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 <asm/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>
101 - if device has no dev->hard_header routine, it adds and removes ll header
102 inside itself. In this case ll header is invisible outside of device,
103 but higher levels still should reserve dev->hard_header_len.
104 Some devices are enough clever to reallocate skb, when header
105 will not fit to reserved space (tunnel), another ones are silly
107 - packet socket receives packets with pulled ll header,
108 so that SOCK_RAW should push it back.
113 Incoming, dev->hard_header!=NULL
114 mac_header -> ll header
117 Outgoing, dev->hard_header!=NULL
118 mac_header -> ll header
121 Incoming, dev->hard_header==NULL
122 mac_header -> UNKNOWN position. It is very likely, that it points to ll
123 header. PPP makes it, that is wrong, because introduce
124 assymetry between rx and tx paths.
127 Outgoing, dev->hard_header==NULL
128 mac_header -> data. ll header is still not built!
132 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
138 dev->hard_header != NULL
139 mac_header -> ll header
142 dev->hard_header == NULL (ll header is added by device, we cannot control it)
146 We should set nh.raw on output to correct posistion,
147 packet classifier depends on it.
150 /* Private packet socket structures. */
152 /* identical to struct packet_mreq except it has
153 * a longer address field.
155 struct packet_mreq_max {
157 unsigned short mr_type;
158 unsigned short mr_alen;
159 unsigned char mr_address[MAX_ADDR_LEN];
163 struct tpacket_hdr *h1;
164 struct tpacket2_hdr *h2;
165 struct tpacket3_hdr *h3;
169 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
170 int closing, int tx_ring);
172 #define V3_ALIGNMENT (8)
174 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
176 #define BLK_PLUS_PRIV(sz_of_priv) \
177 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
179 #define PGV_FROM_VMALLOC 1
181 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
182 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
183 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
184 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
185 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
186 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
187 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
190 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
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_kbdq_core *,
199 struct tpacket_block_desc *);
200 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
201 struct packet_sock *);
202 static void prb_retire_current_block(struct tpacket_kbdq_core *,
203 struct packet_sock *, unsigned int status);
204 static int prb_queue_frozen(struct tpacket_kbdq_core *);
205 static void prb_open_block(struct tpacket_kbdq_core *,
206 struct tpacket_block_desc *);
207 static void prb_retire_rx_blk_timer_expired(unsigned long);
208 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
209 static void prb_init_blk_timer(struct packet_sock *,
210 struct tpacket_kbdq_core *,
211 void (*func) (unsigned long));
212 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
213 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
214 struct tpacket3_hdr *);
215 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
216 struct tpacket3_hdr *);
217 static void packet_flush_mclist(struct sock *sk);
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 netdev_features_t features;
253 struct netdev_queue *txq;
254 int ret = NETDEV_TX_BUSY;
256 if (unlikely(!netif_running(dev) ||
257 !netif_carrier_ok(dev)))
260 features = netif_skb_features(skb);
261 if (skb_needs_linearize(skb, features) &&
262 __skb_linearize(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));
413 WARN(1, "TPACKET version not supported.\n");
420 static int __packet_get_status(struct packet_sock *po, void *frame)
422 union tpacket_uhdr h;
427 switch (po->tp_version) {
429 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
430 return h.h1->tp_status;
432 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
433 return h.h2->tp_status;
436 WARN(1, "TPACKET version not supported.\n");
442 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
445 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
448 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450 return TP_STATUS_TS_RAW_HARDWARE;
452 if (ktime_to_timespec_cond(skb->tstamp, ts))
453 return TP_STATUS_TS_SOFTWARE;
458 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
461 union tpacket_uhdr h;
465 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
469 switch (po->tp_version) {
471 h.h1->tp_sec = ts.tv_sec;
472 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
475 h.h2->tp_sec = ts.tv_sec;
476 h.h2->tp_nsec = ts.tv_nsec;
480 WARN(1, "TPACKET version not supported.\n");
484 /* one flush is safe, as both fields always lie on the same cacheline */
485 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
491 static void *packet_lookup_frame(struct packet_sock *po,
492 struct packet_ring_buffer *rb,
493 unsigned int position,
496 unsigned int pg_vec_pos, frame_offset;
497 union tpacket_uhdr h;
499 pg_vec_pos = position / rb->frames_per_block;
500 frame_offset = position % rb->frames_per_block;
502 h.raw = rb->pg_vec[pg_vec_pos].buffer +
503 (frame_offset * rb->frame_size);
505 if (status != __packet_get_status(po, h.raw))
511 static void *packet_current_frame(struct packet_sock *po,
512 struct packet_ring_buffer *rb,
515 return packet_lookup_frame(po, rb, rb->head, status);
518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
520 del_timer_sync(&pkc->retire_blk_timer);
523 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
524 struct sk_buff_head *rb_queue)
526 struct tpacket_kbdq_core *pkc;
528 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
530 spin_lock_bh(&rb_queue->lock);
531 pkc->delete_blk_timer = 1;
532 spin_unlock_bh(&rb_queue->lock);
534 prb_del_retire_blk_timer(pkc);
537 static void prb_init_blk_timer(struct packet_sock *po,
538 struct tpacket_kbdq_core *pkc,
539 void (*func) (unsigned long))
541 init_timer(&pkc->retire_blk_timer);
542 pkc->retire_blk_timer.data = (long)po;
543 pkc->retire_blk_timer.function = func;
544 pkc->retire_blk_timer.expires = jiffies;
547 static void prb_setup_retire_blk_timer(struct packet_sock *po)
549 struct tpacket_kbdq_core *pkc;
551 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
555 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
556 int blk_size_in_bytes)
558 struct net_device *dev;
559 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
560 struct ethtool_link_ksettings ecmd;
564 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
565 if (unlikely(!dev)) {
567 return DEFAULT_PRB_RETIRE_TOV;
569 err = __ethtool_get_link_ksettings(dev, &ecmd);
573 * If the link speed is so slow you don't really
574 * need to worry about perf anyways
576 if (ecmd.base.speed < SPEED_1000 ||
577 ecmd.base.speed == SPEED_UNKNOWN) {
578 return DEFAULT_PRB_RETIRE_TOV;
581 div = ecmd.base.speed / 1000;
585 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
597 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
598 union tpacket_req_u *req_u)
600 p1->feature_req_word = req_u->req3.tp_feature_req_word;
603 static void init_prb_bdqc(struct packet_sock *po,
604 struct packet_ring_buffer *rb,
606 union tpacket_req_u *req_u)
608 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
609 struct tpacket_block_desc *pbd;
611 memset(p1, 0x0, sizeof(*p1));
613 p1->knxt_seq_num = 1;
615 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
616 p1->pkblk_start = pg_vec[0].buffer;
617 p1->kblk_size = req_u->req3.tp_block_size;
618 p1->knum_blocks = req_u->req3.tp_block_nr;
619 p1->hdrlen = po->tp_hdrlen;
620 p1->version = po->tp_version;
621 p1->last_kactive_blk_num = 0;
622 po->stats.stats3.tp_freeze_q_cnt = 0;
623 if (req_u->req3.tp_retire_blk_tov)
624 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
626 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
627 req_u->req3.tp_block_size);
628 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
629 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
631 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
632 prb_init_ft_ops(p1, req_u);
633 prb_setup_retire_blk_timer(po);
634 prb_open_block(p1, pbd);
637 /* Do NOT update the last_blk_num first.
638 * Assumes sk_buff_head lock is held.
640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
642 mod_timer(&pkc->retire_blk_timer,
643 jiffies + pkc->tov_in_jiffies);
644 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
649 * 1) We refresh the timer only when we open a block.
650 * By doing this we don't waste cycles refreshing the timer
651 * on packet-by-packet basis.
653 * With a 1MB block-size, on a 1Gbps line, it will take
654 * i) ~8 ms to fill a block + ii) memcpy etc.
655 * In this cut we are not accounting for the memcpy time.
657 * So, if the user sets the 'tmo' to 10ms then the timer
658 * will never fire while the block is still getting filled
659 * (which is what we want). However, the user could choose
660 * to close a block early and that's fine.
662 * But when the timer does fire, we check whether or not to refresh it.
663 * Since the tmo granularity is in msecs, it is not too expensive
664 * to refresh the timer, lets say every '8' msecs.
665 * Either the user can set the 'tmo' or we can derive it based on
666 * a) line-speed and b) block-size.
667 * prb_calc_retire_blk_tmo() calculates the tmo.
670 static void prb_retire_rx_blk_timer_expired(unsigned long data)
672 struct packet_sock *po = (struct packet_sock *)data;
673 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
675 struct tpacket_block_desc *pbd;
677 spin_lock(&po->sk.sk_receive_queue.lock);
679 frozen = prb_queue_frozen(pkc);
680 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
682 if (unlikely(pkc->delete_blk_timer))
685 /* We only need to plug the race when the block is partially filled.
687 * lock(); increment BLOCK_NUM_PKTS; unlock()
688 * copy_bits() is in progress ...
689 * timer fires on other cpu:
690 * we can't retire the current block because copy_bits
694 if (BLOCK_NUM_PKTS(pbd)) {
695 while (atomic_read(&pkc->blk_fill_in_prog)) {
696 /* Waiting for skb_copy_bits to finish... */
701 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
703 if (!BLOCK_NUM_PKTS(pbd)) {
704 /* An empty block. Just refresh the timer. */
707 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
708 if (!prb_dispatch_next_block(pkc, po))
713 /* Case 1. Queue was frozen because user-space was
716 if (prb_curr_blk_in_use(pkc, pbd)) {
718 * Ok, user-space is still behind.
719 * So just refresh the timer.
723 /* Case 2. queue was frozen,user-space caught up,
724 * now the link went idle && the timer fired.
725 * We don't have a block to close.So we open this
726 * block and restart the timer.
727 * opening a block thaws the queue,restarts timer
728 * Thawing/timer-refresh is a side effect.
730 prb_open_block(pkc, pbd);
737 _prb_refresh_rx_retire_blk_timer(pkc);
740 spin_unlock(&po->sk.sk_receive_queue.lock);
743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
744 struct tpacket_block_desc *pbd1, __u32 status)
746 /* Flush everything minus the block header */
748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
753 /* Skip the block header(we know header WILL fit in 4K) */
756 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
757 for (; start < end; start += PAGE_SIZE)
758 flush_dcache_page(pgv_to_page(start));
763 /* Now update the block status. */
765 BLOCK_STATUS(pbd1) = status;
767 /* Flush the block header */
769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
771 flush_dcache_page(pgv_to_page(start));
781 * 2) Increment active_blk_num
783 * Note:We DONT refresh the timer on purpose.
784 * Because almost always the next block will be opened.
786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
787 struct tpacket_block_desc *pbd1,
788 struct packet_sock *po, unsigned int stat)
790 __u32 status = TP_STATUS_USER | stat;
792 struct tpacket3_hdr *last_pkt;
793 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
794 struct sock *sk = &po->sk;
796 if (po->stats.stats3.tp_drops)
797 status |= TP_STATUS_LOSING;
799 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
800 last_pkt->tp_next_offset = 0;
802 /* Get the ts of the last pkt */
803 if (BLOCK_NUM_PKTS(pbd1)) {
804 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
805 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
807 /* Ok, we tmo'd - so get the current time.
809 * It shouldn't really happen as we don't close empty
810 * blocks. See prb_retire_rx_blk_timer_expired().
814 h1->ts_last_pkt.ts_sec = ts.tv_sec;
815 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
820 /* Flush the block */
821 prb_flush_block(pkc1, pbd1, status);
823 sk->sk_data_ready(sk);
825 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
828 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
830 pkc->reset_pending_on_curr_blk = 0;
834 * Side effect of opening a block:
836 * 1) prb_queue is thawed.
837 * 2) retire_blk_timer is refreshed.
840 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
841 struct tpacket_block_desc *pbd1)
844 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
848 /* We could have just memset this but we will lose the
849 * flexibility of making the priv area sticky
852 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
853 BLOCK_NUM_PKTS(pbd1) = 0;
854 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
858 h1->ts_first_pkt.ts_sec = ts.tv_sec;
859 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
861 pkc1->pkblk_start = (char *)pbd1;
862 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
867 pbd1->version = pkc1->version;
868 pkc1->prev = pkc1->nxt_offset;
869 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
871 prb_thaw_queue(pkc1);
872 _prb_refresh_rx_retire_blk_timer(pkc1);
878 * Queue freeze logic:
879 * 1) Assume tp_block_nr = 8 blocks.
880 * 2) At time 't0', user opens Rx ring.
881 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882 * 4) user-space is either sleeping or processing block '0'.
883 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884 * it will close block-7,loop around and try to fill block '0'.
886 * __packet_lookup_frame_in_block
887 * prb_retire_current_block()
888 * prb_dispatch_next_block()
889 * |->(BLOCK_STATUS == USER) evaluates to true
890 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
891 * 6) Now there are two cases:
892 * 6.1) Link goes idle right after the queue is frozen.
893 * But remember, the last open_block() refreshed the timer.
894 * When this timer expires,it will refresh itself so that we can
895 * re-open block-0 in near future.
896 * 6.2) Link is busy and keeps on receiving packets. This is a simple
897 * case and __packet_lookup_frame_in_block will check if block-0
898 * is free and can now be re-used.
900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
901 struct packet_sock *po)
903 pkc->reset_pending_on_curr_blk = 1;
904 po->stats.stats3.tp_freeze_q_cnt++;
907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
910 * If the next block is free then we will dispatch it
911 * and return a good offset.
912 * Else, we will freeze the queue.
913 * So, caller must check the return value.
915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
916 struct packet_sock *po)
918 struct tpacket_block_desc *pbd;
922 /* 1. Get current block num */
923 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
925 /* 2. If this block is currently in_use then freeze the queue */
926 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
927 prb_freeze_queue(pkc, po);
933 * open this block and return the offset where the first packet
934 * needs to get stored.
936 prb_open_block(pkc, pbd);
937 return (void *)pkc->nxt_offset;
940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
941 struct packet_sock *po, unsigned int status)
943 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
945 /* retire/close the current block */
946 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
948 * Plug the case where copy_bits() is in progress on
949 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950 * have space to copy the pkt in the current block and
951 * called prb_retire_current_block()
953 * We don't need to worry about the TMO case because
954 * the timer-handler already handled this case.
956 if (!(status & TP_STATUS_BLK_TMO)) {
957 while (atomic_read(&pkc->blk_fill_in_prog)) {
958 /* Waiting for skb_copy_bits to finish... */
962 prb_close_block(pkc, pbd, po, status);
967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
968 struct tpacket_block_desc *pbd)
970 return TP_STATUS_USER & BLOCK_STATUS(pbd);
973 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
975 return pkc->reset_pending_on_curr_blk;
978 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
980 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
981 atomic_dec(&pkc->blk_fill_in_prog);
984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
985 struct tpacket3_hdr *ppd)
987 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
991 struct tpacket3_hdr *ppd)
993 ppd->hv1.tp_rxhash = 0;
996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
997 struct tpacket3_hdr *ppd)
999 if (skb_vlan_tag_present(pkc->skb)) {
1000 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1004 ppd->hv1.tp_vlan_tci = 0;
1005 ppd->hv1.tp_vlan_tpid = 0;
1006 ppd->tp_status = TP_STATUS_AVAILABLE;
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011 struct tpacket3_hdr *ppd)
1013 ppd->hv1.tp_padding = 0;
1014 prb_fill_vlan_info(pkc, ppd);
1016 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017 prb_fill_rxhash(pkc, ppd);
1019 prb_clear_rxhash(pkc, ppd);
1022 static void prb_fill_curr_block(char *curr,
1023 struct tpacket_kbdq_core *pkc,
1024 struct tpacket_block_desc *pbd,
1027 struct tpacket3_hdr *ppd;
1029 ppd = (struct tpacket3_hdr *)curr;
1030 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1032 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 BLOCK_NUM_PKTS(pbd) += 1;
1035 atomic_inc(&pkc->blk_fill_in_prog);
1036 prb_run_all_ft_ops(pkc, ppd);
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041 struct sk_buff *skb,
1046 struct tpacket_kbdq_core *pkc;
1047 struct tpacket_block_desc *pbd;
1050 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1053 /* Queue is frozen when user space is lagging behind */
1054 if (prb_queue_frozen(pkc)) {
1056 * Check if that last block which caused the queue to freeze,
1057 * is still in_use by user-space.
1059 if (prb_curr_blk_in_use(pkc, pbd)) {
1060 /* Can't record this packet */
1064 * Ok, the block was released by user-space.
1065 * Now let's open that block.
1066 * opening a block also thaws the queue.
1067 * Thawing is a side effect.
1069 prb_open_block(pkc, pbd);
1074 curr = pkc->nxt_offset;
1076 end = (char *)pbd + pkc->kblk_size;
1078 /* first try the current block */
1079 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080 prb_fill_curr_block(curr, pkc, pbd, len);
1081 return (void *)curr;
1084 /* Ok, close the current block */
1085 prb_retire_current_block(pkc, po, 0);
1087 /* Now, try to dispatch the next block */
1088 curr = (char *)prb_dispatch_next_block(pkc, po);
1090 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091 prb_fill_curr_block(curr, pkc, pbd, len);
1092 return (void *)curr;
1096 * No free blocks are available.user_space hasn't caught up yet.
1097 * Queue was just frozen and now this packet will get dropped.
1102 static void *packet_current_rx_frame(struct packet_sock *po,
1103 struct sk_buff *skb,
1104 int status, unsigned int len)
1107 switch (po->tp_version) {
1110 curr = packet_lookup_frame(po, &po->rx_ring,
1111 po->rx_ring.head, status);
1114 return __packet_lookup_frame_in_block(po, skb, status, len);
1116 WARN(1, "TPACKET version not supported\n");
1122 static void *prb_lookup_block(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1127 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1128 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1130 if (status != BLOCK_STATUS(pbd))
1135 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1138 if (rb->prb_bdqc.kactive_blk_num)
1139 prev = rb->prb_bdqc.kactive_blk_num-1;
1141 prev = rb->prb_bdqc.knum_blocks-1;
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1150 unsigned int previous = prb_previous_blk_num(rb);
1151 return prb_lookup_block(po, rb, previous, status);
1154 static void *packet_previous_rx_frame(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1158 if (po->tp_version <= TPACKET_V2)
1159 return packet_previous_frame(po, rb, status);
1161 return __prb_previous_block(po, rb, status);
1164 static void packet_increment_rx_head(struct packet_sock *po,
1165 struct packet_ring_buffer *rb)
1167 switch (po->tp_version) {
1170 return packet_increment_head(rb);
1173 WARN(1, "TPACKET version not supported.\n");
1179 static void *packet_previous_frame(struct packet_sock *po,
1180 struct packet_ring_buffer *rb,
1183 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184 return packet_lookup_frame(po, rb, previous, status);
1187 static void packet_increment_head(struct packet_ring_buffer *buff)
1189 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1192 static void packet_inc_pending(struct packet_ring_buffer *rb)
1194 this_cpu_inc(*rb->pending_refcnt);
1197 static void packet_dec_pending(struct packet_ring_buffer *rb)
1199 this_cpu_dec(*rb->pending_refcnt);
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1204 unsigned int refcnt = 0;
1207 /* We don't use pending refcount in rx_ring. */
1208 if (rb->pending_refcnt == NULL)
1211 for_each_possible_cpu(cpu)
1212 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1217 static int packet_alloc_pending(struct packet_sock *po)
1219 po->rx_ring.pending_refcnt = NULL;
1221 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1228 static void packet_free_pending(struct packet_sock *po)
1230 free_percpu(po->tx_ring.pending_refcnt);
1233 #define ROOM_POW_OFF 2
1234 #define ROOM_NONE 0x0
1235 #define ROOM_LOW 0x1
1236 #define ROOM_NORMAL 0x2
1238 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1242 len = po->rx_ring.frame_max + 1;
1243 idx = po->rx_ring.head;
1245 idx += len >> pow_off;
1248 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1251 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1255 len = po->rx_ring.prb_bdqc.knum_blocks;
1256 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1258 idx += len >> pow_off;
1261 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1264 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266 struct sock *sk = &po->sk;
1267 int ret = ROOM_NONE;
1269 if (po->prot_hook.func != tpacket_rcv) {
1270 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271 - (skb ? skb->truesize : 0);
1272 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1280 if (po->tp_version == TPACKET_V3) {
1281 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1283 else if (__tpacket_v3_has_room(po, 0))
1286 if (__tpacket_has_room(po, ROOM_POW_OFF))
1288 else if (__tpacket_has_room(po, 0))
1295 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1300 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301 ret = __packet_rcv_has_room(po, skb);
1302 has_room = ret == ROOM_NORMAL;
1303 if (po->pressure == has_room)
1304 po->pressure = !has_room;
1305 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1310 static void packet_sock_destruct(struct sock *sk)
1312 skb_queue_purge(&sk->sk_error_queue);
1314 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1317 if (!sock_flag(sk, SOCK_DEAD)) {
1318 pr_err("Attempt to release alive packet socket: %p\n", sk);
1322 sk_refcnt_debug_dec(sk);
1325 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1330 rxhash = skb_get_hash(skb);
1331 for (i = 0; i < ROLLOVER_HLEN; i++)
1332 if (po->rollover->history[i] == rxhash)
1335 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336 return count > (ROLLOVER_HLEN >> 1);
1339 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340 struct sk_buff *skb,
1343 return reciprocal_scale(skb_get_hash(skb), num);
1346 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347 struct sk_buff *skb,
1350 unsigned int val = atomic_inc_return(&f->rr_cur);
1355 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356 struct sk_buff *skb,
1359 return smp_processor_id() % num;
1362 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363 struct sk_buff *skb,
1366 return prandom_u32_max(num);
1369 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370 struct sk_buff *skb,
1371 unsigned int idx, bool try_self,
1374 struct packet_sock *po, *po_next, *po_skip = NULL;
1375 unsigned int i, j, room = ROOM_NONE;
1377 po = pkt_sk(f->arr[idx]);
1380 room = packet_rcv_has_room(po, skb);
1381 if (room == ROOM_NORMAL ||
1382 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1387 i = j = min_t(int, po->rollover->sock, num - 1);
1389 po_next = pkt_sk(f->arr[i]);
1390 if (po_next != po_skip && !po_next->pressure &&
1391 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1393 po->rollover->sock = i;
1394 atomic_long_inc(&po->rollover->num);
1395 if (room == ROOM_LOW)
1396 atomic_long_inc(&po->rollover->num_huge);
1404 atomic_long_inc(&po->rollover->num_failed);
1408 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409 struct sk_buff *skb,
1412 return skb_get_queue_mapping(skb) % num;
1415 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416 struct sk_buff *skb,
1419 struct bpf_prog *prog;
1420 unsigned int ret = 0;
1423 prog = rcu_dereference(f->bpf_prog);
1425 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1431 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1433 return f->flags & (flag >> 8);
1436 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437 struct packet_type *pt, struct net_device *orig_dev)
1439 struct packet_fanout *f = pt->af_packet_priv;
1440 unsigned int num = READ_ONCE(f->num_members);
1441 struct net *net = read_pnet(&f->net);
1442 struct packet_sock *po;
1445 if (!net_eq(dev_net(dev), net) || !num) {
1450 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1456 case PACKET_FANOUT_HASH:
1458 idx = fanout_demux_hash(f, skb, num);
1460 case PACKET_FANOUT_LB:
1461 idx = fanout_demux_lb(f, skb, num);
1463 case PACKET_FANOUT_CPU:
1464 idx = fanout_demux_cpu(f, skb, num);
1466 case PACKET_FANOUT_RND:
1467 idx = fanout_demux_rnd(f, skb, num);
1469 case PACKET_FANOUT_QM:
1470 idx = fanout_demux_qm(f, skb, num);
1472 case PACKET_FANOUT_ROLLOVER:
1473 idx = fanout_demux_rollover(f, skb, 0, false, num);
1475 case PACKET_FANOUT_CBPF:
1476 case PACKET_FANOUT_EBPF:
1477 idx = fanout_demux_bpf(f, skb, num);
1481 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482 idx = fanout_demux_rollover(f, skb, idx, true, num);
1484 po = pkt_sk(f->arr[idx]);
1485 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1488 DEFINE_MUTEX(fanout_mutex);
1489 EXPORT_SYMBOL_GPL(fanout_mutex);
1490 static LIST_HEAD(fanout_list);
1492 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1494 struct packet_fanout *f = po->fanout;
1496 spin_lock(&f->lock);
1497 f->arr[f->num_members] = sk;
1500 spin_unlock(&f->lock);
1503 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1505 struct packet_fanout *f = po->fanout;
1508 spin_lock(&f->lock);
1509 for (i = 0; i < f->num_members; i++) {
1510 if (f->arr[i] == sk)
1513 BUG_ON(i >= f->num_members);
1514 f->arr[i] = f->arr[f->num_members - 1];
1516 spin_unlock(&f->lock);
1519 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1521 if (sk->sk_family != PF_PACKET)
1524 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1527 static void fanout_init_data(struct packet_fanout *f)
1530 case PACKET_FANOUT_LB:
1531 atomic_set(&f->rr_cur, 0);
1533 case PACKET_FANOUT_CBPF:
1534 case PACKET_FANOUT_EBPF:
1535 RCU_INIT_POINTER(f->bpf_prog, NULL);
1540 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1542 struct bpf_prog *old;
1544 spin_lock(&f->lock);
1545 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1546 rcu_assign_pointer(f->bpf_prog, new);
1547 spin_unlock(&f->lock);
1551 bpf_prog_destroy(old);
1555 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1558 struct bpf_prog *new;
1559 struct sock_fprog fprog;
1562 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1564 if (len != sizeof(fprog))
1566 if (copy_from_user(&fprog, data, len))
1569 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1573 __fanout_set_data_bpf(po->fanout, new);
1577 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1580 struct bpf_prog *new;
1583 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1585 if (len != sizeof(fd))
1587 if (copy_from_user(&fd, data, len))
1590 new = bpf_prog_get(fd);
1592 return PTR_ERR(new);
1593 if (new->type != BPF_PROG_TYPE_SOCKET_FILTER) {
1598 __fanout_set_data_bpf(po->fanout, new);
1602 static int fanout_set_data(struct packet_sock *po, char __user *data,
1605 switch (po->fanout->type) {
1606 case PACKET_FANOUT_CBPF:
1607 return fanout_set_data_cbpf(po, data, len);
1608 case PACKET_FANOUT_EBPF:
1609 return fanout_set_data_ebpf(po, data, len);
1615 static void fanout_release_data(struct packet_fanout *f)
1618 case PACKET_FANOUT_CBPF:
1619 case PACKET_FANOUT_EBPF:
1620 __fanout_set_data_bpf(f, NULL);
1624 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1626 struct packet_sock *po = pkt_sk(sk);
1627 struct packet_fanout *f, *match;
1628 u8 type = type_flags & 0xff;
1629 u8 flags = type_flags >> 8;
1633 case PACKET_FANOUT_ROLLOVER:
1634 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1636 case PACKET_FANOUT_HASH:
1637 case PACKET_FANOUT_LB:
1638 case PACKET_FANOUT_CPU:
1639 case PACKET_FANOUT_RND:
1640 case PACKET_FANOUT_QM:
1641 case PACKET_FANOUT_CBPF:
1642 case PACKET_FANOUT_EBPF:
1654 if (type == PACKET_FANOUT_ROLLOVER ||
1655 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1656 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1659 atomic_long_set(&po->rollover->num, 0);
1660 atomic_long_set(&po->rollover->num_huge, 0);
1661 atomic_long_set(&po->rollover->num_failed, 0);
1664 mutex_lock(&fanout_mutex);
1666 list_for_each_entry(f, &fanout_list, list) {
1668 read_pnet(&f->net) == sock_net(sk)) {
1674 if (match && match->flags != flags)
1678 match = kzalloc(sizeof(*match), GFP_KERNEL);
1681 write_pnet(&match->net, sock_net(sk));
1684 match->flags = flags;
1685 INIT_LIST_HEAD(&match->list);
1686 spin_lock_init(&match->lock);
1687 atomic_set(&match->sk_ref, 0);
1688 fanout_init_data(match);
1689 match->prot_hook.type = po->prot_hook.type;
1690 match->prot_hook.dev = po->prot_hook.dev;
1691 match->prot_hook.func = packet_rcv_fanout;
1692 match->prot_hook.af_packet_priv = match;
1693 match->prot_hook.id_match = match_fanout_group;
1694 dev_add_pack(&match->prot_hook);
1695 list_add(&match->list, &fanout_list);
1698 if (match->type == type &&
1699 match->prot_hook.type == po->prot_hook.type &&
1700 match->prot_hook.dev == po->prot_hook.dev) {
1702 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1703 __dev_remove_pack(&po->prot_hook);
1705 atomic_inc(&match->sk_ref);
1706 __fanout_link(sk, po);
1711 mutex_unlock(&fanout_mutex);
1713 kfree(po->rollover);
1714 po->rollover = NULL;
1719 static void fanout_release(struct sock *sk)
1721 struct packet_sock *po = pkt_sk(sk);
1722 struct packet_fanout *f;
1728 mutex_lock(&fanout_mutex);
1731 if (atomic_dec_and_test(&f->sk_ref)) {
1733 dev_remove_pack(&f->prot_hook);
1734 fanout_release_data(f);
1737 mutex_unlock(&fanout_mutex);
1740 kfree_rcu(po->rollover, rcu);
1743 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1744 struct sk_buff *skb)
1746 /* Earlier code assumed this would be a VLAN pkt, double-check
1747 * this now that we have the actual packet in hand. We can only
1748 * do this check on Ethernet devices.
1750 if (unlikely(dev->type != ARPHRD_ETHER))
1753 skb_reset_mac_header(skb);
1754 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1757 static const struct proto_ops packet_ops;
1759 static const struct proto_ops packet_ops_spkt;
1761 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1762 struct packet_type *pt, struct net_device *orig_dev)
1765 struct sockaddr_pkt *spkt;
1768 * When we registered the protocol we saved the socket in the data
1769 * field for just this event.
1772 sk = pt->af_packet_priv;
1775 * Yank back the headers [hope the device set this
1776 * right or kerboom...]
1778 * Incoming packets have ll header pulled,
1781 * For outgoing ones skb->data == skb_mac_header(skb)
1782 * so that this procedure is noop.
1785 if (skb->pkt_type == PACKET_LOOPBACK)
1788 if (!net_eq(dev_net(dev), sock_net(sk)))
1791 skb = skb_share_check(skb, GFP_ATOMIC);
1795 /* drop any routing info */
1798 /* drop conntrack reference */
1801 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1803 skb_push(skb, skb->data - skb_mac_header(skb));
1806 * The SOCK_PACKET socket receives _all_ frames.
1809 spkt->spkt_family = dev->type;
1810 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1811 spkt->spkt_protocol = skb->protocol;
1814 * Charge the memory to the socket. This is done specifically
1815 * to prevent sockets using all the memory up.
1818 if (sock_queue_rcv_skb(sk, skb) == 0)
1829 * Output a raw packet to a device layer. This bypasses all the other
1830 * protocol layers and you must therefore supply it with a complete frame
1833 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1836 struct sock *sk = sock->sk;
1837 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1838 struct sk_buff *skb = NULL;
1839 struct net_device *dev;
1840 struct sockcm_cookie sockc;
1846 * Get and verify the address.
1850 if (msg->msg_namelen < sizeof(struct sockaddr))
1852 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1853 proto = saddr->spkt_protocol;
1855 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1858 * Find the device first to size check it
1861 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1864 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1870 if (!(dev->flags & IFF_UP))
1874 * You may not queue a frame bigger than the mtu. This is the lowest level
1875 * raw protocol and you must do your own fragmentation at this level.
1878 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1879 if (!netif_supports_nofcs(dev)) {
1880 err = -EPROTONOSUPPORT;
1883 extra_len = 4; /* We're doing our own CRC */
1887 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1891 size_t reserved = LL_RESERVED_SPACE(dev);
1892 int tlen = dev->needed_tailroom;
1893 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1896 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1899 /* FIXME: Save some space for broken drivers that write a hard
1900 * header at transmission time by themselves. PPP is the notable
1901 * one here. This should really be fixed at the driver level.
1903 skb_reserve(skb, reserved);
1904 skb_reset_network_header(skb);
1906 /* Try to align data part correctly */
1911 skb_reset_network_header(skb);
1913 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1919 if (!dev_validate_header(dev, skb->data, len)) {
1923 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1924 !packet_extra_vlan_len_allowed(dev, skb)) {
1930 if (msg->msg_controllen) {
1931 err = sock_cmsg_send(sk, msg, &sockc);
1932 if (unlikely(err)) {
1938 skb->protocol = proto;
1940 skb->priority = sk->sk_priority;
1941 skb->mark = sk->sk_mark;
1943 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1945 if (unlikely(extra_len == 4))
1948 skb_probe_transport_header(skb, 0);
1950 dev_queue_xmit(skb);
1961 static unsigned int run_filter(struct sk_buff *skb,
1962 const struct sock *sk,
1965 struct sk_filter *filter;
1968 filter = rcu_dereference(sk->sk_filter);
1970 res = bpf_prog_run_clear_cb(filter->prog, skb);
1976 static int __packet_rcv_vnet(const struct sk_buff *skb,
1977 struct virtio_net_hdr *vnet_hdr)
1979 *vnet_hdr = (const struct virtio_net_hdr) { 0 };
1981 if (skb_is_gso(skb)) {
1982 struct skb_shared_info *sinfo = skb_shinfo(skb);
1984 /* This is a hint as to how much should be linear. */
1986 __cpu_to_virtio16(vio_le(), skb_headlen(skb));
1987 vnet_hdr->gso_size =
1988 __cpu_to_virtio16(vio_le(), sinfo->gso_size);
1990 if (sinfo->gso_type & SKB_GSO_TCPV4)
1991 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1992 else if (sinfo->gso_type & SKB_GSO_TCPV6)
1993 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1994 else if (sinfo->gso_type & SKB_GSO_UDP)
1995 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
1996 else if (sinfo->gso_type & SKB_GSO_FCOE)
2001 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2002 vnet_hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2004 vnet_hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
2006 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2007 vnet_hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2008 vnet_hdr->csum_start = __cpu_to_virtio16(vio_le(),
2009 skb_checksum_start_offset(skb));
2010 vnet_hdr->csum_offset = __cpu_to_virtio16(vio_le(),
2012 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2013 vnet_hdr->flags = VIRTIO_NET_HDR_F_DATA_VALID;
2014 } /* else everything is zero */
2019 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2022 struct virtio_net_hdr vnet_hdr;
2024 if (*len < sizeof(vnet_hdr))
2026 *len -= sizeof(vnet_hdr);
2028 if (__packet_rcv_vnet(skb, &vnet_hdr))
2031 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2035 * This function makes lazy skb cloning in hope that most of packets
2036 * are discarded by BPF.
2038 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2039 * and skb->cb are mangled. It works because (and until) packets
2040 * falling here are owned by current CPU. Output packets are cloned
2041 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2042 * sequencially, so that if we return skb to original state on exit,
2043 * we will not harm anyone.
2046 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2047 struct packet_type *pt, struct net_device *orig_dev)
2050 struct sockaddr_ll *sll;
2051 struct packet_sock *po;
2052 u8 *skb_head = skb->data;
2053 int skb_len = skb->len;
2054 unsigned int snaplen, res;
2056 if (skb->pkt_type == PACKET_LOOPBACK)
2059 sk = pt->af_packet_priv;
2062 if (!net_eq(dev_net(dev), sock_net(sk)))
2067 if (dev->header_ops) {
2068 /* The device has an explicit notion of ll header,
2069 * exported to higher levels.
2071 * Otherwise, the device hides details of its frame
2072 * structure, so that corresponding packet head is
2073 * never delivered to user.
2075 if (sk->sk_type != SOCK_DGRAM)
2076 skb_push(skb, skb->data - skb_mac_header(skb));
2077 else if (skb->pkt_type == PACKET_OUTGOING) {
2078 /* Special case: outgoing packets have ll header at head */
2079 skb_pull(skb, skb_network_offset(skb));
2085 res = run_filter(skb, sk, snaplen);
2087 goto drop_n_restore;
2091 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2094 if (skb_shared(skb)) {
2095 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2099 if (skb_head != skb->data) {
2100 skb->data = skb_head;
2107 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2109 sll = &PACKET_SKB_CB(skb)->sa.ll;
2110 sll->sll_hatype = dev->type;
2111 sll->sll_pkttype = skb->pkt_type;
2112 if (unlikely(po->origdev))
2113 sll->sll_ifindex = orig_dev->ifindex;
2115 sll->sll_ifindex = dev->ifindex;
2117 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2119 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2120 * Use their space for storing the original skb length.
2122 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2124 if (pskb_trim(skb, snaplen))
2127 skb_set_owner_r(skb, sk);
2131 /* drop conntrack reference */
2134 spin_lock(&sk->sk_receive_queue.lock);
2135 po->stats.stats1.tp_packets++;
2136 sock_skb_set_dropcount(sk, skb);
2137 __skb_queue_tail(&sk->sk_receive_queue, skb);
2138 spin_unlock(&sk->sk_receive_queue.lock);
2139 sk->sk_data_ready(sk);
2143 spin_lock(&sk->sk_receive_queue.lock);
2144 po->stats.stats1.tp_drops++;
2145 atomic_inc(&sk->sk_drops);
2146 spin_unlock(&sk->sk_receive_queue.lock);
2149 if (skb_head != skb->data && skb_shared(skb)) {
2150 skb->data = skb_head;
2158 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2159 struct packet_type *pt, struct net_device *orig_dev)
2162 struct packet_sock *po;
2163 struct sockaddr_ll *sll;
2164 union tpacket_uhdr h;
2165 u8 *skb_head = skb->data;
2166 int skb_len = skb->len;
2167 unsigned int snaplen, res;
2168 unsigned long status = TP_STATUS_USER;
2169 unsigned short macoff, netoff, hdrlen;
2170 struct sk_buff *copy_skb = NULL;
2174 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2175 * We may add members to them until current aligned size without forcing
2176 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2178 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2179 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2181 if (skb->pkt_type == PACKET_LOOPBACK)
2184 sk = pt->af_packet_priv;
2187 if (!net_eq(dev_net(dev), sock_net(sk)))
2190 if (dev->header_ops) {
2191 if (sk->sk_type != SOCK_DGRAM)
2192 skb_push(skb, skb->data - skb_mac_header(skb));
2193 else if (skb->pkt_type == PACKET_OUTGOING) {
2194 /* Special case: outgoing packets have ll header at head */
2195 skb_pull(skb, skb_network_offset(skb));
2201 res = run_filter(skb, sk, snaplen);
2203 goto drop_n_restore;
2205 if (skb->ip_summed == CHECKSUM_PARTIAL)
2206 status |= TP_STATUS_CSUMNOTREADY;
2207 else if (skb->pkt_type != PACKET_OUTGOING &&
2208 (skb->ip_summed == CHECKSUM_COMPLETE ||
2209 skb_csum_unnecessary(skb)))
2210 status |= TP_STATUS_CSUM_VALID;
2215 if (sk->sk_type == SOCK_DGRAM) {
2216 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2219 unsigned int maclen = skb_network_offset(skb);
2220 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2221 (maclen < 16 ? 16 : maclen)) +
2223 if (po->has_vnet_hdr)
2224 netoff += sizeof(struct virtio_net_hdr);
2225 macoff = netoff - maclen;
2227 if (po->tp_version <= TPACKET_V2) {
2228 if (macoff + snaplen > po->rx_ring.frame_size) {
2229 if (po->copy_thresh &&
2230 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2231 if (skb_shared(skb)) {
2232 copy_skb = skb_clone(skb, GFP_ATOMIC);
2234 copy_skb = skb_get(skb);
2235 skb_head = skb->data;
2238 skb_set_owner_r(copy_skb, sk);
2240 snaplen = po->rx_ring.frame_size - macoff;
2241 if ((int)snaplen < 0)
2244 } else if (unlikely(macoff + snaplen >
2245 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2248 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2249 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2250 snaplen, nval, macoff);
2252 if (unlikely((int)snaplen < 0)) {
2254 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2257 spin_lock(&sk->sk_receive_queue.lock);
2258 h.raw = packet_current_rx_frame(po, skb,
2259 TP_STATUS_KERNEL, (macoff+snaplen));
2261 goto drop_n_account;
2262 if (po->tp_version <= TPACKET_V2) {
2263 packet_increment_rx_head(po, &po->rx_ring);
2265 * LOSING will be reported till you read the stats,
2266 * because it's COR - Clear On Read.
2267 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2270 if (po->stats.stats1.tp_drops)
2271 status |= TP_STATUS_LOSING;
2273 po->stats.stats1.tp_packets++;
2275 status |= TP_STATUS_COPY;
2276 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2278 spin_unlock(&sk->sk_receive_queue.lock);
2280 if (po->has_vnet_hdr) {
2281 if (__packet_rcv_vnet(skb, h.raw + macoff -
2282 sizeof(struct virtio_net_hdr))) {
2283 spin_lock(&sk->sk_receive_queue.lock);
2284 goto drop_n_account;
2288 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2290 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2291 getnstimeofday(&ts);
2293 status |= ts_status;
2295 switch (po->tp_version) {
2297 h.h1->tp_len = skb->len;
2298 h.h1->tp_snaplen = snaplen;
2299 h.h1->tp_mac = macoff;
2300 h.h1->tp_net = netoff;
2301 h.h1->tp_sec = ts.tv_sec;
2302 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2303 hdrlen = sizeof(*h.h1);
2306 h.h2->tp_len = skb->len;
2307 h.h2->tp_snaplen = snaplen;
2308 h.h2->tp_mac = macoff;
2309 h.h2->tp_net = netoff;
2310 h.h2->tp_sec = ts.tv_sec;
2311 h.h2->tp_nsec = ts.tv_nsec;
2312 if (skb_vlan_tag_present(skb)) {
2313 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2314 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2315 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2317 h.h2->tp_vlan_tci = 0;
2318 h.h2->tp_vlan_tpid = 0;
2320 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2321 hdrlen = sizeof(*h.h2);
2324 /* tp_nxt_offset,vlan are already populated above.
2325 * So DONT clear those fields here
2327 h.h3->tp_status |= status;
2328 h.h3->tp_len = skb->len;
2329 h.h3->tp_snaplen = snaplen;
2330 h.h3->tp_mac = macoff;
2331 h.h3->tp_net = netoff;
2332 h.h3->tp_sec = ts.tv_sec;
2333 h.h3->tp_nsec = ts.tv_nsec;
2334 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2335 hdrlen = sizeof(*h.h3);
2341 sll = h.raw + TPACKET_ALIGN(hdrlen);
2342 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2343 sll->sll_family = AF_PACKET;
2344 sll->sll_hatype = dev->type;
2345 sll->sll_protocol = skb->protocol;
2346 sll->sll_pkttype = skb->pkt_type;
2347 if (unlikely(po->origdev))
2348 sll->sll_ifindex = orig_dev->ifindex;
2350 sll->sll_ifindex = dev->ifindex;
2354 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2355 if (po->tp_version <= TPACKET_V2) {
2358 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2361 for (start = h.raw; start < end; start += PAGE_SIZE)
2362 flush_dcache_page(pgv_to_page(start));
2367 if (po->tp_version <= TPACKET_V2) {
2368 __packet_set_status(po, h.raw, status);
2369 sk->sk_data_ready(sk);
2371 prb_clear_blk_fill_status(&po->rx_ring);
2375 if (skb_head != skb->data && skb_shared(skb)) {
2376 skb->data = skb_head;
2384 po->stats.stats1.tp_drops++;
2385 spin_unlock(&sk->sk_receive_queue.lock);
2387 sk->sk_data_ready(sk);
2388 kfree_skb(copy_skb);
2389 goto drop_n_restore;
2392 static void tpacket_destruct_skb(struct sk_buff *skb)
2394 struct packet_sock *po = pkt_sk(skb->sk);
2396 if (likely(po->tx_ring.pg_vec)) {
2400 ph = skb_shinfo(skb)->destructor_arg;
2401 packet_dec_pending(&po->tx_ring);
2403 ts = __packet_set_timestamp(po, ph, skb);
2404 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2410 static void tpacket_set_protocol(const struct net_device *dev,
2411 struct sk_buff *skb)
2413 if (dev->type == ARPHRD_ETHER) {
2414 skb_reset_mac_header(skb);
2415 skb->protocol = eth_hdr(skb)->h_proto;
2419 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2421 unsigned short gso_type = 0;
2423 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2424 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2425 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2426 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2427 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2428 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2429 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2431 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2434 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2435 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2436 case VIRTIO_NET_HDR_GSO_TCPV4:
2437 gso_type = SKB_GSO_TCPV4;
2439 case VIRTIO_NET_HDR_GSO_TCPV6:
2440 gso_type = SKB_GSO_TCPV6;
2442 case VIRTIO_NET_HDR_GSO_UDP:
2443 gso_type = SKB_GSO_UDP;
2449 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
2450 gso_type |= SKB_GSO_TCP_ECN;
2452 if (vnet_hdr->gso_size == 0)
2456 vnet_hdr->gso_type = gso_type; /* changes type, temporary storage */
2460 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2461 struct virtio_net_hdr *vnet_hdr)
2465 if (*len < sizeof(*vnet_hdr))
2467 *len -= sizeof(*vnet_hdr);
2469 n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter);
2470 if (n != sizeof(*vnet_hdr))
2473 return __packet_snd_vnet_parse(vnet_hdr, *len);
2476 static int packet_snd_vnet_gso(struct sk_buff *skb,
2477 struct virtio_net_hdr *vnet_hdr)
2479 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2480 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start);
2481 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset);
2483 if (!skb_partial_csum_set(skb, s, o))
2487 skb_shinfo(skb)->gso_size =
2488 __virtio16_to_cpu(vio_le(), vnet_hdr->gso_size);
2489 skb_shinfo(skb)->gso_type = vnet_hdr->gso_type;
2491 /* Header must be checked, and gso_segs computed. */
2492 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2493 skb_shinfo(skb)->gso_segs = 0;
2497 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2498 void *frame, struct net_device *dev, void *data, int tp_len,
2499 __be16 proto, unsigned char *addr, int hlen, int copylen,
2500 const struct sockcm_cookie *sockc)
2502 union tpacket_uhdr ph;
2503 int to_write, offset, len, nr_frags, len_max;
2504 struct socket *sock = po->sk.sk_socket;
2510 skb->protocol = proto;
2512 skb->priority = po->sk.sk_priority;
2513 skb->mark = po->sk.sk_mark;
2514 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2515 skb_shinfo(skb)->destructor_arg = ph.raw;
2517 skb_reserve(skb, hlen);
2518 skb_reset_network_header(skb);
2522 if (sock->type == SOCK_DGRAM) {
2523 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2525 if (unlikely(err < 0))
2527 } else if (copylen) {
2528 int hdrlen = min_t(int, copylen, tp_len);
2530 skb_push(skb, dev->hard_header_len);
2531 skb_put(skb, copylen - dev->hard_header_len);
2532 err = skb_store_bits(skb, 0, data, hdrlen);
2535 if (!dev_validate_header(dev, skb->data, hdrlen))
2538 tpacket_set_protocol(dev, skb);
2544 offset = offset_in_page(data);
2545 len_max = PAGE_SIZE - offset;
2546 len = ((to_write > len_max) ? len_max : to_write);
2548 skb->data_len = to_write;
2549 skb->len += to_write;
2550 skb->truesize += to_write;
2551 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2553 while (likely(to_write)) {
2554 nr_frags = skb_shinfo(skb)->nr_frags;
2556 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2557 pr_err("Packet exceed the number of skb frags(%lu)\n",
2562 page = pgv_to_page(data);
2564 flush_dcache_page(page);
2566 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2569 len_max = PAGE_SIZE;
2570 len = ((to_write > len_max) ? len_max : to_write);
2573 skb_probe_transport_header(skb, 0);
2578 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2579 int size_max, void **data)
2581 union tpacket_uhdr ph;
2586 switch (po->tp_version) {
2588 tp_len = ph.h2->tp_len;
2591 tp_len = ph.h1->tp_len;
2594 if (unlikely(tp_len > size_max)) {
2595 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2599 if (unlikely(po->tp_tx_has_off)) {
2600 int off_min, off_max;
2602 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2603 off_max = po->tx_ring.frame_size - tp_len;
2604 if (po->sk.sk_type == SOCK_DGRAM) {
2605 switch (po->tp_version) {
2607 off = ph.h2->tp_net;
2610 off = ph.h1->tp_net;
2614 switch (po->tp_version) {
2616 off = ph.h2->tp_mac;
2619 off = ph.h1->tp_mac;
2623 if (unlikely((off < off_min) || (off_max < off)))
2626 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2629 *data = frame + off;
2633 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2635 struct sk_buff *skb;
2636 struct net_device *dev;
2637 struct virtio_net_hdr *vnet_hdr = NULL;
2638 struct sockcm_cookie sockc;
2640 int err, reserve = 0;
2642 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2643 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2644 int tp_len, size_max;
2645 unsigned char *addr;
2648 int status = TP_STATUS_AVAILABLE;
2649 int hlen, tlen, copylen = 0;
2651 mutex_lock(&po->pg_vec_lock);
2653 if (likely(saddr == NULL)) {
2654 dev = packet_cached_dev_get(po);
2659 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2661 if (msg->msg_namelen < (saddr->sll_halen
2662 + offsetof(struct sockaddr_ll,
2665 proto = saddr->sll_protocol;
2666 addr = saddr->sll_addr;
2667 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2671 if (msg->msg_controllen) {
2672 err = sock_cmsg_send(&po->sk, msg, &sockc);
2678 if (unlikely(dev == NULL))
2681 if (unlikely(!(dev->flags & IFF_UP)))
2684 if (po->sk.sk_socket->type == SOCK_RAW)
2685 reserve = dev->hard_header_len;
2686 size_max = po->tx_ring.frame_size
2687 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2689 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2690 size_max = dev->mtu + reserve + VLAN_HLEN;
2693 ph = packet_current_frame(po, &po->tx_ring,
2694 TP_STATUS_SEND_REQUEST);
2695 if (unlikely(ph == NULL)) {
2696 if (need_wait && need_resched())
2702 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2706 status = TP_STATUS_SEND_REQUEST;
2707 hlen = LL_RESERVED_SPACE(dev);
2708 tlen = dev->needed_tailroom;
2709 if (po->has_vnet_hdr) {
2711 data += sizeof(*vnet_hdr);
2712 tp_len -= sizeof(*vnet_hdr);
2714 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2718 copylen = __virtio16_to_cpu(vio_le(),
2721 copylen = max_t(int, copylen, dev->hard_header_len);
2722 skb = sock_alloc_send_skb(&po->sk,
2723 hlen + tlen + sizeof(struct sockaddr_ll) +
2724 (copylen - dev->hard_header_len),
2727 if (unlikely(skb == NULL)) {
2728 /* we assume the socket was initially writeable ... */
2729 if (likely(len_sum > 0))
2733 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2734 addr, hlen, copylen, &sockc);
2735 if (likely(tp_len >= 0) &&
2736 tp_len > dev->mtu + reserve &&
2737 !po->has_vnet_hdr &&
2738 !packet_extra_vlan_len_allowed(dev, skb))
2741 if (unlikely(tp_len < 0)) {
2744 __packet_set_status(po, ph,
2745 TP_STATUS_AVAILABLE);
2746 packet_increment_head(&po->tx_ring);
2750 status = TP_STATUS_WRONG_FORMAT;
2756 if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) {
2761 packet_pick_tx_queue(dev, skb);
2763 skb->destructor = tpacket_destruct_skb;
2764 __packet_set_status(po, ph, TP_STATUS_SENDING);
2765 packet_inc_pending(&po->tx_ring);
2767 status = TP_STATUS_SEND_REQUEST;
2768 err = po->xmit(skb);
2769 if (unlikely(err > 0)) {
2770 err = net_xmit_errno(err);
2771 if (err && __packet_get_status(po, ph) ==
2772 TP_STATUS_AVAILABLE) {
2773 /* skb was destructed already */
2778 * skb was dropped but not destructed yet;
2779 * let's treat it like congestion or err < 0
2783 packet_increment_head(&po->tx_ring);
2785 } while (likely((ph != NULL) ||
2786 /* Note: packet_read_pending() might be slow if we have
2787 * to call it as it's per_cpu variable, but in fast-path
2788 * we already short-circuit the loop with the first
2789 * condition, and luckily don't have to go that path
2792 (need_wait && packet_read_pending(&po->tx_ring))));
2798 __packet_set_status(po, ph, status);
2803 mutex_unlock(&po->pg_vec_lock);
2807 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2808 size_t reserve, size_t len,
2809 size_t linear, int noblock,
2812 struct sk_buff *skb;
2814 /* Under a page? Don't bother with paged skb. */
2815 if (prepad + len < PAGE_SIZE || !linear)
2818 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2823 skb_reserve(skb, reserve);
2824 skb_put(skb, linear);
2825 skb->data_len = len - linear;
2826 skb->len += len - linear;
2831 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2833 struct sock *sk = sock->sk;
2834 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2835 struct sk_buff *skb;
2836 struct net_device *dev;
2838 unsigned char *addr;
2839 int err, reserve = 0;
2840 struct sockcm_cookie sockc;
2841 struct virtio_net_hdr vnet_hdr = { 0 };
2843 struct packet_sock *po = pkt_sk(sk);
2848 * Get and verify the address.
2851 if (likely(saddr == NULL)) {
2852 dev = packet_cached_dev_get(po);
2857 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2859 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2861 proto = saddr->sll_protocol;
2862 addr = saddr->sll_addr;
2863 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2867 if (unlikely(dev == NULL))
2870 if (unlikely(!(dev->flags & IFF_UP)))
2874 sockc.mark = sk->sk_mark;
2875 if (msg->msg_controllen) {
2876 err = sock_cmsg_send(sk, msg, &sockc);
2881 if (sock->type == SOCK_RAW)
2882 reserve = dev->hard_header_len;
2883 if (po->has_vnet_hdr) {
2884 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2889 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2890 if (!netif_supports_nofcs(dev)) {
2891 err = -EPROTONOSUPPORT;
2894 extra_len = 4; /* We're doing our own CRC */
2898 if (!vnet_hdr.gso_type &&
2899 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2903 hlen = LL_RESERVED_SPACE(dev);
2904 tlen = dev->needed_tailroom;
2905 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2906 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2907 msg->msg_flags & MSG_DONTWAIT, &err);
2911 skb_set_network_header(skb, reserve);
2914 if (sock->type == SOCK_DGRAM) {
2915 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2916 if (unlikely(offset < 0))
2920 /* Returns -EFAULT on error */
2921 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2925 if (sock->type == SOCK_RAW &&
2926 !dev_validate_header(dev, skb->data, len)) {
2931 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2933 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2934 !packet_extra_vlan_len_allowed(dev, skb)) {
2939 skb->protocol = proto;
2941 skb->priority = sk->sk_priority;
2942 skb->mark = sockc.mark;
2944 packet_pick_tx_queue(dev, skb);
2946 if (po->has_vnet_hdr) {
2947 err = packet_snd_vnet_gso(skb, &vnet_hdr);
2950 len += sizeof(vnet_hdr);
2953 skb_probe_transport_header(skb, reserve);
2955 if (unlikely(extra_len == 4))
2958 err = po->xmit(skb);
2959 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2975 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2977 struct sock *sk = sock->sk;
2978 struct packet_sock *po = pkt_sk(sk);
2980 if (po->tx_ring.pg_vec)
2981 return tpacket_snd(po, msg);
2983 return packet_snd(sock, msg, len);
2987 * Close a PACKET socket. This is fairly simple. We immediately go
2988 * to 'closed' state and remove our protocol entry in the device list.
2991 static int packet_release(struct socket *sock)
2993 struct sock *sk = sock->sk;
2994 struct packet_sock *po;
2996 union tpacket_req_u req_u;
3004 mutex_lock(&net->packet.sklist_lock);
3005 sk_del_node_init_rcu(sk);
3006 mutex_unlock(&net->packet.sklist_lock);
3009 sock_prot_inuse_add(net, sk->sk_prot, -1);
3012 spin_lock(&po->bind_lock);
3013 unregister_prot_hook(sk, false);
3014 packet_cached_dev_reset(po);
3016 if (po->prot_hook.dev) {
3017 dev_put(po->prot_hook.dev);
3018 po->prot_hook.dev = NULL;
3020 spin_unlock(&po->bind_lock);
3022 packet_flush_mclist(sk);
3024 if (po->rx_ring.pg_vec) {
3025 memset(&req_u, 0, sizeof(req_u));
3026 packet_set_ring(sk, &req_u, 1, 0);
3029 if (po->tx_ring.pg_vec) {
3030 memset(&req_u, 0, sizeof(req_u));
3031 packet_set_ring(sk, &req_u, 1, 1);
3038 * Now the socket is dead. No more input will appear.
3045 skb_queue_purge(&sk->sk_receive_queue);
3046 packet_free_pending(po);
3047 sk_refcnt_debug_release(sk);
3054 * Attach a packet hook.
3057 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3060 struct packet_sock *po = pkt_sk(sk);
3061 struct net_device *dev_curr;
3064 struct net_device *dev = NULL;
3066 bool unlisted = false;
3072 spin_lock(&po->bind_lock);
3076 dev = dev_get_by_name_rcu(sock_net(sk), name);
3081 } else if (ifindex) {
3082 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3092 proto_curr = po->prot_hook.type;
3093 dev_curr = po->prot_hook.dev;
3095 need_rehook = proto_curr != proto || dev_curr != dev;
3100 __unregister_prot_hook(sk, true);
3102 dev_curr = po->prot_hook.dev;
3104 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3109 po->prot_hook.type = proto;
3111 if (unlikely(unlisted)) {
3113 po->prot_hook.dev = NULL;
3115 packet_cached_dev_reset(po);
3117 po->prot_hook.dev = dev;
3118 po->ifindex = dev ? dev->ifindex : 0;
3119 packet_cached_dev_assign(po, dev);
3125 if (proto == 0 || !need_rehook)
3128 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3129 register_prot_hook(sk);
3131 sk->sk_err = ENETDOWN;
3132 if (!sock_flag(sk, SOCK_DEAD))
3133 sk->sk_error_report(sk);
3138 spin_unlock(&po->bind_lock);
3144 * Bind a packet socket to a device
3147 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3150 struct sock *sk = sock->sk;
3157 if (addr_len != sizeof(struct sockaddr))
3159 strlcpy(name, uaddr->sa_data, sizeof(name));
3161 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3164 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3166 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3167 struct sock *sk = sock->sk;
3173 if (addr_len < sizeof(struct sockaddr_ll))
3175 if (sll->sll_family != AF_PACKET)
3178 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3179 sll->sll_protocol ? : pkt_sk(sk)->num);
3182 static struct proto packet_proto = {
3184 .owner = THIS_MODULE,
3185 .obj_size = sizeof(struct packet_sock),
3189 * Create a packet of type SOCK_PACKET.
3192 static int packet_create(struct net *net, struct socket *sock, int protocol,
3196 struct packet_sock *po;
3197 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3200 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3202 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3203 sock->type != SOCK_PACKET)
3204 return -ESOCKTNOSUPPORT;
3206 sock->state = SS_UNCONNECTED;
3209 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3213 sock->ops = &packet_ops;
3214 if (sock->type == SOCK_PACKET)
3215 sock->ops = &packet_ops_spkt;
3217 sock_init_data(sock, sk);
3220 sk->sk_family = PF_PACKET;
3222 po->xmit = dev_queue_xmit;
3224 err = packet_alloc_pending(po);
3228 packet_cached_dev_reset(po);
3230 sk->sk_destruct = packet_sock_destruct;
3231 sk_refcnt_debug_inc(sk);
3234 * Attach a protocol block
3237 spin_lock_init(&po->bind_lock);
3238 mutex_init(&po->pg_vec_lock);
3239 po->rollover = NULL;
3240 po->prot_hook.func = packet_rcv;
3242 if (sock->type == SOCK_PACKET)
3243 po->prot_hook.func = packet_rcv_spkt;
3245 po->prot_hook.af_packet_priv = sk;
3248 po->prot_hook.type = proto;
3249 register_prot_hook(sk);
3252 mutex_lock(&net->packet.sklist_lock);
3253 sk_add_node_rcu(sk, &net->packet.sklist);
3254 mutex_unlock(&net->packet.sklist_lock);
3257 sock_prot_inuse_add(net, &packet_proto, 1);
3268 * Pull a packet from our receive queue and hand it to the user.
3269 * If necessary we block.
3272 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3275 struct sock *sk = sock->sk;
3276 struct sk_buff *skb;
3278 int vnet_hdr_len = 0;
3279 unsigned int origlen = 0;
3282 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3286 /* What error should we return now? EUNATTACH? */
3287 if (pkt_sk(sk)->ifindex < 0)
3291 if (flags & MSG_ERRQUEUE) {
3292 err = sock_recv_errqueue(sk, msg, len,
3293 SOL_PACKET, PACKET_TX_TIMESTAMP);
3298 * Call the generic datagram receiver. This handles all sorts
3299 * of horrible races and re-entrancy so we can forget about it
3300 * in the protocol layers.
3302 * Now it will return ENETDOWN, if device have just gone down,
3303 * but then it will block.
3306 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3309 * An error occurred so return it. Because skb_recv_datagram()
3310 * handles the blocking we don't see and worry about blocking
3317 if (pkt_sk(sk)->pressure)
3318 packet_rcv_has_room(pkt_sk(sk), NULL);
3320 if (pkt_sk(sk)->has_vnet_hdr) {
3321 err = packet_rcv_vnet(msg, skb, &len);
3324 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3327 /* You lose any data beyond the buffer you gave. If it worries
3328 * a user program they can ask the device for its MTU
3334 msg->msg_flags |= MSG_TRUNC;
3337 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3341 if (sock->type != SOCK_PACKET) {
3342 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3344 /* Original length was stored in sockaddr_ll fields */
3345 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3346 sll->sll_family = AF_PACKET;
3347 sll->sll_protocol = skb->protocol;
3350 sock_recv_ts_and_drops(msg, sk, skb);
3352 if (msg->msg_name) {
3353 /* If the address length field is there to be filled
3354 * in, we fill it in now.
3356 if (sock->type == SOCK_PACKET) {
3357 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3358 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3360 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3362 msg->msg_namelen = sll->sll_halen +
3363 offsetof(struct sockaddr_ll, sll_addr);
3365 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3369 if (pkt_sk(sk)->auxdata) {
3370 struct tpacket_auxdata aux;
3372 aux.tp_status = TP_STATUS_USER;
3373 if (skb->ip_summed == CHECKSUM_PARTIAL)
3374 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3375 else if (skb->pkt_type != PACKET_OUTGOING &&
3376 (skb->ip_summed == CHECKSUM_COMPLETE ||
3377 skb_csum_unnecessary(skb)))
3378 aux.tp_status |= TP_STATUS_CSUM_VALID;
3380 aux.tp_len = origlen;
3381 aux.tp_snaplen = skb->len;
3383 aux.tp_net = skb_network_offset(skb);
3384 if (skb_vlan_tag_present(skb)) {
3385 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3386 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3387 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3389 aux.tp_vlan_tci = 0;
3390 aux.tp_vlan_tpid = 0;
3392 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3396 * Free or return the buffer as appropriate. Again this
3397 * hides all the races and re-entrancy issues from us.
3399 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3402 skb_free_datagram(sk, skb);
3407 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3408 int *uaddr_len, int peer)
3410 struct net_device *dev;
3411 struct sock *sk = sock->sk;
3416 uaddr->sa_family = AF_PACKET;
3417 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3419 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3421 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3423 *uaddr_len = sizeof(*uaddr);
3428 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3429 int *uaddr_len, int peer)
3431 struct net_device *dev;
3432 struct sock *sk = sock->sk;
3433 struct packet_sock *po = pkt_sk(sk);
3434 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3439 sll->sll_family = AF_PACKET;
3440 sll->sll_ifindex = po->ifindex;
3441 sll->sll_protocol = po->num;
3442 sll->sll_pkttype = 0;
3444 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3446 sll->sll_hatype = dev->type;
3447 sll->sll_halen = dev->addr_len;
3448 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3450 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3454 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3459 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3463 case PACKET_MR_MULTICAST:
3464 if (i->alen != dev->addr_len)
3467 return dev_mc_add(dev, i->addr);
3469 return dev_mc_del(dev, i->addr);
3471 case PACKET_MR_PROMISC:
3472 return dev_set_promiscuity(dev, what);
3473 case PACKET_MR_ALLMULTI:
3474 return dev_set_allmulti(dev, what);
3475 case PACKET_MR_UNICAST:
3476 if (i->alen != dev->addr_len)
3479 return dev_uc_add(dev, i->addr);
3481 return dev_uc_del(dev, i->addr);
3489 static void packet_dev_mclist_delete(struct net_device *dev,
3490 struct packet_mclist **mlp)
3492 struct packet_mclist *ml;
3494 while ((ml = *mlp) != NULL) {
3495 if (ml->ifindex == dev->ifindex) {
3496 packet_dev_mc(dev, ml, -1);
3504 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3506 struct packet_sock *po = pkt_sk(sk);
3507 struct packet_mclist *ml, *i;
3508 struct net_device *dev;
3514 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3519 if (mreq->mr_alen > dev->addr_len)
3523 i = kmalloc(sizeof(*i), GFP_KERNEL);
3528 for (ml = po->mclist; ml; ml = ml->next) {
3529 if (ml->ifindex == mreq->mr_ifindex &&
3530 ml->type == mreq->mr_type &&
3531 ml->alen == mreq->mr_alen &&
3532 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3534 /* Free the new element ... */
3540 i->type = mreq->mr_type;
3541 i->ifindex = mreq->mr_ifindex;
3542 i->alen = mreq->mr_alen;
3543 memcpy(i->addr, mreq->mr_address, i->alen);
3545 i->next = po->mclist;
3547 err = packet_dev_mc(dev, i, 1);
3549 po->mclist = i->next;
3558 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3560 struct packet_mclist *ml, **mlp;
3564 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3565 if (ml->ifindex == mreq->mr_ifindex &&
3566 ml->type == mreq->mr_type &&
3567 ml->alen == mreq->mr_alen &&
3568 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3569 if (--ml->count == 0) {
3570 struct net_device *dev;
3572 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3574 packet_dev_mc(dev, ml, -1);
3584 static void packet_flush_mclist(struct sock *sk)
3586 struct packet_sock *po = pkt_sk(sk);
3587 struct packet_mclist *ml;
3593 while ((ml = po->mclist) != NULL) {
3594 struct net_device *dev;
3596 po->mclist = ml->next;
3597 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3599 packet_dev_mc(dev, ml, -1);
3606 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3608 struct sock *sk = sock->sk;
3609 struct packet_sock *po = pkt_sk(sk);
3612 if (level != SOL_PACKET)
3613 return -ENOPROTOOPT;
3616 case PACKET_ADD_MEMBERSHIP:
3617 case PACKET_DROP_MEMBERSHIP:
3619 struct packet_mreq_max mreq;
3621 memset(&mreq, 0, sizeof(mreq));
3622 if (len < sizeof(struct packet_mreq))
3624 if (len > sizeof(mreq))
3626 if (copy_from_user(&mreq, optval, len))
3628 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3630 if (optname == PACKET_ADD_MEMBERSHIP)
3631 ret = packet_mc_add(sk, &mreq);
3633 ret = packet_mc_drop(sk, &mreq);
3637 case PACKET_RX_RING:
3638 case PACKET_TX_RING:
3640 union tpacket_req_u req_u;
3643 switch (po->tp_version) {
3646 len = sizeof(req_u.req);
3650 len = sizeof(req_u.req3);
3655 if (copy_from_user(&req_u.req, optval, len))
3657 return packet_set_ring(sk, &req_u, 0,
3658 optname == PACKET_TX_RING);
3660 case PACKET_COPY_THRESH:
3664 if (optlen != sizeof(val))
3666 if (copy_from_user(&val, optval, sizeof(val)))
3669 pkt_sk(sk)->copy_thresh = val;
3672 case PACKET_VERSION:
3676 if (optlen != sizeof(val))
3678 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3680 if (copy_from_user(&val, optval, sizeof(val)))
3686 po->tp_version = val;
3692 case PACKET_RESERVE:
3696 if (optlen != sizeof(val))
3698 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3700 if (copy_from_user(&val, optval, sizeof(val)))
3702 po->tp_reserve = val;
3709 if (optlen != sizeof(val))
3711 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3713 if (copy_from_user(&val, optval, sizeof(val)))
3715 po->tp_loss = !!val;
3718 case PACKET_AUXDATA:
3722 if (optlen < sizeof(val))
3724 if (copy_from_user(&val, optval, sizeof(val)))
3727 po->auxdata = !!val;
3730 case PACKET_ORIGDEV:
3734 if (optlen < sizeof(val))
3736 if (copy_from_user(&val, optval, sizeof(val)))
3739 po->origdev = !!val;
3742 case PACKET_VNET_HDR:
3746 if (sock->type != SOCK_RAW)
3748 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3750 if (optlen < sizeof(val))
3752 if (copy_from_user(&val, optval, sizeof(val)))
3755 po->has_vnet_hdr = !!val;
3758 case PACKET_TIMESTAMP:
3762 if (optlen != sizeof(val))
3764 if (copy_from_user(&val, optval, sizeof(val)))
3767 po->tp_tstamp = val;
3774 if (optlen != sizeof(val))
3776 if (copy_from_user(&val, optval, sizeof(val)))
3779 return fanout_add(sk, val & 0xffff, val >> 16);
3781 case PACKET_FANOUT_DATA:
3786 return fanout_set_data(po, optval, optlen);
3788 case PACKET_TX_HAS_OFF:
3792 if (optlen != sizeof(val))
3794 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3796 if (copy_from_user(&val, optval, sizeof(val)))
3798 po->tp_tx_has_off = !!val;
3801 case PACKET_QDISC_BYPASS:
3805 if (optlen != sizeof(val))
3807 if (copy_from_user(&val, optval, sizeof(val)))
3810 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3814 return -ENOPROTOOPT;
3818 static int packet_getsockopt(struct socket *sock, int level, int optname,
3819 char __user *optval, int __user *optlen)
3822 int val, lv = sizeof(val);
3823 struct sock *sk = sock->sk;
3824 struct packet_sock *po = pkt_sk(sk);
3826 union tpacket_stats_u st;
3827 struct tpacket_rollover_stats rstats;
3829 if (level != SOL_PACKET)
3830 return -ENOPROTOOPT;
3832 if (get_user(len, optlen))
3839 case PACKET_STATISTICS:
3840 spin_lock_bh(&sk->sk_receive_queue.lock);
3841 memcpy(&st, &po->stats, sizeof(st));
3842 memset(&po->stats, 0, sizeof(po->stats));
3843 spin_unlock_bh(&sk->sk_receive_queue.lock);
3845 if (po->tp_version == TPACKET_V3) {
3846 lv = sizeof(struct tpacket_stats_v3);
3847 st.stats3.tp_packets += st.stats3.tp_drops;
3850 lv = sizeof(struct tpacket_stats);
3851 st.stats1.tp_packets += st.stats1.tp_drops;
3856 case PACKET_AUXDATA:
3859 case PACKET_ORIGDEV:
3862 case PACKET_VNET_HDR:
3863 val = po->has_vnet_hdr;
3865 case PACKET_VERSION:
3866 val = po->tp_version;
3869 if (len > sizeof(int))
3871 if (copy_from_user(&val, optval, len))
3875 val = sizeof(struct tpacket_hdr);
3878 val = sizeof(struct tpacket2_hdr);
3881 val = sizeof(struct tpacket3_hdr);
3887 case PACKET_RESERVE:
3888 val = po->tp_reserve;
3893 case PACKET_TIMESTAMP:
3894 val = po->tp_tstamp;
3898 ((u32)po->fanout->id |
3899 ((u32)po->fanout->type << 16) |
3900 ((u32)po->fanout->flags << 24)) :
3903 case PACKET_ROLLOVER_STATS:
3906 rstats.tp_all = atomic_long_read(&po->rollover->num);
3907 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3908 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3910 lv = sizeof(rstats);
3912 case PACKET_TX_HAS_OFF:
3913 val = po->tp_tx_has_off;
3915 case PACKET_QDISC_BYPASS:
3916 val = packet_use_direct_xmit(po);
3919 return -ENOPROTOOPT;
3924 if (put_user(len, optlen))
3926 if (copy_to_user(optval, data, len))
3932 static int packet_notifier(struct notifier_block *this,
3933 unsigned long msg, void *ptr)
3936 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3937 struct net *net = dev_net(dev);
3940 sk_for_each_rcu(sk, &net->packet.sklist) {
3941 struct packet_sock *po = pkt_sk(sk);
3944 case NETDEV_UNREGISTER:
3946 packet_dev_mclist_delete(dev, &po->mclist);
3950 if (dev->ifindex == po->ifindex) {
3951 spin_lock(&po->bind_lock);
3953 __unregister_prot_hook(sk, false);
3954 sk->sk_err = ENETDOWN;
3955 if (!sock_flag(sk, SOCK_DEAD))
3956 sk->sk_error_report(sk);
3958 if (msg == NETDEV_UNREGISTER) {
3959 packet_cached_dev_reset(po);
3961 if (po->prot_hook.dev)
3962 dev_put(po->prot_hook.dev);
3963 po->prot_hook.dev = NULL;
3965 spin_unlock(&po->bind_lock);
3969 if (dev->ifindex == po->ifindex) {
3970 spin_lock(&po->bind_lock);
3972 register_prot_hook(sk);
3973 spin_unlock(&po->bind_lock);
3983 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3986 struct sock *sk = sock->sk;
3991 int amount = sk_wmem_alloc_get(sk);
3993 return put_user(amount, (int __user *)arg);
3997 struct sk_buff *skb;
4000 spin_lock_bh(&sk->sk_receive_queue.lock);
4001 skb = skb_peek(&sk->sk_receive_queue);
4004 spin_unlock_bh(&sk->sk_receive_queue.lock);
4005 return put_user(amount, (int __user *)arg);
4008 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4010 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4020 case SIOCGIFBRDADDR:
4021 case SIOCSIFBRDADDR:
4022 case SIOCGIFNETMASK:
4023 case SIOCSIFNETMASK:
4024 case SIOCGIFDSTADDR:
4025 case SIOCSIFDSTADDR:
4027 return inet_dgram_ops.ioctl(sock, cmd, arg);
4031 return -ENOIOCTLCMD;
4036 static unsigned int packet_poll(struct file *file, struct socket *sock,
4039 struct sock *sk = sock->sk;
4040 struct packet_sock *po = pkt_sk(sk);
4041 unsigned int mask = datagram_poll(file, sock, wait);
4043 spin_lock_bh(&sk->sk_receive_queue.lock);
4044 if (po->rx_ring.pg_vec) {
4045 if (!packet_previous_rx_frame(po, &po->rx_ring,
4047 mask |= POLLIN | POLLRDNORM;
4049 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4051 spin_unlock_bh(&sk->sk_receive_queue.lock);
4052 spin_lock_bh(&sk->sk_write_queue.lock);
4053 if (po->tx_ring.pg_vec) {
4054 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4055 mask |= POLLOUT | POLLWRNORM;
4057 spin_unlock_bh(&sk->sk_write_queue.lock);
4062 /* Dirty? Well, I still did not learn better way to account
4066 static void packet_mm_open(struct vm_area_struct *vma)
4068 struct file *file = vma->vm_file;
4069 struct socket *sock = file->private_data;
4070 struct sock *sk = sock->sk;
4073 atomic_inc(&pkt_sk(sk)->mapped);
4076 static void packet_mm_close(struct vm_area_struct *vma)
4078 struct file *file = vma->vm_file;
4079 struct socket *sock = file->private_data;
4080 struct sock *sk = sock->sk;
4083 atomic_dec(&pkt_sk(sk)->mapped);
4086 static const struct vm_operations_struct packet_mmap_ops = {
4087 .open = packet_mm_open,
4088 .close = packet_mm_close,
4091 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4096 for (i = 0; i < len; i++) {
4097 if (likely(pg_vec[i].buffer)) {
4098 if (is_vmalloc_addr(pg_vec[i].buffer))
4099 vfree(pg_vec[i].buffer);
4101 free_pages((unsigned long)pg_vec[i].buffer,
4103 pg_vec[i].buffer = NULL;
4109 static char *alloc_one_pg_vec_page(unsigned long order)
4112 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4113 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4115 buffer = (char *) __get_free_pages(gfp_flags, order);
4119 /* __get_free_pages failed, fall back to vmalloc */
4120 buffer = vzalloc((1 << order) * PAGE_SIZE);
4124 /* vmalloc failed, lets dig into swap here */
4125 gfp_flags &= ~__GFP_NORETRY;
4126 buffer = (char *) __get_free_pages(gfp_flags, order);
4130 /* complete and utter failure */
4134 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4136 unsigned int block_nr = req->tp_block_nr;
4140 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4141 if (unlikely(!pg_vec))
4144 for (i = 0; i < block_nr; i++) {
4145 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4146 if (unlikely(!pg_vec[i].buffer))
4147 goto out_free_pgvec;
4154 free_pg_vec(pg_vec, order, block_nr);
4159 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4160 int closing, int tx_ring)
4162 struct pgv *pg_vec = NULL;
4163 struct packet_sock *po = pkt_sk(sk);
4164 int was_running, order = 0;
4165 struct packet_ring_buffer *rb;
4166 struct sk_buff_head *rb_queue;
4169 /* Added to avoid minimal code churn */
4170 struct tpacket_req *req = &req_u->req;
4172 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4173 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4174 WARN(1, "Tx-ring is not supported.\n");
4178 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4179 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4183 if (atomic_read(&po->mapped))
4185 if (packet_read_pending(rb))
4189 if (req->tp_block_nr) {
4190 /* Sanity tests and some calculations */
4192 if (unlikely(rb->pg_vec))
4195 switch (po->tp_version) {
4197 po->tp_hdrlen = TPACKET_HDRLEN;
4200 po->tp_hdrlen = TPACKET2_HDRLEN;
4203 po->tp_hdrlen = TPACKET3_HDRLEN;
4208 if (unlikely((int)req->tp_block_size <= 0))
4210 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4212 if (po->tp_version >= TPACKET_V3 &&
4213 (int)(req->tp_block_size -
4214 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4216 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4219 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4222 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4223 if (unlikely(rb->frames_per_block == 0))
4225 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4230 order = get_order(req->tp_block_size);
4231 pg_vec = alloc_pg_vec(req, order);
4232 if (unlikely(!pg_vec))
4234 switch (po->tp_version) {
4236 /* Transmit path is not supported. We checked
4237 * it above but just being paranoid
4240 init_prb_bdqc(po, rb, pg_vec, req_u);
4249 if (unlikely(req->tp_frame_nr))
4255 /* Detach socket from network */
4256 spin_lock(&po->bind_lock);
4257 was_running = po->running;
4261 __unregister_prot_hook(sk, false);
4263 spin_unlock(&po->bind_lock);
4268 mutex_lock(&po->pg_vec_lock);
4269 if (closing || atomic_read(&po->mapped) == 0) {
4271 spin_lock_bh(&rb_queue->lock);
4272 swap(rb->pg_vec, pg_vec);
4273 rb->frame_max = (req->tp_frame_nr - 1);
4275 rb->frame_size = req->tp_frame_size;
4276 spin_unlock_bh(&rb_queue->lock);
4278 swap(rb->pg_vec_order, order);
4279 swap(rb->pg_vec_len, req->tp_block_nr);
4281 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4282 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4283 tpacket_rcv : packet_rcv;
4284 skb_queue_purge(rb_queue);
4285 if (atomic_read(&po->mapped))
4286 pr_err("packet_mmap: vma is busy: %d\n",
4287 atomic_read(&po->mapped));
4289 mutex_unlock(&po->pg_vec_lock);
4291 spin_lock(&po->bind_lock);
4294 register_prot_hook(sk);
4296 spin_unlock(&po->bind_lock);
4297 if (closing && (po->tp_version > TPACKET_V2)) {
4298 /* Because we don't support block-based V3 on tx-ring */
4300 prb_shutdown_retire_blk_timer(po, rb_queue);
4305 free_pg_vec(pg_vec, order, req->tp_block_nr);
4310 static int packet_mmap(struct file *file, struct socket *sock,
4311 struct vm_area_struct *vma)
4313 struct sock *sk = sock->sk;
4314 struct packet_sock *po = pkt_sk(sk);
4315 unsigned long size, expected_size;
4316 struct packet_ring_buffer *rb;
4317 unsigned long start;
4324 mutex_lock(&po->pg_vec_lock);
4327 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4329 expected_size += rb->pg_vec_len
4335 if (expected_size == 0)
4338 size = vma->vm_end - vma->vm_start;
4339 if (size != expected_size)
4342 start = vma->vm_start;
4343 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4344 if (rb->pg_vec == NULL)
4347 for (i = 0; i < rb->pg_vec_len; i++) {
4349 void *kaddr = rb->pg_vec[i].buffer;
4352 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4353 page = pgv_to_page(kaddr);
4354 err = vm_insert_page(vma, start, page);
4363 atomic_inc(&po->mapped);
4364 vma->vm_ops = &packet_mmap_ops;
4368 mutex_unlock(&po->pg_vec_lock);
4372 static const struct proto_ops packet_ops_spkt = {
4373 .family = PF_PACKET,
4374 .owner = THIS_MODULE,
4375 .release = packet_release,
4376 .bind = packet_bind_spkt,
4377 .connect = sock_no_connect,
4378 .socketpair = sock_no_socketpair,
4379 .accept = sock_no_accept,
4380 .getname = packet_getname_spkt,
4381 .poll = datagram_poll,
4382 .ioctl = packet_ioctl,
4383 .listen = sock_no_listen,
4384 .shutdown = sock_no_shutdown,
4385 .setsockopt = sock_no_setsockopt,
4386 .getsockopt = sock_no_getsockopt,
4387 .sendmsg = packet_sendmsg_spkt,
4388 .recvmsg = packet_recvmsg,
4389 .mmap = sock_no_mmap,
4390 .sendpage = sock_no_sendpage,
4393 static const struct proto_ops packet_ops = {
4394 .family = PF_PACKET,
4395 .owner = THIS_MODULE,
4396 .release = packet_release,
4397 .bind = packet_bind,
4398 .connect = sock_no_connect,
4399 .socketpair = sock_no_socketpair,
4400 .accept = sock_no_accept,
4401 .getname = packet_getname,
4402 .poll = packet_poll,
4403 .ioctl = packet_ioctl,
4404 .listen = sock_no_listen,
4405 .shutdown = sock_no_shutdown,
4406 .setsockopt = packet_setsockopt,
4407 .getsockopt = packet_getsockopt,
4408 .sendmsg = packet_sendmsg,
4409 .recvmsg = packet_recvmsg,
4410 .mmap = packet_mmap,
4411 .sendpage = sock_no_sendpage,
4414 static const struct net_proto_family packet_family_ops = {
4415 .family = PF_PACKET,
4416 .create = packet_create,
4417 .owner = THIS_MODULE,
4420 static struct notifier_block packet_netdev_notifier = {
4421 .notifier_call = packet_notifier,
4424 #ifdef CONFIG_PROC_FS
4426 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4429 struct net *net = seq_file_net(seq);
4432 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4435 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4437 struct net *net = seq_file_net(seq);
4438 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4441 static void packet_seq_stop(struct seq_file *seq, void *v)
4447 static int packet_seq_show(struct seq_file *seq, void *v)
4449 if (v == SEQ_START_TOKEN)
4450 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4452 struct sock *s = sk_entry(v);
4453 const struct packet_sock *po = pkt_sk(s);
4456 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4458 atomic_read(&s->sk_refcnt),
4463 atomic_read(&s->sk_rmem_alloc),
4464 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4471 static const struct seq_operations packet_seq_ops = {
4472 .start = packet_seq_start,
4473 .next = packet_seq_next,
4474 .stop = packet_seq_stop,
4475 .show = packet_seq_show,
4478 static int packet_seq_open(struct inode *inode, struct file *file)
4480 return seq_open_net(inode, file, &packet_seq_ops,
4481 sizeof(struct seq_net_private));
4484 static const struct file_operations packet_seq_fops = {
4485 .owner = THIS_MODULE,
4486 .open = packet_seq_open,
4488 .llseek = seq_lseek,
4489 .release = seq_release_net,
4494 static int __net_init packet_net_init(struct net *net)
4496 mutex_init(&net->packet.sklist_lock);
4497 INIT_HLIST_HEAD(&net->packet.sklist);
4499 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4505 static void __net_exit packet_net_exit(struct net *net)
4507 remove_proc_entry("packet", net->proc_net);
4510 static struct pernet_operations packet_net_ops = {
4511 .init = packet_net_init,
4512 .exit = packet_net_exit,
4516 static void __exit packet_exit(void)
4518 unregister_netdevice_notifier(&packet_netdev_notifier);
4519 unregister_pernet_subsys(&packet_net_ops);
4520 sock_unregister(PF_PACKET);
4521 proto_unregister(&packet_proto);
4524 static int __init packet_init(void)
4526 int rc = proto_register(&packet_proto, 0);
4531 sock_register(&packet_family_ops);
4532 register_pernet_subsys(&packet_net_ops);
4533 register_netdevice_notifier(&packet_netdev_notifier);
4538 module_init(packet_init);
4539 module_exit(packet_exit);
4540 MODULE_LICENSE("GPL");
4541 MODULE_ALIAS_NETPROTO(PF_PACKET);