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;
548 return DEFAULT_PRB_RETIRE_TOV;
550 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
562 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
563 union tpacket_req_u *req_u)
565 p1->feature_req_word = req_u->req3.tp_feature_req_word;
568 static void init_prb_bdqc(struct packet_sock *po,
569 struct packet_ring_buffer *rb,
571 union tpacket_req_u *req_u)
573 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
574 struct tpacket_block_desc *pbd;
576 memset(p1, 0x0, sizeof(*p1));
578 p1->knxt_seq_num = 1;
580 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
581 p1->pkblk_start = pg_vec[0].buffer;
582 p1->kblk_size = req_u->req3.tp_block_size;
583 p1->knum_blocks = req_u->req3.tp_block_nr;
584 p1->hdrlen = po->tp_hdrlen;
585 p1->version = po->tp_version;
586 p1->last_kactive_blk_num = 0;
587 po->stats.stats3.tp_freeze_q_cnt = 0;
588 if (req_u->req3.tp_retire_blk_tov)
589 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
591 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
592 req_u->req3.tp_block_size);
593 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
594 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
596 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
597 prb_init_ft_ops(p1, req_u);
598 prb_setup_retire_blk_timer(po);
599 prb_open_block(p1, pbd);
602 /* Do NOT update the last_blk_num first.
603 * Assumes sk_buff_head lock is held.
605 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
607 mod_timer(&pkc->retire_blk_timer,
608 jiffies + pkc->tov_in_jiffies);
609 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
614 * 1) We refresh the timer only when we open a block.
615 * By doing this we don't waste cycles refreshing the timer
616 * on packet-by-packet basis.
618 * With a 1MB block-size, on a 1Gbps line, it will take
619 * i) ~8 ms to fill a block + ii) memcpy etc.
620 * In this cut we are not accounting for the memcpy time.
622 * So, if the user sets the 'tmo' to 10ms then the timer
623 * will never fire while the block is still getting filled
624 * (which is what we want). However, the user could choose
625 * to close a block early and that's fine.
627 * But when the timer does fire, we check whether or not to refresh it.
628 * Since the tmo granularity is in msecs, it is not too expensive
629 * to refresh the timer, lets say every '8' msecs.
630 * Either the user can set the 'tmo' or we can derive it based on
631 * a) line-speed and b) block-size.
632 * prb_calc_retire_blk_tmo() calculates the tmo.
635 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
637 struct packet_sock *po =
638 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
639 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
641 struct tpacket_block_desc *pbd;
643 spin_lock(&po->sk.sk_receive_queue.lock);
645 frozen = prb_queue_frozen(pkc);
646 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
648 if (unlikely(pkc->delete_blk_timer))
651 /* We only need to plug the race when the block is partially filled.
653 * lock(); increment BLOCK_NUM_PKTS; unlock()
654 * copy_bits() is in progress ...
655 * timer fires on other cpu:
656 * we can't retire the current block because copy_bits
660 if (BLOCK_NUM_PKTS(pbd)) {
661 while (atomic_read(&pkc->blk_fill_in_prog)) {
662 /* Waiting for skb_copy_bits to finish... */
667 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
669 if (!BLOCK_NUM_PKTS(pbd)) {
670 /* An empty block. Just refresh the timer. */
673 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
674 if (!prb_dispatch_next_block(pkc, po))
679 /* Case 1. Queue was frozen because user-space was
682 if (prb_curr_blk_in_use(pbd)) {
684 * Ok, user-space is still behind.
685 * So just refresh the timer.
689 /* Case 2. queue was frozen,user-space caught up,
690 * now the link went idle && the timer fired.
691 * We don't have a block to close.So we open this
692 * block and restart the timer.
693 * opening a block thaws the queue,restarts timer
694 * Thawing/timer-refresh is a side effect.
696 prb_open_block(pkc, pbd);
703 _prb_refresh_rx_retire_blk_timer(pkc);
706 spin_unlock(&po->sk.sk_receive_queue.lock);
709 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
710 struct tpacket_block_desc *pbd1, __u32 status)
712 /* Flush everything minus the block header */
714 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
719 /* Skip the block header(we know header WILL fit in 4K) */
722 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
723 for (; start < end; start += PAGE_SIZE)
724 flush_dcache_page(pgv_to_page(start));
729 /* Now update the block status. */
731 BLOCK_STATUS(pbd1) = status;
733 /* Flush the block header */
735 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
737 flush_dcache_page(pgv_to_page(start));
747 * 2) Increment active_blk_num
749 * Note:We DONT refresh the timer on purpose.
750 * Because almost always the next block will be opened.
752 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
753 struct tpacket_block_desc *pbd1,
754 struct packet_sock *po, unsigned int stat)
756 __u32 status = TP_STATUS_USER | stat;
758 struct tpacket3_hdr *last_pkt;
759 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
760 struct sock *sk = &po->sk;
762 if (atomic_read(&po->tp_drops))
763 status |= TP_STATUS_LOSING;
765 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
766 last_pkt->tp_next_offset = 0;
768 /* Get the ts of the last pkt */
769 if (BLOCK_NUM_PKTS(pbd1)) {
770 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
771 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
773 /* Ok, we tmo'd - so get the current time.
775 * It shouldn't really happen as we don't close empty
776 * blocks. See prb_retire_rx_blk_timer_expired().
780 h1->ts_last_pkt.ts_sec = ts.tv_sec;
781 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
786 /* Flush the block */
787 prb_flush_block(pkc1, pbd1, status);
789 sk->sk_data_ready(sk);
791 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
794 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
796 pkc->reset_pending_on_curr_blk = 0;
800 * Side effect of opening a block:
802 * 1) prb_queue is thawed.
803 * 2) retire_blk_timer is refreshed.
806 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
807 struct tpacket_block_desc *pbd1)
810 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
814 /* We could have just memset this but we will lose the
815 * flexibility of making the priv area sticky
818 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
819 BLOCK_NUM_PKTS(pbd1) = 0;
820 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
824 h1->ts_first_pkt.ts_sec = ts.tv_sec;
825 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
827 pkc1->pkblk_start = (char *)pbd1;
828 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
831 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
833 pbd1->version = pkc1->version;
834 pkc1->prev = pkc1->nxt_offset;
835 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
837 prb_thaw_queue(pkc1);
838 _prb_refresh_rx_retire_blk_timer(pkc1);
844 * Queue freeze logic:
845 * 1) Assume tp_block_nr = 8 blocks.
846 * 2) At time 't0', user opens Rx ring.
847 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
848 * 4) user-space is either sleeping or processing block '0'.
849 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
850 * it will close block-7,loop around and try to fill block '0'.
852 * __packet_lookup_frame_in_block
853 * prb_retire_current_block()
854 * prb_dispatch_next_block()
855 * |->(BLOCK_STATUS == USER) evaluates to true
856 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
857 * 6) Now there are two cases:
858 * 6.1) Link goes idle right after the queue is frozen.
859 * But remember, the last open_block() refreshed the timer.
860 * When this timer expires,it will refresh itself so that we can
861 * re-open block-0 in near future.
862 * 6.2) Link is busy and keeps on receiving packets. This is a simple
863 * case and __packet_lookup_frame_in_block will check if block-0
864 * is free and can now be re-used.
866 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
867 struct packet_sock *po)
869 pkc->reset_pending_on_curr_blk = 1;
870 po->stats.stats3.tp_freeze_q_cnt++;
873 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
876 * If the next block is free then we will dispatch it
877 * and return a good offset.
878 * Else, we will freeze the queue.
879 * So, caller must check the return value.
881 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
882 struct packet_sock *po)
884 struct tpacket_block_desc *pbd;
888 /* 1. Get current block num */
889 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
891 /* 2. If this block is currently in_use then freeze the queue */
892 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
893 prb_freeze_queue(pkc, po);
899 * open this block and return the offset where the first packet
900 * needs to get stored.
902 prb_open_block(pkc, pbd);
903 return (void *)pkc->nxt_offset;
906 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
907 struct packet_sock *po, unsigned int status)
909 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
911 /* retire/close the current block */
912 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
914 * Plug the case where copy_bits() is in progress on
915 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
916 * have space to copy the pkt in the current block and
917 * called prb_retire_current_block()
919 * We don't need to worry about the TMO case because
920 * the timer-handler already handled this case.
922 if (!(status & TP_STATUS_BLK_TMO)) {
923 while (atomic_read(&pkc->blk_fill_in_prog)) {
924 /* Waiting for skb_copy_bits to finish... */
928 prb_close_block(pkc, pbd, po, status);
933 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
935 return TP_STATUS_USER & BLOCK_STATUS(pbd);
938 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
940 return pkc->reset_pending_on_curr_blk;
943 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
945 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
946 atomic_dec(&pkc->blk_fill_in_prog);
949 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
950 struct tpacket3_hdr *ppd)
952 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
955 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
956 struct tpacket3_hdr *ppd)
958 ppd->hv1.tp_rxhash = 0;
961 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
962 struct tpacket3_hdr *ppd)
964 if (skb_vlan_tag_present(pkc->skb)) {
965 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
966 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
967 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
969 ppd->hv1.tp_vlan_tci = 0;
970 ppd->hv1.tp_vlan_tpid = 0;
971 ppd->tp_status = TP_STATUS_AVAILABLE;
975 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
976 struct tpacket3_hdr *ppd)
978 ppd->hv1.tp_padding = 0;
979 prb_fill_vlan_info(pkc, ppd);
981 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
982 prb_fill_rxhash(pkc, ppd);
984 prb_clear_rxhash(pkc, ppd);
987 static void prb_fill_curr_block(char *curr,
988 struct tpacket_kbdq_core *pkc,
989 struct tpacket_block_desc *pbd,
992 struct tpacket3_hdr *ppd;
994 ppd = (struct tpacket3_hdr *)curr;
995 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
997 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
998 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
999 BLOCK_NUM_PKTS(pbd) += 1;
1000 atomic_inc(&pkc->blk_fill_in_prog);
1001 prb_run_all_ft_ops(pkc, ppd);
1004 /* Assumes caller has the sk->rx_queue.lock */
1005 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1006 struct sk_buff *skb,
1010 struct tpacket_kbdq_core *pkc;
1011 struct tpacket_block_desc *pbd;
1014 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1015 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1017 /* Queue is frozen when user space is lagging behind */
1018 if (prb_queue_frozen(pkc)) {
1020 * Check if that last block which caused the queue to freeze,
1021 * is still in_use by user-space.
1023 if (prb_curr_blk_in_use(pbd)) {
1024 /* Can't record this packet */
1028 * Ok, the block was released by user-space.
1029 * Now let's open that block.
1030 * opening a block also thaws the queue.
1031 * Thawing is a side effect.
1033 prb_open_block(pkc, pbd);
1038 curr = pkc->nxt_offset;
1040 end = (char *)pbd + pkc->kblk_size;
1042 /* first try the current block */
1043 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1044 prb_fill_curr_block(curr, pkc, pbd, len);
1045 return (void *)curr;
1048 /* Ok, close the current block */
1049 prb_retire_current_block(pkc, po, 0);
1051 /* Now, try to dispatch the next block */
1052 curr = (char *)prb_dispatch_next_block(pkc, po);
1054 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1055 prb_fill_curr_block(curr, pkc, pbd, len);
1056 return (void *)curr;
1060 * No free blocks are available.user_space hasn't caught up yet.
1061 * Queue was just frozen and now this packet will get dropped.
1066 static void *packet_current_rx_frame(struct packet_sock *po,
1067 struct sk_buff *skb,
1068 int status, unsigned int len)
1071 switch (po->tp_version) {
1074 curr = packet_lookup_frame(po, &po->rx_ring,
1075 po->rx_ring.head, status);
1078 return __packet_lookup_frame_in_block(po, skb, len);
1080 WARN(1, "TPACKET version not supported\n");
1086 static void *prb_lookup_block(const struct packet_sock *po,
1087 const struct packet_ring_buffer *rb,
1091 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1092 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1094 if (status != BLOCK_STATUS(pbd))
1099 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1102 if (rb->prb_bdqc.kactive_blk_num)
1103 prev = rb->prb_bdqc.kactive_blk_num-1;
1105 prev = rb->prb_bdqc.knum_blocks-1;
1109 /* Assumes caller has held the rx_queue.lock */
1110 static void *__prb_previous_block(struct packet_sock *po,
1111 struct packet_ring_buffer *rb,
1114 unsigned int previous = prb_previous_blk_num(rb);
1115 return prb_lookup_block(po, rb, previous, status);
1118 static void *packet_previous_rx_frame(struct packet_sock *po,
1119 struct packet_ring_buffer *rb,
1122 if (po->tp_version <= TPACKET_V2)
1123 return packet_previous_frame(po, rb, status);
1125 return __prb_previous_block(po, rb, status);
1128 static void packet_increment_rx_head(struct packet_sock *po,
1129 struct packet_ring_buffer *rb)
1131 switch (po->tp_version) {
1134 return packet_increment_head(rb);
1137 WARN(1, "TPACKET version not supported.\n");
1143 static void *packet_previous_frame(struct packet_sock *po,
1144 struct packet_ring_buffer *rb,
1147 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1148 return packet_lookup_frame(po, rb, previous, status);
1151 static void packet_increment_head(struct packet_ring_buffer *buff)
1153 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1156 static void packet_inc_pending(struct packet_ring_buffer *rb)
1158 this_cpu_inc(*rb->pending_refcnt);
1161 static void packet_dec_pending(struct packet_ring_buffer *rb)
1163 this_cpu_dec(*rb->pending_refcnt);
1166 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1168 unsigned int refcnt = 0;
1171 /* We don't use pending refcount in rx_ring. */
1172 if (rb->pending_refcnt == NULL)
1175 for_each_possible_cpu(cpu)
1176 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1181 static int packet_alloc_pending(struct packet_sock *po)
1183 po->rx_ring.pending_refcnt = NULL;
1185 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1186 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1192 static void packet_free_pending(struct packet_sock *po)
1194 free_percpu(po->tx_ring.pending_refcnt);
1197 #define ROOM_POW_OFF 2
1198 #define ROOM_NONE 0x0
1199 #define ROOM_LOW 0x1
1200 #define ROOM_NORMAL 0x2
1202 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1206 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1207 idx = READ_ONCE(po->rx_ring.head);
1209 idx += len >> pow_off;
1212 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1215 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1219 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1220 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1222 idx += len >> pow_off;
1225 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1228 static int __packet_rcv_has_room(const struct packet_sock *po,
1229 const struct sk_buff *skb)
1231 const struct sock *sk = &po->sk;
1232 int ret = ROOM_NONE;
1234 if (po->prot_hook.func != tpacket_rcv) {
1235 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1236 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1237 - (skb ? skb->truesize : 0);
1239 if (avail > (rcvbuf >> ROOM_POW_OFF))
1247 if (po->tp_version == TPACKET_V3) {
1248 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1250 else if (__tpacket_v3_has_room(po, 0))
1253 if (__tpacket_has_room(po, ROOM_POW_OFF))
1255 else if (__tpacket_has_room(po, 0))
1262 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266 ret = __packet_rcv_has_room(po, skb);
1267 pressure = ret != ROOM_NORMAL;
1269 if (READ_ONCE(po->pressure) != pressure)
1270 WRITE_ONCE(po->pressure, pressure);
1275 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1277 if (READ_ONCE(po->pressure) &&
1278 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1279 WRITE_ONCE(po->pressure, 0);
1282 static void packet_sock_destruct(struct sock *sk)
1284 skb_queue_purge(&sk->sk_error_queue);
1286 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1287 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1289 if (!sock_flag(sk, SOCK_DEAD)) {
1290 pr_err("Attempt to release alive packet socket: %p\n", sk);
1294 sk_refcnt_debug_dec(sk);
1297 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1299 u32 *history = po->rollover->history;
1303 rxhash = skb_get_hash(skb);
1304 for (i = 0; i < ROLLOVER_HLEN; i++)
1305 if (READ_ONCE(history[i]) == rxhash)
1308 victim = prandom_u32() % ROLLOVER_HLEN;
1310 /* Avoid dirtying the cache line if possible */
1311 if (READ_ONCE(history[victim]) != rxhash)
1312 WRITE_ONCE(history[victim], rxhash);
1314 return count > (ROLLOVER_HLEN >> 1);
1317 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1318 struct sk_buff *skb,
1321 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1324 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1325 struct sk_buff *skb,
1328 unsigned int val = atomic_inc_return(&f->rr_cur);
1333 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1334 struct sk_buff *skb,
1337 return smp_processor_id() % num;
1340 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1341 struct sk_buff *skb,
1344 return prandom_u32_max(num);
1347 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1348 struct sk_buff *skb,
1349 unsigned int idx, bool try_self,
1352 struct packet_sock *po, *po_next, *po_skip = NULL;
1353 unsigned int i, j, room = ROOM_NONE;
1355 po = pkt_sk(f->arr[idx]);
1358 room = packet_rcv_has_room(po, skb);
1359 if (room == ROOM_NORMAL ||
1360 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1365 i = j = min_t(int, po->rollover->sock, num - 1);
1367 po_next = pkt_sk(f->arr[i]);
1368 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1369 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1371 po->rollover->sock = i;
1372 atomic_long_inc(&po->rollover->num);
1373 if (room == ROOM_LOW)
1374 atomic_long_inc(&po->rollover->num_huge);
1382 atomic_long_inc(&po->rollover->num_failed);
1386 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1387 struct sk_buff *skb,
1390 return skb_get_queue_mapping(skb) % num;
1393 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1394 struct sk_buff *skb,
1397 struct bpf_prog *prog;
1398 unsigned int ret = 0;
1401 prog = rcu_dereference(f->bpf_prog);
1403 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1409 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1411 return f->flags & (flag >> 8);
1414 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1415 struct packet_type *pt, struct net_device *orig_dev)
1417 struct packet_fanout *f = pt->af_packet_priv;
1418 unsigned int num = READ_ONCE(f->num_members);
1419 struct net *net = read_pnet(&f->net);
1420 struct packet_sock *po;
1423 if (!net_eq(dev_net(dev), net) || !num) {
1428 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1429 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1434 case PACKET_FANOUT_HASH:
1436 idx = fanout_demux_hash(f, skb, num);
1438 case PACKET_FANOUT_LB:
1439 idx = fanout_demux_lb(f, skb, num);
1441 case PACKET_FANOUT_CPU:
1442 idx = fanout_demux_cpu(f, skb, num);
1444 case PACKET_FANOUT_RND:
1445 idx = fanout_demux_rnd(f, skb, num);
1447 case PACKET_FANOUT_QM:
1448 idx = fanout_demux_qm(f, skb, num);
1450 case PACKET_FANOUT_ROLLOVER:
1451 idx = fanout_demux_rollover(f, skb, 0, false, num);
1453 case PACKET_FANOUT_CBPF:
1454 case PACKET_FANOUT_EBPF:
1455 idx = fanout_demux_bpf(f, skb, num);
1459 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1460 idx = fanout_demux_rollover(f, skb, idx, true, num);
1462 po = pkt_sk(f->arr[idx]);
1463 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1466 DEFINE_MUTEX(fanout_mutex);
1467 EXPORT_SYMBOL_GPL(fanout_mutex);
1468 static LIST_HEAD(fanout_list);
1469 static u16 fanout_next_id;
1471 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1473 struct packet_fanout *f = po->fanout;
1475 spin_lock(&f->lock);
1476 f->arr[f->num_members] = sk;
1479 if (f->num_members == 1)
1480 dev_add_pack(&f->prot_hook);
1481 spin_unlock(&f->lock);
1484 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1486 struct packet_fanout *f = po->fanout;
1489 spin_lock(&f->lock);
1490 for (i = 0; i < f->num_members; i++) {
1491 if (f->arr[i] == sk)
1494 BUG_ON(i >= f->num_members);
1495 f->arr[i] = f->arr[f->num_members - 1];
1497 if (f->num_members == 0)
1498 __dev_remove_pack(&f->prot_hook);
1499 spin_unlock(&f->lock);
1502 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1504 if (sk->sk_family != PF_PACKET)
1507 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1510 static void fanout_init_data(struct packet_fanout *f)
1513 case PACKET_FANOUT_LB:
1514 atomic_set(&f->rr_cur, 0);
1516 case PACKET_FANOUT_CBPF:
1517 case PACKET_FANOUT_EBPF:
1518 RCU_INIT_POINTER(f->bpf_prog, NULL);
1523 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1525 struct bpf_prog *old;
1527 spin_lock(&f->lock);
1528 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1529 rcu_assign_pointer(f->bpf_prog, new);
1530 spin_unlock(&f->lock);
1534 bpf_prog_destroy(old);
1538 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1541 struct bpf_prog *new;
1542 struct sock_fprog fprog;
1545 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1547 if (len != sizeof(fprog))
1549 if (copy_from_user(&fprog, data, len))
1552 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1556 __fanout_set_data_bpf(po->fanout, new);
1560 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1563 struct bpf_prog *new;
1566 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1568 if (len != sizeof(fd))
1570 if (copy_from_user(&fd, data, len))
1573 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1575 return PTR_ERR(new);
1577 __fanout_set_data_bpf(po->fanout, new);
1581 static int fanout_set_data(struct packet_sock *po, char __user *data,
1584 switch (po->fanout->type) {
1585 case PACKET_FANOUT_CBPF:
1586 return fanout_set_data_cbpf(po, data, len);
1587 case PACKET_FANOUT_EBPF:
1588 return fanout_set_data_ebpf(po, data, len);
1594 static void fanout_release_data(struct packet_fanout *f)
1597 case PACKET_FANOUT_CBPF:
1598 case PACKET_FANOUT_EBPF:
1599 __fanout_set_data_bpf(f, NULL);
1603 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1605 struct packet_fanout *f;
1607 list_for_each_entry(f, &fanout_list, list) {
1608 if (f->id == candidate_id &&
1609 read_pnet(&f->net) == sock_net(sk)) {
1616 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1618 u16 id = fanout_next_id;
1621 if (__fanout_id_is_free(sk, id)) {
1623 fanout_next_id = id + 1;
1628 } while (id != fanout_next_id);
1633 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1635 struct packet_rollover *rollover = NULL;
1636 struct packet_sock *po = pkt_sk(sk);
1637 struct packet_fanout *f, *match;
1638 u8 type = type_flags & 0xff;
1639 u8 flags = type_flags >> 8;
1643 case PACKET_FANOUT_ROLLOVER:
1644 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1646 case PACKET_FANOUT_HASH:
1647 case PACKET_FANOUT_LB:
1648 case PACKET_FANOUT_CPU:
1649 case PACKET_FANOUT_RND:
1650 case PACKET_FANOUT_QM:
1651 case PACKET_FANOUT_CBPF:
1652 case PACKET_FANOUT_EBPF:
1658 mutex_lock(&fanout_mutex);
1664 if (type == PACKET_FANOUT_ROLLOVER ||
1665 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1667 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1670 atomic_long_set(&rollover->num, 0);
1671 atomic_long_set(&rollover->num_huge, 0);
1672 atomic_long_set(&rollover->num_failed, 0);
1675 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1680 if (!fanout_find_new_id(sk, &id)) {
1684 /* ephemeral flag for the first socket in the group: drop it */
1685 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1689 list_for_each_entry(f, &fanout_list, list) {
1691 read_pnet(&f->net) == sock_net(sk)) {
1697 if (match && match->flags != flags)
1701 match = kzalloc(sizeof(*match), GFP_KERNEL);
1704 write_pnet(&match->net, sock_net(sk));
1707 match->flags = flags;
1708 INIT_LIST_HEAD(&match->list);
1709 spin_lock_init(&match->lock);
1710 refcount_set(&match->sk_ref, 0);
1711 fanout_init_data(match);
1712 match->prot_hook.type = po->prot_hook.type;
1713 match->prot_hook.dev = po->prot_hook.dev;
1714 match->prot_hook.func = packet_rcv_fanout;
1715 match->prot_hook.af_packet_priv = match;
1716 match->prot_hook.id_match = match_fanout_group;
1717 list_add(&match->list, &fanout_list);
1721 spin_lock(&po->bind_lock);
1723 match->type == type &&
1724 match->prot_hook.type == po->prot_hook.type &&
1725 match->prot_hook.dev == po->prot_hook.dev) {
1727 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1728 __dev_remove_pack(&po->prot_hook);
1730 po->rollover = rollover;
1732 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1733 __fanout_link(sk, po);
1737 spin_unlock(&po->bind_lock);
1739 if (err && !refcount_read(&match->sk_ref)) {
1740 list_del(&match->list);
1746 mutex_unlock(&fanout_mutex);
1750 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1751 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1752 * It is the responsibility of the caller to call fanout_release_data() and
1753 * free the returned packet_fanout (after synchronize_net())
1755 static struct packet_fanout *fanout_release(struct sock *sk)
1757 struct packet_sock *po = pkt_sk(sk);
1758 struct packet_fanout *f;
1760 mutex_lock(&fanout_mutex);
1765 if (refcount_dec_and_test(&f->sk_ref))
1770 mutex_unlock(&fanout_mutex);
1775 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1776 struct sk_buff *skb)
1778 /* Earlier code assumed this would be a VLAN pkt, double-check
1779 * this now that we have the actual packet in hand. We can only
1780 * do this check on Ethernet devices.
1782 if (unlikely(dev->type != ARPHRD_ETHER))
1785 skb_reset_mac_header(skb);
1786 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1789 static const struct proto_ops packet_ops;
1791 static const struct proto_ops packet_ops_spkt;
1793 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1794 struct packet_type *pt, struct net_device *orig_dev)
1797 struct sockaddr_pkt *spkt;
1800 * When we registered the protocol we saved the socket in the data
1801 * field for just this event.
1804 sk = pt->af_packet_priv;
1807 * Yank back the headers [hope the device set this
1808 * right or kerboom...]
1810 * Incoming packets have ll header pulled,
1813 * For outgoing ones skb->data == skb_mac_header(skb)
1814 * so that this procedure is noop.
1817 if (skb->pkt_type == PACKET_LOOPBACK)
1820 if (!net_eq(dev_net(dev), sock_net(sk)))
1823 skb = skb_share_check(skb, GFP_ATOMIC);
1827 /* drop any routing info */
1830 /* drop conntrack reference */
1833 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1835 skb_push(skb, skb->data - skb_mac_header(skb));
1838 * The SOCK_PACKET socket receives _all_ frames.
1841 spkt->spkt_family = dev->type;
1842 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1843 spkt->spkt_protocol = skb->protocol;
1846 * Charge the memory to the socket. This is done specifically
1847 * to prevent sockets using all the memory up.
1850 if (sock_queue_rcv_skb(sk, skb) == 0)
1859 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1861 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1862 sock->type == SOCK_RAW) {
1863 skb_reset_mac_header(skb);
1864 skb->protocol = dev_parse_header_protocol(skb);
1867 skb_probe_transport_header(skb);
1871 * Output a raw packet to a device layer. This bypasses all the other
1872 * protocol layers and you must therefore supply it with a complete frame
1875 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1878 struct sock *sk = sock->sk;
1879 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1880 struct sk_buff *skb = NULL;
1881 struct net_device *dev;
1882 struct sockcm_cookie sockc;
1888 * Get and verify the address.
1892 if (msg->msg_namelen < sizeof(struct sockaddr))
1894 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1895 proto = saddr->spkt_protocol;
1897 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1900 * Find the device first to size check it
1903 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1906 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1912 if (!(dev->flags & IFF_UP))
1916 * You may not queue a frame bigger than the mtu. This is the lowest level
1917 * raw protocol and you must do your own fragmentation at this level.
1920 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1921 if (!netif_supports_nofcs(dev)) {
1922 err = -EPROTONOSUPPORT;
1925 extra_len = 4; /* We're doing our own CRC */
1929 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1933 size_t reserved = LL_RESERVED_SPACE(dev);
1934 int tlen = dev->needed_tailroom;
1935 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1938 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1941 /* FIXME: Save some space for broken drivers that write a hard
1942 * header at transmission time by themselves. PPP is the notable
1943 * one here. This should really be fixed at the driver level.
1945 skb_reserve(skb, reserved);
1946 skb_reset_network_header(skb);
1948 /* Try to align data part correctly */
1953 skb_reset_network_header(skb);
1955 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1961 if (!dev_validate_header(dev, skb->data, len)) {
1965 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1966 !packet_extra_vlan_len_allowed(dev, skb)) {
1971 sockcm_init(&sockc, sk);
1972 if (msg->msg_controllen) {
1973 err = sock_cmsg_send(sk, msg, &sockc);
1978 skb->protocol = proto;
1980 skb->priority = sk->sk_priority;
1981 skb->mark = sk->sk_mark;
1982 skb->tstamp = sockc.transmit_time;
1984 skb_setup_tx_timestamp(skb, sockc.tsflags);
1986 if (unlikely(extra_len == 4))
1989 packet_parse_headers(skb, sock);
1991 dev_queue_xmit(skb);
2002 static unsigned int run_filter(struct sk_buff *skb,
2003 const struct sock *sk,
2006 struct sk_filter *filter;
2009 filter = rcu_dereference(sk->sk_filter);
2011 res = bpf_prog_run_clear_cb(filter->prog, skb);
2017 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2020 struct virtio_net_hdr vnet_hdr;
2022 if (*len < sizeof(vnet_hdr))
2024 *len -= sizeof(vnet_hdr);
2026 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2029 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2033 * This function makes lazy skb cloning in hope that most of packets
2034 * are discarded by BPF.
2036 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2037 * and skb->cb are mangled. It works because (and until) packets
2038 * falling here are owned by current CPU. Output packets are cloned
2039 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2040 * sequencially, so that if we return skb to original state on exit,
2041 * we will not harm anyone.
2044 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2045 struct packet_type *pt, struct net_device *orig_dev)
2048 struct sockaddr_ll *sll;
2049 struct packet_sock *po;
2050 u8 *skb_head = skb->data;
2051 int skb_len = skb->len;
2052 unsigned int snaplen, res;
2053 bool is_drop_n_account = false;
2055 if (skb->pkt_type == PACKET_LOOPBACK)
2058 sk = pt->af_packet_priv;
2061 if (!net_eq(dev_net(dev), sock_net(sk)))
2066 if (dev->header_ops) {
2067 /* The device has an explicit notion of ll header,
2068 * exported to higher levels.
2070 * Otherwise, the device hides details of its frame
2071 * structure, so that corresponding packet head is
2072 * never delivered to user.
2074 if (sk->sk_type != SOCK_DGRAM)
2075 skb_push(skb, skb->data - skb_mac_header(skb));
2076 else if (skb->pkt_type == PACKET_OUTGOING) {
2077 /* Special case: outgoing packets have ll header at head */
2078 skb_pull(skb, skb_network_offset(skb));
2084 res = run_filter(skb, sk, snaplen);
2086 goto drop_n_restore;
2090 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2093 if (skb_shared(skb)) {
2094 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2098 if (skb_head != skb->data) {
2099 skb->data = skb_head;
2106 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2108 sll = &PACKET_SKB_CB(skb)->sa.ll;
2109 sll->sll_hatype = dev->type;
2110 sll->sll_pkttype = skb->pkt_type;
2111 if (unlikely(po->origdev))
2112 sll->sll_ifindex = orig_dev->ifindex;
2114 sll->sll_ifindex = dev->ifindex;
2116 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2118 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2119 * Use their space for storing the original skb length.
2121 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2123 if (pskb_trim(skb, snaplen))
2126 skb_set_owner_r(skb, sk);
2130 /* drop conntrack reference */
2133 spin_lock(&sk->sk_receive_queue.lock);
2134 po->stats.stats1.tp_packets++;
2135 sock_skb_set_dropcount(sk, skb);
2136 __skb_queue_tail(&sk->sk_receive_queue, skb);
2137 spin_unlock(&sk->sk_receive_queue.lock);
2138 sk->sk_data_ready(sk);
2142 is_drop_n_account = true;
2143 atomic_inc(&po->tp_drops);
2144 atomic_inc(&sk->sk_drops);
2147 if (skb_head != skb->data && skb_shared(skb)) {
2148 skb->data = skb_head;
2152 if (!is_drop_n_account)
2159 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2160 struct packet_type *pt, struct net_device *orig_dev)
2163 struct packet_sock *po;
2164 struct sockaddr_ll *sll;
2165 union tpacket_uhdr h;
2166 u8 *skb_head = skb->data;
2167 int skb_len = skb->len;
2168 unsigned int snaplen, res;
2169 unsigned long status = TP_STATUS_USER;
2170 unsigned short macoff, netoff, hdrlen;
2171 struct sk_buff *copy_skb = NULL;
2174 bool is_drop_n_account = false;
2175 bool do_vnet = false;
2177 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2178 * We may add members to them until current aligned size without forcing
2179 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2181 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2182 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2184 if (skb->pkt_type == PACKET_LOOPBACK)
2187 sk = pt->af_packet_priv;
2190 if (!net_eq(dev_net(dev), sock_net(sk)))
2193 if (dev->header_ops) {
2194 if (sk->sk_type != SOCK_DGRAM)
2195 skb_push(skb, skb->data - skb_mac_header(skb));
2196 else if (skb->pkt_type == PACKET_OUTGOING) {
2197 /* Special case: outgoing packets have ll header at head */
2198 skb_pull(skb, skb_network_offset(skb));
2204 res = run_filter(skb, sk, snaplen);
2206 goto drop_n_restore;
2208 /* If we are flooded, just give up */
2209 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2210 atomic_inc(&po->tp_drops);
2211 goto drop_n_restore;
2214 if (skb->ip_summed == CHECKSUM_PARTIAL)
2215 status |= TP_STATUS_CSUMNOTREADY;
2216 else if (skb->pkt_type != PACKET_OUTGOING &&
2217 (skb->ip_summed == CHECKSUM_COMPLETE ||
2218 skb_csum_unnecessary(skb)))
2219 status |= TP_STATUS_CSUM_VALID;
2224 if (sk->sk_type == SOCK_DGRAM) {
2225 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2228 unsigned int maclen = skb_network_offset(skb);
2229 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2230 (maclen < 16 ? 16 : maclen)) +
2232 if (po->has_vnet_hdr) {
2233 netoff += sizeof(struct virtio_net_hdr);
2236 macoff = netoff - maclen;
2238 if (po->tp_version <= TPACKET_V2) {
2239 if (macoff + snaplen > po->rx_ring.frame_size) {
2240 if (po->copy_thresh &&
2241 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2242 if (skb_shared(skb)) {
2243 copy_skb = skb_clone(skb, GFP_ATOMIC);
2245 copy_skb = skb_get(skb);
2246 skb_head = skb->data;
2249 skb_set_owner_r(copy_skb, sk);
2251 snaplen = po->rx_ring.frame_size - macoff;
2252 if ((int)snaplen < 0) {
2257 } else if (unlikely(macoff + snaplen >
2258 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2261 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2262 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2263 snaplen, nval, macoff);
2265 if (unlikely((int)snaplen < 0)) {
2267 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2271 spin_lock(&sk->sk_receive_queue.lock);
2272 h.raw = packet_current_rx_frame(po, skb,
2273 TP_STATUS_KERNEL, (macoff+snaplen));
2275 goto drop_n_account;
2276 if (po->tp_version <= TPACKET_V2) {
2277 packet_increment_rx_head(po, &po->rx_ring);
2279 * LOSING will be reported till you read the stats,
2280 * because it's COR - Clear On Read.
2281 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2284 if (atomic_read(&po->tp_drops))
2285 status |= TP_STATUS_LOSING;
2289 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2290 sizeof(struct virtio_net_hdr),
2292 goto drop_n_account;
2294 po->stats.stats1.tp_packets++;
2296 status |= TP_STATUS_COPY;
2297 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2299 spin_unlock(&sk->sk_receive_queue.lock);
2301 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2303 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2304 getnstimeofday(&ts);
2306 status |= ts_status;
2308 switch (po->tp_version) {
2310 h.h1->tp_len = skb->len;
2311 h.h1->tp_snaplen = snaplen;
2312 h.h1->tp_mac = macoff;
2313 h.h1->tp_net = netoff;
2314 h.h1->tp_sec = ts.tv_sec;
2315 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2316 hdrlen = sizeof(*h.h1);
2319 h.h2->tp_len = skb->len;
2320 h.h2->tp_snaplen = snaplen;
2321 h.h2->tp_mac = macoff;
2322 h.h2->tp_net = netoff;
2323 h.h2->tp_sec = ts.tv_sec;
2324 h.h2->tp_nsec = ts.tv_nsec;
2325 if (skb_vlan_tag_present(skb)) {
2326 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2327 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2328 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2330 h.h2->tp_vlan_tci = 0;
2331 h.h2->tp_vlan_tpid = 0;
2333 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2334 hdrlen = sizeof(*h.h2);
2337 /* tp_nxt_offset,vlan are already populated above.
2338 * So DONT clear those fields here
2340 h.h3->tp_status |= status;
2341 h.h3->tp_len = skb->len;
2342 h.h3->tp_snaplen = snaplen;
2343 h.h3->tp_mac = macoff;
2344 h.h3->tp_net = netoff;
2345 h.h3->tp_sec = ts.tv_sec;
2346 h.h3->tp_nsec = ts.tv_nsec;
2347 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2348 hdrlen = sizeof(*h.h3);
2354 sll = h.raw + TPACKET_ALIGN(hdrlen);
2355 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2356 sll->sll_family = AF_PACKET;
2357 sll->sll_hatype = dev->type;
2358 sll->sll_protocol = skb->protocol;
2359 sll->sll_pkttype = skb->pkt_type;
2360 if (unlikely(po->origdev))
2361 sll->sll_ifindex = orig_dev->ifindex;
2363 sll->sll_ifindex = dev->ifindex;
2367 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2368 if (po->tp_version <= TPACKET_V2) {
2371 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2374 for (start = h.raw; start < end; start += PAGE_SIZE)
2375 flush_dcache_page(pgv_to_page(start));
2380 if (po->tp_version <= TPACKET_V2) {
2381 __packet_set_status(po, h.raw, status);
2382 sk->sk_data_ready(sk);
2384 prb_clear_blk_fill_status(&po->rx_ring);
2388 if (skb_head != skb->data && skb_shared(skb)) {
2389 skb->data = skb_head;
2393 if (!is_drop_n_account)
2400 spin_unlock(&sk->sk_receive_queue.lock);
2401 atomic_inc(&po->tp_drops);
2402 is_drop_n_account = true;
2404 sk->sk_data_ready(sk);
2405 kfree_skb(copy_skb);
2406 goto drop_n_restore;
2409 static void tpacket_destruct_skb(struct sk_buff *skb)
2411 struct packet_sock *po = pkt_sk(skb->sk);
2413 if (likely(po->tx_ring.pg_vec)) {
2417 ph = skb_zcopy_get_nouarg(skb);
2418 packet_dec_pending(&po->tx_ring);
2420 ts = __packet_set_timestamp(po, ph, skb);
2421 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2423 if (!packet_read_pending(&po->tx_ring))
2424 complete(&po->skb_completion);
2430 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2432 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2433 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2434 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2435 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2436 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2437 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2438 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2440 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2446 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2447 struct virtio_net_hdr *vnet_hdr)
2449 if (*len < sizeof(*vnet_hdr))
2451 *len -= sizeof(*vnet_hdr);
2453 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2456 return __packet_snd_vnet_parse(vnet_hdr, *len);
2459 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2460 void *frame, struct net_device *dev, void *data, int tp_len,
2461 __be16 proto, unsigned char *addr, int hlen, int copylen,
2462 const struct sockcm_cookie *sockc)
2464 union tpacket_uhdr ph;
2465 int to_write, offset, len, nr_frags, len_max;
2466 struct socket *sock = po->sk.sk_socket;
2472 skb->protocol = proto;
2474 skb->priority = po->sk.sk_priority;
2475 skb->mark = po->sk.sk_mark;
2476 skb->tstamp = sockc->transmit_time;
2477 skb_setup_tx_timestamp(skb, sockc->tsflags);
2478 skb_zcopy_set_nouarg(skb, ph.raw);
2480 skb_reserve(skb, hlen);
2481 skb_reset_network_header(skb);
2485 if (sock->type == SOCK_DGRAM) {
2486 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2488 if (unlikely(err < 0))
2490 } else if (copylen) {
2491 int hdrlen = min_t(int, copylen, tp_len);
2493 skb_push(skb, dev->hard_header_len);
2494 skb_put(skb, copylen - dev->hard_header_len);
2495 err = skb_store_bits(skb, 0, data, hdrlen);
2498 if (!dev_validate_header(dev, skb->data, hdrlen))
2505 offset = offset_in_page(data);
2506 len_max = PAGE_SIZE - offset;
2507 len = ((to_write > len_max) ? len_max : to_write);
2509 skb->data_len = to_write;
2510 skb->len += to_write;
2511 skb->truesize += to_write;
2512 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2514 while (likely(to_write)) {
2515 nr_frags = skb_shinfo(skb)->nr_frags;
2517 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2518 pr_err("Packet exceed the number of skb frags(%lu)\n",
2523 page = pgv_to_page(data);
2525 flush_dcache_page(page);
2527 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2530 len_max = PAGE_SIZE;
2531 len = ((to_write > len_max) ? len_max : to_write);
2534 packet_parse_headers(skb, sock);
2539 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2540 int size_max, void **data)
2542 union tpacket_uhdr ph;
2547 switch (po->tp_version) {
2549 if (ph.h3->tp_next_offset != 0) {
2550 pr_warn_once("variable sized slot not supported");
2553 tp_len = ph.h3->tp_len;
2556 tp_len = ph.h2->tp_len;
2559 tp_len = ph.h1->tp_len;
2562 if (unlikely(tp_len > size_max)) {
2563 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2567 if (unlikely(po->tp_tx_has_off)) {
2568 int off_min, off_max;
2570 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2571 off_max = po->tx_ring.frame_size - tp_len;
2572 if (po->sk.sk_type == SOCK_DGRAM) {
2573 switch (po->tp_version) {
2575 off = ph.h3->tp_net;
2578 off = ph.h2->tp_net;
2581 off = ph.h1->tp_net;
2585 switch (po->tp_version) {
2587 off = ph.h3->tp_mac;
2590 off = ph.h2->tp_mac;
2593 off = ph.h1->tp_mac;
2597 if (unlikely((off < off_min) || (off_max < off)))
2600 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2603 *data = frame + off;
2607 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2609 struct sk_buff *skb = NULL;
2610 struct net_device *dev;
2611 struct virtio_net_hdr *vnet_hdr = NULL;
2612 struct sockcm_cookie sockc;
2614 int err, reserve = 0;
2616 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2617 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2618 unsigned char *addr = NULL;
2619 int tp_len, size_max;
2622 int status = TP_STATUS_AVAILABLE;
2623 int hlen, tlen, copylen = 0;
2626 mutex_lock(&po->pg_vec_lock);
2628 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2629 * we need to confirm it under protection of pg_vec_lock.
2631 if (unlikely(!po->tx_ring.pg_vec)) {
2635 if (likely(saddr == NULL)) {
2636 dev = packet_cached_dev_get(po);
2640 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2642 if (msg->msg_namelen < (saddr->sll_halen
2643 + offsetof(struct sockaddr_ll,
2646 proto = saddr->sll_protocol;
2647 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2648 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2649 if (dev && msg->msg_namelen < dev->addr_len +
2650 offsetof(struct sockaddr_ll, sll_addr))
2652 addr = saddr->sll_addr;
2657 if (unlikely(dev == NULL))
2660 if (unlikely(!(dev->flags & IFF_UP)))
2663 sockcm_init(&sockc, &po->sk);
2664 if (msg->msg_controllen) {
2665 err = sock_cmsg_send(&po->sk, msg, &sockc);
2670 if (po->sk.sk_socket->type == SOCK_RAW)
2671 reserve = dev->hard_header_len;
2672 size_max = po->tx_ring.frame_size
2673 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2675 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2676 size_max = dev->mtu + reserve + VLAN_HLEN;
2678 reinit_completion(&po->skb_completion);
2681 ph = packet_current_frame(po, &po->tx_ring,
2682 TP_STATUS_SEND_REQUEST);
2683 if (unlikely(ph == NULL)) {
2684 if (need_wait && skb) {
2685 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2686 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2688 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2692 /* check for additional frames */
2697 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2701 status = TP_STATUS_SEND_REQUEST;
2702 hlen = LL_RESERVED_SPACE(dev);
2703 tlen = dev->needed_tailroom;
2704 if (po->has_vnet_hdr) {
2706 data += sizeof(*vnet_hdr);
2707 tp_len -= sizeof(*vnet_hdr);
2709 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2713 copylen = __virtio16_to_cpu(vio_le(),
2716 copylen = max_t(int, copylen, dev->hard_header_len);
2717 skb = sock_alloc_send_skb(&po->sk,
2718 hlen + tlen + sizeof(struct sockaddr_ll) +
2719 (copylen - dev->hard_header_len),
2722 if (unlikely(skb == NULL)) {
2723 /* we assume the socket was initially writeable ... */
2724 if (likely(len_sum > 0))
2728 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2729 addr, hlen, copylen, &sockc);
2730 if (likely(tp_len >= 0) &&
2731 tp_len > dev->mtu + reserve &&
2732 !po->has_vnet_hdr &&
2733 !packet_extra_vlan_len_allowed(dev, skb))
2736 if (unlikely(tp_len < 0)) {
2739 __packet_set_status(po, ph,
2740 TP_STATUS_AVAILABLE);
2741 packet_increment_head(&po->tx_ring);
2745 status = TP_STATUS_WRONG_FORMAT;
2751 if (po->has_vnet_hdr) {
2752 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2756 virtio_net_hdr_set_proto(skb, vnet_hdr);
2759 skb->destructor = tpacket_destruct_skb;
2760 __packet_set_status(po, ph, TP_STATUS_SENDING);
2761 packet_inc_pending(&po->tx_ring);
2763 status = TP_STATUS_SEND_REQUEST;
2764 err = po->xmit(skb);
2765 if (unlikely(err > 0)) {
2766 err = net_xmit_errno(err);
2767 if (err && __packet_get_status(po, ph) ==
2768 TP_STATUS_AVAILABLE) {
2769 /* skb was destructed already */
2774 * skb was dropped but not destructed yet;
2775 * let's treat it like congestion or err < 0
2779 packet_increment_head(&po->tx_ring);
2781 } while (likely((ph != NULL) ||
2782 /* Note: packet_read_pending() might be slow if we have
2783 * to call it as it's per_cpu variable, but in fast-path
2784 * we already short-circuit the loop with the first
2785 * condition, and luckily don't have to go that path
2788 (need_wait && packet_read_pending(&po->tx_ring))));
2794 __packet_set_status(po, ph, status);
2799 mutex_unlock(&po->pg_vec_lock);
2803 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2804 size_t reserve, size_t len,
2805 size_t linear, int noblock,
2808 struct sk_buff *skb;
2810 /* Under a page? Don't bother with paged skb. */
2811 if (prepad + len < PAGE_SIZE || !linear)
2814 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2819 skb_reserve(skb, reserve);
2820 skb_put(skb, linear);
2821 skb->data_len = len - linear;
2822 skb->len += len - linear;
2827 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2829 struct sock *sk = sock->sk;
2830 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2831 struct sk_buff *skb;
2832 struct net_device *dev;
2834 unsigned char *addr = NULL;
2835 int err, reserve = 0;
2836 struct sockcm_cookie sockc;
2837 struct virtio_net_hdr vnet_hdr = { 0 };
2839 struct packet_sock *po = pkt_sk(sk);
2840 bool has_vnet_hdr = false;
2841 int hlen, tlen, linear;
2845 * Get and verify the address.
2848 if (likely(saddr == NULL)) {
2849 dev = packet_cached_dev_get(po);
2853 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2855 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2857 proto = saddr->sll_protocol;
2858 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2859 if (sock->type == SOCK_DGRAM) {
2860 if (dev && msg->msg_namelen < dev->addr_len +
2861 offsetof(struct sockaddr_ll, sll_addr))
2863 addr = saddr->sll_addr;
2868 if (unlikely(dev == NULL))
2871 if (unlikely(!(dev->flags & IFF_UP)))
2874 sockcm_init(&sockc, sk);
2875 sockc.mark = sk->sk_mark;
2876 if (msg->msg_controllen) {
2877 err = sock_cmsg_send(sk, msg, &sockc);
2882 if (sock->type == SOCK_RAW)
2883 reserve = dev->hard_header_len;
2884 if (po->has_vnet_hdr) {
2885 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2888 has_vnet_hdr = true;
2891 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2892 if (!netif_supports_nofcs(dev)) {
2893 err = -EPROTONOSUPPORT;
2896 extra_len = 4; /* We're doing our own CRC */
2900 if (!vnet_hdr.gso_type &&
2901 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2905 hlen = LL_RESERVED_SPACE(dev);
2906 tlen = dev->needed_tailroom;
2907 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2908 linear = max(linear, min_t(int, len, dev->hard_header_len));
2909 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2910 msg->msg_flags & MSG_DONTWAIT, &err);
2914 skb_reset_network_header(skb);
2917 if (sock->type == SOCK_DGRAM) {
2918 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2919 if (unlikely(offset < 0))
2921 } else if (reserve) {
2922 skb_reserve(skb, -reserve);
2923 if (len < reserve + sizeof(struct ipv6hdr) &&
2924 dev->min_header_len != dev->hard_header_len)
2925 skb_reset_network_header(skb);
2928 /* Returns -EFAULT on error */
2929 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2933 if (sock->type == SOCK_RAW &&
2934 !dev_validate_header(dev, skb->data, len)) {
2939 skb_setup_tx_timestamp(skb, sockc.tsflags);
2941 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2942 !packet_extra_vlan_len_allowed(dev, skb)) {
2947 skb->protocol = proto;
2949 skb->priority = sk->sk_priority;
2950 skb->mark = sockc.mark;
2951 skb->tstamp = sockc.transmit_time;
2954 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2957 len += sizeof(vnet_hdr);
2958 virtio_net_hdr_set_proto(skb, &vnet_hdr);
2961 packet_parse_headers(skb, sock);
2963 if (unlikely(extra_len == 4))
2966 err = po->xmit(skb);
2967 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2983 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2985 struct sock *sk = sock->sk;
2986 struct packet_sock *po = pkt_sk(sk);
2988 if (po->tx_ring.pg_vec)
2989 return tpacket_snd(po, msg);
2991 return packet_snd(sock, msg, len);
2995 * Close a PACKET socket. This is fairly simple. We immediately go
2996 * to 'closed' state and remove our protocol entry in the device list.
2999 static int packet_release(struct socket *sock)
3001 struct sock *sk = sock->sk;
3002 struct packet_sock *po;
3003 struct packet_fanout *f;
3005 union tpacket_req_u req_u;
3013 mutex_lock(&net->packet.sklist_lock);
3014 sk_del_node_init_rcu(sk);
3015 mutex_unlock(&net->packet.sklist_lock);
3018 sock_prot_inuse_add(net, sk->sk_prot, -1);
3021 spin_lock(&po->bind_lock);
3022 unregister_prot_hook(sk, false);
3023 packet_cached_dev_reset(po);
3025 if (po->prot_hook.dev) {
3026 dev_put(po->prot_hook.dev);
3027 po->prot_hook.dev = NULL;
3029 spin_unlock(&po->bind_lock);
3031 packet_flush_mclist(sk);
3034 if (po->rx_ring.pg_vec) {
3035 memset(&req_u, 0, sizeof(req_u));
3036 packet_set_ring(sk, &req_u, 1, 0);
3039 if (po->tx_ring.pg_vec) {
3040 memset(&req_u, 0, sizeof(req_u));
3041 packet_set_ring(sk, &req_u, 1, 1);
3045 f = fanout_release(sk);
3049 kfree(po->rollover);
3051 fanout_release_data(f);
3055 * Now the socket is dead. No more input will appear.
3062 skb_queue_purge(&sk->sk_receive_queue);
3063 packet_free_pending(po);
3064 sk_refcnt_debug_release(sk);
3071 * Attach a packet hook.
3074 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3077 struct packet_sock *po = pkt_sk(sk);
3078 struct net_device *dev_curr;
3081 struct net_device *dev = NULL;
3083 bool unlisted = false;
3086 spin_lock(&po->bind_lock);
3095 dev = dev_get_by_name_rcu(sock_net(sk), name);
3100 } else if (ifindex) {
3101 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3111 proto_curr = po->prot_hook.type;
3112 dev_curr = po->prot_hook.dev;
3114 need_rehook = proto_curr != proto || dev_curr != dev;
3119 /* prevents packet_notifier() from calling
3120 * register_prot_hook()
3123 __unregister_prot_hook(sk, true);
3125 dev_curr = po->prot_hook.dev;
3127 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3131 BUG_ON(po->running);
3133 po->prot_hook.type = proto;
3135 if (unlikely(unlisted)) {
3137 po->prot_hook.dev = NULL;
3139 packet_cached_dev_reset(po);
3141 po->prot_hook.dev = dev;
3142 po->ifindex = dev ? dev->ifindex : 0;
3143 packet_cached_dev_assign(po, dev);
3149 if (proto == 0 || !need_rehook)
3152 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3153 register_prot_hook(sk);
3155 sk->sk_err = ENETDOWN;
3156 if (!sock_flag(sk, SOCK_DEAD))
3157 sk->sk_error_report(sk);
3162 spin_unlock(&po->bind_lock);
3168 * Bind a packet socket to a device
3171 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3174 struct sock *sk = sock->sk;
3175 char name[sizeof(uaddr->sa_data) + 1];
3181 if (addr_len != sizeof(struct sockaddr))
3183 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3186 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3187 name[sizeof(uaddr->sa_data)] = 0;
3189 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3192 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3194 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3195 struct sock *sk = sock->sk;
3201 if (addr_len < sizeof(struct sockaddr_ll))
3203 if (sll->sll_family != AF_PACKET)
3206 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3207 sll->sll_protocol ? : pkt_sk(sk)->num);
3210 static struct proto packet_proto = {
3212 .owner = THIS_MODULE,
3213 .obj_size = sizeof(struct packet_sock),
3217 * Create a packet of type SOCK_PACKET.
3220 static int packet_create(struct net *net, struct socket *sock, int protocol,
3224 struct packet_sock *po;
3225 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3228 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3230 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3231 sock->type != SOCK_PACKET)
3232 return -ESOCKTNOSUPPORT;
3234 sock->state = SS_UNCONNECTED;
3237 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3241 sock->ops = &packet_ops;
3242 if (sock->type == SOCK_PACKET)
3243 sock->ops = &packet_ops_spkt;
3245 sock_init_data(sock, sk);
3248 init_completion(&po->skb_completion);
3249 sk->sk_family = PF_PACKET;
3251 po->xmit = dev_queue_xmit;
3253 err = packet_alloc_pending(po);
3257 packet_cached_dev_reset(po);
3259 sk->sk_destruct = packet_sock_destruct;
3260 sk_refcnt_debug_inc(sk);
3263 * Attach a protocol block
3266 spin_lock_init(&po->bind_lock);
3267 mutex_init(&po->pg_vec_lock);
3268 po->rollover = NULL;
3269 po->prot_hook.func = packet_rcv;
3271 if (sock->type == SOCK_PACKET)
3272 po->prot_hook.func = packet_rcv_spkt;
3274 po->prot_hook.af_packet_priv = sk;
3277 po->prot_hook.type = proto;
3278 __register_prot_hook(sk);
3281 mutex_lock(&net->packet.sklist_lock);
3282 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3283 mutex_unlock(&net->packet.sklist_lock);
3286 sock_prot_inuse_add(net, &packet_proto, 1);
3297 * Pull a packet from our receive queue and hand it to the user.
3298 * If necessary we block.
3301 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3304 struct sock *sk = sock->sk;
3305 struct sk_buff *skb;
3307 int vnet_hdr_len = 0;
3308 unsigned int origlen = 0;
3311 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3315 /* What error should we return now? EUNATTACH? */
3316 if (pkt_sk(sk)->ifindex < 0)
3320 if (flags & MSG_ERRQUEUE) {
3321 err = sock_recv_errqueue(sk, msg, len,
3322 SOL_PACKET, PACKET_TX_TIMESTAMP);
3327 * Call the generic datagram receiver. This handles all sorts
3328 * of horrible races and re-entrancy so we can forget about it
3329 * in the protocol layers.
3331 * Now it will return ENETDOWN, if device have just gone down,
3332 * but then it will block.
3335 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3338 * An error occurred so return it. Because skb_recv_datagram()
3339 * handles the blocking we don't see and worry about blocking
3346 packet_rcv_try_clear_pressure(pkt_sk(sk));
3348 if (pkt_sk(sk)->has_vnet_hdr) {
3349 err = packet_rcv_vnet(msg, skb, &len);
3352 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3355 /* You lose any data beyond the buffer you gave. If it worries
3356 * a user program they can ask the device for its MTU
3362 msg->msg_flags |= MSG_TRUNC;
3365 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3369 if (sock->type != SOCK_PACKET) {
3370 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3372 /* Original length was stored in sockaddr_ll fields */
3373 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3374 sll->sll_family = AF_PACKET;
3375 sll->sll_protocol = skb->protocol;
3378 sock_recv_ts_and_drops(msg, sk, skb);
3380 if (msg->msg_name) {
3383 /* If the address length field is there to be filled
3384 * in, we fill it in now.
3386 if (sock->type == SOCK_PACKET) {
3387 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3388 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3389 copy_len = msg->msg_namelen;
3391 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3393 msg->msg_namelen = sll->sll_halen +
3394 offsetof(struct sockaddr_ll, sll_addr);
3395 copy_len = msg->msg_namelen;
3396 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3397 memset(msg->msg_name +
3398 offsetof(struct sockaddr_ll, sll_addr),
3399 0, sizeof(sll->sll_addr));
3400 msg->msg_namelen = sizeof(struct sockaddr_ll);
3403 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3406 if (pkt_sk(sk)->auxdata) {
3407 struct tpacket_auxdata aux;
3409 aux.tp_status = TP_STATUS_USER;
3410 if (skb->ip_summed == CHECKSUM_PARTIAL)
3411 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3412 else if (skb->pkt_type != PACKET_OUTGOING &&
3413 (skb->ip_summed == CHECKSUM_COMPLETE ||
3414 skb_csum_unnecessary(skb)))
3415 aux.tp_status |= TP_STATUS_CSUM_VALID;
3417 aux.tp_len = origlen;
3418 aux.tp_snaplen = skb->len;
3420 aux.tp_net = skb_network_offset(skb);
3421 if (skb_vlan_tag_present(skb)) {
3422 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3423 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3424 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3426 aux.tp_vlan_tci = 0;
3427 aux.tp_vlan_tpid = 0;
3429 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3433 * Free or return the buffer as appropriate. Again this
3434 * hides all the races and re-entrancy issues from us.
3436 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3439 skb_free_datagram(sk, skb);
3444 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3447 struct net_device *dev;
3448 struct sock *sk = sock->sk;
3453 uaddr->sa_family = AF_PACKET;
3454 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3456 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3458 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3461 return sizeof(*uaddr);
3464 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3467 struct net_device *dev;
3468 struct sock *sk = sock->sk;
3469 struct packet_sock *po = pkt_sk(sk);
3470 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3475 sll->sll_family = AF_PACKET;
3476 sll->sll_ifindex = po->ifindex;
3477 sll->sll_protocol = po->num;
3478 sll->sll_pkttype = 0;
3480 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3482 sll->sll_hatype = dev->type;
3483 sll->sll_halen = dev->addr_len;
3484 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3486 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3491 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3494 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3498 case PACKET_MR_MULTICAST:
3499 if (i->alen != dev->addr_len)
3502 return dev_mc_add(dev, i->addr);
3504 return dev_mc_del(dev, i->addr);
3506 case PACKET_MR_PROMISC:
3507 return dev_set_promiscuity(dev, what);
3508 case PACKET_MR_ALLMULTI:
3509 return dev_set_allmulti(dev, what);
3510 case PACKET_MR_UNICAST:
3511 if (i->alen != dev->addr_len)
3514 return dev_uc_add(dev, i->addr);
3516 return dev_uc_del(dev, i->addr);
3524 static void packet_dev_mclist_delete(struct net_device *dev,
3525 struct packet_mclist **mlp)
3527 struct packet_mclist *ml;
3529 while ((ml = *mlp) != NULL) {
3530 if (ml->ifindex == dev->ifindex) {
3531 packet_dev_mc(dev, ml, -1);
3539 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3541 struct packet_sock *po = pkt_sk(sk);
3542 struct packet_mclist *ml, *i;
3543 struct net_device *dev;
3549 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3554 if (mreq->mr_alen > dev->addr_len)
3558 i = kmalloc(sizeof(*i), GFP_KERNEL);
3563 for (ml = po->mclist; ml; ml = ml->next) {
3564 if (ml->ifindex == mreq->mr_ifindex &&
3565 ml->type == mreq->mr_type &&
3566 ml->alen == mreq->mr_alen &&
3567 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3569 /* Free the new element ... */
3575 i->type = mreq->mr_type;
3576 i->ifindex = mreq->mr_ifindex;
3577 i->alen = mreq->mr_alen;
3578 memcpy(i->addr, mreq->mr_address, i->alen);
3579 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3581 i->next = po->mclist;
3583 err = packet_dev_mc(dev, i, 1);
3585 po->mclist = i->next;
3594 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3596 struct packet_mclist *ml, **mlp;
3600 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3601 if (ml->ifindex == mreq->mr_ifindex &&
3602 ml->type == mreq->mr_type &&
3603 ml->alen == mreq->mr_alen &&
3604 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3605 if (--ml->count == 0) {
3606 struct net_device *dev;
3608 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3610 packet_dev_mc(dev, ml, -1);
3620 static void packet_flush_mclist(struct sock *sk)
3622 struct packet_sock *po = pkt_sk(sk);
3623 struct packet_mclist *ml;
3629 while ((ml = po->mclist) != NULL) {
3630 struct net_device *dev;
3632 po->mclist = ml->next;
3633 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3635 packet_dev_mc(dev, ml, -1);
3642 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3644 struct sock *sk = sock->sk;
3645 struct packet_sock *po = pkt_sk(sk);
3648 if (level != SOL_PACKET)
3649 return -ENOPROTOOPT;
3652 case PACKET_ADD_MEMBERSHIP:
3653 case PACKET_DROP_MEMBERSHIP:
3655 struct packet_mreq_max mreq;
3657 memset(&mreq, 0, sizeof(mreq));
3658 if (len < sizeof(struct packet_mreq))
3660 if (len > sizeof(mreq))
3662 if (copy_from_user(&mreq, optval, len))
3664 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3666 if (optname == PACKET_ADD_MEMBERSHIP)
3667 ret = packet_mc_add(sk, &mreq);
3669 ret = packet_mc_drop(sk, &mreq);
3673 case PACKET_RX_RING:
3674 case PACKET_TX_RING:
3676 union tpacket_req_u req_u;
3680 switch (po->tp_version) {
3683 len = sizeof(req_u.req);
3687 len = sizeof(req_u.req3);
3693 if (copy_from_user(&req_u.req, optval, len))
3696 ret = packet_set_ring(sk, &req_u, 0,
3697 optname == PACKET_TX_RING);
3702 case PACKET_COPY_THRESH:
3706 if (optlen != sizeof(val))
3708 if (copy_from_user(&val, optval, sizeof(val)))
3711 pkt_sk(sk)->copy_thresh = val;
3714 case PACKET_VERSION:
3718 if (optlen != sizeof(val))
3720 if (copy_from_user(&val, optval, sizeof(val)))
3731 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3734 po->tp_version = val;
3740 case PACKET_RESERVE:
3744 if (optlen != sizeof(val))
3746 if (copy_from_user(&val, optval, sizeof(val)))
3751 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3754 po->tp_reserve = val;
3764 if (optlen != sizeof(val))
3766 if (copy_from_user(&val, optval, sizeof(val)))
3770 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3773 po->tp_loss = !!val;
3779 case PACKET_AUXDATA:
3783 if (optlen < sizeof(val))
3785 if (copy_from_user(&val, optval, sizeof(val)))
3789 po->auxdata = !!val;
3793 case PACKET_ORIGDEV:
3797 if (optlen < sizeof(val))
3799 if (copy_from_user(&val, optval, sizeof(val)))
3803 po->origdev = !!val;
3807 case PACKET_VNET_HDR:
3811 if (sock->type != SOCK_RAW)
3813 if (optlen < sizeof(val))
3815 if (copy_from_user(&val, optval, sizeof(val)))
3819 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3822 po->has_vnet_hdr = !!val;
3828 case PACKET_TIMESTAMP:
3832 if (optlen != sizeof(val))
3834 if (copy_from_user(&val, optval, sizeof(val)))
3837 po->tp_tstamp = val;
3844 if (optlen != sizeof(val))
3846 if (copy_from_user(&val, optval, sizeof(val)))
3849 return fanout_add(sk, val & 0xffff, val >> 16);
3851 case PACKET_FANOUT_DATA:
3856 return fanout_set_data(po, optval, optlen);
3858 case PACKET_IGNORE_OUTGOING:
3862 if (optlen != sizeof(val))
3864 if (copy_from_user(&val, optval, sizeof(val)))
3866 if (val < 0 || val > 1)
3869 po->prot_hook.ignore_outgoing = !!val;
3872 case PACKET_TX_HAS_OFF:
3876 if (optlen != sizeof(val))
3878 if (copy_from_user(&val, optval, sizeof(val)))
3882 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3885 po->tp_tx_has_off = !!val;
3891 case PACKET_QDISC_BYPASS:
3895 if (optlen != sizeof(val))
3897 if (copy_from_user(&val, optval, sizeof(val)))
3900 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3904 return -ENOPROTOOPT;
3908 static int packet_getsockopt(struct socket *sock, int level, int optname,
3909 char __user *optval, int __user *optlen)
3912 int val, lv = sizeof(val);
3913 struct sock *sk = sock->sk;
3914 struct packet_sock *po = pkt_sk(sk);
3916 union tpacket_stats_u st;
3917 struct tpacket_rollover_stats rstats;
3920 if (level != SOL_PACKET)
3921 return -ENOPROTOOPT;
3923 if (get_user(len, optlen))
3930 case PACKET_STATISTICS:
3931 spin_lock_bh(&sk->sk_receive_queue.lock);
3932 memcpy(&st, &po->stats, sizeof(st));
3933 memset(&po->stats, 0, sizeof(po->stats));
3934 spin_unlock_bh(&sk->sk_receive_queue.lock);
3935 drops = atomic_xchg(&po->tp_drops, 0);
3937 if (po->tp_version == TPACKET_V3) {
3938 lv = sizeof(struct tpacket_stats_v3);
3939 st.stats3.tp_drops = drops;
3940 st.stats3.tp_packets += drops;
3943 lv = sizeof(struct tpacket_stats);
3944 st.stats1.tp_drops = drops;
3945 st.stats1.tp_packets += drops;
3950 case PACKET_AUXDATA:
3953 case PACKET_ORIGDEV:
3956 case PACKET_VNET_HDR:
3957 val = po->has_vnet_hdr;
3959 case PACKET_VERSION:
3960 val = po->tp_version;
3963 if (len > sizeof(int))
3965 if (len < sizeof(int))
3967 if (copy_from_user(&val, optval, len))
3971 val = sizeof(struct tpacket_hdr);
3974 val = sizeof(struct tpacket2_hdr);
3977 val = sizeof(struct tpacket3_hdr);
3983 case PACKET_RESERVE:
3984 val = po->tp_reserve;
3989 case PACKET_TIMESTAMP:
3990 val = po->tp_tstamp;
3994 ((u32)po->fanout->id |
3995 ((u32)po->fanout->type << 16) |
3996 ((u32)po->fanout->flags << 24)) :
3999 case PACKET_IGNORE_OUTGOING:
4000 val = po->prot_hook.ignore_outgoing;
4002 case PACKET_ROLLOVER_STATS:
4005 rstats.tp_all = atomic_long_read(&po->rollover->num);
4006 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4007 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4009 lv = sizeof(rstats);
4011 case PACKET_TX_HAS_OFF:
4012 val = po->tp_tx_has_off;
4014 case PACKET_QDISC_BYPASS:
4015 val = packet_use_direct_xmit(po);
4018 return -ENOPROTOOPT;
4023 if (put_user(len, optlen))
4025 if (copy_to_user(optval, data, len))
4031 #ifdef CONFIG_COMPAT
4032 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
4033 char __user *optval, unsigned int optlen)
4035 struct packet_sock *po = pkt_sk(sock->sk);
4037 if (level != SOL_PACKET)
4038 return -ENOPROTOOPT;
4040 if (optname == PACKET_FANOUT_DATA &&
4041 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
4042 optval = (char __user *)get_compat_bpf_fprog(optval);
4045 optlen = sizeof(struct sock_fprog);
4048 return packet_setsockopt(sock, level, optname, optval, optlen);
4052 static int packet_notifier(struct notifier_block *this,
4053 unsigned long msg, void *ptr)
4056 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4057 struct net *net = dev_net(dev);
4060 sk_for_each_rcu(sk, &net->packet.sklist) {
4061 struct packet_sock *po = pkt_sk(sk);
4064 case NETDEV_UNREGISTER:
4066 packet_dev_mclist_delete(dev, &po->mclist);
4070 if (dev->ifindex == po->ifindex) {
4071 spin_lock(&po->bind_lock);
4073 __unregister_prot_hook(sk, false);
4074 sk->sk_err = ENETDOWN;
4075 if (!sock_flag(sk, SOCK_DEAD))
4076 sk->sk_error_report(sk);
4078 if (msg == NETDEV_UNREGISTER) {
4079 packet_cached_dev_reset(po);
4081 if (po->prot_hook.dev)
4082 dev_put(po->prot_hook.dev);
4083 po->prot_hook.dev = NULL;
4085 spin_unlock(&po->bind_lock);
4089 if (dev->ifindex == po->ifindex) {
4090 spin_lock(&po->bind_lock);
4092 register_prot_hook(sk);
4093 spin_unlock(&po->bind_lock);
4103 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4106 struct sock *sk = sock->sk;
4111 int amount = sk_wmem_alloc_get(sk);
4113 return put_user(amount, (int __user *)arg);
4117 struct sk_buff *skb;
4120 spin_lock_bh(&sk->sk_receive_queue.lock);
4121 skb = skb_peek(&sk->sk_receive_queue);
4124 spin_unlock_bh(&sk->sk_receive_queue.lock);
4125 return put_user(amount, (int __user *)arg);
4135 case SIOCGIFBRDADDR:
4136 case SIOCSIFBRDADDR:
4137 case SIOCGIFNETMASK:
4138 case SIOCSIFNETMASK:
4139 case SIOCGIFDSTADDR:
4140 case SIOCSIFDSTADDR:
4142 return inet_dgram_ops.ioctl(sock, cmd, arg);
4146 return -ENOIOCTLCMD;
4151 static __poll_t packet_poll(struct file *file, struct socket *sock,
4154 struct sock *sk = sock->sk;
4155 struct packet_sock *po = pkt_sk(sk);
4156 __poll_t mask = datagram_poll(file, sock, wait);
4158 spin_lock_bh(&sk->sk_receive_queue.lock);
4159 if (po->rx_ring.pg_vec) {
4160 if (!packet_previous_rx_frame(po, &po->rx_ring,
4162 mask |= EPOLLIN | EPOLLRDNORM;
4164 packet_rcv_try_clear_pressure(po);
4165 spin_unlock_bh(&sk->sk_receive_queue.lock);
4166 spin_lock_bh(&sk->sk_write_queue.lock);
4167 if (po->tx_ring.pg_vec) {
4168 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4169 mask |= EPOLLOUT | EPOLLWRNORM;
4171 spin_unlock_bh(&sk->sk_write_queue.lock);
4176 /* Dirty? Well, I still did not learn better way to account
4180 static void packet_mm_open(struct vm_area_struct *vma)
4182 struct file *file = vma->vm_file;
4183 struct socket *sock = file->private_data;
4184 struct sock *sk = sock->sk;
4187 atomic_inc(&pkt_sk(sk)->mapped);
4190 static void packet_mm_close(struct vm_area_struct *vma)
4192 struct file *file = vma->vm_file;
4193 struct socket *sock = file->private_data;
4194 struct sock *sk = sock->sk;
4197 atomic_dec(&pkt_sk(sk)->mapped);
4200 static const struct vm_operations_struct packet_mmap_ops = {
4201 .open = packet_mm_open,
4202 .close = packet_mm_close,
4205 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4210 for (i = 0; i < len; i++) {
4211 if (likely(pg_vec[i].buffer)) {
4212 if (is_vmalloc_addr(pg_vec[i].buffer))
4213 vfree(pg_vec[i].buffer);
4215 free_pages((unsigned long)pg_vec[i].buffer,
4217 pg_vec[i].buffer = NULL;
4223 static char *alloc_one_pg_vec_page(unsigned long order)
4226 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4227 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4229 buffer = (char *) __get_free_pages(gfp_flags, order);
4233 /* __get_free_pages failed, fall back to vmalloc */
4234 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4238 /* vmalloc failed, lets dig into swap here */
4239 gfp_flags &= ~__GFP_NORETRY;
4240 buffer = (char *) __get_free_pages(gfp_flags, order);
4244 /* complete and utter failure */
4248 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4250 unsigned int block_nr = req->tp_block_nr;
4254 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4255 if (unlikely(!pg_vec))
4258 for (i = 0; i < block_nr; i++) {
4259 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4260 if (unlikely(!pg_vec[i].buffer))
4261 goto out_free_pgvec;
4268 free_pg_vec(pg_vec, order, block_nr);
4273 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4274 int closing, int tx_ring)
4276 struct pgv *pg_vec = NULL;
4277 struct packet_sock *po = pkt_sk(sk);
4278 int was_running, order = 0;
4279 struct packet_ring_buffer *rb;
4280 struct sk_buff_head *rb_queue;
4283 /* Added to avoid minimal code churn */
4284 struct tpacket_req *req = &req_u->req;
4286 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4287 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4291 if (atomic_read(&po->mapped))
4293 if (packet_read_pending(rb))
4297 if (req->tp_block_nr) {
4298 unsigned int min_frame_size;
4300 /* Sanity tests and some calculations */
4302 if (unlikely(rb->pg_vec))
4305 switch (po->tp_version) {
4307 po->tp_hdrlen = TPACKET_HDRLEN;
4310 po->tp_hdrlen = TPACKET2_HDRLEN;
4313 po->tp_hdrlen = TPACKET3_HDRLEN;
4318 if (unlikely((int)req->tp_block_size <= 0))
4320 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4322 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4323 if (po->tp_version >= TPACKET_V3 &&
4324 req->tp_block_size <
4325 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4327 if (unlikely(req->tp_frame_size < min_frame_size))
4329 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4332 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4333 if (unlikely(rb->frames_per_block == 0))
4335 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4337 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4342 order = get_order(req->tp_block_size);
4343 pg_vec = alloc_pg_vec(req, order);
4344 if (unlikely(!pg_vec))
4346 switch (po->tp_version) {
4348 /* Block transmit is not supported yet */
4350 init_prb_bdqc(po, rb, pg_vec, req_u);
4352 struct tpacket_req3 *req3 = &req_u->req3;
4354 if (req3->tp_retire_blk_tov ||
4355 req3->tp_sizeof_priv ||
4356 req3->tp_feature_req_word) {
4358 goto out_free_pg_vec;
4369 if (unlikely(req->tp_frame_nr))
4374 /* Detach socket from network */
4375 spin_lock(&po->bind_lock);
4376 was_running = po->running;
4380 __unregister_prot_hook(sk, false);
4382 spin_unlock(&po->bind_lock);
4387 mutex_lock(&po->pg_vec_lock);
4388 if (closing || atomic_read(&po->mapped) == 0) {
4390 spin_lock_bh(&rb_queue->lock);
4391 swap(rb->pg_vec, pg_vec);
4392 rb->frame_max = (req->tp_frame_nr - 1);
4394 rb->frame_size = req->tp_frame_size;
4395 spin_unlock_bh(&rb_queue->lock);
4397 swap(rb->pg_vec_order, order);
4398 swap(rb->pg_vec_len, req->tp_block_nr);
4400 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4401 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4402 tpacket_rcv : packet_rcv;
4403 skb_queue_purge(rb_queue);
4404 if (atomic_read(&po->mapped))
4405 pr_err("packet_mmap: vma is busy: %d\n",
4406 atomic_read(&po->mapped));
4408 mutex_unlock(&po->pg_vec_lock);
4410 spin_lock(&po->bind_lock);
4413 register_prot_hook(sk);
4415 spin_unlock(&po->bind_lock);
4416 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4417 /* Because we don't support block-based V3 on tx-ring */
4419 prb_shutdown_retire_blk_timer(po, rb_queue);
4424 free_pg_vec(pg_vec, order, req->tp_block_nr);
4429 static int packet_mmap(struct file *file, struct socket *sock,
4430 struct vm_area_struct *vma)
4432 struct sock *sk = sock->sk;
4433 struct packet_sock *po = pkt_sk(sk);
4434 unsigned long size, expected_size;
4435 struct packet_ring_buffer *rb;
4436 unsigned long start;
4443 mutex_lock(&po->pg_vec_lock);
4446 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4448 expected_size += rb->pg_vec_len
4454 if (expected_size == 0)
4457 size = vma->vm_end - vma->vm_start;
4458 if (size != expected_size)
4461 start = vma->vm_start;
4462 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4463 if (rb->pg_vec == NULL)
4466 for (i = 0; i < rb->pg_vec_len; i++) {
4468 void *kaddr = rb->pg_vec[i].buffer;
4471 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4472 page = pgv_to_page(kaddr);
4473 err = vm_insert_page(vma, start, page);
4482 atomic_inc(&po->mapped);
4483 vma->vm_ops = &packet_mmap_ops;
4487 mutex_unlock(&po->pg_vec_lock);
4491 static const struct proto_ops packet_ops_spkt = {
4492 .family = PF_PACKET,
4493 .owner = THIS_MODULE,
4494 .release = packet_release,
4495 .bind = packet_bind_spkt,
4496 .connect = sock_no_connect,
4497 .socketpair = sock_no_socketpair,
4498 .accept = sock_no_accept,
4499 .getname = packet_getname_spkt,
4500 .poll = datagram_poll,
4501 .ioctl = packet_ioctl,
4502 .gettstamp = sock_gettstamp,
4503 .listen = sock_no_listen,
4504 .shutdown = sock_no_shutdown,
4505 .setsockopt = sock_no_setsockopt,
4506 .getsockopt = sock_no_getsockopt,
4507 .sendmsg = packet_sendmsg_spkt,
4508 .recvmsg = packet_recvmsg,
4509 .mmap = sock_no_mmap,
4510 .sendpage = sock_no_sendpage,
4513 static const struct proto_ops packet_ops = {
4514 .family = PF_PACKET,
4515 .owner = THIS_MODULE,
4516 .release = packet_release,
4517 .bind = packet_bind,
4518 .connect = sock_no_connect,
4519 .socketpair = sock_no_socketpair,
4520 .accept = sock_no_accept,
4521 .getname = packet_getname,
4522 .poll = packet_poll,
4523 .ioctl = packet_ioctl,
4524 .gettstamp = sock_gettstamp,
4525 .listen = sock_no_listen,
4526 .shutdown = sock_no_shutdown,
4527 .setsockopt = packet_setsockopt,
4528 .getsockopt = packet_getsockopt,
4529 #ifdef CONFIG_COMPAT
4530 .compat_setsockopt = compat_packet_setsockopt,
4532 .sendmsg = packet_sendmsg,
4533 .recvmsg = packet_recvmsg,
4534 .mmap = packet_mmap,
4535 .sendpage = sock_no_sendpage,
4538 static const struct net_proto_family packet_family_ops = {
4539 .family = PF_PACKET,
4540 .create = packet_create,
4541 .owner = THIS_MODULE,
4544 static struct notifier_block packet_netdev_notifier = {
4545 .notifier_call = packet_notifier,
4548 #ifdef CONFIG_PROC_FS
4550 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4553 struct net *net = seq_file_net(seq);
4556 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4559 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4561 struct net *net = seq_file_net(seq);
4562 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4565 static void packet_seq_stop(struct seq_file *seq, void *v)
4571 static int packet_seq_show(struct seq_file *seq, void *v)
4573 if (v == SEQ_START_TOKEN)
4574 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4576 struct sock *s = sk_entry(v);
4577 const struct packet_sock *po = pkt_sk(s);
4580 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4582 refcount_read(&s->sk_refcnt),
4587 atomic_read(&s->sk_rmem_alloc),
4588 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4595 static const struct seq_operations packet_seq_ops = {
4596 .start = packet_seq_start,
4597 .next = packet_seq_next,
4598 .stop = packet_seq_stop,
4599 .show = packet_seq_show,
4603 static int __net_init packet_net_init(struct net *net)
4605 mutex_init(&net->packet.sklist_lock);
4606 INIT_HLIST_HEAD(&net->packet.sklist);
4608 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4609 sizeof(struct seq_net_private)))
4615 static void __net_exit packet_net_exit(struct net *net)
4617 remove_proc_entry("packet", net->proc_net);
4618 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4621 static struct pernet_operations packet_net_ops = {
4622 .init = packet_net_init,
4623 .exit = packet_net_exit,
4627 static void __exit packet_exit(void)
4629 unregister_netdevice_notifier(&packet_netdev_notifier);
4630 unregister_pernet_subsys(&packet_net_ops);
4631 sock_unregister(PF_PACKET);
4632 proto_unregister(&packet_proto);
4635 static int __init packet_init(void)
4639 rc = proto_register(&packet_proto, 0);
4642 rc = sock_register(&packet_family_ops);
4645 rc = register_pernet_subsys(&packet_net_ops);
4648 rc = register_netdevice_notifier(&packet_netdev_notifier);
4655 unregister_pernet_subsys(&packet_net_ops);
4657 sock_unregister(PF_PACKET);
4659 proto_unregister(&packet_proto);
4664 module_init(packet_init);
4665 module_exit(packet_exit);
4666 MODULE_LICENSE("GPL");
4667 MODULE_ALIAS_NETPROTO(PF_PACKET);