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;
2055 bool is_drop_n_account = false;
2057 if (skb->pkt_type == PACKET_LOOPBACK)
2060 sk = pt->af_packet_priv;
2063 if (!net_eq(dev_net(dev), sock_net(sk)))
2068 if (dev->header_ops) {
2069 /* The device has an explicit notion of ll header,
2070 * exported to higher levels.
2072 * Otherwise, the device hides details of its frame
2073 * structure, so that corresponding packet head is
2074 * never delivered to user.
2076 if (sk->sk_type != SOCK_DGRAM)
2077 skb_push(skb, skb->data - skb_mac_header(skb));
2078 else if (skb->pkt_type == PACKET_OUTGOING) {
2079 /* Special case: outgoing packets have ll header at head */
2080 skb_pull(skb, skb_network_offset(skb));
2086 res = run_filter(skb, sk, snaplen);
2088 goto drop_n_restore;
2092 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2095 if (skb_shared(skb)) {
2096 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2100 if (skb_head != skb->data) {
2101 skb->data = skb_head;
2108 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2110 sll = &PACKET_SKB_CB(skb)->sa.ll;
2111 sll->sll_hatype = dev->type;
2112 sll->sll_pkttype = skb->pkt_type;
2113 if (unlikely(po->origdev))
2114 sll->sll_ifindex = orig_dev->ifindex;
2116 sll->sll_ifindex = dev->ifindex;
2118 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2120 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2121 * Use their space for storing the original skb length.
2123 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2125 if (pskb_trim(skb, snaplen))
2128 skb_set_owner_r(skb, sk);
2132 /* drop conntrack reference */
2135 spin_lock(&sk->sk_receive_queue.lock);
2136 po->stats.stats1.tp_packets++;
2137 sock_skb_set_dropcount(sk, skb);
2138 __skb_queue_tail(&sk->sk_receive_queue, skb);
2139 spin_unlock(&sk->sk_receive_queue.lock);
2140 sk->sk_data_ready(sk);
2144 is_drop_n_account = true;
2145 spin_lock(&sk->sk_receive_queue.lock);
2146 po->stats.stats1.tp_drops++;
2147 atomic_inc(&sk->sk_drops);
2148 spin_unlock(&sk->sk_receive_queue.lock);
2151 if (skb_head != skb->data && skb_shared(skb)) {
2152 skb->data = skb_head;
2156 if (!is_drop_n_account)
2163 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2164 struct packet_type *pt, struct net_device *orig_dev)
2167 struct packet_sock *po;
2168 struct sockaddr_ll *sll;
2169 union tpacket_uhdr h;
2170 u8 *skb_head = skb->data;
2171 int skb_len = skb->len;
2172 unsigned int snaplen, res;
2173 unsigned long status = TP_STATUS_USER;
2174 unsigned short macoff, netoff, hdrlen;
2175 struct sk_buff *copy_skb = NULL;
2178 bool is_drop_n_account = false;
2180 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2181 * We may add members to them until current aligned size without forcing
2182 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2184 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2185 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2187 if (skb->pkt_type == PACKET_LOOPBACK)
2190 sk = pt->af_packet_priv;
2193 if (!net_eq(dev_net(dev), sock_net(sk)))
2196 if (dev->header_ops) {
2197 if (sk->sk_type != SOCK_DGRAM)
2198 skb_push(skb, skb->data - skb_mac_header(skb));
2199 else if (skb->pkt_type == PACKET_OUTGOING) {
2200 /* Special case: outgoing packets have ll header at head */
2201 skb_pull(skb, skb_network_offset(skb));
2207 res = run_filter(skb, sk, snaplen);
2209 goto drop_n_restore;
2211 if (skb->ip_summed == CHECKSUM_PARTIAL)
2212 status |= TP_STATUS_CSUMNOTREADY;
2213 else if (skb->pkt_type != PACKET_OUTGOING &&
2214 (skb->ip_summed == CHECKSUM_COMPLETE ||
2215 skb_csum_unnecessary(skb)))
2216 status |= TP_STATUS_CSUM_VALID;
2221 if (sk->sk_type == SOCK_DGRAM) {
2222 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2225 unsigned int maclen = skb_network_offset(skb);
2226 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2227 (maclen < 16 ? 16 : maclen)) +
2229 if (po->has_vnet_hdr)
2230 netoff += sizeof(struct virtio_net_hdr);
2231 macoff = netoff - maclen;
2233 if (po->tp_version <= TPACKET_V2) {
2234 if (macoff + snaplen > po->rx_ring.frame_size) {
2235 if (po->copy_thresh &&
2236 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2237 if (skb_shared(skb)) {
2238 copy_skb = skb_clone(skb, GFP_ATOMIC);
2240 copy_skb = skb_get(skb);
2241 skb_head = skb->data;
2244 skb_set_owner_r(copy_skb, sk);
2246 snaplen = po->rx_ring.frame_size - macoff;
2247 if ((int)snaplen < 0)
2250 } else if (unlikely(macoff + snaplen >
2251 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2254 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2255 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2256 snaplen, nval, macoff);
2258 if (unlikely((int)snaplen < 0)) {
2260 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2263 spin_lock(&sk->sk_receive_queue.lock);
2264 h.raw = packet_current_rx_frame(po, skb,
2265 TP_STATUS_KERNEL, (macoff+snaplen));
2267 goto drop_n_account;
2268 if (po->tp_version <= TPACKET_V2) {
2269 packet_increment_rx_head(po, &po->rx_ring);
2271 * LOSING will be reported till you read the stats,
2272 * because it's COR - Clear On Read.
2273 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2276 if (po->stats.stats1.tp_drops)
2277 status |= TP_STATUS_LOSING;
2279 po->stats.stats1.tp_packets++;
2281 status |= TP_STATUS_COPY;
2282 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2284 spin_unlock(&sk->sk_receive_queue.lock);
2286 if (po->has_vnet_hdr) {
2287 if (__packet_rcv_vnet(skb, h.raw + macoff -
2288 sizeof(struct virtio_net_hdr))) {
2289 spin_lock(&sk->sk_receive_queue.lock);
2290 goto drop_n_account;
2294 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2296 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2297 getnstimeofday(&ts);
2299 status |= ts_status;
2301 switch (po->tp_version) {
2303 h.h1->tp_len = skb->len;
2304 h.h1->tp_snaplen = snaplen;
2305 h.h1->tp_mac = macoff;
2306 h.h1->tp_net = netoff;
2307 h.h1->tp_sec = ts.tv_sec;
2308 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2309 hdrlen = sizeof(*h.h1);
2312 h.h2->tp_len = skb->len;
2313 h.h2->tp_snaplen = snaplen;
2314 h.h2->tp_mac = macoff;
2315 h.h2->tp_net = netoff;
2316 h.h2->tp_sec = ts.tv_sec;
2317 h.h2->tp_nsec = ts.tv_nsec;
2318 if (skb_vlan_tag_present(skb)) {
2319 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2320 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2321 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2323 h.h2->tp_vlan_tci = 0;
2324 h.h2->tp_vlan_tpid = 0;
2326 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2327 hdrlen = sizeof(*h.h2);
2330 /* tp_nxt_offset,vlan are already populated above.
2331 * So DONT clear those fields here
2333 h.h3->tp_status |= status;
2334 h.h3->tp_len = skb->len;
2335 h.h3->tp_snaplen = snaplen;
2336 h.h3->tp_mac = macoff;
2337 h.h3->tp_net = netoff;
2338 h.h3->tp_sec = ts.tv_sec;
2339 h.h3->tp_nsec = ts.tv_nsec;
2340 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2341 hdrlen = sizeof(*h.h3);
2347 sll = h.raw + TPACKET_ALIGN(hdrlen);
2348 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2349 sll->sll_family = AF_PACKET;
2350 sll->sll_hatype = dev->type;
2351 sll->sll_protocol = skb->protocol;
2352 sll->sll_pkttype = skb->pkt_type;
2353 if (unlikely(po->origdev))
2354 sll->sll_ifindex = orig_dev->ifindex;
2356 sll->sll_ifindex = dev->ifindex;
2360 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2361 if (po->tp_version <= TPACKET_V2) {
2364 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2367 for (start = h.raw; start < end; start += PAGE_SIZE)
2368 flush_dcache_page(pgv_to_page(start));
2373 if (po->tp_version <= TPACKET_V2) {
2374 __packet_set_status(po, h.raw, status);
2375 sk->sk_data_ready(sk);
2377 prb_clear_blk_fill_status(&po->rx_ring);
2381 if (skb_head != skb->data && skb_shared(skb)) {
2382 skb->data = skb_head;
2386 if (!is_drop_n_account)
2393 is_drop_n_account = true;
2394 po->stats.stats1.tp_drops++;
2395 spin_unlock(&sk->sk_receive_queue.lock);
2397 sk->sk_data_ready(sk);
2398 kfree_skb(copy_skb);
2399 goto drop_n_restore;
2402 static void tpacket_destruct_skb(struct sk_buff *skb)
2404 struct packet_sock *po = pkt_sk(skb->sk);
2406 if (likely(po->tx_ring.pg_vec)) {
2410 ph = skb_shinfo(skb)->destructor_arg;
2411 packet_dec_pending(&po->tx_ring);
2413 ts = __packet_set_timestamp(po, ph, skb);
2414 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2420 static void tpacket_set_protocol(const struct net_device *dev,
2421 struct sk_buff *skb)
2423 if (dev->type == ARPHRD_ETHER) {
2424 skb_reset_mac_header(skb);
2425 skb->protocol = eth_hdr(skb)->h_proto;
2429 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2431 unsigned short gso_type = 0;
2433 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2434 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2435 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2436 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2437 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2438 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2439 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2441 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2444 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2445 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2446 case VIRTIO_NET_HDR_GSO_TCPV4:
2447 gso_type = SKB_GSO_TCPV4;
2449 case VIRTIO_NET_HDR_GSO_TCPV6:
2450 gso_type = SKB_GSO_TCPV6;
2452 case VIRTIO_NET_HDR_GSO_UDP:
2453 gso_type = SKB_GSO_UDP;
2459 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
2460 gso_type |= SKB_GSO_TCP_ECN;
2462 if (vnet_hdr->gso_size == 0)
2466 vnet_hdr->gso_type = gso_type; /* changes type, temporary storage */
2470 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2471 struct virtio_net_hdr *vnet_hdr)
2475 if (*len < sizeof(*vnet_hdr))
2477 *len -= sizeof(*vnet_hdr);
2479 n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter);
2480 if (n != sizeof(*vnet_hdr))
2483 return __packet_snd_vnet_parse(vnet_hdr, *len);
2486 static int packet_snd_vnet_gso(struct sk_buff *skb,
2487 struct virtio_net_hdr *vnet_hdr)
2489 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2490 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start);
2491 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset);
2493 if (!skb_partial_csum_set(skb, s, o))
2497 skb_shinfo(skb)->gso_size =
2498 __virtio16_to_cpu(vio_le(), vnet_hdr->gso_size);
2499 skb_shinfo(skb)->gso_type = vnet_hdr->gso_type;
2501 /* Header must be checked, and gso_segs computed. */
2502 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2503 skb_shinfo(skb)->gso_segs = 0;
2507 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2508 void *frame, struct net_device *dev, void *data, int tp_len,
2509 __be16 proto, unsigned char *addr, int hlen, int copylen,
2510 const struct sockcm_cookie *sockc)
2512 union tpacket_uhdr ph;
2513 int to_write, offset, len, nr_frags, len_max;
2514 struct socket *sock = po->sk.sk_socket;
2520 skb->protocol = proto;
2522 skb->priority = po->sk.sk_priority;
2523 skb->mark = po->sk.sk_mark;
2524 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2525 skb_shinfo(skb)->destructor_arg = ph.raw;
2527 skb_reserve(skb, hlen);
2528 skb_reset_network_header(skb);
2532 if (sock->type == SOCK_DGRAM) {
2533 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2535 if (unlikely(err < 0))
2537 } else if (copylen) {
2538 int hdrlen = min_t(int, copylen, tp_len);
2540 skb_push(skb, dev->hard_header_len);
2541 skb_put(skb, copylen - dev->hard_header_len);
2542 err = skb_store_bits(skb, 0, data, hdrlen);
2545 if (!dev_validate_header(dev, skb->data, hdrlen))
2548 tpacket_set_protocol(dev, skb);
2554 offset = offset_in_page(data);
2555 len_max = PAGE_SIZE - offset;
2556 len = ((to_write > len_max) ? len_max : to_write);
2558 skb->data_len = to_write;
2559 skb->len += to_write;
2560 skb->truesize += to_write;
2561 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2563 while (likely(to_write)) {
2564 nr_frags = skb_shinfo(skb)->nr_frags;
2566 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2567 pr_err("Packet exceed the number of skb frags(%lu)\n",
2572 page = pgv_to_page(data);
2574 flush_dcache_page(page);
2576 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2579 len_max = PAGE_SIZE;
2580 len = ((to_write > len_max) ? len_max : to_write);
2583 skb_probe_transport_header(skb, 0);
2588 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2589 int size_max, void **data)
2591 union tpacket_uhdr ph;
2596 switch (po->tp_version) {
2598 tp_len = ph.h2->tp_len;
2601 tp_len = ph.h1->tp_len;
2604 if (unlikely(tp_len > size_max)) {
2605 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2609 if (unlikely(po->tp_tx_has_off)) {
2610 int off_min, off_max;
2612 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2613 off_max = po->tx_ring.frame_size - tp_len;
2614 if (po->sk.sk_type == SOCK_DGRAM) {
2615 switch (po->tp_version) {
2617 off = ph.h2->tp_net;
2620 off = ph.h1->tp_net;
2624 switch (po->tp_version) {
2626 off = ph.h2->tp_mac;
2629 off = ph.h1->tp_mac;
2633 if (unlikely((off < off_min) || (off_max < off)))
2636 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2639 *data = frame + off;
2643 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2645 struct sk_buff *skb;
2646 struct net_device *dev;
2647 struct virtio_net_hdr *vnet_hdr = NULL;
2648 struct sockcm_cookie sockc;
2650 int err, reserve = 0;
2652 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2653 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2654 int tp_len, size_max;
2655 unsigned char *addr;
2658 int status = TP_STATUS_AVAILABLE;
2659 int hlen, tlen, copylen = 0;
2661 mutex_lock(&po->pg_vec_lock);
2663 if (likely(saddr == NULL)) {
2664 dev = packet_cached_dev_get(po);
2669 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2671 if (msg->msg_namelen < (saddr->sll_halen
2672 + offsetof(struct sockaddr_ll,
2675 proto = saddr->sll_protocol;
2676 addr = saddr->sll_addr;
2677 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2681 if (msg->msg_controllen) {
2682 err = sock_cmsg_send(&po->sk, msg, &sockc);
2688 if (unlikely(dev == NULL))
2691 if (unlikely(!(dev->flags & IFF_UP)))
2694 if (po->sk.sk_socket->type == SOCK_RAW)
2695 reserve = dev->hard_header_len;
2696 size_max = po->tx_ring.frame_size
2697 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2699 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2700 size_max = dev->mtu + reserve + VLAN_HLEN;
2703 ph = packet_current_frame(po, &po->tx_ring,
2704 TP_STATUS_SEND_REQUEST);
2705 if (unlikely(ph == NULL)) {
2706 if (need_wait && need_resched())
2712 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2716 status = TP_STATUS_SEND_REQUEST;
2717 hlen = LL_RESERVED_SPACE(dev);
2718 tlen = dev->needed_tailroom;
2719 if (po->has_vnet_hdr) {
2721 data += sizeof(*vnet_hdr);
2722 tp_len -= sizeof(*vnet_hdr);
2724 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2728 copylen = __virtio16_to_cpu(vio_le(),
2731 copylen = max_t(int, copylen, dev->hard_header_len);
2732 skb = sock_alloc_send_skb(&po->sk,
2733 hlen + tlen + sizeof(struct sockaddr_ll) +
2734 (copylen - dev->hard_header_len),
2737 if (unlikely(skb == NULL)) {
2738 /* we assume the socket was initially writeable ... */
2739 if (likely(len_sum > 0))
2743 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2744 addr, hlen, copylen, &sockc);
2745 if (likely(tp_len >= 0) &&
2746 tp_len > dev->mtu + reserve &&
2747 !po->has_vnet_hdr &&
2748 !packet_extra_vlan_len_allowed(dev, skb))
2751 if (unlikely(tp_len < 0)) {
2754 __packet_set_status(po, ph,
2755 TP_STATUS_AVAILABLE);
2756 packet_increment_head(&po->tx_ring);
2760 status = TP_STATUS_WRONG_FORMAT;
2766 if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) {
2771 packet_pick_tx_queue(dev, skb);
2773 skb->destructor = tpacket_destruct_skb;
2774 __packet_set_status(po, ph, TP_STATUS_SENDING);
2775 packet_inc_pending(&po->tx_ring);
2777 status = TP_STATUS_SEND_REQUEST;
2778 err = po->xmit(skb);
2779 if (unlikely(err > 0)) {
2780 err = net_xmit_errno(err);
2781 if (err && __packet_get_status(po, ph) ==
2782 TP_STATUS_AVAILABLE) {
2783 /* skb was destructed already */
2788 * skb was dropped but not destructed yet;
2789 * let's treat it like congestion or err < 0
2793 packet_increment_head(&po->tx_ring);
2795 } while (likely((ph != NULL) ||
2796 /* Note: packet_read_pending() might be slow if we have
2797 * to call it as it's per_cpu variable, but in fast-path
2798 * we already short-circuit the loop with the first
2799 * condition, and luckily don't have to go that path
2802 (need_wait && packet_read_pending(&po->tx_ring))));
2808 __packet_set_status(po, ph, status);
2813 mutex_unlock(&po->pg_vec_lock);
2817 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2818 size_t reserve, size_t len,
2819 size_t linear, int noblock,
2822 struct sk_buff *skb;
2824 /* Under a page? Don't bother with paged skb. */
2825 if (prepad + len < PAGE_SIZE || !linear)
2828 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2833 skb_reserve(skb, reserve);
2834 skb_put(skb, linear);
2835 skb->data_len = len - linear;
2836 skb->len += len - linear;
2841 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2843 struct sock *sk = sock->sk;
2844 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2845 struct sk_buff *skb;
2846 struct net_device *dev;
2848 unsigned char *addr;
2849 int err, reserve = 0;
2850 struct sockcm_cookie sockc;
2851 struct virtio_net_hdr vnet_hdr = { 0 };
2853 struct packet_sock *po = pkt_sk(sk);
2858 * Get and verify the address.
2861 if (likely(saddr == NULL)) {
2862 dev = packet_cached_dev_get(po);
2867 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2869 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2871 proto = saddr->sll_protocol;
2872 addr = saddr->sll_addr;
2873 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2877 if (unlikely(dev == NULL))
2880 if (unlikely(!(dev->flags & IFF_UP)))
2884 sockc.mark = sk->sk_mark;
2885 if (msg->msg_controllen) {
2886 err = sock_cmsg_send(sk, msg, &sockc);
2891 if (sock->type == SOCK_RAW)
2892 reserve = dev->hard_header_len;
2893 if (po->has_vnet_hdr) {
2894 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2899 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2900 if (!netif_supports_nofcs(dev)) {
2901 err = -EPROTONOSUPPORT;
2904 extra_len = 4; /* We're doing our own CRC */
2908 if (!vnet_hdr.gso_type &&
2909 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2913 hlen = LL_RESERVED_SPACE(dev);
2914 tlen = dev->needed_tailroom;
2915 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2916 __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len),
2917 msg->msg_flags & MSG_DONTWAIT, &err);
2921 skb_set_network_header(skb, reserve);
2924 if (sock->type == SOCK_DGRAM) {
2925 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2926 if (unlikely(offset < 0))
2930 /* Returns -EFAULT on error */
2931 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2935 if (sock->type == SOCK_RAW &&
2936 !dev_validate_header(dev, skb->data, len)) {
2941 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2943 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2944 !packet_extra_vlan_len_allowed(dev, skb)) {
2949 skb->protocol = proto;
2951 skb->priority = sk->sk_priority;
2952 skb->mark = sockc.mark;
2954 packet_pick_tx_queue(dev, skb);
2956 if (po->has_vnet_hdr) {
2957 err = packet_snd_vnet_gso(skb, &vnet_hdr);
2960 len += sizeof(vnet_hdr);
2963 skb_probe_transport_header(skb, reserve);
2965 if (unlikely(extra_len == 4))
2968 err = po->xmit(skb);
2969 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2985 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2987 struct sock *sk = sock->sk;
2988 struct packet_sock *po = pkt_sk(sk);
2990 if (po->tx_ring.pg_vec)
2991 return tpacket_snd(po, msg);
2993 return packet_snd(sock, msg, len);
2997 * Close a PACKET socket. This is fairly simple. We immediately go
2998 * to 'closed' state and remove our protocol entry in the device list.
3001 static int packet_release(struct socket *sock)
3003 struct sock *sk = sock->sk;
3004 struct packet_sock *po;
3006 union tpacket_req_u req_u;
3014 mutex_lock(&net->packet.sklist_lock);
3015 sk_del_node_init_rcu(sk);
3016 mutex_unlock(&net->packet.sklist_lock);
3019 sock_prot_inuse_add(net, sk->sk_prot, -1);
3022 spin_lock(&po->bind_lock);
3023 unregister_prot_hook(sk, false);
3024 packet_cached_dev_reset(po);
3026 if (po->prot_hook.dev) {
3027 dev_put(po->prot_hook.dev);
3028 po->prot_hook.dev = NULL;
3030 spin_unlock(&po->bind_lock);
3032 packet_flush_mclist(sk);
3034 if (po->rx_ring.pg_vec) {
3035 memset(&req_u, 0, sizeof(req_u));
3036 packet_set_ring(sk, &req_u, 1, 0);
3039 if (po->tx_ring.pg_vec) {
3040 memset(&req_u, 0, sizeof(req_u));
3041 packet_set_ring(sk, &req_u, 1, 1);
3048 * Now the socket is dead. No more input will appear.
3055 skb_queue_purge(&sk->sk_receive_queue);
3056 packet_free_pending(po);
3057 sk_refcnt_debug_release(sk);
3064 * Attach a packet hook.
3067 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3070 struct packet_sock *po = pkt_sk(sk);
3071 struct net_device *dev_curr;
3074 struct net_device *dev = NULL;
3076 bool unlisted = false;
3082 spin_lock(&po->bind_lock);
3086 dev = dev_get_by_name_rcu(sock_net(sk), name);
3091 } else if (ifindex) {
3092 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3102 proto_curr = po->prot_hook.type;
3103 dev_curr = po->prot_hook.dev;
3105 need_rehook = proto_curr != proto || dev_curr != dev;
3110 __unregister_prot_hook(sk, true);
3112 dev_curr = po->prot_hook.dev;
3114 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3119 po->prot_hook.type = proto;
3121 if (unlikely(unlisted)) {
3123 po->prot_hook.dev = NULL;
3125 packet_cached_dev_reset(po);
3127 po->prot_hook.dev = dev;
3128 po->ifindex = dev ? dev->ifindex : 0;
3129 packet_cached_dev_assign(po, dev);
3135 if (proto == 0 || !need_rehook)
3138 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3139 register_prot_hook(sk);
3141 sk->sk_err = ENETDOWN;
3142 if (!sock_flag(sk, SOCK_DEAD))
3143 sk->sk_error_report(sk);
3148 spin_unlock(&po->bind_lock);
3154 * Bind a packet socket to a device
3157 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3160 struct sock *sk = sock->sk;
3167 if (addr_len != sizeof(struct sockaddr))
3169 strlcpy(name, uaddr->sa_data, sizeof(name));
3171 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3174 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3176 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3177 struct sock *sk = sock->sk;
3183 if (addr_len < sizeof(struct sockaddr_ll))
3185 if (sll->sll_family != AF_PACKET)
3188 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3189 sll->sll_protocol ? : pkt_sk(sk)->num);
3192 static struct proto packet_proto = {
3194 .owner = THIS_MODULE,
3195 .obj_size = sizeof(struct packet_sock),
3199 * Create a packet of type SOCK_PACKET.
3202 static int packet_create(struct net *net, struct socket *sock, int protocol,
3206 struct packet_sock *po;
3207 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3210 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3212 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3213 sock->type != SOCK_PACKET)
3214 return -ESOCKTNOSUPPORT;
3216 sock->state = SS_UNCONNECTED;
3219 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3223 sock->ops = &packet_ops;
3224 if (sock->type == SOCK_PACKET)
3225 sock->ops = &packet_ops_spkt;
3227 sock_init_data(sock, sk);
3230 sk->sk_family = PF_PACKET;
3232 po->xmit = dev_queue_xmit;
3234 err = packet_alloc_pending(po);
3238 packet_cached_dev_reset(po);
3240 sk->sk_destruct = packet_sock_destruct;
3241 sk_refcnt_debug_inc(sk);
3244 * Attach a protocol block
3247 spin_lock_init(&po->bind_lock);
3248 mutex_init(&po->pg_vec_lock);
3249 po->rollover = NULL;
3250 po->prot_hook.func = packet_rcv;
3252 if (sock->type == SOCK_PACKET)
3253 po->prot_hook.func = packet_rcv_spkt;
3255 po->prot_hook.af_packet_priv = sk;
3258 po->prot_hook.type = proto;
3259 register_prot_hook(sk);
3262 mutex_lock(&net->packet.sklist_lock);
3263 sk_add_node_rcu(sk, &net->packet.sklist);
3264 mutex_unlock(&net->packet.sklist_lock);
3267 sock_prot_inuse_add(net, &packet_proto, 1);
3278 * Pull a packet from our receive queue and hand it to the user.
3279 * If necessary we block.
3282 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3285 struct sock *sk = sock->sk;
3286 struct sk_buff *skb;
3288 int vnet_hdr_len = 0;
3289 unsigned int origlen = 0;
3292 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3296 /* What error should we return now? EUNATTACH? */
3297 if (pkt_sk(sk)->ifindex < 0)
3301 if (flags & MSG_ERRQUEUE) {
3302 err = sock_recv_errqueue(sk, msg, len,
3303 SOL_PACKET, PACKET_TX_TIMESTAMP);
3308 * Call the generic datagram receiver. This handles all sorts
3309 * of horrible races and re-entrancy so we can forget about it
3310 * in the protocol layers.
3312 * Now it will return ENETDOWN, if device have just gone down,
3313 * but then it will block.
3316 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3319 * An error occurred so return it. Because skb_recv_datagram()
3320 * handles the blocking we don't see and worry about blocking
3327 if (pkt_sk(sk)->pressure)
3328 packet_rcv_has_room(pkt_sk(sk), NULL);
3330 if (pkt_sk(sk)->has_vnet_hdr) {
3331 err = packet_rcv_vnet(msg, skb, &len);
3334 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3337 /* You lose any data beyond the buffer you gave. If it worries
3338 * a user program they can ask the device for its MTU
3344 msg->msg_flags |= MSG_TRUNC;
3347 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3351 if (sock->type != SOCK_PACKET) {
3352 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3354 /* Original length was stored in sockaddr_ll fields */
3355 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3356 sll->sll_family = AF_PACKET;
3357 sll->sll_protocol = skb->protocol;
3360 sock_recv_ts_and_drops(msg, sk, skb);
3362 if (msg->msg_name) {
3363 /* If the address length field is there to be filled
3364 * in, we fill it in now.
3366 if (sock->type == SOCK_PACKET) {
3367 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3368 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3370 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3372 msg->msg_namelen = sll->sll_halen +
3373 offsetof(struct sockaddr_ll, sll_addr);
3375 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3379 if (pkt_sk(sk)->auxdata) {
3380 struct tpacket_auxdata aux;
3382 aux.tp_status = TP_STATUS_USER;
3383 if (skb->ip_summed == CHECKSUM_PARTIAL)
3384 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3385 else if (skb->pkt_type != PACKET_OUTGOING &&
3386 (skb->ip_summed == CHECKSUM_COMPLETE ||
3387 skb_csum_unnecessary(skb)))
3388 aux.tp_status |= TP_STATUS_CSUM_VALID;
3390 aux.tp_len = origlen;
3391 aux.tp_snaplen = skb->len;
3393 aux.tp_net = skb_network_offset(skb);
3394 if (skb_vlan_tag_present(skb)) {
3395 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3396 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3397 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3399 aux.tp_vlan_tci = 0;
3400 aux.tp_vlan_tpid = 0;
3402 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3406 * Free or return the buffer as appropriate. Again this
3407 * hides all the races and re-entrancy issues from us.
3409 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3412 skb_free_datagram(sk, skb);
3417 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3418 int *uaddr_len, int peer)
3420 struct net_device *dev;
3421 struct sock *sk = sock->sk;
3426 uaddr->sa_family = AF_PACKET;
3427 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3429 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3431 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3433 *uaddr_len = sizeof(*uaddr);
3438 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3439 int *uaddr_len, int peer)
3441 struct net_device *dev;
3442 struct sock *sk = sock->sk;
3443 struct packet_sock *po = pkt_sk(sk);
3444 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3449 sll->sll_family = AF_PACKET;
3450 sll->sll_ifindex = po->ifindex;
3451 sll->sll_protocol = po->num;
3452 sll->sll_pkttype = 0;
3454 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3456 sll->sll_hatype = dev->type;
3457 sll->sll_halen = dev->addr_len;
3458 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3460 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3464 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3469 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3473 case PACKET_MR_MULTICAST:
3474 if (i->alen != dev->addr_len)
3477 return dev_mc_add(dev, i->addr);
3479 return dev_mc_del(dev, i->addr);
3481 case PACKET_MR_PROMISC:
3482 return dev_set_promiscuity(dev, what);
3483 case PACKET_MR_ALLMULTI:
3484 return dev_set_allmulti(dev, what);
3485 case PACKET_MR_UNICAST:
3486 if (i->alen != dev->addr_len)
3489 return dev_uc_add(dev, i->addr);
3491 return dev_uc_del(dev, i->addr);
3499 static void packet_dev_mclist_delete(struct net_device *dev,
3500 struct packet_mclist **mlp)
3502 struct packet_mclist *ml;
3504 while ((ml = *mlp) != NULL) {
3505 if (ml->ifindex == dev->ifindex) {
3506 packet_dev_mc(dev, ml, -1);
3514 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3516 struct packet_sock *po = pkt_sk(sk);
3517 struct packet_mclist *ml, *i;
3518 struct net_device *dev;
3524 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3529 if (mreq->mr_alen > dev->addr_len)
3533 i = kmalloc(sizeof(*i), GFP_KERNEL);
3538 for (ml = po->mclist; ml; ml = ml->next) {
3539 if (ml->ifindex == mreq->mr_ifindex &&
3540 ml->type == mreq->mr_type &&
3541 ml->alen == mreq->mr_alen &&
3542 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3544 /* Free the new element ... */
3550 i->type = mreq->mr_type;
3551 i->ifindex = mreq->mr_ifindex;
3552 i->alen = mreq->mr_alen;
3553 memcpy(i->addr, mreq->mr_address, i->alen);
3554 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3556 i->next = po->mclist;
3558 err = packet_dev_mc(dev, i, 1);
3560 po->mclist = i->next;
3569 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3571 struct packet_mclist *ml, **mlp;
3575 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3576 if (ml->ifindex == mreq->mr_ifindex &&
3577 ml->type == mreq->mr_type &&
3578 ml->alen == mreq->mr_alen &&
3579 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3580 if (--ml->count == 0) {
3581 struct net_device *dev;
3583 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3585 packet_dev_mc(dev, ml, -1);
3595 static void packet_flush_mclist(struct sock *sk)
3597 struct packet_sock *po = pkt_sk(sk);
3598 struct packet_mclist *ml;
3604 while ((ml = po->mclist) != NULL) {
3605 struct net_device *dev;
3607 po->mclist = ml->next;
3608 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3610 packet_dev_mc(dev, ml, -1);
3617 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3619 struct sock *sk = sock->sk;
3620 struct packet_sock *po = pkt_sk(sk);
3623 if (level != SOL_PACKET)
3624 return -ENOPROTOOPT;
3627 case PACKET_ADD_MEMBERSHIP:
3628 case PACKET_DROP_MEMBERSHIP:
3630 struct packet_mreq_max mreq;
3632 memset(&mreq, 0, sizeof(mreq));
3633 if (len < sizeof(struct packet_mreq))
3635 if (len > sizeof(mreq))
3637 if (copy_from_user(&mreq, optval, len))
3639 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3641 if (optname == PACKET_ADD_MEMBERSHIP)
3642 ret = packet_mc_add(sk, &mreq);
3644 ret = packet_mc_drop(sk, &mreq);
3648 case PACKET_RX_RING:
3649 case PACKET_TX_RING:
3651 union tpacket_req_u req_u;
3654 switch (po->tp_version) {
3657 len = sizeof(req_u.req);
3661 len = sizeof(req_u.req3);
3666 if (copy_from_user(&req_u.req, optval, len))
3668 return packet_set_ring(sk, &req_u, 0,
3669 optname == PACKET_TX_RING);
3671 case PACKET_COPY_THRESH:
3675 if (optlen != sizeof(val))
3677 if (copy_from_user(&val, optval, sizeof(val)))
3680 pkt_sk(sk)->copy_thresh = val;
3683 case PACKET_VERSION:
3687 if (optlen != sizeof(val))
3689 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3691 if (copy_from_user(&val, optval, sizeof(val)))
3697 po->tp_version = val;
3703 case PACKET_RESERVE:
3707 if (optlen != sizeof(val))
3709 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3711 if (copy_from_user(&val, optval, sizeof(val)))
3713 po->tp_reserve = val;
3720 if (optlen != sizeof(val))
3722 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3724 if (copy_from_user(&val, optval, sizeof(val)))
3726 po->tp_loss = !!val;
3729 case PACKET_AUXDATA:
3733 if (optlen < sizeof(val))
3735 if (copy_from_user(&val, optval, sizeof(val)))
3738 po->auxdata = !!val;
3741 case PACKET_ORIGDEV:
3745 if (optlen < sizeof(val))
3747 if (copy_from_user(&val, optval, sizeof(val)))
3750 po->origdev = !!val;
3753 case PACKET_VNET_HDR:
3757 if (sock->type != SOCK_RAW)
3759 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3761 if (optlen < sizeof(val))
3763 if (copy_from_user(&val, optval, sizeof(val)))
3766 po->has_vnet_hdr = !!val;
3769 case PACKET_TIMESTAMP:
3773 if (optlen != sizeof(val))
3775 if (copy_from_user(&val, optval, sizeof(val)))
3778 po->tp_tstamp = val;
3785 if (optlen != sizeof(val))
3787 if (copy_from_user(&val, optval, sizeof(val)))
3790 return fanout_add(sk, val & 0xffff, val >> 16);
3792 case PACKET_FANOUT_DATA:
3797 return fanout_set_data(po, optval, optlen);
3799 case PACKET_TX_HAS_OFF:
3803 if (optlen != sizeof(val))
3805 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3807 if (copy_from_user(&val, optval, sizeof(val)))
3809 po->tp_tx_has_off = !!val;
3812 case PACKET_QDISC_BYPASS:
3816 if (optlen != sizeof(val))
3818 if (copy_from_user(&val, optval, sizeof(val)))
3821 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3825 return -ENOPROTOOPT;
3829 static int packet_getsockopt(struct socket *sock, int level, int optname,
3830 char __user *optval, int __user *optlen)
3833 int val, lv = sizeof(val);
3834 struct sock *sk = sock->sk;
3835 struct packet_sock *po = pkt_sk(sk);
3837 union tpacket_stats_u st;
3838 struct tpacket_rollover_stats rstats;
3840 if (level != SOL_PACKET)
3841 return -ENOPROTOOPT;
3843 if (get_user(len, optlen))
3850 case PACKET_STATISTICS:
3851 spin_lock_bh(&sk->sk_receive_queue.lock);
3852 memcpy(&st, &po->stats, sizeof(st));
3853 memset(&po->stats, 0, sizeof(po->stats));
3854 spin_unlock_bh(&sk->sk_receive_queue.lock);
3856 if (po->tp_version == TPACKET_V3) {
3857 lv = sizeof(struct tpacket_stats_v3);
3858 st.stats3.tp_packets += st.stats3.tp_drops;
3861 lv = sizeof(struct tpacket_stats);
3862 st.stats1.tp_packets += st.stats1.tp_drops;
3867 case PACKET_AUXDATA:
3870 case PACKET_ORIGDEV:
3873 case PACKET_VNET_HDR:
3874 val = po->has_vnet_hdr;
3876 case PACKET_VERSION:
3877 val = po->tp_version;
3880 if (len > sizeof(int))
3882 if (copy_from_user(&val, optval, len))
3886 val = sizeof(struct tpacket_hdr);
3889 val = sizeof(struct tpacket2_hdr);
3892 val = sizeof(struct tpacket3_hdr);
3898 case PACKET_RESERVE:
3899 val = po->tp_reserve;
3904 case PACKET_TIMESTAMP:
3905 val = po->tp_tstamp;
3909 ((u32)po->fanout->id |
3910 ((u32)po->fanout->type << 16) |
3911 ((u32)po->fanout->flags << 24)) :
3914 case PACKET_ROLLOVER_STATS:
3917 rstats.tp_all = atomic_long_read(&po->rollover->num);
3918 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3919 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3921 lv = sizeof(rstats);
3923 case PACKET_TX_HAS_OFF:
3924 val = po->tp_tx_has_off;
3926 case PACKET_QDISC_BYPASS:
3927 val = packet_use_direct_xmit(po);
3930 return -ENOPROTOOPT;
3935 if (put_user(len, optlen))
3937 if (copy_to_user(optval, data, len))
3943 static int packet_notifier(struct notifier_block *this,
3944 unsigned long msg, void *ptr)
3947 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3948 struct net *net = dev_net(dev);
3951 sk_for_each_rcu(sk, &net->packet.sklist) {
3952 struct packet_sock *po = pkt_sk(sk);
3955 case NETDEV_UNREGISTER:
3957 packet_dev_mclist_delete(dev, &po->mclist);
3961 if (dev->ifindex == po->ifindex) {
3962 spin_lock(&po->bind_lock);
3964 __unregister_prot_hook(sk, false);
3965 sk->sk_err = ENETDOWN;
3966 if (!sock_flag(sk, SOCK_DEAD))
3967 sk->sk_error_report(sk);
3969 if (msg == NETDEV_UNREGISTER) {
3970 packet_cached_dev_reset(po);
3972 if (po->prot_hook.dev)
3973 dev_put(po->prot_hook.dev);
3974 po->prot_hook.dev = NULL;
3976 spin_unlock(&po->bind_lock);
3980 if (dev->ifindex == po->ifindex) {
3981 spin_lock(&po->bind_lock);
3983 register_prot_hook(sk);
3984 spin_unlock(&po->bind_lock);
3994 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3997 struct sock *sk = sock->sk;
4002 int amount = sk_wmem_alloc_get(sk);
4004 return put_user(amount, (int __user *)arg);
4008 struct sk_buff *skb;
4011 spin_lock_bh(&sk->sk_receive_queue.lock);
4012 skb = skb_peek(&sk->sk_receive_queue);
4015 spin_unlock_bh(&sk->sk_receive_queue.lock);
4016 return put_user(amount, (int __user *)arg);
4019 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4021 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4031 case SIOCGIFBRDADDR:
4032 case SIOCSIFBRDADDR:
4033 case SIOCGIFNETMASK:
4034 case SIOCSIFNETMASK:
4035 case SIOCGIFDSTADDR:
4036 case SIOCSIFDSTADDR:
4038 return inet_dgram_ops.ioctl(sock, cmd, arg);
4042 return -ENOIOCTLCMD;
4047 static unsigned int packet_poll(struct file *file, struct socket *sock,
4050 struct sock *sk = sock->sk;
4051 struct packet_sock *po = pkt_sk(sk);
4052 unsigned int mask = datagram_poll(file, sock, wait);
4054 spin_lock_bh(&sk->sk_receive_queue.lock);
4055 if (po->rx_ring.pg_vec) {
4056 if (!packet_previous_rx_frame(po, &po->rx_ring,
4058 mask |= POLLIN | POLLRDNORM;
4060 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4062 spin_unlock_bh(&sk->sk_receive_queue.lock);
4063 spin_lock_bh(&sk->sk_write_queue.lock);
4064 if (po->tx_ring.pg_vec) {
4065 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4066 mask |= POLLOUT | POLLWRNORM;
4068 spin_unlock_bh(&sk->sk_write_queue.lock);
4073 /* Dirty? Well, I still did not learn better way to account
4077 static void packet_mm_open(struct vm_area_struct *vma)
4079 struct file *file = vma->vm_file;
4080 struct socket *sock = file->private_data;
4081 struct sock *sk = sock->sk;
4084 atomic_inc(&pkt_sk(sk)->mapped);
4087 static void packet_mm_close(struct vm_area_struct *vma)
4089 struct file *file = vma->vm_file;
4090 struct socket *sock = file->private_data;
4091 struct sock *sk = sock->sk;
4094 atomic_dec(&pkt_sk(sk)->mapped);
4097 static const struct vm_operations_struct packet_mmap_ops = {
4098 .open = packet_mm_open,
4099 .close = packet_mm_close,
4102 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4107 for (i = 0; i < len; i++) {
4108 if (likely(pg_vec[i].buffer)) {
4109 if (is_vmalloc_addr(pg_vec[i].buffer))
4110 vfree(pg_vec[i].buffer);
4112 free_pages((unsigned long)pg_vec[i].buffer,
4114 pg_vec[i].buffer = NULL;
4120 static char *alloc_one_pg_vec_page(unsigned long order)
4123 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4124 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4126 buffer = (char *) __get_free_pages(gfp_flags, order);
4130 /* __get_free_pages failed, fall back to vmalloc */
4131 buffer = vzalloc((1 << order) * PAGE_SIZE);
4135 /* vmalloc failed, lets dig into swap here */
4136 gfp_flags &= ~__GFP_NORETRY;
4137 buffer = (char *) __get_free_pages(gfp_flags, order);
4141 /* complete and utter failure */
4145 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4147 unsigned int block_nr = req->tp_block_nr;
4151 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4152 if (unlikely(!pg_vec))
4155 for (i = 0; i < block_nr; i++) {
4156 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4157 if (unlikely(!pg_vec[i].buffer))
4158 goto out_free_pgvec;
4165 free_pg_vec(pg_vec, order, block_nr);
4170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4171 int closing, int tx_ring)
4173 struct pgv *pg_vec = NULL;
4174 struct packet_sock *po = pkt_sk(sk);
4175 int was_running, order = 0;
4176 struct packet_ring_buffer *rb;
4177 struct sk_buff_head *rb_queue;
4180 /* Added to avoid minimal code churn */
4181 struct tpacket_req *req = &req_u->req;
4183 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4184 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4185 net_warn_ratelimited("Tx-ring is not supported.\n");
4189 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4190 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4194 if (atomic_read(&po->mapped))
4196 if (packet_read_pending(rb))
4200 if (req->tp_block_nr) {
4201 /* Sanity tests and some calculations */
4203 if (unlikely(rb->pg_vec))
4206 switch (po->tp_version) {
4208 po->tp_hdrlen = TPACKET_HDRLEN;
4211 po->tp_hdrlen = TPACKET2_HDRLEN;
4214 po->tp_hdrlen = TPACKET3_HDRLEN;
4219 if (unlikely((int)req->tp_block_size <= 0))
4221 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4223 if (po->tp_version >= TPACKET_V3 &&
4224 (int)(req->tp_block_size -
4225 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4227 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4230 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4233 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4234 if (unlikely(rb->frames_per_block == 0))
4236 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4241 order = get_order(req->tp_block_size);
4242 pg_vec = alloc_pg_vec(req, order);
4243 if (unlikely(!pg_vec))
4245 switch (po->tp_version) {
4247 /* Transmit path is not supported. We checked
4248 * it above but just being paranoid
4251 init_prb_bdqc(po, rb, pg_vec, req_u);
4260 if (unlikely(req->tp_frame_nr))
4266 /* Detach socket from network */
4267 spin_lock(&po->bind_lock);
4268 was_running = po->running;
4272 __unregister_prot_hook(sk, false);
4274 spin_unlock(&po->bind_lock);
4279 mutex_lock(&po->pg_vec_lock);
4280 if (closing || atomic_read(&po->mapped) == 0) {
4282 spin_lock_bh(&rb_queue->lock);
4283 swap(rb->pg_vec, pg_vec);
4284 rb->frame_max = (req->tp_frame_nr - 1);
4286 rb->frame_size = req->tp_frame_size;
4287 spin_unlock_bh(&rb_queue->lock);
4289 swap(rb->pg_vec_order, order);
4290 swap(rb->pg_vec_len, req->tp_block_nr);
4292 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4293 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4294 tpacket_rcv : packet_rcv;
4295 skb_queue_purge(rb_queue);
4296 if (atomic_read(&po->mapped))
4297 pr_err("packet_mmap: vma is busy: %d\n",
4298 atomic_read(&po->mapped));
4300 mutex_unlock(&po->pg_vec_lock);
4302 spin_lock(&po->bind_lock);
4305 register_prot_hook(sk);
4307 spin_unlock(&po->bind_lock);
4308 if (closing && (po->tp_version > TPACKET_V2)) {
4309 /* Because we don't support block-based V3 on tx-ring */
4311 prb_shutdown_retire_blk_timer(po, rb_queue);
4316 free_pg_vec(pg_vec, order, req->tp_block_nr);
4321 static int packet_mmap(struct file *file, struct socket *sock,
4322 struct vm_area_struct *vma)
4324 struct sock *sk = sock->sk;
4325 struct packet_sock *po = pkt_sk(sk);
4326 unsigned long size, expected_size;
4327 struct packet_ring_buffer *rb;
4328 unsigned long start;
4335 mutex_lock(&po->pg_vec_lock);
4338 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4340 expected_size += rb->pg_vec_len
4346 if (expected_size == 0)
4349 size = vma->vm_end - vma->vm_start;
4350 if (size != expected_size)
4353 start = vma->vm_start;
4354 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4355 if (rb->pg_vec == NULL)
4358 for (i = 0; i < rb->pg_vec_len; i++) {
4360 void *kaddr = rb->pg_vec[i].buffer;
4363 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4364 page = pgv_to_page(kaddr);
4365 err = vm_insert_page(vma, start, page);
4374 atomic_inc(&po->mapped);
4375 vma->vm_ops = &packet_mmap_ops;
4379 mutex_unlock(&po->pg_vec_lock);
4383 static const struct proto_ops packet_ops_spkt = {
4384 .family = PF_PACKET,
4385 .owner = THIS_MODULE,
4386 .release = packet_release,
4387 .bind = packet_bind_spkt,
4388 .connect = sock_no_connect,
4389 .socketpair = sock_no_socketpair,
4390 .accept = sock_no_accept,
4391 .getname = packet_getname_spkt,
4392 .poll = datagram_poll,
4393 .ioctl = packet_ioctl,
4394 .listen = sock_no_listen,
4395 .shutdown = sock_no_shutdown,
4396 .setsockopt = sock_no_setsockopt,
4397 .getsockopt = sock_no_getsockopt,
4398 .sendmsg = packet_sendmsg_spkt,
4399 .recvmsg = packet_recvmsg,
4400 .mmap = sock_no_mmap,
4401 .sendpage = sock_no_sendpage,
4404 static const struct proto_ops packet_ops = {
4405 .family = PF_PACKET,
4406 .owner = THIS_MODULE,
4407 .release = packet_release,
4408 .bind = packet_bind,
4409 .connect = sock_no_connect,
4410 .socketpair = sock_no_socketpair,
4411 .accept = sock_no_accept,
4412 .getname = packet_getname,
4413 .poll = packet_poll,
4414 .ioctl = packet_ioctl,
4415 .listen = sock_no_listen,
4416 .shutdown = sock_no_shutdown,
4417 .setsockopt = packet_setsockopt,
4418 .getsockopt = packet_getsockopt,
4419 .sendmsg = packet_sendmsg,
4420 .recvmsg = packet_recvmsg,
4421 .mmap = packet_mmap,
4422 .sendpage = sock_no_sendpage,
4425 static const struct net_proto_family packet_family_ops = {
4426 .family = PF_PACKET,
4427 .create = packet_create,
4428 .owner = THIS_MODULE,
4431 static struct notifier_block packet_netdev_notifier = {
4432 .notifier_call = packet_notifier,
4435 #ifdef CONFIG_PROC_FS
4437 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4440 struct net *net = seq_file_net(seq);
4443 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4446 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4448 struct net *net = seq_file_net(seq);
4449 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4452 static void packet_seq_stop(struct seq_file *seq, void *v)
4458 static int packet_seq_show(struct seq_file *seq, void *v)
4460 if (v == SEQ_START_TOKEN)
4461 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4463 struct sock *s = sk_entry(v);
4464 const struct packet_sock *po = pkt_sk(s);
4467 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4469 atomic_read(&s->sk_refcnt),
4474 atomic_read(&s->sk_rmem_alloc),
4475 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4482 static const struct seq_operations packet_seq_ops = {
4483 .start = packet_seq_start,
4484 .next = packet_seq_next,
4485 .stop = packet_seq_stop,
4486 .show = packet_seq_show,
4489 static int packet_seq_open(struct inode *inode, struct file *file)
4491 return seq_open_net(inode, file, &packet_seq_ops,
4492 sizeof(struct seq_net_private));
4495 static const struct file_operations packet_seq_fops = {
4496 .owner = THIS_MODULE,
4497 .open = packet_seq_open,
4499 .llseek = seq_lseek,
4500 .release = seq_release_net,
4505 static int __net_init packet_net_init(struct net *net)
4507 mutex_init(&net->packet.sklist_lock);
4508 INIT_HLIST_HEAD(&net->packet.sklist);
4510 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4516 static void __net_exit packet_net_exit(struct net *net)
4518 remove_proc_entry("packet", net->proc_net);
4521 static struct pernet_operations packet_net_ops = {
4522 .init = packet_net_init,
4523 .exit = packet_net_exit,
4527 static void __exit packet_exit(void)
4529 unregister_netdevice_notifier(&packet_netdev_notifier);
4530 unregister_pernet_subsys(&packet_net_ops);
4531 sock_unregister(PF_PACKET);
4532 proto_unregister(&packet_proto);
4535 static int __init packet_init(void)
4537 int rc = proto_register(&packet_proto, 0);
4542 sock_register(&packet_family_ops);
4543 register_pernet_subsys(&packet_net_ops);
4544 register_netdevice_notifier(&packet_netdev_notifier);
4549 module_init(packet_init);
4550 module_exit(packet_exit);
4551 MODULE_LICENSE("GPL");
4552 MODULE_ALIAS_NETPROTO(PF_PACKET);