2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <asm/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
96 #include <net/compat.h>
102 - if device has no dev->hard_header routine, it adds and removes ll header
103 inside itself. In this case ll header is invisible outside of device,
104 but higher levels still should reserve dev->hard_header_len.
105 Some devices are enough clever to reallocate skb, when header
106 will not fit to reserved space (tunnel), another ones are silly
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
122 Incoming, dev->hard_header==NULL
123 mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 header. PPP makes it, that is wrong, because introduce
125 assymetry between rx and tx paths.
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
139 dev->hard_header != NULL
140 mac_header -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
156 struct packet_mreq_max {
158 unsigned short mr_type;
159 unsigned short mr_alen;
160 unsigned char mr_address[MAX_ADDR_LEN];
164 struct tpacket_hdr *h1;
165 struct tpacket2_hdr *h2;
166 struct tpacket3_hdr *h3;
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 int closing, int tx_ring);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define PGV_FROM_VMALLOC 1
182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
193 struct packet_type *pt, struct net_device *orig_dev);
195 static void *packet_previous_frame(struct packet_sock *po,
196 struct packet_ring_buffer *rb,
198 static void packet_increment_head(struct packet_ring_buffer *buff);
199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
200 struct tpacket_block_desc *);
201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
202 struct packet_sock *);
203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
204 struct packet_sock *, unsigned int status);
205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
206 static void prb_open_block(struct tpacket_kbdq_core *,
207 struct tpacket_block_desc *);
208 static void prb_retire_rx_blk_timer_expired(unsigned long);
209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
210 static void prb_init_blk_timer(struct packet_sock *,
211 struct tpacket_kbdq_core *,
212 void (*func) (unsigned long));
213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
214 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
217 struct tpacket3_hdr *);
218 static void packet_flush_mclist(struct sock *sk);
220 struct packet_skb_cb {
222 struct sockaddr_pkt pkt;
224 /* Trick: alias skb original length with
225 * ll.sll_family and ll.protocol in order
228 unsigned int origlen;
229 struct sockaddr_ll ll;
234 #define vio_le() virtio_legacy_is_little_endian()
236 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
238 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
239 #define GET_PBLOCK_DESC(x, bid) \
240 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
242 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
243 #define GET_NEXT_PRB_BLK_NUM(x) \
244 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
245 ((x)->kactive_blk_num+1) : 0)
247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
248 static void __fanout_link(struct sock *sk, struct packet_sock *po);
250 static int packet_direct_xmit(struct sk_buff *skb)
252 struct net_device *dev = skb->dev;
253 struct sk_buff *orig_skb = skb;
254 struct netdev_queue *txq;
255 int ret = NETDEV_TX_BUSY;
257 if (unlikely(!netif_running(dev) ||
258 !netif_carrier_ok(dev)))
261 skb = validate_xmit_skb_list(skb, dev);
265 txq = skb_get_tx_queue(dev, skb);
269 HARD_TX_LOCK(dev, txq, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq))
271 ret = netdev_start_xmit(skb, dev, txq, false);
272 HARD_TX_UNLOCK(dev, txq);
276 if (!dev_xmit_complete(ret))
281 atomic_long_inc(&dev->tx_dropped);
283 return NET_XMIT_DROP;
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
288 struct net_device *dev;
291 dev = rcu_dereference(po->cached_dev);
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
302 rcu_assign_pointer(po->cached_dev, dev);
305 static void packet_cached_dev_reset(struct packet_sock *po)
307 RCU_INIT_POINTER(po->cached_dev, NULL);
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
312 return po->xmit == packet_direct_xmit;
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
322 const struct net_device_ops *ops = dev->netdev_ops;
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
330 queue_index = __packet_pick_tx_queue(dev, skb);
333 skb_set_queue_mapping(skb, queue_index);
336 /* __register_prot_hook must be invoked through register_prot_hook
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
340 static void __register_prot_hook(struct sock *sk)
342 struct packet_sock *po = pkt_sk(sk);
346 __fanout_link(sk, po);
348 dev_add_pack(&po->prot_hook);
355 static void register_prot_hook(struct sock *sk)
357 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
358 __register_prot_hook(sk);
361 /* If the sync parameter is true, we will temporarily drop
362 * the po->bind_lock and do a synchronize_net to make sure no
363 * asynchronous packet processing paths still refer to the elements
364 * of po->prot_hook. If the sync parameter is false, it is the
365 * callers responsibility to take care of this.
367 static void __unregister_prot_hook(struct sock *sk, bool sync)
369 struct packet_sock *po = pkt_sk(sk);
371 lockdep_assert_held_once(&po->bind_lock);
376 __fanout_unlink(sk, po);
378 __dev_remove_pack(&po->prot_hook);
383 spin_unlock(&po->bind_lock);
385 spin_lock(&po->bind_lock);
389 static void unregister_prot_hook(struct sock *sk, bool sync)
391 struct packet_sock *po = pkt_sk(sk);
394 __unregister_prot_hook(sk, sync);
397 static inline struct page * __pure pgv_to_page(void *addr)
399 if (is_vmalloc_addr(addr))
400 return vmalloc_to_page(addr);
401 return virt_to_page(addr);
404 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
406 union tpacket_uhdr h;
409 switch (po->tp_version) {
411 h.h1->tp_status = status;
412 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
415 h.h2->tp_status = status;
416 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
420 WARN(1, "TPACKET version not supported.\n");
427 static int __packet_get_status(struct packet_sock *po, void *frame)
429 union tpacket_uhdr h;
434 switch (po->tp_version) {
436 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
437 return h.h1->tp_status;
439 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
440 return h.h2->tp_status;
443 WARN(1, "TPACKET version not supported.\n");
449 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
452 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
455 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
456 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
457 return TP_STATUS_TS_RAW_HARDWARE;
459 if (ktime_to_timespec_cond(skb->tstamp, ts))
460 return TP_STATUS_TS_SOFTWARE;
465 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
468 union tpacket_uhdr h;
472 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
476 switch (po->tp_version) {
478 h.h1->tp_sec = ts.tv_sec;
479 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
482 h.h2->tp_sec = ts.tv_sec;
483 h.h2->tp_nsec = ts.tv_nsec;
487 WARN(1, "TPACKET version not supported.\n");
491 /* one flush is safe, as both fields always lie on the same cacheline */
492 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
498 static void *packet_lookup_frame(struct packet_sock *po,
499 struct packet_ring_buffer *rb,
500 unsigned int position,
503 unsigned int pg_vec_pos, frame_offset;
504 union tpacket_uhdr h;
506 pg_vec_pos = position / rb->frames_per_block;
507 frame_offset = position % rb->frames_per_block;
509 h.raw = rb->pg_vec[pg_vec_pos].buffer +
510 (frame_offset * rb->frame_size);
512 if (status != __packet_get_status(po, h.raw))
518 static void *packet_current_frame(struct packet_sock *po,
519 struct packet_ring_buffer *rb,
522 return packet_lookup_frame(po, rb, rb->head, status);
525 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
527 del_timer_sync(&pkc->retire_blk_timer);
530 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
531 struct sk_buff_head *rb_queue)
533 struct tpacket_kbdq_core *pkc;
535 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
537 spin_lock_bh(&rb_queue->lock);
538 pkc->delete_blk_timer = 1;
539 spin_unlock_bh(&rb_queue->lock);
541 prb_del_retire_blk_timer(pkc);
544 static void prb_init_blk_timer(struct packet_sock *po,
545 struct tpacket_kbdq_core *pkc,
546 void (*func) (unsigned long))
548 init_timer(&pkc->retire_blk_timer);
549 pkc->retire_blk_timer.data = (long)po;
550 pkc->retire_blk_timer.function = func;
551 pkc->retire_blk_timer.expires = jiffies;
554 static void prb_setup_retire_blk_timer(struct packet_sock *po)
556 struct tpacket_kbdq_core *pkc;
558 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
559 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
562 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
563 int blk_size_in_bytes)
565 struct net_device *dev;
566 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
567 struct ethtool_link_ksettings ecmd;
571 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
572 if (unlikely(!dev)) {
574 return DEFAULT_PRB_RETIRE_TOV;
576 err = __ethtool_get_link_ksettings(dev, &ecmd);
580 * If the link speed is so slow you don't really
581 * need to worry about perf anyways
583 if (ecmd.base.speed < SPEED_1000 ||
584 ecmd.base.speed == SPEED_UNKNOWN) {
585 return DEFAULT_PRB_RETIRE_TOV;
588 div = ecmd.base.speed / 1000;
591 return DEFAULT_PRB_RETIRE_TOV;
593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
606 union tpacket_req_u *req_u)
608 p1->feature_req_word = req_u->req3.tp_feature_req_word;
611 static void init_prb_bdqc(struct packet_sock *po,
612 struct packet_ring_buffer *rb,
614 union tpacket_req_u *req_u)
616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
617 struct tpacket_block_desc *pbd;
619 memset(p1, 0x0, sizeof(*p1));
621 p1->knxt_seq_num = 1;
623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
624 p1->pkblk_start = pg_vec[0].buffer;
625 p1->kblk_size = req_u->req3.tp_block_size;
626 p1->knum_blocks = req_u->req3.tp_block_nr;
627 p1->hdrlen = po->tp_hdrlen;
628 p1->version = po->tp_version;
629 p1->last_kactive_blk_num = 0;
630 po->stats.stats3.tp_freeze_q_cnt = 0;
631 if (req_u->req3.tp_retire_blk_tov)
632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 req_u->req3.tp_block_size);
636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
639 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
640 prb_init_ft_ops(p1, req_u);
641 prb_setup_retire_blk_timer(po);
642 prb_open_block(p1, pbd);
645 /* Do NOT update the last_blk_num first.
646 * Assumes sk_buff_head lock is held.
648 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
650 mod_timer(&pkc->retire_blk_timer,
651 jiffies + pkc->tov_in_jiffies);
652 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
657 * 1) We refresh the timer only when we open a block.
658 * By doing this we don't waste cycles refreshing the timer
659 * on packet-by-packet basis.
661 * With a 1MB block-size, on a 1Gbps line, it will take
662 * i) ~8 ms to fill a block + ii) memcpy etc.
663 * In this cut we are not accounting for the memcpy time.
665 * So, if the user sets the 'tmo' to 10ms then the timer
666 * will never fire while the block is still getting filled
667 * (which is what we want). However, the user could choose
668 * to close a block early and that's fine.
670 * But when the timer does fire, we check whether or not to refresh it.
671 * Since the tmo granularity is in msecs, it is not too expensive
672 * to refresh the timer, lets say every '8' msecs.
673 * Either the user can set the 'tmo' or we can derive it based on
674 * a) line-speed and b) block-size.
675 * prb_calc_retire_blk_tmo() calculates the tmo.
678 static void prb_retire_rx_blk_timer_expired(unsigned long data)
680 struct packet_sock *po = (struct packet_sock *)data;
681 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
683 struct tpacket_block_desc *pbd;
685 spin_lock(&po->sk.sk_receive_queue.lock);
687 frozen = prb_queue_frozen(pkc);
688 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
690 if (unlikely(pkc->delete_blk_timer))
693 /* We only need to plug the race when the block is partially filled.
695 * lock(); increment BLOCK_NUM_PKTS; unlock()
696 * copy_bits() is in progress ...
697 * timer fires on other cpu:
698 * we can't retire the current block because copy_bits
702 if (BLOCK_NUM_PKTS(pbd)) {
703 while (atomic_read(&pkc->blk_fill_in_prog)) {
704 /* Waiting for skb_copy_bits to finish... */
709 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
711 if (!BLOCK_NUM_PKTS(pbd)) {
712 /* An empty block. Just refresh the timer. */
715 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
716 if (!prb_dispatch_next_block(pkc, po))
721 /* Case 1. Queue was frozen because user-space was
724 if (prb_curr_blk_in_use(pkc, pbd)) {
726 * Ok, user-space is still behind.
727 * So just refresh the timer.
731 /* Case 2. queue was frozen,user-space caught up,
732 * now the link went idle && the timer fired.
733 * We don't have a block to close.So we open this
734 * block and restart the timer.
735 * opening a block thaws the queue,restarts timer
736 * Thawing/timer-refresh is a side effect.
738 prb_open_block(pkc, pbd);
745 _prb_refresh_rx_retire_blk_timer(pkc);
748 spin_unlock(&po->sk.sk_receive_queue.lock);
751 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
752 struct tpacket_block_desc *pbd1, __u32 status)
754 /* Flush everything minus the block header */
756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
761 /* Skip the block header(we know header WILL fit in 4K) */
764 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
765 for (; start < end; start += PAGE_SIZE)
766 flush_dcache_page(pgv_to_page(start));
771 /* Now update the block status. */
773 BLOCK_STATUS(pbd1) = status;
775 /* Flush the block header */
777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
779 flush_dcache_page(pgv_to_page(start));
789 * 2) Increment active_blk_num
791 * Note:We DONT refresh the timer on purpose.
792 * Because almost always the next block will be opened.
794 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
795 struct tpacket_block_desc *pbd1,
796 struct packet_sock *po, unsigned int stat)
798 __u32 status = TP_STATUS_USER | stat;
800 struct tpacket3_hdr *last_pkt;
801 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
802 struct sock *sk = &po->sk;
804 if (po->stats.stats3.tp_drops)
805 status |= TP_STATUS_LOSING;
807 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
808 last_pkt->tp_next_offset = 0;
810 /* Get the ts of the last pkt */
811 if (BLOCK_NUM_PKTS(pbd1)) {
812 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
813 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
815 /* Ok, we tmo'd - so get the current time.
817 * It shouldn't really happen as we don't close empty
818 * blocks. See prb_retire_rx_blk_timer_expired().
822 h1->ts_last_pkt.ts_sec = ts.tv_sec;
823 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
828 /* Flush the block */
829 prb_flush_block(pkc1, pbd1, status);
831 sk->sk_data_ready(sk);
833 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
836 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
838 pkc->reset_pending_on_curr_blk = 0;
842 * Side effect of opening a block:
844 * 1) prb_queue is thawed.
845 * 2) retire_blk_timer is refreshed.
848 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
849 struct tpacket_block_desc *pbd1)
852 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
856 /* We could have just memset this but we will lose the
857 * flexibility of making the priv area sticky
860 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
861 BLOCK_NUM_PKTS(pbd1) = 0;
862 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 h1->ts_first_pkt.ts_sec = ts.tv_sec;
867 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
869 pkc1->pkblk_start = (char *)pbd1;
870 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
872 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
873 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
875 pbd1->version = pkc1->version;
876 pkc1->prev = pkc1->nxt_offset;
877 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
879 prb_thaw_queue(pkc1);
880 _prb_refresh_rx_retire_blk_timer(pkc1);
886 * Queue freeze logic:
887 * 1) Assume tp_block_nr = 8 blocks.
888 * 2) At time 't0', user opens Rx ring.
889 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
890 * 4) user-space is either sleeping or processing block '0'.
891 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
892 * it will close block-7,loop around and try to fill block '0'.
894 * __packet_lookup_frame_in_block
895 * prb_retire_current_block()
896 * prb_dispatch_next_block()
897 * |->(BLOCK_STATUS == USER) evaluates to true
898 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
899 * 6) Now there are two cases:
900 * 6.1) Link goes idle right after the queue is frozen.
901 * But remember, the last open_block() refreshed the timer.
902 * When this timer expires,it will refresh itself so that we can
903 * re-open block-0 in near future.
904 * 6.2) Link is busy and keeps on receiving packets. This is a simple
905 * case and __packet_lookup_frame_in_block will check if block-0
906 * is free and can now be re-used.
908 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
909 struct packet_sock *po)
911 pkc->reset_pending_on_curr_blk = 1;
912 po->stats.stats3.tp_freeze_q_cnt++;
915 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
918 * If the next block is free then we will dispatch it
919 * and return a good offset.
920 * Else, we will freeze the queue.
921 * So, caller must check the return value.
923 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
924 struct packet_sock *po)
926 struct tpacket_block_desc *pbd;
930 /* 1. Get current block num */
931 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
933 /* 2. If this block is currently in_use then freeze the queue */
934 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
935 prb_freeze_queue(pkc, po);
941 * open this block and return the offset where the first packet
942 * needs to get stored.
944 prb_open_block(pkc, pbd);
945 return (void *)pkc->nxt_offset;
948 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
949 struct packet_sock *po, unsigned int status)
951 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
953 /* retire/close the current block */
954 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
956 * Plug the case where copy_bits() is in progress on
957 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
958 * have space to copy the pkt in the current block and
959 * called prb_retire_current_block()
961 * We don't need to worry about the TMO case because
962 * the timer-handler already handled this case.
964 if (!(status & TP_STATUS_BLK_TMO)) {
965 while (atomic_read(&pkc->blk_fill_in_prog)) {
966 /* Waiting for skb_copy_bits to finish... */
970 prb_close_block(pkc, pbd, po, status);
975 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
976 struct tpacket_block_desc *pbd)
978 return TP_STATUS_USER & BLOCK_STATUS(pbd);
981 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
983 return pkc->reset_pending_on_curr_blk;
986 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
988 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
989 atomic_dec(&pkc->blk_fill_in_prog);
992 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
993 struct tpacket3_hdr *ppd)
995 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
998 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
999 struct tpacket3_hdr *ppd)
1001 ppd->hv1.tp_rxhash = 0;
1004 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1005 struct tpacket3_hdr *ppd)
1007 if (skb_vlan_tag_present(pkc->skb)) {
1008 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1009 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1010 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1012 ppd->hv1.tp_vlan_tci = 0;
1013 ppd->hv1.tp_vlan_tpid = 0;
1014 ppd->tp_status = TP_STATUS_AVAILABLE;
1018 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1019 struct tpacket3_hdr *ppd)
1021 ppd->hv1.tp_padding = 0;
1022 prb_fill_vlan_info(pkc, ppd);
1024 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1025 prb_fill_rxhash(pkc, ppd);
1027 prb_clear_rxhash(pkc, ppd);
1030 static void prb_fill_curr_block(char *curr,
1031 struct tpacket_kbdq_core *pkc,
1032 struct tpacket_block_desc *pbd,
1035 struct tpacket3_hdr *ppd;
1037 ppd = (struct tpacket3_hdr *)curr;
1038 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1040 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1042 BLOCK_NUM_PKTS(pbd) += 1;
1043 atomic_inc(&pkc->blk_fill_in_prog);
1044 prb_run_all_ft_ops(pkc, ppd);
1047 /* Assumes caller has the sk->rx_queue.lock */
1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1049 struct sk_buff *skb,
1054 struct tpacket_kbdq_core *pkc;
1055 struct tpacket_block_desc *pbd;
1058 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1059 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1061 /* Queue is frozen when user space is lagging behind */
1062 if (prb_queue_frozen(pkc)) {
1064 * Check if that last block which caused the queue to freeze,
1065 * is still in_use by user-space.
1067 if (prb_curr_blk_in_use(pkc, pbd)) {
1068 /* Can't record this packet */
1072 * Ok, the block was released by user-space.
1073 * Now let's open that block.
1074 * opening a block also thaws the queue.
1075 * Thawing is a side effect.
1077 prb_open_block(pkc, pbd);
1082 curr = pkc->nxt_offset;
1084 end = (char *)pbd + pkc->kblk_size;
1086 /* first try the current block */
1087 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1088 prb_fill_curr_block(curr, pkc, pbd, len);
1089 return (void *)curr;
1092 /* Ok, close the current block */
1093 prb_retire_current_block(pkc, po, 0);
1095 /* Now, try to dispatch the next block */
1096 curr = (char *)prb_dispatch_next_block(pkc, po);
1098 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1099 prb_fill_curr_block(curr, pkc, pbd, len);
1100 return (void *)curr;
1104 * No free blocks are available.user_space hasn't caught up yet.
1105 * Queue was just frozen and now this packet will get dropped.
1110 static void *packet_current_rx_frame(struct packet_sock *po,
1111 struct sk_buff *skb,
1112 int status, unsigned int len)
1115 switch (po->tp_version) {
1118 curr = packet_lookup_frame(po, &po->rx_ring,
1119 po->rx_ring.head, status);
1122 return __packet_lookup_frame_in_block(po, skb, status, len);
1124 WARN(1, "TPACKET version not supported\n");
1130 static void *prb_lookup_block(struct packet_sock *po,
1131 struct packet_ring_buffer *rb,
1135 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1136 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1138 if (status != BLOCK_STATUS(pbd))
1143 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1146 if (rb->prb_bdqc.kactive_blk_num)
1147 prev = rb->prb_bdqc.kactive_blk_num-1;
1149 prev = rb->prb_bdqc.knum_blocks-1;
1153 /* Assumes caller has held the rx_queue.lock */
1154 static void *__prb_previous_block(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1158 unsigned int previous = prb_previous_blk_num(rb);
1159 return prb_lookup_block(po, rb, previous, status);
1162 static void *packet_previous_rx_frame(struct packet_sock *po,
1163 struct packet_ring_buffer *rb,
1166 if (po->tp_version <= TPACKET_V2)
1167 return packet_previous_frame(po, rb, status);
1169 return __prb_previous_block(po, rb, status);
1172 static void packet_increment_rx_head(struct packet_sock *po,
1173 struct packet_ring_buffer *rb)
1175 switch (po->tp_version) {
1178 return packet_increment_head(rb);
1181 WARN(1, "TPACKET version not supported.\n");
1187 static void *packet_previous_frame(struct packet_sock *po,
1188 struct packet_ring_buffer *rb,
1191 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1192 return packet_lookup_frame(po, rb, previous, status);
1195 static void packet_increment_head(struct packet_ring_buffer *buff)
1197 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1200 static void packet_inc_pending(struct packet_ring_buffer *rb)
1202 this_cpu_inc(*rb->pending_refcnt);
1205 static void packet_dec_pending(struct packet_ring_buffer *rb)
1207 this_cpu_dec(*rb->pending_refcnt);
1210 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1212 unsigned int refcnt = 0;
1215 /* We don't use pending refcount in rx_ring. */
1216 if (rb->pending_refcnt == NULL)
1219 for_each_possible_cpu(cpu)
1220 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1225 static int packet_alloc_pending(struct packet_sock *po)
1227 po->rx_ring.pending_refcnt = NULL;
1229 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1230 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1236 static void packet_free_pending(struct packet_sock *po)
1238 free_percpu(po->tx_ring.pending_refcnt);
1241 #define ROOM_POW_OFF 2
1242 #define ROOM_NONE 0x0
1243 #define ROOM_LOW 0x1
1244 #define ROOM_NORMAL 0x2
1246 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1250 len = po->rx_ring.frame_max + 1;
1251 idx = po->rx_ring.head;
1253 idx += len >> pow_off;
1256 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1259 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1263 len = po->rx_ring.prb_bdqc.knum_blocks;
1264 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1266 idx += len >> pow_off;
1269 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1272 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1274 struct sock *sk = &po->sk;
1275 int ret = ROOM_NONE;
1277 if (po->prot_hook.func != tpacket_rcv) {
1278 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1279 - (skb ? skb->truesize : 0);
1280 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1288 if (po->tp_version == TPACKET_V3) {
1289 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1291 else if (__tpacket_v3_has_room(po, 0))
1294 if (__tpacket_has_room(po, ROOM_POW_OFF))
1296 else if (__tpacket_has_room(po, 0))
1303 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1308 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1309 ret = __packet_rcv_has_room(po, skb);
1310 has_room = ret == ROOM_NORMAL;
1311 if (po->pressure == has_room)
1312 po->pressure = !has_room;
1313 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1318 static void packet_sock_destruct(struct sock *sk)
1320 skb_queue_purge(&sk->sk_error_queue);
1322 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1323 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1325 if (!sock_flag(sk, SOCK_DEAD)) {
1326 pr_err("Attempt to release alive packet socket: %p\n", sk);
1330 sk_refcnt_debug_dec(sk);
1333 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1335 u32 *history = po->rollover->history;
1339 rxhash = skb_get_hash(skb);
1340 for (i = 0; i < ROLLOVER_HLEN; i++)
1341 if (READ_ONCE(history[i]) == rxhash)
1344 victim = prandom_u32() % ROLLOVER_HLEN;
1346 /* Avoid dirtying the cache line if possible */
1347 if (READ_ONCE(history[victim]) != rxhash)
1348 WRITE_ONCE(history[victim], rxhash);
1350 return count > (ROLLOVER_HLEN >> 1);
1353 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1354 struct sk_buff *skb,
1357 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1360 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1361 struct sk_buff *skb,
1364 unsigned int val = atomic_inc_return(&f->rr_cur);
1369 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1370 struct sk_buff *skb,
1373 return smp_processor_id() % num;
1376 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1377 struct sk_buff *skb,
1380 return prandom_u32_max(num);
1383 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1384 struct sk_buff *skb,
1385 unsigned int idx, bool try_self,
1388 struct packet_sock *po, *po_next, *po_skip = NULL;
1389 unsigned int i, j, room = ROOM_NONE;
1391 po = pkt_sk(f->arr[idx]);
1394 room = packet_rcv_has_room(po, skb);
1395 if (room == ROOM_NORMAL ||
1396 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1401 i = j = min_t(int, po->rollover->sock, num - 1);
1403 po_next = pkt_sk(f->arr[i]);
1404 if (po_next != po_skip && !po_next->pressure &&
1405 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1407 po->rollover->sock = i;
1408 atomic_long_inc(&po->rollover->num);
1409 if (room == ROOM_LOW)
1410 atomic_long_inc(&po->rollover->num_huge);
1418 atomic_long_inc(&po->rollover->num_failed);
1422 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1423 struct sk_buff *skb,
1426 return skb_get_queue_mapping(skb) % num;
1429 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1430 struct sk_buff *skb,
1433 struct bpf_prog *prog;
1434 unsigned int ret = 0;
1437 prog = rcu_dereference(f->bpf_prog);
1439 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1445 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1447 return f->flags & (flag >> 8);
1450 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1451 struct packet_type *pt, struct net_device *orig_dev)
1453 struct packet_fanout *f = pt->af_packet_priv;
1454 unsigned int num = READ_ONCE(f->num_members);
1455 struct net *net = read_pnet(&f->net);
1456 struct packet_sock *po;
1459 if (!net_eq(dev_net(dev), net) || !num) {
1464 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1465 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1470 case PACKET_FANOUT_HASH:
1472 idx = fanout_demux_hash(f, skb, num);
1474 case PACKET_FANOUT_LB:
1475 idx = fanout_demux_lb(f, skb, num);
1477 case PACKET_FANOUT_CPU:
1478 idx = fanout_demux_cpu(f, skb, num);
1480 case PACKET_FANOUT_RND:
1481 idx = fanout_demux_rnd(f, skb, num);
1483 case PACKET_FANOUT_QM:
1484 idx = fanout_demux_qm(f, skb, num);
1486 case PACKET_FANOUT_ROLLOVER:
1487 idx = fanout_demux_rollover(f, skb, 0, false, num);
1489 case PACKET_FANOUT_CBPF:
1490 case PACKET_FANOUT_EBPF:
1491 idx = fanout_demux_bpf(f, skb, num);
1495 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1496 idx = fanout_demux_rollover(f, skb, idx, true, num);
1498 po = pkt_sk(f->arr[idx]);
1499 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1502 DEFINE_MUTEX(fanout_mutex);
1503 EXPORT_SYMBOL_GPL(fanout_mutex);
1504 static LIST_HEAD(fanout_list);
1506 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1508 struct packet_fanout *f = po->fanout;
1510 spin_lock(&f->lock);
1511 f->arr[f->num_members] = sk;
1514 if (f->num_members == 1)
1515 dev_add_pack(&f->prot_hook);
1516 spin_unlock(&f->lock);
1519 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1521 struct packet_fanout *f = po->fanout;
1524 spin_lock(&f->lock);
1525 for (i = 0; i < f->num_members; i++) {
1526 if (f->arr[i] == sk)
1529 BUG_ON(i >= f->num_members);
1530 f->arr[i] = f->arr[f->num_members - 1];
1532 if (f->num_members == 0)
1533 __dev_remove_pack(&f->prot_hook);
1534 spin_unlock(&f->lock);
1537 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1539 if (sk->sk_family != PF_PACKET)
1542 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1545 static void fanout_init_data(struct packet_fanout *f)
1548 case PACKET_FANOUT_LB:
1549 atomic_set(&f->rr_cur, 0);
1551 case PACKET_FANOUT_CBPF:
1552 case PACKET_FANOUT_EBPF:
1553 RCU_INIT_POINTER(f->bpf_prog, NULL);
1558 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1560 struct bpf_prog *old;
1562 spin_lock(&f->lock);
1563 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1564 rcu_assign_pointer(f->bpf_prog, new);
1565 spin_unlock(&f->lock);
1569 bpf_prog_destroy(old);
1573 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1576 struct bpf_prog *new;
1577 struct sock_fprog fprog;
1580 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1582 if (len != sizeof(fprog))
1584 if (copy_from_user(&fprog, data, len))
1587 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1591 __fanout_set_data_bpf(po->fanout, new);
1595 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1598 struct bpf_prog *new;
1601 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1603 if (len != sizeof(fd))
1605 if (copy_from_user(&fd, data, len))
1608 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1610 return PTR_ERR(new);
1612 __fanout_set_data_bpf(po->fanout, new);
1616 static int fanout_set_data(struct packet_sock *po, char __user *data,
1619 switch (po->fanout->type) {
1620 case PACKET_FANOUT_CBPF:
1621 return fanout_set_data_cbpf(po, data, len);
1622 case PACKET_FANOUT_EBPF:
1623 return fanout_set_data_ebpf(po, data, len);
1629 static void fanout_release_data(struct packet_fanout *f)
1632 case PACKET_FANOUT_CBPF:
1633 case PACKET_FANOUT_EBPF:
1634 __fanout_set_data_bpf(f, NULL);
1638 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1640 struct packet_rollover *rollover = NULL;
1641 struct packet_sock *po = pkt_sk(sk);
1642 struct packet_fanout *f, *match;
1643 u8 type = type_flags & 0xff;
1644 u8 flags = type_flags >> 8;
1648 case PACKET_FANOUT_ROLLOVER:
1649 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1651 case PACKET_FANOUT_HASH:
1652 case PACKET_FANOUT_LB:
1653 case PACKET_FANOUT_CPU:
1654 case PACKET_FANOUT_RND:
1655 case PACKET_FANOUT_QM:
1656 case PACKET_FANOUT_CBPF:
1657 case PACKET_FANOUT_EBPF:
1663 mutex_lock(&fanout_mutex);
1669 if (type == PACKET_FANOUT_ROLLOVER ||
1670 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1672 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1675 atomic_long_set(&rollover->num, 0);
1676 atomic_long_set(&rollover->num_huge, 0);
1677 atomic_long_set(&rollover->num_failed, 0);
1681 list_for_each_entry(f, &fanout_list, list) {
1683 read_pnet(&f->net) == sock_net(sk)) {
1689 if (match && match->flags != flags)
1693 match = kzalloc(sizeof(*match), GFP_KERNEL);
1696 write_pnet(&match->net, sock_net(sk));
1699 match->flags = flags;
1700 INIT_LIST_HEAD(&match->list);
1701 spin_lock_init(&match->lock);
1702 atomic_set(&match->sk_ref, 0);
1703 fanout_init_data(match);
1704 match->prot_hook.type = po->prot_hook.type;
1705 match->prot_hook.dev = po->prot_hook.dev;
1706 match->prot_hook.func = packet_rcv_fanout;
1707 match->prot_hook.af_packet_priv = match;
1708 match->prot_hook.id_match = match_fanout_group;
1709 list_add(&match->list, &fanout_list);
1713 spin_lock(&po->bind_lock);
1715 match->type == type &&
1716 match->prot_hook.type == po->prot_hook.type &&
1717 match->prot_hook.dev == po->prot_hook.dev) {
1719 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1720 __dev_remove_pack(&po->prot_hook);
1722 po->rollover = rollover;
1724 atomic_inc(&match->sk_ref);
1725 __fanout_link(sk, po);
1729 spin_unlock(&po->bind_lock);
1731 if (err && !atomic_read(&match->sk_ref)) {
1732 list_del(&match->list);
1738 mutex_unlock(&fanout_mutex);
1742 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1743 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1744 * It is the responsibility of the caller to call fanout_release_data() and
1745 * free the returned packet_fanout (after synchronize_net())
1747 static struct packet_fanout *fanout_release(struct sock *sk)
1749 struct packet_sock *po = pkt_sk(sk);
1750 struct packet_fanout *f;
1752 mutex_lock(&fanout_mutex);
1757 if (atomic_dec_and_test(&f->sk_ref))
1762 mutex_unlock(&fanout_mutex);
1767 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1768 struct sk_buff *skb)
1770 /* Earlier code assumed this would be a VLAN pkt, double-check
1771 * this now that we have the actual packet in hand. We can only
1772 * do this check on Ethernet devices.
1774 if (unlikely(dev->type != ARPHRD_ETHER))
1777 skb_reset_mac_header(skb);
1778 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1781 static const struct proto_ops packet_ops;
1783 static const struct proto_ops packet_ops_spkt;
1785 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1786 struct packet_type *pt, struct net_device *orig_dev)
1789 struct sockaddr_pkt *spkt;
1792 * When we registered the protocol we saved the socket in the data
1793 * field for just this event.
1796 sk = pt->af_packet_priv;
1799 * Yank back the headers [hope the device set this
1800 * right or kerboom...]
1802 * Incoming packets have ll header pulled,
1805 * For outgoing ones skb->data == skb_mac_header(skb)
1806 * so that this procedure is noop.
1809 if (skb->pkt_type == PACKET_LOOPBACK)
1812 if (!net_eq(dev_net(dev), sock_net(sk)))
1815 skb = skb_share_check(skb, GFP_ATOMIC);
1819 /* drop any routing info */
1822 /* drop conntrack reference */
1825 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1827 skb_push(skb, skb->data - skb_mac_header(skb));
1830 * The SOCK_PACKET socket receives _all_ frames.
1833 spkt->spkt_family = dev->type;
1834 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1835 spkt->spkt_protocol = skb->protocol;
1838 * Charge the memory to the socket. This is done specifically
1839 * to prevent sockets using all the memory up.
1842 if (sock_queue_rcv_skb(sk, skb) == 0)
1853 * Output a raw packet to a device layer. This bypasses all the other
1854 * protocol layers and you must therefore supply it with a complete frame
1857 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1860 struct sock *sk = sock->sk;
1861 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1862 struct sk_buff *skb = NULL;
1863 struct net_device *dev;
1864 struct sockcm_cookie sockc;
1870 * Get and verify the address.
1874 if (msg->msg_namelen < sizeof(struct sockaddr))
1876 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1877 proto = saddr->spkt_protocol;
1879 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1882 * Find the device first to size check it
1885 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1888 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1894 if (!(dev->flags & IFF_UP))
1898 * You may not queue a frame bigger than the mtu. This is the lowest level
1899 * raw protocol and you must do your own fragmentation at this level.
1902 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1903 if (!netif_supports_nofcs(dev)) {
1904 err = -EPROTONOSUPPORT;
1907 extra_len = 4; /* We're doing our own CRC */
1911 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1915 size_t reserved = LL_RESERVED_SPACE(dev);
1916 int tlen = dev->needed_tailroom;
1917 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1920 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1923 /* FIXME: Save some space for broken drivers that write a hard
1924 * header at transmission time by themselves. PPP is the notable
1925 * one here. This should really be fixed at the driver level.
1927 skb_reserve(skb, reserved);
1928 skb_reset_network_header(skb);
1930 /* Try to align data part correctly */
1935 skb_reset_network_header(skb);
1937 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1943 if (!dev_validate_header(dev, skb->data, len)) {
1947 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1948 !packet_extra_vlan_len_allowed(dev, skb)) {
1953 sockc.tsflags = sk->sk_tsflags;
1954 if (msg->msg_controllen) {
1955 err = sock_cmsg_send(sk, msg, &sockc);
1960 skb->protocol = proto;
1962 skb->priority = sk->sk_priority;
1963 skb->mark = sk->sk_mark;
1965 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1967 if (unlikely(extra_len == 4))
1970 skb_probe_transport_header(skb, 0);
1972 dev_queue_xmit(skb);
1983 static unsigned int run_filter(struct sk_buff *skb,
1984 const struct sock *sk,
1987 struct sk_filter *filter;
1990 filter = rcu_dereference(sk->sk_filter);
1992 res = bpf_prog_run_clear_cb(filter->prog, skb);
1998 static int __packet_rcv_vnet(const struct sk_buff *skb,
1999 struct virtio_net_hdr *vnet_hdr)
2001 *vnet_hdr = (const struct virtio_net_hdr) { 0 };
2003 if (virtio_net_hdr_from_skb(skb, vnet_hdr, vio_le(), true))
2009 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2012 struct virtio_net_hdr vnet_hdr;
2014 if (*len < sizeof(vnet_hdr))
2016 *len -= sizeof(vnet_hdr);
2018 if (__packet_rcv_vnet(skb, &vnet_hdr))
2021 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2025 * This function makes lazy skb cloning in hope that most of packets
2026 * are discarded by BPF.
2028 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2029 * and skb->cb are mangled. It works because (and until) packets
2030 * falling here are owned by current CPU. Output packets are cloned
2031 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2032 * sequencially, so that if we return skb to original state on exit,
2033 * we will not harm anyone.
2036 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2037 struct packet_type *pt, struct net_device *orig_dev)
2040 struct sockaddr_ll *sll;
2041 struct packet_sock *po;
2042 u8 *skb_head = skb->data;
2043 int skb_len = skb->len;
2044 unsigned int snaplen, res;
2045 bool is_drop_n_account = false;
2047 if (skb->pkt_type == PACKET_LOOPBACK)
2050 sk = pt->af_packet_priv;
2053 if (!net_eq(dev_net(dev), sock_net(sk)))
2058 if (dev->header_ops) {
2059 /* The device has an explicit notion of ll header,
2060 * exported to higher levels.
2062 * Otherwise, the device hides details of its frame
2063 * structure, so that corresponding packet head is
2064 * never delivered to user.
2066 if (sk->sk_type != SOCK_DGRAM)
2067 skb_push(skb, skb->data - skb_mac_header(skb));
2068 else if (skb->pkt_type == PACKET_OUTGOING) {
2069 /* Special case: outgoing packets have ll header at head */
2070 skb_pull(skb, skb_network_offset(skb));
2076 res = run_filter(skb, sk, snaplen);
2078 goto drop_n_restore;
2082 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2085 if (skb_shared(skb)) {
2086 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2090 if (skb_head != skb->data) {
2091 skb->data = skb_head;
2098 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2100 sll = &PACKET_SKB_CB(skb)->sa.ll;
2101 sll->sll_hatype = dev->type;
2102 sll->sll_pkttype = skb->pkt_type;
2103 if (unlikely(po->origdev))
2104 sll->sll_ifindex = orig_dev->ifindex;
2106 sll->sll_ifindex = dev->ifindex;
2108 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2110 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2111 * Use their space for storing the original skb length.
2113 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2115 if (pskb_trim(skb, snaplen))
2118 skb_set_owner_r(skb, sk);
2122 /* drop conntrack reference */
2125 spin_lock(&sk->sk_receive_queue.lock);
2126 po->stats.stats1.tp_packets++;
2127 sock_skb_set_dropcount(sk, skb);
2128 __skb_queue_tail(&sk->sk_receive_queue, skb);
2129 spin_unlock(&sk->sk_receive_queue.lock);
2130 sk->sk_data_ready(sk);
2134 is_drop_n_account = true;
2135 spin_lock(&sk->sk_receive_queue.lock);
2136 po->stats.stats1.tp_drops++;
2137 atomic_inc(&sk->sk_drops);
2138 spin_unlock(&sk->sk_receive_queue.lock);
2141 if (skb_head != skb->data && skb_shared(skb)) {
2142 skb->data = skb_head;
2146 if (!is_drop_n_account)
2153 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2154 struct packet_type *pt, struct net_device *orig_dev)
2157 struct packet_sock *po;
2158 struct sockaddr_ll *sll;
2159 union tpacket_uhdr h;
2160 u8 *skb_head = skb->data;
2161 int skb_len = skb->len;
2162 unsigned int snaplen, res;
2163 unsigned long status = TP_STATUS_USER;
2164 unsigned short macoff, hdrlen;
2165 unsigned int netoff;
2166 struct sk_buff *copy_skb = NULL;
2169 bool is_drop_n_account = false;
2170 bool do_vnet = false;
2172 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2173 * We may add members to them until current aligned size without forcing
2174 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2176 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2177 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2179 if (skb->pkt_type == PACKET_LOOPBACK)
2182 sk = pt->af_packet_priv;
2185 if (!net_eq(dev_net(dev), sock_net(sk)))
2188 if (dev->header_ops) {
2189 if (sk->sk_type != SOCK_DGRAM)
2190 skb_push(skb, skb->data - skb_mac_header(skb));
2191 else if (skb->pkt_type == PACKET_OUTGOING) {
2192 /* Special case: outgoing packets have ll header at head */
2193 skb_pull(skb, skb_network_offset(skb));
2199 res = run_filter(skb, sk, snaplen);
2201 goto drop_n_restore;
2203 if (skb->ip_summed == CHECKSUM_PARTIAL)
2204 status |= TP_STATUS_CSUMNOTREADY;
2205 else if (skb->pkt_type != PACKET_OUTGOING &&
2206 (skb->ip_summed == CHECKSUM_COMPLETE ||
2207 skb_csum_unnecessary(skb)))
2208 status |= TP_STATUS_CSUM_VALID;
2213 if (sk->sk_type == SOCK_DGRAM) {
2214 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2217 unsigned int maclen = skb_network_offset(skb);
2218 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2219 (maclen < 16 ? 16 : maclen)) +
2221 if (po->has_vnet_hdr) {
2222 netoff += sizeof(struct virtio_net_hdr);
2225 macoff = netoff - maclen;
2227 if (netoff > USHRT_MAX) {
2228 spin_lock(&sk->sk_receive_queue.lock);
2229 po->stats.stats1.tp_drops++;
2230 spin_unlock(&sk->sk_receive_queue.lock);
2231 goto drop_n_restore;
2233 if (po->tp_version <= TPACKET_V2) {
2234 if (macoff + snaplen > po->rx_ring.frame_size) {
2235 if (po->copy_thresh &&
2236 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2237 if (skb_shared(skb)) {
2238 copy_skb = skb_clone(skb, GFP_ATOMIC);
2240 copy_skb = skb_get(skb);
2241 skb_head = skb->data;
2244 skb_set_owner_r(copy_skb, sk);
2246 snaplen = po->rx_ring.frame_size - macoff;
2247 if ((int)snaplen < 0) {
2252 } else if (unlikely(macoff + snaplen >
2253 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2256 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2257 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2258 snaplen, nval, macoff);
2260 if (unlikely((int)snaplen < 0)) {
2262 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2266 spin_lock(&sk->sk_receive_queue.lock);
2267 h.raw = packet_current_rx_frame(po, skb,
2268 TP_STATUS_KERNEL, (macoff+snaplen));
2270 goto drop_n_account;
2271 if (po->tp_version <= TPACKET_V2) {
2272 packet_increment_rx_head(po, &po->rx_ring);
2274 * LOSING will be reported till you read the stats,
2275 * because it's COR - Clear On Read.
2276 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2279 if (po->stats.stats1.tp_drops)
2280 status |= TP_STATUS_LOSING;
2284 __packet_rcv_vnet(skb, h.raw + macoff -
2285 sizeof(struct virtio_net_hdr)))
2286 goto drop_n_account;
2288 po->stats.stats1.tp_packets++;
2290 status |= TP_STATUS_COPY;
2291 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2293 spin_unlock(&sk->sk_receive_queue.lock);
2295 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2297 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2298 getnstimeofday(&ts);
2300 status |= ts_status;
2302 switch (po->tp_version) {
2304 h.h1->tp_len = skb->len;
2305 h.h1->tp_snaplen = snaplen;
2306 h.h1->tp_mac = macoff;
2307 h.h1->tp_net = netoff;
2308 h.h1->tp_sec = ts.tv_sec;
2309 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2310 hdrlen = sizeof(*h.h1);
2313 h.h2->tp_len = skb->len;
2314 h.h2->tp_snaplen = snaplen;
2315 h.h2->tp_mac = macoff;
2316 h.h2->tp_net = netoff;
2317 h.h2->tp_sec = ts.tv_sec;
2318 h.h2->tp_nsec = ts.tv_nsec;
2319 if (skb_vlan_tag_present(skb)) {
2320 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2321 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2322 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2324 h.h2->tp_vlan_tci = 0;
2325 h.h2->tp_vlan_tpid = 0;
2327 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2328 hdrlen = sizeof(*h.h2);
2331 /* tp_nxt_offset,vlan are already populated above.
2332 * So DONT clear those fields here
2334 h.h3->tp_status |= status;
2335 h.h3->tp_len = skb->len;
2336 h.h3->tp_snaplen = snaplen;
2337 h.h3->tp_mac = macoff;
2338 h.h3->tp_net = netoff;
2339 h.h3->tp_sec = ts.tv_sec;
2340 h.h3->tp_nsec = ts.tv_nsec;
2341 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2342 hdrlen = sizeof(*h.h3);
2348 sll = h.raw + TPACKET_ALIGN(hdrlen);
2349 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2350 sll->sll_family = AF_PACKET;
2351 sll->sll_hatype = dev->type;
2352 sll->sll_protocol = skb->protocol;
2353 sll->sll_pkttype = skb->pkt_type;
2354 if (unlikely(po->origdev))
2355 sll->sll_ifindex = orig_dev->ifindex;
2357 sll->sll_ifindex = dev->ifindex;
2361 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2362 if (po->tp_version <= TPACKET_V2) {
2365 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2368 for (start = h.raw; start < end; start += PAGE_SIZE)
2369 flush_dcache_page(pgv_to_page(start));
2374 if (po->tp_version <= TPACKET_V2) {
2375 __packet_set_status(po, h.raw, status);
2376 sk->sk_data_ready(sk);
2378 prb_clear_blk_fill_status(&po->rx_ring);
2382 if (skb_head != skb->data && skb_shared(skb)) {
2383 skb->data = skb_head;
2387 if (!is_drop_n_account)
2394 is_drop_n_account = true;
2395 po->stats.stats1.tp_drops++;
2396 spin_unlock(&sk->sk_receive_queue.lock);
2398 sk->sk_data_ready(sk);
2399 kfree_skb(copy_skb);
2400 goto drop_n_restore;
2403 static void tpacket_destruct_skb(struct sk_buff *skb)
2405 struct packet_sock *po = pkt_sk(skb->sk);
2407 if (likely(po->tx_ring.pg_vec)) {
2411 ph = skb_shinfo(skb)->destructor_arg;
2412 packet_dec_pending(&po->tx_ring);
2414 ts = __packet_set_timestamp(po, ph, skb);
2415 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2417 if (!packet_read_pending(&po->tx_ring))
2418 complete(&po->skb_completion);
2424 static void tpacket_set_protocol(const struct net_device *dev,
2425 struct sk_buff *skb)
2427 if (dev->type == ARPHRD_ETHER) {
2428 skb_reset_mac_header(skb);
2429 skb->protocol = eth_hdr(skb)->h_proto;
2433 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2435 unsigned short gso_type = 0;
2437 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2438 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2439 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2440 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2441 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2442 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2443 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2445 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2448 if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2449 switch (vnet_hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2450 case VIRTIO_NET_HDR_GSO_TCPV4:
2451 gso_type = SKB_GSO_TCPV4;
2453 case VIRTIO_NET_HDR_GSO_TCPV6:
2454 gso_type = SKB_GSO_TCPV6;
2456 case VIRTIO_NET_HDR_GSO_UDP:
2457 gso_type = SKB_GSO_UDP;
2463 if (vnet_hdr->gso_type & VIRTIO_NET_HDR_GSO_ECN)
2464 gso_type |= SKB_GSO_TCP_ECN;
2466 if (vnet_hdr->gso_size == 0)
2470 vnet_hdr->gso_type = gso_type; /* changes type, temporary storage */
2474 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2475 struct virtio_net_hdr *vnet_hdr)
2479 if (*len < sizeof(*vnet_hdr))
2481 *len -= sizeof(*vnet_hdr);
2483 n = copy_from_iter(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter);
2484 if (n != sizeof(*vnet_hdr))
2487 return __packet_snd_vnet_parse(vnet_hdr, *len);
2490 static int packet_snd_vnet_gso(struct sk_buff *skb,
2491 struct virtio_net_hdr *vnet_hdr)
2493 if (vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2494 u16 s = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start);
2495 u16 o = __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset);
2497 if (!skb_partial_csum_set(skb, s, o))
2501 skb_shinfo(skb)->gso_size =
2502 __virtio16_to_cpu(vio_le(), vnet_hdr->gso_size);
2503 skb_shinfo(skb)->gso_type = vnet_hdr->gso_type;
2505 /* Header must be checked, and gso_segs computed. */
2506 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2507 skb_shinfo(skb)->gso_segs = 0;
2511 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2512 void *frame, struct net_device *dev, void *data, int tp_len,
2513 __be16 proto, unsigned char *addr, int hlen, int copylen,
2514 const struct sockcm_cookie *sockc)
2516 union tpacket_uhdr ph;
2517 int to_write, offset, len, nr_frags, len_max;
2518 struct socket *sock = po->sk.sk_socket;
2524 skb->protocol = proto;
2526 skb->priority = po->sk.sk_priority;
2527 skb->mark = po->sk.sk_mark;
2528 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2529 skb_shinfo(skb)->destructor_arg = ph.raw;
2531 skb_reserve(skb, hlen);
2532 skb_reset_network_header(skb);
2536 if (sock->type == SOCK_DGRAM) {
2537 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2539 if (unlikely(err < 0))
2541 } else if (copylen) {
2542 int hdrlen = min_t(int, copylen, tp_len);
2544 skb_push(skb, dev->hard_header_len);
2545 skb_put(skb, copylen - dev->hard_header_len);
2546 err = skb_store_bits(skb, 0, data, hdrlen);
2549 if (!dev_validate_header(dev, skb->data, hdrlen))
2552 tpacket_set_protocol(dev, skb);
2558 offset = offset_in_page(data);
2559 len_max = PAGE_SIZE - offset;
2560 len = ((to_write > len_max) ? len_max : to_write);
2562 skb->data_len = to_write;
2563 skb->len += to_write;
2564 skb->truesize += to_write;
2565 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2567 while (likely(to_write)) {
2568 nr_frags = skb_shinfo(skb)->nr_frags;
2570 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2571 pr_err("Packet exceed the number of skb frags(%lu)\n",
2576 page = pgv_to_page(data);
2578 flush_dcache_page(page);
2580 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2583 len_max = PAGE_SIZE;
2584 len = ((to_write > len_max) ? len_max : to_write);
2587 skb_probe_transport_header(skb, 0);
2592 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2593 int size_max, void **data)
2595 union tpacket_uhdr ph;
2600 switch (po->tp_version) {
2602 tp_len = ph.h2->tp_len;
2605 tp_len = ph.h1->tp_len;
2608 if (unlikely(tp_len > size_max)) {
2609 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2613 if (unlikely(po->tp_tx_has_off)) {
2614 int off_min, off_max;
2616 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2617 off_max = po->tx_ring.frame_size - tp_len;
2618 if (po->sk.sk_type == SOCK_DGRAM) {
2619 switch (po->tp_version) {
2621 off = ph.h2->tp_net;
2624 off = ph.h1->tp_net;
2628 switch (po->tp_version) {
2630 off = ph.h2->tp_mac;
2633 off = ph.h1->tp_mac;
2637 if (unlikely((off < off_min) || (off_max < off)))
2640 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2643 *data = frame + off;
2647 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2649 struct sk_buff *skb = NULL;
2650 struct net_device *dev;
2651 struct virtio_net_hdr *vnet_hdr = NULL;
2652 struct sockcm_cookie sockc;
2654 int err, reserve = 0;
2656 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2657 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2658 unsigned char *addr = NULL;
2659 int tp_len, size_max;
2662 int status = TP_STATUS_AVAILABLE;
2663 int hlen, tlen, copylen = 0;
2666 mutex_lock(&po->pg_vec_lock);
2668 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2669 * we need to confirm it under protection of pg_vec_lock.
2671 if (unlikely(!po->tx_ring.pg_vec)) {
2675 if (likely(saddr == NULL)) {
2676 dev = packet_cached_dev_get(po);
2677 proto = READ_ONCE(po->num);
2680 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2682 if (msg->msg_namelen < (saddr->sll_halen
2683 + offsetof(struct sockaddr_ll,
2686 proto = saddr->sll_protocol;
2687 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2688 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2689 if (dev && msg->msg_namelen < dev->addr_len +
2690 offsetof(struct sockaddr_ll, sll_addr))
2692 addr = saddr->sll_addr;
2697 if (unlikely(dev == NULL))
2700 if (unlikely(!(dev->flags & IFF_UP)))
2703 sockc.tsflags = po->sk.sk_tsflags;
2704 if (msg->msg_controllen) {
2705 err = sock_cmsg_send(&po->sk, msg, &sockc);
2710 if (po->sk.sk_socket->type == SOCK_RAW)
2711 reserve = dev->hard_header_len;
2712 size_max = po->tx_ring.frame_size
2713 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2715 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2716 size_max = dev->mtu + reserve + VLAN_HLEN;
2718 reinit_completion(&po->skb_completion);
2721 ph = packet_current_frame(po, &po->tx_ring,
2722 TP_STATUS_SEND_REQUEST);
2723 if (unlikely(ph == NULL)) {
2724 if (need_wait && skb) {
2725 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2726 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2728 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2732 /* check for additional frames */
2737 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2741 status = TP_STATUS_SEND_REQUEST;
2742 hlen = LL_RESERVED_SPACE(dev);
2743 tlen = dev->needed_tailroom;
2744 if (po->has_vnet_hdr) {
2746 data += sizeof(*vnet_hdr);
2747 tp_len -= sizeof(*vnet_hdr);
2749 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2753 copylen = __virtio16_to_cpu(vio_le(),
2756 copylen = max_t(int, copylen, dev->hard_header_len);
2757 skb = sock_alloc_send_skb(&po->sk,
2758 hlen + tlen + sizeof(struct sockaddr_ll) +
2759 (copylen - dev->hard_header_len),
2762 if (unlikely(skb == NULL)) {
2763 /* we assume the socket was initially writeable ... */
2764 if (likely(len_sum > 0))
2768 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2769 addr, hlen, copylen, &sockc);
2770 if (likely(tp_len >= 0) &&
2771 tp_len > dev->mtu + reserve &&
2772 !po->has_vnet_hdr &&
2773 !packet_extra_vlan_len_allowed(dev, skb))
2776 if (unlikely(tp_len < 0)) {
2779 __packet_set_status(po, ph,
2780 TP_STATUS_AVAILABLE);
2781 packet_increment_head(&po->tx_ring);
2785 status = TP_STATUS_WRONG_FORMAT;
2791 if (po->has_vnet_hdr && packet_snd_vnet_gso(skb, vnet_hdr)) {
2796 packet_pick_tx_queue(dev, skb);
2798 skb->destructor = tpacket_destruct_skb;
2799 __packet_set_status(po, ph, TP_STATUS_SENDING);
2800 packet_inc_pending(&po->tx_ring);
2802 status = TP_STATUS_SEND_REQUEST;
2803 err = po->xmit(skb);
2804 if (unlikely(err > 0)) {
2805 err = net_xmit_errno(err);
2806 if (err && __packet_get_status(po, ph) ==
2807 TP_STATUS_AVAILABLE) {
2808 /* skb was destructed already */
2813 * skb was dropped but not destructed yet;
2814 * let's treat it like congestion or err < 0
2818 packet_increment_head(&po->tx_ring);
2820 } while (likely((ph != NULL) ||
2821 /* Note: packet_read_pending() might be slow if we have
2822 * to call it as it's per_cpu variable, but in fast-path
2823 * we already short-circuit the loop with the first
2824 * condition, and luckily don't have to go that path
2827 (need_wait && packet_read_pending(&po->tx_ring))));
2833 __packet_set_status(po, ph, status);
2838 mutex_unlock(&po->pg_vec_lock);
2842 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2843 size_t reserve, size_t len,
2844 size_t linear, int noblock,
2847 struct sk_buff *skb;
2849 /* Under a page? Don't bother with paged skb. */
2850 if (prepad + len < PAGE_SIZE || !linear)
2853 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2858 skb_reserve(skb, reserve);
2859 skb_put(skb, linear);
2860 skb->data_len = len - linear;
2861 skb->len += len - linear;
2866 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2868 struct sock *sk = sock->sk;
2869 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2870 struct sk_buff *skb;
2871 struct net_device *dev;
2873 unsigned char *addr = NULL;
2874 int err, reserve = 0;
2875 struct sockcm_cookie sockc;
2876 struct virtio_net_hdr vnet_hdr = { 0 };
2878 struct packet_sock *po = pkt_sk(sk);
2879 bool has_vnet_hdr = false;
2880 int hlen, tlen, linear;
2884 * Get and verify the address.
2887 if (likely(saddr == NULL)) {
2888 dev = packet_cached_dev_get(po);
2889 proto = READ_ONCE(po->num);
2892 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2894 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2896 proto = saddr->sll_protocol;
2897 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2898 if (sock->type == SOCK_DGRAM) {
2899 if (dev && msg->msg_namelen < dev->addr_len +
2900 offsetof(struct sockaddr_ll, sll_addr))
2902 addr = saddr->sll_addr;
2907 if (unlikely(dev == NULL))
2910 if (unlikely(!(dev->flags & IFF_UP)))
2913 sockc.tsflags = sk->sk_tsflags;
2914 sockc.mark = sk->sk_mark;
2915 if (msg->msg_controllen) {
2916 err = sock_cmsg_send(sk, msg, &sockc);
2921 if (sock->type == SOCK_RAW)
2922 reserve = dev->hard_header_len;
2923 if (po->has_vnet_hdr) {
2924 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2927 has_vnet_hdr = true;
2930 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2931 if (!netif_supports_nofcs(dev)) {
2932 err = -EPROTONOSUPPORT;
2935 extra_len = 4; /* We're doing our own CRC */
2939 if (!vnet_hdr.gso_type &&
2940 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2944 hlen = LL_RESERVED_SPACE(dev);
2945 tlen = dev->needed_tailroom;
2946 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2947 linear = max(linear, min_t(int, len, dev->hard_header_len));
2948 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2949 msg->msg_flags & MSG_DONTWAIT, &err);
2953 skb_reset_network_header(skb);
2956 if (sock->type == SOCK_DGRAM) {
2957 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2958 if (unlikely(offset < 0))
2960 } else if (reserve) {
2961 skb_reserve(skb, -reserve);
2963 skb_reset_network_header(skb);
2966 /* Returns -EFAULT on error */
2967 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2971 if (sock->type == SOCK_RAW &&
2972 !dev_validate_header(dev, skb->data, len)) {
2977 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2979 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2980 !packet_extra_vlan_len_allowed(dev, skb)) {
2985 skb->protocol = proto;
2987 skb->priority = sk->sk_priority;
2988 skb->mark = sockc.mark;
2990 packet_pick_tx_queue(dev, skb);
2993 err = packet_snd_vnet_gso(skb, &vnet_hdr);
2996 len += sizeof(vnet_hdr);
2999 skb_probe_transport_header(skb, reserve);
3001 if (unlikely(extra_len == 4))
3004 err = po->xmit(skb);
3005 if (err > 0 && (err = net_xmit_errno(err)) != 0)
3021 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3023 struct sock *sk = sock->sk;
3024 struct packet_sock *po = pkt_sk(sk);
3026 if (po->tx_ring.pg_vec)
3027 return tpacket_snd(po, msg);
3029 return packet_snd(sock, msg, len);
3033 * Close a PACKET socket. This is fairly simple. We immediately go
3034 * to 'closed' state and remove our protocol entry in the device list.
3037 static int packet_release(struct socket *sock)
3039 struct sock *sk = sock->sk;
3040 struct packet_sock *po;
3041 struct packet_fanout *f;
3043 union tpacket_req_u req_u;
3051 mutex_lock(&net->packet.sklist_lock);
3052 sk_del_node_init_rcu(sk);
3053 mutex_unlock(&net->packet.sklist_lock);
3056 sock_prot_inuse_add(net, sk->sk_prot, -1);
3059 spin_lock(&po->bind_lock);
3060 unregister_prot_hook(sk, false);
3061 packet_cached_dev_reset(po);
3063 if (po->prot_hook.dev) {
3064 dev_put(po->prot_hook.dev);
3065 po->prot_hook.dev = NULL;
3067 spin_unlock(&po->bind_lock);
3069 packet_flush_mclist(sk);
3072 if (po->rx_ring.pg_vec) {
3073 memset(&req_u, 0, sizeof(req_u));
3074 packet_set_ring(sk, &req_u, 1, 0);
3077 if (po->tx_ring.pg_vec) {
3078 memset(&req_u, 0, sizeof(req_u));
3079 packet_set_ring(sk, &req_u, 1, 1);
3083 f = fanout_release(sk);
3088 kfree(po->rollover);
3089 fanout_release_data(f);
3093 * Now the socket is dead. No more input will appear.
3100 skb_queue_purge(&sk->sk_receive_queue);
3101 packet_free_pending(po);
3102 sk_refcnt_debug_release(sk);
3109 * Attach a packet hook.
3112 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3115 struct packet_sock *po = pkt_sk(sk);
3116 struct net_device *dev_curr;
3119 struct net_device *dev = NULL;
3121 bool unlisted = false;
3124 spin_lock(&po->bind_lock);
3133 dev = dev_get_by_name_rcu(sock_net(sk), name);
3138 } else if (ifindex) {
3139 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3149 proto_curr = po->prot_hook.type;
3150 dev_curr = po->prot_hook.dev;
3152 need_rehook = proto_curr != proto || dev_curr != dev;
3157 /* prevents packet_notifier() from calling
3158 * register_prot_hook()
3160 WRITE_ONCE(po->num, 0);
3161 __unregister_prot_hook(sk, true);
3163 dev_curr = po->prot_hook.dev;
3165 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3169 BUG_ON(po->running);
3170 WRITE_ONCE(po->num, proto);
3171 po->prot_hook.type = proto;
3173 if (unlikely(unlisted)) {
3175 po->prot_hook.dev = NULL;
3176 WRITE_ONCE(po->ifindex, -1);
3177 packet_cached_dev_reset(po);
3179 po->prot_hook.dev = dev;
3180 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3181 packet_cached_dev_assign(po, dev);
3187 if (proto == 0 || !need_rehook)
3190 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3191 register_prot_hook(sk);
3193 sk->sk_err = ENETDOWN;
3194 if (!sock_flag(sk, SOCK_DEAD))
3195 sk->sk_error_report(sk);
3200 spin_unlock(&po->bind_lock);
3206 * Bind a packet socket to a device
3209 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3212 struct sock *sk = sock->sk;
3213 char name[sizeof(uaddr->sa_data) + 1];
3219 if (addr_len != sizeof(struct sockaddr))
3221 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3224 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3225 name[sizeof(uaddr->sa_data)] = 0;
3227 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3230 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3232 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3233 struct sock *sk = sock->sk;
3239 if (addr_len < sizeof(struct sockaddr_ll))
3241 if (sll->sll_family != AF_PACKET)
3244 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3245 sll->sll_protocol ? : pkt_sk(sk)->num);
3248 static struct proto packet_proto = {
3250 .owner = THIS_MODULE,
3251 .obj_size = sizeof(struct packet_sock),
3255 * Create a packet of type SOCK_PACKET.
3258 static int packet_create(struct net *net, struct socket *sock, int protocol,
3262 struct packet_sock *po;
3263 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3266 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3268 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3269 sock->type != SOCK_PACKET)
3270 return -ESOCKTNOSUPPORT;
3272 sock->state = SS_UNCONNECTED;
3275 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3279 sock->ops = &packet_ops;
3280 if (sock->type == SOCK_PACKET)
3281 sock->ops = &packet_ops_spkt;
3283 sock_init_data(sock, sk);
3286 init_completion(&po->skb_completion);
3287 sk->sk_family = PF_PACKET;
3289 po->xmit = dev_queue_xmit;
3291 err = packet_alloc_pending(po);
3295 packet_cached_dev_reset(po);
3297 sk->sk_destruct = packet_sock_destruct;
3298 sk_refcnt_debug_inc(sk);
3301 * Attach a protocol block
3304 spin_lock_init(&po->bind_lock);
3305 mutex_init(&po->pg_vec_lock);
3306 po->rollover = NULL;
3307 po->prot_hook.func = packet_rcv;
3309 if (sock->type == SOCK_PACKET)
3310 po->prot_hook.func = packet_rcv_spkt;
3312 po->prot_hook.af_packet_priv = sk;
3315 po->prot_hook.type = proto;
3316 __register_prot_hook(sk);
3319 mutex_lock(&net->packet.sklist_lock);
3320 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3321 mutex_unlock(&net->packet.sklist_lock);
3324 sock_prot_inuse_add(net, &packet_proto, 1);
3335 * Pull a packet from our receive queue and hand it to the user.
3336 * If necessary we block.
3339 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3342 struct sock *sk = sock->sk;
3343 struct sk_buff *skb;
3345 int vnet_hdr_len = 0;
3346 unsigned int origlen = 0;
3349 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3353 /* What error should we return now? EUNATTACH? */
3354 if (pkt_sk(sk)->ifindex < 0)
3358 if (flags & MSG_ERRQUEUE) {
3359 err = sock_recv_errqueue(sk, msg, len,
3360 SOL_PACKET, PACKET_TX_TIMESTAMP);
3365 * Call the generic datagram receiver. This handles all sorts
3366 * of horrible races and re-entrancy so we can forget about it
3367 * in the protocol layers.
3369 * Now it will return ENETDOWN, if device have just gone down,
3370 * but then it will block.
3373 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3376 * An error occurred so return it. Because skb_recv_datagram()
3377 * handles the blocking we don't see and worry about blocking
3384 if (pkt_sk(sk)->pressure)
3385 packet_rcv_has_room(pkt_sk(sk), NULL);
3387 if (pkt_sk(sk)->has_vnet_hdr) {
3388 err = packet_rcv_vnet(msg, skb, &len);
3391 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3394 /* You lose any data beyond the buffer you gave. If it worries
3395 * a user program they can ask the device for its MTU
3401 msg->msg_flags |= MSG_TRUNC;
3404 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3408 if (sock->type != SOCK_PACKET) {
3409 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3411 /* Original length was stored in sockaddr_ll fields */
3412 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3413 sll->sll_family = AF_PACKET;
3414 sll->sll_protocol = skb->protocol;
3417 sock_recv_ts_and_drops(msg, sk, skb);
3419 if (msg->msg_name) {
3422 /* If the address length field is there to be filled
3423 * in, we fill it in now.
3425 if (sock->type == SOCK_PACKET) {
3426 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3427 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3428 copy_len = msg->msg_namelen;
3430 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3432 msg->msg_namelen = sll->sll_halen +
3433 offsetof(struct sockaddr_ll, sll_addr);
3434 copy_len = msg->msg_namelen;
3435 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3436 memset(msg->msg_name +
3437 offsetof(struct sockaddr_ll, sll_addr),
3438 0, sizeof(sll->sll_addr));
3439 msg->msg_namelen = sizeof(struct sockaddr_ll);
3442 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3445 if (pkt_sk(sk)->auxdata) {
3446 struct tpacket_auxdata aux;
3448 aux.tp_status = TP_STATUS_USER;
3449 if (skb->ip_summed == CHECKSUM_PARTIAL)
3450 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3451 else if (skb->pkt_type != PACKET_OUTGOING &&
3452 (skb->ip_summed == CHECKSUM_COMPLETE ||
3453 skb_csum_unnecessary(skb)))
3454 aux.tp_status |= TP_STATUS_CSUM_VALID;
3456 aux.tp_len = origlen;
3457 aux.tp_snaplen = skb->len;
3459 aux.tp_net = skb_network_offset(skb);
3460 if (skb_vlan_tag_present(skb)) {
3461 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3462 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3463 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3465 aux.tp_vlan_tci = 0;
3466 aux.tp_vlan_tpid = 0;
3468 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3472 * Free or return the buffer as appropriate. Again this
3473 * hides all the races and re-entrancy issues from us.
3475 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3478 skb_free_datagram(sk, skb);
3483 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3484 int *uaddr_len, int peer)
3486 struct net_device *dev;
3487 struct sock *sk = sock->sk;
3492 uaddr->sa_family = AF_PACKET;
3493 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3495 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3497 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3499 *uaddr_len = sizeof(*uaddr);
3504 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3505 int *uaddr_len, int peer)
3507 struct net_device *dev;
3508 struct sock *sk = sock->sk;
3509 struct packet_sock *po = pkt_sk(sk);
3510 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3516 ifindex = READ_ONCE(po->ifindex);
3517 sll->sll_family = AF_PACKET;
3518 sll->sll_ifindex = ifindex;
3519 sll->sll_protocol = READ_ONCE(po->num);
3520 sll->sll_pkttype = 0;
3522 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3524 sll->sll_hatype = dev->type;
3525 sll->sll_halen = dev->addr_len;
3526 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3528 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3532 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3537 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3541 case PACKET_MR_MULTICAST:
3542 if (i->alen != dev->addr_len)
3545 return dev_mc_add(dev, i->addr);
3547 return dev_mc_del(dev, i->addr);
3549 case PACKET_MR_PROMISC:
3550 return dev_set_promiscuity(dev, what);
3551 case PACKET_MR_ALLMULTI:
3552 return dev_set_allmulti(dev, what);
3553 case PACKET_MR_UNICAST:
3554 if (i->alen != dev->addr_len)
3557 return dev_uc_add(dev, i->addr);
3559 return dev_uc_del(dev, i->addr);
3567 static void packet_dev_mclist_delete(struct net_device *dev,
3568 struct packet_mclist **mlp)
3570 struct packet_mclist *ml;
3572 while ((ml = *mlp) != NULL) {
3573 if (ml->ifindex == dev->ifindex) {
3574 packet_dev_mc(dev, ml, -1);
3582 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3584 struct packet_sock *po = pkt_sk(sk);
3585 struct packet_mclist *ml, *i;
3586 struct net_device *dev;
3592 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3597 if (mreq->mr_alen > dev->addr_len)
3601 i = kmalloc(sizeof(*i), GFP_KERNEL);
3606 for (ml = po->mclist; ml; ml = ml->next) {
3607 if (ml->ifindex == mreq->mr_ifindex &&
3608 ml->type == mreq->mr_type &&
3609 ml->alen == mreq->mr_alen &&
3610 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3612 /* Free the new element ... */
3618 i->type = mreq->mr_type;
3619 i->ifindex = mreq->mr_ifindex;
3620 i->alen = mreq->mr_alen;
3621 memcpy(i->addr, mreq->mr_address, i->alen);
3622 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3624 i->next = po->mclist;
3626 err = packet_dev_mc(dev, i, 1);
3628 po->mclist = i->next;
3637 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3639 struct packet_mclist *ml, **mlp;
3643 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3644 if (ml->ifindex == mreq->mr_ifindex &&
3645 ml->type == mreq->mr_type &&
3646 ml->alen == mreq->mr_alen &&
3647 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3648 if (--ml->count == 0) {
3649 struct net_device *dev;
3651 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3653 packet_dev_mc(dev, ml, -1);
3663 static void packet_flush_mclist(struct sock *sk)
3665 struct packet_sock *po = pkt_sk(sk);
3666 struct packet_mclist *ml;
3672 while ((ml = po->mclist) != NULL) {
3673 struct net_device *dev;
3675 po->mclist = ml->next;
3676 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3678 packet_dev_mc(dev, ml, -1);
3685 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3687 struct sock *sk = sock->sk;
3688 struct packet_sock *po = pkt_sk(sk);
3691 if (level != SOL_PACKET)
3692 return -ENOPROTOOPT;
3695 case PACKET_ADD_MEMBERSHIP:
3696 case PACKET_DROP_MEMBERSHIP:
3698 struct packet_mreq_max mreq;
3700 memset(&mreq, 0, sizeof(mreq));
3701 if (len < sizeof(struct packet_mreq))
3703 if (len > sizeof(mreq))
3705 if (copy_from_user(&mreq, optval, len))
3707 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3709 if (optname == PACKET_ADD_MEMBERSHIP)
3710 ret = packet_mc_add(sk, &mreq);
3712 ret = packet_mc_drop(sk, &mreq);
3716 case PACKET_RX_RING:
3717 case PACKET_TX_RING:
3719 union tpacket_req_u req_u;
3723 switch (po->tp_version) {
3726 len = sizeof(req_u.req);
3730 len = sizeof(req_u.req3);
3736 if (copy_from_user(&req_u.req, optval, len))
3739 ret = packet_set_ring(sk, &req_u, 0,
3740 optname == PACKET_TX_RING);
3745 case PACKET_COPY_THRESH:
3749 if (optlen != sizeof(val))
3751 if (copy_from_user(&val, optval, sizeof(val)))
3754 pkt_sk(sk)->copy_thresh = val;
3757 case PACKET_VERSION:
3761 if (optlen != sizeof(val))
3763 if (copy_from_user(&val, optval, sizeof(val)))
3774 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3777 po->tp_version = val;
3783 case PACKET_RESERVE:
3787 if (optlen != sizeof(val))
3789 if (copy_from_user(&val, optval, sizeof(val)))
3794 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3797 po->tp_reserve = val;
3807 if (optlen != sizeof(val))
3809 if (copy_from_user(&val, optval, sizeof(val)))
3813 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3816 po->tp_loss = !!val;
3822 case PACKET_AUXDATA:
3826 if (optlen < sizeof(val))
3828 if (copy_from_user(&val, optval, sizeof(val)))
3832 po->auxdata = !!val;
3836 case PACKET_ORIGDEV:
3840 if (optlen < sizeof(val))
3842 if (copy_from_user(&val, optval, sizeof(val)))
3846 po->origdev = !!val;
3850 case PACKET_VNET_HDR:
3854 if (sock->type != SOCK_RAW)
3856 if (optlen < sizeof(val))
3858 if (copy_from_user(&val, optval, sizeof(val)))
3862 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3865 po->has_vnet_hdr = !!val;
3871 case PACKET_TIMESTAMP:
3875 if (optlen != sizeof(val))
3877 if (copy_from_user(&val, optval, sizeof(val)))
3880 po->tp_tstamp = val;
3887 if (optlen != sizeof(val))
3889 if (copy_from_user(&val, optval, sizeof(val)))
3892 return fanout_add(sk, val & 0xffff, val >> 16);
3894 case PACKET_FANOUT_DATA:
3899 return fanout_set_data(po, optval, optlen);
3901 case PACKET_TX_HAS_OFF:
3905 if (optlen != sizeof(val))
3907 if (copy_from_user(&val, optval, sizeof(val)))
3911 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3914 po->tp_tx_has_off = !!val;
3920 case PACKET_QDISC_BYPASS:
3924 if (optlen != sizeof(val))
3926 if (copy_from_user(&val, optval, sizeof(val)))
3929 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3933 return -ENOPROTOOPT;
3937 static int packet_getsockopt(struct socket *sock, int level, int optname,
3938 char __user *optval, int __user *optlen)
3941 int val, lv = sizeof(val);
3942 struct sock *sk = sock->sk;
3943 struct packet_sock *po = pkt_sk(sk);
3945 union tpacket_stats_u st;
3946 struct tpacket_rollover_stats rstats;
3948 if (level != SOL_PACKET)
3949 return -ENOPROTOOPT;
3951 if (get_user(len, optlen))
3958 case PACKET_STATISTICS:
3959 spin_lock_bh(&sk->sk_receive_queue.lock);
3960 memcpy(&st, &po->stats, sizeof(st));
3961 memset(&po->stats, 0, sizeof(po->stats));
3962 spin_unlock_bh(&sk->sk_receive_queue.lock);
3964 if (po->tp_version == TPACKET_V3) {
3965 lv = sizeof(struct tpacket_stats_v3);
3966 st.stats3.tp_packets += st.stats3.tp_drops;
3969 lv = sizeof(struct tpacket_stats);
3970 st.stats1.tp_packets += st.stats1.tp_drops;
3975 case PACKET_AUXDATA:
3978 case PACKET_ORIGDEV:
3981 case PACKET_VNET_HDR:
3982 val = po->has_vnet_hdr;
3984 case PACKET_VERSION:
3985 val = po->tp_version;
3988 if (len > sizeof(int))
3990 if (len < sizeof(int))
3992 if (copy_from_user(&val, optval, len))
3996 val = sizeof(struct tpacket_hdr);
3999 val = sizeof(struct tpacket2_hdr);
4002 val = sizeof(struct tpacket3_hdr);
4008 case PACKET_RESERVE:
4009 val = po->tp_reserve;
4014 case PACKET_TIMESTAMP:
4015 val = po->tp_tstamp;
4019 ((u32)po->fanout->id |
4020 ((u32)po->fanout->type << 16) |
4021 ((u32)po->fanout->flags << 24)) :
4024 case PACKET_ROLLOVER_STATS:
4027 rstats.tp_all = atomic_long_read(&po->rollover->num);
4028 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4029 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4031 lv = sizeof(rstats);
4033 case PACKET_TX_HAS_OFF:
4034 val = po->tp_tx_has_off;
4036 case PACKET_QDISC_BYPASS:
4037 val = packet_use_direct_xmit(po);
4040 return -ENOPROTOOPT;
4045 if (put_user(len, optlen))
4047 if (copy_to_user(optval, data, len))
4053 #ifdef CONFIG_COMPAT
4054 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4055 char __user *optval, unsigned int optlen)
4057 struct packet_sock *po = pkt_sk(sock->sk);
4059 if (level != SOL_PACKET)
4060 return -ENOPROTOOPT;
4062 if (optname == PACKET_FANOUT_DATA &&
4063 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4064 optval = (char __user *)get_compat_bpf_fprog(optval);
4067 optlen = sizeof(struct sock_fprog);
4070 return packet_setsockopt(sock, level, optname, optval, optlen);
4074 static int packet_notifier(struct notifier_block *this,
4075 unsigned long msg, void *ptr)
4078 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4079 struct net *net = dev_net(dev);
4082 sk_for_each_rcu(sk, &net->packet.sklist) {
4083 struct packet_sock *po = pkt_sk(sk);
4086 case NETDEV_UNREGISTER:
4088 packet_dev_mclist_delete(dev, &po->mclist);
4092 if (dev->ifindex == po->ifindex) {
4093 spin_lock(&po->bind_lock);
4095 __unregister_prot_hook(sk, false);
4096 sk->sk_err = ENETDOWN;
4097 if (!sock_flag(sk, SOCK_DEAD))
4098 sk->sk_error_report(sk);
4100 if (msg == NETDEV_UNREGISTER) {
4101 packet_cached_dev_reset(po);
4102 WRITE_ONCE(po->ifindex, -1);
4103 if (po->prot_hook.dev)
4104 dev_put(po->prot_hook.dev);
4105 po->prot_hook.dev = NULL;
4107 spin_unlock(&po->bind_lock);
4111 if (dev->ifindex == po->ifindex) {
4112 spin_lock(&po->bind_lock);
4114 register_prot_hook(sk);
4115 spin_unlock(&po->bind_lock);
4125 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4128 struct sock *sk = sock->sk;
4133 int amount = sk_wmem_alloc_get(sk);
4135 return put_user(amount, (int __user *)arg);
4139 struct sk_buff *skb;
4142 spin_lock_bh(&sk->sk_receive_queue.lock);
4143 skb = skb_peek(&sk->sk_receive_queue);
4146 spin_unlock_bh(&sk->sk_receive_queue.lock);
4147 return put_user(amount, (int __user *)arg);
4150 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4152 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4162 case SIOCGIFBRDADDR:
4163 case SIOCSIFBRDADDR:
4164 case SIOCGIFNETMASK:
4165 case SIOCSIFNETMASK:
4166 case SIOCGIFDSTADDR:
4167 case SIOCSIFDSTADDR:
4169 return inet_dgram_ops.ioctl(sock, cmd, arg);
4173 return -ENOIOCTLCMD;
4178 static unsigned int packet_poll(struct file *file, struct socket *sock,
4181 struct sock *sk = sock->sk;
4182 struct packet_sock *po = pkt_sk(sk);
4183 unsigned int mask = datagram_poll(file, sock, wait);
4185 spin_lock_bh(&sk->sk_receive_queue.lock);
4186 if (po->rx_ring.pg_vec) {
4187 if (!packet_previous_rx_frame(po, &po->rx_ring,
4189 mask |= POLLIN | POLLRDNORM;
4191 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4193 spin_unlock_bh(&sk->sk_receive_queue.lock);
4194 spin_lock_bh(&sk->sk_write_queue.lock);
4195 if (po->tx_ring.pg_vec) {
4196 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4197 mask |= POLLOUT | POLLWRNORM;
4199 spin_unlock_bh(&sk->sk_write_queue.lock);
4204 /* Dirty? Well, I still did not learn better way to account
4208 static void packet_mm_open(struct vm_area_struct *vma)
4210 struct file *file = vma->vm_file;
4211 struct socket *sock = file->private_data;
4212 struct sock *sk = sock->sk;
4215 atomic_inc(&pkt_sk(sk)->mapped);
4218 static void packet_mm_close(struct vm_area_struct *vma)
4220 struct file *file = vma->vm_file;
4221 struct socket *sock = file->private_data;
4222 struct sock *sk = sock->sk;
4225 atomic_dec(&pkt_sk(sk)->mapped);
4228 static const struct vm_operations_struct packet_mmap_ops = {
4229 .open = packet_mm_open,
4230 .close = packet_mm_close,
4233 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4238 for (i = 0; i < len; i++) {
4239 if (likely(pg_vec[i].buffer)) {
4240 if (is_vmalloc_addr(pg_vec[i].buffer))
4241 vfree(pg_vec[i].buffer);
4243 free_pages((unsigned long)pg_vec[i].buffer,
4245 pg_vec[i].buffer = NULL;
4251 static char *alloc_one_pg_vec_page(unsigned long order)
4254 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4255 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4257 buffer = (char *) __get_free_pages(gfp_flags, order);
4261 /* __get_free_pages failed, fall back to vmalloc */
4262 buffer = vzalloc((1 << order) * PAGE_SIZE);
4266 /* vmalloc failed, lets dig into swap here */
4267 gfp_flags &= ~__GFP_NORETRY;
4268 buffer = (char *) __get_free_pages(gfp_flags, order);
4272 /* complete and utter failure */
4276 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4278 unsigned int block_nr = req->tp_block_nr;
4282 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4283 if (unlikely(!pg_vec))
4286 for (i = 0; i < block_nr; i++) {
4287 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4288 if (unlikely(!pg_vec[i].buffer))
4289 goto out_free_pgvec;
4296 free_pg_vec(pg_vec, order, block_nr);
4301 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4302 int closing, int tx_ring)
4304 struct pgv *pg_vec = NULL;
4305 struct packet_sock *po = pkt_sk(sk);
4306 int was_running, order = 0;
4307 struct packet_ring_buffer *rb;
4308 struct sk_buff_head *rb_queue;
4311 /* Added to avoid minimal code churn */
4312 struct tpacket_req *req = &req_u->req;
4314 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4315 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4316 net_warn_ratelimited("Tx-ring is not supported.\n");
4320 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4321 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4325 if (atomic_read(&po->mapped))
4327 if (packet_read_pending(rb))
4331 if (req->tp_block_nr) {
4332 unsigned int min_frame_size;
4334 /* Sanity tests and some calculations */
4336 if (unlikely(rb->pg_vec))
4339 switch (po->tp_version) {
4341 po->tp_hdrlen = TPACKET_HDRLEN;
4344 po->tp_hdrlen = TPACKET2_HDRLEN;
4347 po->tp_hdrlen = TPACKET3_HDRLEN;
4352 if (unlikely((int)req->tp_block_size <= 0))
4354 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4356 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4357 if (po->tp_version >= TPACKET_V3 &&
4358 req->tp_block_size <
4359 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4361 if (unlikely(req->tp_frame_size < min_frame_size))
4363 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4366 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4367 if (unlikely(rb->frames_per_block == 0))
4369 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4371 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4376 order = get_order(req->tp_block_size);
4377 pg_vec = alloc_pg_vec(req, order);
4378 if (unlikely(!pg_vec))
4380 switch (po->tp_version) {
4382 /* Transmit path is not supported. We checked
4383 * it above but just being paranoid
4386 init_prb_bdqc(po, rb, pg_vec, req_u);
4395 if (unlikely(req->tp_frame_nr))
4400 /* Detach socket from network */
4401 spin_lock(&po->bind_lock);
4402 was_running = po->running;
4405 WRITE_ONCE(po->num, 0);
4406 __unregister_prot_hook(sk, false);
4408 spin_unlock(&po->bind_lock);
4413 mutex_lock(&po->pg_vec_lock);
4414 if (closing || atomic_read(&po->mapped) == 0) {
4416 spin_lock_bh(&rb_queue->lock);
4417 swap(rb->pg_vec, pg_vec);
4418 rb->frame_max = (req->tp_frame_nr - 1);
4420 rb->frame_size = req->tp_frame_size;
4421 spin_unlock_bh(&rb_queue->lock);
4423 swap(rb->pg_vec_order, order);
4424 swap(rb->pg_vec_len, req->tp_block_nr);
4426 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4427 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4428 tpacket_rcv : packet_rcv;
4429 skb_queue_purge(rb_queue);
4430 if (atomic_read(&po->mapped))
4431 pr_err("packet_mmap: vma is busy: %d\n",
4432 atomic_read(&po->mapped));
4434 mutex_unlock(&po->pg_vec_lock);
4436 spin_lock(&po->bind_lock);
4438 WRITE_ONCE(po->num, num);
4439 register_prot_hook(sk);
4441 spin_unlock(&po->bind_lock);
4442 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4443 /* Because we don't support block-based V3 on tx-ring */
4445 prb_shutdown_retire_blk_timer(po, rb_queue);
4449 free_pg_vec(pg_vec, order, req->tp_block_nr);
4454 static int packet_mmap(struct file *file, struct socket *sock,
4455 struct vm_area_struct *vma)
4457 struct sock *sk = sock->sk;
4458 struct packet_sock *po = pkt_sk(sk);
4459 unsigned long size, expected_size;
4460 struct packet_ring_buffer *rb;
4461 unsigned long start;
4468 mutex_lock(&po->pg_vec_lock);
4471 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4473 expected_size += rb->pg_vec_len
4479 if (expected_size == 0)
4482 size = vma->vm_end - vma->vm_start;
4483 if (size != expected_size)
4486 start = vma->vm_start;
4487 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4488 if (rb->pg_vec == NULL)
4491 for (i = 0; i < rb->pg_vec_len; i++) {
4493 void *kaddr = rb->pg_vec[i].buffer;
4496 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4497 page = pgv_to_page(kaddr);
4498 err = vm_insert_page(vma, start, page);
4507 atomic_inc(&po->mapped);
4508 vma->vm_ops = &packet_mmap_ops;
4512 mutex_unlock(&po->pg_vec_lock);
4516 static const struct proto_ops packet_ops_spkt = {
4517 .family = PF_PACKET,
4518 .owner = THIS_MODULE,
4519 .release = packet_release,
4520 .bind = packet_bind_spkt,
4521 .connect = sock_no_connect,
4522 .socketpair = sock_no_socketpair,
4523 .accept = sock_no_accept,
4524 .getname = packet_getname_spkt,
4525 .poll = datagram_poll,
4526 .ioctl = packet_ioctl,
4527 .listen = sock_no_listen,
4528 .shutdown = sock_no_shutdown,
4529 .setsockopt = sock_no_setsockopt,
4530 .getsockopt = sock_no_getsockopt,
4531 .sendmsg = packet_sendmsg_spkt,
4532 .recvmsg = packet_recvmsg,
4533 .mmap = sock_no_mmap,
4534 .sendpage = sock_no_sendpage,
4537 static const struct proto_ops packet_ops = {
4538 .family = PF_PACKET,
4539 .owner = THIS_MODULE,
4540 .release = packet_release,
4541 .bind = packet_bind,
4542 .connect = sock_no_connect,
4543 .socketpair = sock_no_socketpair,
4544 .accept = sock_no_accept,
4545 .getname = packet_getname,
4546 .poll = packet_poll,
4547 .ioctl = packet_ioctl,
4548 .listen = sock_no_listen,
4549 .shutdown = sock_no_shutdown,
4550 .setsockopt = packet_setsockopt,
4551 .getsockopt = packet_getsockopt,
4552 #ifdef CONFIG_COMPAT
4553 .compat_setsockopt = compat_packet_setsockopt,
4555 .sendmsg = packet_sendmsg,
4556 .recvmsg = packet_recvmsg,
4557 .mmap = packet_mmap,
4558 .sendpage = sock_no_sendpage,
4561 static const struct net_proto_family packet_family_ops = {
4562 .family = PF_PACKET,
4563 .create = packet_create,
4564 .owner = THIS_MODULE,
4567 static struct notifier_block packet_netdev_notifier = {
4568 .notifier_call = packet_notifier,
4571 #ifdef CONFIG_PROC_FS
4573 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4576 struct net *net = seq_file_net(seq);
4579 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4582 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4584 struct net *net = seq_file_net(seq);
4585 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4588 static void packet_seq_stop(struct seq_file *seq, void *v)
4594 static int packet_seq_show(struct seq_file *seq, void *v)
4596 if (v == SEQ_START_TOKEN)
4597 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4599 struct sock *s = sk_entry(v);
4600 const struct packet_sock *po = pkt_sk(s);
4603 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4605 atomic_read(&s->sk_refcnt),
4607 ntohs(READ_ONCE(po->num)),
4608 READ_ONCE(po->ifindex),
4610 atomic_read(&s->sk_rmem_alloc),
4611 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4618 static const struct seq_operations packet_seq_ops = {
4619 .start = packet_seq_start,
4620 .next = packet_seq_next,
4621 .stop = packet_seq_stop,
4622 .show = packet_seq_show,
4625 static int packet_seq_open(struct inode *inode, struct file *file)
4627 return seq_open_net(inode, file, &packet_seq_ops,
4628 sizeof(struct seq_net_private));
4631 static const struct file_operations packet_seq_fops = {
4632 .owner = THIS_MODULE,
4633 .open = packet_seq_open,
4635 .llseek = seq_lseek,
4636 .release = seq_release_net,
4641 static int __net_init packet_net_init(struct net *net)
4643 mutex_init(&net->packet.sklist_lock);
4644 INIT_HLIST_HEAD(&net->packet.sklist);
4646 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4652 static void __net_exit packet_net_exit(struct net *net)
4654 remove_proc_entry("packet", net->proc_net);
4657 static struct pernet_operations packet_net_ops = {
4658 .init = packet_net_init,
4659 .exit = packet_net_exit,
4663 static void __exit packet_exit(void)
4665 unregister_netdevice_notifier(&packet_netdev_notifier);
4666 unregister_pernet_subsys(&packet_net_ops);
4667 sock_unregister(PF_PACKET);
4668 proto_unregister(&packet_proto);
4671 static int __init packet_init(void)
4675 rc = proto_register(&packet_proto, 0);
4678 rc = sock_register(&packet_family_ops);
4681 rc = register_pernet_subsys(&packet_net_ops);
4684 rc = register_netdevice_notifier(&packet_netdev_notifier);
4691 unregister_pernet_subsys(&packet_net_ops);
4693 sock_unregister(PF_PACKET);
4695 proto_unregister(&packet_proto);
4700 module_init(packet_init);
4701 module_exit(packet_exit);
4702 MODULE_LICENSE("GPL");
4703 MODULE_ALIAS_NETPROTO(PF_PACKET);