1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PACKET - implements raw packet sockets.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
49 #include <linux/types.h>
51 #include <linux/capability.h>
52 #include <linux/fcntl.h>
53 #include <linux/socket.h>
55 #include <linux/inet.h>
56 #include <linux/netdevice.h>
57 #include <linux/if_packet.h>
58 #include <linux/wireless.h>
59 #include <linux/kernel.h>
60 #include <linux/kmod.h>
61 #include <linux/slab.h>
62 #include <linux/vmalloc.h>
63 #include <net/net_namespace.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.h>
68 #include <linux/errno.h>
69 #include <linux/timer.h>
70 #include <linux/uaccess.h>
71 #include <asm/ioctls.h>
73 #include <asm/cacheflush.h>
75 #include <linux/proc_fs.h>
76 #include <linux/seq_file.h>
77 #include <linux/poll.h>
78 #include <linux/module.h>
79 #include <linux/init.h>
80 #include <linux/mutex.h>
81 #include <linux/if_vlan.h>
82 #include <linux/virtio_net.h>
83 #include <linux/errqueue.h>
84 #include <linux/net_tstamp.h>
85 #include <linux/percpu.h>
87 #include <net/inet_common.h>
89 #include <linux/bpf.h>
90 #include <net/compat.h>
96 - if device has no dev->hard_header routine, it adds and removes ll header
97 inside itself. In this case ll header is invisible outside of device,
98 but higher levels still should reserve dev->hard_header_len.
99 Some devices are enough clever to reallocate skb, when header
100 will not fit to reserved space (tunnel), another ones are silly
102 - packet socket receives packets with pulled ll header,
103 so that SOCK_RAW should push it back.
108 Incoming, dev->hard_header!=NULL
109 mac_header -> ll header
112 Outgoing, dev->hard_header!=NULL
113 mac_header -> ll header
116 Incoming, dev->hard_header==NULL
117 mac_header -> UNKNOWN position. It is very likely, that it points to ll
118 header. PPP makes it, that is wrong, because introduce
119 assymetry between rx and tx paths.
122 Outgoing, dev->hard_header==NULL
123 mac_header -> data. ll header is still not built!
127 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
133 dev->hard_header != NULL
134 mac_header -> ll header
137 dev->hard_header == NULL (ll header is added by device, we cannot control it)
141 We should set nh.raw on output to correct posistion,
142 packet classifier depends on it.
145 /* Private packet socket structures. */
147 /* identical to struct packet_mreq except it has
148 * a longer address field.
150 struct packet_mreq_max {
152 unsigned short mr_type;
153 unsigned short mr_alen;
154 unsigned char mr_address[MAX_ADDR_LEN];
158 struct tpacket_hdr *h1;
159 struct tpacket2_hdr *h2;
160 struct tpacket3_hdr *h3;
164 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
165 int closing, int tx_ring);
167 #define V3_ALIGNMENT (8)
169 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
171 #define BLK_PLUS_PRIV(sz_of_priv) \
172 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
174 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
175 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
176 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
177 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
178 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
179 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
180 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
183 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
184 struct packet_type *pt, struct net_device *orig_dev);
186 static void *packet_previous_frame(struct packet_sock *po,
187 struct packet_ring_buffer *rb,
189 static void packet_increment_head(struct packet_ring_buffer *buff);
190 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
191 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
192 struct packet_sock *);
193 static void prb_retire_current_block(struct tpacket_kbdq_core *,
194 struct packet_sock *, unsigned int status);
195 static int prb_queue_frozen(struct tpacket_kbdq_core *);
196 static void prb_open_block(struct tpacket_kbdq_core *,
197 struct tpacket_block_desc *);
198 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
199 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
200 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
201 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
202 struct tpacket3_hdr *);
203 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
204 struct tpacket3_hdr *);
205 static void packet_flush_mclist(struct sock *sk);
206 static u16 packet_pick_tx_queue(struct sk_buff *skb);
208 struct packet_skb_cb {
210 struct sockaddr_pkt pkt;
212 /* Trick: alias skb original length with
213 * ll.sll_family and ll.protocol in order
216 unsigned int origlen;
217 struct sockaddr_ll ll;
222 #define vio_le() virtio_legacy_is_little_endian()
224 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
226 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
227 #define GET_PBLOCK_DESC(x, bid) \
228 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
229 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
230 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
231 #define GET_NEXT_PRB_BLK_NUM(x) \
232 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
233 ((x)->kactive_blk_num+1) : 0)
235 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
236 static void __fanout_link(struct sock *sk, struct packet_sock *po);
238 static int packet_direct_xmit(struct sk_buff *skb)
240 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
243 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
245 struct net_device *dev;
248 dev = rcu_dereference(po->cached_dev);
256 static void packet_cached_dev_assign(struct packet_sock *po,
257 struct net_device *dev)
259 rcu_assign_pointer(po->cached_dev, dev);
262 static void packet_cached_dev_reset(struct packet_sock *po)
264 RCU_INIT_POINTER(po->cached_dev, NULL);
267 static bool packet_use_direct_xmit(const struct packet_sock *po)
269 return po->xmit == packet_direct_xmit;
272 static u16 packet_pick_tx_queue(struct sk_buff *skb)
274 struct net_device *dev = skb->dev;
275 const struct net_device_ops *ops = dev->netdev_ops;
276 int cpu = raw_smp_processor_id();
280 skb->sender_cpu = cpu + 1;
282 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
283 if (ops->ndo_select_queue) {
284 queue_index = ops->ndo_select_queue(dev, skb, NULL);
285 queue_index = netdev_cap_txqueue(dev, queue_index);
287 queue_index = netdev_pick_tx(dev, skb, NULL);
293 /* __register_prot_hook must be invoked through register_prot_hook
294 * or from a context in which asynchronous accesses to the packet
295 * socket is not possible (packet_create()).
297 static void __register_prot_hook(struct sock *sk)
299 struct packet_sock *po = pkt_sk(sk);
303 __fanout_link(sk, po);
305 dev_add_pack(&po->prot_hook);
312 static void register_prot_hook(struct sock *sk)
314 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
315 __register_prot_hook(sk);
318 /* If the sync parameter is true, we will temporarily drop
319 * the po->bind_lock and do a synchronize_net to make sure no
320 * asynchronous packet processing paths still refer to the elements
321 * of po->prot_hook. If the sync parameter is false, it is the
322 * callers responsibility to take care of this.
324 static void __unregister_prot_hook(struct sock *sk, bool sync)
326 struct packet_sock *po = pkt_sk(sk);
328 lockdep_assert_held_once(&po->bind_lock);
333 __fanout_unlink(sk, po);
335 __dev_remove_pack(&po->prot_hook);
340 spin_unlock(&po->bind_lock);
342 spin_lock(&po->bind_lock);
346 static void unregister_prot_hook(struct sock *sk, bool sync)
348 struct packet_sock *po = pkt_sk(sk);
351 __unregister_prot_hook(sk, sync);
354 static inline struct page * __pure pgv_to_page(void *addr)
356 if (is_vmalloc_addr(addr))
357 return vmalloc_to_page(addr);
358 return virt_to_page(addr);
361 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
363 union tpacket_uhdr h;
366 switch (po->tp_version) {
368 h.h1->tp_status = status;
369 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
372 h.h2->tp_status = status;
373 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
376 h.h3->tp_status = status;
377 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
380 WARN(1, "TPACKET version not supported.\n");
387 static int __packet_get_status(const struct packet_sock *po, void *frame)
389 union tpacket_uhdr h;
394 switch (po->tp_version) {
396 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
397 return h.h1->tp_status;
399 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
400 return h.h2->tp_status;
402 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
403 return h.h3->tp_status;
405 WARN(1, "TPACKET version not supported.\n");
411 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
414 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
417 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
418 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
419 return TP_STATUS_TS_RAW_HARDWARE;
421 if (ktime_to_timespec_cond(skb->tstamp, ts))
422 return TP_STATUS_TS_SOFTWARE;
427 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
430 union tpacket_uhdr h;
434 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
438 switch (po->tp_version) {
440 h.h1->tp_sec = ts.tv_sec;
441 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
444 h.h2->tp_sec = ts.tv_sec;
445 h.h2->tp_nsec = ts.tv_nsec;
448 h.h3->tp_sec = ts.tv_sec;
449 h.h3->tp_nsec = ts.tv_nsec;
452 WARN(1, "TPACKET version not supported.\n");
456 /* one flush is safe, as both fields always lie on the same cacheline */
457 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
463 static void *packet_lookup_frame(const struct packet_sock *po,
464 const struct packet_ring_buffer *rb,
465 unsigned int position,
468 unsigned int pg_vec_pos, frame_offset;
469 union tpacket_uhdr h;
471 pg_vec_pos = position / rb->frames_per_block;
472 frame_offset = position % rb->frames_per_block;
474 h.raw = rb->pg_vec[pg_vec_pos].buffer +
475 (frame_offset * rb->frame_size);
477 if (status != __packet_get_status(po, h.raw))
483 static void *packet_current_frame(struct packet_sock *po,
484 struct packet_ring_buffer *rb,
487 return packet_lookup_frame(po, rb, rb->head, status);
490 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
492 del_timer_sync(&pkc->retire_blk_timer);
495 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
496 struct sk_buff_head *rb_queue)
498 struct tpacket_kbdq_core *pkc;
500 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
502 spin_lock_bh(&rb_queue->lock);
503 pkc->delete_blk_timer = 1;
504 spin_unlock_bh(&rb_queue->lock);
506 prb_del_retire_blk_timer(pkc);
509 static void prb_setup_retire_blk_timer(struct packet_sock *po)
511 struct tpacket_kbdq_core *pkc;
513 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
514 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
516 pkc->retire_blk_timer.expires = jiffies;
519 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
520 int blk_size_in_bytes)
522 struct net_device *dev;
523 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
524 struct ethtool_link_ksettings ecmd;
528 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
529 if (unlikely(!dev)) {
531 return DEFAULT_PRB_RETIRE_TOV;
533 err = __ethtool_get_link_ksettings(dev, &ecmd);
537 * If the link speed is so slow you don't really
538 * need to worry about perf anyways
540 if (ecmd.base.speed < SPEED_1000 ||
541 ecmd.base.speed == SPEED_UNKNOWN) {
542 return DEFAULT_PRB_RETIRE_TOV;
545 div = ecmd.base.speed / 1000;
549 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
561 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
562 union tpacket_req_u *req_u)
564 p1->feature_req_word = req_u->req3.tp_feature_req_word;
567 static void init_prb_bdqc(struct packet_sock *po,
568 struct packet_ring_buffer *rb,
570 union tpacket_req_u *req_u)
572 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
573 struct tpacket_block_desc *pbd;
575 memset(p1, 0x0, sizeof(*p1));
577 p1->knxt_seq_num = 1;
579 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
580 p1->pkblk_start = pg_vec[0].buffer;
581 p1->kblk_size = req_u->req3.tp_block_size;
582 p1->knum_blocks = req_u->req3.tp_block_nr;
583 p1->hdrlen = po->tp_hdrlen;
584 p1->version = po->tp_version;
585 p1->last_kactive_blk_num = 0;
586 po->stats.stats3.tp_freeze_q_cnt = 0;
587 if (req_u->req3.tp_retire_blk_tov)
588 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
590 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
591 req_u->req3.tp_block_size);
592 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
593 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
595 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
596 prb_init_ft_ops(p1, req_u);
597 prb_setup_retire_blk_timer(po);
598 prb_open_block(p1, pbd);
601 /* Do NOT update the last_blk_num first.
602 * Assumes sk_buff_head lock is held.
604 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
606 mod_timer(&pkc->retire_blk_timer,
607 jiffies + pkc->tov_in_jiffies);
608 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
613 * 1) We refresh the timer only when we open a block.
614 * By doing this we don't waste cycles refreshing the timer
615 * on packet-by-packet basis.
617 * With a 1MB block-size, on a 1Gbps line, it will take
618 * i) ~8 ms to fill a block + ii) memcpy etc.
619 * In this cut we are not accounting for the memcpy time.
621 * So, if the user sets the 'tmo' to 10ms then the timer
622 * will never fire while the block is still getting filled
623 * (which is what we want). However, the user could choose
624 * to close a block early and that's fine.
626 * But when the timer does fire, we check whether or not to refresh it.
627 * Since the tmo granularity is in msecs, it is not too expensive
628 * to refresh the timer, lets say every '8' msecs.
629 * Either the user can set the 'tmo' or we can derive it based on
630 * a) line-speed and b) block-size.
631 * prb_calc_retire_blk_tmo() calculates the tmo.
634 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
636 struct packet_sock *po =
637 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
638 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
640 struct tpacket_block_desc *pbd;
642 spin_lock(&po->sk.sk_receive_queue.lock);
644 frozen = prb_queue_frozen(pkc);
645 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
647 if (unlikely(pkc->delete_blk_timer))
650 /* We only need to plug the race when the block is partially filled.
652 * lock(); increment BLOCK_NUM_PKTS; unlock()
653 * copy_bits() is in progress ...
654 * timer fires on other cpu:
655 * we can't retire the current block because copy_bits
659 if (BLOCK_NUM_PKTS(pbd)) {
660 while (atomic_read(&pkc->blk_fill_in_prog)) {
661 /* Waiting for skb_copy_bits to finish... */
666 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
668 if (!BLOCK_NUM_PKTS(pbd)) {
669 /* An empty block. Just refresh the timer. */
672 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
673 if (!prb_dispatch_next_block(pkc, po))
678 /* Case 1. Queue was frozen because user-space was
681 if (prb_curr_blk_in_use(pbd)) {
683 * Ok, user-space is still behind.
684 * So just refresh the timer.
688 /* Case 2. queue was frozen,user-space caught up,
689 * now the link went idle && the timer fired.
690 * We don't have a block to close.So we open this
691 * block and restart the timer.
692 * opening a block thaws the queue,restarts timer
693 * Thawing/timer-refresh is a side effect.
695 prb_open_block(pkc, pbd);
702 _prb_refresh_rx_retire_blk_timer(pkc);
705 spin_unlock(&po->sk.sk_receive_queue.lock);
708 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
709 struct tpacket_block_desc *pbd1, __u32 status)
711 /* Flush everything minus the block header */
713 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
718 /* Skip the block header(we know header WILL fit in 4K) */
721 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
722 for (; start < end; start += PAGE_SIZE)
723 flush_dcache_page(pgv_to_page(start));
728 /* Now update the block status. */
730 BLOCK_STATUS(pbd1) = status;
732 /* Flush the block header */
734 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
736 flush_dcache_page(pgv_to_page(start));
746 * 2) Increment active_blk_num
748 * Note:We DONT refresh the timer on purpose.
749 * Because almost always the next block will be opened.
751 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
752 struct tpacket_block_desc *pbd1,
753 struct packet_sock *po, unsigned int stat)
755 __u32 status = TP_STATUS_USER | stat;
757 struct tpacket3_hdr *last_pkt;
758 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
759 struct sock *sk = &po->sk;
761 if (atomic_read(&po->tp_drops))
762 status |= TP_STATUS_LOSING;
764 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
765 last_pkt->tp_next_offset = 0;
767 /* Get the ts of the last pkt */
768 if (BLOCK_NUM_PKTS(pbd1)) {
769 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
770 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
772 /* Ok, we tmo'd - so get the current time.
774 * It shouldn't really happen as we don't close empty
775 * blocks. See prb_retire_rx_blk_timer_expired().
779 h1->ts_last_pkt.ts_sec = ts.tv_sec;
780 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
785 /* Flush the block */
786 prb_flush_block(pkc1, pbd1, status);
788 sk->sk_data_ready(sk);
790 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
793 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
795 pkc->reset_pending_on_curr_blk = 0;
799 * Side effect of opening a block:
801 * 1) prb_queue is thawed.
802 * 2) retire_blk_timer is refreshed.
805 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
806 struct tpacket_block_desc *pbd1)
809 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
813 /* We could have just memset this but we will lose the
814 * flexibility of making the priv area sticky
817 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
818 BLOCK_NUM_PKTS(pbd1) = 0;
819 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
823 h1->ts_first_pkt.ts_sec = ts.tv_sec;
824 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
826 pkc1->pkblk_start = (char *)pbd1;
827 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
829 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
832 pbd1->version = pkc1->version;
833 pkc1->prev = pkc1->nxt_offset;
834 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
836 prb_thaw_queue(pkc1);
837 _prb_refresh_rx_retire_blk_timer(pkc1);
843 * Queue freeze logic:
844 * 1) Assume tp_block_nr = 8 blocks.
845 * 2) At time 't0', user opens Rx ring.
846 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
847 * 4) user-space is either sleeping or processing block '0'.
848 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
849 * it will close block-7,loop around and try to fill block '0'.
851 * __packet_lookup_frame_in_block
852 * prb_retire_current_block()
853 * prb_dispatch_next_block()
854 * |->(BLOCK_STATUS == USER) evaluates to true
855 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
856 * 6) Now there are two cases:
857 * 6.1) Link goes idle right after the queue is frozen.
858 * But remember, the last open_block() refreshed the timer.
859 * When this timer expires,it will refresh itself so that we can
860 * re-open block-0 in near future.
861 * 6.2) Link is busy and keeps on receiving packets. This is a simple
862 * case and __packet_lookup_frame_in_block will check if block-0
863 * is free and can now be re-used.
865 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
866 struct packet_sock *po)
868 pkc->reset_pending_on_curr_blk = 1;
869 po->stats.stats3.tp_freeze_q_cnt++;
872 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
875 * If the next block is free then we will dispatch it
876 * and return a good offset.
877 * Else, we will freeze the queue.
878 * So, caller must check the return value.
880 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
881 struct packet_sock *po)
883 struct tpacket_block_desc *pbd;
887 /* 1. Get current block num */
888 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
890 /* 2. If this block is currently in_use then freeze the queue */
891 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
892 prb_freeze_queue(pkc, po);
898 * open this block and return the offset where the first packet
899 * needs to get stored.
901 prb_open_block(pkc, pbd);
902 return (void *)pkc->nxt_offset;
905 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
906 struct packet_sock *po, unsigned int status)
908 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
910 /* retire/close the current block */
911 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
913 * Plug the case where copy_bits() is in progress on
914 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
915 * have space to copy the pkt in the current block and
916 * called prb_retire_current_block()
918 * We don't need to worry about the TMO case because
919 * the timer-handler already handled this case.
921 if (!(status & TP_STATUS_BLK_TMO)) {
922 while (atomic_read(&pkc->blk_fill_in_prog)) {
923 /* Waiting for skb_copy_bits to finish... */
927 prb_close_block(pkc, pbd, po, status);
932 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
934 return TP_STATUS_USER & BLOCK_STATUS(pbd);
937 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
939 return pkc->reset_pending_on_curr_blk;
942 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
944 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
945 atomic_dec(&pkc->blk_fill_in_prog);
948 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
949 struct tpacket3_hdr *ppd)
951 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
954 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
955 struct tpacket3_hdr *ppd)
957 ppd->hv1.tp_rxhash = 0;
960 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
961 struct tpacket3_hdr *ppd)
963 if (skb_vlan_tag_present(pkc->skb)) {
964 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
965 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
966 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
968 ppd->hv1.tp_vlan_tci = 0;
969 ppd->hv1.tp_vlan_tpid = 0;
970 ppd->tp_status = TP_STATUS_AVAILABLE;
974 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
975 struct tpacket3_hdr *ppd)
977 ppd->hv1.tp_padding = 0;
978 prb_fill_vlan_info(pkc, ppd);
980 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
981 prb_fill_rxhash(pkc, ppd);
983 prb_clear_rxhash(pkc, ppd);
986 static void prb_fill_curr_block(char *curr,
987 struct tpacket_kbdq_core *pkc,
988 struct tpacket_block_desc *pbd,
991 struct tpacket3_hdr *ppd;
993 ppd = (struct tpacket3_hdr *)curr;
994 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
996 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
997 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
998 BLOCK_NUM_PKTS(pbd) += 1;
999 atomic_inc(&pkc->blk_fill_in_prog);
1000 prb_run_all_ft_ops(pkc, ppd);
1003 /* Assumes caller has the sk->rx_queue.lock */
1004 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1005 struct sk_buff *skb,
1009 struct tpacket_kbdq_core *pkc;
1010 struct tpacket_block_desc *pbd;
1013 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1014 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1016 /* Queue is frozen when user space is lagging behind */
1017 if (prb_queue_frozen(pkc)) {
1019 * Check if that last block which caused the queue to freeze,
1020 * is still in_use by user-space.
1022 if (prb_curr_blk_in_use(pbd)) {
1023 /* Can't record this packet */
1027 * Ok, the block was released by user-space.
1028 * Now let's open that block.
1029 * opening a block also thaws the queue.
1030 * Thawing is a side effect.
1032 prb_open_block(pkc, pbd);
1037 curr = pkc->nxt_offset;
1039 end = (char *)pbd + pkc->kblk_size;
1041 /* first try the current block */
1042 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1043 prb_fill_curr_block(curr, pkc, pbd, len);
1044 return (void *)curr;
1047 /* Ok, close the current block */
1048 prb_retire_current_block(pkc, po, 0);
1050 /* Now, try to dispatch the next block */
1051 curr = (char *)prb_dispatch_next_block(pkc, po);
1053 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1054 prb_fill_curr_block(curr, pkc, pbd, len);
1055 return (void *)curr;
1059 * No free blocks are available.user_space hasn't caught up yet.
1060 * Queue was just frozen and now this packet will get dropped.
1065 static void *packet_current_rx_frame(struct packet_sock *po,
1066 struct sk_buff *skb,
1067 int status, unsigned int len)
1070 switch (po->tp_version) {
1073 curr = packet_lookup_frame(po, &po->rx_ring,
1074 po->rx_ring.head, status);
1077 return __packet_lookup_frame_in_block(po, skb, len);
1079 WARN(1, "TPACKET version not supported\n");
1085 static void *prb_lookup_block(const struct packet_sock *po,
1086 const struct packet_ring_buffer *rb,
1090 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1091 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1093 if (status != BLOCK_STATUS(pbd))
1098 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1101 if (rb->prb_bdqc.kactive_blk_num)
1102 prev = rb->prb_bdqc.kactive_blk_num-1;
1104 prev = rb->prb_bdqc.knum_blocks-1;
1108 /* Assumes caller has held the rx_queue.lock */
1109 static void *__prb_previous_block(struct packet_sock *po,
1110 struct packet_ring_buffer *rb,
1113 unsigned int previous = prb_previous_blk_num(rb);
1114 return prb_lookup_block(po, rb, previous, status);
1117 static void *packet_previous_rx_frame(struct packet_sock *po,
1118 struct packet_ring_buffer *rb,
1121 if (po->tp_version <= TPACKET_V2)
1122 return packet_previous_frame(po, rb, status);
1124 return __prb_previous_block(po, rb, status);
1127 static void packet_increment_rx_head(struct packet_sock *po,
1128 struct packet_ring_buffer *rb)
1130 switch (po->tp_version) {
1133 return packet_increment_head(rb);
1136 WARN(1, "TPACKET version not supported.\n");
1142 static void *packet_previous_frame(struct packet_sock *po,
1143 struct packet_ring_buffer *rb,
1146 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1147 return packet_lookup_frame(po, rb, previous, status);
1150 static void packet_increment_head(struct packet_ring_buffer *buff)
1152 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1155 static void packet_inc_pending(struct packet_ring_buffer *rb)
1157 this_cpu_inc(*rb->pending_refcnt);
1160 static void packet_dec_pending(struct packet_ring_buffer *rb)
1162 this_cpu_dec(*rb->pending_refcnt);
1165 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1167 unsigned int refcnt = 0;
1170 /* We don't use pending refcount in rx_ring. */
1171 if (rb->pending_refcnt == NULL)
1174 for_each_possible_cpu(cpu)
1175 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1180 static int packet_alloc_pending(struct packet_sock *po)
1182 po->rx_ring.pending_refcnt = NULL;
1184 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1185 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1191 static void packet_free_pending(struct packet_sock *po)
1193 free_percpu(po->tx_ring.pending_refcnt);
1196 #define ROOM_POW_OFF 2
1197 #define ROOM_NONE 0x0
1198 #define ROOM_LOW 0x1
1199 #define ROOM_NORMAL 0x2
1201 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1205 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1206 idx = READ_ONCE(po->rx_ring.head);
1208 idx += len >> pow_off;
1211 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1214 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1218 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1219 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1221 idx += len >> pow_off;
1224 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1227 static int __packet_rcv_has_room(const struct packet_sock *po,
1228 const struct sk_buff *skb)
1230 const struct sock *sk = &po->sk;
1231 int ret = ROOM_NONE;
1233 if (po->prot_hook.func != tpacket_rcv) {
1234 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1235 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1236 - (skb ? skb->truesize : 0);
1238 if (avail > (rcvbuf >> ROOM_POW_OFF))
1246 if (po->tp_version == TPACKET_V3) {
1247 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1249 else if (__tpacket_v3_has_room(po, 0))
1252 if (__tpacket_has_room(po, ROOM_POW_OFF))
1254 else if (__tpacket_has_room(po, 0))
1261 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1265 ret = __packet_rcv_has_room(po, skb);
1266 pressure = ret != ROOM_NORMAL;
1268 if (READ_ONCE(po->pressure) != pressure)
1269 WRITE_ONCE(po->pressure, pressure);
1274 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1276 if (READ_ONCE(po->pressure) &&
1277 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1278 WRITE_ONCE(po->pressure, 0);
1281 static void packet_sock_destruct(struct sock *sk)
1283 skb_queue_purge(&sk->sk_error_queue);
1285 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1286 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1288 if (!sock_flag(sk, SOCK_DEAD)) {
1289 pr_err("Attempt to release alive packet socket: %p\n", sk);
1293 sk_refcnt_debug_dec(sk);
1296 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1301 rxhash = skb_get_hash(skb);
1302 for (i = 0; i < ROLLOVER_HLEN; i++)
1303 if (po->rollover->history[i] == rxhash)
1306 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1307 return count > (ROLLOVER_HLEN >> 1);
1310 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1311 struct sk_buff *skb,
1314 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1317 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1318 struct sk_buff *skb,
1321 unsigned int val = atomic_inc_return(&f->rr_cur);
1326 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1327 struct sk_buff *skb,
1330 return smp_processor_id() % num;
1333 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1334 struct sk_buff *skb,
1337 return prandom_u32_max(num);
1340 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1341 struct sk_buff *skb,
1342 unsigned int idx, bool try_self,
1345 struct packet_sock *po, *po_next, *po_skip = NULL;
1346 unsigned int i, j, room = ROOM_NONE;
1348 po = pkt_sk(f->arr[idx]);
1351 room = packet_rcv_has_room(po, skb);
1352 if (room == ROOM_NORMAL ||
1353 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1358 i = j = min_t(int, po->rollover->sock, num - 1);
1360 po_next = pkt_sk(f->arr[i]);
1361 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1362 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1364 po->rollover->sock = i;
1365 atomic_long_inc(&po->rollover->num);
1366 if (room == ROOM_LOW)
1367 atomic_long_inc(&po->rollover->num_huge);
1375 atomic_long_inc(&po->rollover->num_failed);
1379 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1380 struct sk_buff *skb,
1383 return skb_get_queue_mapping(skb) % num;
1386 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1387 struct sk_buff *skb,
1390 struct bpf_prog *prog;
1391 unsigned int ret = 0;
1394 prog = rcu_dereference(f->bpf_prog);
1396 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1402 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1404 return f->flags & (flag >> 8);
1407 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1408 struct packet_type *pt, struct net_device *orig_dev)
1410 struct packet_fanout *f = pt->af_packet_priv;
1411 unsigned int num = READ_ONCE(f->num_members);
1412 struct net *net = read_pnet(&f->net);
1413 struct packet_sock *po;
1416 if (!net_eq(dev_net(dev), net) || !num) {
1421 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1422 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1427 case PACKET_FANOUT_HASH:
1429 idx = fanout_demux_hash(f, skb, num);
1431 case PACKET_FANOUT_LB:
1432 idx = fanout_demux_lb(f, skb, num);
1434 case PACKET_FANOUT_CPU:
1435 idx = fanout_demux_cpu(f, skb, num);
1437 case PACKET_FANOUT_RND:
1438 idx = fanout_demux_rnd(f, skb, num);
1440 case PACKET_FANOUT_QM:
1441 idx = fanout_demux_qm(f, skb, num);
1443 case PACKET_FANOUT_ROLLOVER:
1444 idx = fanout_demux_rollover(f, skb, 0, false, num);
1446 case PACKET_FANOUT_CBPF:
1447 case PACKET_FANOUT_EBPF:
1448 idx = fanout_demux_bpf(f, skb, num);
1452 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1453 idx = fanout_demux_rollover(f, skb, idx, true, num);
1455 po = pkt_sk(f->arr[idx]);
1456 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1459 DEFINE_MUTEX(fanout_mutex);
1460 EXPORT_SYMBOL_GPL(fanout_mutex);
1461 static LIST_HEAD(fanout_list);
1462 static u16 fanout_next_id;
1464 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1466 struct packet_fanout *f = po->fanout;
1468 spin_lock(&f->lock);
1469 f->arr[f->num_members] = sk;
1472 if (f->num_members == 1)
1473 dev_add_pack(&f->prot_hook);
1474 spin_unlock(&f->lock);
1477 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1479 struct packet_fanout *f = po->fanout;
1482 spin_lock(&f->lock);
1483 for (i = 0; i < f->num_members; i++) {
1484 if (f->arr[i] == sk)
1487 BUG_ON(i >= f->num_members);
1488 f->arr[i] = f->arr[f->num_members - 1];
1490 if (f->num_members == 0)
1491 __dev_remove_pack(&f->prot_hook);
1492 spin_unlock(&f->lock);
1495 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1497 if (sk->sk_family != PF_PACKET)
1500 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1503 static void fanout_init_data(struct packet_fanout *f)
1506 case PACKET_FANOUT_LB:
1507 atomic_set(&f->rr_cur, 0);
1509 case PACKET_FANOUT_CBPF:
1510 case PACKET_FANOUT_EBPF:
1511 RCU_INIT_POINTER(f->bpf_prog, NULL);
1516 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1518 struct bpf_prog *old;
1520 spin_lock(&f->lock);
1521 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1522 rcu_assign_pointer(f->bpf_prog, new);
1523 spin_unlock(&f->lock);
1527 bpf_prog_destroy(old);
1531 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1534 struct bpf_prog *new;
1535 struct sock_fprog fprog;
1538 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1540 if (len != sizeof(fprog))
1542 if (copy_from_user(&fprog, data, len))
1545 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1549 __fanout_set_data_bpf(po->fanout, new);
1553 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1556 struct bpf_prog *new;
1559 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1561 if (len != sizeof(fd))
1563 if (copy_from_user(&fd, data, len))
1566 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1568 return PTR_ERR(new);
1570 __fanout_set_data_bpf(po->fanout, new);
1574 static int fanout_set_data(struct packet_sock *po, char __user *data,
1577 switch (po->fanout->type) {
1578 case PACKET_FANOUT_CBPF:
1579 return fanout_set_data_cbpf(po, data, len);
1580 case PACKET_FANOUT_EBPF:
1581 return fanout_set_data_ebpf(po, data, len);
1587 static void fanout_release_data(struct packet_fanout *f)
1590 case PACKET_FANOUT_CBPF:
1591 case PACKET_FANOUT_EBPF:
1592 __fanout_set_data_bpf(f, NULL);
1596 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1598 struct packet_fanout *f;
1600 list_for_each_entry(f, &fanout_list, list) {
1601 if (f->id == candidate_id &&
1602 read_pnet(&f->net) == sock_net(sk)) {
1609 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1611 u16 id = fanout_next_id;
1614 if (__fanout_id_is_free(sk, id)) {
1616 fanout_next_id = id + 1;
1621 } while (id != fanout_next_id);
1626 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1628 struct packet_rollover *rollover = NULL;
1629 struct packet_sock *po = pkt_sk(sk);
1630 struct packet_fanout *f, *match;
1631 u8 type = type_flags & 0xff;
1632 u8 flags = type_flags >> 8;
1636 case PACKET_FANOUT_ROLLOVER:
1637 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1639 case PACKET_FANOUT_HASH:
1640 case PACKET_FANOUT_LB:
1641 case PACKET_FANOUT_CPU:
1642 case PACKET_FANOUT_RND:
1643 case PACKET_FANOUT_QM:
1644 case PACKET_FANOUT_CBPF:
1645 case PACKET_FANOUT_EBPF:
1651 mutex_lock(&fanout_mutex);
1657 if (type == PACKET_FANOUT_ROLLOVER ||
1658 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1660 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1663 atomic_long_set(&rollover->num, 0);
1664 atomic_long_set(&rollover->num_huge, 0);
1665 atomic_long_set(&rollover->num_failed, 0);
1668 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1673 if (!fanout_find_new_id(sk, &id)) {
1677 /* ephemeral flag for the first socket in the group: drop it */
1678 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1682 list_for_each_entry(f, &fanout_list, list) {
1684 read_pnet(&f->net) == sock_net(sk)) {
1690 if (match && match->flags != flags)
1694 match = kzalloc(sizeof(*match), GFP_KERNEL);
1697 write_pnet(&match->net, sock_net(sk));
1700 match->flags = flags;
1701 INIT_LIST_HEAD(&match->list);
1702 spin_lock_init(&match->lock);
1703 refcount_set(&match->sk_ref, 0);
1704 fanout_init_data(match);
1705 match->prot_hook.type = po->prot_hook.type;
1706 match->prot_hook.dev = po->prot_hook.dev;
1707 match->prot_hook.func = packet_rcv_fanout;
1708 match->prot_hook.af_packet_priv = match;
1709 match->prot_hook.id_match = match_fanout_group;
1710 list_add(&match->list, &fanout_list);
1714 spin_lock(&po->bind_lock);
1716 match->type == type &&
1717 match->prot_hook.type == po->prot_hook.type &&
1718 match->prot_hook.dev == po->prot_hook.dev) {
1720 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1721 __dev_remove_pack(&po->prot_hook);
1723 po->rollover = rollover;
1725 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1726 __fanout_link(sk, po);
1730 spin_unlock(&po->bind_lock);
1732 if (err && !refcount_read(&match->sk_ref)) {
1733 list_del(&match->list);
1739 mutex_unlock(&fanout_mutex);
1743 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1744 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1745 * It is the responsibility of the caller to call fanout_release_data() and
1746 * free the returned packet_fanout (after synchronize_net())
1748 static struct packet_fanout *fanout_release(struct sock *sk)
1750 struct packet_sock *po = pkt_sk(sk);
1751 struct packet_fanout *f;
1753 mutex_lock(&fanout_mutex);
1758 if (refcount_dec_and_test(&f->sk_ref))
1763 mutex_unlock(&fanout_mutex);
1768 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1769 struct sk_buff *skb)
1771 /* Earlier code assumed this would be a VLAN pkt, double-check
1772 * this now that we have the actual packet in hand. We can only
1773 * do this check on Ethernet devices.
1775 if (unlikely(dev->type != ARPHRD_ETHER))
1778 skb_reset_mac_header(skb);
1779 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1782 static const struct proto_ops packet_ops;
1784 static const struct proto_ops packet_ops_spkt;
1786 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1787 struct packet_type *pt, struct net_device *orig_dev)
1790 struct sockaddr_pkt *spkt;
1793 * When we registered the protocol we saved the socket in the data
1794 * field for just this event.
1797 sk = pt->af_packet_priv;
1800 * Yank back the headers [hope the device set this
1801 * right or kerboom...]
1803 * Incoming packets have ll header pulled,
1806 * For outgoing ones skb->data == skb_mac_header(skb)
1807 * so that this procedure is noop.
1810 if (skb->pkt_type == PACKET_LOOPBACK)
1813 if (!net_eq(dev_net(dev), sock_net(sk)))
1816 skb = skb_share_check(skb, GFP_ATOMIC);
1820 /* drop any routing info */
1823 /* drop conntrack reference */
1826 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1828 skb_push(skb, skb->data - skb_mac_header(skb));
1831 * The SOCK_PACKET socket receives _all_ frames.
1834 spkt->spkt_family = dev->type;
1835 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1836 spkt->spkt_protocol = skb->protocol;
1839 * Charge the memory to the socket. This is done specifically
1840 * to prevent sockets using all the memory up.
1843 if (sock_queue_rcv_skb(sk, skb) == 0)
1852 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1854 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1855 sock->type == SOCK_RAW) {
1856 skb_reset_mac_header(skb);
1857 skb->protocol = dev_parse_header_protocol(skb);
1860 skb_probe_transport_header(skb);
1864 * Output a raw packet to a device layer. This bypasses all the other
1865 * protocol layers and you must therefore supply it with a complete frame
1868 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1871 struct sock *sk = sock->sk;
1872 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1873 struct sk_buff *skb = NULL;
1874 struct net_device *dev;
1875 struct sockcm_cookie sockc;
1881 * Get and verify the address.
1885 if (msg->msg_namelen < sizeof(struct sockaddr))
1887 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1888 proto = saddr->spkt_protocol;
1890 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1893 * Find the device first to size check it
1896 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1899 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1905 if (!(dev->flags & IFF_UP))
1909 * You may not queue a frame bigger than the mtu. This is the lowest level
1910 * raw protocol and you must do your own fragmentation at this level.
1913 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1914 if (!netif_supports_nofcs(dev)) {
1915 err = -EPROTONOSUPPORT;
1918 extra_len = 4; /* We're doing our own CRC */
1922 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1926 size_t reserved = LL_RESERVED_SPACE(dev);
1927 int tlen = dev->needed_tailroom;
1928 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1931 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1934 /* FIXME: Save some space for broken drivers that write a hard
1935 * header at transmission time by themselves. PPP is the notable
1936 * one here. This should really be fixed at the driver level.
1938 skb_reserve(skb, reserved);
1939 skb_reset_network_header(skb);
1941 /* Try to align data part correctly */
1946 skb_reset_network_header(skb);
1948 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1954 if (!dev_validate_header(dev, skb->data, len)) {
1958 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1959 !packet_extra_vlan_len_allowed(dev, skb)) {
1964 sockcm_init(&sockc, sk);
1965 if (msg->msg_controllen) {
1966 err = sock_cmsg_send(sk, msg, &sockc);
1971 skb->protocol = proto;
1973 skb->priority = sk->sk_priority;
1974 skb->mark = sk->sk_mark;
1975 skb->tstamp = sockc.transmit_time;
1977 skb_setup_tx_timestamp(skb, sockc.tsflags);
1979 if (unlikely(extra_len == 4))
1982 packet_parse_headers(skb, sock);
1984 dev_queue_xmit(skb);
1995 static unsigned int run_filter(struct sk_buff *skb,
1996 const struct sock *sk,
1999 struct sk_filter *filter;
2002 filter = rcu_dereference(sk->sk_filter);
2004 res = bpf_prog_run_clear_cb(filter->prog, skb);
2010 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2013 struct virtio_net_hdr vnet_hdr;
2015 if (*len < sizeof(vnet_hdr))
2017 *len -= sizeof(vnet_hdr);
2019 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2022 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2026 * This function makes lazy skb cloning in hope that most of packets
2027 * are discarded by BPF.
2029 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2030 * and skb->cb are mangled. It works because (and until) packets
2031 * falling here are owned by current CPU. Output packets are cloned
2032 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2033 * sequencially, so that if we return skb to original state on exit,
2034 * we will not harm anyone.
2037 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2038 struct packet_type *pt, struct net_device *orig_dev)
2041 struct sockaddr_ll *sll;
2042 struct packet_sock *po;
2043 u8 *skb_head = skb->data;
2044 int skb_len = skb->len;
2045 unsigned int snaplen, res;
2046 bool is_drop_n_account = false;
2048 if (skb->pkt_type == PACKET_LOOPBACK)
2051 sk = pt->af_packet_priv;
2054 if (!net_eq(dev_net(dev), sock_net(sk)))
2059 if (dev->header_ops) {
2060 /* The device has an explicit notion of ll header,
2061 * exported to higher levels.
2063 * Otherwise, the device hides details of its frame
2064 * structure, so that corresponding packet head is
2065 * never delivered to user.
2067 if (sk->sk_type != SOCK_DGRAM)
2068 skb_push(skb, skb->data - skb_mac_header(skb));
2069 else if (skb->pkt_type == PACKET_OUTGOING) {
2070 /* Special case: outgoing packets have ll header at head */
2071 skb_pull(skb, skb_network_offset(skb));
2077 res = run_filter(skb, sk, snaplen);
2079 goto drop_n_restore;
2083 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2086 if (skb_shared(skb)) {
2087 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2091 if (skb_head != skb->data) {
2092 skb->data = skb_head;
2099 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2101 sll = &PACKET_SKB_CB(skb)->sa.ll;
2102 sll->sll_hatype = dev->type;
2103 sll->sll_pkttype = skb->pkt_type;
2104 if (unlikely(po->origdev))
2105 sll->sll_ifindex = orig_dev->ifindex;
2107 sll->sll_ifindex = dev->ifindex;
2109 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2111 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2112 * Use their space for storing the original skb length.
2114 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2116 if (pskb_trim(skb, snaplen))
2119 skb_set_owner_r(skb, sk);
2123 /* drop conntrack reference */
2126 spin_lock(&sk->sk_receive_queue.lock);
2127 po->stats.stats1.tp_packets++;
2128 sock_skb_set_dropcount(sk, skb);
2129 __skb_queue_tail(&sk->sk_receive_queue, skb);
2130 spin_unlock(&sk->sk_receive_queue.lock);
2131 sk->sk_data_ready(sk);
2135 is_drop_n_account = true;
2136 atomic_inc(&po->tp_drops);
2137 atomic_inc(&sk->sk_drops);
2140 if (skb_head != skb->data && skb_shared(skb)) {
2141 skb->data = skb_head;
2145 if (!is_drop_n_account)
2152 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2153 struct packet_type *pt, struct net_device *orig_dev)
2156 struct packet_sock *po;
2157 struct sockaddr_ll *sll;
2158 union tpacket_uhdr h;
2159 u8 *skb_head = skb->data;
2160 int skb_len = skb->len;
2161 unsigned int snaplen, res;
2162 unsigned long status = TP_STATUS_USER;
2163 unsigned short macoff, netoff, hdrlen;
2164 struct sk_buff *copy_skb = NULL;
2167 bool is_drop_n_account = false;
2168 bool do_vnet = false;
2170 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2171 * We may add members to them until current aligned size without forcing
2172 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2174 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2175 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2177 if (skb->pkt_type == PACKET_LOOPBACK)
2180 sk = pt->af_packet_priv;
2183 if (!net_eq(dev_net(dev), sock_net(sk)))
2186 if (dev->header_ops) {
2187 if (sk->sk_type != SOCK_DGRAM)
2188 skb_push(skb, skb->data - skb_mac_header(skb));
2189 else if (skb->pkt_type == PACKET_OUTGOING) {
2190 /* Special case: outgoing packets have ll header at head */
2191 skb_pull(skb, skb_network_offset(skb));
2197 res = run_filter(skb, sk, snaplen);
2199 goto drop_n_restore;
2201 /* If we are flooded, just give up */
2202 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2203 atomic_inc(&po->tp_drops);
2204 goto drop_n_restore;
2207 if (skb->ip_summed == CHECKSUM_PARTIAL)
2208 status |= TP_STATUS_CSUMNOTREADY;
2209 else if (skb->pkt_type != PACKET_OUTGOING &&
2210 (skb->ip_summed == CHECKSUM_COMPLETE ||
2211 skb_csum_unnecessary(skb)))
2212 status |= TP_STATUS_CSUM_VALID;
2217 if (sk->sk_type == SOCK_DGRAM) {
2218 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2221 unsigned int maclen = skb_network_offset(skb);
2222 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2223 (maclen < 16 ? 16 : maclen)) +
2225 if (po->has_vnet_hdr) {
2226 netoff += sizeof(struct virtio_net_hdr);
2229 macoff = netoff - maclen;
2231 if (po->tp_version <= TPACKET_V2) {
2232 if (macoff + snaplen > po->rx_ring.frame_size) {
2233 if (po->copy_thresh &&
2234 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2235 if (skb_shared(skb)) {
2236 copy_skb = skb_clone(skb, GFP_ATOMIC);
2238 copy_skb = skb_get(skb);
2239 skb_head = skb->data;
2242 skb_set_owner_r(copy_skb, sk);
2244 snaplen = po->rx_ring.frame_size - macoff;
2245 if ((int)snaplen < 0) {
2250 } else if (unlikely(macoff + snaplen >
2251 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2254 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2255 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2256 snaplen, nval, macoff);
2258 if (unlikely((int)snaplen < 0)) {
2260 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2264 spin_lock(&sk->sk_receive_queue.lock);
2265 h.raw = packet_current_rx_frame(po, skb,
2266 TP_STATUS_KERNEL, (macoff+snaplen));
2268 goto drop_n_account;
2269 if (po->tp_version <= TPACKET_V2) {
2270 packet_increment_rx_head(po, &po->rx_ring);
2272 * LOSING will be reported till you read the stats,
2273 * because it's COR - Clear On Read.
2274 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2277 if (atomic_read(&po->tp_drops))
2278 status |= TP_STATUS_LOSING;
2282 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2283 sizeof(struct virtio_net_hdr),
2285 goto drop_n_account;
2287 po->stats.stats1.tp_packets++;
2289 status |= TP_STATUS_COPY;
2290 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2292 spin_unlock(&sk->sk_receive_queue.lock);
2294 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2296 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2297 getnstimeofday(&ts);
2299 status |= ts_status;
2301 switch (po->tp_version) {
2303 h.h1->tp_len = skb->len;
2304 h.h1->tp_snaplen = snaplen;
2305 h.h1->tp_mac = macoff;
2306 h.h1->tp_net = netoff;
2307 h.h1->tp_sec = ts.tv_sec;
2308 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2309 hdrlen = sizeof(*h.h1);
2312 h.h2->tp_len = skb->len;
2313 h.h2->tp_snaplen = snaplen;
2314 h.h2->tp_mac = macoff;
2315 h.h2->tp_net = netoff;
2316 h.h2->tp_sec = ts.tv_sec;
2317 h.h2->tp_nsec = ts.tv_nsec;
2318 if (skb_vlan_tag_present(skb)) {
2319 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2320 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2321 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2323 h.h2->tp_vlan_tci = 0;
2324 h.h2->tp_vlan_tpid = 0;
2326 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2327 hdrlen = sizeof(*h.h2);
2330 /* tp_nxt_offset,vlan are already populated above.
2331 * So DONT clear those fields here
2333 h.h3->tp_status |= status;
2334 h.h3->tp_len = skb->len;
2335 h.h3->tp_snaplen = snaplen;
2336 h.h3->tp_mac = macoff;
2337 h.h3->tp_net = netoff;
2338 h.h3->tp_sec = ts.tv_sec;
2339 h.h3->tp_nsec = ts.tv_nsec;
2340 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2341 hdrlen = sizeof(*h.h3);
2347 sll = h.raw + TPACKET_ALIGN(hdrlen);
2348 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2349 sll->sll_family = AF_PACKET;
2350 sll->sll_hatype = dev->type;
2351 sll->sll_protocol = skb->protocol;
2352 sll->sll_pkttype = skb->pkt_type;
2353 if (unlikely(po->origdev))
2354 sll->sll_ifindex = orig_dev->ifindex;
2356 sll->sll_ifindex = dev->ifindex;
2360 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2361 if (po->tp_version <= TPACKET_V2) {
2364 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2367 for (start = h.raw; start < end; start += PAGE_SIZE)
2368 flush_dcache_page(pgv_to_page(start));
2373 if (po->tp_version <= TPACKET_V2) {
2374 __packet_set_status(po, h.raw, status);
2375 sk->sk_data_ready(sk);
2377 prb_clear_blk_fill_status(&po->rx_ring);
2381 if (skb_head != skb->data && skb_shared(skb)) {
2382 skb->data = skb_head;
2386 if (!is_drop_n_account)
2393 spin_unlock(&sk->sk_receive_queue.lock);
2394 atomic_inc(&po->tp_drops);
2395 is_drop_n_account = true;
2397 sk->sk_data_ready(sk);
2398 kfree_skb(copy_skb);
2399 goto drop_n_restore;
2402 static void tpacket_destruct_skb(struct sk_buff *skb)
2404 struct packet_sock *po = pkt_sk(skb->sk);
2406 if (likely(po->tx_ring.pg_vec)) {
2410 ph = skb_zcopy_get_nouarg(skb);
2411 packet_dec_pending(&po->tx_ring);
2413 ts = __packet_set_timestamp(po, ph, skb);
2414 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2416 if (!packet_read_pending(&po->tx_ring))
2417 complete(&po->skb_completion);
2423 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2425 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2426 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2427 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2428 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2429 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2430 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2431 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2433 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2439 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2440 struct virtio_net_hdr *vnet_hdr)
2442 if (*len < sizeof(*vnet_hdr))
2444 *len -= sizeof(*vnet_hdr);
2446 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2449 return __packet_snd_vnet_parse(vnet_hdr, *len);
2452 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2453 void *frame, struct net_device *dev, void *data, int tp_len,
2454 __be16 proto, unsigned char *addr, int hlen, int copylen,
2455 const struct sockcm_cookie *sockc)
2457 union tpacket_uhdr ph;
2458 int to_write, offset, len, nr_frags, len_max;
2459 struct socket *sock = po->sk.sk_socket;
2465 skb->protocol = proto;
2467 skb->priority = po->sk.sk_priority;
2468 skb->mark = po->sk.sk_mark;
2469 skb->tstamp = sockc->transmit_time;
2470 skb_setup_tx_timestamp(skb, sockc->tsflags);
2471 skb_zcopy_set_nouarg(skb, ph.raw);
2473 skb_reserve(skb, hlen);
2474 skb_reset_network_header(skb);
2478 if (sock->type == SOCK_DGRAM) {
2479 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2481 if (unlikely(err < 0))
2483 } else if (copylen) {
2484 int hdrlen = min_t(int, copylen, tp_len);
2486 skb_push(skb, dev->hard_header_len);
2487 skb_put(skb, copylen - dev->hard_header_len);
2488 err = skb_store_bits(skb, 0, data, hdrlen);
2491 if (!dev_validate_header(dev, skb->data, hdrlen))
2498 offset = offset_in_page(data);
2499 len_max = PAGE_SIZE - offset;
2500 len = ((to_write > len_max) ? len_max : to_write);
2502 skb->data_len = to_write;
2503 skb->len += to_write;
2504 skb->truesize += to_write;
2505 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2507 while (likely(to_write)) {
2508 nr_frags = skb_shinfo(skb)->nr_frags;
2510 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2511 pr_err("Packet exceed the number of skb frags(%lu)\n",
2516 page = pgv_to_page(data);
2518 flush_dcache_page(page);
2520 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2523 len_max = PAGE_SIZE;
2524 len = ((to_write > len_max) ? len_max : to_write);
2527 packet_parse_headers(skb, sock);
2532 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2533 int size_max, void **data)
2535 union tpacket_uhdr ph;
2540 switch (po->tp_version) {
2542 if (ph.h3->tp_next_offset != 0) {
2543 pr_warn_once("variable sized slot not supported");
2546 tp_len = ph.h3->tp_len;
2549 tp_len = ph.h2->tp_len;
2552 tp_len = ph.h1->tp_len;
2555 if (unlikely(tp_len > size_max)) {
2556 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2560 if (unlikely(po->tp_tx_has_off)) {
2561 int off_min, off_max;
2563 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2564 off_max = po->tx_ring.frame_size - tp_len;
2565 if (po->sk.sk_type == SOCK_DGRAM) {
2566 switch (po->tp_version) {
2568 off = ph.h3->tp_net;
2571 off = ph.h2->tp_net;
2574 off = ph.h1->tp_net;
2578 switch (po->tp_version) {
2580 off = ph.h3->tp_mac;
2583 off = ph.h2->tp_mac;
2586 off = ph.h1->tp_mac;
2590 if (unlikely((off < off_min) || (off_max < off)))
2593 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2596 *data = frame + off;
2600 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2602 struct sk_buff *skb = NULL;
2603 struct net_device *dev;
2604 struct virtio_net_hdr *vnet_hdr = NULL;
2605 struct sockcm_cookie sockc;
2607 int err, reserve = 0;
2609 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2610 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2611 unsigned char *addr = NULL;
2612 int tp_len, size_max;
2615 int status = TP_STATUS_AVAILABLE;
2616 int hlen, tlen, copylen = 0;
2619 mutex_lock(&po->pg_vec_lock);
2621 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2622 * we need to confirm it under protection of pg_vec_lock.
2624 if (unlikely(!po->tx_ring.pg_vec)) {
2628 if (likely(saddr == NULL)) {
2629 dev = packet_cached_dev_get(po);
2633 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2635 if (msg->msg_namelen < (saddr->sll_halen
2636 + offsetof(struct sockaddr_ll,
2639 proto = saddr->sll_protocol;
2640 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2641 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2642 if (dev && msg->msg_namelen < dev->addr_len +
2643 offsetof(struct sockaddr_ll, sll_addr))
2645 addr = saddr->sll_addr;
2650 if (unlikely(dev == NULL))
2653 if (unlikely(!(dev->flags & IFF_UP)))
2656 sockcm_init(&sockc, &po->sk);
2657 if (msg->msg_controllen) {
2658 err = sock_cmsg_send(&po->sk, msg, &sockc);
2663 if (po->sk.sk_socket->type == SOCK_RAW)
2664 reserve = dev->hard_header_len;
2665 size_max = po->tx_ring.frame_size
2666 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2668 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2669 size_max = dev->mtu + reserve + VLAN_HLEN;
2671 reinit_completion(&po->skb_completion);
2674 ph = packet_current_frame(po, &po->tx_ring,
2675 TP_STATUS_SEND_REQUEST);
2676 if (unlikely(ph == NULL)) {
2677 if (need_wait && skb) {
2678 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2679 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2681 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2685 /* check for additional frames */
2690 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2694 status = TP_STATUS_SEND_REQUEST;
2695 hlen = LL_RESERVED_SPACE(dev);
2696 tlen = dev->needed_tailroom;
2697 if (po->has_vnet_hdr) {
2699 data += sizeof(*vnet_hdr);
2700 tp_len -= sizeof(*vnet_hdr);
2702 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2706 copylen = __virtio16_to_cpu(vio_le(),
2709 copylen = max_t(int, copylen, dev->hard_header_len);
2710 skb = sock_alloc_send_skb(&po->sk,
2711 hlen + tlen + sizeof(struct sockaddr_ll) +
2712 (copylen - dev->hard_header_len),
2715 if (unlikely(skb == NULL)) {
2716 /* we assume the socket was initially writeable ... */
2717 if (likely(len_sum > 0))
2721 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2722 addr, hlen, copylen, &sockc);
2723 if (likely(tp_len >= 0) &&
2724 tp_len > dev->mtu + reserve &&
2725 !po->has_vnet_hdr &&
2726 !packet_extra_vlan_len_allowed(dev, skb))
2729 if (unlikely(tp_len < 0)) {
2732 __packet_set_status(po, ph,
2733 TP_STATUS_AVAILABLE);
2734 packet_increment_head(&po->tx_ring);
2738 status = TP_STATUS_WRONG_FORMAT;
2744 if (po->has_vnet_hdr) {
2745 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2749 virtio_net_hdr_set_proto(skb, vnet_hdr);
2752 skb->destructor = tpacket_destruct_skb;
2753 __packet_set_status(po, ph, TP_STATUS_SENDING);
2754 packet_inc_pending(&po->tx_ring);
2756 status = TP_STATUS_SEND_REQUEST;
2757 err = po->xmit(skb);
2758 if (unlikely(err > 0)) {
2759 err = net_xmit_errno(err);
2760 if (err && __packet_get_status(po, ph) ==
2761 TP_STATUS_AVAILABLE) {
2762 /* skb was destructed already */
2767 * skb was dropped but not destructed yet;
2768 * let's treat it like congestion or err < 0
2772 packet_increment_head(&po->tx_ring);
2774 } while (likely((ph != NULL) ||
2775 /* Note: packet_read_pending() might be slow if we have
2776 * to call it as it's per_cpu variable, but in fast-path
2777 * we already short-circuit the loop with the first
2778 * condition, and luckily don't have to go that path
2781 (need_wait && packet_read_pending(&po->tx_ring))));
2787 __packet_set_status(po, ph, status);
2792 mutex_unlock(&po->pg_vec_lock);
2796 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2797 size_t reserve, size_t len,
2798 size_t linear, int noblock,
2801 struct sk_buff *skb;
2803 /* Under a page? Don't bother with paged skb. */
2804 if (prepad + len < PAGE_SIZE || !linear)
2807 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2812 skb_reserve(skb, reserve);
2813 skb_put(skb, linear);
2814 skb->data_len = len - linear;
2815 skb->len += len - linear;
2820 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2822 struct sock *sk = sock->sk;
2823 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2824 struct sk_buff *skb;
2825 struct net_device *dev;
2827 unsigned char *addr = NULL;
2828 int err, reserve = 0;
2829 struct sockcm_cookie sockc;
2830 struct virtio_net_hdr vnet_hdr = { 0 };
2832 struct packet_sock *po = pkt_sk(sk);
2833 bool has_vnet_hdr = false;
2834 int hlen, tlen, linear;
2838 * Get and verify the address.
2841 if (likely(saddr == NULL)) {
2842 dev = packet_cached_dev_get(po);
2846 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2848 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2850 proto = saddr->sll_protocol;
2851 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2852 if (sock->type == SOCK_DGRAM) {
2853 if (dev && msg->msg_namelen < dev->addr_len +
2854 offsetof(struct sockaddr_ll, sll_addr))
2856 addr = saddr->sll_addr;
2861 if (unlikely(dev == NULL))
2864 if (unlikely(!(dev->flags & IFF_UP)))
2867 sockcm_init(&sockc, sk);
2868 sockc.mark = sk->sk_mark;
2869 if (msg->msg_controllen) {
2870 err = sock_cmsg_send(sk, msg, &sockc);
2875 if (sock->type == SOCK_RAW)
2876 reserve = dev->hard_header_len;
2877 if (po->has_vnet_hdr) {
2878 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2881 has_vnet_hdr = true;
2884 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2885 if (!netif_supports_nofcs(dev)) {
2886 err = -EPROTONOSUPPORT;
2889 extra_len = 4; /* We're doing our own CRC */
2893 if (!vnet_hdr.gso_type &&
2894 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2898 hlen = LL_RESERVED_SPACE(dev);
2899 tlen = dev->needed_tailroom;
2900 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2901 linear = max(linear, min_t(int, len, dev->hard_header_len));
2902 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2903 msg->msg_flags & MSG_DONTWAIT, &err);
2907 skb_reset_network_header(skb);
2910 if (sock->type == SOCK_DGRAM) {
2911 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2912 if (unlikely(offset < 0))
2914 } else if (reserve) {
2915 skb_reserve(skb, -reserve);
2916 if (len < reserve + sizeof(struct ipv6hdr) &&
2917 dev->min_header_len != dev->hard_header_len)
2918 skb_reset_network_header(skb);
2921 /* Returns -EFAULT on error */
2922 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2926 if (sock->type == SOCK_RAW &&
2927 !dev_validate_header(dev, skb->data, len)) {
2932 skb_setup_tx_timestamp(skb, sockc.tsflags);
2934 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2935 !packet_extra_vlan_len_allowed(dev, skb)) {
2940 skb->protocol = proto;
2942 skb->priority = sk->sk_priority;
2943 skb->mark = sockc.mark;
2944 skb->tstamp = sockc.transmit_time;
2947 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2950 len += sizeof(vnet_hdr);
2951 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2954 packet_parse_headers(skb, sock);
2956 if (unlikely(extra_len == 4))
2959 err = po->xmit(skb);
2960 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2976 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2978 struct sock *sk = sock->sk;
2979 struct packet_sock *po = pkt_sk(sk);
2981 if (po->tx_ring.pg_vec)
2982 return tpacket_snd(po, msg);
2984 return packet_snd(sock, msg, len);
2988 * Close a PACKET socket. This is fairly simple. We immediately go
2989 * to 'closed' state and remove our protocol entry in the device list.
2992 static int packet_release(struct socket *sock)
2994 struct sock *sk = sock->sk;
2995 struct packet_sock *po;
2996 struct packet_fanout *f;
2998 union tpacket_req_u req_u;
3006 mutex_lock(&net->packet.sklist_lock);
3007 sk_del_node_init_rcu(sk);
3008 mutex_unlock(&net->packet.sklist_lock);
3011 sock_prot_inuse_add(net, sk->sk_prot, -1);
3014 spin_lock(&po->bind_lock);
3015 unregister_prot_hook(sk, false);
3016 packet_cached_dev_reset(po);
3018 if (po->prot_hook.dev) {
3019 dev_put(po->prot_hook.dev);
3020 po->prot_hook.dev = NULL;
3022 spin_unlock(&po->bind_lock);
3024 packet_flush_mclist(sk);
3027 if (po->rx_ring.pg_vec) {
3028 memset(&req_u, 0, sizeof(req_u));
3029 packet_set_ring(sk, &req_u, 1, 0);
3032 if (po->tx_ring.pg_vec) {
3033 memset(&req_u, 0, sizeof(req_u));
3034 packet_set_ring(sk, &req_u, 1, 1);
3038 f = fanout_release(sk);
3042 kfree(po->rollover);
3044 fanout_release_data(f);
3048 * Now the socket is dead. No more input will appear.
3055 skb_queue_purge(&sk->sk_receive_queue);
3056 packet_free_pending(po);
3057 sk_refcnt_debug_release(sk);
3064 * Attach a packet hook.
3067 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3070 struct packet_sock *po = pkt_sk(sk);
3071 struct net_device *dev_curr;
3074 struct net_device *dev = NULL;
3076 bool unlisted = false;
3079 spin_lock(&po->bind_lock);
3088 dev = dev_get_by_name_rcu(sock_net(sk), name);
3093 } else if (ifindex) {
3094 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3104 proto_curr = po->prot_hook.type;
3105 dev_curr = po->prot_hook.dev;
3107 need_rehook = proto_curr != proto || dev_curr != dev;
3112 /* prevents packet_notifier() from calling
3113 * register_prot_hook()
3116 __unregister_prot_hook(sk, true);
3118 dev_curr = po->prot_hook.dev;
3120 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3124 BUG_ON(po->running);
3126 po->prot_hook.type = proto;
3128 if (unlikely(unlisted)) {
3130 po->prot_hook.dev = NULL;
3132 packet_cached_dev_reset(po);
3134 po->prot_hook.dev = dev;
3135 po->ifindex = dev ? dev->ifindex : 0;
3136 packet_cached_dev_assign(po, dev);
3142 if (proto == 0 || !need_rehook)
3145 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3146 register_prot_hook(sk);
3148 sk->sk_err = ENETDOWN;
3149 if (!sock_flag(sk, SOCK_DEAD))
3150 sk->sk_error_report(sk);
3155 spin_unlock(&po->bind_lock);
3161 * Bind a packet socket to a device
3164 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3167 struct sock *sk = sock->sk;
3168 char name[sizeof(uaddr->sa_data) + 1];
3174 if (addr_len != sizeof(struct sockaddr))
3176 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3179 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3180 name[sizeof(uaddr->sa_data)] = 0;
3182 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3185 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3187 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3188 struct sock *sk = sock->sk;
3194 if (addr_len < sizeof(struct sockaddr_ll))
3196 if (sll->sll_family != AF_PACKET)
3199 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3200 sll->sll_protocol ? : pkt_sk(sk)->num);
3203 static struct proto packet_proto = {
3205 .owner = THIS_MODULE,
3206 .obj_size = sizeof(struct packet_sock),
3210 * Create a packet of type SOCK_PACKET.
3213 static int packet_create(struct net *net, struct socket *sock, int protocol,
3217 struct packet_sock *po;
3218 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3221 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3223 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3224 sock->type != SOCK_PACKET)
3225 return -ESOCKTNOSUPPORT;
3227 sock->state = SS_UNCONNECTED;
3230 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3234 sock->ops = &packet_ops;
3235 if (sock->type == SOCK_PACKET)
3236 sock->ops = &packet_ops_spkt;
3238 sock_init_data(sock, sk);
3241 init_completion(&po->skb_completion);
3242 sk->sk_family = PF_PACKET;
3244 po->xmit = dev_queue_xmit;
3246 err = packet_alloc_pending(po);
3250 packet_cached_dev_reset(po);
3252 sk->sk_destruct = packet_sock_destruct;
3253 sk_refcnt_debug_inc(sk);
3256 * Attach a protocol block
3259 spin_lock_init(&po->bind_lock);
3260 mutex_init(&po->pg_vec_lock);
3261 po->rollover = NULL;
3262 po->prot_hook.func = packet_rcv;
3264 if (sock->type == SOCK_PACKET)
3265 po->prot_hook.func = packet_rcv_spkt;
3267 po->prot_hook.af_packet_priv = sk;
3270 po->prot_hook.type = proto;
3271 __register_prot_hook(sk);
3274 mutex_lock(&net->packet.sklist_lock);
3275 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3276 mutex_unlock(&net->packet.sklist_lock);
3279 sock_prot_inuse_add(net, &packet_proto, 1);
3290 * Pull a packet from our receive queue and hand it to the user.
3291 * If necessary we block.
3294 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3297 struct sock *sk = sock->sk;
3298 struct sk_buff *skb;
3300 int vnet_hdr_len = 0;
3301 unsigned int origlen = 0;
3304 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3308 /* What error should we return now? EUNATTACH? */
3309 if (pkt_sk(sk)->ifindex < 0)
3313 if (flags & MSG_ERRQUEUE) {
3314 err = sock_recv_errqueue(sk, msg, len,
3315 SOL_PACKET, PACKET_TX_TIMESTAMP);
3320 * Call the generic datagram receiver. This handles all sorts
3321 * of horrible races and re-entrancy so we can forget about it
3322 * in the protocol layers.
3324 * Now it will return ENETDOWN, if device have just gone down,
3325 * but then it will block.
3328 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3331 * An error occurred so return it. Because skb_recv_datagram()
3332 * handles the blocking we don't see and worry about blocking
3339 packet_rcv_try_clear_pressure(pkt_sk(sk));
3341 if (pkt_sk(sk)->has_vnet_hdr) {
3342 err = packet_rcv_vnet(msg, skb, &len);
3345 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3348 /* You lose any data beyond the buffer you gave. If it worries
3349 * a user program they can ask the device for its MTU
3355 msg->msg_flags |= MSG_TRUNC;
3358 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3362 if (sock->type != SOCK_PACKET) {
3363 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3365 /* Original length was stored in sockaddr_ll fields */
3366 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3367 sll->sll_family = AF_PACKET;
3368 sll->sll_protocol = skb->protocol;
3371 sock_recv_ts_and_drops(msg, sk, skb);
3373 if (msg->msg_name) {
3376 /* If the address length field is there to be filled
3377 * in, we fill it in now.
3379 if (sock->type == SOCK_PACKET) {
3380 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3381 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3382 copy_len = msg->msg_namelen;
3384 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3386 msg->msg_namelen = sll->sll_halen +
3387 offsetof(struct sockaddr_ll, sll_addr);
3388 copy_len = msg->msg_namelen;
3389 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3390 memset(msg->msg_name +
3391 offsetof(struct sockaddr_ll, sll_addr),
3392 0, sizeof(sll->sll_addr));
3393 msg->msg_namelen = sizeof(struct sockaddr_ll);
3396 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3399 if (pkt_sk(sk)->auxdata) {
3400 struct tpacket_auxdata aux;
3402 aux.tp_status = TP_STATUS_USER;
3403 if (skb->ip_summed == CHECKSUM_PARTIAL)
3404 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3405 else if (skb->pkt_type != PACKET_OUTGOING &&
3406 (skb->ip_summed == CHECKSUM_COMPLETE ||
3407 skb_csum_unnecessary(skb)))
3408 aux.tp_status |= TP_STATUS_CSUM_VALID;
3410 aux.tp_len = origlen;
3411 aux.tp_snaplen = skb->len;
3413 aux.tp_net = skb_network_offset(skb);
3414 if (skb_vlan_tag_present(skb)) {
3415 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3416 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3417 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3419 aux.tp_vlan_tci = 0;
3420 aux.tp_vlan_tpid = 0;
3422 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3426 * Free or return the buffer as appropriate. Again this
3427 * hides all the races and re-entrancy issues from us.
3429 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3432 skb_free_datagram(sk, skb);
3437 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3440 struct net_device *dev;
3441 struct sock *sk = sock->sk;
3446 uaddr->sa_family = AF_PACKET;
3447 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3449 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3451 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3454 return sizeof(*uaddr);
3457 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3460 struct net_device *dev;
3461 struct sock *sk = sock->sk;
3462 struct packet_sock *po = pkt_sk(sk);
3463 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3468 sll->sll_family = AF_PACKET;
3469 sll->sll_ifindex = po->ifindex;
3470 sll->sll_protocol = po->num;
3471 sll->sll_pkttype = 0;
3473 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3475 sll->sll_hatype = dev->type;
3476 sll->sll_halen = dev->addr_len;
3477 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3479 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3484 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3487 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3491 case PACKET_MR_MULTICAST:
3492 if (i->alen != dev->addr_len)
3495 return dev_mc_add(dev, i->addr);
3497 return dev_mc_del(dev, i->addr);
3499 case PACKET_MR_PROMISC:
3500 return dev_set_promiscuity(dev, what);
3501 case PACKET_MR_ALLMULTI:
3502 return dev_set_allmulti(dev, what);
3503 case PACKET_MR_UNICAST:
3504 if (i->alen != dev->addr_len)
3507 return dev_uc_add(dev, i->addr);
3509 return dev_uc_del(dev, i->addr);
3517 static void packet_dev_mclist_delete(struct net_device *dev,
3518 struct packet_mclist **mlp)
3520 struct packet_mclist *ml;
3522 while ((ml = *mlp) != NULL) {
3523 if (ml->ifindex == dev->ifindex) {
3524 packet_dev_mc(dev, ml, -1);
3532 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3534 struct packet_sock *po = pkt_sk(sk);
3535 struct packet_mclist *ml, *i;
3536 struct net_device *dev;
3542 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3547 if (mreq->mr_alen > dev->addr_len)
3551 i = kmalloc(sizeof(*i), GFP_KERNEL);
3556 for (ml = po->mclist; ml; ml = ml->next) {
3557 if (ml->ifindex == mreq->mr_ifindex &&
3558 ml->type == mreq->mr_type &&
3559 ml->alen == mreq->mr_alen &&
3560 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3562 /* Free the new element ... */
3568 i->type = mreq->mr_type;
3569 i->ifindex = mreq->mr_ifindex;
3570 i->alen = mreq->mr_alen;
3571 memcpy(i->addr, mreq->mr_address, i->alen);
3572 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3574 i->next = po->mclist;
3576 err = packet_dev_mc(dev, i, 1);
3578 po->mclist = i->next;
3587 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3589 struct packet_mclist *ml, **mlp;
3593 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3594 if (ml->ifindex == mreq->mr_ifindex &&
3595 ml->type == mreq->mr_type &&
3596 ml->alen == mreq->mr_alen &&
3597 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3598 if (--ml->count == 0) {
3599 struct net_device *dev;
3601 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3603 packet_dev_mc(dev, ml, -1);
3613 static void packet_flush_mclist(struct sock *sk)
3615 struct packet_sock *po = pkt_sk(sk);
3616 struct packet_mclist *ml;
3622 while ((ml = po->mclist) != NULL) {
3623 struct net_device *dev;
3625 po->mclist = ml->next;
3626 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3628 packet_dev_mc(dev, ml, -1);
3635 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3637 struct sock *sk = sock->sk;
3638 struct packet_sock *po = pkt_sk(sk);
3641 if (level != SOL_PACKET)
3642 return -ENOPROTOOPT;
3645 case PACKET_ADD_MEMBERSHIP:
3646 case PACKET_DROP_MEMBERSHIP:
3648 struct packet_mreq_max mreq;
3650 memset(&mreq, 0, sizeof(mreq));
3651 if (len < sizeof(struct packet_mreq))
3653 if (len > sizeof(mreq))
3655 if (copy_from_user(&mreq, optval, len))
3657 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3659 if (optname == PACKET_ADD_MEMBERSHIP)
3660 ret = packet_mc_add(sk, &mreq);
3662 ret = packet_mc_drop(sk, &mreq);
3666 case PACKET_RX_RING:
3667 case PACKET_TX_RING:
3669 union tpacket_req_u req_u;
3673 switch (po->tp_version) {
3676 len = sizeof(req_u.req);
3680 len = sizeof(req_u.req3);
3686 if (copy_from_user(&req_u.req, optval, len))
3689 ret = packet_set_ring(sk, &req_u, 0,
3690 optname == PACKET_TX_RING);
3695 case PACKET_COPY_THRESH:
3699 if (optlen != sizeof(val))
3701 if (copy_from_user(&val, optval, sizeof(val)))
3704 pkt_sk(sk)->copy_thresh = val;
3707 case PACKET_VERSION:
3711 if (optlen != sizeof(val))
3713 if (copy_from_user(&val, optval, sizeof(val)))
3724 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3727 po->tp_version = val;
3733 case PACKET_RESERVE:
3737 if (optlen != sizeof(val))
3739 if (copy_from_user(&val, optval, sizeof(val)))
3744 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3747 po->tp_reserve = val;
3757 if (optlen != sizeof(val))
3759 if (copy_from_user(&val, optval, sizeof(val)))
3763 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3766 po->tp_loss = !!val;
3772 case PACKET_AUXDATA:
3776 if (optlen < sizeof(val))
3778 if (copy_from_user(&val, optval, sizeof(val)))
3782 po->auxdata = !!val;
3786 case PACKET_ORIGDEV:
3790 if (optlen < sizeof(val))
3792 if (copy_from_user(&val, optval, sizeof(val)))
3796 po->origdev = !!val;
3800 case PACKET_VNET_HDR:
3804 if (sock->type != SOCK_RAW)
3806 if (optlen < sizeof(val))
3808 if (copy_from_user(&val, optval, sizeof(val)))
3812 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3815 po->has_vnet_hdr = !!val;
3821 case PACKET_TIMESTAMP:
3825 if (optlen != sizeof(val))
3827 if (copy_from_user(&val, optval, sizeof(val)))
3830 po->tp_tstamp = val;
3837 if (optlen != sizeof(val))
3839 if (copy_from_user(&val, optval, sizeof(val)))
3842 return fanout_add(sk, val & 0xffff, val >> 16);
3844 case PACKET_FANOUT_DATA:
3849 return fanout_set_data(po, optval, optlen);
3851 case PACKET_IGNORE_OUTGOING:
3855 if (optlen != sizeof(val))
3857 if (copy_from_user(&val, optval, sizeof(val)))
3859 if (val < 0 || val > 1)
3862 po->prot_hook.ignore_outgoing = !!val;
3865 case PACKET_TX_HAS_OFF:
3869 if (optlen != sizeof(val))
3871 if (copy_from_user(&val, optval, sizeof(val)))
3875 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3878 po->tp_tx_has_off = !!val;
3884 case PACKET_QDISC_BYPASS:
3888 if (optlen != sizeof(val))
3890 if (copy_from_user(&val, optval, sizeof(val)))
3893 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3897 return -ENOPROTOOPT;
3901 static int packet_getsockopt(struct socket *sock, int level, int optname,
3902 char __user *optval, int __user *optlen)
3905 int val, lv = sizeof(val);
3906 struct sock *sk = sock->sk;
3907 struct packet_sock *po = pkt_sk(sk);
3909 union tpacket_stats_u st;
3910 struct tpacket_rollover_stats rstats;
3913 if (level != SOL_PACKET)
3914 return -ENOPROTOOPT;
3916 if (get_user(len, optlen))
3923 case PACKET_STATISTICS:
3924 spin_lock_bh(&sk->sk_receive_queue.lock);
3925 memcpy(&st, &po->stats, sizeof(st));
3926 memset(&po->stats, 0, sizeof(po->stats));
3927 spin_unlock_bh(&sk->sk_receive_queue.lock);
3928 drops = atomic_xchg(&po->tp_drops, 0);
3930 if (po->tp_version == TPACKET_V3) {
3931 lv = sizeof(struct tpacket_stats_v3);
3932 st.stats3.tp_drops = drops;
3933 st.stats3.tp_packets += drops;
3936 lv = sizeof(struct tpacket_stats);
3937 st.stats1.tp_drops = drops;
3938 st.stats1.tp_packets += drops;
3943 case PACKET_AUXDATA:
3946 case PACKET_ORIGDEV:
3949 case PACKET_VNET_HDR:
3950 val = po->has_vnet_hdr;
3952 case PACKET_VERSION:
3953 val = po->tp_version;
3956 if (len > sizeof(int))
3958 if (len < sizeof(int))
3960 if (copy_from_user(&val, optval, len))
3964 val = sizeof(struct tpacket_hdr);
3967 val = sizeof(struct tpacket2_hdr);
3970 val = sizeof(struct tpacket3_hdr);
3976 case PACKET_RESERVE:
3977 val = po->tp_reserve;
3982 case PACKET_TIMESTAMP:
3983 val = po->tp_tstamp;
3987 ((u32)po->fanout->id |
3988 ((u32)po->fanout->type << 16) |
3989 ((u32)po->fanout->flags << 24)) :
3992 case PACKET_IGNORE_OUTGOING:
3993 val = po->prot_hook.ignore_outgoing;
3995 case PACKET_ROLLOVER_STATS:
3998 rstats.tp_all = atomic_long_read(&po->rollover->num);
3999 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4000 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4002 lv = sizeof(rstats);
4004 case PACKET_TX_HAS_OFF:
4005 val = po->tp_tx_has_off;
4007 case PACKET_QDISC_BYPASS:
4008 val = packet_use_direct_xmit(po);
4011 return -ENOPROTOOPT;
4016 if (put_user(len, optlen))
4018 if (copy_to_user(optval, data, len))
4024 #ifdef CONFIG_COMPAT
4025 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4026 char __user *optval, unsigned int optlen)
4028 struct packet_sock *po = pkt_sk(sock->sk);
4030 if (level != SOL_PACKET)
4031 return -ENOPROTOOPT;
4033 if (optname == PACKET_FANOUT_DATA &&
4034 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4035 optval = (char __user *)get_compat_bpf_fprog(optval);
4038 optlen = sizeof(struct sock_fprog);
4041 return packet_setsockopt(sock, level, optname, optval, optlen);
4045 static int packet_notifier(struct notifier_block *this,
4046 unsigned long msg, void *ptr)
4049 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4050 struct net *net = dev_net(dev);
4053 sk_for_each_rcu(sk, &net->packet.sklist) {
4054 struct packet_sock *po = pkt_sk(sk);
4057 case NETDEV_UNREGISTER:
4059 packet_dev_mclist_delete(dev, &po->mclist);
4063 if (dev->ifindex == po->ifindex) {
4064 spin_lock(&po->bind_lock);
4066 __unregister_prot_hook(sk, false);
4067 sk->sk_err = ENETDOWN;
4068 if (!sock_flag(sk, SOCK_DEAD))
4069 sk->sk_error_report(sk);
4071 if (msg == NETDEV_UNREGISTER) {
4072 packet_cached_dev_reset(po);
4074 if (po->prot_hook.dev)
4075 dev_put(po->prot_hook.dev);
4076 po->prot_hook.dev = NULL;
4078 spin_unlock(&po->bind_lock);
4082 if (dev->ifindex == po->ifindex) {
4083 spin_lock(&po->bind_lock);
4085 register_prot_hook(sk);
4086 spin_unlock(&po->bind_lock);
4096 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4099 struct sock *sk = sock->sk;
4104 int amount = sk_wmem_alloc_get(sk);
4106 return put_user(amount, (int __user *)arg);
4110 struct sk_buff *skb;
4113 spin_lock_bh(&sk->sk_receive_queue.lock);
4114 skb = skb_peek(&sk->sk_receive_queue);
4117 spin_unlock_bh(&sk->sk_receive_queue.lock);
4118 return put_user(amount, (int __user *)arg);
4128 case SIOCGIFBRDADDR:
4129 case SIOCSIFBRDADDR:
4130 case SIOCGIFNETMASK:
4131 case SIOCSIFNETMASK:
4132 case SIOCGIFDSTADDR:
4133 case SIOCSIFDSTADDR:
4135 return inet_dgram_ops.ioctl(sock, cmd, arg);
4139 return -ENOIOCTLCMD;
4144 static __poll_t packet_poll(struct file *file, struct socket *sock,
4147 struct sock *sk = sock->sk;
4148 struct packet_sock *po = pkt_sk(sk);
4149 __poll_t mask = datagram_poll(file, sock, wait);
4151 spin_lock_bh(&sk->sk_receive_queue.lock);
4152 if (po->rx_ring.pg_vec) {
4153 if (!packet_previous_rx_frame(po, &po->rx_ring,
4155 mask |= EPOLLIN | EPOLLRDNORM;
4157 packet_rcv_try_clear_pressure(po);
4158 spin_unlock_bh(&sk->sk_receive_queue.lock);
4159 spin_lock_bh(&sk->sk_write_queue.lock);
4160 if (po->tx_ring.pg_vec) {
4161 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4162 mask |= EPOLLOUT | EPOLLWRNORM;
4164 spin_unlock_bh(&sk->sk_write_queue.lock);
4169 /* Dirty? Well, I still did not learn better way to account
4173 static void packet_mm_open(struct vm_area_struct *vma)
4175 struct file *file = vma->vm_file;
4176 struct socket *sock = file->private_data;
4177 struct sock *sk = sock->sk;
4180 atomic_inc(&pkt_sk(sk)->mapped);
4183 static void packet_mm_close(struct vm_area_struct *vma)
4185 struct file *file = vma->vm_file;
4186 struct socket *sock = file->private_data;
4187 struct sock *sk = sock->sk;
4190 atomic_dec(&pkt_sk(sk)->mapped);
4193 static const struct vm_operations_struct packet_mmap_ops = {
4194 .open = packet_mm_open,
4195 .close = packet_mm_close,
4198 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4203 for (i = 0; i < len; i++) {
4204 if (likely(pg_vec[i].buffer)) {
4205 if (is_vmalloc_addr(pg_vec[i].buffer))
4206 vfree(pg_vec[i].buffer);
4208 free_pages((unsigned long)pg_vec[i].buffer,
4210 pg_vec[i].buffer = NULL;
4216 static char *alloc_one_pg_vec_page(unsigned long order)
4219 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4220 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4222 buffer = (char *) __get_free_pages(gfp_flags, order);
4226 /* __get_free_pages failed, fall back to vmalloc */
4227 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4231 /* vmalloc failed, lets dig into swap here */
4232 gfp_flags &= ~__GFP_NORETRY;
4233 buffer = (char *) __get_free_pages(gfp_flags, order);
4237 /* complete and utter failure */
4241 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4243 unsigned int block_nr = req->tp_block_nr;
4247 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4248 if (unlikely(!pg_vec))
4251 for (i = 0; i < block_nr; i++) {
4252 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4253 if (unlikely(!pg_vec[i].buffer))
4254 goto out_free_pgvec;
4261 free_pg_vec(pg_vec, order, block_nr);
4266 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4267 int closing, int tx_ring)
4269 struct pgv *pg_vec = NULL;
4270 struct packet_sock *po = pkt_sk(sk);
4271 int was_running, order = 0;
4272 struct packet_ring_buffer *rb;
4273 struct sk_buff_head *rb_queue;
4276 /* Added to avoid minimal code churn */
4277 struct tpacket_req *req = &req_u->req;
4279 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4280 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4284 if (atomic_read(&po->mapped))
4286 if (packet_read_pending(rb))
4290 if (req->tp_block_nr) {
4291 unsigned int min_frame_size;
4293 /* Sanity tests and some calculations */
4295 if (unlikely(rb->pg_vec))
4298 switch (po->tp_version) {
4300 po->tp_hdrlen = TPACKET_HDRLEN;
4303 po->tp_hdrlen = TPACKET2_HDRLEN;
4306 po->tp_hdrlen = TPACKET3_HDRLEN;
4311 if (unlikely((int)req->tp_block_size <= 0))
4313 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4315 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4316 if (po->tp_version >= TPACKET_V3 &&
4317 req->tp_block_size <
4318 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4320 if (unlikely(req->tp_frame_size < min_frame_size))
4322 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4325 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4326 if (unlikely(rb->frames_per_block == 0))
4328 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4330 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4335 order = get_order(req->tp_block_size);
4336 pg_vec = alloc_pg_vec(req, order);
4337 if (unlikely(!pg_vec))
4339 switch (po->tp_version) {
4341 /* Block transmit is not supported yet */
4343 init_prb_bdqc(po, rb, pg_vec, req_u);
4345 struct tpacket_req3 *req3 = &req_u->req3;
4347 if (req3->tp_retire_blk_tov ||
4348 req3->tp_sizeof_priv ||
4349 req3->tp_feature_req_word) {
4351 goto out_free_pg_vec;
4362 if (unlikely(req->tp_frame_nr))
4367 /* Detach socket from network */
4368 spin_lock(&po->bind_lock);
4369 was_running = po->running;
4373 __unregister_prot_hook(sk, false);
4375 spin_unlock(&po->bind_lock);
4380 mutex_lock(&po->pg_vec_lock);
4381 if (closing || atomic_read(&po->mapped) == 0) {
4383 spin_lock_bh(&rb_queue->lock);
4384 swap(rb->pg_vec, pg_vec);
4385 rb->frame_max = (req->tp_frame_nr - 1);
4387 rb->frame_size = req->tp_frame_size;
4388 spin_unlock_bh(&rb_queue->lock);
4390 swap(rb->pg_vec_order, order);
4391 swap(rb->pg_vec_len, req->tp_block_nr);
4393 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4394 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4395 tpacket_rcv : packet_rcv;
4396 skb_queue_purge(rb_queue);
4397 if (atomic_read(&po->mapped))
4398 pr_err("packet_mmap: vma is busy: %d\n",
4399 atomic_read(&po->mapped));
4401 mutex_unlock(&po->pg_vec_lock);
4403 spin_lock(&po->bind_lock);
4406 register_prot_hook(sk);
4408 spin_unlock(&po->bind_lock);
4409 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4410 /* Because we don't support block-based V3 on tx-ring */
4412 prb_shutdown_retire_blk_timer(po, rb_queue);
4417 free_pg_vec(pg_vec, order, req->tp_block_nr);
4422 static int packet_mmap(struct file *file, struct socket *sock,
4423 struct vm_area_struct *vma)
4425 struct sock *sk = sock->sk;
4426 struct packet_sock *po = pkt_sk(sk);
4427 unsigned long size, expected_size;
4428 struct packet_ring_buffer *rb;
4429 unsigned long start;
4436 mutex_lock(&po->pg_vec_lock);
4439 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4441 expected_size += rb->pg_vec_len
4447 if (expected_size == 0)
4450 size = vma->vm_end - vma->vm_start;
4451 if (size != expected_size)
4454 start = vma->vm_start;
4455 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4456 if (rb->pg_vec == NULL)
4459 for (i = 0; i < rb->pg_vec_len; i++) {
4461 void *kaddr = rb->pg_vec[i].buffer;
4464 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4465 page = pgv_to_page(kaddr);
4466 err = vm_insert_page(vma, start, page);
4475 atomic_inc(&po->mapped);
4476 vma->vm_ops = &packet_mmap_ops;
4480 mutex_unlock(&po->pg_vec_lock);
4484 static const struct proto_ops packet_ops_spkt = {
4485 .family = PF_PACKET,
4486 .owner = THIS_MODULE,
4487 .release = packet_release,
4488 .bind = packet_bind_spkt,
4489 .connect = sock_no_connect,
4490 .socketpair = sock_no_socketpair,
4491 .accept = sock_no_accept,
4492 .getname = packet_getname_spkt,
4493 .poll = datagram_poll,
4494 .ioctl = packet_ioctl,
4495 .gettstamp = sock_gettstamp,
4496 .listen = sock_no_listen,
4497 .shutdown = sock_no_shutdown,
4498 .setsockopt = sock_no_setsockopt,
4499 .getsockopt = sock_no_getsockopt,
4500 .sendmsg = packet_sendmsg_spkt,
4501 .recvmsg = packet_recvmsg,
4502 .mmap = sock_no_mmap,
4503 .sendpage = sock_no_sendpage,
4506 static const struct proto_ops packet_ops = {
4507 .family = PF_PACKET,
4508 .owner = THIS_MODULE,
4509 .release = packet_release,
4510 .bind = packet_bind,
4511 .connect = sock_no_connect,
4512 .socketpair = sock_no_socketpair,
4513 .accept = sock_no_accept,
4514 .getname = packet_getname,
4515 .poll = packet_poll,
4516 .ioctl = packet_ioctl,
4517 .gettstamp = sock_gettstamp,
4518 .listen = sock_no_listen,
4519 .shutdown = sock_no_shutdown,
4520 .setsockopt = packet_setsockopt,
4521 .getsockopt = packet_getsockopt,
4522 #ifdef CONFIG_COMPAT
4523 .compat_setsockopt = compat_packet_setsockopt,
4525 .sendmsg = packet_sendmsg,
4526 .recvmsg = packet_recvmsg,
4527 .mmap = packet_mmap,
4528 .sendpage = sock_no_sendpage,
4531 static const struct net_proto_family packet_family_ops = {
4532 .family = PF_PACKET,
4533 .create = packet_create,
4534 .owner = THIS_MODULE,
4537 static struct notifier_block packet_netdev_notifier = {
4538 .notifier_call = packet_notifier,
4541 #ifdef CONFIG_PROC_FS
4543 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4546 struct net *net = seq_file_net(seq);
4549 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4552 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4554 struct net *net = seq_file_net(seq);
4555 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4558 static void packet_seq_stop(struct seq_file *seq, void *v)
4564 static int packet_seq_show(struct seq_file *seq, void *v)
4566 if (v == SEQ_START_TOKEN)
4567 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4569 struct sock *s = sk_entry(v);
4570 const struct packet_sock *po = pkt_sk(s);
4573 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4575 refcount_read(&s->sk_refcnt),
4580 atomic_read(&s->sk_rmem_alloc),
4581 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4588 static const struct seq_operations packet_seq_ops = {
4589 .start = packet_seq_start,
4590 .next = packet_seq_next,
4591 .stop = packet_seq_stop,
4592 .show = packet_seq_show,
4596 static int __net_init packet_net_init(struct net *net)
4598 mutex_init(&net->packet.sklist_lock);
4599 INIT_HLIST_HEAD(&net->packet.sklist);
4601 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4602 sizeof(struct seq_net_private)))
4608 static void __net_exit packet_net_exit(struct net *net)
4610 remove_proc_entry("packet", net->proc_net);
4611 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4614 static struct pernet_operations packet_net_ops = {
4615 .init = packet_net_init,
4616 .exit = packet_net_exit,
4620 static void __exit packet_exit(void)
4622 unregister_netdevice_notifier(&packet_netdev_notifier);
4623 unregister_pernet_subsys(&packet_net_ops);
4624 sock_unregister(PF_PACKET);
4625 proto_unregister(&packet_proto);
4628 static int __init packet_init(void)
4632 rc = proto_register(&packet_proto, 0);
4635 rc = sock_register(&packet_family_ops);
4638 rc = register_pernet_subsys(&packet_net_ops);
4641 rc = register_netdevice_notifier(&packet_netdev_notifier);
4648 unregister_pernet_subsys(&packet_net_ops);
4650 sock_unregister(PF_PACKET);
4652 proto_unregister(&packet_proto);
4657 module_init(packet_init);
4658 module_exit(packet_exit);
4659 MODULE_LICENSE("GPL");
4660 MODULE_ALIAS_NETPROTO(PF_PACKET);