1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PACKET - implements raw packet sockets.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
49 #include <linux/types.h>
51 #include <linux/capability.h>
52 #include <linux/fcntl.h>
53 #include <linux/socket.h>
55 #include <linux/inet.h>
56 #include <linux/netdevice.h>
57 #include <linux/if_packet.h>
58 #include <linux/wireless.h>
59 #include <linux/kernel.h>
60 #include <linux/kmod.h>
61 #include <linux/slab.h>
62 #include <linux/vmalloc.h>
63 #include <net/net_namespace.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.h>
68 #include <linux/errno.h>
69 #include <linux/timer.h>
70 #include <linux/uaccess.h>
71 #include <asm/ioctls.h>
73 #include <asm/cacheflush.h>
75 #include <linux/proc_fs.h>
76 #include <linux/seq_file.h>
77 #include <linux/poll.h>
78 #include <linux/module.h>
79 #include <linux/init.h>
80 #include <linux/mutex.h>
81 #include <linux/if_vlan.h>
82 #include <linux/virtio_net.h>
83 #include <linux/errqueue.h>
84 #include <linux/net_tstamp.h>
85 #include <linux/percpu.h>
87 #include <net/inet_common.h>
89 #include <linux/bpf.h>
90 #include <net/compat.h>
96 - if device has no dev->hard_header routine, it adds and removes ll header
97 inside itself. In this case ll header is invisible outside of device,
98 but higher levels still should reserve dev->hard_header_len.
99 Some devices are enough clever to reallocate skb, when header
100 will not fit to reserved space (tunnel), another ones are silly
102 - packet socket receives packets with pulled ll header,
103 so that SOCK_RAW should push it back.
108 Incoming, dev->hard_header!=NULL
109 mac_header -> ll header
112 Outgoing, dev->hard_header!=NULL
113 mac_header -> ll header
116 Incoming, dev->hard_header==NULL
117 mac_header -> UNKNOWN position. It is very likely, that it points to ll
118 header. PPP makes it, that is wrong, because introduce
119 assymetry between rx and tx paths.
122 Outgoing, dev->hard_header==NULL
123 mac_header -> data. ll header is still not built!
127 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
133 dev->hard_header != NULL
134 mac_header -> ll header
137 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141 We should set nh.raw on output to correct posistion,
142 packet classifier depends on it.
145 /* Private packet socket structures. */
147 /* identical to struct packet_mreq except it has
148 * a longer address field.
150 struct packet_mreq_max {
152 unsigned short mr_type;
153 unsigned short mr_alen;
154 unsigned char mr_address[MAX_ADDR_LEN];
158 struct tpacket_hdr *h1;
159 struct tpacket2_hdr *h2;
160 struct tpacket3_hdr *h3;
164 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
165 int closing, int tx_ring);
167 #define V3_ALIGNMENT (8)
169 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
171 #define BLK_PLUS_PRIV(sz_of_priv) \
172 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
175 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
176 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
177 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
178 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
180 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
184 struct packet_type *pt, struct net_device *orig_dev);
186 static void *packet_previous_frame(struct packet_sock *po,
187 struct packet_ring_buffer *rb,
189 static void packet_increment_head(struct packet_ring_buffer *buff);
190 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
192 struct packet_sock *);
193 static void prb_retire_current_block(struct tpacket_kbdq_core *,
194 struct packet_sock *, unsigned int status);
195 static int prb_queue_frozen(struct tpacket_kbdq_core *);
196 static void prb_open_block(struct tpacket_kbdq_core *,
197 struct tpacket_block_desc *);
198 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
200 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
201 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
202 struct tpacket3_hdr *);
203 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
204 struct tpacket3_hdr *);
205 static void packet_flush_mclist(struct sock *sk);
206 static u16 packet_pick_tx_queue(struct sk_buff *skb);
208 struct packet_skb_cb {
210 struct sockaddr_pkt pkt;
212 /* Trick: alias skb original length with
213 * ll.sll_family and ll.protocol in order
216 unsigned int origlen;
217 struct sockaddr_ll ll;
222 #define vio_le() virtio_legacy_is_little_endian()
224 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
226 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
227 #define GET_PBLOCK_DESC(x, bid) \
228 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
229 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
231 #define GET_NEXT_PRB_BLK_NUM(x) \
232 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
233 ((x)->kactive_blk_num+1) : 0)
235 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
236 static void __fanout_link(struct sock *sk, struct packet_sock *po);
238 static int packet_direct_xmit(struct sk_buff *skb)
240 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
243 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
245 struct net_device *dev;
248 dev = rcu_dereference(po->cached_dev);
256 static void packet_cached_dev_assign(struct packet_sock *po,
257 struct net_device *dev)
259 rcu_assign_pointer(po->cached_dev, dev);
262 static void packet_cached_dev_reset(struct packet_sock *po)
264 RCU_INIT_POINTER(po->cached_dev, NULL);
267 static bool packet_use_direct_xmit(const struct packet_sock *po)
269 return po->xmit == packet_direct_xmit;
272 static u16 packet_pick_tx_queue(struct sk_buff *skb)
274 struct net_device *dev = skb->dev;
275 const struct net_device_ops *ops = dev->netdev_ops;
276 int cpu = raw_smp_processor_id();
280 skb->sender_cpu = cpu + 1;
282 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
283 if (ops->ndo_select_queue) {
284 queue_index = ops->ndo_select_queue(dev, skb, NULL);
285 queue_index = netdev_cap_txqueue(dev, queue_index);
287 queue_index = netdev_pick_tx(dev, skb, NULL);
293 /* __register_prot_hook must be invoked through register_prot_hook
294 * or from a context in which asynchronous accesses to the packet
295 * socket is not possible (packet_create()).
297 static void __register_prot_hook(struct sock *sk)
299 struct packet_sock *po = pkt_sk(sk);
303 __fanout_link(sk, po);
305 dev_add_pack(&po->prot_hook);
312 static void register_prot_hook(struct sock *sk)
314 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
315 __register_prot_hook(sk);
318 /* If the sync parameter is true, we will temporarily drop
319 * the po->bind_lock and do a synchronize_net to make sure no
320 * asynchronous packet processing paths still refer to the elements
321 * of po->prot_hook. If the sync parameter is false, it is the
322 * callers responsibility to take care of this.
324 static void __unregister_prot_hook(struct sock *sk, bool sync)
326 struct packet_sock *po = pkt_sk(sk);
328 lockdep_assert_held_once(&po->bind_lock);
333 __fanout_unlink(sk, po);
335 __dev_remove_pack(&po->prot_hook);
340 spin_unlock(&po->bind_lock);
342 spin_lock(&po->bind_lock);
346 static void unregister_prot_hook(struct sock *sk, bool sync)
348 struct packet_sock *po = pkt_sk(sk);
351 __unregister_prot_hook(sk, sync);
354 static inline struct page * __pure pgv_to_page(void *addr)
356 if (is_vmalloc_addr(addr))
357 return vmalloc_to_page(addr);
358 return virt_to_page(addr);
361 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
363 union tpacket_uhdr h;
366 switch (po->tp_version) {
368 h.h1->tp_status = status;
369 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
372 h.h2->tp_status = status;
373 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
376 h.h3->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
380 WARN(1, "TPACKET version not supported.\n");
387 static int __packet_get_status(const struct packet_sock *po, void *frame)
389 union tpacket_uhdr h;
394 switch (po->tp_version) {
396 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
397 return h.h1->tp_status;
399 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
400 return h.h2->tp_status;
402 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
403 return h.h3->tp_status;
405 WARN(1, "TPACKET version not supported.\n");
411 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
414 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
417 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
418 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
419 return TP_STATUS_TS_RAW_HARDWARE;
421 if (ktime_to_timespec_cond(skb->tstamp, ts))
422 return TP_STATUS_TS_SOFTWARE;
427 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
430 union tpacket_uhdr h;
434 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
438 switch (po->tp_version) {
440 h.h1->tp_sec = ts.tv_sec;
441 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
444 h.h2->tp_sec = ts.tv_sec;
445 h.h2->tp_nsec = ts.tv_nsec;
448 h.h3->tp_sec = ts.tv_sec;
449 h.h3->tp_nsec = ts.tv_nsec;
452 WARN(1, "TPACKET version not supported.\n");
456 /* one flush is safe, as both fields always lie on the same cacheline */
457 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
463 static void *packet_lookup_frame(const struct packet_sock *po,
464 const struct packet_ring_buffer *rb,
465 unsigned int position,
468 unsigned int pg_vec_pos, frame_offset;
469 union tpacket_uhdr h;
471 pg_vec_pos = position / rb->frames_per_block;
472 frame_offset = position % rb->frames_per_block;
474 h.raw = rb->pg_vec[pg_vec_pos].buffer +
475 (frame_offset * rb->frame_size);
477 if (status != __packet_get_status(po, h.raw))
483 static void *packet_current_frame(struct packet_sock *po,
484 struct packet_ring_buffer *rb,
487 return packet_lookup_frame(po, rb, rb->head, status);
490 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
492 del_timer_sync(&pkc->retire_blk_timer);
495 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
496 struct sk_buff_head *rb_queue)
498 struct tpacket_kbdq_core *pkc;
500 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
502 spin_lock_bh(&rb_queue->lock);
503 pkc->delete_blk_timer = 1;
504 spin_unlock_bh(&rb_queue->lock);
506 prb_del_retire_blk_timer(pkc);
509 static void prb_setup_retire_blk_timer(struct packet_sock *po)
511 struct tpacket_kbdq_core *pkc;
513 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
514 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
516 pkc->retire_blk_timer.expires = jiffies;
519 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
520 int blk_size_in_bytes)
522 struct net_device *dev;
523 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
524 struct ethtool_link_ksettings ecmd;
528 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
529 if (unlikely(!dev)) {
531 return DEFAULT_PRB_RETIRE_TOV;
533 err = __ethtool_get_link_ksettings(dev, &ecmd);
537 * If the link speed is so slow you don't really
538 * need to worry about perf anyways
540 if (ecmd.base.speed < SPEED_1000 ||
541 ecmd.base.speed == SPEED_UNKNOWN) {
542 return DEFAULT_PRB_RETIRE_TOV;
545 div = ecmd.base.speed / 1000;
549 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
561 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
562 union tpacket_req_u *req_u)
564 p1->feature_req_word = req_u->req3.tp_feature_req_word;
567 static void init_prb_bdqc(struct packet_sock *po,
568 struct packet_ring_buffer *rb,
570 union tpacket_req_u *req_u)
572 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
573 struct tpacket_block_desc *pbd;
575 memset(p1, 0x0, sizeof(*p1));
577 p1->knxt_seq_num = 1;
579 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
580 p1->pkblk_start = pg_vec[0].buffer;
581 p1->kblk_size = req_u->req3.tp_block_size;
582 p1->knum_blocks = req_u->req3.tp_block_nr;
583 p1->hdrlen = po->tp_hdrlen;
584 p1->version = po->tp_version;
585 p1->last_kactive_blk_num = 0;
586 po->stats.stats3.tp_freeze_q_cnt = 0;
587 if (req_u->req3.tp_retire_blk_tov)
588 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
590 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
591 req_u->req3.tp_block_size);
592 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
593 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
595 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
596 prb_init_ft_ops(p1, req_u);
597 prb_setup_retire_blk_timer(po);
598 prb_open_block(p1, pbd);
601 /* Do NOT update the last_blk_num first.
602 * Assumes sk_buff_head lock is held.
604 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
606 mod_timer(&pkc->retire_blk_timer,
607 jiffies + pkc->tov_in_jiffies);
608 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
613 * 1) We refresh the timer only when we open a block.
614 * By doing this we don't waste cycles refreshing the timer
615 * on packet-by-packet basis.
617 * With a 1MB block-size, on a 1Gbps line, it will take
618 * i) ~8 ms to fill a block + ii) memcpy etc.
619 * In this cut we are not accounting for the memcpy time.
621 * So, if the user sets the 'tmo' to 10ms then the timer
622 * will never fire while the block is still getting filled
623 * (which is what we want). However, the user could choose
624 * to close a block early and that's fine.
626 * But when the timer does fire, we check whether or not to refresh it.
627 * Since the tmo granularity is in msecs, it is not too expensive
628 * to refresh the timer, lets say every '8' msecs.
629 * Either the user can set the 'tmo' or we can derive it based on
630 * a) line-speed and b) block-size.
631 * prb_calc_retire_blk_tmo() calculates the tmo.
634 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
636 struct packet_sock *po =
637 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
638 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
640 struct tpacket_block_desc *pbd;
642 spin_lock(&po->sk.sk_receive_queue.lock);
644 frozen = prb_queue_frozen(pkc);
645 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
647 if (unlikely(pkc->delete_blk_timer))
650 /* We only need to plug the race when the block is partially filled.
652 * lock(); increment BLOCK_NUM_PKTS; unlock()
653 * copy_bits() is in progress ...
654 * timer fires on other cpu:
655 * we can't retire the current block because copy_bits
659 if (BLOCK_NUM_PKTS(pbd)) {
660 while (atomic_read(&pkc->blk_fill_in_prog)) {
661 /* Waiting for skb_copy_bits to finish... */
666 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
668 if (!BLOCK_NUM_PKTS(pbd)) {
669 /* An empty block. Just refresh the timer. */
672 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
673 if (!prb_dispatch_next_block(pkc, po))
678 /* Case 1. Queue was frozen because user-space was
681 if (prb_curr_blk_in_use(pbd)) {
683 * Ok, user-space is still behind.
684 * So just refresh the timer.
688 /* Case 2. queue was frozen,user-space caught up,
689 * now the link went idle && the timer fired.
690 * We don't have a block to close.So we open this
691 * block and restart the timer.
692 * opening a block thaws the queue,restarts timer
693 * Thawing/timer-refresh is a side effect.
695 prb_open_block(pkc, pbd);
702 _prb_refresh_rx_retire_blk_timer(pkc);
705 spin_unlock(&po->sk.sk_receive_queue.lock);
708 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
709 struct tpacket_block_desc *pbd1, __u32 status)
711 /* Flush everything minus the block header */
713 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
718 /* Skip the block header(we know header WILL fit in 4K) */
721 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
722 for (; start < end; start += PAGE_SIZE)
723 flush_dcache_page(pgv_to_page(start));
728 /* Now update the block status. */
730 BLOCK_STATUS(pbd1) = status;
732 /* Flush the block header */
734 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
736 flush_dcache_page(pgv_to_page(start));
746 * 2) Increment active_blk_num
748 * Note:We DONT refresh the timer on purpose.
749 * Because almost always the next block will be opened.
751 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
752 struct tpacket_block_desc *pbd1,
753 struct packet_sock *po, unsigned int stat)
755 __u32 status = TP_STATUS_USER | stat;
757 struct tpacket3_hdr *last_pkt;
758 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
759 struct sock *sk = &po->sk;
761 if (atomic_read(&po->tp_drops))
762 status |= TP_STATUS_LOSING;
764 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
765 last_pkt->tp_next_offset = 0;
767 /* Get the ts of the last pkt */
768 if (BLOCK_NUM_PKTS(pbd1)) {
769 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
770 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
772 /* Ok, we tmo'd - so get the current time.
774 * It shouldn't really happen as we don't close empty
775 * blocks. See prb_retire_rx_blk_timer_expired().
779 h1->ts_last_pkt.ts_sec = ts.tv_sec;
780 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
785 /* Flush the block */
786 prb_flush_block(pkc1, pbd1, status);
788 sk->sk_data_ready(sk);
790 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
793 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
795 pkc->reset_pending_on_curr_blk = 0;
799 * Side effect of opening a block:
801 * 1) prb_queue is thawed.
802 * 2) retire_blk_timer is refreshed.
805 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
806 struct tpacket_block_desc *pbd1)
809 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
813 /* We could have just memset this but we will lose the
814 * flexibility of making the priv area sticky
817 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
818 BLOCK_NUM_PKTS(pbd1) = 0;
819 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
823 h1->ts_first_pkt.ts_sec = ts.tv_sec;
824 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
826 pkc1->pkblk_start = (char *)pbd1;
827 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
829 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
832 pbd1->version = pkc1->version;
833 pkc1->prev = pkc1->nxt_offset;
834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
836 prb_thaw_queue(pkc1);
837 _prb_refresh_rx_retire_blk_timer(pkc1);
843 * Queue freeze logic:
844 * 1) Assume tp_block_nr = 8 blocks.
845 * 2) At time 't0', user opens Rx ring.
846 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
847 * 4) user-space is either sleeping or processing block '0'.
848 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
849 * it will close block-7,loop around and try to fill block '0'.
851 * __packet_lookup_frame_in_block
852 * prb_retire_current_block()
853 * prb_dispatch_next_block()
854 * |->(BLOCK_STATUS == USER) evaluates to true
855 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
856 * 6) Now there are two cases:
857 * 6.1) Link goes idle right after the queue is frozen.
858 * But remember, the last open_block() refreshed the timer.
859 * When this timer expires,it will refresh itself so that we can
860 * re-open block-0 in near future.
861 * 6.2) Link is busy and keeps on receiving packets. This is a simple
862 * case and __packet_lookup_frame_in_block will check if block-0
863 * is free and can now be re-used.
865 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
866 struct packet_sock *po)
868 pkc->reset_pending_on_curr_blk = 1;
869 po->stats.stats3.tp_freeze_q_cnt++;
872 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
875 * If the next block is free then we will dispatch it
876 * and return a good offset.
877 * Else, we will freeze the queue.
878 * So, caller must check the return value.
880 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
881 struct packet_sock *po)
883 struct tpacket_block_desc *pbd;
887 /* 1. Get current block num */
888 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
890 /* 2. If this block is currently in_use then freeze the queue */
891 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
892 prb_freeze_queue(pkc, po);
898 * open this block and return the offset where the first packet
899 * needs to get stored.
901 prb_open_block(pkc, pbd);
902 return (void *)pkc->nxt_offset;
905 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
906 struct packet_sock *po, unsigned int status)
908 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
910 /* retire/close the current block */
911 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
913 * Plug the case where copy_bits() is in progress on
914 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
915 * have space to copy the pkt in the current block and
916 * called prb_retire_current_block()
918 * We don't need to worry about the TMO case because
919 * the timer-handler already handled this case.
921 if (!(status & TP_STATUS_BLK_TMO)) {
922 while (atomic_read(&pkc->blk_fill_in_prog)) {
923 /* Waiting for skb_copy_bits to finish... */
927 prb_close_block(pkc, pbd, po, status);
932 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
934 return TP_STATUS_USER & BLOCK_STATUS(pbd);
937 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
939 return pkc->reset_pending_on_curr_blk;
942 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
944 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
945 atomic_dec(&pkc->blk_fill_in_prog);
948 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
949 struct tpacket3_hdr *ppd)
951 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
954 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
955 struct tpacket3_hdr *ppd)
957 ppd->hv1.tp_rxhash = 0;
960 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
963 if (skb_vlan_tag_present(pkc->skb)) {
964 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
965 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
966 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
968 ppd->hv1.tp_vlan_tci = 0;
969 ppd->hv1.tp_vlan_tpid = 0;
970 ppd->tp_status = TP_STATUS_AVAILABLE;
974 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
975 struct tpacket3_hdr *ppd)
977 ppd->hv1.tp_padding = 0;
978 prb_fill_vlan_info(pkc, ppd);
980 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
981 prb_fill_rxhash(pkc, ppd);
983 prb_clear_rxhash(pkc, ppd);
986 static void prb_fill_curr_block(char *curr,
987 struct tpacket_kbdq_core *pkc,
988 struct tpacket_block_desc *pbd,
991 struct tpacket3_hdr *ppd;
993 ppd = (struct tpacket3_hdr *)curr;
994 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
996 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
997 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
998 BLOCK_NUM_PKTS(pbd) += 1;
999 atomic_inc(&pkc->blk_fill_in_prog);
1000 prb_run_all_ft_ops(pkc, ppd);
1003 /* Assumes caller has the sk->rx_queue.lock */
1004 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1005 struct sk_buff *skb,
1009 struct tpacket_kbdq_core *pkc;
1010 struct tpacket_block_desc *pbd;
1013 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1014 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1016 /* Queue is frozen when user space is lagging behind */
1017 if (prb_queue_frozen(pkc)) {
1019 * Check if that last block which caused the queue to freeze,
1020 * is still in_use by user-space.
1022 if (prb_curr_blk_in_use(pbd)) {
1023 /* Can't record this packet */
1027 * Ok, the block was released by user-space.
1028 * Now let's open that block.
1029 * opening a block also thaws the queue.
1030 * Thawing is a side effect.
1032 prb_open_block(pkc, pbd);
1037 curr = pkc->nxt_offset;
1039 end = (char *)pbd + pkc->kblk_size;
1041 /* first try the current block */
1042 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1043 prb_fill_curr_block(curr, pkc, pbd, len);
1044 return (void *)curr;
1047 /* Ok, close the current block */
1048 prb_retire_current_block(pkc, po, 0);
1050 /* Now, try to dispatch the next block */
1051 curr = (char *)prb_dispatch_next_block(pkc, po);
1053 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1054 prb_fill_curr_block(curr, pkc, pbd, len);
1055 return (void *)curr;
1059 * No free blocks are available.user_space hasn't caught up yet.
1060 * Queue was just frozen and now this packet will get dropped.
1065 static void *packet_current_rx_frame(struct packet_sock *po,
1066 struct sk_buff *skb,
1067 int status, unsigned int len)
1070 switch (po->tp_version) {
1073 curr = packet_lookup_frame(po, &po->rx_ring,
1074 po->rx_ring.head, status);
1077 return __packet_lookup_frame_in_block(po, skb, len);
1079 WARN(1, "TPACKET version not supported\n");
1085 static void *prb_lookup_block(const struct packet_sock *po,
1086 const struct packet_ring_buffer *rb,
1090 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1091 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1093 if (status != BLOCK_STATUS(pbd))
1098 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1101 if (rb->prb_bdqc.kactive_blk_num)
1102 prev = rb->prb_bdqc.kactive_blk_num-1;
1104 prev = rb->prb_bdqc.knum_blocks-1;
1108 /* Assumes caller has held the rx_queue.lock */
1109 static void *__prb_previous_block(struct packet_sock *po,
1110 struct packet_ring_buffer *rb,
1113 unsigned int previous = prb_previous_blk_num(rb);
1114 return prb_lookup_block(po, rb, previous, status);
1117 static void *packet_previous_rx_frame(struct packet_sock *po,
1118 struct packet_ring_buffer *rb,
1121 if (po->tp_version <= TPACKET_V2)
1122 return packet_previous_frame(po, rb, status);
1124 return __prb_previous_block(po, rb, status);
1127 static void packet_increment_rx_head(struct packet_sock *po,
1128 struct packet_ring_buffer *rb)
1130 switch (po->tp_version) {
1133 return packet_increment_head(rb);
1136 WARN(1, "TPACKET version not supported.\n");
1142 static void *packet_previous_frame(struct packet_sock *po,
1143 struct packet_ring_buffer *rb,
1146 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1147 return packet_lookup_frame(po, rb, previous, status);
1150 static void packet_increment_head(struct packet_ring_buffer *buff)
1152 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1155 static void packet_inc_pending(struct packet_ring_buffer *rb)
1157 this_cpu_inc(*rb->pending_refcnt);
1160 static void packet_dec_pending(struct packet_ring_buffer *rb)
1162 this_cpu_dec(*rb->pending_refcnt);
1165 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1167 unsigned int refcnt = 0;
1170 /* We don't use pending refcount in rx_ring. */
1171 if (rb->pending_refcnt == NULL)
1174 for_each_possible_cpu(cpu)
1175 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1180 static int packet_alloc_pending(struct packet_sock *po)
1182 po->rx_ring.pending_refcnt = NULL;
1184 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1185 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1191 static void packet_free_pending(struct packet_sock *po)
1193 free_percpu(po->tx_ring.pending_refcnt);
1196 #define ROOM_POW_OFF 2
1197 #define ROOM_NONE 0x0
1198 #define ROOM_LOW 0x1
1199 #define ROOM_NORMAL 0x2
1201 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1205 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1206 idx = READ_ONCE(po->rx_ring.head);
1208 idx += len >> pow_off;
1211 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1214 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1218 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1219 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1221 idx += len >> pow_off;
1224 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1227 static int __packet_rcv_has_room(const struct packet_sock *po,
1228 const struct sk_buff *skb)
1230 const struct sock *sk = &po->sk;
1231 int ret = ROOM_NONE;
1233 if (po->prot_hook.func != tpacket_rcv) {
1234 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1235 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1236 - (skb ? skb->truesize : 0);
1238 if (avail > (rcvbuf >> ROOM_POW_OFF))
1246 if (po->tp_version == TPACKET_V3) {
1247 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1249 else if (__tpacket_v3_has_room(po, 0))
1252 if (__tpacket_has_room(po, ROOM_POW_OFF))
1254 else if (__tpacket_has_room(po, 0))
1261 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1265 ret = __packet_rcv_has_room(po, skb);
1266 pressure = ret != ROOM_NORMAL;
1268 if (READ_ONCE(po->pressure) != pressure)
1269 WRITE_ONCE(po->pressure, pressure);
1274 static void packet_sock_destruct(struct sock *sk)
1276 skb_queue_purge(&sk->sk_error_queue);
1278 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1279 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1281 if (!sock_flag(sk, SOCK_DEAD)) {
1282 pr_err("Attempt to release alive packet socket: %p\n", sk);
1286 sk_refcnt_debug_dec(sk);
1289 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1294 rxhash = skb_get_hash(skb);
1295 for (i = 0; i < ROLLOVER_HLEN; i++)
1296 if (po->rollover->history[i] == rxhash)
1299 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1300 return count > (ROLLOVER_HLEN >> 1);
1303 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1304 struct sk_buff *skb,
1307 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1310 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1311 struct sk_buff *skb,
1314 unsigned int val = atomic_inc_return(&f->rr_cur);
1319 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1320 struct sk_buff *skb,
1323 return smp_processor_id() % num;
1326 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1327 struct sk_buff *skb,
1330 return prandom_u32_max(num);
1333 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1334 struct sk_buff *skb,
1335 unsigned int idx, bool try_self,
1338 struct packet_sock *po, *po_next, *po_skip = NULL;
1339 unsigned int i, j, room = ROOM_NONE;
1341 po = pkt_sk(f->arr[idx]);
1344 room = packet_rcv_has_room(po, skb);
1345 if (room == ROOM_NORMAL ||
1346 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1351 i = j = min_t(int, po->rollover->sock, num - 1);
1353 po_next = pkt_sk(f->arr[i]);
1354 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1355 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1357 po->rollover->sock = i;
1358 atomic_long_inc(&po->rollover->num);
1359 if (room == ROOM_LOW)
1360 atomic_long_inc(&po->rollover->num_huge);
1368 atomic_long_inc(&po->rollover->num_failed);
1372 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1373 struct sk_buff *skb,
1376 return skb_get_queue_mapping(skb) % num;
1379 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1380 struct sk_buff *skb,
1383 struct bpf_prog *prog;
1384 unsigned int ret = 0;
1387 prog = rcu_dereference(f->bpf_prog);
1389 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1395 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1397 return f->flags & (flag >> 8);
1400 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1401 struct packet_type *pt, struct net_device *orig_dev)
1403 struct packet_fanout *f = pt->af_packet_priv;
1404 unsigned int num = READ_ONCE(f->num_members);
1405 struct net *net = read_pnet(&f->net);
1406 struct packet_sock *po;
1409 if (!net_eq(dev_net(dev), net) || !num) {
1414 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1415 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1420 case PACKET_FANOUT_HASH:
1422 idx = fanout_demux_hash(f, skb, num);
1424 case PACKET_FANOUT_LB:
1425 idx = fanout_demux_lb(f, skb, num);
1427 case PACKET_FANOUT_CPU:
1428 idx = fanout_demux_cpu(f, skb, num);
1430 case PACKET_FANOUT_RND:
1431 idx = fanout_demux_rnd(f, skb, num);
1433 case PACKET_FANOUT_QM:
1434 idx = fanout_demux_qm(f, skb, num);
1436 case PACKET_FANOUT_ROLLOVER:
1437 idx = fanout_demux_rollover(f, skb, 0, false, num);
1439 case PACKET_FANOUT_CBPF:
1440 case PACKET_FANOUT_EBPF:
1441 idx = fanout_demux_bpf(f, skb, num);
1445 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1446 idx = fanout_demux_rollover(f, skb, idx, true, num);
1448 po = pkt_sk(f->arr[idx]);
1449 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1452 DEFINE_MUTEX(fanout_mutex);
1453 EXPORT_SYMBOL_GPL(fanout_mutex);
1454 static LIST_HEAD(fanout_list);
1455 static u16 fanout_next_id;
1457 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1459 struct packet_fanout *f = po->fanout;
1461 spin_lock(&f->lock);
1462 f->arr[f->num_members] = sk;
1465 if (f->num_members == 1)
1466 dev_add_pack(&f->prot_hook);
1467 spin_unlock(&f->lock);
1470 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1472 struct packet_fanout *f = po->fanout;
1475 spin_lock(&f->lock);
1476 for (i = 0; i < f->num_members; i++) {
1477 if (f->arr[i] == sk)
1480 BUG_ON(i >= f->num_members);
1481 f->arr[i] = f->arr[f->num_members - 1];
1483 if (f->num_members == 0)
1484 __dev_remove_pack(&f->prot_hook);
1485 spin_unlock(&f->lock);
1488 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1490 if (sk->sk_family != PF_PACKET)
1493 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1496 static void fanout_init_data(struct packet_fanout *f)
1499 case PACKET_FANOUT_LB:
1500 atomic_set(&f->rr_cur, 0);
1502 case PACKET_FANOUT_CBPF:
1503 case PACKET_FANOUT_EBPF:
1504 RCU_INIT_POINTER(f->bpf_prog, NULL);
1509 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1511 struct bpf_prog *old;
1513 spin_lock(&f->lock);
1514 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1515 rcu_assign_pointer(f->bpf_prog, new);
1516 spin_unlock(&f->lock);
1520 bpf_prog_destroy(old);
1524 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1527 struct bpf_prog *new;
1528 struct sock_fprog fprog;
1531 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1533 if (len != sizeof(fprog))
1535 if (copy_from_user(&fprog, data, len))
1538 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1542 __fanout_set_data_bpf(po->fanout, new);
1546 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1549 struct bpf_prog *new;
1552 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1554 if (len != sizeof(fd))
1556 if (copy_from_user(&fd, data, len))
1559 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1561 return PTR_ERR(new);
1563 __fanout_set_data_bpf(po->fanout, new);
1567 static int fanout_set_data(struct packet_sock *po, char __user *data,
1570 switch (po->fanout->type) {
1571 case PACKET_FANOUT_CBPF:
1572 return fanout_set_data_cbpf(po, data, len);
1573 case PACKET_FANOUT_EBPF:
1574 return fanout_set_data_ebpf(po, data, len);
1580 static void fanout_release_data(struct packet_fanout *f)
1583 case PACKET_FANOUT_CBPF:
1584 case PACKET_FANOUT_EBPF:
1585 __fanout_set_data_bpf(f, NULL);
1589 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1591 struct packet_fanout *f;
1593 list_for_each_entry(f, &fanout_list, list) {
1594 if (f->id == candidate_id &&
1595 read_pnet(&f->net) == sock_net(sk)) {
1602 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1604 u16 id = fanout_next_id;
1607 if (__fanout_id_is_free(sk, id)) {
1609 fanout_next_id = id + 1;
1614 } while (id != fanout_next_id);
1619 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1621 struct packet_rollover *rollover = NULL;
1622 struct packet_sock *po = pkt_sk(sk);
1623 struct packet_fanout *f, *match;
1624 u8 type = type_flags & 0xff;
1625 u8 flags = type_flags >> 8;
1629 case PACKET_FANOUT_ROLLOVER:
1630 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1632 case PACKET_FANOUT_HASH:
1633 case PACKET_FANOUT_LB:
1634 case PACKET_FANOUT_CPU:
1635 case PACKET_FANOUT_RND:
1636 case PACKET_FANOUT_QM:
1637 case PACKET_FANOUT_CBPF:
1638 case PACKET_FANOUT_EBPF:
1644 mutex_lock(&fanout_mutex);
1650 if (type == PACKET_FANOUT_ROLLOVER ||
1651 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1653 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1656 atomic_long_set(&rollover->num, 0);
1657 atomic_long_set(&rollover->num_huge, 0);
1658 atomic_long_set(&rollover->num_failed, 0);
1661 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1666 if (!fanout_find_new_id(sk, &id)) {
1670 /* ephemeral flag for the first socket in the group: drop it */
1671 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1675 list_for_each_entry(f, &fanout_list, list) {
1677 read_pnet(&f->net) == sock_net(sk)) {
1683 if (match && match->flags != flags)
1687 match = kzalloc(sizeof(*match), GFP_KERNEL);
1690 write_pnet(&match->net, sock_net(sk));
1693 match->flags = flags;
1694 INIT_LIST_HEAD(&match->list);
1695 spin_lock_init(&match->lock);
1696 refcount_set(&match->sk_ref, 0);
1697 fanout_init_data(match);
1698 match->prot_hook.type = po->prot_hook.type;
1699 match->prot_hook.dev = po->prot_hook.dev;
1700 match->prot_hook.func = packet_rcv_fanout;
1701 match->prot_hook.af_packet_priv = match;
1702 match->prot_hook.id_match = match_fanout_group;
1703 list_add(&match->list, &fanout_list);
1707 spin_lock(&po->bind_lock);
1709 match->type == type &&
1710 match->prot_hook.type == po->prot_hook.type &&
1711 match->prot_hook.dev == po->prot_hook.dev) {
1713 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1714 __dev_remove_pack(&po->prot_hook);
1716 po->rollover = rollover;
1718 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1719 __fanout_link(sk, po);
1723 spin_unlock(&po->bind_lock);
1725 if (err && !refcount_read(&match->sk_ref)) {
1726 list_del(&match->list);
1732 mutex_unlock(&fanout_mutex);
1736 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1737 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1738 * It is the responsibility of the caller to call fanout_release_data() and
1739 * free the returned packet_fanout (after synchronize_net())
1741 static struct packet_fanout *fanout_release(struct sock *sk)
1743 struct packet_sock *po = pkt_sk(sk);
1744 struct packet_fanout *f;
1746 mutex_lock(&fanout_mutex);
1751 if (refcount_dec_and_test(&f->sk_ref))
1756 mutex_unlock(&fanout_mutex);
1761 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1762 struct sk_buff *skb)
1764 /* Earlier code assumed this would be a VLAN pkt, double-check
1765 * this now that we have the actual packet in hand. We can only
1766 * do this check on Ethernet devices.
1768 if (unlikely(dev->type != ARPHRD_ETHER))
1771 skb_reset_mac_header(skb);
1772 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1775 static const struct proto_ops packet_ops;
1777 static const struct proto_ops packet_ops_spkt;
1779 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1780 struct packet_type *pt, struct net_device *orig_dev)
1783 struct sockaddr_pkt *spkt;
1786 * When we registered the protocol we saved the socket in the data
1787 * field for just this event.
1790 sk = pt->af_packet_priv;
1793 * Yank back the headers [hope the device set this
1794 * right or kerboom...]
1796 * Incoming packets have ll header pulled,
1799 * For outgoing ones skb->data == skb_mac_header(skb)
1800 * so that this procedure is noop.
1803 if (skb->pkt_type == PACKET_LOOPBACK)
1806 if (!net_eq(dev_net(dev), sock_net(sk)))
1809 skb = skb_share_check(skb, GFP_ATOMIC);
1813 /* drop any routing info */
1816 /* drop conntrack reference */
1819 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1821 skb_push(skb, skb->data - skb_mac_header(skb));
1824 * The SOCK_PACKET socket receives _all_ frames.
1827 spkt->spkt_family = dev->type;
1828 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1829 spkt->spkt_protocol = skb->protocol;
1832 * Charge the memory to the socket. This is done specifically
1833 * to prevent sockets using all the memory up.
1836 if (sock_queue_rcv_skb(sk, skb) == 0)
1845 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1847 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1848 sock->type == SOCK_RAW) {
1849 skb_reset_mac_header(skb);
1850 skb->protocol = dev_parse_header_protocol(skb);
1853 skb_probe_transport_header(skb);
1857 * Output a raw packet to a device layer. This bypasses all the other
1858 * protocol layers and you must therefore supply it with a complete frame
1861 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1864 struct sock *sk = sock->sk;
1865 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1866 struct sk_buff *skb = NULL;
1867 struct net_device *dev;
1868 struct sockcm_cookie sockc;
1874 * Get and verify the address.
1878 if (msg->msg_namelen < sizeof(struct sockaddr))
1880 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1881 proto = saddr->spkt_protocol;
1883 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1886 * Find the device first to size check it
1889 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1892 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1898 if (!(dev->flags & IFF_UP))
1902 * You may not queue a frame bigger than the mtu. This is the lowest level
1903 * raw protocol and you must do your own fragmentation at this level.
1906 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1907 if (!netif_supports_nofcs(dev)) {
1908 err = -EPROTONOSUPPORT;
1911 extra_len = 4; /* We're doing our own CRC */
1915 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1919 size_t reserved = LL_RESERVED_SPACE(dev);
1920 int tlen = dev->needed_tailroom;
1921 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1924 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1927 /* FIXME: Save some space for broken drivers that write a hard
1928 * header at transmission time by themselves. PPP is the notable
1929 * one here. This should really be fixed at the driver level.
1931 skb_reserve(skb, reserved);
1932 skb_reset_network_header(skb);
1934 /* Try to align data part correctly */
1939 skb_reset_network_header(skb);
1941 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1947 if (!dev_validate_header(dev, skb->data, len)) {
1951 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1952 !packet_extra_vlan_len_allowed(dev, skb)) {
1957 sockcm_init(&sockc, sk);
1958 if (msg->msg_controllen) {
1959 err = sock_cmsg_send(sk, msg, &sockc);
1964 skb->protocol = proto;
1966 skb->priority = sk->sk_priority;
1967 skb->mark = sk->sk_mark;
1968 skb->tstamp = sockc.transmit_time;
1970 skb_setup_tx_timestamp(skb, sockc.tsflags);
1972 if (unlikely(extra_len == 4))
1975 packet_parse_headers(skb, sock);
1977 dev_queue_xmit(skb);
1988 static unsigned int run_filter(struct sk_buff *skb,
1989 const struct sock *sk,
1992 struct sk_filter *filter;
1995 filter = rcu_dereference(sk->sk_filter);
1997 res = bpf_prog_run_clear_cb(filter->prog, skb);
2003 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2006 struct virtio_net_hdr vnet_hdr;
2008 if (*len < sizeof(vnet_hdr))
2010 *len -= sizeof(vnet_hdr);
2012 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2015 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2019 * This function makes lazy skb cloning in hope that most of packets
2020 * are discarded by BPF.
2022 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2023 * and skb->cb are mangled. It works because (and until) packets
2024 * falling here are owned by current CPU. Output packets are cloned
2025 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2026 * sequencially, so that if we return skb to original state on exit,
2027 * we will not harm anyone.
2030 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2031 struct packet_type *pt, struct net_device *orig_dev)
2034 struct sockaddr_ll *sll;
2035 struct packet_sock *po;
2036 u8 *skb_head = skb->data;
2037 int skb_len = skb->len;
2038 unsigned int snaplen, res;
2039 bool is_drop_n_account = false;
2041 if (skb->pkt_type == PACKET_LOOPBACK)
2044 sk = pt->af_packet_priv;
2047 if (!net_eq(dev_net(dev), sock_net(sk)))
2052 if (dev->header_ops) {
2053 /* The device has an explicit notion of ll header,
2054 * exported to higher levels.
2056 * Otherwise, the device hides details of its frame
2057 * structure, so that corresponding packet head is
2058 * never delivered to user.
2060 if (sk->sk_type != SOCK_DGRAM)
2061 skb_push(skb, skb->data - skb_mac_header(skb));
2062 else if (skb->pkt_type == PACKET_OUTGOING) {
2063 /* Special case: outgoing packets have ll header at head */
2064 skb_pull(skb, skb_network_offset(skb));
2070 res = run_filter(skb, sk, snaplen);
2072 goto drop_n_restore;
2076 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2079 if (skb_shared(skb)) {
2080 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2084 if (skb_head != skb->data) {
2085 skb->data = skb_head;
2092 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2094 sll = &PACKET_SKB_CB(skb)->sa.ll;
2095 sll->sll_hatype = dev->type;
2096 sll->sll_pkttype = skb->pkt_type;
2097 if (unlikely(po->origdev))
2098 sll->sll_ifindex = orig_dev->ifindex;
2100 sll->sll_ifindex = dev->ifindex;
2102 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2104 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2105 * Use their space for storing the original skb length.
2107 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2109 if (pskb_trim(skb, snaplen))
2112 skb_set_owner_r(skb, sk);
2116 /* drop conntrack reference */
2119 spin_lock(&sk->sk_receive_queue.lock);
2120 po->stats.stats1.tp_packets++;
2121 sock_skb_set_dropcount(sk, skb);
2122 __skb_queue_tail(&sk->sk_receive_queue, skb);
2123 spin_unlock(&sk->sk_receive_queue.lock);
2124 sk->sk_data_ready(sk);
2128 is_drop_n_account = true;
2129 atomic_inc(&po->tp_drops);
2130 atomic_inc(&sk->sk_drops);
2133 if (skb_head != skb->data && skb_shared(skb)) {
2134 skb->data = skb_head;
2138 if (!is_drop_n_account)
2145 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2146 struct packet_type *pt, struct net_device *orig_dev)
2149 struct packet_sock *po;
2150 struct sockaddr_ll *sll;
2151 union tpacket_uhdr h;
2152 u8 *skb_head = skb->data;
2153 int skb_len = skb->len;
2154 unsigned int snaplen, res;
2155 unsigned long status = TP_STATUS_USER;
2156 unsigned short macoff, netoff, hdrlen;
2157 struct sk_buff *copy_skb = NULL;
2160 bool is_drop_n_account = false;
2161 bool do_vnet = false;
2163 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2164 * We may add members to them until current aligned size without forcing
2165 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2167 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2168 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2170 if (skb->pkt_type == PACKET_LOOPBACK)
2173 sk = pt->af_packet_priv;
2176 if (!net_eq(dev_net(dev), sock_net(sk)))
2179 if (dev->header_ops) {
2180 if (sk->sk_type != SOCK_DGRAM)
2181 skb_push(skb, skb->data - skb_mac_header(skb));
2182 else if (skb->pkt_type == PACKET_OUTGOING) {
2183 /* Special case: outgoing packets have ll header at head */
2184 skb_pull(skb, skb_network_offset(skb));
2190 res = run_filter(skb, sk, snaplen);
2192 goto drop_n_restore;
2194 /* If we are flooded, just give up */
2195 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2196 atomic_inc(&po->tp_drops);
2197 goto drop_n_restore;
2200 if (skb->ip_summed == CHECKSUM_PARTIAL)
2201 status |= TP_STATUS_CSUMNOTREADY;
2202 else if (skb->pkt_type != PACKET_OUTGOING &&
2203 (skb->ip_summed == CHECKSUM_COMPLETE ||
2204 skb_csum_unnecessary(skb)))
2205 status |= TP_STATUS_CSUM_VALID;
2210 if (sk->sk_type == SOCK_DGRAM) {
2211 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2214 unsigned int maclen = skb_network_offset(skb);
2215 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2216 (maclen < 16 ? 16 : maclen)) +
2218 if (po->has_vnet_hdr) {
2219 netoff += sizeof(struct virtio_net_hdr);
2222 macoff = netoff - maclen;
2224 if (po->tp_version <= TPACKET_V2) {
2225 if (macoff + snaplen > po->rx_ring.frame_size) {
2226 if (po->copy_thresh &&
2227 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2228 if (skb_shared(skb)) {
2229 copy_skb = skb_clone(skb, GFP_ATOMIC);
2231 copy_skb = skb_get(skb);
2232 skb_head = skb->data;
2235 skb_set_owner_r(copy_skb, sk);
2237 snaplen = po->rx_ring.frame_size - macoff;
2238 if ((int)snaplen < 0) {
2243 } else if (unlikely(macoff + snaplen >
2244 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2247 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2248 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2249 snaplen, nval, macoff);
2251 if (unlikely((int)snaplen < 0)) {
2253 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2257 spin_lock(&sk->sk_receive_queue.lock);
2258 h.raw = packet_current_rx_frame(po, skb,
2259 TP_STATUS_KERNEL, (macoff+snaplen));
2261 goto drop_n_account;
2262 if (po->tp_version <= TPACKET_V2) {
2263 packet_increment_rx_head(po, &po->rx_ring);
2265 * LOSING will be reported till you read the stats,
2266 * because it's COR - Clear On Read.
2267 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2270 if (atomic_read(&po->tp_drops))
2271 status |= TP_STATUS_LOSING;
2275 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2276 sizeof(struct virtio_net_hdr),
2278 goto drop_n_account;
2280 po->stats.stats1.tp_packets++;
2282 status |= TP_STATUS_COPY;
2283 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2285 spin_unlock(&sk->sk_receive_queue.lock);
2287 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2289 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2290 getnstimeofday(&ts);
2292 status |= ts_status;
2294 switch (po->tp_version) {
2296 h.h1->tp_len = skb->len;
2297 h.h1->tp_snaplen = snaplen;
2298 h.h1->tp_mac = macoff;
2299 h.h1->tp_net = netoff;
2300 h.h1->tp_sec = ts.tv_sec;
2301 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2302 hdrlen = sizeof(*h.h1);
2305 h.h2->tp_len = skb->len;
2306 h.h2->tp_snaplen = snaplen;
2307 h.h2->tp_mac = macoff;
2308 h.h2->tp_net = netoff;
2309 h.h2->tp_sec = ts.tv_sec;
2310 h.h2->tp_nsec = ts.tv_nsec;
2311 if (skb_vlan_tag_present(skb)) {
2312 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2313 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2314 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2316 h.h2->tp_vlan_tci = 0;
2317 h.h2->tp_vlan_tpid = 0;
2319 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2320 hdrlen = sizeof(*h.h2);
2323 /* tp_nxt_offset,vlan are already populated above.
2324 * So DONT clear those fields here
2326 h.h3->tp_status |= status;
2327 h.h3->tp_len = skb->len;
2328 h.h3->tp_snaplen = snaplen;
2329 h.h3->tp_mac = macoff;
2330 h.h3->tp_net = netoff;
2331 h.h3->tp_sec = ts.tv_sec;
2332 h.h3->tp_nsec = ts.tv_nsec;
2333 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2334 hdrlen = sizeof(*h.h3);
2340 sll = h.raw + TPACKET_ALIGN(hdrlen);
2341 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2342 sll->sll_family = AF_PACKET;
2343 sll->sll_hatype = dev->type;
2344 sll->sll_protocol = skb->protocol;
2345 sll->sll_pkttype = skb->pkt_type;
2346 if (unlikely(po->origdev))
2347 sll->sll_ifindex = orig_dev->ifindex;
2349 sll->sll_ifindex = dev->ifindex;
2353 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2354 if (po->tp_version <= TPACKET_V2) {
2357 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2360 for (start = h.raw; start < end; start += PAGE_SIZE)
2361 flush_dcache_page(pgv_to_page(start));
2366 if (po->tp_version <= TPACKET_V2) {
2367 __packet_set_status(po, h.raw, status);
2368 sk->sk_data_ready(sk);
2370 prb_clear_blk_fill_status(&po->rx_ring);
2374 if (skb_head != skb->data && skb_shared(skb)) {
2375 skb->data = skb_head;
2379 if (!is_drop_n_account)
2386 spin_unlock(&sk->sk_receive_queue.lock);
2387 atomic_inc(&po->tp_drops);
2388 is_drop_n_account = true;
2390 sk->sk_data_ready(sk);
2391 kfree_skb(copy_skb);
2392 goto drop_n_restore;
2395 static void tpacket_destruct_skb(struct sk_buff *skb)
2397 struct packet_sock *po = pkt_sk(skb->sk);
2399 if (likely(po->tx_ring.pg_vec)) {
2403 ph = skb_zcopy_get_nouarg(skb);
2404 packet_dec_pending(&po->tx_ring);
2406 ts = __packet_set_timestamp(po, ph, skb);
2407 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2413 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2415 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2416 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2417 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2418 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2419 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2420 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2421 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2423 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2429 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2430 struct virtio_net_hdr *vnet_hdr)
2432 if (*len < sizeof(*vnet_hdr))
2434 *len -= sizeof(*vnet_hdr);
2436 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2439 return __packet_snd_vnet_parse(vnet_hdr, *len);
2442 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2443 void *frame, struct net_device *dev, void *data, int tp_len,
2444 __be16 proto, unsigned char *addr, int hlen, int copylen,
2445 const struct sockcm_cookie *sockc)
2447 union tpacket_uhdr ph;
2448 int to_write, offset, len, nr_frags, len_max;
2449 struct socket *sock = po->sk.sk_socket;
2455 skb->protocol = proto;
2457 skb->priority = po->sk.sk_priority;
2458 skb->mark = po->sk.sk_mark;
2459 skb->tstamp = sockc->transmit_time;
2460 skb_setup_tx_timestamp(skb, sockc->tsflags);
2461 skb_zcopy_set_nouarg(skb, ph.raw);
2463 skb_reserve(skb, hlen);
2464 skb_reset_network_header(skb);
2468 if (sock->type == SOCK_DGRAM) {
2469 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2471 if (unlikely(err < 0))
2473 } else if (copylen) {
2474 int hdrlen = min_t(int, copylen, tp_len);
2476 skb_push(skb, dev->hard_header_len);
2477 skb_put(skb, copylen - dev->hard_header_len);
2478 err = skb_store_bits(skb, 0, data, hdrlen);
2481 if (!dev_validate_header(dev, skb->data, hdrlen))
2488 offset = offset_in_page(data);
2489 len_max = PAGE_SIZE - offset;
2490 len = ((to_write > len_max) ? len_max : to_write);
2492 skb->data_len = to_write;
2493 skb->len += to_write;
2494 skb->truesize += to_write;
2495 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2497 while (likely(to_write)) {
2498 nr_frags = skb_shinfo(skb)->nr_frags;
2500 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2501 pr_err("Packet exceed the number of skb frags(%lu)\n",
2506 page = pgv_to_page(data);
2508 flush_dcache_page(page);
2510 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2513 len_max = PAGE_SIZE;
2514 len = ((to_write > len_max) ? len_max : to_write);
2517 packet_parse_headers(skb, sock);
2522 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2523 int size_max, void **data)
2525 union tpacket_uhdr ph;
2530 switch (po->tp_version) {
2532 if (ph.h3->tp_next_offset != 0) {
2533 pr_warn_once("variable sized slot not supported");
2536 tp_len = ph.h3->tp_len;
2539 tp_len = ph.h2->tp_len;
2542 tp_len = ph.h1->tp_len;
2545 if (unlikely(tp_len > size_max)) {
2546 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2550 if (unlikely(po->tp_tx_has_off)) {
2551 int off_min, off_max;
2553 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2554 off_max = po->tx_ring.frame_size - tp_len;
2555 if (po->sk.sk_type == SOCK_DGRAM) {
2556 switch (po->tp_version) {
2558 off = ph.h3->tp_net;
2561 off = ph.h2->tp_net;
2564 off = ph.h1->tp_net;
2568 switch (po->tp_version) {
2570 off = ph.h3->tp_mac;
2573 off = ph.h2->tp_mac;
2576 off = ph.h1->tp_mac;
2580 if (unlikely((off < off_min) || (off_max < off)))
2583 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2586 *data = frame + off;
2590 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2592 struct sk_buff *skb;
2593 struct net_device *dev;
2594 struct virtio_net_hdr *vnet_hdr = NULL;
2595 struct sockcm_cookie sockc;
2597 int err, reserve = 0;
2599 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2600 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2601 unsigned char *addr = NULL;
2602 int tp_len, size_max;
2605 int status = TP_STATUS_AVAILABLE;
2606 int hlen, tlen, copylen = 0;
2608 mutex_lock(&po->pg_vec_lock);
2610 if (likely(saddr == NULL)) {
2611 dev = packet_cached_dev_get(po);
2615 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2617 if (msg->msg_namelen < (saddr->sll_halen
2618 + offsetof(struct sockaddr_ll,
2621 proto = saddr->sll_protocol;
2622 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2623 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2624 if (dev && msg->msg_namelen < dev->addr_len +
2625 offsetof(struct sockaddr_ll, sll_addr))
2627 addr = saddr->sll_addr;
2632 if (unlikely(dev == NULL))
2635 if (unlikely(!(dev->flags & IFF_UP)))
2638 sockcm_init(&sockc, &po->sk);
2639 if (msg->msg_controllen) {
2640 err = sock_cmsg_send(&po->sk, msg, &sockc);
2645 if (po->sk.sk_socket->type == SOCK_RAW)
2646 reserve = dev->hard_header_len;
2647 size_max = po->tx_ring.frame_size
2648 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2650 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2651 size_max = dev->mtu + reserve + VLAN_HLEN;
2654 ph = packet_current_frame(po, &po->tx_ring,
2655 TP_STATUS_SEND_REQUEST);
2656 if (unlikely(ph == NULL)) {
2657 if (need_wait && need_resched())
2663 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2667 status = TP_STATUS_SEND_REQUEST;
2668 hlen = LL_RESERVED_SPACE(dev);
2669 tlen = dev->needed_tailroom;
2670 if (po->has_vnet_hdr) {
2672 data += sizeof(*vnet_hdr);
2673 tp_len -= sizeof(*vnet_hdr);
2675 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2679 copylen = __virtio16_to_cpu(vio_le(),
2682 copylen = max_t(int, copylen, dev->hard_header_len);
2683 skb = sock_alloc_send_skb(&po->sk,
2684 hlen + tlen + sizeof(struct sockaddr_ll) +
2685 (copylen - dev->hard_header_len),
2688 if (unlikely(skb == NULL)) {
2689 /* we assume the socket was initially writeable ... */
2690 if (likely(len_sum > 0))
2694 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2695 addr, hlen, copylen, &sockc);
2696 if (likely(tp_len >= 0) &&
2697 tp_len > dev->mtu + reserve &&
2698 !po->has_vnet_hdr &&
2699 !packet_extra_vlan_len_allowed(dev, skb))
2702 if (unlikely(tp_len < 0)) {
2705 __packet_set_status(po, ph,
2706 TP_STATUS_AVAILABLE);
2707 packet_increment_head(&po->tx_ring);
2711 status = TP_STATUS_WRONG_FORMAT;
2717 if (po->has_vnet_hdr) {
2718 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2722 virtio_net_hdr_set_proto(skb, vnet_hdr);
2725 skb->destructor = tpacket_destruct_skb;
2726 __packet_set_status(po, ph, TP_STATUS_SENDING);
2727 packet_inc_pending(&po->tx_ring);
2729 status = TP_STATUS_SEND_REQUEST;
2730 err = po->xmit(skb);
2731 if (unlikely(err > 0)) {
2732 err = net_xmit_errno(err);
2733 if (err && __packet_get_status(po, ph) ==
2734 TP_STATUS_AVAILABLE) {
2735 /* skb was destructed already */
2740 * skb was dropped but not destructed yet;
2741 * let's treat it like congestion or err < 0
2745 packet_increment_head(&po->tx_ring);
2747 } while (likely((ph != NULL) ||
2748 /* Note: packet_read_pending() might be slow if we have
2749 * to call it as it's per_cpu variable, but in fast-path
2750 * we already short-circuit the loop with the first
2751 * condition, and luckily don't have to go that path
2754 (need_wait && packet_read_pending(&po->tx_ring))));
2760 __packet_set_status(po, ph, status);
2765 mutex_unlock(&po->pg_vec_lock);
2769 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2770 size_t reserve, size_t len,
2771 size_t linear, int noblock,
2774 struct sk_buff *skb;
2776 /* Under a page? Don't bother with paged skb. */
2777 if (prepad + len < PAGE_SIZE || !linear)
2780 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2785 skb_reserve(skb, reserve);
2786 skb_put(skb, linear);
2787 skb->data_len = len - linear;
2788 skb->len += len - linear;
2793 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2795 struct sock *sk = sock->sk;
2796 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2797 struct sk_buff *skb;
2798 struct net_device *dev;
2800 unsigned char *addr = NULL;
2801 int err, reserve = 0;
2802 struct sockcm_cookie sockc;
2803 struct virtio_net_hdr vnet_hdr = { 0 };
2805 struct packet_sock *po = pkt_sk(sk);
2806 bool has_vnet_hdr = false;
2807 int hlen, tlen, linear;
2811 * Get and verify the address.
2814 if (likely(saddr == NULL)) {
2815 dev = packet_cached_dev_get(po);
2819 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2821 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2823 proto = saddr->sll_protocol;
2824 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2825 if (sock->type == SOCK_DGRAM) {
2826 if (dev && msg->msg_namelen < dev->addr_len +
2827 offsetof(struct sockaddr_ll, sll_addr))
2829 addr = saddr->sll_addr;
2834 if (unlikely(dev == NULL))
2837 if (unlikely(!(dev->flags & IFF_UP)))
2840 sockcm_init(&sockc, sk);
2841 sockc.mark = sk->sk_mark;
2842 if (msg->msg_controllen) {
2843 err = sock_cmsg_send(sk, msg, &sockc);
2848 if (sock->type == SOCK_RAW)
2849 reserve = dev->hard_header_len;
2850 if (po->has_vnet_hdr) {
2851 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2854 has_vnet_hdr = true;
2857 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2858 if (!netif_supports_nofcs(dev)) {
2859 err = -EPROTONOSUPPORT;
2862 extra_len = 4; /* We're doing our own CRC */
2866 if (!vnet_hdr.gso_type &&
2867 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2871 hlen = LL_RESERVED_SPACE(dev);
2872 tlen = dev->needed_tailroom;
2873 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2874 linear = max(linear, min_t(int, len, dev->hard_header_len));
2875 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2876 msg->msg_flags & MSG_DONTWAIT, &err);
2880 skb_reset_network_header(skb);
2883 if (sock->type == SOCK_DGRAM) {
2884 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2885 if (unlikely(offset < 0))
2887 } else if (reserve) {
2888 skb_reserve(skb, -reserve);
2889 if (len < reserve + sizeof(struct ipv6hdr) &&
2890 dev->min_header_len != dev->hard_header_len)
2891 skb_reset_network_header(skb);
2894 /* Returns -EFAULT on error */
2895 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2899 if (sock->type == SOCK_RAW &&
2900 !dev_validate_header(dev, skb->data, len)) {
2905 skb_setup_tx_timestamp(skb, sockc.tsflags);
2907 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2908 !packet_extra_vlan_len_allowed(dev, skb)) {
2913 skb->protocol = proto;
2915 skb->priority = sk->sk_priority;
2916 skb->mark = sockc.mark;
2917 skb->tstamp = sockc.transmit_time;
2920 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2923 len += sizeof(vnet_hdr);
2924 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2927 packet_parse_headers(skb, sock);
2929 if (unlikely(extra_len == 4))
2932 err = po->xmit(skb);
2933 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2949 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2951 struct sock *sk = sock->sk;
2952 struct packet_sock *po = pkt_sk(sk);
2954 if (po->tx_ring.pg_vec)
2955 return tpacket_snd(po, msg);
2957 return packet_snd(sock, msg, len);
2961 * Close a PACKET socket. This is fairly simple. We immediately go
2962 * to 'closed' state and remove our protocol entry in the device list.
2965 static int packet_release(struct socket *sock)
2967 struct sock *sk = sock->sk;
2968 struct packet_sock *po;
2969 struct packet_fanout *f;
2971 union tpacket_req_u req_u;
2979 mutex_lock(&net->packet.sklist_lock);
2980 sk_del_node_init_rcu(sk);
2981 mutex_unlock(&net->packet.sklist_lock);
2984 sock_prot_inuse_add(net, sk->sk_prot, -1);
2987 spin_lock(&po->bind_lock);
2988 unregister_prot_hook(sk, false);
2989 packet_cached_dev_reset(po);
2991 if (po->prot_hook.dev) {
2992 dev_put(po->prot_hook.dev);
2993 po->prot_hook.dev = NULL;
2995 spin_unlock(&po->bind_lock);
2997 packet_flush_mclist(sk);
3000 if (po->rx_ring.pg_vec) {
3001 memset(&req_u, 0, sizeof(req_u));
3002 packet_set_ring(sk, &req_u, 1, 0);
3005 if (po->tx_ring.pg_vec) {
3006 memset(&req_u, 0, sizeof(req_u));
3007 packet_set_ring(sk, &req_u, 1, 1);
3011 f = fanout_release(sk);
3015 kfree(po->rollover);
3017 fanout_release_data(f);
3021 * Now the socket is dead. No more input will appear.
3028 skb_queue_purge(&sk->sk_receive_queue);
3029 packet_free_pending(po);
3030 sk_refcnt_debug_release(sk);
3037 * Attach a packet hook.
3040 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3043 struct packet_sock *po = pkt_sk(sk);
3044 struct net_device *dev_curr;
3047 struct net_device *dev = NULL;
3049 bool unlisted = false;
3052 spin_lock(&po->bind_lock);
3061 dev = dev_get_by_name_rcu(sock_net(sk), name);
3066 } else if (ifindex) {
3067 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3077 proto_curr = po->prot_hook.type;
3078 dev_curr = po->prot_hook.dev;
3080 need_rehook = proto_curr != proto || dev_curr != dev;
3085 /* prevents packet_notifier() from calling
3086 * register_prot_hook()
3089 __unregister_prot_hook(sk, true);
3091 dev_curr = po->prot_hook.dev;
3093 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3097 BUG_ON(po->running);
3099 po->prot_hook.type = proto;
3101 if (unlikely(unlisted)) {
3103 po->prot_hook.dev = NULL;
3105 packet_cached_dev_reset(po);
3107 po->prot_hook.dev = dev;
3108 po->ifindex = dev ? dev->ifindex : 0;
3109 packet_cached_dev_assign(po, dev);
3115 if (proto == 0 || !need_rehook)
3118 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3119 register_prot_hook(sk);
3121 sk->sk_err = ENETDOWN;
3122 if (!sock_flag(sk, SOCK_DEAD))
3123 sk->sk_error_report(sk);
3128 spin_unlock(&po->bind_lock);
3134 * Bind a packet socket to a device
3137 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3140 struct sock *sk = sock->sk;
3141 char name[sizeof(uaddr->sa_data) + 1];
3147 if (addr_len != sizeof(struct sockaddr))
3149 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3152 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3153 name[sizeof(uaddr->sa_data)] = 0;
3155 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3158 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3160 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3161 struct sock *sk = sock->sk;
3167 if (addr_len < sizeof(struct sockaddr_ll))
3169 if (sll->sll_family != AF_PACKET)
3172 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3173 sll->sll_protocol ? : pkt_sk(sk)->num);
3176 static struct proto packet_proto = {
3178 .owner = THIS_MODULE,
3179 .obj_size = sizeof(struct packet_sock),
3183 * Create a packet of type SOCK_PACKET.
3186 static int packet_create(struct net *net, struct socket *sock, int protocol,
3190 struct packet_sock *po;
3191 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3194 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3196 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3197 sock->type != SOCK_PACKET)
3198 return -ESOCKTNOSUPPORT;
3200 sock->state = SS_UNCONNECTED;
3203 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3207 sock->ops = &packet_ops;
3208 if (sock->type == SOCK_PACKET)
3209 sock->ops = &packet_ops_spkt;
3211 sock_init_data(sock, sk);
3214 sk->sk_family = PF_PACKET;
3216 po->xmit = dev_queue_xmit;
3218 err = packet_alloc_pending(po);
3222 packet_cached_dev_reset(po);
3224 sk->sk_destruct = packet_sock_destruct;
3225 sk_refcnt_debug_inc(sk);
3228 * Attach a protocol block
3231 spin_lock_init(&po->bind_lock);
3232 mutex_init(&po->pg_vec_lock);
3233 po->rollover = NULL;
3234 po->prot_hook.func = packet_rcv;
3236 if (sock->type == SOCK_PACKET)
3237 po->prot_hook.func = packet_rcv_spkt;
3239 po->prot_hook.af_packet_priv = sk;
3242 po->prot_hook.type = proto;
3243 __register_prot_hook(sk);
3246 mutex_lock(&net->packet.sklist_lock);
3247 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3248 mutex_unlock(&net->packet.sklist_lock);
3251 sock_prot_inuse_add(net, &packet_proto, 1);
3262 * Pull a packet from our receive queue and hand it to the user.
3263 * If necessary we block.
3266 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3269 struct sock *sk = sock->sk;
3270 struct sk_buff *skb;
3272 int vnet_hdr_len = 0;
3273 unsigned int origlen = 0;
3276 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3280 /* What error should we return now? EUNATTACH? */
3281 if (pkt_sk(sk)->ifindex < 0)
3285 if (flags & MSG_ERRQUEUE) {
3286 err = sock_recv_errqueue(sk, msg, len,
3287 SOL_PACKET, PACKET_TX_TIMESTAMP);
3292 * Call the generic datagram receiver. This handles all sorts
3293 * of horrible races and re-entrancy so we can forget about it
3294 * in the protocol layers.
3296 * Now it will return ENETDOWN, if device have just gone down,
3297 * but then it will block.
3300 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3303 * An error occurred so return it. Because skb_recv_datagram()
3304 * handles the blocking we don't see and worry about blocking
3311 if (READ_ONCE(pkt_sk(sk)->pressure))
3312 packet_rcv_has_room(pkt_sk(sk), NULL);
3314 if (pkt_sk(sk)->has_vnet_hdr) {
3315 err = packet_rcv_vnet(msg, skb, &len);
3318 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3321 /* You lose any data beyond the buffer you gave. If it worries
3322 * a user program they can ask the device for its MTU
3328 msg->msg_flags |= MSG_TRUNC;
3331 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3335 if (sock->type != SOCK_PACKET) {
3336 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3338 /* Original length was stored in sockaddr_ll fields */
3339 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3340 sll->sll_family = AF_PACKET;
3341 sll->sll_protocol = skb->protocol;
3344 sock_recv_ts_and_drops(msg, sk, skb);
3346 if (msg->msg_name) {
3349 /* If the address length field is there to be filled
3350 * in, we fill it in now.
3352 if (sock->type == SOCK_PACKET) {
3353 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3354 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3355 copy_len = msg->msg_namelen;
3357 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3359 msg->msg_namelen = sll->sll_halen +
3360 offsetof(struct sockaddr_ll, sll_addr);
3361 copy_len = msg->msg_namelen;
3362 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3363 memset(msg->msg_name +
3364 offsetof(struct sockaddr_ll, sll_addr),
3365 0, sizeof(sll->sll_addr));
3366 msg->msg_namelen = sizeof(struct sockaddr_ll);
3369 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3372 if (pkt_sk(sk)->auxdata) {
3373 struct tpacket_auxdata aux;
3375 aux.tp_status = TP_STATUS_USER;
3376 if (skb->ip_summed == CHECKSUM_PARTIAL)
3377 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3378 else if (skb->pkt_type != PACKET_OUTGOING &&
3379 (skb->ip_summed == CHECKSUM_COMPLETE ||
3380 skb_csum_unnecessary(skb)))
3381 aux.tp_status |= TP_STATUS_CSUM_VALID;
3383 aux.tp_len = origlen;
3384 aux.tp_snaplen = skb->len;
3386 aux.tp_net = skb_network_offset(skb);
3387 if (skb_vlan_tag_present(skb)) {
3388 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3389 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3390 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3392 aux.tp_vlan_tci = 0;
3393 aux.tp_vlan_tpid = 0;
3395 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3399 * Free or return the buffer as appropriate. Again this
3400 * hides all the races and re-entrancy issues from us.
3402 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3405 skb_free_datagram(sk, skb);
3410 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3413 struct net_device *dev;
3414 struct sock *sk = sock->sk;
3419 uaddr->sa_family = AF_PACKET;
3420 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3422 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3424 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3427 return sizeof(*uaddr);
3430 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3433 struct net_device *dev;
3434 struct sock *sk = sock->sk;
3435 struct packet_sock *po = pkt_sk(sk);
3436 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3441 sll->sll_family = AF_PACKET;
3442 sll->sll_ifindex = po->ifindex;
3443 sll->sll_protocol = po->num;
3444 sll->sll_pkttype = 0;
3446 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3448 sll->sll_hatype = dev->type;
3449 sll->sll_halen = dev->addr_len;
3450 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3452 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3457 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3460 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3464 case PACKET_MR_MULTICAST:
3465 if (i->alen != dev->addr_len)
3468 return dev_mc_add(dev, i->addr);
3470 return dev_mc_del(dev, i->addr);
3472 case PACKET_MR_PROMISC:
3473 return dev_set_promiscuity(dev, what);
3474 case PACKET_MR_ALLMULTI:
3475 return dev_set_allmulti(dev, what);
3476 case PACKET_MR_UNICAST:
3477 if (i->alen != dev->addr_len)
3480 return dev_uc_add(dev, i->addr);
3482 return dev_uc_del(dev, i->addr);
3490 static void packet_dev_mclist_delete(struct net_device *dev,
3491 struct packet_mclist **mlp)
3493 struct packet_mclist *ml;
3495 while ((ml = *mlp) != NULL) {
3496 if (ml->ifindex == dev->ifindex) {
3497 packet_dev_mc(dev, ml, -1);
3505 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3507 struct packet_sock *po = pkt_sk(sk);
3508 struct packet_mclist *ml, *i;
3509 struct net_device *dev;
3515 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3520 if (mreq->mr_alen > dev->addr_len)
3524 i = kmalloc(sizeof(*i), GFP_KERNEL);
3529 for (ml = po->mclist; ml; ml = ml->next) {
3530 if (ml->ifindex == mreq->mr_ifindex &&
3531 ml->type == mreq->mr_type &&
3532 ml->alen == mreq->mr_alen &&
3533 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3535 /* Free the new element ... */
3541 i->type = mreq->mr_type;
3542 i->ifindex = mreq->mr_ifindex;
3543 i->alen = mreq->mr_alen;
3544 memcpy(i->addr, mreq->mr_address, i->alen);
3545 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3547 i->next = po->mclist;
3549 err = packet_dev_mc(dev, i, 1);
3551 po->mclist = i->next;
3560 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3562 struct packet_mclist *ml, **mlp;
3566 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3567 if (ml->ifindex == mreq->mr_ifindex &&
3568 ml->type == mreq->mr_type &&
3569 ml->alen == mreq->mr_alen &&
3570 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3571 if (--ml->count == 0) {
3572 struct net_device *dev;
3574 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3576 packet_dev_mc(dev, ml, -1);
3586 static void packet_flush_mclist(struct sock *sk)
3588 struct packet_sock *po = pkt_sk(sk);
3589 struct packet_mclist *ml;
3595 while ((ml = po->mclist) != NULL) {
3596 struct net_device *dev;
3598 po->mclist = ml->next;
3599 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3601 packet_dev_mc(dev, ml, -1);
3608 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3610 struct sock *sk = sock->sk;
3611 struct packet_sock *po = pkt_sk(sk);
3614 if (level != SOL_PACKET)
3615 return -ENOPROTOOPT;
3618 case PACKET_ADD_MEMBERSHIP:
3619 case PACKET_DROP_MEMBERSHIP:
3621 struct packet_mreq_max mreq;
3623 memset(&mreq, 0, sizeof(mreq));
3624 if (len < sizeof(struct packet_mreq))
3626 if (len > sizeof(mreq))
3628 if (copy_from_user(&mreq, optval, len))
3630 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3632 if (optname == PACKET_ADD_MEMBERSHIP)
3633 ret = packet_mc_add(sk, &mreq);
3635 ret = packet_mc_drop(sk, &mreq);
3639 case PACKET_RX_RING:
3640 case PACKET_TX_RING:
3642 union tpacket_req_u req_u;
3646 switch (po->tp_version) {
3649 len = sizeof(req_u.req);
3653 len = sizeof(req_u.req3);
3659 if (copy_from_user(&req_u.req, optval, len))
3662 ret = packet_set_ring(sk, &req_u, 0,
3663 optname == PACKET_TX_RING);
3668 case PACKET_COPY_THRESH:
3672 if (optlen != sizeof(val))
3674 if (copy_from_user(&val, optval, sizeof(val)))
3677 pkt_sk(sk)->copy_thresh = val;
3680 case PACKET_VERSION:
3684 if (optlen != sizeof(val))
3686 if (copy_from_user(&val, optval, sizeof(val)))
3697 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3700 po->tp_version = val;
3706 case PACKET_RESERVE:
3710 if (optlen != sizeof(val))
3712 if (copy_from_user(&val, optval, sizeof(val)))
3717 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3720 po->tp_reserve = val;
3730 if (optlen != sizeof(val))
3732 if (copy_from_user(&val, optval, sizeof(val)))
3736 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3739 po->tp_loss = !!val;
3745 case PACKET_AUXDATA:
3749 if (optlen < sizeof(val))
3751 if (copy_from_user(&val, optval, sizeof(val)))
3755 po->auxdata = !!val;
3759 case PACKET_ORIGDEV:
3763 if (optlen < sizeof(val))
3765 if (copy_from_user(&val, optval, sizeof(val)))
3769 po->origdev = !!val;
3773 case PACKET_VNET_HDR:
3777 if (sock->type != SOCK_RAW)
3779 if (optlen < sizeof(val))
3781 if (copy_from_user(&val, optval, sizeof(val)))
3785 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3788 po->has_vnet_hdr = !!val;
3794 case PACKET_TIMESTAMP:
3798 if (optlen != sizeof(val))
3800 if (copy_from_user(&val, optval, sizeof(val)))
3803 po->tp_tstamp = val;
3810 if (optlen != sizeof(val))
3812 if (copy_from_user(&val, optval, sizeof(val)))
3815 return fanout_add(sk, val & 0xffff, val >> 16);
3817 case PACKET_FANOUT_DATA:
3822 return fanout_set_data(po, optval, optlen);
3824 case PACKET_IGNORE_OUTGOING:
3828 if (optlen != sizeof(val))
3830 if (copy_from_user(&val, optval, sizeof(val)))
3832 if (val < 0 || val > 1)
3835 po->prot_hook.ignore_outgoing = !!val;
3838 case PACKET_TX_HAS_OFF:
3842 if (optlen != sizeof(val))
3844 if (copy_from_user(&val, optval, sizeof(val)))
3848 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3851 po->tp_tx_has_off = !!val;
3857 case PACKET_QDISC_BYPASS:
3861 if (optlen != sizeof(val))
3863 if (copy_from_user(&val, optval, sizeof(val)))
3866 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3870 return -ENOPROTOOPT;
3874 static int packet_getsockopt(struct socket *sock, int level, int optname,
3875 char __user *optval, int __user *optlen)
3878 int val, lv = sizeof(val);
3879 struct sock *sk = sock->sk;
3880 struct packet_sock *po = pkt_sk(sk);
3882 union tpacket_stats_u st;
3883 struct tpacket_rollover_stats rstats;
3886 if (level != SOL_PACKET)
3887 return -ENOPROTOOPT;
3889 if (get_user(len, optlen))
3896 case PACKET_STATISTICS:
3897 spin_lock_bh(&sk->sk_receive_queue.lock);
3898 memcpy(&st, &po->stats, sizeof(st));
3899 memset(&po->stats, 0, sizeof(po->stats));
3900 spin_unlock_bh(&sk->sk_receive_queue.lock);
3901 drops = atomic_xchg(&po->tp_drops, 0);
3903 if (po->tp_version == TPACKET_V3) {
3904 lv = sizeof(struct tpacket_stats_v3);
3905 st.stats3.tp_drops = drops;
3906 st.stats3.tp_packets += drops;
3909 lv = sizeof(struct tpacket_stats);
3910 st.stats1.tp_drops = drops;
3911 st.stats1.tp_packets += drops;
3916 case PACKET_AUXDATA:
3919 case PACKET_ORIGDEV:
3922 case PACKET_VNET_HDR:
3923 val = po->has_vnet_hdr;
3925 case PACKET_VERSION:
3926 val = po->tp_version;
3929 if (len > sizeof(int))
3931 if (len < sizeof(int))
3933 if (copy_from_user(&val, optval, len))
3937 val = sizeof(struct tpacket_hdr);
3940 val = sizeof(struct tpacket2_hdr);
3943 val = sizeof(struct tpacket3_hdr);
3949 case PACKET_RESERVE:
3950 val = po->tp_reserve;
3955 case PACKET_TIMESTAMP:
3956 val = po->tp_tstamp;
3960 ((u32)po->fanout->id |
3961 ((u32)po->fanout->type << 16) |
3962 ((u32)po->fanout->flags << 24)) :
3965 case PACKET_IGNORE_OUTGOING:
3966 val = po->prot_hook.ignore_outgoing;
3968 case PACKET_ROLLOVER_STATS:
3971 rstats.tp_all = atomic_long_read(&po->rollover->num);
3972 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3973 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3975 lv = sizeof(rstats);
3977 case PACKET_TX_HAS_OFF:
3978 val = po->tp_tx_has_off;
3980 case PACKET_QDISC_BYPASS:
3981 val = packet_use_direct_xmit(po);
3984 return -ENOPROTOOPT;
3989 if (put_user(len, optlen))
3991 if (copy_to_user(optval, data, len))
3997 #ifdef CONFIG_COMPAT
3998 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3999 char __user *optval, unsigned int optlen)
4001 struct packet_sock *po = pkt_sk(sock->sk);
4003 if (level != SOL_PACKET)
4004 return -ENOPROTOOPT;
4006 if (optname == PACKET_FANOUT_DATA &&
4007 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4008 optval = (char __user *)get_compat_bpf_fprog(optval);
4011 optlen = sizeof(struct sock_fprog);
4014 return packet_setsockopt(sock, level, optname, optval, optlen);
4018 static int packet_notifier(struct notifier_block *this,
4019 unsigned long msg, void *ptr)
4022 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4023 struct net *net = dev_net(dev);
4026 sk_for_each_rcu(sk, &net->packet.sklist) {
4027 struct packet_sock *po = pkt_sk(sk);
4030 case NETDEV_UNREGISTER:
4032 packet_dev_mclist_delete(dev, &po->mclist);
4036 if (dev->ifindex == po->ifindex) {
4037 spin_lock(&po->bind_lock);
4039 __unregister_prot_hook(sk, false);
4040 sk->sk_err = ENETDOWN;
4041 if (!sock_flag(sk, SOCK_DEAD))
4042 sk->sk_error_report(sk);
4044 if (msg == NETDEV_UNREGISTER) {
4045 packet_cached_dev_reset(po);
4047 if (po->prot_hook.dev)
4048 dev_put(po->prot_hook.dev);
4049 po->prot_hook.dev = NULL;
4051 spin_unlock(&po->bind_lock);
4055 if (dev->ifindex == po->ifindex) {
4056 spin_lock(&po->bind_lock);
4058 register_prot_hook(sk);
4059 spin_unlock(&po->bind_lock);
4069 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4072 struct sock *sk = sock->sk;
4077 int amount = sk_wmem_alloc_get(sk);
4079 return put_user(amount, (int __user *)arg);
4083 struct sk_buff *skb;
4086 spin_lock_bh(&sk->sk_receive_queue.lock);
4087 skb = skb_peek(&sk->sk_receive_queue);
4090 spin_unlock_bh(&sk->sk_receive_queue.lock);
4091 return put_user(amount, (int __user *)arg);
4101 case SIOCGIFBRDADDR:
4102 case SIOCSIFBRDADDR:
4103 case SIOCGIFNETMASK:
4104 case SIOCSIFNETMASK:
4105 case SIOCGIFDSTADDR:
4106 case SIOCSIFDSTADDR:
4108 return inet_dgram_ops.ioctl(sock, cmd, arg);
4112 return -ENOIOCTLCMD;
4117 static __poll_t packet_poll(struct file *file, struct socket *sock,
4120 struct sock *sk = sock->sk;
4121 struct packet_sock *po = pkt_sk(sk);
4122 __poll_t mask = datagram_poll(file, sock, wait);
4124 spin_lock_bh(&sk->sk_receive_queue.lock);
4125 if (po->rx_ring.pg_vec) {
4126 if (!packet_previous_rx_frame(po, &po->rx_ring,
4128 mask |= EPOLLIN | EPOLLRDNORM;
4130 if (READ_ONCE(po->pressure) && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4131 WRITE_ONCE(po->pressure, 0);
4132 spin_unlock_bh(&sk->sk_receive_queue.lock);
4133 spin_lock_bh(&sk->sk_write_queue.lock);
4134 if (po->tx_ring.pg_vec) {
4135 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4136 mask |= EPOLLOUT | EPOLLWRNORM;
4138 spin_unlock_bh(&sk->sk_write_queue.lock);
4143 /* Dirty? Well, I still did not learn better way to account
4147 static void packet_mm_open(struct vm_area_struct *vma)
4149 struct file *file = vma->vm_file;
4150 struct socket *sock = file->private_data;
4151 struct sock *sk = sock->sk;
4154 atomic_inc(&pkt_sk(sk)->mapped);
4157 static void packet_mm_close(struct vm_area_struct *vma)
4159 struct file *file = vma->vm_file;
4160 struct socket *sock = file->private_data;
4161 struct sock *sk = sock->sk;
4164 atomic_dec(&pkt_sk(sk)->mapped);
4167 static const struct vm_operations_struct packet_mmap_ops = {
4168 .open = packet_mm_open,
4169 .close = packet_mm_close,
4172 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4177 for (i = 0; i < len; i++) {
4178 if (likely(pg_vec[i].buffer)) {
4179 if (is_vmalloc_addr(pg_vec[i].buffer))
4180 vfree(pg_vec[i].buffer);
4182 free_pages((unsigned long)pg_vec[i].buffer,
4184 pg_vec[i].buffer = NULL;
4190 static char *alloc_one_pg_vec_page(unsigned long order)
4193 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4194 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4196 buffer = (char *) __get_free_pages(gfp_flags, order);
4200 /* __get_free_pages failed, fall back to vmalloc */
4201 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4205 /* vmalloc failed, lets dig into swap here */
4206 gfp_flags &= ~__GFP_NORETRY;
4207 buffer = (char *) __get_free_pages(gfp_flags, order);
4211 /* complete and utter failure */
4215 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4217 unsigned int block_nr = req->tp_block_nr;
4221 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4222 if (unlikely(!pg_vec))
4225 for (i = 0; i < block_nr; i++) {
4226 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4227 if (unlikely(!pg_vec[i].buffer))
4228 goto out_free_pgvec;
4235 free_pg_vec(pg_vec, order, block_nr);
4240 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4241 int closing, int tx_ring)
4243 struct pgv *pg_vec = NULL;
4244 struct packet_sock *po = pkt_sk(sk);
4245 int was_running, order = 0;
4246 struct packet_ring_buffer *rb;
4247 struct sk_buff_head *rb_queue;
4250 /* Added to avoid minimal code churn */
4251 struct tpacket_req *req = &req_u->req;
4253 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4254 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4258 if (atomic_read(&po->mapped))
4260 if (packet_read_pending(rb))
4264 if (req->tp_block_nr) {
4265 unsigned int min_frame_size;
4267 /* Sanity tests and some calculations */
4269 if (unlikely(rb->pg_vec))
4272 switch (po->tp_version) {
4274 po->tp_hdrlen = TPACKET_HDRLEN;
4277 po->tp_hdrlen = TPACKET2_HDRLEN;
4280 po->tp_hdrlen = TPACKET3_HDRLEN;
4285 if (unlikely((int)req->tp_block_size <= 0))
4287 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4289 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4290 if (po->tp_version >= TPACKET_V3 &&
4291 req->tp_block_size <
4292 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4294 if (unlikely(req->tp_frame_size < min_frame_size))
4296 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4299 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4300 if (unlikely(rb->frames_per_block == 0))
4302 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4304 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4309 order = get_order(req->tp_block_size);
4310 pg_vec = alloc_pg_vec(req, order);
4311 if (unlikely(!pg_vec))
4313 switch (po->tp_version) {
4315 /* Block transmit is not supported yet */
4317 init_prb_bdqc(po, rb, pg_vec, req_u);
4319 struct tpacket_req3 *req3 = &req_u->req3;
4321 if (req3->tp_retire_blk_tov ||
4322 req3->tp_sizeof_priv ||
4323 req3->tp_feature_req_word) {
4336 if (unlikely(req->tp_frame_nr))
4341 /* Detach socket from network */
4342 spin_lock(&po->bind_lock);
4343 was_running = po->running;
4347 __unregister_prot_hook(sk, false);
4349 spin_unlock(&po->bind_lock);
4354 mutex_lock(&po->pg_vec_lock);
4355 if (closing || atomic_read(&po->mapped) == 0) {
4357 spin_lock_bh(&rb_queue->lock);
4358 swap(rb->pg_vec, pg_vec);
4359 rb->frame_max = (req->tp_frame_nr - 1);
4361 rb->frame_size = req->tp_frame_size;
4362 spin_unlock_bh(&rb_queue->lock);
4364 swap(rb->pg_vec_order, order);
4365 swap(rb->pg_vec_len, req->tp_block_nr);
4367 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4368 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4369 tpacket_rcv : packet_rcv;
4370 skb_queue_purge(rb_queue);
4371 if (atomic_read(&po->mapped))
4372 pr_err("packet_mmap: vma is busy: %d\n",
4373 atomic_read(&po->mapped));
4375 mutex_unlock(&po->pg_vec_lock);
4377 spin_lock(&po->bind_lock);
4380 register_prot_hook(sk);
4382 spin_unlock(&po->bind_lock);
4383 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4384 /* Because we don't support block-based V3 on tx-ring */
4386 prb_shutdown_retire_blk_timer(po, rb_queue);
4390 free_pg_vec(pg_vec, order, req->tp_block_nr);
4395 static int packet_mmap(struct file *file, struct socket *sock,
4396 struct vm_area_struct *vma)
4398 struct sock *sk = sock->sk;
4399 struct packet_sock *po = pkt_sk(sk);
4400 unsigned long size, expected_size;
4401 struct packet_ring_buffer *rb;
4402 unsigned long start;
4409 mutex_lock(&po->pg_vec_lock);
4412 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4414 expected_size += rb->pg_vec_len
4420 if (expected_size == 0)
4423 size = vma->vm_end - vma->vm_start;
4424 if (size != expected_size)
4427 start = vma->vm_start;
4428 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4429 if (rb->pg_vec == NULL)
4432 for (i = 0; i < rb->pg_vec_len; i++) {
4434 void *kaddr = rb->pg_vec[i].buffer;
4437 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4438 page = pgv_to_page(kaddr);
4439 err = vm_insert_page(vma, start, page);
4448 atomic_inc(&po->mapped);
4449 vma->vm_ops = &packet_mmap_ops;
4453 mutex_unlock(&po->pg_vec_lock);
4457 static const struct proto_ops packet_ops_spkt = {
4458 .family = PF_PACKET,
4459 .owner = THIS_MODULE,
4460 .release = packet_release,
4461 .bind = packet_bind_spkt,
4462 .connect = sock_no_connect,
4463 .socketpair = sock_no_socketpair,
4464 .accept = sock_no_accept,
4465 .getname = packet_getname_spkt,
4466 .poll = datagram_poll,
4467 .ioctl = packet_ioctl,
4468 .gettstamp = sock_gettstamp,
4469 .listen = sock_no_listen,
4470 .shutdown = sock_no_shutdown,
4471 .setsockopt = sock_no_setsockopt,
4472 .getsockopt = sock_no_getsockopt,
4473 .sendmsg = packet_sendmsg_spkt,
4474 .recvmsg = packet_recvmsg,
4475 .mmap = sock_no_mmap,
4476 .sendpage = sock_no_sendpage,
4479 static const struct proto_ops packet_ops = {
4480 .family = PF_PACKET,
4481 .owner = THIS_MODULE,
4482 .release = packet_release,
4483 .bind = packet_bind,
4484 .connect = sock_no_connect,
4485 .socketpair = sock_no_socketpair,
4486 .accept = sock_no_accept,
4487 .getname = packet_getname,
4488 .poll = packet_poll,
4489 .ioctl = packet_ioctl,
4490 .gettstamp = sock_gettstamp,
4491 .listen = sock_no_listen,
4492 .shutdown = sock_no_shutdown,
4493 .setsockopt = packet_setsockopt,
4494 .getsockopt = packet_getsockopt,
4495 #ifdef CONFIG_COMPAT
4496 .compat_setsockopt = compat_packet_setsockopt,
4498 .sendmsg = packet_sendmsg,
4499 .recvmsg = packet_recvmsg,
4500 .mmap = packet_mmap,
4501 .sendpage = sock_no_sendpage,
4504 static const struct net_proto_family packet_family_ops = {
4505 .family = PF_PACKET,
4506 .create = packet_create,
4507 .owner = THIS_MODULE,
4510 static struct notifier_block packet_netdev_notifier = {
4511 .notifier_call = packet_notifier,
4514 #ifdef CONFIG_PROC_FS
4516 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4519 struct net *net = seq_file_net(seq);
4522 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4525 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4527 struct net *net = seq_file_net(seq);
4528 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4531 static void packet_seq_stop(struct seq_file *seq, void *v)
4537 static int packet_seq_show(struct seq_file *seq, void *v)
4539 if (v == SEQ_START_TOKEN)
4540 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4542 struct sock *s = sk_entry(v);
4543 const struct packet_sock *po = pkt_sk(s);
4546 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4548 refcount_read(&s->sk_refcnt),
4553 atomic_read(&s->sk_rmem_alloc),
4554 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4561 static const struct seq_operations packet_seq_ops = {
4562 .start = packet_seq_start,
4563 .next = packet_seq_next,
4564 .stop = packet_seq_stop,
4565 .show = packet_seq_show,
4569 static int __net_init packet_net_init(struct net *net)
4571 mutex_init(&net->packet.sklist_lock);
4572 INIT_HLIST_HEAD(&net->packet.sklist);
4574 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4575 sizeof(struct seq_net_private)))
4581 static void __net_exit packet_net_exit(struct net *net)
4583 remove_proc_entry("packet", net->proc_net);
4584 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4587 static struct pernet_operations packet_net_ops = {
4588 .init = packet_net_init,
4589 .exit = packet_net_exit,
4593 static void __exit packet_exit(void)
4595 unregister_netdevice_notifier(&packet_netdev_notifier);
4596 unregister_pernet_subsys(&packet_net_ops);
4597 sock_unregister(PF_PACKET);
4598 proto_unregister(&packet_proto);
4601 static int __init packet_init(void)
4605 rc = proto_register(&packet_proto, 0);
4608 rc = sock_register(&packet_family_ops);
4611 rc = register_pernet_subsys(&packet_net_ops);
4614 rc = register_netdevice_notifier(&packet_netdev_notifier);
4621 unregister_pernet_subsys(&packet_net_ops);
4623 sock_unregister(PF_PACKET);
4625 proto_unregister(&packet_proto);
4630 module_init(packet_init);
4631 module_exit(packet_exit);
4632 MODULE_LICENSE("GPL");
4633 MODULE_ALIAS_NETPROTO(PF_PACKET);