACKET_ROLLOVER_STATS:
3915 rstats.tp_all = atomic_long_read(&po->rollover->num);
3916 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3917 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3919 lv = sizeof(rstats);
3921 case PACKET_TX_HAS_OFF:
3922 val = po->tp_tx_has_off;
3924 case PACKET_QDISC_BYPASS:
3925 val = packet_use_direct_xmit(po);
3928 return -ENOPROTOOPT;
3933 if (put_user(len, optlen))
3935 if (copy_to_user(optval, data, len))
3941 #ifdef CONFIG_COMPAT
3942 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3943 char __user *optval, unsigned int optlen)
3945 struct packet_sock *po = pkt_sk(sock->sk);
3947 if (level != SOL_PACKET)
3948 return -ENOPROTOOPT;
3950 if (optname == PACKET_FANOUT_DATA &&
3951 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3952 optval = (char __user *)get_compat_bpf_fprog(optval);
3955 optlen = sizeof(struct sock_fprog);
3958 return packet_setsockopt(sock, level, optname, optval, optlen);
3962 static int packet_notifier(struct notifier_block *this,
3963 unsigned long msg, void *ptr)
3966 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3967 struct net *net = dev_net(dev);
3970 sk_for_each_rcu(sk, &net->packet.sklist) {
3971 struct packet_sock *po = pkt_sk(sk);
3974 case NETDEV_UNREGISTER:
3976 packet_dev_mclist_delete(dev, &po->mclist);
3980 if (dev->ifindex == po->ifindex) {
3981 spin_lock(&po->bind_lock);
3983 __unregister_prot_hook(sk, false);
3984 sk->sk_err = ENETDOWN;
3985 if (!sock_flag(sk, SOCK_DEAD))
3986 sk->sk_error_report(sk);
3988 if (msg == NETDEV_UNREGISTER) {
3989 packet_cached_dev_reset(po);
3991 if (po->prot_hook.dev)
3992 dev_put(po->prot_hook.dev);
3993 po->prot_hook.dev = NULL;
3995 spin_unlock(&po->bind_lock);
3999 if (dev->ifindex == po->ifindex) {
4000 spin_lock(&po->bind_lock);
4002 register_prot_hook(sk);
4003 spin_unlock(&po->bind_lock);
4013 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4016 struct sock *sk = sock->sk;
4021 int amount = sk_wmem_alloc_get(sk);
4023 return put_user(amount, (int __user *)arg);
4027 struct sk_buff *skb;
4030 spin_lock_bh(&sk->sk_receive_queue.lock);
4031 skb = skb_peek(&sk->sk_receive_queue);
4034 spin_unlock_bh(&sk->sk_receive_queue.lock);
4035 return put_user(amount, (int __user *)arg);
4038 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4040 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4050 case SIOCGIFBRDADDR:
4051 case SIOCSIFBRDADDR:
4052 case SIOCGIFNETMASK:
4053 case SIOCSIFNETMASK:
4054 case SIOCGIFDSTADDR:
4055 case SIOCSIFDSTADDR:
4057 return inet_dgram_ops.ioctl(sock, cmd, arg);
4061 return -ENOIOCTLCMD;
4066 static unsigned int packet_poll(struct file *file, struct socket *sock,
4069 struct sock *sk = sock->sk;
4070 struct packet_sock *po = pkt_sk(sk);
4071 unsigned int mask = datagram_poll(file, sock, wait);
4073 spin_lock_bh(&sk->sk_receive_queue.lock);
4074 if (po->rx_ring.pg_vec) {
4075 if (!packet_previous_rx_frame(po, &po->rx_ring,
4077 mask |= POLLIN | POLLRDNORM;
4079 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4081 spin_unlock_bh(&sk->sk_receive_queue.lock);
4082 spin_lock_bh(&sk->sk_write_queue.lock);
4083 if (po->tx_ring.pg_vec) {
4084 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4085 mask |= POLLOUT | POLLWRNORM;
4087 spin_unlock_bh(&sk->sk_write_queue.lock);
4092 /* Dirty? Well, I still did not learn better way to account
4096 static void packet_mm_open(struct vm_area_struct *vma)
4098 struct file *file = vma->vm_file;
4099 struct socket *sock = file->private_data;
4100 struct sock *sk = sock->sk;
4103 atomic_inc(&pkt_sk(sk)->mapped);
4106 static void packet_mm_close(struct vm_area_struct *vma)
4108 struct file *file = vma->vm_file;
4109 struct socket *sock = file->private_data;
4110 struct sock *sk = sock->sk;
4113 atomic_dec(&pkt_sk(sk)->mapped);
4116 static const struct vm_operations_struct packet_mmap_ops = {
4117 .open = packet_mm_open,
4118 .close = packet_mm_close,
4121 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4126 for (i = 0; i < len; i++) {
4127 if (likely(pg_vec[i].buffer)) {
4128 if (is_vmalloc_addr(pg_vec[i].buffer))
4129 vfree(pg_vec[i].buffer);
4131 free_pages((unsigned long)pg_vec[i].buffer,
4133 pg_vec[i].buffer = NULL;
4139 static char *alloc_one_pg_vec_page(unsigned long order)
4142 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4143 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4145 buffer = (char *) __get_free_pages(gfp_flags, order);
4149 /* __get_free_pages failed, fall back to vmalloc */
4150 buffer = vzalloc((1 << order) * PAGE_SIZE);
4154 /* vmalloc failed, lets dig into swap here */
4155 gfp_flags &= ~__GFP_NORETRY;
4156 buffer = (char *) __get_free_pages(gfp_flags, order);
4160 /* complete and utter failure */
4164 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4166 unsigned int block_nr = req->tp_block_nr;
4170 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4171 if (unlikely(!pg_vec))
4174 for (i = 0; i < block_nr; i++) {
4175 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4176 if (unlikely(!pg_vec[i].buffer))
4177 goto out_free_pgvec;
4184 free_pg_vec(pg_vec, order, block_nr);
4189 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4190 int closing, int tx_ring)
4192 struct pgv *pg_vec = NULL;
4193 struct packet_sock *po = pkt_sk(sk);
4194 int was_running, order = 0;
4195 struct packet_ring_buffer *rb;
4196 struct sk_buff_head *rb_queue;
4199 /* Added to avoid minimal code churn */
4200 struct tpacket_req *req = &req_u->req;
4204 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4205 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4209 if (atomic_read(&po->mapped))
4211 if (packet_read_pending(rb))
4215 if (req->tp_block_nr) {
4216 /* Sanity tests and some calculations */
4218 if (unlikely(rb->pg_vec))
4221 switch (po->tp_version) {
4223 po->tp_hdrlen = TPACKET_HDRLEN;
4226 po->tp_hdrlen = TPACKET2_HDRLEN;
4229 po->tp_hdrlen = TPACKET3_HDRLEN;
4234 if (unlikely((int)req->tp_block_size <= 0))
4236 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4238 if (po->tp_version >= TPACKET_V3 &&
4239 req->tp_block_size <=
4240 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv))
4242 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4245 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4248 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4249 if (unlikely(rb->frames_per_block == 0))
4251 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4253 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4258 order = get_order(req->tp_block_size);
4259 pg_vec = alloc_pg_vec(req, order);
4260 if (unlikely(!pg_vec))
4262 switch (po->tp_version) {
4264 /* Block transmit is not supported yet */
4266 init_prb_bdqc(po, rb, pg_vec, req_u);
4268 struct tpacket_req3 *req3 = &req_u->req3;
4270 if (req3->tp_retire_blk_tov ||
4271 req3->tp_sizeof_priv ||
4272 req3->tp_feature_req_word) {
4285 if (unlikely(req->tp_frame_nr))
4290 /* Detach socket from network */
4291 spin_lock(&po->bind_lock);
4292 was_running = po->running;
4296 __unregister_prot_hook(sk, false);
4298 spin_unlock(&po->bind_lock);
4303 mutex_lock(&po->pg_vec_lock);
4304 if (closing || atomic_read(&po->mapped) == 0) {
4306 spin_lock_bh(&rb_queue->lock);
4307 swap(rb->pg_vec, pg_vec);
4308 rb->frame_max = (req->tp_frame_nr - 1);
4310 rb->frame_size = req->tp_frame_size;
4311 spin_unlock_bh(&rb_queue->lock);
4313 swap(rb->pg_vec_order, order);
4314 swap(rb->pg_vec_len, req->tp_block_nr);
4316 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4317 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4318 tpacket_rcv : packet_rcv;
4319 skb_queue_purge(rb_queue);
4320 if (atomic_read(&po->mapped))
4321 pr_err("packet_mmap: vma is busy: %d\n",
4322 atomic_read(&po->mapped));
4324 mutex_unlock(&po->pg_vec_lock);
4326 spin_lock(&po->bind_lock);
4329 register_prot_hook(sk);
4331 spin_unlock(&po->bind_lock);
4332 if (closing && (po->tp_version > TPACKET_V2)) {
4333 /* Because we don't support block-based V3 on tx-ring */
4335 prb_shutdown_retire_blk_timer(po, rb_queue);
4339 free_pg_vec(pg_vec, order, req->tp_block_nr);
4345 static int packet_mmap(struct file *file, struct socket *sock,
4346 struct vm_area_struct *vma)
4348 struct sock *sk = sock->sk;
4349 struct packet_sock *po = pkt_sk(sk);
4350 unsigned long size, expected_size;
4351 struct packet_ring_buffer *rb;
4352 unsigned long start;
4359 mutex_lock(&po->pg_vec_lock);
4362 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4364 expected_size += rb->pg_vec_len
4370 if (expected_size == 0)
4373 size = vma->vm_end - vma->vm_start;
4374 if (size != expected_size)
4377 start = vma->vm_start;
4378 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4379 if (rb->pg_vec == NULL)
4382 for (i = 0; i < rb->pg_vec_len; i++) {
4384 void *kaddr = rb->pg_vec[i].buffer;
4387 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4388 page = pgv_to_page(kaddr);
4389 err = vm_insert_page(vma, start, page);
4398 atomic_inc(&po->mapped);
4399 vma->vm_ops = &packet_mmap_ops;
4403 mutex_unlock(&po->pg_vec_lock);
4407 static const struct proto_ops packet_ops_spkt = {
4408 .family = PF_PACKET,
4409 .owner = THIS_MODULE,
4410 .release = packet_release,
4411 .bind = packet_bind_spkt,
4412 .connect = sock_no_connect,
4413 .socketpair = sock_no_socketpair,
4414 .accept = sock_no_accept,
4415 .getname = packet_getname_spkt,
4416 .poll = datagram_poll,
4417 .ioctl = packet_ioctl,
4418 .listen = sock_no_listen,
4419 .shutdown = sock_no_shutdown,
4420 .setsockopt = sock_no_setsockopt,
4421 .getsockopt = sock_no_getsockopt,
4422 .sendmsg = packet_sendmsg_spkt,
4423 .recvmsg = packet_recvmsg,
4424 .mmap = sock_no_mmap,
4425 .sendpage = sock_no_sendpage,
4428 static const struct proto_ops packet_ops = {
4429 .family = PF_PACKET,
4430 .owner = THIS_MODULE,
4431 .release = packet_release,
4432 .bind = packet_bind,
4433 .connect = sock_no_connect,
4434 .socketpair = sock_no_socketpair,
4435 .accept = sock_no_accept,
4436 .getname = packet_getname,
4437 .poll = packet_poll,
4438 .ioctl = packet_ioctl,
4439 .listen = sock_no_listen,
4440 .shutdown = sock_no_shutdown,
4441 .setsockopt = packet_setsockopt,
4442 .getsockopt = packet_getsockopt,
4443 #ifdef CONFIG_COMPAT
4444 .compat_setsockopt = compat_packet_setsockopt,
4446 .sendmsg = packet_sendmsg,
4447 .recvmsg = packet_recvmsg,
4448 .mmap = packet_mmap,
4449 .sendpage = sock_no_sendpage,
4452 static const struct net_proto_family packet_family_ops = {
4453 .family = PF_PACKET,
4454 .create = packet_create,
4455 .owner = THIS_MODULE,
4458 static struct notifier_block packet_netdev_notifier = {
4459 .notifier_call = packet_notifier,
4462 #ifdef CONFIG_PROC_FS
4464 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4467 struct net *net = seq_file_net(seq);
4470 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4473 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4475 struct net *net = seq_file_net(seq);
4476 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4479 static void packet_seq_stop(struct seq_file *seq, void *v)
4485 static int packet_seq_show(struct seq_file *seq, void *v)
4487 if (v == SEQ_START_TOKEN)
4488 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4490 struct sock *s = sk_entry(v);
4491 const struct packet_sock *po = pkt_sk(s);
4494 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4496 refcount_read(&s->sk_refcnt),
4501 atomic_read(&s->sk_rmem_alloc),
4502 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4509 static const struct seq_operations packet_seq_ops = {
4510 .start = packet_seq_start,
4511 .next = packet_seq_next,
4512 .stop = packet_seq_stop,
4513 .show = packet_seq_show,
4516 static int packet_seq_open(struct inode *inode, struct file *file)
4518 return seq_open_net(inode, file, &packet_seq_ops,
4519 sizeof(struct seq_net_private));
4522 static const struct file_operations packet_seq_fops = {
4523 .owner = THIS_MODULE,
4524 .open = packet_seq_open,
4526 .llseek = seq_lseek,
4527 .release = seq_release_net,
4532 static int __net_init packet_net_init(struct net *net)
4534 mutex_init(&net->packet.sklist_lock);
4535 INIT_HLIST_HEAD(&net->packet.sklist);
4537 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4543 static void __net_exit packet_net_exit(struct net *net)
4545 remove_proc_entry("packet", net->proc_net);
4548 static struct pernet_operations packet_net_ops = {
4549 .init = packet_net_init,
4550 .exit = packet_net_exit,
4554 static void __exit packet_exit(void)
4556 unregister_netdevice_notifier(&packet_netdev_notifier);
4557 unregister_pernet_subsys(&packet_net_ops);
4558 sock_unregister(PF_PACKET);
4559 proto_unregister(&packet_proto);
4562 static int __init packet_init(void)
4564 int rc = proto_register(&packet_proto, 0);
4569 sock_register(&packet_family_ops);
4570 register_pernet_subsys(&packet_net_ops);
4571 register_netdevice_notifier(&packet_netdev_notifier);
4576 module_init(packet_init);
4577 module_exit(packet_exit);
4578 MODULE_LICENSE("GPL");
4579 MODULE_ALIAS_NETPROTO(PF_PACKET);