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)
1298 u32 *history = po->rollover->history;
1302 rxhash = skb_get_hash(skb);
1303 for (i = 0; i < ROLLOVER_HLEN; i++)
1304 if (READ_ONCE(history[i]) == rxhash)
1307 victim = prandom_u32() % ROLLOVER_HLEN;
1309 /* Avoid dirtying the cache line if possible */
1310 if (READ_ONCE(history[victim]) != rxhash)
1311 WRITE_ONCE(history[victim], rxhash);
1313 return count > (ROLLOVER_HLEN >> 1);
1316 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1317 struct sk_buff *skb,
1320 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1323 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1324 struct sk_buff *skb,
1327 unsigned int val = atomic_inc_return(&f->rr_cur);
1332 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1333 struct sk_buff *skb,
1336 return smp_processor_id() % num;
1339 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1340 struct sk_buff *skb,
1343 return prandom_u32_max(num);
1346 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1347 struct sk_buff *skb,
1348 unsigned int idx, bool try_self,
1351 struct packet_sock *po, *po_next, *po_skip = NULL;
1352 unsigned int i, j, room = ROOM_NONE;
1354 po = pkt_sk(f->arr[idx]);
1357 room = packet_rcv_has_room(po, skb);
1358 if (room == ROOM_NORMAL ||
1359 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1364 i = j = min_t(int, po->rollover->sock, num - 1);
1366 po_next = pkt_sk(f->arr[i]);
1367 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1368 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1370 po->rollover->sock = i;
1371 atomic_long_inc(&po->rollover->num);
1372 if (room == ROOM_LOW)
1373 atomic_long_inc(&po->rollover->num_huge);
1381 atomic_long_inc(&po->rollover->num_failed);
1385 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1386 struct sk_buff *skb,
1389 return skb_get_queue_mapping(skb) % num;
1392 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1393 struct sk_buff *skb,
1396 struct bpf_prog *prog;
1397 unsigned int ret = 0;
1400 prog = rcu_dereference(f->bpf_prog);
1402 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1408 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1410 return f->flags & (flag >> 8);
1413 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1414 struct packet_type *pt, struct net_device *orig_dev)
1416 struct packet_fanout *f = pt->af_packet_priv;
1417 unsigned int num = READ_ONCE(f->num_members);
1418 struct net *net = read_pnet(&f->net);
1419 struct packet_sock *po;
1422 if (!net_eq(dev_net(dev), net) || !num) {
1427 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1428 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1433 case PACKET_FANOUT_HASH:
1435 idx = fanout_demux_hash(f, skb, num);
1437 case PACKET_FANOUT_LB:
1438 idx = fanout_demux_lb(f, skb, num);
1440 case PACKET_FANOUT_CPU:
1441 idx = fanout_demux_cpu(f, skb, num);
1443 case PACKET_FANOUT_RND:
1444 idx = fanout_demux_rnd(f, skb, num);
1446 case PACKET_FANOUT_QM:
1447 idx = fanout_demux_qm(f, skb, num);
1449 case PACKET_FANOUT_ROLLOVER:
1450 idx = fanout_demux_rollover(f, skb, 0, false, num);
1452 case PACKET_FANOUT_CBPF:
1453 case PACKET_FANOUT_EBPF:
1454 idx = fanout_demux_bpf(f, skb, num);
1458 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1459 idx = fanout_demux_rollover(f, skb, idx, true, num);
1461 po = pkt_sk(f->arr[idx]);
1462 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1465 DEFINE_MUTEX(fanout_mutex);
1466 EXPORT_SYMBOL_GPL(fanout_mutex);
1467 static LIST_HEAD(fanout_list);
1468 static u16 fanout_next_id;
1470 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1472 struct packet_fanout *f = po->fanout;
1474 spin_lock(&f->lock);
1475 f->arr[f->num_members] = sk;
1478 if (f->num_members == 1)
1479 dev_add_pack(&f->prot_hook);
1480 spin_unlock(&f->lock);
1483 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1485 struct packet_fanout *f = po->fanout;
1488 spin_lock(&f->lock);
1489 for (i = 0; i < f->num_members; i++) {
1490 if (f->arr[i] == sk)
1493 BUG_ON(i >= f->num_members);
1494 f->arr[i] = f->arr[f->num_members - 1];
1496 if (f->num_members == 0)
1497 __dev_remove_pack(&f->prot_hook);
1498 spin_unlock(&f->lock);
1501 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1503 if (sk->sk_family != PF_PACKET)
1506 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1509 static void fanout_init_data(struct packet_fanout *f)
1512 case PACKET_FANOUT_LB:
1513 atomic_set(&f->rr_cur, 0);
1515 case PACKET_FANOUT_CBPF:
1516 case PACKET_FANOUT_EBPF:
1517 RCU_INIT_POINTER(f->bpf_prog, NULL);
1522 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1524 struct bpf_prog *old;
1526 spin_lock(&f->lock);
1527 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1528 rcu_assign_pointer(f->bpf_prog, new);
1529 spin_unlock(&f->lock);
1533 bpf_prog_destroy(old);
1537 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1540 struct bpf_prog *new;
1541 struct sock_fprog fprog;
1544 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1546 if (len != sizeof(fprog))
1548 if (copy_from_user(&fprog, data, len))
1551 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1555 __fanout_set_data_bpf(po->fanout, new);
1559 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1562 struct bpf_prog *new;
1565 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1567 if (len != sizeof(fd))
1569 if (copy_from_user(&fd, data, len))
1572 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1574 return PTR_ERR(new);
1576 __fanout_set_data_bpf(po->fanout, new);
1580 static int fanout_set_data(struct packet_sock *po, char __user *data,
1583 switch (po->fanout->type) {
1584 case PACKET_FANOUT_CBPF:
1585 return fanout_set_data_cbpf(po, data, len);
1586 case PACKET_FANOUT_EBPF:
1587 return fanout_set_data_ebpf(po, data, len);
1593 static void fanout_release_data(struct packet_fanout *f)
1596 case PACKET_FANOUT_CBPF:
1597 case PACKET_FANOUT_EBPF:
1598 __fanout_set_data_bpf(f, NULL);
1602 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1604 struct packet_fanout *f;
1606 list_for_each_entry(f, &fanout_list, list) {
1607 if (f->id == candidate_id &&
1608 read_pnet(&f->net) == sock_net(sk)) {
1615 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1617 u16 id = fanout_next_id;
1620 if (__fanout_id_is_free(sk, id)) {
1622 fanout_next_id = id + 1;
1627 } while (id != fanout_next_id);
1632 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1634 struct packet_rollover *rollover = NULL;
1635 struct packet_sock *po = pkt_sk(sk);
1636 struct packet_fanout *f, *match;
1637 u8 type = type_flags & 0xff;
1638 u8 flags = type_flags >> 8;
1642 case PACKET_FANOUT_ROLLOVER:
1643 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1645 case PACKET_FANOUT_HASH:
1646 case PACKET_FANOUT_LB:
1647 case PACKET_FANOUT_CPU:
1648 case PACKET_FANOUT_RND:
1649 case PACKET_FANOUT_QM:
1650 case PACKET_FANOUT_CBPF:
1651 case PACKET_FANOUT_EBPF:
1657 mutex_lock(&fanout_mutex);
1663 if (type == PACKET_FANOUT_ROLLOVER ||
1664 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1666 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1669 atomic_long_set(&rollover->num, 0);
1670 atomic_long_set(&rollover->num_huge, 0);
1671 atomic_long_set(&rollover->num_failed, 0);
1674 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1679 if (!fanout_find_new_id(sk, &id)) {
1683 /* ephemeral flag for the first socket in the group: drop it */
1684 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1688 list_for_each_entry(f, &fanout_list, list) {
1690 read_pnet(&f->net) == sock_net(sk)) {
1696 if (match && match->flags != flags)
1700 match = kzalloc(sizeof(*match), GFP_KERNEL);
1703 write_pnet(&match->net, sock_net(sk));
1706 match->flags = flags;
1707 INIT_LIST_HEAD(&match->list);
1708 spin_lock_init(&match->lock);
1709 refcount_set(&match->sk_ref, 0);
1710 fanout_init_data(match);
1711 match->prot_hook.type = po->prot_hook.type;
1712 match->prot_hook.dev = po->prot_hook.dev;
1713 match->prot_hook.func = packet_rcv_fanout;
1714 match->prot_hook.af_packet_priv = match;
1715 match->prot_hook.id_match = match_fanout_group;
1716 list_add(&match->list, &fanout_list);
1720 spin_lock(&po->bind_lock);
1722 match->type == type &&
1723 match->prot_hook.type == po->prot_hook.type &&
1724 match->prot_hook.dev == po->prot_hook.dev) {
1726 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1727 __dev_remove_pack(&po->prot_hook);
1729 po->rollover = rollover;
1731 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1732 __fanout_link(sk, po);
1736 spin_unlock(&po->bind_lock);
1738 if (err && !refcount_read(&match->sk_ref)) {
1739 list_del(&match->list);
1745 mutex_unlock(&fanout_mutex);
1749 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1750 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1751 * It is the responsibility of the caller to call fanout_release_data() and
1752 * free the returned packet_fanout (after synchronize_net())
1754 static struct packet_fanout *fanout_release(struct sock *sk)
1756 struct packet_sock *po = pkt_sk(sk);
1757 struct packet_fanout *f;
1759 mutex_lock(&fanout_mutex);
1764 if (refcount_dec_and_test(&f->sk_ref))
1769 mutex_unlock(&fanout_mutex);
1774 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1775 struct sk_buff *skb)
1777 /* Earlier code assumed this would be a VLAN pkt, double-check
1778 * this now that we have the actual packet in hand. We can only
1779 * do this check on Ethernet devices.
1781 if (unlikely(dev->type != ARPHRD_ETHER))
1784 skb_reset_mac_header(skb);
1785 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1788 static const struct proto_ops packet_ops;
1790 static const struct proto_ops packet_ops_spkt;
1792 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1793 struct packet_type *pt, struct net_device *orig_dev)
1796 struct sockaddr_pkt *spkt;
1799 * When we registered the protocol we saved the socket in the data
1800 * field for just this event.
1803 sk = pt->af_packet_priv;
1806 * Yank back the headers [hope the device set this
1807 * right or kerboom...]
1809 * Incoming packets have ll header pulled,
1812 * For outgoing ones skb->data == skb_mac_header(skb)
1813 * so that this procedure is noop.
1816 if (skb->pkt_type == PACKET_LOOPBACK)
1819 if (!net_eq(dev_net(dev), sock_net(sk)))
1822 skb = skb_share_check(skb, GFP_ATOMIC);
1826 /* drop any routing info */
1829 /* drop conntrack reference */
1832 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1834 skb_push(skb, skb->data - skb_mac_header(skb));
1837 * The SOCK_PACKET socket receives _all_ frames.
1840 spkt->spkt_family = dev->type;
1841 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1842 spkt->spkt_protocol = skb->protocol;
1845 * Charge the memory to the socket. This is done specifically
1846 * to prevent sockets using all the memory up.
1849 if (sock_queue_rcv_skb(sk, skb) == 0)
1858 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1860 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1861 sock->type == SOCK_RAW) {
1862 skb_reset_mac_header(skb);
1863 skb->protocol = dev_parse_header_protocol(skb);
1866 skb_probe_transport_header(skb);
1870 * Output a raw packet to a device layer. This bypasses all the other
1871 * protocol layers and you must therefore supply it with a complete frame
1874 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1877 struct sock *sk = sock->sk;
1878 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1879 struct sk_buff *skb = NULL;
1880 struct net_device *dev;
1881 struct sockcm_cookie sockc;
1887 * Get and verify the address.
1891 if (msg->msg_namelen < sizeof(struct sockaddr))
1893 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1894 proto = saddr->spkt_protocol;
1896 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1899 * Find the device first to size check it
1902 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1905 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1911 if (!(dev->flags & IFF_UP))
1915 * You may not queue a frame bigger than the mtu. This is the lowest level
1916 * raw protocol and you must do your own fragmentation at this level.
1919 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1920 if (!netif_supports_nofcs(dev)) {
1921 err = -EPROTONOSUPPORT;
1924 extra_len = 4; /* We're doing our own CRC */
1928 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1932 size_t reserved = LL_RESERVED_SPACE(dev);
1933 int tlen = dev->needed_tailroom;
1934 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1937 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1940 /* FIXME: Save some space for broken drivers that write a hard
1941 * header at transmission time by themselves. PPP is the notable
1942 * one here. This should really be fixed at the driver level.
1944 skb_reserve(skb, reserved);
1945 skb_reset_network_header(skb);
1947 /* Try to align data part correctly */
1952 skb_reset_network_header(skb);
1954 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1960 if (!dev_validate_header(dev, skb->data, len)) {
1964 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1965 !packet_extra_vlan_len_allowed(dev, skb)) {
1970 sockcm_init(&sockc, sk);
1971 if (msg->msg_controllen) {
1972 err = sock_cmsg_send(sk, msg, &sockc);
1977 skb->protocol = proto;
1979 skb->priority = sk->sk_priority;
1980 skb->mark = sk->sk_mark;
1981 skb->tstamp = sockc.transmit_time;
1983 skb_setup_tx_timestamp(skb, sockc.tsflags);
1985 if (unlikely(extra_len == 4))
1988 packet_parse_headers(skb, sock);
1990 dev_queue_xmit(skb);
2001 static unsigned int run_filter(struct sk_buff *skb,
2002 const struct sock *sk,
2005 struct sk_filter *filter;
2008 filter = rcu_dereference(sk->sk_filter);
2010 res = bpf_prog_run_clear_cb(filter->prog, skb);
2016 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2019 struct virtio_net_hdr vnet_hdr;
2021 if (*len < sizeof(vnet_hdr))
2023 *len -= sizeof(vnet_hdr);
2025 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2028 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2032 * This function makes lazy skb cloning in hope that most of packets
2033 * are discarded by BPF.
2035 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2036 * and skb->cb are mangled. It works because (and until) packets
2037 * falling here are owned by current CPU. Output packets are cloned
2038 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2039 * sequencially, so that if we return skb to original state on exit,
2040 * we will not harm anyone.
2043 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2044 struct packet_type *pt, struct net_device *orig_dev)
2047 struct sockaddr_ll *sll;
2048 struct packet_sock *po;
2049 u8 *skb_head = skb->data;
2050 int skb_len = skb->len;
2051 unsigned int snaplen, res;
2052 bool is_drop_n_account = false;
2054 if (skb->pkt_type == PACKET_LOOPBACK)
2057 sk = pt->af_packet_priv;
2060 if (!net_eq(dev_net(dev), sock_net(sk)))
2065 if (dev->header_ops) {
2066 /* The device has an explicit notion of ll header,
2067 * exported to higher levels.
2069 * Otherwise, the device hides details of its frame
2070 * structure, so that corresponding packet head is
2071 * never delivered to user.
2073 if (sk->sk_type != SOCK_DGRAM)
2074 skb_push(skb, skb->data - skb_mac_header(skb));
2075 else if (skb->pkt_type == PACKET_OUTGOING) {
2076 /* Special case: outgoing packets have ll header at head */
2077 skb_pull(skb, skb_network_offset(skb));
2083 res = run_filter(skb, sk, snaplen);
2085 goto drop_n_restore;
2089 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2092 if (skb_shared(skb)) {
2093 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2097 if (skb_head != skb->data) {
2098 skb->data = skb_head;
2105 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2107 sll = &PACKET_SKB_CB(skb)->sa.ll;
2108 sll->sll_hatype = dev->type;
2109 sll->sll_pkttype = skb->pkt_type;
2110 if (unlikely(po->origdev))
2111 sll->sll_ifindex = orig_dev->ifindex;
2113 sll->sll_ifindex = dev->ifindex;
2115 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2117 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2118 * Use their space for storing the original skb length.
2120 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2122 if (pskb_trim(skb, snaplen))
2125 skb_set_owner_r(skb, sk);
2129 /* drop conntrack reference */
2132 spin_lock(&sk->sk_receive_queue.lock);
2133 po->stats.stats1.tp_packets++;
2134 sock_skb_set_dropcount(sk, skb);
2135 __skb_queue_tail(&sk->sk_receive_queue, skb);
2136 spin_unlock(&sk->sk_receive_queue.lock);
2137 sk->sk_data_ready(sk);
2141 is_drop_n_account = true;
2142 atomic_inc(&po->tp_drops);
2143 atomic_inc(&sk->sk_drops);
2146 if (skb_head != skb->data && skb_shared(skb)) {
2147 skb->data = skb_head;
2151 if (!is_drop_n_account)
2158 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2159 struct packet_type *pt, struct net_device *orig_dev)
2162 struct packet_sock *po;
2163 struct sockaddr_ll *sll;
2164 union tpacket_uhdr h;
2165 u8 *skb_head = skb->data;
2166 int skb_len = skb->len;
2167 unsigned int snaplen, res;
2168 unsigned long status = TP_STATUS_USER;
2169 unsigned short macoff, netoff, hdrlen;
2170 struct sk_buff *copy_skb = NULL;
2173 bool is_drop_n_account = false;
2174 bool do_vnet = false;
2176 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2177 * We may add members to them until current aligned size without forcing
2178 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2180 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2181 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2183 if (skb->pkt_type == PACKET_LOOPBACK)
2186 sk = pt->af_packet_priv;
2189 if (!net_eq(dev_net(dev), sock_net(sk)))
2192 if (dev->header_ops) {
2193 if (sk->sk_type != SOCK_DGRAM)
2194 skb_push(skb, skb->data - skb_mac_header(skb));
2195 else if (skb->pkt_type == PACKET_OUTGOING) {
2196 /* Special case: outgoing packets have ll header at head */
2197 skb_pull(skb, skb_network_offset(skb));
2203 res = run_filter(skb, sk, snaplen);
2205 goto drop_n_restore;
2207 /* If we are flooded, just give up */
2208 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2209 atomic_inc(&po->tp_drops);
2210 goto drop_n_restore;
2213 if (skb->ip_summed == CHECKSUM_PARTIAL)
2214 status |= TP_STATUS_CSUMNOTREADY;
2215 else if (skb->pkt_type != PACKET_OUTGOING &&
2216 (skb->ip_summed == CHECKSUM_COMPLETE ||
2217 skb_csum_unnecessary(skb)))
2218 status |= TP_STATUS_CSUM_VALID;
2223 if (sk->sk_type == SOCK_DGRAM) {
2224 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2227 unsigned int maclen = skb_network_offset(skb);
2228 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2229 (maclen < 16 ? 16 : maclen)) +
2231 if (po->has_vnet_hdr) {
2232 netoff += sizeof(struct virtio_net_hdr);
2235 macoff = netoff - maclen;
2237 if (po->tp_version <= TPACKET_V2) {
2238 if (macoff + snaplen > po->rx_ring.frame_size) {
2239 if (po->copy_thresh &&
2240 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2241 if (skb_shared(skb)) {
2242 copy_skb = skb_clone(skb, GFP_ATOMIC);
2244 copy_skb = skb_get(skb);
2245 skb_head = skb->data;
2248 skb_set_owner_r(copy_skb, sk);
2250 snaplen = po->rx_ring.frame_size - macoff;
2251 if ((int)snaplen < 0) {
2256 } else if (unlikely(macoff + snaplen >
2257 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2260 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2261 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2262 snaplen, nval, macoff);
2264 if (unlikely((int)snaplen < 0)) {
2266 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2270 spin_lock(&sk->sk_receive_queue.lock);
2271 h.raw = packet_current_rx_frame(po, skb,
2272 TP_STATUS_KERNEL, (macoff+snaplen));
2274 goto drop_n_account;
2275 if (po->tp_version <= TPACKET_V2) {
2276 packet_increment_rx_head(po, &po->rx_ring);
2278 * LOSING will be reported till you read the stats,
2279 * because it's COR - Clear On Read.
2280 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2283 if (atomic_read(&po->tp_drops))
2284 status |= TP_STATUS_LOSING;
2288 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2289 sizeof(struct virtio_net_hdr),
2291 goto drop_n_account;
2293 po->stats.stats1.tp_packets++;
2295 status |= TP_STATUS_COPY;
2296 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2298 spin_unlock(&sk->sk_receive_queue.lock);
2300 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2302 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2303 getnstimeofday(&ts);
2305 status |= ts_status;
2307 switch (po->tp_version) {
2309 h.h1->tp_len = skb->len;
2310 h.h1->tp_snaplen = snaplen;
2311 h.h1->tp_mac = macoff;
2312 h.h1->tp_net = netoff;
2313 h.h1->tp_sec = ts.tv_sec;
2314 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2315 hdrlen = sizeof(*h.h1);
2318 h.h2->tp_len = skb->len;
2319 h.h2->tp_snaplen = snaplen;
2320 h.h2->tp_mac = macoff;
2321 h.h2->tp_net = netoff;
2322 h.h2->tp_sec = ts.tv_sec;
2323 h.h2->tp_nsec = ts.tv_nsec;
2324 if (skb_vlan_tag_present(skb)) {
2325 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2326 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2327 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2329 h.h2->tp_vlan_tci = 0;
2330 h.h2->tp_vlan_tpid = 0;
2332 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2333 hdrlen = sizeof(*h.h2);
2336 /* tp_nxt_offset,vlan are already populated above.
2337 * So DONT clear those fields here
2339 h.h3->tp_status |= status;
2340 h.h3->tp_len = skb->len;
2341 h.h3->tp_snaplen = snaplen;
2342 h.h3->tp_mac = macoff;
2343 h.h3->tp_net = netoff;
2344 h.h3->tp_sec = ts.tv_sec;
2345 h.h3->tp_nsec = ts.tv_nsec;
2346 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2347 hdrlen = sizeof(*h.h3);
2353 sll = h.raw + TPACKET_ALIGN(hdrlen);
2354 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2355 sll->sll_family = AF_PACKET;
2356 sll->sll_hatype = dev->type;
2357 sll->sll_protocol = skb->protocol;
2358 sll->sll_pkttype = skb->pkt_type;
2359 if (unlikely(po->origdev))
2360 sll->sll_ifindex = orig_dev->ifindex;
2362 sll->sll_ifindex = dev->ifindex;
2366 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2367 if (po->tp_version <= TPACKET_V2) {
2370 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2373 for (start = h.raw; start < end; start += PAGE_SIZE)
2374 flush_dcache_page(pgv_to_page(start));
2379 if (po->tp_version <= TPACKET_V2) {
2380 __packet_set_status(po, h.raw, status);
2381 sk->sk_data_ready(sk);
2383 prb_clear_blk_fill_status(&po->rx_ring);
2387 if (skb_head != skb->data && skb_shared(skb)) {
2388 skb->data = skb_head;
2392 if (!is_drop_n_account)
2399 spin_unlock(&sk->sk_receive_queue.lock);
2400 atomic_inc(&po->tp_drops);
2401 is_drop_n_account = true;
2403 sk->sk_data_ready(sk);
2404 kfree_skb(copy_skb);
2405 goto drop_n_restore;
2408 static void tpacket_destruct_skb(struct sk_buff *skb)
2410 struct packet_sock *po = pkt_sk(skb->sk);
2412 if (likely(po->tx_ring.pg_vec)) {
2416 ph = skb_zcopy_get_nouarg(skb);
2417 packet_dec_pending(&po->tx_ring);
2419 ts = __packet_set_timestamp(po, ph, skb);
2420 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2422 if (!packet_read_pending(&po->tx_ring))
2423 complete(&po->skb_completion);
2429 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2431 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2432 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2433 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2434 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2435 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2436 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2437 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2439 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2445 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2446 struct virtio_net_hdr *vnet_hdr)
2448 if (*len < sizeof(*vnet_hdr))
2450 *len -= sizeof(*vnet_hdr);
2452 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2455 return __packet_snd_vnet_parse(vnet_hdr, *len);
2458 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2459 void *frame, struct net_device *dev, void *data, int tp_len,
2460 __be16 proto, unsigned char *addr, int hlen, int copylen,
2461 const struct sockcm_cookie *sockc)
2463 union tpacket_uhdr ph;
2464 int to_write, offset, len, nr_frags, len_max;
2465 struct socket *sock = po->sk.sk_socket;
2471 skb->protocol = proto;
2473 skb->priority = po->sk.sk_priority;
2474 skb->mark = po->sk.sk_mark;
2475 skb->tstamp = sockc->transmit_time;
2476 skb_setup_tx_timestamp(skb, sockc->tsflags);
2477 skb_zcopy_set_nouarg(skb, ph.raw);
2479 skb_reserve(skb, hlen);
2480 skb_reset_network_header(skb);
2484 if (sock->type == SOCK_DGRAM) {
2485 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2487 if (unlikely(err < 0))
2489 } else if (copylen) {
2490 int hdrlen = min_t(int, copylen, tp_len);
2492 skb_push(skb, dev->hard_header_len);
2493 skb_put(skb, copylen - dev->hard_header_len);
2494 err = skb_store_bits(skb, 0, data, hdrlen);
2497 if (!dev_validate_header(dev, skb->data, hdrlen))
2504 offset = offset_in_page(data);
2505 len_max = PAGE_SIZE - offset;
2506 len = ((to_write > len_max) ? len_max : to_write);
2508 skb->data_len = to_write;
2509 skb->len += to_write;
2510 skb->truesize += to_write;
2511 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2513 while (likely(to_write)) {
2514 nr_frags = skb_shinfo(skb)->nr_frags;
2516 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2517 pr_err("Packet exceed the number of skb frags(%lu)\n",
2522 page = pgv_to_page(data);
2524 flush_dcache_page(page);
2526 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2529 len_max = PAGE_SIZE;
2530 len = ((to_write > len_max) ? len_max : to_write);
2533 packet_parse_headers(skb, sock);
2538 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2539 int size_max, void **data)
2541 union tpacket_uhdr ph;
2546 switch (po->tp_version) {
2548 if (ph.h3->tp_next_offset != 0) {
2549 pr_warn_once("variable sized slot not supported");
2552 tp_len = ph.h3->tp_len;
2555 tp_len = ph.h2->tp_len;
2558 tp_len = ph.h1->tp_len;
2561 if (unlikely(tp_len > size_max)) {
2562 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2566 if (unlikely(po->tp_tx_has_off)) {
2567 int off_min, off_max;
2569 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2570 off_max = po->tx_ring.frame_size - tp_len;
2571 if (po->sk.sk_type == SOCK_DGRAM) {
2572 switch (po->tp_version) {
2574 off = ph.h3->tp_net;
2577 off = ph.h2->tp_net;
2580 off = ph.h1->tp_net;
2584 switch (po->tp_version) {
2586 off = ph.h3->tp_mac;
2589 off = ph.h2->tp_mac;
2592 off = ph.h1->tp_mac;
2596 if (unlikely((off < off_min) || (off_max < off)))
2599 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2602 *data = frame + off;
2606 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2608 struct sk_buff *skb = NULL;
2609 struct net_device *dev;
2610 struct virtio_net_hdr *vnet_hdr = NULL;
2611 struct sockcm_cookie sockc;
2613 int err, reserve = 0;
2615 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2616 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2617 unsigned char *addr = NULL;
2618 int tp_len, size_max;
2621 int status = TP_STATUS_AVAILABLE;
2622 int hlen, tlen, copylen = 0;
2625 mutex_lock(&po->pg_vec_lock);
2627 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2628 * we need to confirm it under protection of pg_vec_lock.
2630 if (unlikely(!po->tx_ring.pg_vec)) {
2634 if (likely(saddr == NULL)) {
2635 dev = packet_cached_dev_get(po);
2639 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2641 if (msg->msg_namelen < (saddr->sll_halen
2642 + offsetof(struct sockaddr_ll,
2645 proto = saddr->sll_protocol;
2646 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2647 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2648 if (dev && msg->msg_namelen < dev->addr_len +
2649 offsetof(struct sockaddr_ll, sll_addr))
2651 addr = saddr->sll_addr;
2656 if (unlikely(dev == NULL))
2659 if (unlikely(!(dev->flags & IFF_UP)))
2662 sockcm_init(&sockc, &po->sk);
2663 if (msg->msg_controllen) {
2664 err = sock_cmsg_send(&po->sk, msg, &sockc);
2669 if (po->sk.sk_socket->type == SOCK_RAW)
2670 reserve = dev->hard_header_len;
2671 size_max = po->tx_ring.frame_size
2672 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2674 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2675 size_max = dev->mtu + reserve + VLAN_HLEN;
2677 reinit_completion(&po->skb_completion);
2680 ph = packet_current_frame(po, &po->tx_ring,
2681 TP_STATUS_SEND_REQUEST);
2682 if (unlikely(ph == NULL)) {
2683 if (need_wait && skb) {
2684 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2685 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2687 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2691 /* check for additional frames */
2696 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2700 status = TP_STATUS_SEND_REQUEST;
2701 hlen = LL_RESERVED_SPACE(dev);
2702 tlen = dev->needed_tailroom;
2703 if (po->has_vnet_hdr) {
2705 data += sizeof(*vnet_hdr);
2706 tp_len -= sizeof(*vnet_hdr);
2708 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2712 copylen = __virtio16_to_cpu(vio_le(),
2715 copylen = max_t(int, copylen, dev->hard_header_len);
2716 skb = sock_alloc_send_skb(&po->sk,
2717 hlen + tlen + sizeof(struct sockaddr_ll) +
2718 (copylen - dev->hard_header_len),
2721 if (unlikely(skb == NULL)) {
2722 /* we assume the socket was initially writeable ... */
2723 if (likely(len_sum > 0))
2727 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2728 addr, hlen, copylen, &sockc);
2729 if (likely(tp_len >= 0) &&
2730 tp_len > dev->mtu + reserve &&
2731 !po->has_vnet_hdr &&
2732 !packet_extra_vlan_len_allowed(dev, skb))
2735 if (unlikely(tp_len < 0)) {
2738 __packet_set_status(po, ph,
2739 TP_STATUS_AVAILABLE);
2740 packet_increment_head(&po->tx_ring);
2744 status = TP_STATUS_WRONG_FORMAT;
2750 if (po->has_vnet_hdr) {
2751 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2755 virtio_net_hdr_set_proto(skb, vnet_hdr);
2758 skb->destructor = tpacket_destruct_skb;
2759 __packet_set_status(po, ph, TP_STATUS_SENDING);
2760 packet_inc_pending(&po->tx_ring);
2762 status = TP_STATUS_SEND_REQUEST;
2763 err = po->xmit(skb);
2764 if (unlikely(err > 0)) {
2765 err = net_xmit_errno(err);
2766 if (err && __packet_get_status(po, ph) ==
2767 TP_STATUS_AVAILABLE) {
2768 /* skb was destructed already */
2773 * skb was dropped but not destructed yet;
2774 * let's treat it like congestion or err < 0
2778 packet_increment_head(&po->tx_ring);
2780 } while (likely((ph != NULL) ||
2781 /* Note: packet_read_pending() might be slow if we have
2782 * to call it as it's per_cpu variable, but in fast-path
2783 * we already short-circuit the loop with the first
2784 * condition, and luckily don't have to go that path
2787 (need_wait && packet_read_pending(&po->tx_ring))));
2793 __packet_set_status(po, ph, status);
2798 mutex_unlock(&po->pg_vec_lock);
2802 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2803 size_t reserve, size_t len,
2804 size_t linear, int noblock,
2807 struct sk_buff *skb;
2809 /* Under a page? Don't bother with paged skb. */
2810 if (prepad + len < PAGE_SIZE || !linear)
2813 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2818 skb_reserve(skb, reserve);
2819 skb_put(skb, linear);
2820 skb->data_len = len - linear;
2821 skb->len += len - linear;
2826 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2828 struct sock *sk = sock->sk;
2829 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2830 struct sk_buff *skb;
2831 struct net_device *dev;
2833 unsigned char *addr = NULL;
2834 int err, reserve = 0;
2835 struct sockcm_cookie sockc;
2836 struct virtio_net_hdr vnet_hdr = { 0 };
2838 struct packet_sock *po = pkt_sk(sk);
2839 bool has_vnet_hdr = false;
2840 int hlen, tlen, linear;
2844 * Get and verify the address.
2847 if (likely(saddr == NULL)) {
2848 dev = packet_cached_dev_get(po);
2852 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2854 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2856 proto = saddr->sll_protocol;
2857 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2858 if (sock->type == SOCK_DGRAM) {
2859 if (dev && msg->msg_namelen < dev->addr_len +
2860 offsetof(struct sockaddr_ll, sll_addr))
2862 addr = saddr->sll_addr;
2867 if (unlikely(dev == NULL))
2870 if (unlikely(!(dev->flags & IFF_UP)))
2873 sockcm_init(&sockc, sk);
2874 sockc.mark = sk->sk_mark;
2875 if (msg->msg_controllen) {
2876 err = sock_cmsg_send(sk, msg, &sockc);
2881 if (sock->type == SOCK_RAW)
2882 reserve = dev->hard_header_len;
2883 if (po->has_vnet_hdr) {
2884 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2887 has_vnet_hdr = true;
2890 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2891 if (!netif_supports_nofcs(dev)) {
2892 err = -EPROTONOSUPPORT;
2895 extra_len = 4; /* We're doing our own CRC */
2899 if (!vnet_hdr.gso_type &&
2900 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2904 hlen = LL_RESERVED_SPACE(dev);
2905 tlen = dev->needed_tailroom;
2906 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2907 linear = max(linear, min_t(int, len, dev->hard_header_len));
2908 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2909 msg->msg_flags & MSG_DONTWAIT, &err);
2913 skb_reset_network_header(skb);
2916 if (sock->type == SOCK_DGRAM) {
2917 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2918 if (unlikely(offset < 0))
2920 } else if (reserve) {
2921 skb_reserve(skb, -reserve);
2922 if (len < reserve + sizeof(struct ipv6hdr) &&
2923 dev->min_header_len != dev->hard_header_len)
2924 skb_reset_network_header(skb);
2927 /* Returns -EFAULT on error */
2928 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2932 if (sock->type == SOCK_RAW &&
2933 !dev_validate_header(dev, skb->data, len)) {
2938 skb_setup_tx_timestamp(skb, sockc.tsflags);
2940 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2941 !packet_extra_vlan_len_allowed(dev, skb)) {
2946 skb->protocol = proto;
2948 skb->priority = sk->sk_priority;
2949 skb->mark = sockc.mark;
2950 skb->tstamp = sockc.transmit_time;
2953 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2956 len += sizeof(vnet_hdr);
2957 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2960 packet_parse_headers(skb, sock);
2962 if (unlikely(extra_len == 4))
2965 err = po->xmit(skb);
2966 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2982 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2984 struct sock *sk = sock->sk;
2985 struct packet_sock *po = pkt_sk(sk);
2987 if (po->tx_ring.pg_vec)
2988 return tpacket_snd(po, msg);
2990 return packet_snd(sock, msg, len);
2994 * Close a PACKET socket. This is fairly simple. We immediately go
2995 * to 'closed' state and remove our protocol entry in the device list.
2998 static int packet_release(struct socket *sock)
3000 struct sock *sk = sock->sk;
3001 struct packet_sock *po;
3002 struct packet_fanout *f;
3004 union tpacket_req_u req_u;
3012 mutex_lock(&net->packet.sklist_lock);
3013 sk_del_node_init_rcu(sk);
3014 mutex_unlock(&net->packet.sklist_lock);
3017 sock_prot_inuse_add(net, sk->sk_prot, -1);
3020 spin_lock(&po->bind_lock);
3021 unregister_prot_hook(sk, false);
3022 packet_cached_dev_reset(po);
3024 if (po->prot_hook.dev) {
3025 dev_put(po->prot_hook.dev);
3026 po->prot_hook.dev = NULL;
3028 spin_unlock(&po->bind_lock);
3030 packet_flush_mclist(sk);
3033 if (po->rx_ring.pg_vec) {
3034 memset(&req_u, 0, sizeof(req_u));
3035 packet_set_ring(sk, &req_u, 1, 0);
3038 if (po->tx_ring.pg_vec) {
3039 memset(&req_u, 0, sizeof(req_u));
3040 packet_set_ring(sk, &req_u, 1, 1);
3044 f = fanout_release(sk);
3048 kfree(po->rollover);
3050 fanout_release_data(f);
3054 * Now the socket is dead. No more input will appear.
3061 skb_queue_purge(&sk->sk_receive_queue);
3062 packet_free_pending(po);
3063 sk_refcnt_debug_release(sk);
3070 * Attach a packet hook.
3073 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3076 struct packet_sock *po = pkt_sk(sk);
3077 struct net_device *dev_curr;
3080 struct net_device *dev = NULL;
3082 bool unlisted = false;
3085 spin_lock(&po->bind_lock);
3094 dev = dev_get_by_name_rcu(sock_net(sk), name);
3099 } else if (ifindex) {
3100 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3110 proto_curr = po->prot_hook.type;
3111 dev_curr = po->prot_hook.dev;
3113 need_rehook = proto_curr != proto || dev_curr != dev;
3118 /* prevents packet_notifier() from calling
3119 * register_prot_hook()
3122 __unregister_prot_hook(sk, true);
3124 dev_curr = po->prot_hook.dev;
3126 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3130 BUG_ON(po->running);
3132 po->prot_hook.type = proto;
3134 if (unlikely(unlisted)) {
3136 po->prot_hook.dev = NULL;
3138 packet_cached_dev_reset(po);
3140 po->prot_hook.dev = dev;
3141 po->ifindex = dev ? dev->ifindex : 0;
3142 packet_cached_dev_assign(po, dev);
3148 if (proto == 0 || !need_rehook)
3151 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3152 register_prot_hook(sk);
3154 sk->sk_err = ENETDOWN;
3155 if (!sock_flag(sk, SOCK_DEAD))
3156 sk->sk_error_report(sk);
3161 spin_unlock(&po->bind_lock);
3167 * Bind a packet socket to a device
3170 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3173 struct sock *sk = sock->sk;
3174 char name[sizeof(uaddr->sa_data) + 1];
3180 if (addr_len != sizeof(struct sockaddr))
3182 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3185 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3186 name[sizeof(uaddr->sa_data)] = 0;
3188 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3191 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3193 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3194 struct sock *sk = sock->sk;
3200 if (addr_len < sizeof(struct sockaddr_ll))
3202 if (sll->sll_family != AF_PACKET)
3205 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3206 sll->sll_protocol ? : pkt_sk(sk)->num);
3209 static struct proto packet_proto = {
3211 .owner = THIS_MODULE,
3212 .obj_size = sizeof(struct packet_sock),
3216 * Create a packet of type SOCK_PACKET.
3219 static int packet_create(struct net *net, struct socket *sock, int protocol,
3223 struct packet_sock *po;
3224 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3227 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3229 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3230 sock->type != SOCK_PACKET)
3231 return -ESOCKTNOSUPPORT;
3233 sock->state = SS_UNCONNECTED;
3236 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3240 sock->ops = &packet_ops;
3241 if (sock->type == SOCK_PACKET)
3242 sock->ops = &packet_ops_spkt;
3244 sock_init_data(sock, sk);
3247 init_completion(&po->skb_completion);
3248 sk->sk_family = PF_PACKET;
3250 po->xmit = dev_queue_xmit;
3252 err = packet_alloc_pending(po);
3256 packet_cached_dev_reset(po);
3258 sk->sk_destruct = packet_sock_destruct;
3259 sk_refcnt_debug_inc(sk);
3262 * Attach a protocol block
3265 spin_lock_init(&po->bind_lock);
3266 mutex_init(&po->pg_vec_lock);
3267 po->rollover = NULL;
3268 po->prot_hook.func = packet_rcv;
3270 if (sock->type == SOCK_PACKET)
3271 po->prot_hook.func = packet_rcv_spkt;
3273 po->prot_hook.af_packet_priv = sk;
3276 po->prot_hook.type = proto;
3277 __register_prot_hook(sk);
3280 mutex_lock(&net->packet.sklist_lock);
3281 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3282 mutex_unlock(&net->packet.sklist_lock);
3285 sock_prot_inuse_add(net, &packet_proto, 1);
3296 * Pull a packet from our receive queue and hand it to the user.
3297 * If necessary we block.
3300 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3303 struct sock *sk = sock->sk;
3304 struct sk_buff *skb;
3306 int vnet_hdr_len = 0;
3307 unsigned int origlen = 0;
3310 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3314 /* What error should we return now? EUNATTACH? */
3315 if (pkt_sk(sk)->ifindex < 0)
3319 if (flags & MSG_ERRQUEUE) {
3320 err = sock_recv_errqueue(sk, msg, len,
3321 SOL_PACKET, PACKET_TX_TIMESTAMP);
3326 * Call the generic datagram receiver. This handles all sorts
3327 * of horrible races and re-entrancy so we can forget about it
3328 * in the protocol layers.
3330 * Now it will return ENETDOWN, if device have just gone down,
3331 * but then it will block.
3334 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3337 * An error occurred so return it. Because skb_recv_datagram()
3338 * handles the blocking we don't see and worry about blocking
3345 packet_rcv_try_clear_pressure(pkt_sk(sk));
3347 if (pkt_sk(sk)->has_vnet_hdr) {
3348 err = packet_rcv_vnet(msg, skb, &len);
3351 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3354 /* You lose any data beyond the buffer you gave. If it worries
3355 * a user program they can ask the device for its MTU
3361 msg->msg_flags |= MSG_TRUNC;
3364 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3368 if (sock->type != SOCK_PACKET) {
3369 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3371 /* Original length was stored in sockaddr_ll fields */
3372 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3373 sll->sll_family = AF_PACKET;
3374 sll->sll_protocol = skb->protocol;
3377 sock_recv_ts_and_drops(msg, sk, skb);
3379 if (msg->msg_name) {
3382 /* If the address length field is there to be filled
3383 * in, we fill it in now.
3385 if (sock->type == SOCK_PACKET) {
3386 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3387 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3388 copy_len = msg->msg_namelen;
3390 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3392 msg->msg_namelen = sll->sll_halen +
3393 offsetof(struct sockaddr_ll, sll_addr);
3394 copy_len = msg->msg_namelen;
3395 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3396 memset(msg->msg_name +
3397 offsetof(struct sockaddr_ll, sll_addr),
3398 0, sizeof(sll->sll_addr));
3399 msg->msg_namelen = sizeof(struct sockaddr_ll);
3402 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3405 if (pkt_sk(sk)->auxdata) {
3406 struct tpacket_auxdata aux;
3408 aux.tp_status = TP_STATUS_USER;
3409 if (skb->ip_summed == CHECKSUM_PARTIAL)
3410 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3411 else if (skb->pkt_type != PACKET_OUTGOING &&
3412 (skb->ip_summed == CHECKSUM_COMPLETE ||
3413 skb_csum_unnecessary(skb)))
3414 aux.tp_status |= TP_STATUS_CSUM_VALID;
3416 aux.tp_len = origlen;
3417 aux.tp_snaplen = skb->len;
3419 aux.tp_net = skb_network_offset(skb);
3420 if (skb_vlan_tag_present(skb)) {
3421 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3422 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3423 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3425 aux.tp_vlan_tci = 0;
3426 aux.tp_vlan_tpid = 0;
3428 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3432 * Free or return the buffer as appropriate. Again this
3433 * hides all the races and re-entrancy issues from us.
3435 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3438 skb_free_datagram(sk, skb);
3443 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3446 struct net_device *dev;
3447 struct sock *sk = sock->sk;
3452 uaddr->sa_family = AF_PACKET;
3453 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3455 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3457 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3460 return sizeof(*uaddr);
3463 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3466 struct net_device *dev;
3467 struct sock *sk = sock->sk;
3468 struct packet_sock *po = pkt_sk(sk);
3469 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3474 sll->sll_family = AF_PACKET;
3475 sll->sll_ifindex = po->ifindex;
3476 sll->sll_protocol = po->num;
3477 sll->sll_pkttype = 0;
3479 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3481 sll->sll_hatype = dev->type;
3482 sll->sll_halen = dev->addr_len;
3483 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3485 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3490 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3493 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3497 case PACKET_MR_MULTICAST:
3498 if (i->alen != dev->addr_len)
3501 return dev_mc_add(dev, i->addr);
3503 return dev_mc_del(dev, i->addr);
3505 case PACKET_MR_PROMISC:
3506 return dev_set_promiscuity(dev, what);
3507 case PACKET_MR_ALLMULTI:
3508 return dev_set_allmulti(dev, what);
3509 case PACKET_MR_UNICAST:
3510 if (i->alen != dev->addr_len)
3513 return dev_uc_add(dev, i->addr);
3515 return dev_uc_del(dev, i->addr);
3523 static void packet_dev_mclist_delete(struct net_device *dev,
3524 struct packet_mclist **mlp)
3526 struct packet_mclist *ml;
3528 while ((ml = *mlp) != NULL) {
3529 if (ml->ifindex == dev->ifindex) {
3530 packet_dev_mc(dev, ml, -1);
3538 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3540 struct packet_sock *po = pkt_sk(sk);
3541 struct packet_mclist *ml, *i;
3542 struct net_device *dev;
3548 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3553 if (mreq->mr_alen > dev->addr_len)
3557 i = kmalloc(sizeof(*i), GFP_KERNEL);
3562 for (ml = po->mclist; ml; ml = ml->next) {
3563 if (ml->ifindex == mreq->mr_ifindex &&
3564 ml->type == mreq->mr_type &&
3565 ml->alen == mreq->mr_alen &&
3566 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3568 /* Free the new element ... */
3574 i->type = mreq->mr_type;
3575 i->ifindex = mreq->mr_ifindex;
3576 i->alen = mreq->mr_alen;
3577 memcpy(i->addr, mreq->mr_address, i->alen);
3578 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3580 i->next = po->mclist;
3582 err = packet_dev_mc(dev, i, 1);
3584 po->mclist = i->next;
3593 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3595 struct packet_mclist *ml, **mlp;
3599 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3600 if (ml->ifindex == mreq->mr_ifindex &&
3601 ml->type == mreq->mr_type &&
3602 ml->alen == mreq->mr_alen &&
3603 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3604 if (--ml->count == 0) {
3605 struct net_device *dev;
3607 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3609 packet_dev_mc(dev, ml, -1);
3619 static void packet_flush_mclist(struct sock *sk)
3621 struct packet_sock *po = pkt_sk(sk);
3622 struct packet_mclist *ml;
3628 while ((ml = po->mclist) != NULL) {
3629 struct net_device *dev;
3631 po->mclist = ml->next;
3632 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3634 packet_dev_mc(dev, ml, -1);
3641 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3643 struct sock *sk = sock->sk;
3644 struct packet_sock *po = pkt_sk(sk);
3647 if (level != SOL_PACKET)
3648 return -ENOPROTOOPT;
3651 case PACKET_ADD_MEMBERSHIP:
3652 case PACKET_DROP_MEMBERSHIP:
3654 struct packet_mreq_max mreq;
3656 memset(&mreq, 0, sizeof(mreq));
3657 if (len < sizeof(struct packet_mreq))
3659 if (len > sizeof(mreq))
3661 if (copy_from_user(&mreq, optval, len))
3663 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3665 if (optname == PACKET_ADD_MEMBERSHIP)
3666 ret = packet_mc_add(sk, &mreq);
3668 ret = packet_mc_drop(sk, &mreq);
3672 case PACKET_RX_RING:
3673 case PACKET_TX_RING:
3675 union tpacket_req_u req_u;
3679 switch (po->tp_version) {
3682 len = sizeof(req_u.req);
3686 len = sizeof(req_u.req3);
3692 if (copy_from_user(&req_u.req, optval, len))
3695 ret = packet_set_ring(sk, &req_u, 0,
3696 optname == PACKET_TX_RING);
3701 case PACKET_COPY_THRESH:
3705 if (optlen != sizeof(val))
3707 if (copy_from_user(&val, optval, sizeof(val)))
3710 pkt_sk(sk)->copy_thresh = val;
3713 case PACKET_VERSION:
3717 if (optlen != sizeof(val))
3719 if (copy_from_user(&val, optval, sizeof(val)))
3730 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3733 po->tp_version = val;
3739 case PACKET_RESERVE:
3743 if (optlen != sizeof(val))
3745 if (copy_from_user(&val, optval, sizeof(val)))
3750 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3753 po->tp_reserve = val;
3763 if (optlen != sizeof(val))
3765 if (copy_from_user(&val, optval, sizeof(val)))
3769 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3772 po->tp_loss = !!val;
3778 case PACKET_AUXDATA:
3782 if (optlen < sizeof(val))
3784 if (copy_from_user(&val, optval, sizeof(val)))
3788 po->auxdata = !!val;
3792 case PACKET_ORIGDEV:
3796 if (optlen < sizeof(val))
3798 if (copy_from_user(&val, optval, sizeof(val)))
3802 po->origdev = !!val;
3806 case PACKET_VNET_HDR:
3810 if (sock->type != SOCK_RAW)
3812 if (optlen < sizeof(val))
3814 if (copy_from_user(&val, optval, sizeof(val)))
3818 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3821 po->has_vnet_hdr = !!val;
3827 case PACKET_TIMESTAMP:
3831 if (optlen != sizeof(val))
3833 if (copy_from_user(&val, optval, sizeof(val)))
3836 po->tp_tstamp = val;
3843 if (optlen != sizeof(val))
3845 if (copy_from_user(&val, optval, sizeof(val)))
3848 return fanout_add(sk, val & 0xffff, val >> 16);
3850 case PACKET_FANOUT_DATA:
3855 return fanout_set_data(po, optval, optlen);
3857 case PACKET_IGNORE_OUTGOING:
3861 if (optlen != sizeof(val))
3863 if (copy_from_user(&val, optval, sizeof(val)))
3865 if (val < 0 || val > 1)
3868 po->prot_hook.ignore_outgoing = !!val;
3871 case PACKET_TX_HAS_OFF:
3875 if (optlen != sizeof(val))
3877 if (copy_from_user(&val, optval, sizeof(val)))
3881 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3884 po->tp_tx_has_off = !!val;
3890 case PACKET_QDISC_BYPASS:
3894 if (optlen != sizeof(val))
3896 if (copy_from_user(&val, optval, sizeof(val)))
3899 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3903 return -ENOPROTOOPT;
3907 static int packet_getsockopt(struct socket *sock, int level, int optname,
3908 char __user *optval, int __user *optlen)
3911 int val, lv = sizeof(val);
3912 struct sock *sk = sock->sk;
3913 struct packet_sock *po = pkt_sk(sk);
3915 union tpacket_stats_u st;
3916 struct tpacket_rollover_stats rstats;
3919 if (level != SOL_PACKET)
3920 return -ENOPROTOOPT;
3922 if (get_user(len, optlen))
3929 case PACKET_STATISTICS:
3930 spin_lock_bh(&sk->sk_receive_queue.lock);
3931 memcpy(&st, &po->stats, sizeof(st));
3932 memset(&po->stats, 0, sizeof(po->stats));
3933 spin_unlock_bh(&sk->sk_receive_queue.lock);
3934 drops = atomic_xchg(&po->tp_drops, 0);
3936 if (po->tp_version == TPACKET_V3) {
3937 lv = sizeof(struct tpacket_stats_v3);
3938 st.stats3.tp_drops = drops;
3939 st.stats3.tp_packets += drops;
3942 lv = sizeof(struct tpacket_stats);
3943 st.stats1.tp_drops = drops;
3944 st.stats1.tp_packets += drops;
3949 case PACKET_AUXDATA:
3952 case PACKET_ORIGDEV:
3955 case PACKET_VNET_HDR:
3956 val = po->has_vnet_hdr;
3958 case PACKET_VERSION:
3959 val = po->tp_version;
3962 if (len > sizeof(int))
3964 if (len < sizeof(int))
3966 if (copy_from_user(&val, optval, len))
3970 val = sizeof(struct tpacket_hdr);
3973 val = sizeof(struct tpacket2_hdr);
3976 val = sizeof(struct tpacket3_hdr);
3982 case PACKET_RESERVE:
3983 val = po->tp_reserve;
3988 case PACKET_TIMESTAMP:
3989 val = po->tp_tstamp;
3993 ((u32)po->fanout->id |
3994 ((u32)po->fanout->type << 16) |
3995 ((u32)po->fanout->flags << 24)) :
3998 case PACKET_IGNORE_OUTGOING:
3999 val = po->prot_hook.ignore_outgoing;
4001 case PACKET_ROLLOVER_STATS:
4004 rstats.tp_all = atomic_long_read(&po->rollover->num);
4005 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4006 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4008 lv = sizeof(rstats);
4010 case PACKET_TX_HAS_OFF:
4011 val = po->tp_tx_has_off;
4013 case PACKET_QDISC_BYPASS:
4014 val = packet_use_direct_xmit(po);
4017 return -ENOPROTOOPT;
4022 if (put_user(len, optlen))
4024 if (copy_to_user(optval, data, len))
4030 #ifdef CONFIG_COMPAT
4031 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4032 char __user *optval, unsigned int optlen)
4034 struct packet_sock *po = pkt_sk(sock->sk);
4036 if (level != SOL_PACKET)
4037 return -ENOPROTOOPT;
4039 if (optname == PACKET_FANOUT_DATA &&
4040 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4041 optval = (char __user *)get_compat_bpf_fprog(optval);
4044 optlen = sizeof(struct sock_fprog);
4047 return packet_setsockopt(sock, level, optname, optval, optlen);
4051 static int packet_notifier(struct notifier_block *this,
4052 unsigned long msg, void *ptr)
4055 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4056 struct net *net = dev_net(dev);
4059 sk_for_each_rcu(sk, &net->packet.sklist) {
4060 struct packet_sock *po = pkt_sk(sk);
4063 case NETDEV_UNREGISTER:
4065 packet_dev_mclist_delete(dev, &po->mclist);
4069 if (dev->ifindex == po->ifindex) {
4070 spin_lock(&po->bind_lock);
4072 __unregister_prot_hook(sk, false);
4073 sk->sk_err = ENETDOWN;
4074 if (!sock_flag(sk, SOCK_DEAD))
4075 sk->sk_error_report(sk);
4077 if (msg == NETDEV_UNREGISTER) {
4078 packet_cached_dev_reset(po);
4080 if (po->prot_hook.dev)
4081 dev_put(po->prot_hook.dev);
4082 po->prot_hook.dev = NULL;
4084 spin_unlock(&po->bind_lock);
4088 if (dev->ifindex == po->ifindex) {
4089 spin_lock(&po->bind_lock);
4091 register_prot_hook(sk);
4092 spin_unlock(&po->bind_lock);
4102 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4105 struct sock *sk = sock->sk;
4110 int amount = sk_wmem_alloc_get(sk);
4112 return put_user(amount, (int __user *)arg);
4116 struct sk_buff *skb;
4119 spin_lock_bh(&sk->sk_receive_queue.lock);
4120 skb = skb_peek(&sk->sk_receive_queue);
4123 spin_unlock_bh(&sk->sk_receive_queue.lock);
4124 return put_user(amount, (int __user *)arg);
4134 case SIOCGIFBRDADDR:
4135 case SIOCSIFBRDADDR:
4136 case SIOCGIFNETMASK:
4137 case SIOCSIFNETMASK:
4138 case SIOCGIFDSTADDR:
4139 case SIOCSIFDSTADDR:
4141 return inet_dgram_ops.ioctl(sock, cmd, arg);
4145 return -ENOIOCTLCMD;
4150 static __poll_t packet_poll(struct file *file, struct socket *sock,
4153 struct sock *sk = sock->sk;
4154 struct packet_sock *po = pkt_sk(sk);
4155 __poll_t mask = datagram_poll(file, sock, wait);
4157 spin_lock_bh(&sk->sk_receive_queue.lock);
4158 if (po->rx_ring.pg_vec) {
4159 if (!packet_previous_rx_frame(po, &po->rx_ring,
4161 mask |= EPOLLIN | EPOLLRDNORM;
4163 packet_rcv_try_clear_pressure(po);
4164 spin_unlock_bh(&sk->sk_receive_queue.lock);
4165 spin_lock_bh(&sk->sk_write_queue.lock);
4166 if (po->tx_ring.pg_vec) {
4167 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4168 mask |= EPOLLOUT | EPOLLWRNORM;
4170 spin_unlock_bh(&sk->sk_write_queue.lock);
4175 /* Dirty? Well, I still did not learn better way to account
4179 static void packet_mm_open(struct vm_area_struct *vma)
4181 struct file *file = vma->vm_file;
4182 struct socket *sock = file->private_data;
4183 struct sock *sk = sock->sk;
4186 atomic_inc(&pkt_sk(sk)->mapped);
4189 static void packet_mm_close(struct vm_area_struct *vma)
4191 struct file *file = vma->vm_file;
4192 struct socket *sock = file->private_data;
4193 struct sock *sk = sock->sk;
4196 atomic_dec(&pkt_sk(sk)->mapped);
4199 static const struct vm_operations_struct packet_mmap_ops = {
4200 .open = packet_mm_open,
4201 .close = packet_mm_close,
4204 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4209 for (i = 0; i < len; i++) {
4210 if (likely(pg_vec[i].buffer)) {
4211 if (is_vmalloc_addr(pg_vec[i].buffer))
4212 vfree(pg_vec[i].buffer);
4214 free_pages((unsigned long)pg_vec[i].buffer,
4216 pg_vec[i].buffer = NULL;
4222 static char *alloc_one_pg_vec_page(unsigned long order)
4225 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4226 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4228 buffer = (char *) __get_free_pages(gfp_flags, order);
4232 /* __get_free_pages failed, fall back to vmalloc */
4233 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4237 /* vmalloc failed, lets dig into swap here */
4238 gfp_flags &= ~__GFP_NORETRY;
4239 buffer = (char *) __get_free_pages(gfp_flags, order);
4243 /* complete and utter failure */
4247 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4249 unsigned int block_nr = req->tp_block_nr;
4253 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4254 if (unlikely(!pg_vec))
4257 for (i = 0; i < block_nr; i++) {
4258 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4259 if (unlikely(!pg_vec[i].buffer))
4260 goto out_free_pgvec;
4267 free_pg_vec(pg_vec, order, block_nr);
4272 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4273 int closing, int tx_ring)
4275 struct pgv *pg_vec = NULL;
4276 struct packet_sock *po = pkt_sk(sk);
4277 int was_running, order = 0;
4278 struct packet_ring_buffer *rb;
4279 struct sk_buff_head *rb_queue;
4282 /* Added to avoid minimal code churn */
4283 struct tpacket_req *req = &req_u->req;
4285 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4286 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4290 if (atomic_read(&po->mapped))
4292 if (packet_read_pending(rb))
4296 if (req->tp_block_nr) {
4297 unsigned int min_frame_size;
4299 /* Sanity tests and some calculations */
4301 if (unlikely(rb->pg_vec))
4304 switch (po->tp_version) {
4306 po->tp_hdrlen = TPACKET_HDRLEN;
4309 po->tp_hdrlen = TPACKET2_HDRLEN;
4312 po->tp_hdrlen = TPACKET3_HDRLEN;
4317 if (unlikely((int)req->tp_block_size <= 0))
4319 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4321 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4322 if (po->tp_version >= TPACKET_V3 &&
4323 req->tp_block_size <
4324 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4326 if (unlikely(req->tp_frame_size < min_frame_size))
4328 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4331 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4332 if (unlikely(rb->frames_per_block == 0))
4334 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4336 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4341 order = get_order(req->tp_block_size);
4342 pg_vec = alloc_pg_vec(req, order);
4343 if (unlikely(!pg_vec))
4345 switch (po->tp_version) {
4347 /* Block transmit is not supported yet */
4349 init_prb_bdqc(po, rb, pg_vec, req_u);
4351 struct tpacket_req3 *req3 = &req_u->req3;
4353 if (req3->tp_retire_blk_tov ||
4354 req3->tp_sizeof_priv ||
4355 req3->tp_feature_req_word) {
4357 goto out_free_pg_vec;
4368 if (unlikely(req->tp_frame_nr))
4373 /* Detach socket from network */
4374 spin_lock(&po->bind_lock);
4375 was_running = po->running;
4379 __unregister_prot_hook(sk, false);
4381 spin_unlock(&po->bind_lock);
4386 mutex_lock(&po->pg_vec_lock);
4387 if (closing || atomic_read(&po->mapped) == 0) {
4389 spin_lock_bh(&rb_queue->lock);
4390 swap(rb->pg_vec, pg_vec);
4391 rb->frame_max = (req->tp_frame_nr - 1);
4393 rb->frame_size = req->tp_frame_size;
4394 spin_unlock_bh(&rb_queue->lock);
4396 swap(rb->pg_vec_order, order);
4397 swap(rb->pg_vec_len, req->tp_block_nr);
4399 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4400 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4401 tpacket_rcv : packet_rcv;
4402 skb_queue_purge(rb_queue);
4403 if (atomic_read(&po->mapped))
4404 pr_err("packet_mmap: vma is busy: %d\n",
4405 atomic_read(&po->mapped));
4407 mutex_unlock(&po->pg_vec_lock);
4409 spin_lock(&po->bind_lock);
4412 register_prot_hook(sk);
4414 spin_unlock(&po->bind_lock);
4415 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4416 /* Because we don't support block-based V3 on tx-ring */
4418 prb_shutdown_retire_blk_timer(po, rb_queue);
4423 free_pg_vec(pg_vec, order, req->tp_block_nr);
4428 static int packet_mmap(struct file *file, struct socket *sock,
4429 struct vm_area_struct *vma)
4431 struct sock *sk = sock->sk;
4432 struct packet_sock *po = pkt_sk(sk);
4433 unsigned long size, expected_size;
4434 struct packet_ring_buffer *rb;
4435 unsigned long start;
4442 mutex_lock(&po->pg_vec_lock);
4445 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4447 expected_size += rb->pg_vec_len
4453 if (expected_size == 0)
4456 size = vma->vm_end - vma->vm_start;
4457 if (size != expected_size)
4460 start = vma->vm_start;
4461 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4462 if (rb->pg_vec == NULL)
4465 for (i = 0; i < rb->pg_vec_len; i++) {
4467 void *kaddr = rb->pg_vec[i].buffer;
4470 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4471 page = pgv_to_page(kaddr);
4472 err = vm_insert_page(vma, start, page);
4481 atomic_inc(&po->mapped);
4482 vma->vm_ops = &packet_mmap_ops;
4486 mutex_unlock(&po->pg_vec_lock);
4490 static const struct proto_ops packet_ops_spkt = {
4491 .family = PF_PACKET,
4492 .owner = THIS_MODULE,
4493 .release = packet_release,
4494 .bind = packet_bind_spkt,
4495 .connect = sock_no_connect,
4496 .socketpair = sock_no_socketpair,
4497 .accept = sock_no_accept,
4498 .getname = packet_getname_spkt,
4499 .poll = datagram_poll,
4500 .ioctl = packet_ioctl,
4501 .gettstamp = sock_gettstamp,
4502 .listen = sock_no_listen,
4503 .shutdown = sock_no_shutdown,
4504 .setsockopt = sock_no_setsockopt,
4505 .getsockopt = sock_no_getsockopt,
4506 .sendmsg = packet_sendmsg_spkt,
4507 .recvmsg = packet_recvmsg,
4508 .mmap = sock_no_mmap,
4509 .sendpage = sock_no_sendpage,
4512 static const struct proto_ops packet_ops = {
4513 .family = PF_PACKET,
4514 .owner = THIS_MODULE,
4515 .release = packet_release,
4516 .bind = packet_bind,
4517 .connect = sock_no_connect,
4518 .socketpair = sock_no_socketpair,
4519 .accept = sock_no_accept,
4520 .getname = packet_getname,
4521 .poll = packet_poll,
4522 .ioctl = packet_ioctl,
4523 .gettstamp = sock_gettstamp,
4524 .listen = sock_no_listen,
4525 .shutdown = sock_no_shutdown,
4526 .setsockopt = packet_setsockopt,
4527 .getsockopt = packet_getsockopt,
4528 #ifdef CONFIG_COMPAT
4529 .compat_setsockopt = compat_packet_setsockopt,
4531 .sendmsg = packet_sendmsg,
4532 .recvmsg = packet_recvmsg,
4533 .mmap = packet_mmap,
4534 .sendpage = sock_no_sendpage,
4537 static const struct net_proto_family packet_family_ops = {
4538 .family = PF_PACKET,
4539 .create = packet_create,
4540 .owner = THIS_MODULE,
4543 static struct notifier_block packet_netdev_notifier = {
4544 .notifier_call = packet_notifier,
4547 #ifdef CONFIG_PROC_FS
4549 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4552 struct net *net = seq_file_net(seq);
4555 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4558 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4560 struct net *net = seq_file_net(seq);
4561 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4564 static void packet_seq_stop(struct seq_file *seq, void *v)
4570 static int packet_seq_show(struct seq_file *seq, void *v)
4572 if (v == SEQ_START_TOKEN)
4573 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4575 struct sock *s = sk_entry(v);
4576 const struct packet_sock *po = pkt_sk(s);
4579 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4581 refcount_read(&s->sk_refcnt),
4586 atomic_read(&s->sk_rmem_alloc),
4587 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4594 static const struct seq_operations packet_seq_ops = {
4595 .start = packet_seq_start,
4596 .next = packet_seq_next,
4597 .stop = packet_seq_stop,
4598 .show = packet_seq_show,
4602 static int __net_init packet_net_init(struct net *net)
4604 mutex_init(&net->packet.sklist_lock);
4605 INIT_HLIST_HEAD(&net->packet.sklist);
4607 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4608 sizeof(struct seq_net_private)))
4614 static void __net_exit packet_net_exit(struct net *net)
4616 remove_proc_entry("packet", net->proc_net);
4617 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4620 static struct pernet_operations packet_net_ops = {
4621 .init = packet_net_init,
4622 .exit = packet_net_exit,
4626 static void __exit packet_exit(void)
4628 unregister_netdevice_notifier(&packet_netdev_notifier);
4629 unregister_pernet_subsys(&packet_net_ops);
4630 sock_unregister(PF_PACKET);
4631 proto_unregister(&packet_proto);
4634 static int __init packet_init(void)
4638 rc = proto_register(&packet_proto, 0);
4641 rc = sock_register(&packet_family_ops);
4644 rc = register_pernet_subsys(&packet_net_ops);
4647 rc = register_netdevice_notifier(&packet_netdev_notifier);
4654 unregister_pernet_subsys(&packet_net_ops);
4656 sock_unregister(PF_PACKET);
4658 proto_unregister(&packet_proto);
4663 module_init(packet_init);
4664 module_exit(packet_exit);
4665 MODULE_LICENSE("GPL");
4666 MODULE_ALIAS_NETPROTO(PF_PACKET);