1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PACKET - implements raw packet sockets.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
49 #include <linux/types.h>
51 #include <linux/capability.h>
52 #include <linux/fcntl.h>
53 #include <linux/socket.h>
55 #include <linux/inet.h>
56 #include <linux/netdevice.h>
57 #include <linux/if_packet.h>
58 #include <linux/wireless.h>
59 #include <linux/kernel.h>
60 #include <linux/kmod.h>
61 #include <linux/slab.h>
62 #include <linux/vmalloc.h>
63 #include <net/net_namespace.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.h>
68 #include <linux/errno.h>
69 #include <linux/timer.h>
70 #include <linux/uaccess.h>
71 #include <asm/ioctls.h>
73 #include <asm/cacheflush.h>
75 #include <linux/proc_fs.h>
76 #include <linux/seq_file.h>
77 #include <linux/poll.h>
78 #include <linux/module.h>
79 #include <linux/init.h>
80 #include <linux/mutex.h>
81 #include <linux/if_vlan.h>
82 #include <linux/virtio_net.h>
83 #include <linux/errqueue.h>
84 #include <linux/net_tstamp.h>
85 #include <linux/percpu.h>
87 #include <net/inet_common.h>
89 #include <linux/bpf.h>
90 #include <net/compat.h>
96 - if device has no dev->hard_header routine, it adds and removes ll header
97 inside itself. In this case ll header is invisible outside of device,
98 but higher levels still should reserve dev->hard_header_len.
99 Some devices are enough clever to reallocate skb, when header
100 will not fit to reserved space (tunnel), another ones are silly
102 - packet socket receives packets with pulled ll header,
103 so that SOCK_RAW should push it back.
108 Incoming, dev->hard_header!=NULL
109 mac_header -> ll header
112 Outgoing, dev->hard_header!=NULL
113 mac_header -> ll header
116 Incoming, dev->hard_header==NULL
117 mac_header -> UNKNOWN position. It is very likely, that it points to ll
118 header. PPP makes it, that is wrong, because introduce
119 assymetry between rx and tx paths.
122 Outgoing, dev->hard_header==NULL
123 mac_header -> data. ll header is still not built!
127 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
133 dev->hard_header != NULL
134 mac_header -> ll header
137 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141 We should set nh.raw on output to correct posistion,
142 packet classifier depends on it.
145 /* Private packet socket structures. */
147 /* identical to struct packet_mreq except it has
148 * a longer address field.
150 struct packet_mreq_max {
152 unsigned short mr_type;
153 unsigned short mr_alen;
154 unsigned char mr_address[MAX_ADDR_LEN];
158 struct tpacket_hdr *h1;
159 struct tpacket2_hdr *h2;
160 struct tpacket3_hdr *h3;
164 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
165 int closing, int tx_ring);
167 #define V3_ALIGNMENT (8)
169 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
171 #define BLK_PLUS_PRIV(sz_of_priv) \
172 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
175 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
176 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
177 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
178 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
180 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
184 struct packet_type *pt, struct net_device *orig_dev);
186 static void *packet_previous_frame(struct packet_sock *po,
187 struct packet_ring_buffer *rb,
189 static void packet_increment_head(struct packet_ring_buffer *buff);
190 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
192 struct packet_sock *);
193 static void prb_retire_current_block(struct tpacket_kbdq_core *,
194 struct packet_sock *, unsigned int status);
195 static int prb_queue_frozen(struct tpacket_kbdq_core *);
196 static void prb_open_block(struct tpacket_kbdq_core *,
197 struct tpacket_block_desc *);
198 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
200 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
201 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
202 struct tpacket3_hdr *);
203 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
204 struct tpacket3_hdr *);
205 static void packet_flush_mclist(struct sock *sk);
206 static u16 packet_pick_tx_queue(struct sk_buff *skb);
208 struct packet_skb_cb {
210 struct sockaddr_pkt pkt;
212 /* Trick: alias skb original length with
213 * ll.sll_family and ll.protocol in order
216 unsigned int origlen;
217 struct sockaddr_ll ll;
222 #define vio_le() virtio_legacy_is_little_endian()
224 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
226 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
227 #define GET_PBLOCK_DESC(x, bid) \
228 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
229 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
231 #define GET_NEXT_PRB_BLK_NUM(x) \
232 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
233 ((x)->kactive_blk_num+1) : 0)
235 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
236 static void __fanout_link(struct sock *sk, struct packet_sock *po);
238 static int packet_direct_xmit(struct sk_buff *skb)
240 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
243 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
245 struct net_device *dev;
248 dev = rcu_dereference(po->cached_dev);
256 static void packet_cached_dev_assign(struct packet_sock *po,
257 struct net_device *dev)
259 rcu_assign_pointer(po->cached_dev, dev);
262 static void packet_cached_dev_reset(struct packet_sock *po)
264 RCU_INIT_POINTER(po->cached_dev, NULL);
267 static bool packet_use_direct_xmit(const struct packet_sock *po)
269 return po->xmit == packet_direct_xmit;
272 static u16 packet_pick_tx_queue(struct sk_buff *skb)
274 struct net_device *dev = skb->dev;
275 const struct net_device_ops *ops = dev->netdev_ops;
276 int cpu = raw_smp_processor_id();
280 skb->sender_cpu = cpu + 1;
282 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
283 if (ops->ndo_select_queue) {
284 queue_index = ops->ndo_select_queue(dev, skb, NULL);
285 queue_index = netdev_cap_txqueue(dev, queue_index);
287 queue_index = netdev_pick_tx(dev, skb, NULL);
293 /* __register_prot_hook must be invoked through register_prot_hook
294 * or from a context in which asynchronous accesses to the packet
295 * socket is not possible (packet_create()).
297 static void __register_prot_hook(struct sock *sk)
299 struct packet_sock *po = pkt_sk(sk);
303 __fanout_link(sk, po);
305 dev_add_pack(&po->prot_hook);
312 static void register_prot_hook(struct sock *sk)
314 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
315 __register_prot_hook(sk);
318 /* If the sync parameter is true, we will temporarily drop
319 * the po->bind_lock and do a synchronize_net to make sure no
320 * asynchronous packet processing paths still refer to the elements
321 * of po->prot_hook. If the sync parameter is false, it is the
322 * callers responsibility to take care of this.
324 static void __unregister_prot_hook(struct sock *sk, bool sync)
326 struct packet_sock *po = pkt_sk(sk);
328 lockdep_assert_held_once(&po->bind_lock);
333 __fanout_unlink(sk, po);
335 __dev_remove_pack(&po->prot_hook);
340 spin_unlock(&po->bind_lock);
342 spin_lock(&po->bind_lock);
346 static void unregister_prot_hook(struct sock *sk, bool sync)
348 struct packet_sock *po = pkt_sk(sk);
351 __unregister_prot_hook(sk, sync);
354 static inline struct page * __pure pgv_to_page(void *addr)
356 if (is_vmalloc_addr(addr))
357 return vmalloc_to_page(addr);
358 return virt_to_page(addr);
361 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
363 union tpacket_uhdr h;
366 switch (po->tp_version) {
368 h.h1->tp_status = status;
369 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
372 h.h2->tp_status = status;
373 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
376 h.h3->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
380 WARN(1, "TPACKET version not supported.\n");
387 static int __packet_get_status(const struct packet_sock *po, void *frame)
389 union tpacket_uhdr h;
394 switch (po->tp_version) {
396 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
397 return h.h1->tp_status;
399 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
400 return h.h2->tp_status;
402 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
403 return h.h3->tp_status;
405 WARN(1, "TPACKET version not supported.\n");
411 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
414 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
417 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
418 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
419 return TP_STATUS_TS_RAW_HARDWARE;
421 if (ktime_to_timespec_cond(skb->tstamp, ts))
422 return TP_STATUS_TS_SOFTWARE;
427 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
430 union tpacket_uhdr h;
434 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
438 switch (po->tp_version) {
440 h.h1->tp_sec = ts.tv_sec;
441 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
444 h.h2->tp_sec = ts.tv_sec;
445 h.h2->tp_nsec = ts.tv_nsec;
448 h.h3->tp_sec = ts.tv_sec;
449 h.h3->tp_nsec = ts.tv_nsec;
452 WARN(1, "TPACKET version not supported.\n");
456 /* one flush is safe, as both fields always lie on the same cacheline */
457 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
463 static void *packet_lookup_frame(const struct packet_sock *po,
464 const struct packet_ring_buffer *rb,
465 unsigned int position,
468 unsigned int pg_vec_pos, frame_offset;
469 union tpacket_uhdr h;
471 pg_vec_pos = position / rb->frames_per_block;
472 frame_offset = position % rb->frames_per_block;
474 h.raw = rb->pg_vec[pg_vec_pos].buffer +
475 (frame_offset * rb->frame_size);
477 if (status != __packet_get_status(po, h.raw))
483 static void *packet_current_frame(struct packet_sock *po,
484 struct packet_ring_buffer *rb,
487 return packet_lookup_frame(po, rb, rb->head, status);
490 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
492 del_timer_sync(&pkc->retire_blk_timer);
495 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
496 struct sk_buff_head *rb_queue)
498 struct tpacket_kbdq_core *pkc;
500 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
502 spin_lock_bh(&rb_queue->lock);
503 pkc->delete_blk_timer = 1;
504 spin_unlock_bh(&rb_queue->lock);
506 prb_del_retire_blk_timer(pkc);
509 static void prb_setup_retire_blk_timer(struct packet_sock *po)
511 struct tpacket_kbdq_core *pkc;
513 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
514 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
516 pkc->retire_blk_timer.expires = jiffies;
519 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
520 int blk_size_in_bytes)
522 struct net_device *dev;
523 unsigned int mbits, div;
524 struct ethtool_link_ksettings ecmd;
528 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
529 if (unlikely(!dev)) {
531 return DEFAULT_PRB_RETIRE_TOV;
533 err = __ethtool_get_link_ksettings(dev, &ecmd);
536 return DEFAULT_PRB_RETIRE_TOV;
538 /* If the link speed is so slow you don't really
539 * need to worry about perf anyways
541 if (ecmd.base.speed < SPEED_1000 ||
542 ecmd.base.speed == SPEED_UNKNOWN)
543 return DEFAULT_PRB_RETIRE_TOV;
545 div = ecmd.base.speed / 1000;
546 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
556 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
557 union tpacket_req_u *req_u)
559 p1->feature_req_word = req_u->req3.tp_feature_req_word;
562 static void init_prb_bdqc(struct packet_sock *po,
563 struct packet_ring_buffer *rb,
565 union tpacket_req_u *req_u)
567 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
568 struct tpacket_block_desc *pbd;
570 memset(p1, 0x0, sizeof(*p1));
572 p1->knxt_seq_num = 1;
574 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
575 p1->pkblk_start = pg_vec[0].buffer;
576 p1->kblk_size = req_u->req3.tp_block_size;
577 p1->knum_blocks = req_u->req3.tp_block_nr;
578 p1->hdrlen = po->tp_hdrlen;
579 p1->version = po->tp_version;
580 p1->last_kactive_blk_num = 0;
581 po->stats.stats3.tp_freeze_q_cnt = 0;
582 if (req_u->req3.tp_retire_blk_tov)
583 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
585 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
586 req_u->req3.tp_block_size);
587 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
588 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
590 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
591 prb_init_ft_ops(p1, req_u);
592 prb_setup_retire_blk_timer(po);
593 prb_open_block(p1, pbd);
596 /* Do NOT update the last_blk_num first.
597 * Assumes sk_buff_head lock is held.
599 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
601 mod_timer(&pkc->retire_blk_timer,
602 jiffies + pkc->tov_in_jiffies);
603 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
608 * 1) We refresh the timer only when we open a block.
609 * By doing this we don't waste cycles refreshing the timer
610 * on packet-by-packet basis.
612 * With a 1MB block-size, on a 1Gbps line, it will take
613 * i) ~8 ms to fill a block + ii) memcpy etc.
614 * In this cut we are not accounting for the memcpy time.
616 * So, if the user sets the 'tmo' to 10ms then the timer
617 * will never fire while the block is still getting filled
618 * (which is what we want). However, the user could choose
619 * to close a block early and that's fine.
621 * But when the timer does fire, we check whether or not to refresh it.
622 * Since the tmo granularity is in msecs, it is not too expensive
623 * to refresh the timer, lets say every '8' msecs.
624 * Either the user can set the 'tmo' or we can derive it based on
625 * a) line-speed and b) block-size.
626 * prb_calc_retire_blk_tmo() calculates the tmo.
629 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
631 struct packet_sock *po =
632 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
633 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
635 struct tpacket_block_desc *pbd;
637 spin_lock(&po->sk.sk_receive_queue.lock);
639 frozen = prb_queue_frozen(pkc);
640 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
642 if (unlikely(pkc->delete_blk_timer))
645 /* We only need to plug the race when the block is partially filled.
647 * lock(); increment BLOCK_NUM_PKTS; unlock()
648 * copy_bits() is in progress ...
649 * timer fires on other cpu:
650 * we can't retire the current block because copy_bits
654 if (BLOCK_NUM_PKTS(pbd)) {
655 while (atomic_read(&pkc->blk_fill_in_prog)) {
656 /* Waiting for skb_copy_bits to finish... */
661 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
663 if (!BLOCK_NUM_PKTS(pbd)) {
664 /* An empty block. Just refresh the timer. */
667 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
668 if (!prb_dispatch_next_block(pkc, po))
673 /* Case 1. Queue was frozen because user-space was
676 if (prb_curr_blk_in_use(pbd)) {
678 * Ok, user-space is still behind.
679 * So just refresh the timer.
683 /* Case 2. queue was frozen,user-space caught up,
684 * now the link went idle && the timer fired.
685 * We don't have a block to close.So we open this
686 * block and restart the timer.
687 * opening a block thaws the queue,restarts timer
688 * Thawing/timer-refresh is a side effect.
690 prb_open_block(pkc, pbd);
697 _prb_refresh_rx_retire_blk_timer(pkc);
700 spin_unlock(&po->sk.sk_receive_queue.lock);
703 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
704 struct tpacket_block_desc *pbd1, __u32 status)
706 /* Flush everything minus the block header */
708 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
713 /* Skip the block header(we know header WILL fit in 4K) */
716 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
717 for (; start < end; start += PAGE_SIZE)
718 flush_dcache_page(pgv_to_page(start));
723 /* Now update the block status. */
725 BLOCK_STATUS(pbd1) = status;
727 /* Flush the block header */
729 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
731 flush_dcache_page(pgv_to_page(start));
741 * 2) Increment active_blk_num
743 * Note:We DONT refresh the timer on purpose.
744 * Because almost always the next block will be opened.
746 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
747 struct tpacket_block_desc *pbd1,
748 struct packet_sock *po, unsigned int stat)
750 __u32 status = TP_STATUS_USER | stat;
752 struct tpacket3_hdr *last_pkt;
753 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
754 struct sock *sk = &po->sk;
756 if (atomic_read(&po->tp_drops))
757 status |= TP_STATUS_LOSING;
759 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
760 last_pkt->tp_next_offset = 0;
762 /* Get the ts of the last pkt */
763 if (BLOCK_NUM_PKTS(pbd1)) {
764 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
765 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
767 /* Ok, we tmo'd - so get the current time.
769 * It shouldn't really happen as we don't close empty
770 * blocks. See prb_retire_rx_blk_timer_expired().
774 h1->ts_last_pkt.ts_sec = ts.tv_sec;
775 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
780 /* Flush the block */
781 prb_flush_block(pkc1, pbd1, status);
783 sk->sk_data_ready(sk);
785 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
788 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
790 pkc->reset_pending_on_curr_blk = 0;
794 * Side effect of opening a block:
796 * 1) prb_queue is thawed.
797 * 2) retire_blk_timer is refreshed.
800 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
801 struct tpacket_block_desc *pbd1)
804 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
808 /* We could have just memset this but we will lose the
809 * flexibility of making the priv area sticky
812 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
813 BLOCK_NUM_PKTS(pbd1) = 0;
814 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
818 h1->ts_first_pkt.ts_sec = ts.tv_sec;
819 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
821 pkc1->pkblk_start = (char *)pbd1;
822 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
824 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
825 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
827 pbd1->version = pkc1->version;
828 pkc1->prev = pkc1->nxt_offset;
829 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
831 prb_thaw_queue(pkc1);
832 _prb_refresh_rx_retire_blk_timer(pkc1);
838 * Queue freeze logic:
839 * 1) Assume tp_block_nr = 8 blocks.
840 * 2) At time 't0', user opens Rx ring.
841 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
842 * 4) user-space is either sleeping or processing block '0'.
843 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
844 * it will close block-7,loop around and try to fill block '0'.
846 * __packet_lookup_frame_in_block
847 * prb_retire_current_block()
848 * prb_dispatch_next_block()
849 * |->(BLOCK_STATUS == USER) evaluates to true
850 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
851 * 6) Now there are two cases:
852 * 6.1) Link goes idle right after the queue is frozen.
853 * But remember, the last open_block() refreshed the timer.
854 * When this timer expires,it will refresh itself so that we can
855 * re-open block-0 in near future.
856 * 6.2) Link is busy and keeps on receiving packets. This is a simple
857 * case and __packet_lookup_frame_in_block will check if block-0
858 * is free and can now be re-used.
860 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
861 struct packet_sock *po)
863 pkc->reset_pending_on_curr_blk = 1;
864 po->stats.stats3.tp_freeze_q_cnt++;
867 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
870 * If the next block is free then we will dispatch it
871 * and return a good offset.
872 * Else, we will freeze the queue.
873 * So, caller must check the return value.
875 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
876 struct packet_sock *po)
878 struct tpacket_block_desc *pbd;
882 /* 1. Get current block num */
883 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
885 /* 2. If this block is currently in_use then freeze the queue */
886 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
887 prb_freeze_queue(pkc, po);
893 * open this block and return the offset where the first packet
894 * needs to get stored.
896 prb_open_block(pkc, pbd);
897 return (void *)pkc->nxt_offset;
900 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
901 struct packet_sock *po, unsigned int status)
903 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
905 /* retire/close the current block */
906 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
908 * Plug the case where copy_bits() is in progress on
909 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
910 * have space to copy the pkt in the current block and
911 * called prb_retire_current_block()
913 * We don't need to worry about the TMO case because
914 * the timer-handler already handled this case.
916 if (!(status & TP_STATUS_BLK_TMO)) {
917 while (atomic_read(&pkc->blk_fill_in_prog)) {
918 /* Waiting for skb_copy_bits to finish... */
922 prb_close_block(pkc, pbd, po, status);
927 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
929 return TP_STATUS_USER & BLOCK_STATUS(pbd);
932 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
934 return pkc->reset_pending_on_curr_blk;
937 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
939 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
940 atomic_dec(&pkc->blk_fill_in_prog);
943 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
944 struct tpacket3_hdr *ppd)
946 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
949 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
950 struct tpacket3_hdr *ppd)
952 ppd->hv1.tp_rxhash = 0;
955 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
956 struct tpacket3_hdr *ppd)
958 if (skb_vlan_tag_present(pkc->skb)) {
959 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
960 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
961 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
963 ppd->hv1.tp_vlan_tci = 0;
964 ppd->hv1.tp_vlan_tpid = 0;
965 ppd->tp_status = TP_STATUS_AVAILABLE;
969 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
970 struct tpacket3_hdr *ppd)
972 ppd->hv1.tp_padding = 0;
973 prb_fill_vlan_info(pkc, ppd);
975 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
976 prb_fill_rxhash(pkc, ppd);
978 prb_clear_rxhash(pkc, ppd);
981 static void prb_fill_curr_block(char *curr,
982 struct tpacket_kbdq_core *pkc,
983 struct tpacket_block_desc *pbd,
986 struct tpacket3_hdr *ppd;
988 ppd = (struct tpacket3_hdr *)curr;
989 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
991 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
992 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
993 BLOCK_NUM_PKTS(pbd) += 1;
994 atomic_inc(&pkc->blk_fill_in_prog);
995 prb_run_all_ft_ops(pkc, ppd);
998 /* Assumes caller has the sk->rx_queue.lock */
999 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1000 struct sk_buff *skb,
1004 struct tpacket_kbdq_core *pkc;
1005 struct tpacket_block_desc *pbd;
1008 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1009 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1011 /* Queue is frozen when user space is lagging behind */
1012 if (prb_queue_frozen(pkc)) {
1014 * Check if that last block which caused the queue to freeze,
1015 * is still in_use by user-space.
1017 if (prb_curr_blk_in_use(pbd)) {
1018 /* Can't record this packet */
1022 * Ok, the block was released by user-space.
1023 * Now let's open that block.
1024 * opening a block also thaws the queue.
1025 * Thawing is a side effect.
1027 prb_open_block(pkc, pbd);
1032 curr = pkc->nxt_offset;
1034 end = (char *)pbd + pkc->kblk_size;
1036 /* first try the current block */
1037 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1038 prb_fill_curr_block(curr, pkc, pbd, len);
1039 return (void *)curr;
1042 /* Ok, close the current block */
1043 prb_retire_current_block(pkc, po, 0);
1045 /* Now, try to dispatch the next block */
1046 curr = (char *)prb_dispatch_next_block(pkc, po);
1048 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1049 prb_fill_curr_block(curr, pkc, pbd, len);
1050 return (void *)curr;
1054 * No free blocks are available.user_space hasn't caught up yet.
1055 * Queue was just frozen and now this packet will get dropped.
1060 static void *packet_current_rx_frame(struct packet_sock *po,
1061 struct sk_buff *skb,
1062 int status, unsigned int len)
1065 switch (po->tp_version) {
1068 curr = packet_lookup_frame(po, &po->rx_ring,
1069 po->rx_ring.head, status);
1072 return __packet_lookup_frame_in_block(po, skb, len);
1074 WARN(1, "TPACKET version not supported\n");
1080 static void *prb_lookup_block(const struct packet_sock *po,
1081 const struct packet_ring_buffer *rb,
1085 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1086 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1088 if (status != BLOCK_STATUS(pbd))
1093 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1096 if (rb->prb_bdqc.kactive_blk_num)
1097 prev = rb->prb_bdqc.kactive_blk_num-1;
1099 prev = rb->prb_bdqc.knum_blocks-1;
1103 /* Assumes caller has held the rx_queue.lock */
1104 static void *__prb_previous_block(struct packet_sock *po,
1105 struct packet_ring_buffer *rb,
1108 unsigned int previous = prb_previous_blk_num(rb);
1109 return prb_lookup_block(po, rb, previous, status);
1112 static void *packet_previous_rx_frame(struct packet_sock *po,
1113 struct packet_ring_buffer *rb,
1116 if (po->tp_version <= TPACKET_V2)
1117 return packet_previous_frame(po, rb, status);
1119 return __prb_previous_block(po, rb, status);
1122 static void packet_increment_rx_head(struct packet_sock *po,
1123 struct packet_ring_buffer *rb)
1125 switch (po->tp_version) {
1128 return packet_increment_head(rb);
1131 WARN(1, "TPACKET version not supported.\n");
1137 static void *packet_previous_frame(struct packet_sock *po,
1138 struct packet_ring_buffer *rb,
1141 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1142 return packet_lookup_frame(po, rb, previous, status);
1145 static void packet_increment_head(struct packet_ring_buffer *buff)
1147 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1150 static void packet_inc_pending(struct packet_ring_buffer *rb)
1152 this_cpu_inc(*rb->pending_refcnt);
1155 static void packet_dec_pending(struct packet_ring_buffer *rb)
1157 this_cpu_dec(*rb->pending_refcnt);
1160 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1162 unsigned int refcnt = 0;
1165 /* We don't use pending refcount in rx_ring. */
1166 if (rb->pending_refcnt == NULL)
1169 for_each_possible_cpu(cpu)
1170 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1175 static int packet_alloc_pending(struct packet_sock *po)
1177 po->rx_ring.pending_refcnt = NULL;
1179 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1180 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1186 static void packet_free_pending(struct packet_sock *po)
1188 free_percpu(po->tx_ring.pending_refcnt);
1191 #define ROOM_POW_OFF 2
1192 #define ROOM_NONE 0x0
1193 #define ROOM_LOW 0x1
1194 #define ROOM_NORMAL 0x2
1196 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1200 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1201 idx = READ_ONCE(po->rx_ring.head);
1203 idx += len >> pow_off;
1206 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1209 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1213 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1214 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1216 idx += len >> pow_off;
1219 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1222 static int __packet_rcv_has_room(const struct packet_sock *po,
1223 const struct sk_buff *skb)
1225 const struct sock *sk = &po->sk;
1226 int ret = ROOM_NONE;
1228 if (po->prot_hook.func != tpacket_rcv) {
1229 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1230 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1231 - (skb ? skb->truesize : 0);
1233 if (avail > (rcvbuf >> ROOM_POW_OFF))
1241 if (po->tp_version == TPACKET_V3) {
1242 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1244 else if (__tpacket_v3_has_room(po, 0))
1247 if (__tpacket_has_room(po, ROOM_POW_OFF))
1249 else if (__tpacket_has_room(po, 0))
1256 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1260 ret = __packet_rcv_has_room(po, skb);
1261 pressure = ret != ROOM_NORMAL;
1263 if (READ_ONCE(po->pressure) != pressure)
1264 WRITE_ONCE(po->pressure, pressure);
1269 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1271 if (READ_ONCE(po->pressure) &&
1272 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1273 WRITE_ONCE(po->pressure, 0);
1276 static void packet_sock_destruct(struct sock *sk)
1278 skb_queue_purge(&sk->sk_error_queue);
1280 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1281 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1283 if (!sock_flag(sk, SOCK_DEAD)) {
1284 pr_err("Attempt to release alive packet socket: %p\n", sk);
1288 sk_refcnt_debug_dec(sk);
1291 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1293 u32 *history = po->rollover->history;
1297 rxhash = skb_get_hash(skb);
1298 for (i = 0; i < ROLLOVER_HLEN; i++)
1299 if (READ_ONCE(history[i]) == rxhash)
1302 victim = prandom_u32() % ROLLOVER_HLEN;
1304 /* Avoid dirtying the cache line if possible */
1305 if (READ_ONCE(history[victim]) != rxhash)
1306 WRITE_ONCE(history[victim], rxhash);
1308 return count > (ROLLOVER_HLEN >> 1);
1311 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1312 struct sk_buff *skb,
1315 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1318 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1319 struct sk_buff *skb,
1322 unsigned int val = atomic_inc_return(&f->rr_cur);
1327 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1328 struct sk_buff *skb,
1331 return smp_processor_id() % num;
1334 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1335 struct sk_buff *skb,
1338 return prandom_u32_max(num);
1341 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1342 struct sk_buff *skb,
1343 unsigned int idx, bool try_self,
1346 struct packet_sock *po, *po_next, *po_skip = NULL;
1347 unsigned int i, j, room = ROOM_NONE;
1349 po = pkt_sk(f->arr[idx]);
1352 room = packet_rcv_has_room(po, skb);
1353 if (room == ROOM_NORMAL ||
1354 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1359 i = j = min_t(int, po->rollover->sock, num - 1);
1361 po_next = pkt_sk(f->arr[i]);
1362 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1363 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1365 po->rollover->sock = i;
1366 atomic_long_inc(&po->rollover->num);
1367 if (room == ROOM_LOW)
1368 atomic_long_inc(&po->rollover->num_huge);
1376 atomic_long_inc(&po->rollover->num_failed);
1380 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1381 struct sk_buff *skb,
1384 return skb_get_queue_mapping(skb) % num;
1387 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1388 struct sk_buff *skb,
1391 struct bpf_prog *prog;
1392 unsigned int ret = 0;
1395 prog = rcu_dereference(f->bpf_prog);
1397 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1403 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1405 return f->flags & (flag >> 8);
1408 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1409 struct packet_type *pt, struct net_device *orig_dev)
1411 struct packet_fanout *f = pt->af_packet_priv;
1412 unsigned int num = READ_ONCE(f->num_members);
1413 struct net *net = read_pnet(&f->net);
1414 struct packet_sock *po;
1417 if (!net_eq(dev_net(dev), net) || !num) {
1422 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1423 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1428 case PACKET_FANOUT_HASH:
1430 idx = fanout_demux_hash(f, skb, num);
1432 case PACKET_FANOUT_LB:
1433 idx = fanout_demux_lb(f, skb, num);
1435 case PACKET_FANOUT_CPU:
1436 idx = fanout_demux_cpu(f, skb, num);
1438 case PACKET_FANOUT_RND:
1439 idx = fanout_demux_rnd(f, skb, num);
1441 case PACKET_FANOUT_QM:
1442 idx = fanout_demux_qm(f, skb, num);
1444 case PACKET_FANOUT_ROLLOVER:
1445 idx = fanout_demux_rollover(f, skb, 0, false, num);
1447 case PACKET_FANOUT_CBPF:
1448 case PACKET_FANOUT_EBPF:
1449 idx = fanout_demux_bpf(f, skb, num);
1453 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1454 idx = fanout_demux_rollover(f, skb, idx, true, num);
1456 po = pkt_sk(f->arr[idx]);
1457 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1460 DEFINE_MUTEX(fanout_mutex);
1461 EXPORT_SYMBOL_GPL(fanout_mutex);
1462 static LIST_HEAD(fanout_list);
1463 static u16 fanout_next_id;
1465 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1467 struct packet_fanout *f = po->fanout;
1469 spin_lock(&f->lock);
1470 f->arr[f->num_members] = sk;
1473 if (f->num_members == 1)
1474 dev_add_pack(&f->prot_hook);
1475 spin_unlock(&f->lock);
1478 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1480 struct packet_fanout *f = po->fanout;
1483 spin_lock(&f->lock);
1484 for (i = 0; i < f->num_members; i++) {
1485 if (f->arr[i] == sk)
1488 BUG_ON(i >= f->num_members);
1489 f->arr[i] = f->arr[f->num_members - 1];
1491 if (f->num_members == 0)
1492 __dev_remove_pack(&f->prot_hook);
1493 spin_unlock(&f->lock);
1496 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1498 if (sk->sk_family != PF_PACKET)
1501 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1504 static void fanout_init_data(struct packet_fanout *f)
1507 case PACKET_FANOUT_LB:
1508 atomic_set(&f->rr_cur, 0);
1510 case PACKET_FANOUT_CBPF:
1511 case PACKET_FANOUT_EBPF:
1512 RCU_INIT_POINTER(f->bpf_prog, NULL);
1517 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1519 struct bpf_prog *old;
1521 spin_lock(&f->lock);
1522 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1523 rcu_assign_pointer(f->bpf_prog, new);
1524 spin_unlock(&f->lock);
1528 bpf_prog_destroy(old);
1532 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1535 struct bpf_prog *new;
1536 struct sock_fprog fprog;
1539 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1541 if (len != sizeof(fprog))
1543 if (copy_from_user(&fprog, data, len))
1546 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1550 __fanout_set_data_bpf(po->fanout, new);
1554 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1557 struct bpf_prog *new;
1560 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1562 if (len != sizeof(fd))
1564 if (copy_from_user(&fd, data, len))
1567 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1569 return PTR_ERR(new);
1571 __fanout_set_data_bpf(po->fanout, new);
1575 static int fanout_set_data(struct packet_sock *po, char __user *data,
1578 switch (po->fanout->type) {
1579 case PACKET_FANOUT_CBPF:
1580 return fanout_set_data_cbpf(po, data, len);
1581 case PACKET_FANOUT_EBPF:
1582 return fanout_set_data_ebpf(po, data, len);
1588 static void fanout_release_data(struct packet_fanout *f)
1591 case PACKET_FANOUT_CBPF:
1592 case PACKET_FANOUT_EBPF:
1593 __fanout_set_data_bpf(f, NULL);
1597 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1599 struct packet_fanout *f;
1601 list_for_each_entry(f, &fanout_list, list) {
1602 if (f->id == candidate_id &&
1603 read_pnet(&f->net) == sock_net(sk)) {
1610 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1612 u16 id = fanout_next_id;
1615 if (__fanout_id_is_free(sk, id)) {
1617 fanout_next_id = id + 1;
1622 } while (id != fanout_next_id);
1627 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1629 struct packet_rollover *rollover = NULL;
1630 struct packet_sock *po = pkt_sk(sk);
1631 struct packet_fanout *f, *match;
1632 u8 type = type_flags & 0xff;
1633 u8 flags = type_flags >> 8;
1637 case PACKET_FANOUT_ROLLOVER:
1638 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1640 case PACKET_FANOUT_HASH:
1641 case PACKET_FANOUT_LB:
1642 case PACKET_FANOUT_CPU:
1643 case PACKET_FANOUT_RND:
1644 case PACKET_FANOUT_QM:
1645 case PACKET_FANOUT_CBPF:
1646 case PACKET_FANOUT_EBPF:
1652 mutex_lock(&fanout_mutex);
1658 if (type == PACKET_FANOUT_ROLLOVER ||
1659 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1661 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1664 atomic_long_set(&rollover->num, 0);
1665 atomic_long_set(&rollover->num_huge, 0);
1666 atomic_long_set(&rollover->num_failed, 0);
1669 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1674 if (!fanout_find_new_id(sk, &id)) {
1678 /* ephemeral flag for the first socket in the group: drop it */
1679 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1683 list_for_each_entry(f, &fanout_list, list) {
1685 read_pnet(&f->net) == sock_net(sk)) {
1691 if (match && match->flags != flags)
1695 match = kzalloc(sizeof(*match), GFP_KERNEL);
1698 write_pnet(&match->net, sock_net(sk));
1701 match->flags = flags;
1702 INIT_LIST_HEAD(&match->list);
1703 spin_lock_init(&match->lock);
1704 refcount_set(&match->sk_ref, 0);
1705 fanout_init_data(match);
1706 match->prot_hook.type = po->prot_hook.type;
1707 match->prot_hook.dev = po->prot_hook.dev;
1708 match->prot_hook.func = packet_rcv_fanout;
1709 match->prot_hook.af_packet_priv = match;
1710 match->prot_hook.id_match = match_fanout_group;
1711 list_add(&match->list, &fanout_list);
1715 spin_lock(&po->bind_lock);
1717 match->type == type &&
1718 match->prot_hook.type == po->prot_hook.type &&
1719 match->prot_hook.dev == po->prot_hook.dev) {
1721 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1722 __dev_remove_pack(&po->prot_hook);
1724 po->rollover = rollover;
1726 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1727 __fanout_link(sk, po);
1731 spin_unlock(&po->bind_lock);
1733 if (err && !refcount_read(&match->sk_ref)) {
1734 list_del(&match->list);
1740 mutex_unlock(&fanout_mutex);
1744 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1745 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1746 * It is the responsibility of the caller to call fanout_release_data() and
1747 * free the returned packet_fanout (after synchronize_net())
1749 static struct packet_fanout *fanout_release(struct sock *sk)
1751 struct packet_sock *po = pkt_sk(sk);
1752 struct packet_fanout *f;
1754 mutex_lock(&fanout_mutex);
1759 if (refcount_dec_and_test(&f->sk_ref))
1764 mutex_unlock(&fanout_mutex);
1769 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1770 struct sk_buff *skb)
1772 /* Earlier code assumed this would be a VLAN pkt, double-check
1773 * this now that we have the actual packet in hand. We can only
1774 * do this check on Ethernet devices.
1776 if (unlikely(dev->type != ARPHRD_ETHER))
1779 skb_reset_mac_header(skb);
1780 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1783 static const struct proto_ops packet_ops;
1785 static const struct proto_ops packet_ops_spkt;
1787 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1788 struct packet_type *pt, struct net_device *orig_dev)
1791 struct sockaddr_pkt *spkt;
1794 * When we registered the protocol we saved the socket in the data
1795 * field for just this event.
1798 sk = pt->af_packet_priv;
1801 * Yank back the headers [hope the device set this
1802 * right or kerboom...]
1804 * Incoming packets have ll header pulled,
1807 * For outgoing ones skb->data == skb_mac_header(skb)
1808 * so that this procedure is noop.
1811 if (skb->pkt_type == PACKET_LOOPBACK)
1814 if (!net_eq(dev_net(dev), sock_net(sk)))
1817 skb = skb_share_check(skb, GFP_ATOMIC);
1821 /* drop any routing info */
1824 /* drop conntrack reference */
1827 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1829 skb_push(skb, skb->data - skb_mac_header(skb));
1832 * The SOCK_PACKET socket receives _all_ frames.
1835 spkt->spkt_family = dev->type;
1836 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1837 spkt->spkt_protocol = skb->protocol;
1840 * Charge the memory to the socket. This is done specifically
1841 * to prevent sockets using all the memory up.
1844 if (sock_queue_rcv_skb(sk, skb) == 0)
1853 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1855 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1856 sock->type == SOCK_RAW) {
1857 skb_reset_mac_header(skb);
1858 skb->protocol = dev_parse_header_protocol(skb);
1861 skb_probe_transport_header(skb);
1865 * Output a raw packet to a device layer. This bypasses all the other
1866 * protocol layers and you must therefore supply it with a complete frame
1869 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1872 struct sock *sk = sock->sk;
1873 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1874 struct sk_buff *skb = NULL;
1875 struct net_device *dev;
1876 struct sockcm_cookie sockc;
1882 * Get and verify the address.
1886 if (msg->msg_namelen < sizeof(struct sockaddr))
1888 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1889 proto = saddr->spkt_protocol;
1891 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1894 * Find the device first to size check it
1897 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1900 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1906 if (!(dev->flags & IFF_UP))
1910 * You may not queue a frame bigger than the mtu. This is the lowest level
1911 * raw protocol and you must do your own fragmentation at this level.
1914 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1915 if (!netif_supports_nofcs(dev)) {
1916 err = -EPROTONOSUPPORT;
1919 extra_len = 4; /* We're doing our own CRC */
1923 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1927 size_t reserved = LL_RESERVED_SPACE(dev);
1928 int tlen = dev->needed_tailroom;
1929 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1932 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1935 /* FIXME: Save some space for broken drivers that write a hard
1936 * header at transmission time by themselves. PPP is the notable
1937 * one here. This should really be fixed at the driver level.
1939 skb_reserve(skb, reserved);
1940 skb_reset_network_header(skb);
1942 /* Try to align data part correctly */
1947 skb_reset_network_header(skb);
1949 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1955 if (!dev_validate_header(dev, skb->data, len)) {
1959 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1960 !packet_extra_vlan_len_allowed(dev, skb)) {
1965 sockcm_init(&sockc, sk);
1966 if (msg->msg_controllen) {
1967 err = sock_cmsg_send(sk, msg, &sockc);
1972 skb->protocol = proto;
1974 skb->priority = sk->sk_priority;
1975 skb->mark = sk->sk_mark;
1976 skb->tstamp = sockc.transmit_time;
1978 skb_setup_tx_timestamp(skb, sockc.tsflags);
1980 if (unlikely(extra_len == 4))
1983 packet_parse_headers(skb, sock);
1985 dev_queue_xmit(skb);
1996 static unsigned int run_filter(struct sk_buff *skb,
1997 const struct sock *sk,
2000 struct sk_filter *filter;
2003 filter = rcu_dereference(sk->sk_filter);
2005 res = bpf_prog_run_clear_cb(filter->prog, skb);
2011 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2014 struct virtio_net_hdr vnet_hdr;
2016 if (*len < sizeof(vnet_hdr))
2018 *len -= sizeof(vnet_hdr);
2020 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2023 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2027 * This function makes lazy skb cloning in hope that most of packets
2028 * are discarded by BPF.
2030 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2031 * and skb->cb are mangled. It works because (and until) packets
2032 * falling here are owned by current CPU. Output packets are cloned
2033 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2034 * sequencially, so that if we return skb to original state on exit,
2035 * we will not harm anyone.
2038 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2039 struct packet_type *pt, struct net_device *orig_dev)
2042 struct sockaddr_ll *sll;
2043 struct packet_sock *po;
2044 u8 *skb_head = skb->data;
2045 int skb_len = skb->len;
2046 unsigned int snaplen, res;
2047 bool is_drop_n_account = false;
2049 if (skb->pkt_type == PACKET_LOOPBACK)
2052 sk = pt->af_packet_priv;
2055 if (!net_eq(dev_net(dev), sock_net(sk)))
2060 if (dev->header_ops) {
2061 /* The device has an explicit notion of ll header,
2062 * exported to higher levels.
2064 * Otherwise, the device hides details of its frame
2065 * structure, so that corresponding packet head is
2066 * never delivered to user.
2068 if (sk->sk_type != SOCK_DGRAM)
2069 skb_push(skb, skb->data - skb_mac_header(skb));
2070 else if (skb->pkt_type == PACKET_OUTGOING) {
2071 /* Special case: outgoing packets have ll header at head */
2072 skb_pull(skb, skb_network_offset(skb));
2078 res = run_filter(skb, sk, snaplen);
2080 goto drop_n_restore;
2084 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2087 if (skb_shared(skb)) {
2088 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2092 if (skb_head != skb->data) {
2093 skb->data = skb_head;
2100 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2102 sll = &PACKET_SKB_CB(skb)->sa.ll;
2103 sll->sll_hatype = dev->type;
2104 sll->sll_pkttype = skb->pkt_type;
2105 if (unlikely(po->origdev))
2106 sll->sll_ifindex = orig_dev->ifindex;
2108 sll->sll_ifindex = dev->ifindex;
2110 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2112 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2113 * Use their space for storing the original skb length.
2115 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2117 if (pskb_trim(skb, snaplen))
2120 skb_set_owner_r(skb, sk);
2124 /* drop conntrack reference */
2127 spin_lock(&sk->sk_receive_queue.lock);
2128 po->stats.stats1.tp_packets++;
2129 sock_skb_set_dropcount(sk, skb);
2130 __skb_queue_tail(&sk->sk_receive_queue, skb);
2131 spin_unlock(&sk->sk_receive_queue.lock);
2132 sk->sk_data_ready(sk);
2136 is_drop_n_account = true;
2137 atomic_inc(&po->tp_drops);
2138 atomic_inc(&sk->sk_drops);
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, netoff, hdrlen;
2165 struct sk_buff *copy_skb = NULL;
2168 bool is_drop_n_account = false;
2169 bool do_vnet = false;
2171 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2172 * We may add members to them until current aligned size without forcing
2173 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2175 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2176 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2178 if (skb->pkt_type == PACKET_LOOPBACK)
2181 sk = pt->af_packet_priv;
2184 if (!net_eq(dev_net(dev), sock_net(sk)))
2187 if (dev->header_ops) {
2188 if (sk->sk_type != SOCK_DGRAM)
2189 skb_push(skb, skb->data - skb_mac_header(skb));
2190 else if (skb->pkt_type == PACKET_OUTGOING) {
2191 /* Special case: outgoing packets have ll header at head */
2192 skb_pull(skb, skb_network_offset(skb));
2198 res = run_filter(skb, sk, snaplen);
2200 goto drop_n_restore;
2202 /* If we are flooded, just give up */
2203 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2204 atomic_inc(&po->tp_drops);
2205 goto drop_n_restore;
2208 if (skb->ip_summed == CHECKSUM_PARTIAL)
2209 status |= TP_STATUS_CSUMNOTREADY;
2210 else if (skb->pkt_type != PACKET_OUTGOING &&
2211 (skb->ip_summed == CHECKSUM_COMPLETE ||
2212 skb_csum_unnecessary(skb)))
2213 status |= TP_STATUS_CSUM_VALID;
2218 if (sk->sk_type == SOCK_DGRAM) {
2219 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2222 unsigned int maclen = skb_network_offset(skb);
2223 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2224 (maclen < 16 ? 16 : maclen)) +
2226 if (po->has_vnet_hdr) {
2227 netoff += sizeof(struct virtio_net_hdr);
2230 macoff = netoff - maclen;
2232 if (po->tp_version <= TPACKET_V2) {
2233 if (macoff + snaplen > po->rx_ring.frame_size) {
2234 if (po->copy_thresh &&
2235 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2236 if (skb_shared(skb)) {
2237 copy_skb = skb_clone(skb, GFP_ATOMIC);
2239 copy_skb = skb_get(skb);
2240 skb_head = skb->data;
2243 skb_set_owner_r(copy_skb, sk);
2245 snaplen = po->rx_ring.frame_size - macoff;
2246 if ((int)snaplen < 0) {
2251 } else if (unlikely(macoff + snaplen >
2252 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2255 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2256 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2257 snaplen, nval, macoff);
2259 if (unlikely((int)snaplen < 0)) {
2261 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2265 spin_lock(&sk->sk_receive_queue.lock);
2266 h.raw = packet_current_rx_frame(po, skb,
2267 TP_STATUS_KERNEL, (macoff+snaplen));
2269 goto drop_n_account;
2270 if (po->tp_version <= TPACKET_V2) {
2271 packet_increment_rx_head(po, &po->rx_ring);
2273 * LOSING will be reported till you read the stats,
2274 * because it's COR - Clear On Read.
2275 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2278 if (atomic_read(&po->tp_drops))
2279 status |= TP_STATUS_LOSING;
2283 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2284 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 spin_unlock(&sk->sk_receive_queue.lock);
2395 atomic_inc(&po->tp_drops);
2396 is_drop_n_account = true;
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_zcopy_get_nouarg(skb);
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 int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2426 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2427 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2428 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2429 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2430 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2431 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2432 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2434 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2440 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2441 struct virtio_net_hdr *vnet_hdr)
2443 if (*len < sizeof(*vnet_hdr))
2445 *len -= sizeof(*vnet_hdr);
2447 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2450 return __packet_snd_vnet_parse(vnet_hdr, *len);
2453 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2454 void *frame, struct net_device *dev, void *data, int tp_len,
2455 __be16 proto, unsigned char *addr, int hlen, int copylen,
2456 const struct sockcm_cookie *sockc)
2458 union tpacket_uhdr ph;
2459 int to_write, offset, len, nr_frags, len_max;
2460 struct socket *sock = po->sk.sk_socket;
2466 skb->protocol = proto;
2468 skb->priority = po->sk.sk_priority;
2469 skb->mark = po->sk.sk_mark;
2470 skb->tstamp = sockc->transmit_time;
2471 skb_setup_tx_timestamp(skb, sockc->tsflags);
2472 skb_zcopy_set_nouarg(skb, ph.raw);
2474 skb_reserve(skb, hlen);
2475 skb_reset_network_header(skb);
2479 if (sock->type == SOCK_DGRAM) {
2480 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2482 if (unlikely(err < 0))
2484 } else if (copylen) {
2485 int hdrlen = min_t(int, copylen, tp_len);
2487 skb_push(skb, dev->hard_header_len);
2488 skb_put(skb, copylen - dev->hard_header_len);
2489 err = skb_store_bits(skb, 0, data, hdrlen);
2492 if (!dev_validate_header(dev, skb->data, hdrlen))
2499 offset = offset_in_page(data);
2500 len_max = PAGE_SIZE - offset;
2501 len = ((to_write > len_max) ? len_max : to_write);
2503 skb->data_len = to_write;
2504 skb->len += to_write;
2505 skb->truesize += to_write;
2506 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2508 while (likely(to_write)) {
2509 nr_frags = skb_shinfo(skb)->nr_frags;
2511 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2512 pr_err("Packet exceed the number of skb frags(%lu)\n",
2517 page = pgv_to_page(data);
2519 flush_dcache_page(page);
2521 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2524 len_max = PAGE_SIZE;
2525 len = ((to_write > len_max) ? len_max : to_write);
2528 packet_parse_headers(skb, sock);
2533 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2534 int size_max, void **data)
2536 union tpacket_uhdr ph;
2541 switch (po->tp_version) {
2543 if (ph.h3->tp_next_offset != 0) {
2544 pr_warn_once("variable sized slot not supported");
2547 tp_len = ph.h3->tp_len;
2550 tp_len = ph.h2->tp_len;
2553 tp_len = ph.h1->tp_len;
2556 if (unlikely(tp_len > size_max)) {
2557 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2561 if (unlikely(po->tp_tx_has_off)) {
2562 int off_min, off_max;
2564 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2565 off_max = po->tx_ring.frame_size - tp_len;
2566 if (po->sk.sk_type == SOCK_DGRAM) {
2567 switch (po->tp_version) {
2569 off = ph.h3->tp_net;
2572 off = ph.h2->tp_net;
2575 off = ph.h1->tp_net;
2579 switch (po->tp_version) {
2581 off = ph.h3->tp_mac;
2584 off = ph.h2->tp_mac;
2587 off = ph.h1->tp_mac;
2591 if (unlikely((off < off_min) || (off_max < off)))
2594 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2597 *data = frame + off;
2601 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2603 struct sk_buff *skb = NULL;
2604 struct net_device *dev;
2605 struct virtio_net_hdr *vnet_hdr = NULL;
2606 struct sockcm_cookie sockc;
2608 int err, reserve = 0;
2610 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2611 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2612 unsigned char *addr = NULL;
2613 int tp_len, size_max;
2616 int status = TP_STATUS_AVAILABLE;
2617 int hlen, tlen, copylen = 0;
2620 mutex_lock(&po->pg_vec_lock);
2622 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2623 * we need to confirm it under protection of pg_vec_lock.
2625 if (unlikely(!po->tx_ring.pg_vec)) {
2629 if (likely(saddr == NULL)) {
2630 dev = packet_cached_dev_get(po);
2634 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2636 if (msg->msg_namelen < (saddr->sll_halen
2637 + offsetof(struct sockaddr_ll,
2640 proto = saddr->sll_protocol;
2641 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2642 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2643 if (dev && msg->msg_namelen < dev->addr_len +
2644 offsetof(struct sockaddr_ll, sll_addr))
2646 addr = saddr->sll_addr;
2651 if (unlikely(dev == NULL))
2654 if (unlikely(!(dev->flags & IFF_UP)))
2657 sockcm_init(&sockc, &po->sk);
2658 if (msg->msg_controllen) {
2659 err = sock_cmsg_send(&po->sk, msg, &sockc);
2664 if (po->sk.sk_socket->type == SOCK_RAW)
2665 reserve = dev->hard_header_len;
2666 size_max = po->tx_ring.frame_size
2667 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2669 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2670 size_max = dev->mtu + reserve + VLAN_HLEN;
2672 reinit_completion(&po->skb_completion);
2675 ph = packet_current_frame(po, &po->tx_ring,
2676 TP_STATUS_SEND_REQUEST);
2677 if (unlikely(ph == NULL)) {
2678 if (need_wait && skb) {
2679 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2680 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2682 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2686 /* check for additional frames */
2691 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2695 status = TP_STATUS_SEND_REQUEST;
2696 hlen = LL_RESERVED_SPACE(dev);
2697 tlen = dev->needed_tailroom;
2698 if (po->has_vnet_hdr) {
2700 data += sizeof(*vnet_hdr);
2701 tp_len -= sizeof(*vnet_hdr);
2703 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2707 copylen = __virtio16_to_cpu(vio_le(),
2710 copylen = max_t(int, copylen, dev->hard_header_len);
2711 skb = sock_alloc_send_skb(&po->sk,
2712 hlen + tlen + sizeof(struct sockaddr_ll) +
2713 (copylen - dev->hard_header_len),
2716 if (unlikely(skb == NULL)) {
2717 /* we assume the socket was initially writeable ... */
2718 if (likely(len_sum > 0))
2722 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2723 addr, hlen, copylen, &sockc);
2724 if (likely(tp_len >= 0) &&
2725 tp_len > dev->mtu + reserve &&
2726 !po->has_vnet_hdr &&
2727 !packet_extra_vlan_len_allowed(dev, skb))
2730 if (unlikely(tp_len < 0)) {
2733 __packet_set_status(po, ph,
2734 TP_STATUS_AVAILABLE);
2735 packet_increment_head(&po->tx_ring);
2739 status = TP_STATUS_WRONG_FORMAT;
2745 if (po->has_vnet_hdr) {
2746 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2750 virtio_net_hdr_set_proto(skb, vnet_hdr);
2753 skb->destructor = tpacket_destruct_skb;
2754 __packet_set_status(po, ph, TP_STATUS_SENDING);
2755 packet_inc_pending(&po->tx_ring);
2757 status = TP_STATUS_SEND_REQUEST;
2758 err = po->xmit(skb);
2759 if (unlikely(err > 0)) {
2760 err = net_xmit_errno(err);
2761 if (err && __packet_get_status(po, ph) ==
2762 TP_STATUS_AVAILABLE) {
2763 /* skb was destructed already */
2768 * skb was dropped but not destructed yet;
2769 * let's treat it like congestion or err < 0
2773 packet_increment_head(&po->tx_ring);
2775 } while (likely((ph != NULL) ||
2776 /* Note: packet_read_pending() might be slow if we have
2777 * to call it as it's per_cpu variable, but in fast-path
2778 * we already short-circuit the loop with the first
2779 * condition, and luckily don't have to go that path
2782 (need_wait && packet_read_pending(&po->tx_ring))));
2788 __packet_set_status(po, ph, status);
2793 mutex_unlock(&po->pg_vec_lock);
2797 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2798 size_t reserve, size_t len,
2799 size_t linear, int noblock,
2802 struct sk_buff *skb;
2804 /* Under a page? Don't bother with paged skb. */
2805 if (prepad + len < PAGE_SIZE || !linear)
2808 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2813 skb_reserve(skb, reserve);
2814 skb_put(skb, linear);
2815 skb->data_len = len - linear;
2816 skb->len += len - linear;
2821 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2823 struct sock *sk = sock->sk;
2824 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2825 struct sk_buff *skb;
2826 struct net_device *dev;
2828 unsigned char *addr = NULL;
2829 int err, reserve = 0;
2830 struct sockcm_cookie sockc;
2831 struct virtio_net_hdr vnet_hdr = { 0 };
2833 struct packet_sock *po = pkt_sk(sk);
2834 bool has_vnet_hdr = false;
2835 int hlen, tlen, linear;
2839 * Get and verify the address.
2842 if (likely(saddr == NULL)) {
2843 dev = packet_cached_dev_get(po);
2847 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2849 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2851 proto = saddr->sll_protocol;
2852 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2853 if (sock->type == SOCK_DGRAM) {
2854 if (dev && msg->msg_namelen < dev->addr_len +
2855 offsetof(struct sockaddr_ll, sll_addr))
2857 addr = saddr->sll_addr;
2862 if (unlikely(dev == NULL))
2865 if (unlikely(!(dev->flags & IFF_UP)))
2868 sockcm_init(&sockc, sk);
2869 sockc.mark = sk->sk_mark;
2870 if (msg->msg_controllen) {
2871 err = sock_cmsg_send(sk, msg, &sockc);
2876 if (sock->type == SOCK_RAW)
2877 reserve = dev->hard_header_len;
2878 if (po->has_vnet_hdr) {
2879 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2882 has_vnet_hdr = true;
2885 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2886 if (!netif_supports_nofcs(dev)) {
2887 err = -EPROTONOSUPPORT;
2890 extra_len = 4; /* We're doing our own CRC */
2894 if (!vnet_hdr.gso_type &&
2895 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2899 hlen = LL_RESERVED_SPACE(dev);
2900 tlen = dev->needed_tailroom;
2901 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2902 linear = max(linear, min_t(int, len, dev->hard_header_len));
2903 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2904 msg->msg_flags & MSG_DONTWAIT, &err);
2908 skb_reset_network_header(skb);
2911 if (sock->type == SOCK_DGRAM) {
2912 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2913 if (unlikely(offset < 0))
2915 } else if (reserve) {
2916 skb_reserve(skb, -reserve);
2917 if (len < reserve + sizeof(struct ipv6hdr) &&
2918 dev->min_header_len != dev->hard_header_len)
2919 skb_reset_network_header(skb);
2922 /* Returns -EFAULT on error */
2923 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2927 if (sock->type == SOCK_RAW &&
2928 !dev_validate_header(dev, skb->data, len)) {
2933 skb_setup_tx_timestamp(skb, sockc.tsflags);
2935 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2936 !packet_extra_vlan_len_allowed(dev, skb)) {
2941 skb->protocol = proto;
2943 skb->priority = sk->sk_priority;
2944 skb->mark = sockc.mark;
2945 skb->tstamp = sockc.transmit_time;
2948 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2951 len += sizeof(vnet_hdr);
2952 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2955 packet_parse_headers(skb, sock);
2957 if (unlikely(extra_len == 4))
2960 err = po->xmit(skb);
2961 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2977 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2979 struct sock *sk = sock->sk;
2980 struct packet_sock *po = pkt_sk(sk);
2982 if (po->tx_ring.pg_vec)
2983 return tpacket_snd(po, msg);
2985 return packet_snd(sock, msg, len);
2989 * Close a PACKET socket. This is fairly simple. We immediately go
2990 * to 'closed' state and remove our protocol entry in the device list.
2993 static int packet_release(struct socket *sock)
2995 struct sock *sk = sock->sk;
2996 struct packet_sock *po;
2997 struct packet_fanout *f;
2999 union tpacket_req_u req_u;
3007 mutex_lock(&net->packet.sklist_lock);
3008 sk_del_node_init_rcu(sk);
3009 mutex_unlock(&net->packet.sklist_lock);
3012 sock_prot_inuse_add(net, sk->sk_prot, -1);
3015 spin_lock(&po->bind_lock);
3016 unregister_prot_hook(sk, false);
3017 packet_cached_dev_reset(po);
3019 if (po->prot_hook.dev) {
3020 dev_put(po->prot_hook.dev);
3021 po->prot_hook.dev = NULL;
3023 spin_unlock(&po->bind_lock);
3025 packet_flush_mclist(sk);
3028 if (po->rx_ring.pg_vec) {
3029 memset(&req_u, 0, sizeof(req_u));
3030 packet_set_ring(sk, &req_u, 1, 0);
3033 if (po->tx_ring.pg_vec) {
3034 memset(&req_u, 0, sizeof(req_u));
3035 packet_set_ring(sk, &req_u, 1, 1);
3039 f = fanout_release(sk);
3043 kfree(po->rollover);
3045 fanout_release_data(f);
3049 * Now the socket is dead. No more input will appear.
3056 skb_queue_purge(&sk->sk_receive_queue);
3057 packet_free_pending(po);
3058 sk_refcnt_debug_release(sk);
3065 * Attach a packet hook.
3068 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3071 struct packet_sock *po = pkt_sk(sk);
3072 struct net_device *dev_curr;
3075 struct net_device *dev = NULL;
3077 bool unlisted = false;
3080 spin_lock(&po->bind_lock);
3089 dev = dev_get_by_name_rcu(sock_net(sk), name);
3094 } else if (ifindex) {
3095 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3105 proto_curr = po->prot_hook.type;
3106 dev_curr = po->prot_hook.dev;
3108 need_rehook = proto_curr != proto || dev_curr != dev;
3113 /* prevents packet_notifier() from calling
3114 * register_prot_hook()
3117 __unregister_prot_hook(sk, true);
3119 dev_curr = po->prot_hook.dev;
3121 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3125 BUG_ON(po->running);
3127 po->prot_hook.type = proto;
3129 if (unlikely(unlisted)) {
3131 po->prot_hook.dev = NULL;
3133 packet_cached_dev_reset(po);
3135 po->prot_hook.dev = dev;
3136 po->ifindex = dev ? dev->ifindex : 0;
3137 packet_cached_dev_assign(po, dev);
3143 if (proto == 0 || !need_rehook)
3146 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3147 register_prot_hook(sk);
3149 sk->sk_err = ENETDOWN;
3150 if (!sock_flag(sk, SOCK_DEAD))
3151 sk->sk_error_report(sk);
3156 spin_unlock(&po->bind_lock);
3162 * Bind a packet socket to a device
3165 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3168 struct sock *sk = sock->sk;
3169 char name[sizeof(uaddr->sa_data) + 1];
3175 if (addr_len != sizeof(struct sockaddr))
3177 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3180 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3181 name[sizeof(uaddr->sa_data)] = 0;
3183 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3186 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3188 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3189 struct sock *sk = sock->sk;
3195 if (addr_len < sizeof(struct sockaddr_ll))
3197 if (sll->sll_family != AF_PACKET)
3200 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3201 sll->sll_protocol ? : pkt_sk(sk)->num);
3204 static struct proto packet_proto = {
3206 .owner = THIS_MODULE,
3207 .obj_size = sizeof(struct packet_sock),
3211 * Create a packet of type SOCK_PACKET.
3214 static int packet_create(struct net *net, struct socket *sock, int protocol,
3218 struct packet_sock *po;
3219 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3222 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3224 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3225 sock->type != SOCK_PACKET)
3226 return -ESOCKTNOSUPPORT;
3228 sock->state = SS_UNCONNECTED;
3231 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3235 sock->ops = &packet_ops;
3236 if (sock->type == SOCK_PACKET)
3237 sock->ops = &packet_ops_spkt;
3239 sock_init_data(sock, sk);
3242 init_completion(&po->skb_completion);
3243 sk->sk_family = PF_PACKET;
3245 po->xmit = dev_queue_xmit;
3247 err = packet_alloc_pending(po);
3251 packet_cached_dev_reset(po);
3253 sk->sk_destruct = packet_sock_destruct;
3254 sk_refcnt_debug_inc(sk);
3257 * Attach a protocol block
3260 spin_lock_init(&po->bind_lock);
3261 mutex_init(&po->pg_vec_lock);
3262 po->rollover = NULL;
3263 po->prot_hook.func = packet_rcv;
3265 if (sock->type == SOCK_PACKET)
3266 po->prot_hook.func = packet_rcv_spkt;
3268 po->prot_hook.af_packet_priv = sk;
3271 po->prot_hook.type = proto;
3272 __register_prot_hook(sk);
3275 mutex_lock(&net->packet.sklist_lock);
3276 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3277 mutex_unlock(&net->packet.sklist_lock);
3280 sock_prot_inuse_add(net, &packet_proto, 1);
3291 * Pull a packet from our receive queue and hand it to the user.
3292 * If necessary we block.
3295 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3298 struct sock *sk = sock->sk;
3299 struct sk_buff *skb;
3301 int vnet_hdr_len = 0;
3302 unsigned int origlen = 0;
3305 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3309 /* What error should we return now? EUNATTACH? */
3310 if (pkt_sk(sk)->ifindex < 0)
3314 if (flags & MSG_ERRQUEUE) {
3315 err = sock_recv_errqueue(sk, msg, len,
3316 SOL_PACKET, PACKET_TX_TIMESTAMP);
3321 * Call the generic datagram receiver. This handles all sorts
3322 * of horrible races and re-entrancy so we can forget about it
3323 * in the protocol layers.
3325 * Now it will return ENETDOWN, if device have just gone down,
3326 * but then it will block.
3329 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3332 * An error occurred so return it. Because skb_recv_datagram()
3333 * handles the blocking we don't see and worry about blocking
3340 packet_rcv_try_clear_pressure(pkt_sk(sk));
3342 if (pkt_sk(sk)->has_vnet_hdr) {
3343 err = packet_rcv_vnet(msg, skb, &len);
3346 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3349 /* You lose any data beyond the buffer you gave. If it worries
3350 * a user program they can ask the device for its MTU
3356 msg->msg_flags |= MSG_TRUNC;
3359 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3363 if (sock->type != SOCK_PACKET) {
3364 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3366 /* Original length was stored in sockaddr_ll fields */
3367 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3368 sll->sll_family = AF_PACKET;
3369 sll->sll_protocol = skb->protocol;
3372 sock_recv_ts_and_drops(msg, sk, skb);
3374 if (msg->msg_name) {
3377 /* If the address length field is there to be filled
3378 * in, we fill it in now.
3380 if (sock->type == SOCK_PACKET) {
3381 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3382 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3383 copy_len = msg->msg_namelen;
3385 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3387 msg->msg_namelen = sll->sll_halen +
3388 offsetof(struct sockaddr_ll, sll_addr);
3389 copy_len = msg->msg_namelen;
3390 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3391 memset(msg->msg_name +
3392 offsetof(struct sockaddr_ll, sll_addr),
3393 0, sizeof(sll->sll_addr));
3394 msg->msg_namelen = sizeof(struct sockaddr_ll);
3397 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3400 if (pkt_sk(sk)->auxdata) {
3401 struct tpacket_auxdata aux;
3403 aux.tp_status = TP_STATUS_USER;
3404 if (skb->ip_summed == CHECKSUM_PARTIAL)
3405 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3406 else if (skb->pkt_type != PACKET_OUTGOING &&
3407 (skb->ip_summed == CHECKSUM_COMPLETE ||
3408 skb_csum_unnecessary(skb)))
3409 aux.tp_status |= TP_STATUS_CSUM_VALID;
3411 aux.tp_len = origlen;
3412 aux.tp_snaplen = skb->len;
3414 aux.tp_net = skb_network_offset(skb);
3415 if (skb_vlan_tag_present(skb)) {
3416 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3417 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3418 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3420 aux.tp_vlan_tci = 0;
3421 aux.tp_vlan_tpid = 0;
3423 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3427 * Free or return the buffer as appropriate. Again this
3428 * hides all the races and re-entrancy issues from us.
3430 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3433 skb_free_datagram(sk, skb);
3438 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3441 struct net_device *dev;
3442 struct sock *sk = sock->sk;
3447 uaddr->sa_family = AF_PACKET;
3448 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3450 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3452 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3455 return sizeof(*uaddr);
3458 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3461 struct net_device *dev;
3462 struct sock *sk = sock->sk;
3463 struct packet_sock *po = pkt_sk(sk);
3464 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3469 sll->sll_family = AF_PACKET;
3470 sll->sll_ifindex = po->ifindex;
3471 sll->sll_protocol = po->num;
3472 sll->sll_pkttype = 0;
3474 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3476 sll->sll_hatype = dev->type;
3477 sll->sll_halen = dev->addr_len;
3478 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3480 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3485 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3488 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3492 case PACKET_MR_MULTICAST:
3493 if (i->alen != dev->addr_len)
3496 return dev_mc_add(dev, i->addr);
3498 return dev_mc_del(dev, i->addr);
3500 case PACKET_MR_PROMISC:
3501 return dev_set_promiscuity(dev, what);
3502 case PACKET_MR_ALLMULTI:
3503 return dev_set_allmulti(dev, what);
3504 case PACKET_MR_UNICAST:
3505 if (i->alen != dev->addr_len)
3508 return dev_uc_add(dev, i->addr);
3510 return dev_uc_del(dev, i->addr);
3518 static void packet_dev_mclist_delete(struct net_device *dev,
3519 struct packet_mclist **mlp)
3521 struct packet_mclist *ml;
3523 while ((ml = *mlp) != NULL) {
3524 if (ml->ifindex == dev->ifindex) {
3525 packet_dev_mc(dev, ml, -1);
3533 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3535 struct packet_sock *po = pkt_sk(sk);
3536 struct packet_mclist *ml, *i;
3537 struct net_device *dev;
3543 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3548 if (mreq->mr_alen > dev->addr_len)
3552 i = kmalloc(sizeof(*i), GFP_KERNEL);
3557 for (ml = po->mclist; ml; ml = ml->next) {
3558 if (ml->ifindex == mreq->mr_ifindex &&
3559 ml->type == mreq->mr_type &&
3560 ml->alen == mreq->mr_alen &&
3561 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3563 /* Free the new element ... */
3569 i->type = mreq->mr_type;
3570 i->ifindex = mreq->mr_ifindex;
3571 i->alen = mreq->mr_alen;
3572 memcpy(i->addr, mreq->mr_address, i->alen);
3573 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3575 i->next = po->mclist;
3577 err = packet_dev_mc(dev, i, 1);
3579 po->mclist = i->next;
3588 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3590 struct packet_mclist *ml, **mlp;
3594 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3595 if (ml->ifindex == mreq->mr_ifindex &&
3596 ml->type == mreq->mr_type &&
3597 ml->alen == mreq->mr_alen &&
3598 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3599 if (--ml->count == 0) {
3600 struct net_device *dev;
3602 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3604 packet_dev_mc(dev, ml, -1);
3614 static void packet_flush_mclist(struct sock *sk)
3616 struct packet_sock *po = pkt_sk(sk);
3617 struct packet_mclist *ml;
3623 while ((ml = po->mclist) != NULL) {
3624 struct net_device *dev;
3626 po->mclist = ml->next;
3627 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3629 packet_dev_mc(dev, ml, -1);
3636 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3638 struct sock *sk = sock->sk;
3639 struct packet_sock *po = pkt_sk(sk);
3642 if (level != SOL_PACKET)
3643 return -ENOPROTOOPT;
3646 case PACKET_ADD_MEMBERSHIP:
3647 case PACKET_DROP_MEMBERSHIP:
3649 struct packet_mreq_max mreq;
3651 memset(&mreq, 0, sizeof(mreq));
3652 if (len < sizeof(struct packet_mreq))
3654 if (len > sizeof(mreq))
3656 if (copy_from_user(&mreq, optval, len))
3658 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3660 if (optname == PACKET_ADD_MEMBERSHIP)
3661 ret = packet_mc_add(sk, &mreq);
3663 ret = packet_mc_drop(sk, &mreq);
3667 case PACKET_RX_RING:
3668 case PACKET_TX_RING:
3670 union tpacket_req_u req_u;
3674 switch (po->tp_version) {
3677 len = sizeof(req_u.req);
3681 len = sizeof(req_u.req3);
3687 if (copy_from_user(&req_u.req, optval, len))
3690 ret = packet_set_ring(sk, &req_u, 0,
3691 optname == PACKET_TX_RING);
3696 case PACKET_COPY_THRESH:
3700 if (optlen != sizeof(val))
3702 if (copy_from_user(&val, optval, sizeof(val)))
3705 pkt_sk(sk)->copy_thresh = val;
3708 case PACKET_VERSION:
3712 if (optlen != sizeof(val))
3714 if (copy_from_user(&val, optval, sizeof(val)))
3725 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3728 po->tp_version = val;
3734 case PACKET_RESERVE:
3738 if (optlen != sizeof(val))
3740 if (copy_from_user(&val, optval, sizeof(val)))
3745 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3748 po->tp_reserve = val;
3758 if (optlen != sizeof(val))
3760 if (copy_from_user(&val, optval, sizeof(val)))
3764 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3767 po->tp_loss = !!val;
3773 case PACKET_AUXDATA:
3777 if (optlen < sizeof(val))
3779 if (copy_from_user(&val, optval, sizeof(val)))
3783 po->auxdata = !!val;
3787 case PACKET_ORIGDEV:
3791 if (optlen < sizeof(val))
3793 if (copy_from_user(&val, optval, sizeof(val)))
3797 po->origdev = !!val;
3801 case PACKET_VNET_HDR:
3805 if (sock->type != SOCK_RAW)
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->has_vnet_hdr = !!val;
3822 case PACKET_TIMESTAMP:
3826 if (optlen != sizeof(val))
3828 if (copy_from_user(&val, optval, sizeof(val)))
3831 po->tp_tstamp = val;
3838 if (optlen != sizeof(val))
3840 if (copy_from_user(&val, optval, sizeof(val)))
3843 return fanout_add(sk, val & 0xffff, val >> 16);
3845 case PACKET_FANOUT_DATA:
3850 return fanout_set_data(po, optval, optlen);
3852 case PACKET_IGNORE_OUTGOING:
3856 if (optlen != sizeof(val))
3858 if (copy_from_user(&val, optval, sizeof(val)))
3860 if (val < 0 || val > 1)
3863 po->prot_hook.ignore_outgoing = !!val;
3866 case PACKET_TX_HAS_OFF:
3870 if (optlen != sizeof(val))
3872 if (copy_from_user(&val, optval, sizeof(val)))
3876 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3879 po->tp_tx_has_off = !!val;
3885 case PACKET_QDISC_BYPASS:
3889 if (optlen != sizeof(val))
3891 if (copy_from_user(&val, optval, sizeof(val)))
3894 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3898 return -ENOPROTOOPT;
3902 static int packet_getsockopt(struct socket *sock, int level, int optname,
3903 char __user *optval, int __user *optlen)
3906 int val, lv = sizeof(val);
3907 struct sock *sk = sock->sk;
3908 struct packet_sock *po = pkt_sk(sk);
3910 union tpacket_stats_u st;
3911 struct tpacket_rollover_stats rstats;
3914 if (level != SOL_PACKET)
3915 return -ENOPROTOOPT;
3917 if (get_user(len, optlen))
3924 case PACKET_STATISTICS:
3925 spin_lock_bh(&sk->sk_receive_queue.lock);
3926 memcpy(&st, &po->stats, sizeof(st));
3927 memset(&po->stats, 0, sizeof(po->stats));
3928 spin_unlock_bh(&sk->sk_receive_queue.lock);
3929 drops = atomic_xchg(&po->tp_drops, 0);
3931 if (po->tp_version == TPACKET_V3) {
3932 lv = sizeof(struct tpacket_stats_v3);
3933 st.stats3.tp_drops = drops;
3934 st.stats3.tp_packets += drops;
3937 lv = sizeof(struct tpacket_stats);
3938 st.stats1.tp_drops = drops;
3939 st.stats1.tp_packets += drops;
3944 case PACKET_AUXDATA:
3947 case PACKET_ORIGDEV:
3950 case PACKET_VNET_HDR:
3951 val = po->has_vnet_hdr;
3953 case PACKET_VERSION:
3954 val = po->tp_version;
3957 if (len > sizeof(int))
3959 if (len < sizeof(int))
3961 if (copy_from_user(&val, optval, len))
3965 val = sizeof(struct tpacket_hdr);
3968 val = sizeof(struct tpacket2_hdr);
3971 val = sizeof(struct tpacket3_hdr);
3977 case PACKET_RESERVE:
3978 val = po->tp_reserve;
3983 case PACKET_TIMESTAMP:
3984 val = po->tp_tstamp;
3988 ((u32)po->fanout->id |
3989 ((u32)po->fanout->type << 16) |
3990 ((u32)po->fanout->flags << 24)) :
3993 case PACKET_IGNORE_OUTGOING:
3994 val = po->prot_hook.ignore_outgoing;
3996 case PACKET_ROLLOVER_STATS:
3999 rstats.tp_all = atomic_long_read(&po->rollover->num);
4000 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4001 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4003 lv = sizeof(rstats);
4005 case PACKET_TX_HAS_OFF:
4006 val = po->tp_tx_has_off;
4008 case PACKET_QDISC_BYPASS:
4009 val = packet_use_direct_xmit(po);
4012 return -ENOPROTOOPT;
4017 if (put_user(len, optlen))
4019 if (copy_to_user(optval, data, len))
4025 #ifdef CONFIG_COMPAT
4026 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4027 char __user *optval, unsigned int optlen)
4029 struct packet_sock *po = pkt_sk(sock->sk);
4031 if (level != SOL_PACKET)
4032 return -ENOPROTOOPT;
4034 if (optname == PACKET_FANOUT_DATA &&
4035 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4036 optval = (char __user *)get_compat_bpf_fprog(optval);
4039 optlen = sizeof(struct sock_fprog);
4042 return packet_setsockopt(sock, level, optname, optval, optlen);
4046 static int packet_notifier(struct notifier_block *this,
4047 unsigned long msg, void *ptr)
4050 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4051 struct net *net = dev_net(dev);
4054 sk_for_each_rcu(sk, &net->packet.sklist) {
4055 struct packet_sock *po = pkt_sk(sk);
4058 case NETDEV_UNREGISTER:
4060 packet_dev_mclist_delete(dev, &po->mclist);
4064 if (dev->ifindex == po->ifindex) {
4065 spin_lock(&po->bind_lock);
4067 __unregister_prot_hook(sk, false);
4068 sk->sk_err = ENETDOWN;
4069 if (!sock_flag(sk, SOCK_DEAD))
4070 sk->sk_error_report(sk);
4072 if (msg == NETDEV_UNREGISTER) {
4073 packet_cached_dev_reset(po);
4075 if (po->prot_hook.dev)
4076 dev_put(po->prot_hook.dev);
4077 po->prot_hook.dev = NULL;
4079 spin_unlock(&po->bind_lock);
4083 if (dev->ifindex == po->ifindex) {
4084 spin_lock(&po->bind_lock);
4086 register_prot_hook(sk);
4087 spin_unlock(&po->bind_lock);
4097 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4100 struct sock *sk = sock->sk;
4105 int amount = sk_wmem_alloc_get(sk);
4107 return put_user(amount, (int __user *)arg);
4111 struct sk_buff *skb;
4114 spin_lock_bh(&sk->sk_receive_queue.lock);
4115 skb = skb_peek(&sk->sk_receive_queue);
4118 spin_unlock_bh(&sk->sk_receive_queue.lock);
4119 return put_user(amount, (int __user *)arg);
4129 case SIOCGIFBRDADDR:
4130 case SIOCSIFBRDADDR:
4131 case SIOCGIFNETMASK:
4132 case SIOCSIFNETMASK:
4133 case SIOCGIFDSTADDR:
4134 case SIOCSIFDSTADDR:
4136 return inet_dgram_ops.ioctl(sock, cmd, arg);
4140 return -ENOIOCTLCMD;
4145 static __poll_t packet_poll(struct file *file, struct socket *sock,
4148 struct sock *sk = sock->sk;
4149 struct packet_sock *po = pkt_sk(sk);
4150 __poll_t mask = datagram_poll(file, sock, wait);
4152 spin_lock_bh(&sk->sk_receive_queue.lock);
4153 if (po->rx_ring.pg_vec) {
4154 if (!packet_previous_rx_frame(po, &po->rx_ring,
4156 mask |= EPOLLIN | EPOLLRDNORM;
4158 packet_rcv_try_clear_pressure(po);
4159 spin_unlock_bh(&sk->sk_receive_queue.lock);
4160 spin_lock_bh(&sk->sk_write_queue.lock);
4161 if (po->tx_ring.pg_vec) {
4162 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4163 mask |= EPOLLOUT | EPOLLWRNORM;
4165 spin_unlock_bh(&sk->sk_write_queue.lock);
4170 /* Dirty? Well, I still did not learn better way to account
4174 static void packet_mm_open(struct vm_area_struct *vma)
4176 struct file *file = vma->vm_file;
4177 struct socket *sock = file->private_data;
4178 struct sock *sk = sock->sk;
4181 atomic_inc(&pkt_sk(sk)->mapped);
4184 static void packet_mm_close(struct vm_area_struct *vma)
4186 struct file *file = vma->vm_file;
4187 struct socket *sock = file->private_data;
4188 struct sock *sk = sock->sk;
4191 atomic_dec(&pkt_sk(sk)->mapped);
4194 static const struct vm_operations_struct packet_mmap_ops = {
4195 .open = packet_mm_open,
4196 .close = packet_mm_close,
4199 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4204 for (i = 0; i < len; i++) {
4205 if (likely(pg_vec[i].buffer)) {
4206 if (is_vmalloc_addr(pg_vec[i].buffer))
4207 vfree(pg_vec[i].buffer);
4209 free_pages((unsigned long)pg_vec[i].buffer,
4211 pg_vec[i].buffer = NULL;
4217 static char *alloc_one_pg_vec_page(unsigned long order)
4220 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4221 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4223 buffer = (char *) __get_free_pages(gfp_flags, order);
4227 /* __get_free_pages failed, fall back to vmalloc */
4228 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4232 /* vmalloc failed, lets dig into swap here */
4233 gfp_flags &= ~__GFP_NORETRY;
4234 buffer = (char *) __get_free_pages(gfp_flags, order);
4238 /* complete and utter failure */
4242 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4244 unsigned int block_nr = req->tp_block_nr;
4248 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4249 if (unlikely(!pg_vec))
4252 for (i = 0; i < block_nr; i++) {
4253 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4254 if (unlikely(!pg_vec[i].buffer))
4255 goto out_free_pgvec;
4262 free_pg_vec(pg_vec, order, block_nr);
4267 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4268 int closing, int tx_ring)
4270 struct pgv *pg_vec = NULL;
4271 struct packet_sock *po = pkt_sk(sk);
4272 int was_running, order = 0;
4273 struct packet_ring_buffer *rb;
4274 struct sk_buff_head *rb_queue;
4277 /* Added to avoid minimal code churn */
4278 struct tpacket_req *req = &req_u->req;
4280 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4281 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4285 if (atomic_read(&po->mapped))
4287 if (packet_read_pending(rb))
4291 if (req->tp_block_nr) {
4292 unsigned int min_frame_size;
4294 /* Sanity tests and some calculations */
4296 if (unlikely(rb->pg_vec))
4299 switch (po->tp_version) {
4301 po->tp_hdrlen = TPACKET_HDRLEN;
4304 po->tp_hdrlen = TPACKET2_HDRLEN;
4307 po->tp_hdrlen = TPACKET3_HDRLEN;
4312 if (unlikely((int)req->tp_block_size <= 0))
4314 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4316 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4317 if (po->tp_version >= TPACKET_V3 &&
4318 req->tp_block_size <
4319 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4321 if (unlikely(req->tp_frame_size < min_frame_size))
4323 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4326 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4327 if (unlikely(rb->frames_per_block == 0))
4329 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4331 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4336 order = get_order(req->tp_block_size);
4337 pg_vec = alloc_pg_vec(req, order);
4338 if (unlikely(!pg_vec))
4340 switch (po->tp_version) {
4342 /* Block transmit is not supported yet */
4344 init_prb_bdqc(po, rb, pg_vec, req_u);
4346 struct tpacket_req3 *req3 = &req_u->req3;
4348 if (req3->tp_retire_blk_tov ||
4349 req3->tp_sizeof_priv ||
4350 req3->tp_feature_req_word) {
4352 goto out_free_pg_vec;
4363 if (unlikely(req->tp_frame_nr))
4368 /* Detach socket from network */
4369 spin_lock(&po->bind_lock);
4370 was_running = po->running;
4374 __unregister_prot_hook(sk, false);
4376 spin_unlock(&po->bind_lock);
4381 mutex_lock(&po->pg_vec_lock);
4382 if (closing || atomic_read(&po->mapped) == 0) {
4384 spin_lock_bh(&rb_queue->lock);
4385 swap(rb->pg_vec, pg_vec);
4386 rb->frame_max = (req->tp_frame_nr - 1);
4388 rb->frame_size = req->tp_frame_size;
4389 spin_unlock_bh(&rb_queue->lock);
4391 swap(rb->pg_vec_order, order);
4392 swap(rb->pg_vec_len, req->tp_block_nr);
4394 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4395 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4396 tpacket_rcv : packet_rcv;
4397 skb_queue_purge(rb_queue);
4398 if (atomic_read(&po->mapped))
4399 pr_err("packet_mmap: vma is busy: %d\n",
4400 atomic_read(&po->mapped));
4402 mutex_unlock(&po->pg_vec_lock);
4404 spin_lock(&po->bind_lock);
4407 register_prot_hook(sk);
4409 spin_unlock(&po->bind_lock);
4410 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4411 /* Because we don't support block-based V3 on tx-ring */
4413 prb_shutdown_retire_blk_timer(po, rb_queue);
4418 free_pg_vec(pg_vec, order, req->tp_block_nr);
4423 static int packet_mmap(struct file *file, struct socket *sock,
4424 struct vm_area_struct *vma)
4426 struct sock *sk = sock->sk;
4427 struct packet_sock *po = pkt_sk(sk);
4428 unsigned long size, expected_size;
4429 struct packet_ring_buffer *rb;
4430 unsigned long start;
4437 mutex_lock(&po->pg_vec_lock);
4440 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4442 expected_size += rb->pg_vec_len
4448 if (expected_size == 0)
4451 size = vma->vm_end - vma->vm_start;
4452 if (size != expected_size)
4455 start = vma->vm_start;
4456 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4457 if (rb->pg_vec == NULL)
4460 for (i = 0; i < rb->pg_vec_len; i++) {
4462 void *kaddr = rb->pg_vec[i].buffer;
4465 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4466 page = pgv_to_page(kaddr);
4467 err = vm_insert_page(vma, start, page);
4476 atomic_inc(&po->mapped);
4477 vma->vm_ops = &packet_mmap_ops;
4481 mutex_unlock(&po->pg_vec_lock);
4485 static const struct proto_ops packet_ops_spkt = {
4486 .family = PF_PACKET,
4487 .owner = THIS_MODULE,
4488 .release = packet_release,
4489 .bind = packet_bind_spkt,
4490 .connect = sock_no_connect,
4491 .socketpair = sock_no_socketpair,
4492 .accept = sock_no_accept,
4493 .getname = packet_getname_spkt,
4494 .poll = datagram_poll,
4495 .ioctl = packet_ioctl,
4496 .gettstamp = sock_gettstamp,
4497 .listen = sock_no_listen,
4498 .shutdown = sock_no_shutdown,
4499 .setsockopt = sock_no_setsockopt,
4500 .getsockopt = sock_no_getsockopt,
4501 .sendmsg = packet_sendmsg_spkt,
4502 .recvmsg = packet_recvmsg,
4503 .mmap = sock_no_mmap,
4504 .sendpage = sock_no_sendpage,
4507 static const struct proto_ops packet_ops = {
4508 .family = PF_PACKET,
4509 .owner = THIS_MODULE,
4510 .release = packet_release,
4511 .bind = packet_bind,
4512 .connect = sock_no_connect,
4513 .socketpair = sock_no_socketpair,
4514 .accept = sock_no_accept,
4515 .getname = packet_getname,
4516 .poll = packet_poll,
4517 .ioctl = packet_ioctl,
4518 .gettstamp = sock_gettstamp,
4519 .listen = sock_no_listen,
4520 .shutdown = sock_no_shutdown,
4521 .setsockopt = packet_setsockopt,
4522 .getsockopt = packet_getsockopt,
4523 #ifdef CONFIG_COMPAT
4524 .compat_setsockopt = compat_packet_setsockopt,
4526 .sendmsg = packet_sendmsg,
4527 .recvmsg = packet_recvmsg,
4528 .mmap = packet_mmap,
4529 .sendpage = sock_no_sendpage,
4532 static const struct net_proto_family packet_family_ops = {
4533 .family = PF_PACKET,
4534 .create = packet_create,
4535 .owner = THIS_MODULE,
4538 static struct notifier_block packet_netdev_notifier = {
4539 .notifier_call = packet_notifier,
4542 #ifdef CONFIG_PROC_FS
4544 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4547 struct net *net = seq_file_net(seq);
4550 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4553 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4555 struct net *net = seq_file_net(seq);
4556 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4559 static void packet_seq_stop(struct seq_file *seq, void *v)
4565 static int packet_seq_show(struct seq_file *seq, void *v)
4567 if (v == SEQ_START_TOKEN)
4568 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4570 struct sock *s = sk_entry(v);
4571 const struct packet_sock *po = pkt_sk(s);
4574 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4576 refcount_read(&s->sk_refcnt),
4581 atomic_read(&s->sk_rmem_alloc),
4582 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4589 static const struct seq_operations packet_seq_ops = {
4590 .start = packet_seq_start,
4591 .next = packet_seq_next,
4592 .stop = packet_seq_stop,
4593 .show = packet_seq_show,
4597 static int __net_init packet_net_init(struct net *net)
4599 mutex_init(&net->packet.sklist_lock);
4600 INIT_HLIST_HEAD(&net->packet.sklist);
4602 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4603 sizeof(struct seq_net_private)))
4609 static void __net_exit packet_net_exit(struct net *net)
4611 remove_proc_entry("packet", net->proc_net);
4612 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4615 static struct pernet_operations packet_net_ops = {
4616 .init = packet_net_init,
4617 .exit = packet_net_exit,
4621 static void __exit packet_exit(void)
4623 unregister_netdevice_notifier(&packet_netdev_notifier);
4624 unregister_pernet_subsys(&packet_net_ops);
4625 sock_unregister(PF_PACKET);
4626 proto_unregister(&packet_proto);
4629 static int __init packet_init(void)
4633 rc = proto_register(&packet_proto, 0);
4636 rc = sock_register(&packet_family_ops);
4639 rc = register_pernet_subsys(&packet_net_ops);
4642 rc = register_netdevice_notifier(&packet_netdev_notifier);
4649 unregister_pernet_subsys(&packet_net_ops);
4651 sock_unregister(PF_PACKET);
4653 proto_unregister(&packet_proto);
4658 module_init(packet_init);
4659 module_exit(packet_exit);
4660 MODULE_LICENSE("GPL");
4661 MODULE_ALIAS_NETPROTO(PF_PACKET);