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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
51 #include <linux/ethtool.h>
52 #include <linux/filter.h>
53 #include <linux/types.h>
55 #include <linux/capability.h>
56 #include <linux/fcntl.h>
57 #include <linux/socket.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/if_packet.h>
62 #include <linux/wireless.h>
63 #include <linux/kernel.h>
64 #include <linux/kmod.h>
65 #include <linux/slab.h>
66 #include <linux/vmalloc.h>
67 #include <net/net_namespace.h>
69 #include <net/protocol.h>
70 #include <linux/skbuff.h>
72 #include <linux/errno.h>
73 #include <linux/timer.h>
74 #include <linux/uaccess.h>
75 #include <asm/ioctls.h>
77 #include <asm/cacheflush.h>
79 #include <linux/proc_fs.h>
80 #include <linux/seq_file.h>
81 #include <linux/poll.h>
82 #include <linux/module.h>
83 #include <linux/init.h>
84 #include <linux/mutex.h>
85 #include <linux/if_vlan.h>
86 #include <linux/virtio_net.h>
87 #include <linux/errqueue.h>
88 #include <linux/net_tstamp.h>
89 #include <linux/percpu.h>
91 #include <net/inet_common.h>
93 #include <linux/bpf.h>
94 #include <net/compat.h>
95 #include <linux/netfilter_netdev.h>
101 - If the device has no dev->header_ops->create, there is no LL header
102 visible above the device. In this case, its hard_header_len should be 0.
103 The device may prepend its own header internally. In this case, its
104 needed_headroom should be set to the space needed for it to add its
106 For example, a WiFi driver pretending to be an Ethernet driver should
107 set its hard_header_len to be the Ethernet header length, and set its
108 needed_headroom to be (the real WiFi header length - the fake Ethernet
110 - packet socket receives packets with pulled ll header,
111 so that SOCK_RAW should push it back.
116 Incoming, dev_has_header(dev) == true
117 mac_header -> ll header
120 Outgoing, dev_has_header(dev) == true
121 mac_header -> ll header
124 Incoming, dev_has_header(dev) == false
126 However drivers often make it point to the ll header.
127 This is incorrect because the ll header should be invisible to us.
130 Outgoing, dev_has_header(dev) == false
131 mac_header -> data. ll header is invisible to us.
135 If dev_has_header(dev) == false we are unable to restore the ll header,
136 because it is invisible to us.
142 dev_has_header(dev) == true
143 mac_header -> ll header
146 dev_has_header(dev) == false (ll header is invisible to us)
150 We should set network_header on output to the correct position,
151 packet classifier depends on it.
154 /* Private packet socket structures. */
156 /* identical to struct packet_mreq except it has
157 * a longer address field.
159 struct packet_mreq_max {
161 unsigned short mr_type;
162 unsigned short mr_alen;
163 unsigned char mr_address[MAX_ADDR_LEN];
167 struct tpacket_hdr *h1;
168 struct tpacket2_hdr *h2;
169 struct tpacket3_hdr *h3;
173 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
174 int closing, int tx_ring);
176 #define V3_ALIGNMENT (8)
178 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
180 #define BLK_PLUS_PRIV(sz_of_priv) \
181 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
183 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
184 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
185 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
186 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
187 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
188 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
194 static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
209 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
210 struct tpacket3_hdr *);
211 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
212 struct tpacket3_hdr *);
213 static void packet_flush_mclist(struct sock *sk);
214 static u16 packet_pick_tx_queue(struct sk_buff *skb);
216 struct packet_skb_cb {
218 struct sockaddr_pkt pkt;
220 /* Trick: alias skb original length with
221 * ll.sll_family and ll.protocol in order
224 unsigned int origlen;
225 struct sockaddr_ll ll;
230 #define vio_le() virtio_legacy_is_little_endian()
232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246 #ifdef CONFIG_NETFILTER_EGRESS
247 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
249 struct sk_buff *next, *head = NULL, *tail;
253 for (; skb != NULL; skb = next) {
255 skb_mark_not_on_list(skb);
257 if (!nf_hook_egress(skb, &rc, skb->dev))
273 static int packet_direct_xmit(struct sk_buff *skb)
275 #ifdef CONFIG_NETFILTER_EGRESS
276 if (nf_hook_egress_active()) {
277 skb = nf_hook_direct_egress(skb);
279 return NET_XMIT_DROP;
282 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
285 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287 struct net_device *dev;
290 dev = rcu_dereference(po->cached_dev);
297 static void packet_cached_dev_assign(struct packet_sock *po,
298 struct net_device *dev)
300 rcu_assign_pointer(po->cached_dev, dev);
303 static void packet_cached_dev_reset(struct packet_sock *po)
305 RCU_INIT_POINTER(po->cached_dev, NULL);
308 static bool packet_use_direct_xmit(const struct packet_sock *po)
310 return po->xmit == packet_direct_xmit;
313 static u16 packet_pick_tx_queue(struct sk_buff *skb)
315 struct net_device *dev = skb->dev;
316 const struct net_device_ops *ops = dev->netdev_ops;
317 int cpu = raw_smp_processor_id();
321 skb->sender_cpu = cpu + 1;
323 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
324 if (ops->ndo_select_queue) {
325 queue_index = ops->ndo_select_queue(dev, skb, NULL);
326 queue_index = netdev_cap_txqueue(dev, queue_index);
328 queue_index = netdev_pick_tx(dev, skb, NULL);
334 /* __register_prot_hook must be invoked through register_prot_hook
335 * or from a context in which asynchronous accesses to the packet
336 * socket is not possible (packet_create()).
338 static void __register_prot_hook(struct sock *sk)
340 struct packet_sock *po = pkt_sk(sk);
344 __fanout_link(sk, po);
346 dev_add_pack(&po->prot_hook);
353 static void register_prot_hook(struct sock *sk)
355 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
356 __register_prot_hook(sk);
359 /* If the sync parameter is true, we will temporarily drop
360 * the po->bind_lock and do a synchronize_net to make sure no
361 * asynchronous packet processing paths still refer to the elements
362 * of po->prot_hook. If the sync parameter is false, it is the
363 * callers responsibility to take care of this.
365 static void __unregister_prot_hook(struct sock *sk, bool sync)
367 struct packet_sock *po = pkt_sk(sk);
369 lockdep_assert_held_once(&po->bind_lock);
374 __fanout_unlink(sk, po);
376 __dev_remove_pack(&po->prot_hook);
381 spin_unlock(&po->bind_lock);
383 spin_lock(&po->bind_lock);
387 static void unregister_prot_hook(struct sock *sk, bool sync)
389 struct packet_sock *po = pkt_sk(sk);
392 __unregister_prot_hook(sk, sync);
395 static inline struct page * __pure pgv_to_page(void *addr)
397 if (is_vmalloc_addr(addr))
398 return vmalloc_to_page(addr);
399 return virt_to_page(addr);
402 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
404 union tpacket_uhdr h;
407 switch (po->tp_version) {
409 h.h1->tp_status = status;
410 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
413 h.h2->tp_status = status;
414 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
417 h.h3->tp_status = status;
418 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
421 WARN(1, "TPACKET version not supported.\n");
428 static int __packet_get_status(const struct packet_sock *po, void *frame)
430 union tpacket_uhdr h;
435 switch (po->tp_version) {
437 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
438 return h.h1->tp_status;
440 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
441 return h.h2->tp_status;
443 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
444 return h.h3->tp_status;
446 WARN(1, "TPACKET version not supported.\n");
452 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
455 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
458 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
459 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
460 return TP_STATUS_TS_RAW_HARDWARE;
462 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
463 ktime_to_timespec64_cond(skb_tstamp(skb), ts))
464 return TP_STATUS_TS_SOFTWARE;
469 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
472 union tpacket_uhdr h;
473 struct timespec64 ts;
476 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
481 * versions 1 through 3 overflow the timestamps in y2106, since they
482 * all store the seconds in a 32-bit unsigned integer.
483 * If we create a version 4, that should have a 64-bit timestamp,
484 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
487 switch (po->tp_version) {
489 h.h1->tp_sec = ts.tv_sec;
490 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
493 h.h2->tp_sec = ts.tv_sec;
494 h.h2->tp_nsec = ts.tv_nsec;
497 h.h3->tp_sec = ts.tv_sec;
498 h.h3->tp_nsec = ts.tv_nsec;
501 WARN(1, "TPACKET version not supported.\n");
505 /* one flush is safe, as both fields always lie on the same cacheline */
506 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
512 static void *packet_lookup_frame(const struct packet_sock *po,
513 const struct packet_ring_buffer *rb,
514 unsigned int position,
517 unsigned int pg_vec_pos, frame_offset;
518 union tpacket_uhdr h;
520 pg_vec_pos = position / rb->frames_per_block;
521 frame_offset = position % rb->frames_per_block;
523 h.raw = rb->pg_vec[pg_vec_pos].buffer +
524 (frame_offset * rb->frame_size);
526 if (status != __packet_get_status(po, h.raw))
532 static void *packet_current_frame(struct packet_sock *po,
533 struct packet_ring_buffer *rb,
536 return packet_lookup_frame(po, rb, rb->head, status);
539 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
541 del_timer_sync(&pkc->retire_blk_timer);
544 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
545 struct sk_buff_head *rb_queue)
547 struct tpacket_kbdq_core *pkc;
549 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
551 spin_lock_bh(&rb_queue->lock);
552 pkc->delete_blk_timer = 1;
553 spin_unlock_bh(&rb_queue->lock);
555 prb_del_retire_blk_timer(pkc);
558 static void prb_setup_retire_blk_timer(struct packet_sock *po)
560 struct tpacket_kbdq_core *pkc;
562 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
563 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
565 pkc->retire_blk_timer.expires = jiffies;
568 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
569 int blk_size_in_bytes)
571 struct net_device *dev;
572 unsigned int mbits, div;
573 struct ethtool_link_ksettings ecmd;
577 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
578 if (unlikely(!dev)) {
580 return DEFAULT_PRB_RETIRE_TOV;
582 err = __ethtool_get_link_ksettings(dev, &ecmd);
585 return DEFAULT_PRB_RETIRE_TOV;
587 /* If the link speed is so slow you don't really
588 * need to worry about perf anyways
590 if (ecmd.base.speed < SPEED_1000 ||
591 ecmd.base.speed == SPEED_UNKNOWN)
592 return DEFAULT_PRB_RETIRE_TOV;
594 div = ecmd.base.speed / 1000;
595 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
606 union tpacket_req_u *req_u)
608 p1->feature_req_word = req_u->req3.tp_feature_req_word;
611 static void init_prb_bdqc(struct packet_sock *po,
612 struct packet_ring_buffer *rb,
614 union tpacket_req_u *req_u)
616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
617 struct tpacket_block_desc *pbd;
619 memset(p1, 0x0, sizeof(*p1));
621 p1->knxt_seq_num = 1;
623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
624 p1->pkblk_start = pg_vec[0].buffer;
625 p1->kblk_size = req_u->req3.tp_block_size;
626 p1->knum_blocks = req_u->req3.tp_block_nr;
627 p1->hdrlen = po->tp_hdrlen;
628 p1->version = po->tp_version;
629 p1->last_kactive_blk_num = 0;
630 po->stats.stats3.tp_freeze_q_cnt = 0;
631 if (req_u->req3.tp_retire_blk_tov)
632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 req_u->req3.tp_block_size);
636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
638 rwlock_init(&p1->blk_fill_in_prog_lock);
640 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
641 prb_init_ft_ops(p1, req_u);
642 prb_setup_retire_blk_timer(po);
643 prb_open_block(p1, pbd);
646 /* Do NOT update the last_blk_num first.
647 * Assumes sk_buff_head lock is held.
649 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
651 mod_timer(&pkc->retire_blk_timer,
652 jiffies + pkc->tov_in_jiffies);
653 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
658 * 1) We refresh the timer only when we open a block.
659 * By doing this we don't waste cycles refreshing the timer
660 * on packet-by-packet basis.
662 * With a 1MB block-size, on a 1Gbps line, it will take
663 * i) ~8 ms to fill a block + ii) memcpy etc.
664 * In this cut we are not accounting for the memcpy time.
666 * So, if the user sets the 'tmo' to 10ms then the timer
667 * will never fire while the block is still getting filled
668 * (which is what we want). However, the user could choose
669 * to close a block early and that's fine.
671 * But when the timer does fire, we check whether or not to refresh it.
672 * Since the tmo granularity is in msecs, it is not too expensive
673 * to refresh the timer, lets say every '8' msecs.
674 * Either the user can set the 'tmo' or we can derive it based on
675 * a) line-speed and b) block-size.
676 * prb_calc_retire_blk_tmo() calculates the tmo.
679 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
681 struct packet_sock *po =
682 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
683 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
685 struct tpacket_block_desc *pbd;
687 spin_lock(&po->sk.sk_receive_queue.lock);
689 frozen = prb_queue_frozen(pkc);
690 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
692 if (unlikely(pkc->delete_blk_timer))
695 /* We only need to plug the race when the block is partially filled.
697 * lock(); increment BLOCK_NUM_PKTS; unlock()
698 * copy_bits() is in progress ...
699 * timer fires on other cpu:
700 * we can't retire the current block because copy_bits
704 if (BLOCK_NUM_PKTS(pbd)) {
705 /* Waiting for skb_copy_bits to finish... */
706 write_lock(&pkc->blk_fill_in_prog_lock);
707 write_unlock(&pkc->blk_fill_in_prog_lock);
710 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
712 if (!BLOCK_NUM_PKTS(pbd)) {
713 /* An empty block. Just refresh the timer. */
716 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
717 if (!prb_dispatch_next_block(pkc, po))
722 /* Case 1. Queue was frozen because user-space was
725 if (prb_curr_blk_in_use(pbd)) {
727 * Ok, user-space is still behind.
728 * So just refresh the timer.
732 /* Case 2. queue was frozen,user-space caught up,
733 * now the link went idle && the timer fired.
734 * We don't have a block to close.So we open this
735 * block and restart the timer.
736 * opening a block thaws the queue,restarts timer
737 * Thawing/timer-refresh is a side effect.
739 prb_open_block(pkc, pbd);
746 _prb_refresh_rx_retire_blk_timer(pkc);
749 spin_unlock(&po->sk.sk_receive_queue.lock);
752 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
753 struct tpacket_block_desc *pbd1, __u32 status)
755 /* Flush everything minus the block header */
757 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
762 /* Skip the block header(we know header WILL fit in 4K) */
765 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
766 for (; start < end; start += PAGE_SIZE)
767 flush_dcache_page(pgv_to_page(start));
772 /* Now update the block status. */
774 BLOCK_STATUS(pbd1) = status;
776 /* Flush the block header */
778 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
780 flush_dcache_page(pgv_to_page(start));
790 * 2) Increment active_blk_num
792 * Note:We DONT refresh the timer on purpose.
793 * Because almost always the next block will be opened.
795 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
796 struct tpacket_block_desc *pbd1,
797 struct packet_sock *po, unsigned int stat)
799 __u32 status = TP_STATUS_USER | stat;
801 struct tpacket3_hdr *last_pkt;
802 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
803 struct sock *sk = &po->sk;
805 if (atomic_read(&po->tp_drops))
806 status |= TP_STATUS_LOSING;
808 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
809 last_pkt->tp_next_offset = 0;
811 /* Get the ts of the last pkt */
812 if (BLOCK_NUM_PKTS(pbd1)) {
813 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
814 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
816 /* Ok, we tmo'd - so get the current time.
818 * It shouldn't really happen as we don't close empty
819 * blocks. See prb_retire_rx_blk_timer_expired().
821 struct timespec64 ts;
822 ktime_get_real_ts64(&ts);
823 h1->ts_last_pkt.ts_sec = ts.tv_sec;
824 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
829 /* Flush the block */
830 prb_flush_block(pkc1, pbd1, status);
832 sk->sk_data_ready(sk);
834 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
837 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
839 pkc->reset_pending_on_curr_blk = 0;
843 * Side effect of opening a block:
845 * 1) prb_queue is thawed.
846 * 2) retire_blk_timer is refreshed.
849 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
850 struct tpacket_block_desc *pbd1)
852 struct timespec64 ts;
853 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
857 /* We could have just memset this but we will lose the
858 * flexibility of making the priv area sticky
861 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
862 BLOCK_NUM_PKTS(pbd1) = 0;
863 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 ktime_get_real_ts64(&ts);
867 h1->ts_first_pkt.ts_sec = ts.tv_sec;
868 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
870 pkc1->pkblk_start = (char *)pbd1;
871 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
873 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
874 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
876 pbd1->version = pkc1->version;
877 pkc1->prev = pkc1->nxt_offset;
878 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
880 prb_thaw_queue(pkc1);
881 _prb_refresh_rx_retire_blk_timer(pkc1);
887 * Queue freeze logic:
888 * 1) Assume tp_block_nr = 8 blocks.
889 * 2) At time 't0', user opens Rx ring.
890 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
891 * 4) user-space is either sleeping or processing block '0'.
892 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
893 * it will close block-7,loop around and try to fill block '0'.
895 * __packet_lookup_frame_in_block
896 * prb_retire_current_block()
897 * prb_dispatch_next_block()
898 * |->(BLOCK_STATUS == USER) evaluates to true
899 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
900 * 6) Now there are two cases:
901 * 6.1) Link goes idle right after the queue is frozen.
902 * But remember, the last open_block() refreshed the timer.
903 * When this timer expires,it will refresh itself so that we can
904 * re-open block-0 in near future.
905 * 6.2) Link is busy and keeps on receiving packets. This is a simple
906 * case and __packet_lookup_frame_in_block will check if block-0
907 * is free and can now be re-used.
909 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
910 struct packet_sock *po)
912 pkc->reset_pending_on_curr_blk = 1;
913 po->stats.stats3.tp_freeze_q_cnt++;
916 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
919 * If the next block is free then we will dispatch it
920 * and return a good offset.
921 * Else, we will freeze the queue.
922 * So, caller must check the return value.
924 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
925 struct packet_sock *po)
927 struct tpacket_block_desc *pbd;
931 /* 1. Get current block num */
932 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
934 /* 2. If this block is currently in_use then freeze the queue */
935 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
936 prb_freeze_queue(pkc, po);
942 * open this block and return the offset where the first packet
943 * needs to get stored.
945 prb_open_block(pkc, pbd);
946 return (void *)pkc->nxt_offset;
949 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
950 struct packet_sock *po, unsigned int status)
952 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
954 /* retire/close the current block */
955 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
957 * Plug the case where copy_bits() is in progress on
958 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
959 * have space to copy the pkt in the current block and
960 * called prb_retire_current_block()
962 * We don't need to worry about the TMO case because
963 * the timer-handler already handled this case.
965 if (!(status & TP_STATUS_BLK_TMO)) {
966 /* Waiting for skb_copy_bits to finish... */
967 write_lock(&pkc->blk_fill_in_prog_lock);
968 write_unlock(&pkc->blk_fill_in_prog_lock);
970 prb_close_block(pkc, pbd, po, status);
975 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
977 return TP_STATUS_USER & BLOCK_STATUS(pbd);
980 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
982 return pkc->reset_pending_on_curr_blk;
985 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
986 __releases(&pkc->blk_fill_in_prog_lock)
988 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
990 read_unlock(&pkc->blk_fill_in_prog_lock);
993 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
994 struct tpacket3_hdr *ppd)
996 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
999 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1000 struct tpacket3_hdr *ppd)
1002 ppd->hv1.tp_rxhash = 0;
1005 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1006 struct tpacket3_hdr *ppd)
1008 if (skb_vlan_tag_present(pkc->skb)) {
1009 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1010 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1011 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1013 ppd->hv1.tp_vlan_tci = 0;
1014 ppd->hv1.tp_vlan_tpid = 0;
1015 ppd->tp_status = TP_STATUS_AVAILABLE;
1019 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1020 struct tpacket3_hdr *ppd)
1022 ppd->hv1.tp_padding = 0;
1023 prb_fill_vlan_info(pkc, ppd);
1025 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1026 prb_fill_rxhash(pkc, ppd);
1028 prb_clear_rxhash(pkc, ppd);
1031 static void prb_fill_curr_block(char *curr,
1032 struct tpacket_kbdq_core *pkc,
1033 struct tpacket_block_desc *pbd,
1035 __acquires(&pkc->blk_fill_in_prog_lock)
1037 struct tpacket3_hdr *ppd;
1039 ppd = (struct tpacket3_hdr *)curr;
1040 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1042 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1043 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1044 BLOCK_NUM_PKTS(pbd) += 1;
1045 read_lock(&pkc->blk_fill_in_prog_lock);
1046 prb_run_all_ft_ops(pkc, ppd);
1049 /* Assumes caller has the sk->rx_queue.lock */
1050 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1051 struct sk_buff *skb,
1055 struct tpacket_kbdq_core *pkc;
1056 struct tpacket_block_desc *pbd;
1059 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1060 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1062 /* Queue is frozen when user space is lagging behind */
1063 if (prb_queue_frozen(pkc)) {
1065 * Check if that last block which caused the queue to freeze,
1066 * is still in_use by user-space.
1068 if (prb_curr_blk_in_use(pbd)) {
1069 /* Can't record this packet */
1073 * Ok, the block was released by user-space.
1074 * Now let's open that block.
1075 * opening a block also thaws the queue.
1076 * Thawing is a side effect.
1078 prb_open_block(pkc, pbd);
1083 curr = pkc->nxt_offset;
1085 end = (char *)pbd + pkc->kblk_size;
1087 /* first try the current block */
1088 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1089 prb_fill_curr_block(curr, pkc, pbd, len);
1090 return (void *)curr;
1093 /* Ok, close the current block */
1094 prb_retire_current_block(pkc, po, 0);
1096 /* Now, try to dispatch the next block */
1097 curr = (char *)prb_dispatch_next_block(pkc, po);
1099 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1100 prb_fill_curr_block(curr, pkc, pbd, len);
1101 return (void *)curr;
1105 * No free blocks are available.user_space hasn't caught up yet.
1106 * Queue was just frozen and now this packet will get dropped.
1111 static void *packet_current_rx_frame(struct packet_sock *po,
1112 struct sk_buff *skb,
1113 int status, unsigned int len)
1116 switch (po->tp_version) {
1119 curr = packet_lookup_frame(po, &po->rx_ring,
1120 po->rx_ring.head, status);
1123 return __packet_lookup_frame_in_block(po, skb, len);
1125 WARN(1, "TPACKET version not supported\n");
1131 static void *prb_lookup_block(const struct packet_sock *po,
1132 const struct packet_ring_buffer *rb,
1136 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1137 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1139 if (status != BLOCK_STATUS(pbd))
1144 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1147 if (rb->prb_bdqc.kactive_blk_num)
1148 prev = rb->prb_bdqc.kactive_blk_num-1;
1150 prev = rb->prb_bdqc.knum_blocks-1;
1154 /* Assumes caller has held the rx_queue.lock */
1155 static void *__prb_previous_block(struct packet_sock *po,
1156 struct packet_ring_buffer *rb,
1159 unsigned int previous = prb_previous_blk_num(rb);
1160 return prb_lookup_block(po, rb, previous, status);
1163 static void *packet_previous_rx_frame(struct packet_sock *po,
1164 struct packet_ring_buffer *rb,
1167 if (po->tp_version <= TPACKET_V2)
1168 return packet_previous_frame(po, rb, status);
1170 return __prb_previous_block(po, rb, status);
1173 static void packet_increment_rx_head(struct packet_sock *po,
1174 struct packet_ring_buffer *rb)
1176 switch (po->tp_version) {
1179 return packet_increment_head(rb);
1182 WARN(1, "TPACKET version not supported.\n");
1188 static void *packet_previous_frame(struct packet_sock *po,
1189 struct packet_ring_buffer *rb,
1192 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1193 return packet_lookup_frame(po, rb, previous, status);
1196 static void packet_increment_head(struct packet_ring_buffer *buff)
1198 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1201 static void packet_inc_pending(struct packet_ring_buffer *rb)
1203 this_cpu_inc(*rb->pending_refcnt);
1206 static void packet_dec_pending(struct packet_ring_buffer *rb)
1208 this_cpu_dec(*rb->pending_refcnt);
1211 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1213 unsigned int refcnt = 0;
1216 /* We don't use pending refcount in rx_ring. */
1217 if (rb->pending_refcnt == NULL)
1220 for_each_possible_cpu(cpu)
1221 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1226 static int packet_alloc_pending(struct packet_sock *po)
1228 po->rx_ring.pending_refcnt = NULL;
1230 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1231 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1237 static void packet_free_pending(struct packet_sock *po)
1239 free_percpu(po->tx_ring.pending_refcnt);
1242 #define ROOM_POW_OFF 2
1243 #define ROOM_NONE 0x0
1244 #define ROOM_LOW 0x1
1245 #define ROOM_NORMAL 0x2
1247 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1251 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1252 idx = READ_ONCE(po->rx_ring.head);
1254 idx += len >> pow_off;
1257 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1260 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1264 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1265 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1267 idx += len >> pow_off;
1270 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1273 static int __packet_rcv_has_room(const struct packet_sock *po,
1274 const struct sk_buff *skb)
1276 const struct sock *sk = &po->sk;
1277 int ret = ROOM_NONE;
1279 if (po->prot_hook.func != tpacket_rcv) {
1280 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1281 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1282 - (skb ? skb->truesize : 0);
1284 if (avail > (rcvbuf >> ROOM_POW_OFF))
1292 if (po->tp_version == TPACKET_V3) {
1293 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1295 else if (__tpacket_v3_has_room(po, 0))
1298 if (__tpacket_has_room(po, ROOM_POW_OFF))
1300 else if (__tpacket_has_room(po, 0))
1307 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1311 ret = __packet_rcv_has_room(po, skb);
1312 pressure = ret != ROOM_NORMAL;
1314 if (READ_ONCE(po->pressure) != pressure)
1315 WRITE_ONCE(po->pressure, pressure);
1320 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1322 if (READ_ONCE(po->pressure) &&
1323 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1324 WRITE_ONCE(po->pressure, 0);
1327 static void packet_sock_destruct(struct sock *sk)
1329 skb_queue_purge(&sk->sk_error_queue);
1331 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1332 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1334 if (!sock_flag(sk, SOCK_DEAD)) {
1335 pr_err("Attempt to release alive packet socket: %p\n", sk);
1339 sk_refcnt_debug_dec(sk);
1342 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1344 u32 *history = po->rollover->history;
1348 rxhash = skb_get_hash(skb);
1349 for (i = 0; i < ROLLOVER_HLEN; i++)
1350 if (READ_ONCE(history[i]) == rxhash)
1353 victim = prandom_u32_max(ROLLOVER_HLEN);
1355 /* Avoid dirtying the cache line if possible */
1356 if (READ_ONCE(history[victim]) != rxhash)
1357 WRITE_ONCE(history[victim], rxhash);
1359 return count > (ROLLOVER_HLEN >> 1);
1362 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1363 struct sk_buff *skb,
1366 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1369 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1370 struct sk_buff *skb,
1373 unsigned int val = atomic_inc_return(&f->rr_cur);
1378 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1379 struct sk_buff *skb,
1382 return smp_processor_id() % num;
1385 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1386 struct sk_buff *skb,
1389 return prandom_u32_max(num);
1392 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1393 struct sk_buff *skb,
1394 unsigned int idx, bool try_self,
1397 struct packet_sock *po, *po_next, *po_skip = NULL;
1398 unsigned int i, j, room = ROOM_NONE;
1400 po = pkt_sk(rcu_dereference(f->arr[idx]));
1403 room = packet_rcv_has_room(po, skb);
1404 if (room == ROOM_NORMAL ||
1405 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1410 i = j = min_t(int, po->rollover->sock, num - 1);
1412 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1413 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1414 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1416 po->rollover->sock = i;
1417 atomic_long_inc(&po->rollover->num);
1418 if (room == ROOM_LOW)
1419 atomic_long_inc(&po->rollover->num_huge);
1427 atomic_long_inc(&po->rollover->num_failed);
1431 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1432 struct sk_buff *skb,
1435 return skb_get_queue_mapping(skb) % num;
1438 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1439 struct sk_buff *skb,
1442 struct bpf_prog *prog;
1443 unsigned int ret = 0;
1446 prog = rcu_dereference(f->bpf_prog);
1448 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1454 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1456 return f->flags & (flag >> 8);
1459 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1460 struct packet_type *pt, struct net_device *orig_dev)
1462 struct packet_fanout *f = pt->af_packet_priv;
1463 unsigned int num = READ_ONCE(f->num_members);
1464 struct net *net = read_pnet(&f->net);
1465 struct packet_sock *po;
1468 if (!net_eq(dev_net(dev), net) || !num) {
1473 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1474 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1479 case PACKET_FANOUT_HASH:
1481 idx = fanout_demux_hash(f, skb, num);
1483 case PACKET_FANOUT_LB:
1484 idx = fanout_demux_lb(f, skb, num);
1486 case PACKET_FANOUT_CPU:
1487 idx = fanout_demux_cpu(f, skb, num);
1489 case PACKET_FANOUT_RND:
1490 idx = fanout_demux_rnd(f, skb, num);
1492 case PACKET_FANOUT_QM:
1493 idx = fanout_demux_qm(f, skb, num);
1495 case PACKET_FANOUT_ROLLOVER:
1496 idx = fanout_demux_rollover(f, skb, 0, false, num);
1498 case PACKET_FANOUT_CBPF:
1499 case PACKET_FANOUT_EBPF:
1500 idx = fanout_demux_bpf(f, skb, num);
1504 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1505 idx = fanout_demux_rollover(f, skb, idx, true, num);
1507 po = pkt_sk(rcu_dereference(f->arr[idx]));
1508 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1511 DEFINE_MUTEX(fanout_mutex);
1512 EXPORT_SYMBOL_GPL(fanout_mutex);
1513 static LIST_HEAD(fanout_list);
1514 static u16 fanout_next_id;
1516 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1518 struct packet_fanout *f = po->fanout;
1520 spin_lock(&f->lock);
1521 rcu_assign_pointer(f->arr[f->num_members], sk);
1524 if (f->num_members == 1)
1525 dev_add_pack(&f->prot_hook);
1526 spin_unlock(&f->lock);
1529 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1531 struct packet_fanout *f = po->fanout;
1534 spin_lock(&f->lock);
1535 for (i = 0; i < f->num_members; i++) {
1536 if (rcu_dereference_protected(f->arr[i],
1537 lockdep_is_held(&f->lock)) == sk)
1540 BUG_ON(i >= f->num_members);
1541 rcu_assign_pointer(f->arr[i],
1542 rcu_dereference_protected(f->arr[f->num_members - 1],
1543 lockdep_is_held(&f->lock)));
1545 if (f->num_members == 0)
1546 __dev_remove_pack(&f->prot_hook);
1547 spin_unlock(&f->lock);
1550 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1552 if (sk->sk_family != PF_PACKET)
1555 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1558 static void fanout_init_data(struct packet_fanout *f)
1561 case PACKET_FANOUT_LB:
1562 atomic_set(&f->rr_cur, 0);
1564 case PACKET_FANOUT_CBPF:
1565 case PACKET_FANOUT_EBPF:
1566 RCU_INIT_POINTER(f->bpf_prog, NULL);
1571 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1573 struct bpf_prog *old;
1575 spin_lock(&f->lock);
1576 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1577 rcu_assign_pointer(f->bpf_prog, new);
1578 spin_unlock(&f->lock);
1582 bpf_prog_destroy(old);
1586 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1589 struct bpf_prog *new;
1590 struct sock_fprog fprog;
1593 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1596 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1600 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1604 __fanout_set_data_bpf(po->fanout, new);
1608 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1611 struct bpf_prog *new;
1614 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1616 if (len != sizeof(fd))
1618 if (copy_from_sockptr(&fd, data, len))
1621 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1623 return PTR_ERR(new);
1625 __fanout_set_data_bpf(po->fanout, new);
1629 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1632 switch (po->fanout->type) {
1633 case PACKET_FANOUT_CBPF:
1634 return fanout_set_data_cbpf(po, data, len);
1635 case PACKET_FANOUT_EBPF:
1636 return fanout_set_data_ebpf(po, data, len);
1642 static void fanout_release_data(struct packet_fanout *f)
1645 case PACKET_FANOUT_CBPF:
1646 case PACKET_FANOUT_EBPF:
1647 __fanout_set_data_bpf(f, NULL);
1651 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1653 struct packet_fanout *f;
1655 list_for_each_entry(f, &fanout_list, list) {
1656 if (f->id == candidate_id &&
1657 read_pnet(&f->net) == sock_net(sk)) {
1664 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1666 u16 id = fanout_next_id;
1669 if (__fanout_id_is_free(sk, id)) {
1671 fanout_next_id = id + 1;
1676 } while (id != fanout_next_id);
1681 static int fanout_add(struct sock *sk, struct fanout_args *args)
1683 struct packet_rollover *rollover = NULL;
1684 struct packet_sock *po = pkt_sk(sk);
1685 u16 type_flags = args->type_flags;
1686 struct packet_fanout *f, *match;
1687 u8 type = type_flags & 0xff;
1688 u8 flags = type_flags >> 8;
1693 case PACKET_FANOUT_ROLLOVER:
1694 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1697 case PACKET_FANOUT_HASH:
1698 case PACKET_FANOUT_LB:
1699 case PACKET_FANOUT_CPU:
1700 case PACKET_FANOUT_RND:
1701 case PACKET_FANOUT_QM:
1702 case PACKET_FANOUT_CBPF:
1703 case PACKET_FANOUT_EBPF:
1709 mutex_lock(&fanout_mutex);
1715 if (type == PACKET_FANOUT_ROLLOVER ||
1716 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1718 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1721 atomic_long_set(&rollover->num, 0);
1722 atomic_long_set(&rollover->num_huge, 0);
1723 atomic_long_set(&rollover->num_failed, 0);
1726 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1731 if (!fanout_find_new_id(sk, &id)) {
1735 /* ephemeral flag for the first socket in the group: drop it */
1736 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1740 list_for_each_entry(f, &fanout_list, list) {
1742 read_pnet(&f->net) == sock_net(sk)) {
1749 if (match->flags != flags)
1751 if (args->max_num_members &&
1752 args->max_num_members != match->max_num_members)
1755 if (args->max_num_members > PACKET_FANOUT_MAX)
1757 if (!args->max_num_members)
1758 /* legacy PACKET_FANOUT_MAX */
1759 args->max_num_members = 256;
1761 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1765 write_pnet(&match->net, sock_net(sk));
1768 match->flags = flags;
1769 INIT_LIST_HEAD(&match->list);
1770 spin_lock_init(&match->lock);
1771 refcount_set(&match->sk_ref, 0);
1772 fanout_init_data(match);
1773 match->prot_hook.type = po->prot_hook.type;
1774 match->prot_hook.dev = po->prot_hook.dev;
1775 match->prot_hook.func = packet_rcv_fanout;
1776 match->prot_hook.af_packet_priv = match;
1777 match->prot_hook.af_packet_net = read_pnet(&match->net);
1778 match->prot_hook.id_match = match_fanout_group;
1779 match->max_num_members = args->max_num_members;
1780 list_add(&match->list, &fanout_list);
1784 spin_lock(&po->bind_lock);
1786 match->type == type &&
1787 match->prot_hook.type == po->prot_hook.type &&
1788 match->prot_hook.dev == po->prot_hook.dev) {
1790 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1791 __dev_remove_pack(&po->prot_hook);
1793 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1794 WRITE_ONCE(po->fanout, match);
1796 po->rollover = rollover;
1798 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1799 __fanout_link(sk, po);
1803 spin_unlock(&po->bind_lock);
1805 if (err && !refcount_read(&match->sk_ref)) {
1806 list_del(&match->list);
1812 mutex_unlock(&fanout_mutex);
1816 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1817 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1818 * It is the responsibility of the caller to call fanout_release_data() and
1819 * free the returned packet_fanout (after synchronize_net())
1821 static struct packet_fanout *fanout_release(struct sock *sk)
1823 struct packet_sock *po = pkt_sk(sk);
1824 struct packet_fanout *f;
1826 mutex_lock(&fanout_mutex);
1831 if (refcount_dec_and_test(&f->sk_ref))
1836 mutex_unlock(&fanout_mutex);
1841 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1842 struct sk_buff *skb)
1844 /* Earlier code assumed this would be a VLAN pkt, double-check
1845 * this now that we have the actual packet in hand. We can only
1846 * do this check on Ethernet devices.
1848 if (unlikely(dev->type != ARPHRD_ETHER))
1851 skb_reset_mac_header(skb);
1852 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1855 static const struct proto_ops packet_ops;
1857 static const struct proto_ops packet_ops_spkt;
1859 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1860 struct packet_type *pt, struct net_device *orig_dev)
1863 struct sockaddr_pkt *spkt;
1866 * When we registered the protocol we saved the socket in the data
1867 * field for just this event.
1870 sk = pt->af_packet_priv;
1873 * Yank back the headers [hope the device set this
1874 * right or kerboom...]
1876 * Incoming packets have ll header pulled,
1879 * For outgoing ones skb->data == skb_mac_header(skb)
1880 * so that this procedure is noop.
1883 if (skb->pkt_type == PACKET_LOOPBACK)
1886 if (!net_eq(dev_net(dev), sock_net(sk)))
1889 skb = skb_share_check(skb, GFP_ATOMIC);
1893 /* drop any routing info */
1896 /* drop conntrack reference */
1899 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1901 skb_push(skb, skb->data - skb_mac_header(skb));
1904 * The SOCK_PACKET socket receives _all_ frames.
1907 spkt->spkt_family = dev->type;
1908 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1909 spkt->spkt_protocol = skb->protocol;
1912 * Charge the memory to the socket. This is done specifically
1913 * to prevent sockets using all the memory up.
1916 if (sock_queue_rcv_skb(sk, skb) == 0)
1925 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1929 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1930 sock->type == SOCK_RAW) {
1931 skb_reset_mac_header(skb);
1932 skb->protocol = dev_parse_header_protocol(skb);
1935 /* Move network header to the right position for VLAN tagged packets */
1936 if (likely(skb->dev->type == ARPHRD_ETHER) &&
1937 eth_type_vlan(skb->protocol) &&
1938 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) {
1939 if (pskb_may_pull(skb, depth))
1940 skb_set_network_header(skb, depth);
1943 skb_probe_transport_header(skb);
1947 * Output a raw packet to a device layer. This bypasses all the other
1948 * protocol layers and you must therefore supply it with a complete frame
1951 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1954 struct sock *sk = sock->sk;
1955 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1956 struct sk_buff *skb = NULL;
1957 struct net_device *dev;
1958 struct sockcm_cookie sockc;
1964 * Get and verify the address.
1968 if (msg->msg_namelen < sizeof(struct sockaddr))
1970 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1971 proto = saddr->spkt_protocol;
1973 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1976 * Find the device first to size check it
1979 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1982 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1988 if (!(dev->flags & IFF_UP))
1992 * You may not queue a frame bigger than the mtu. This is the lowest level
1993 * raw protocol and you must do your own fragmentation at this level.
1996 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1997 if (!netif_supports_nofcs(dev)) {
1998 err = -EPROTONOSUPPORT;
2001 extra_len = 4; /* We're doing our own CRC */
2005 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2009 size_t reserved = LL_RESERVED_SPACE(dev);
2010 int tlen = dev->needed_tailroom;
2011 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2014 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2017 /* FIXME: Save some space for broken drivers that write a hard
2018 * header at transmission time by themselves. PPP is the notable
2019 * one here. This should really be fixed at the driver level.
2021 skb_reserve(skb, reserved);
2022 skb_reset_network_header(skb);
2024 /* Try to align data part correctly */
2029 skb_reset_network_header(skb);
2031 err = memcpy_from_msg(skb_put(skb, len), msg, len);
2037 if (!dev_validate_header(dev, skb->data, len)) {
2041 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2042 !packet_extra_vlan_len_allowed(dev, skb)) {
2047 sockcm_init(&sockc, sk);
2048 if (msg->msg_controllen) {
2049 err = sock_cmsg_send(sk, msg, &sockc);
2054 skb->protocol = proto;
2056 skb->priority = sk->sk_priority;
2057 skb->mark = sk->sk_mark;
2058 skb->tstamp = sockc.transmit_time;
2060 skb_setup_tx_timestamp(skb, sockc.tsflags);
2062 if (unlikely(extra_len == 4))
2065 packet_parse_headers(skb, sock);
2067 dev_queue_xmit(skb);
2078 static unsigned int run_filter(struct sk_buff *skb,
2079 const struct sock *sk,
2082 struct sk_filter *filter;
2085 filter = rcu_dereference(sk->sk_filter);
2087 res = bpf_prog_run_clear_cb(filter->prog, skb);
2093 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2096 struct virtio_net_hdr vnet_hdr;
2098 if (*len < sizeof(vnet_hdr))
2100 *len -= sizeof(vnet_hdr);
2102 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2105 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2109 * This function makes lazy skb cloning in hope that most of packets
2110 * are discarded by BPF.
2112 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2113 * and skb->cb are mangled. It works because (and until) packets
2114 * falling here are owned by current CPU. Output packets are cloned
2115 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2116 * sequentially, so that if we return skb to original state on exit,
2117 * we will not harm anyone.
2120 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2121 struct packet_type *pt, struct net_device *orig_dev)
2124 struct sockaddr_ll *sll;
2125 struct packet_sock *po;
2126 u8 *skb_head = skb->data;
2127 int skb_len = skb->len;
2128 unsigned int snaplen, res;
2129 bool is_drop_n_account = false;
2131 if (skb->pkt_type == PACKET_LOOPBACK)
2134 sk = pt->af_packet_priv;
2137 if (!net_eq(dev_net(dev), sock_net(sk)))
2142 if (dev_has_header(dev)) {
2143 /* The device has an explicit notion of ll header,
2144 * exported to higher levels.
2146 * Otherwise, the device hides details of its frame
2147 * structure, so that corresponding packet head is
2148 * never delivered to user.
2150 if (sk->sk_type != SOCK_DGRAM)
2151 skb_push(skb, skb->data - skb_mac_header(skb));
2152 else if (skb->pkt_type == PACKET_OUTGOING) {
2153 /* Special case: outgoing packets have ll header at head */
2154 skb_pull(skb, skb_network_offset(skb));
2160 res = run_filter(skb, sk, snaplen);
2162 goto drop_n_restore;
2166 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2169 if (skb_shared(skb)) {
2170 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2174 if (skb_head != skb->data) {
2175 skb->data = skb_head;
2182 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2184 sll = &PACKET_SKB_CB(skb)->sa.ll;
2185 sll->sll_hatype = dev->type;
2186 sll->sll_pkttype = skb->pkt_type;
2187 if (unlikely(po->origdev))
2188 sll->sll_ifindex = orig_dev->ifindex;
2190 sll->sll_ifindex = dev->ifindex;
2192 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2194 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2195 * Use their space for storing the original skb length.
2197 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2199 if (pskb_trim(skb, snaplen))
2202 skb_set_owner_r(skb, sk);
2206 /* drop conntrack reference */
2209 spin_lock(&sk->sk_receive_queue.lock);
2210 po->stats.stats1.tp_packets++;
2211 sock_skb_set_dropcount(sk, skb);
2212 skb_clear_delivery_time(skb);
2213 __skb_queue_tail(&sk->sk_receive_queue, skb);
2214 spin_unlock(&sk->sk_receive_queue.lock);
2215 sk->sk_data_ready(sk);
2219 is_drop_n_account = true;
2220 atomic_inc(&po->tp_drops);
2221 atomic_inc(&sk->sk_drops);
2224 if (skb_head != skb->data && skb_shared(skb)) {
2225 skb->data = skb_head;
2229 if (!is_drop_n_account)
2236 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2237 struct packet_type *pt, struct net_device *orig_dev)
2240 struct packet_sock *po;
2241 struct sockaddr_ll *sll;
2242 union tpacket_uhdr h;
2243 u8 *skb_head = skb->data;
2244 int skb_len = skb->len;
2245 unsigned int snaplen, res;
2246 unsigned long status = TP_STATUS_USER;
2247 unsigned short macoff, hdrlen;
2248 unsigned int netoff;
2249 struct sk_buff *copy_skb = NULL;
2250 struct timespec64 ts;
2252 bool is_drop_n_account = false;
2253 unsigned int slot_id = 0;
2254 bool do_vnet = false;
2256 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2257 * We may add members to them until current aligned size without forcing
2258 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2260 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2261 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2263 if (skb->pkt_type == PACKET_LOOPBACK)
2266 sk = pt->af_packet_priv;
2269 if (!net_eq(dev_net(dev), sock_net(sk)))
2272 if (dev_has_header(dev)) {
2273 if (sk->sk_type != SOCK_DGRAM)
2274 skb_push(skb, skb->data - skb_mac_header(skb));
2275 else if (skb->pkt_type == PACKET_OUTGOING) {
2276 /* Special case: outgoing packets have ll header at head */
2277 skb_pull(skb, skb_network_offset(skb));
2283 res = run_filter(skb, sk, snaplen);
2285 goto drop_n_restore;
2287 /* If we are flooded, just give up */
2288 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2289 atomic_inc(&po->tp_drops);
2290 goto drop_n_restore;
2293 if (skb->ip_summed == CHECKSUM_PARTIAL)
2294 status |= TP_STATUS_CSUMNOTREADY;
2295 else if (skb->pkt_type != PACKET_OUTGOING &&
2296 skb_csum_unnecessary(skb))
2297 status |= TP_STATUS_CSUM_VALID;
2302 if (sk->sk_type == SOCK_DGRAM) {
2303 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2306 unsigned int maclen = skb_network_offset(skb);
2307 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2308 (maclen < 16 ? 16 : maclen)) +
2310 if (po->has_vnet_hdr) {
2311 netoff += sizeof(struct virtio_net_hdr);
2314 macoff = netoff - maclen;
2316 if (netoff > USHRT_MAX) {
2317 atomic_inc(&po->tp_drops);
2318 goto drop_n_restore;
2320 if (po->tp_version <= TPACKET_V2) {
2321 if (macoff + snaplen > po->rx_ring.frame_size) {
2322 if (po->copy_thresh &&
2323 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2324 if (skb_shared(skb)) {
2325 copy_skb = skb_clone(skb, GFP_ATOMIC);
2327 copy_skb = skb_get(skb);
2328 skb_head = skb->data;
2331 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2332 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2333 skb_set_owner_r(copy_skb, sk);
2336 snaplen = po->rx_ring.frame_size - macoff;
2337 if ((int)snaplen < 0) {
2342 } else if (unlikely(macoff + snaplen >
2343 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2346 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2347 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2348 snaplen, nval, macoff);
2350 if (unlikely((int)snaplen < 0)) {
2352 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2356 spin_lock(&sk->sk_receive_queue.lock);
2357 h.raw = packet_current_rx_frame(po, skb,
2358 TP_STATUS_KERNEL, (macoff+snaplen));
2360 goto drop_n_account;
2362 if (po->tp_version <= TPACKET_V2) {
2363 slot_id = po->rx_ring.head;
2364 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2365 goto drop_n_account;
2366 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2370 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2371 sizeof(struct virtio_net_hdr),
2372 vio_le(), true, 0)) {
2373 if (po->tp_version == TPACKET_V3)
2374 prb_clear_blk_fill_status(&po->rx_ring);
2375 goto drop_n_account;
2378 if (po->tp_version <= TPACKET_V2) {
2379 packet_increment_rx_head(po, &po->rx_ring);
2381 * LOSING will be reported till you read the stats,
2382 * because it's COR - Clear On Read.
2383 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2386 if (atomic_read(&po->tp_drops))
2387 status |= TP_STATUS_LOSING;
2390 po->stats.stats1.tp_packets++;
2392 status |= TP_STATUS_COPY;
2393 skb_clear_delivery_time(copy_skb);
2394 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2396 spin_unlock(&sk->sk_receive_queue.lock);
2398 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2400 /* Always timestamp; prefer an existing software timestamp taken
2401 * closer to the time of capture.
2403 ts_status = tpacket_get_timestamp(skb, &ts,
2404 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2406 ktime_get_real_ts64(&ts);
2408 status |= ts_status;
2410 switch (po->tp_version) {
2412 h.h1->tp_len = skb->len;
2413 h.h1->tp_snaplen = snaplen;
2414 h.h1->tp_mac = macoff;
2415 h.h1->tp_net = netoff;
2416 h.h1->tp_sec = ts.tv_sec;
2417 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2418 hdrlen = sizeof(*h.h1);
2421 h.h2->tp_len = skb->len;
2422 h.h2->tp_snaplen = snaplen;
2423 h.h2->tp_mac = macoff;
2424 h.h2->tp_net = netoff;
2425 h.h2->tp_sec = ts.tv_sec;
2426 h.h2->tp_nsec = ts.tv_nsec;
2427 if (skb_vlan_tag_present(skb)) {
2428 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2429 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2430 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2432 h.h2->tp_vlan_tci = 0;
2433 h.h2->tp_vlan_tpid = 0;
2435 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2436 hdrlen = sizeof(*h.h2);
2439 /* tp_nxt_offset,vlan are already populated above.
2440 * So DONT clear those fields here
2442 h.h3->tp_status |= status;
2443 h.h3->tp_len = skb->len;
2444 h.h3->tp_snaplen = snaplen;
2445 h.h3->tp_mac = macoff;
2446 h.h3->tp_net = netoff;
2447 h.h3->tp_sec = ts.tv_sec;
2448 h.h3->tp_nsec = ts.tv_nsec;
2449 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2450 hdrlen = sizeof(*h.h3);
2456 sll = h.raw + TPACKET_ALIGN(hdrlen);
2457 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2458 sll->sll_family = AF_PACKET;
2459 sll->sll_hatype = dev->type;
2460 sll->sll_protocol = skb->protocol;
2461 sll->sll_pkttype = skb->pkt_type;
2462 if (unlikely(po->origdev))
2463 sll->sll_ifindex = orig_dev->ifindex;
2465 sll->sll_ifindex = dev->ifindex;
2469 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2470 if (po->tp_version <= TPACKET_V2) {
2473 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2476 for (start = h.raw; start < end; start += PAGE_SIZE)
2477 flush_dcache_page(pgv_to_page(start));
2482 if (po->tp_version <= TPACKET_V2) {
2483 spin_lock(&sk->sk_receive_queue.lock);
2484 __packet_set_status(po, h.raw, status);
2485 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2486 spin_unlock(&sk->sk_receive_queue.lock);
2487 sk->sk_data_ready(sk);
2488 } else if (po->tp_version == TPACKET_V3) {
2489 prb_clear_blk_fill_status(&po->rx_ring);
2493 if (skb_head != skb->data && skb_shared(skb)) {
2494 skb->data = skb_head;
2498 if (!is_drop_n_account)
2505 spin_unlock(&sk->sk_receive_queue.lock);
2506 atomic_inc(&po->tp_drops);
2507 is_drop_n_account = true;
2509 sk->sk_data_ready(sk);
2510 kfree_skb(copy_skb);
2511 goto drop_n_restore;
2514 static void tpacket_destruct_skb(struct sk_buff *skb)
2516 struct packet_sock *po = pkt_sk(skb->sk);
2518 if (likely(po->tx_ring.pg_vec)) {
2522 ph = skb_zcopy_get_nouarg(skb);
2523 packet_dec_pending(&po->tx_ring);
2525 ts = __packet_set_timestamp(po, ph, skb);
2526 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2528 if (!packet_read_pending(&po->tx_ring))
2529 complete(&po->skb_completion);
2535 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2537 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2538 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2539 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2540 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2541 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2542 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2543 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2545 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2551 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2552 struct virtio_net_hdr *vnet_hdr)
2554 if (*len < sizeof(*vnet_hdr))
2556 *len -= sizeof(*vnet_hdr);
2558 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2561 return __packet_snd_vnet_parse(vnet_hdr, *len);
2564 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2565 void *frame, struct net_device *dev, void *data, int tp_len,
2566 __be16 proto, unsigned char *addr, int hlen, int copylen,
2567 const struct sockcm_cookie *sockc)
2569 union tpacket_uhdr ph;
2570 int to_write, offset, len, nr_frags, len_max;
2571 struct socket *sock = po->sk.sk_socket;
2577 skb->protocol = proto;
2579 skb->priority = po->sk.sk_priority;
2580 skb->mark = po->sk.sk_mark;
2581 skb->tstamp = sockc->transmit_time;
2582 skb_setup_tx_timestamp(skb, sockc->tsflags);
2583 skb_zcopy_set_nouarg(skb, ph.raw);
2585 skb_reserve(skb, hlen);
2586 skb_reset_network_header(skb);
2590 if (sock->type == SOCK_DGRAM) {
2591 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2593 if (unlikely(err < 0))
2595 } else if (copylen) {
2596 int hdrlen = min_t(int, copylen, tp_len);
2598 skb_push(skb, dev->hard_header_len);
2599 skb_put(skb, copylen - dev->hard_header_len);
2600 err = skb_store_bits(skb, 0, data, hdrlen);
2603 if (!dev_validate_header(dev, skb->data, hdrlen))
2610 offset = offset_in_page(data);
2611 len_max = PAGE_SIZE - offset;
2612 len = ((to_write > len_max) ? len_max : to_write);
2614 skb->data_len = to_write;
2615 skb->len += to_write;
2616 skb->truesize += to_write;
2617 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2619 while (likely(to_write)) {
2620 nr_frags = skb_shinfo(skb)->nr_frags;
2622 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2623 pr_err("Packet exceed the number of skb frags(%lu)\n",
2628 page = pgv_to_page(data);
2630 flush_dcache_page(page);
2632 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2635 len_max = PAGE_SIZE;
2636 len = ((to_write > len_max) ? len_max : to_write);
2639 packet_parse_headers(skb, sock);
2644 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2645 int size_max, void **data)
2647 union tpacket_uhdr ph;
2652 switch (po->tp_version) {
2654 if (ph.h3->tp_next_offset != 0) {
2655 pr_warn_once("variable sized slot not supported");
2658 tp_len = ph.h3->tp_len;
2661 tp_len = ph.h2->tp_len;
2664 tp_len = ph.h1->tp_len;
2667 if (unlikely(tp_len > size_max)) {
2668 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2672 if (unlikely(po->tp_tx_has_off)) {
2673 int off_min, off_max;
2675 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2676 off_max = po->tx_ring.frame_size - tp_len;
2677 if (po->sk.sk_type == SOCK_DGRAM) {
2678 switch (po->tp_version) {
2680 off = ph.h3->tp_net;
2683 off = ph.h2->tp_net;
2686 off = ph.h1->tp_net;
2690 switch (po->tp_version) {
2692 off = ph.h3->tp_mac;
2695 off = ph.h2->tp_mac;
2698 off = ph.h1->tp_mac;
2702 if (unlikely((off < off_min) || (off_max < off)))
2705 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2708 *data = frame + off;
2712 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2714 struct sk_buff *skb = NULL;
2715 struct net_device *dev;
2716 struct virtio_net_hdr *vnet_hdr = NULL;
2717 struct sockcm_cookie sockc;
2719 int err, reserve = 0;
2721 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2722 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2723 unsigned char *addr = NULL;
2724 int tp_len, size_max;
2727 int status = TP_STATUS_AVAILABLE;
2728 int hlen, tlen, copylen = 0;
2731 mutex_lock(&po->pg_vec_lock);
2733 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2734 * we need to confirm it under protection of pg_vec_lock.
2736 if (unlikely(!po->tx_ring.pg_vec)) {
2740 if (likely(saddr == NULL)) {
2741 dev = packet_cached_dev_get(po);
2742 proto = READ_ONCE(po->num);
2745 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2747 if (msg->msg_namelen < (saddr->sll_halen
2748 + offsetof(struct sockaddr_ll,
2751 proto = saddr->sll_protocol;
2752 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2753 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2754 if (dev && msg->msg_namelen < dev->addr_len +
2755 offsetof(struct sockaddr_ll, sll_addr))
2757 addr = saddr->sll_addr;
2762 if (unlikely(dev == NULL))
2765 if (unlikely(!(dev->flags & IFF_UP)))
2768 sockcm_init(&sockc, &po->sk);
2769 if (msg->msg_controllen) {
2770 err = sock_cmsg_send(&po->sk, msg, &sockc);
2775 if (po->sk.sk_socket->type == SOCK_RAW)
2776 reserve = dev->hard_header_len;
2777 size_max = po->tx_ring.frame_size
2778 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2780 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2781 size_max = dev->mtu + reserve + VLAN_HLEN;
2783 reinit_completion(&po->skb_completion);
2786 ph = packet_current_frame(po, &po->tx_ring,
2787 TP_STATUS_SEND_REQUEST);
2788 if (unlikely(ph == NULL)) {
2789 if (need_wait && skb) {
2790 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2791 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2793 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2797 /* check for additional frames */
2802 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2806 status = TP_STATUS_SEND_REQUEST;
2807 hlen = LL_RESERVED_SPACE(dev);
2808 tlen = dev->needed_tailroom;
2809 if (po->has_vnet_hdr) {
2811 data += sizeof(*vnet_hdr);
2812 tp_len -= sizeof(*vnet_hdr);
2814 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2818 copylen = __virtio16_to_cpu(vio_le(),
2821 copylen = max_t(int, copylen, dev->hard_header_len);
2822 skb = sock_alloc_send_skb(&po->sk,
2823 hlen + tlen + sizeof(struct sockaddr_ll) +
2824 (copylen - dev->hard_header_len),
2827 if (unlikely(skb == NULL)) {
2828 /* we assume the socket was initially writeable ... */
2829 if (likely(len_sum > 0))
2833 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2834 addr, hlen, copylen, &sockc);
2835 if (likely(tp_len >= 0) &&
2836 tp_len > dev->mtu + reserve &&
2837 !po->has_vnet_hdr &&
2838 !packet_extra_vlan_len_allowed(dev, skb))
2841 if (unlikely(tp_len < 0)) {
2844 __packet_set_status(po, ph,
2845 TP_STATUS_AVAILABLE);
2846 packet_increment_head(&po->tx_ring);
2850 status = TP_STATUS_WRONG_FORMAT;
2856 if (po->has_vnet_hdr) {
2857 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2861 virtio_net_hdr_set_proto(skb, vnet_hdr);
2864 skb->destructor = tpacket_destruct_skb;
2865 __packet_set_status(po, ph, TP_STATUS_SENDING);
2866 packet_inc_pending(&po->tx_ring);
2868 status = TP_STATUS_SEND_REQUEST;
2869 err = po->xmit(skb);
2870 if (unlikely(err != 0)) {
2872 err = net_xmit_errno(err);
2873 if (err && __packet_get_status(po, ph) ==
2874 TP_STATUS_AVAILABLE) {
2875 /* skb was destructed already */
2880 * skb was dropped but not destructed yet;
2881 * let's treat it like congestion or err < 0
2885 packet_increment_head(&po->tx_ring);
2887 } while (likely((ph != NULL) ||
2888 /* Note: packet_read_pending() might be slow if we have
2889 * to call it as it's per_cpu variable, but in fast-path
2890 * we already short-circuit the loop with the first
2891 * condition, and luckily don't have to go that path
2894 (need_wait && packet_read_pending(&po->tx_ring))));
2900 __packet_set_status(po, ph, status);
2905 mutex_unlock(&po->pg_vec_lock);
2909 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2910 size_t reserve, size_t len,
2911 size_t linear, int noblock,
2914 struct sk_buff *skb;
2916 /* Under a page? Don't bother with paged skb. */
2917 if (prepad + len < PAGE_SIZE || !linear)
2920 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2925 skb_reserve(skb, reserve);
2926 skb_put(skb, linear);
2927 skb->data_len = len - linear;
2928 skb->len += len - linear;
2933 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2935 struct sock *sk = sock->sk;
2936 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2937 struct sk_buff *skb;
2938 struct net_device *dev;
2940 unsigned char *addr = NULL;
2941 int err, reserve = 0;
2942 struct sockcm_cookie sockc;
2943 struct virtio_net_hdr vnet_hdr = { 0 };
2945 struct packet_sock *po = pkt_sk(sk);
2946 bool has_vnet_hdr = false;
2947 int hlen, tlen, linear;
2951 * Get and verify the address.
2954 if (likely(saddr == NULL)) {
2955 dev = packet_cached_dev_get(po);
2956 proto = READ_ONCE(po->num);
2959 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2961 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2963 proto = saddr->sll_protocol;
2964 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2965 if (sock->type == SOCK_DGRAM) {
2966 if (dev && msg->msg_namelen < dev->addr_len +
2967 offsetof(struct sockaddr_ll, sll_addr))
2969 addr = saddr->sll_addr;
2974 if (unlikely(dev == NULL))
2977 if (unlikely(!(dev->flags & IFF_UP)))
2980 sockcm_init(&sockc, sk);
2981 sockc.mark = sk->sk_mark;
2982 if (msg->msg_controllen) {
2983 err = sock_cmsg_send(sk, msg, &sockc);
2988 if (sock->type == SOCK_RAW)
2989 reserve = dev->hard_header_len;
2990 if (po->has_vnet_hdr) {
2991 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2994 has_vnet_hdr = true;
2997 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2998 if (!netif_supports_nofcs(dev)) {
2999 err = -EPROTONOSUPPORT;
3002 extra_len = 4; /* We're doing our own CRC */
3006 if (!vnet_hdr.gso_type &&
3007 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3011 hlen = LL_RESERVED_SPACE(dev);
3012 tlen = dev->needed_tailroom;
3013 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3014 linear = max(linear, min_t(int, len, dev->hard_header_len));
3015 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3016 msg->msg_flags & MSG_DONTWAIT, &err);
3020 skb_reset_network_header(skb);
3023 if (sock->type == SOCK_DGRAM) {
3024 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3025 if (unlikely(offset < 0))
3027 } else if (reserve) {
3028 skb_reserve(skb, -reserve);
3029 if (len < reserve + sizeof(struct ipv6hdr) &&
3030 dev->min_header_len != dev->hard_header_len)
3031 skb_reset_network_header(skb);
3034 /* Returns -EFAULT on error */
3035 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3039 if ((sock->type == SOCK_RAW &&
3040 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3045 skb_setup_tx_timestamp(skb, sockc.tsflags);
3047 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3048 !packet_extra_vlan_len_allowed(dev, skb)) {
3053 skb->protocol = proto;
3055 skb->priority = sk->sk_priority;
3056 skb->mark = sockc.mark;
3057 skb->tstamp = sockc.transmit_time;
3059 if (unlikely(extra_len == 4))
3062 packet_parse_headers(skb, sock);
3065 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3068 len += sizeof(vnet_hdr);
3069 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3072 err = po->xmit(skb);
3073 if (unlikely(err != 0)) {
3075 err = net_xmit_errno(err);
3092 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3094 struct sock *sk = sock->sk;
3095 struct packet_sock *po = pkt_sk(sk);
3097 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3098 * tpacket_snd() will redo the check safely.
3100 if (data_race(po->tx_ring.pg_vec))
3101 return tpacket_snd(po, msg);
3103 return packet_snd(sock, msg, len);
3107 * Close a PACKET socket. This is fairly simple. We immediately go
3108 * to 'closed' state and remove our protocol entry in the device list.
3111 static int packet_release(struct socket *sock)
3113 struct sock *sk = sock->sk;
3114 struct packet_sock *po;
3115 struct packet_fanout *f;
3117 union tpacket_req_u req_u;
3125 mutex_lock(&net->packet.sklist_lock);
3126 sk_del_node_init_rcu(sk);
3127 mutex_unlock(&net->packet.sklist_lock);
3129 sock_prot_inuse_add(net, sk->sk_prot, -1);
3131 spin_lock(&po->bind_lock);
3132 unregister_prot_hook(sk, false);
3133 packet_cached_dev_reset(po);
3135 if (po->prot_hook.dev) {
3136 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3137 po->prot_hook.dev = NULL;
3139 spin_unlock(&po->bind_lock);
3141 packet_flush_mclist(sk);
3144 if (po->rx_ring.pg_vec) {
3145 memset(&req_u, 0, sizeof(req_u));
3146 packet_set_ring(sk, &req_u, 1, 0);
3149 if (po->tx_ring.pg_vec) {
3150 memset(&req_u, 0, sizeof(req_u));
3151 packet_set_ring(sk, &req_u, 1, 1);
3155 f = fanout_release(sk);
3159 kfree(po->rollover);
3161 fanout_release_data(f);
3165 * Now the socket is dead. No more input will appear.
3172 skb_queue_purge(&sk->sk_receive_queue);
3173 packet_free_pending(po);
3174 sk_refcnt_debug_release(sk);
3181 * Attach a packet hook.
3184 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3187 struct packet_sock *po = pkt_sk(sk);
3188 struct net_device *dev = NULL;
3189 bool unlisted = false;
3194 spin_lock(&po->bind_lock);
3203 dev = dev_get_by_name_rcu(sock_net(sk), name);
3208 } else if (ifindex) {
3209 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3216 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3222 /* prevents packet_notifier() from calling
3223 * register_prot_hook()
3225 WRITE_ONCE(po->num, 0);
3226 __unregister_prot_hook(sk, true);
3229 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3233 BUG_ON(po->running);
3234 WRITE_ONCE(po->num, proto);
3235 po->prot_hook.type = proto;
3237 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3239 if (unlikely(unlisted)) {
3240 po->prot_hook.dev = NULL;
3241 WRITE_ONCE(po->ifindex, -1);
3242 packet_cached_dev_reset(po);
3244 netdev_hold(dev, &po->prot_hook.dev_tracker,
3246 po->prot_hook.dev = dev;
3247 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3248 packet_cached_dev_assign(po, dev);
3253 if (proto == 0 || !need_rehook)
3256 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3257 register_prot_hook(sk);
3259 sk->sk_err = ENETDOWN;
3260 if (!sock_flag(sk, SOCK_DEAD))
3261 sk_error_report(sk);
3266 spin_unlock(&po->bind_lock);
3272 * Bind a packet socket to a device
3275 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3278 struct sock *sk = sock->sk;
3279 char name[sizeof(uaddr->sa_data) + 1];
3285 if (addr_len != sizeof(struct sockaddr))
3287 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3290 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3291 name[sizeof(uaddr->sa_data)] = 0;
3293 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3296 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3298 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3299 struct sock *sk = sock->sk;
3305 if (addr_len < sizeof(struct sockaddr_ll))
3307 if (sll->sll_family != AF_PACKET)
3310 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3311 sll->sll_protocol ? : pkt_sk(sk)->num);
3314 static struct proto packet_proto = {
3316 .owner = THIS_MODULE,
3317 .obj_size = sizeof(struct packet_sock),
3321 * Create a packet of type SOCK_PACKET.
3324 static int packet_create(struct net *net, struct socket *sock, int protocol,
3328 struct packet_sock *po;
3329 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3332 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3334 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3335 sock->type != SOCK_PACKET)
3336 return -ESOCKTNOSUPPORT;
3338 sock->state = SS_UNCONNECTED;
3341 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3345 sock->ops = &packet_ops;
3346 if (sock->type == SOCK_PACKET)
3347 sock->ops = &packet_ops_spkt;
3349 sock_init_data(sock, sk);
3352 init_completion(&po->skb_completion);
3353 sk->sk_family = PF_PACKET;
3355 po->xmit = dev_queue_xmit;
3357 err = packet_alloc_pending(po);
3361 packet_cached_dev_reset(po);
3363 sk->sk_destruct = packet_sock_destruct;
3364 sk_refcnt_debug_inc(sk);
3367 * Attach a protocol block
3370 spin_lock_init(&po->bind_lock);
3371 mutex_init(&po->pg_vec_lock);
3372 po->rollover = NULL;
3373 po->prot_hook.func = packet_rcv;
3375 if (sock->type == SOCK_PACKET)
3376 po->prot_hook.func = packet_rcv_spkt;
3378 po->prot_hook.af_packet_priv = sk;
3379 po->prot_hook.af_packet_net = sock_net(sk);
3382 po->prot_hook.type = proto;
3383 __register_prot_hook(sk);
3386 mutex_lock(&net->packet.sklist_lock);
3387 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3388 mutex_unlock(&net->packet.sklist_lock);
3390 sock_prot_inuse_add(net, &packet_proto, 1);
3400 * Pull a packet from our receive queue and hand it to the user.
3401 * If necessary we block.
3404 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3407 struct sock *sk = sock->sk;
3408 struct sk_buff *skb;
3410 int vnet_hdr_len = 0;
3411 unsigned int origlen = 0;
3414 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3418 /* What error should we return now? EUNATTACH? */
3419 if (pkt_sk(sk)->ifindex < 0)
3423 if (flags & MSG_ERRQUEUE) {
3424 err = sock_recv_errqueue(sk, msg, len,
3425 SOL_PACKET, PACKET_TX_TIMESTAMP);
3430 * Call the generic datagram receiver. This handles all sorts
3431 * of horrible races and re-entrancy so we can forget about it
3432 * in the protocol layers.
3434 * Now it will return ENETDOWN, if device have just gone down,
3435 * but then it will block.
3438 skb = skb_recv_datagram(sk, flags, &err);
3441 * An error occurred so return it. Because skb_recv_datagram()
3442 * handles the blocking we don't see and worry about blocking
3449 packet_rcv_try_clear_pressure(pkt_sk(sk));
3451 if (pkt_sk(sk)->has_vnet_hdr) {
3452 err = packet_rcv_vnet(msg, skb, &len);
3455 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3458 /* You lose any data beyond the buffer you gave. If it worries
3459 * a user program they can ask the device for its MTU
3465 msg->msg_flags |= MSG_TRUNC;
3468 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3472 if (sock->type != SOCK_PACKET) {
3473 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3475 /* Original length was stored in sockaddr_ll fields */
3476 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3477 sll->sll_family = AF_PACKET;
3478 sll->sll_protocol = skb->protocol;
3481 sock_recv_cmsgs(msg, sk, skb);
3483 if (msg->msg_name) {
3484 const size_t max_len = min(sizeof(skb->cb),
3485 sizeof(struct sockaddr_storage));
3488 /* If the address length field is there to be filled
3489 * in, we fill it in now.
3491 if (sock->type == SOCK_PACKET) {
3492 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3493 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3494 copy_len = msg->msg_namelen;
3496 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3498 msg->msg_namelen = sll->sll_halen +
3499 offsetof(struct sockaddr_ll, sll_addr);
3500 copy_len = msg->msg_namelen;
3501 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3502 memset(msg->msg_name +
3503 offsetof(struct sockaddr_ll, sll_addr),
3504 0, sizeof(sll->sll_addr));
3505 msg->msg_namelen = sizeof(struct sockaddr_ll);
3508 if (WARN_ON_ONCE(copy_len > max_len)) {
3510 msg->msg_namelen = copy_len;
3512 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3515 if (pkt_sk(sk)->auxdata) {
3516 struct tpacket_auxdata aux;
3518 aux.tp_status = TP_STATUS_USER;
3519 if (skb->ip_summed == CHECKSUM_PARTIAL)
3520 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3521 else if (skb->pkt_type != PACKET_OUTGOING &&
3522 skb_csum_unnecessary(skb))
3523 aux.tp_status |= TP_STATUS_CSUM_VALID;
3525 aux.tp_len = origlen;
3526 aux.tp_snaplen = skb->len;
3528 aux.tp_net = skb_network_offset(skb);
3529 if (skb_vlan_tag_present(skb)) {
3530 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3531 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3532 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3534 aux.tp_vlan_tci = 0;
3535 aux.tp_vlan_tpid = 0;
3537 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3541 * Free or return the buffer as appropriate. Again this
3542 * hides all the races and re-entrancy issues from us.
3544 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3547 skb_free_datagram(sk, skb);
3552 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3555 struct net_device *dev;
3556 struct sock *sk = sock->sk;
3561 uaddr->sa_family = AF_PACKET;
3562 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3564 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3566 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3569 return sizeof(*uaddr);
3572 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3575 struct net_device *dev;
3576 struct sock *sk = sock->sk;
3577 struct packet_sock *po = pkt_sk(sk);
3578 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3584 ifindex = READ_ONCE(po->ifindex);
3585 sll->sll_family = AF_PACKET;
3586 sll->sll_ifindex = ifindex;
3587 sll->sll_protocol = READ_ONCE(po->num);
3588 sll->sll_pkttype = 0;
3590 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3592 sll->sll_hatype = dev->type;
3593 sll->sll_halen = dev->addr_len;
3594 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3596 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3601 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3604 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3608 case PACKET_MR_MULTICAST:
3609 if (i->alen != dev->addr_len)
3612 return dev_mc_add(dev, i->addr);
3614 return dev_mc_del(dev, i->addr);
3616 case PACKET_MR_PROMISC:
3617 return dev_set_promiscuity(dev, what);
3618 case PACKET_MR_ALLMULTI:
3619 return dev_set_allmulti(dev, what);
3620 case PACKET_MR_UNICAST:
3621 if (i->alen != dev->addr_len)
3624 return dev_uc_add(dev, i->addr);
3626 return dev_uc_del(dev, i->addr);
3634 static void packet_dev_mclist_delete(struct net_device *dev,
3635 struct packet_mclist **mlp)
3637 struct packet_mclist *ml;
3639 while ((ml = *mlp) != NULL) {
3640 if (ml->ifindex == dev->ifindex) {
3641 packet_dev_mc(dev, ml, -1);
3649 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3651 struct packet_sock *po = pkt_sk(sk);
3652 struct packet_mclist *ml, *i;
3653 struct net_device *dev;
3659 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3664 if (mreq->mr_alen > dev->addr_len)
3668 i = kmalloc(sizeof(*i), GFP_KERNEL);
3673 for (ml = po->mclist; ml; ml = ml->next) {
3674 if (ml->ifindex == mreq->mr_ifindex &&
3675 ml->type == mreq->mr_type &&
3676 ml->alen == mreq->mr_alen &&
3677 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3679 /* Free the new element ... */
3685 i->type = mreq->mr_type;
3686 i->ifindex = mreq->mr_ifindex;
3687 i->alen = mreq->mr_alen;
3688 memcpy(i->addr, mreq->mr_address, i->alen);
3689 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3691 i->next = po->mclist;
3693 err = packet_dev_mc(dev, i, 1);
3695 po->mclist = i->next;
3704 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3706 struct packet_mclist *ml, **mlp;
3710 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3711 if (ml->ifindex == mreq->mr_ifindex &&
3712 ml->type == mreq->mr_type &&
3713 ml->alen == mreq->mr_alen &&
3714 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3715 if (--ml->count == 0) {
3716 struct net_device *dev;
3718 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3720 packet_dev_mc(dev, ml, -1);
3730 static void packet_flush_mclist(struct sock *sk)
3732 struct packet_sock *po = pkt_sk(sk);
3733 struct packet_mclist *ml;
3739 while ((ml = po->mclist) != NULL) {
3740 struct net_device *dev;
3742 po->mclist = ml->next;
3743 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3745 packet_dev_mc(dev, ml, -1);
3752 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3753 unsigned int optlen)
3755 struct sock *sk = sock->sk;
3756 struct packet_sock *po = pkt_sk(sk);
3759 if (level != SOL_PACKET)
3760 return -ENOPROTOOPT;
3763 case PACKET_ADD_MEMBERSHIP:
3764 case PACKET_DROP_MEMBERSHIP:
3766 struct packet_mreq_max mreq;
3768 memset(&mreq, 0, sizeof(mreq));
3769 if (len < sizeof(struct packet_mreq))
3771 if (len > sizeof(mreq))
3773 if (copy_from_sockptr(&mreq, optval, len))
3775 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3777 if (optname == PACKET_ADD_MEMBERSHIP)
3778 ret = packet_mc_add(sk, &mreq);
3780 ret = packet_mc_drop(sk, &mreq);
3784 case PACKET_RX_RING:
3785 case PACKET_TX_RING:
3787 union tpacket_req_u req_u;
3791 switch (po->tp_version) {
3794 len = sizeof(req_u.req);
3798 len = sizeof(req_u.req3);
3804 if (copy_from_sockptr(&req_u.req, optval, len))
3807 ret = packet_set_ring(sk, &req_u, 0,
3808 optname == PACKET_TX_RING);
3813 case PACKET_COPY_THRESH:
3817 if (optlen != sizeof(val))
3819 if (copy_from_sockptr(&val, optval, sizeof(val)))
3822 pkt_sk(sk)->copy_thresh = val;
3825 case PACKET_VERSION:
3829 if (optlen != sizeof(val))
3831 if (copy_from_sockptr(&val, optval, sizeof(val)))
3842 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3845 po->tp_version = val;
3851 case PACKET_RESERVE:
3855 if (optlen != sizeof(val))
3857 if (copy_from_sockptr(&val, optval, sizeof(val)))
3862 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3865 po->tp_reserve = val;
3875 if (optlen != sizeof(val))
3877 if (copy_from_sockptr(&val, optval, sizeof(val)))
3881 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3884 po->tp_loss = !!val;
3890 case PACKET_AUXDATA:
3894 if (optlen < sizeof(val))
3896 if (copy_from_sockptr(&val, optval, sizeof(val)))
3900 po->auxdata = !!val;
3904 case PACKET_ORIGDEV:
3908 if (optlen < sizeof(val))
3910 if (copy_from_sockptr(&val, optval, sizeof(val)))
3914 po->origdev = !!val;
3918 case PACKET_VNET_HDR:
3922 if (sock->type != SOCK_RAW)
3924 if (optlen < sizeof(val))
3926 if (copy_from_sockptr(&val, optval, sizeof(val)))
3930 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3933 po->has_vnet_hdr = !!val;
3939 case PACKET_TIMESTAMP:
3943 if (optlen != sizeof(val))
3945 if (copy_from_sockptr(&val, optval, sizeof(val)))
3948 po->tp_tstamp = val;
3953 struct fanout_args args = { 0 };
3955 if (optlen != sizeof(int) && optlen != sizeof(args))
3957 if (copy_from_sockptr(&args, optval, optlen))
3960 return fanout_add(sk, &args);
3962 case PACKET_FANOUT_DATA:
3964 /* Paired with the WRITE_ONCE() in fanout_add() */
3965 if (!READ_ONCE(po->fanout))
3968 return fanout_set_data(po, optval, optlen);
3970 case PACKET_IGNORE_OUTGOING:
3974 if (optlen != sizeof(val))
3976 if (copy_from_sockptr(&val, optval, sizeof(val)))
3978 if (val < 0 || val > 1)
3981 po->prot_hook.ignore_outgoing = !!val;
3984 case PACKET_TX_HAS_OFF:
3988 if (optlen != sizeof(val))
3990 if (copy_from_sockptr(&val, optval, sizeof(val)))
3994 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
3995 po->tp_tx_has_off = !!val;
4000 case PACKET_QDISC_BYPASS:
4004 if (optlen != sizeof(val))
4006 if (copy_from_sockptr(&val, optval, sizeof(val)))
4009 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
4013 return -ENOPROTOOPT;
4017 static int packet_getsockopt(struct socket *sock, int level, int optname,
4018 char __user *optval, int __user *optlen)
4021 int val, lv = sizeof(val);
4022 struct sock *sk = sock->sk;
4023 struct packet_sock *po = pkt_sk(sk);
4025 union tpacket_stats_u st;
4026 struct tpacket_rollover_stats rstats;
4029 if (level != SOL_PACKET)
4030 return -ENOPROTOOPT;
4032 if (get_user(len, optlen))
4039 case PACKET_STATISTICS:
4040 spin_lock_bh(&sk->sk_receive_queue.lock);
4041 memcpy(&st, &po->stats, sizeof(st));
4042 memset(&po->stats, 0, sizeof(po->stats));
4043 spin_unlock_bh(&sk->sk_receive_queue.lock);
4044 drops = atomic_xchg(&po->tp_drops, 0);
4046 if (po->tp_version == TPACKET_V3) {
4047 lv = sizeof(struct tpacket_stats_v3);
4048 st.stats3.tp_drops = drops;
4049 st.stats3.tp_packets += drops;
4052 lv = sizeof(struct tpacket_stats);
4053 st.stats1.tp_drops = drops;
4054 st.stats1.tp_packets += drops;
4059 case PACKET_AUXDATA:
4062 case PACKET_ORIGDEV:
4065 case PACKET_VNET_HDR:
4066 val = po->has_vnet_hdr;
4068 case PACKET_VERSION:
4069 val = po->tp_version;
4072 if (len > sizeof(int))
4074 if (len < sizeof(int))
4076 if (copy_from_user(&val, optval, len))
4080 val = sizeof(struct tpacket_hdr);
4083 val = sizeof(struct tpacket2_hdr);
4086 val = sizeof(struct tpacket3_hdr);
4092 case PACKET_RESERVE:
4093 val = po->tp_reserve;
4098 case PACKET_TIMESTAMP:
4099 val = po->tp_tstamp;
4103 ((u32)po->fanout->id |
4104 ((u32)po->fanout->type << 16) |
4105 ((u32)po->fanout->flags << 24)) :
4108 case PACKET_IGNORE_OUTGOING:
4109 val = po->prot_hook.ignore_outgoing;
4111 case PACKET_ROLLOVER_STATS:
4114 rstats.tp_all = atomic_long_read(&po->rollover->num);
4115 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4116 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4118 lv = sizeof(rstats);
4120 case PACKET_TX_HAS_OFF:
4121 val = po->tp_tx_has_off;
4123 case PACKET_QDISC_BYPASS:
4124 val = packet_use_direct_xmit(po);
4127 return -ENOPROTOOPT;
4132 if (put_user(len, optlen))
4134 if (copy_to_user(optval, data, len))
4139 static int packet_notifier(struct notifier_block *this,
4140 unsigned long msg, void *ptr)
4143 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4144 struct net *net = dev_net(dev);
4147 sk_for_each_rcu(sk, &net->packet.sklist) {
4148 struct packet_sock *po = pkt_sk(sk);
4151 case NETDEV_UNREGISTER:
4153 packet_dev_mclist_delete(dev, &po->mclist);
4157 if (dev->ifindex == po->ifindex) {
4158 spin_lock(&po->bind_lock);
4160 __unregister_prot_hook(sk, false);
4161 sk->sk_err = ENETDOWN;
4162 if (!sock_flag(sk, SOCK_DEAD))
4163 sk_error_report(sk);
4165 if (msg == NETDEV_UNREGISTER) {
4166 packet_cached_dev_reset(po);
4167 WRITE_ONCE(po->ifindex, -1);
4168 netdev_put(po->prot_hook.dev,
4169 &po->prot_hook.dev_tracker);
4170 po->prot_hook.dev = NULL;
4172 spin_unlock(&po->bind_lock);
4176 if (dev->ifindex == po->ifindex) {
4177 spin_lock(&po->bind_lock);
4179 register_prot_hook(sk);
4180 spin_unlock(&po->bind_lock);
4190 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4193 struct sock *sk = sock->sk;
4198 int amount = sk_wmem_alloc_get(sk);
4200 return put_user(amount, (int __user *)arg);
4204 struct sk_buff *skb;
4207 spin_lock_bh(&sk->sk_receive_queue.lock);
4208 skb = skb_peek(&sk->sk_receive_queue);
4211 spin_unlock_bh(&sk->sk_receive_queue.lock);
4212 return put_user(amount, (int __user *)arg);
4222 case SIOCGIFBRDADDR:
4223 case SIOCSIFBRDADDR:
4224 case SIOCGIFNETMASK:
4225 case SIOCSIFNETMASK:
4226 case SIOCGIFDSTADDR:
4227 case SIOCSIFDSTADDR:
4229 return inet_dgram_ops.ioctl(sock, cmd, arg);
4233 return -ENOIOCTLCMD;
4238 static __poll_t packet_poll(struct file *file, struct socket *sock,
4241 struct sock *sk = sock->sk;
4242 struct packet_sock *po = pkt_sk(sk);
4243 __poll_t mask = datagram_poll(file, sock, wait);
4245 spin_lock_bh(&sk->sk_receive_queue.lock);
4246 if (po->rx_ring.pg_vec) {
4247 if (!packet_previous_rx_frame(po, &po->rx_ring,
4249 mask |= EPOLLIN | EPOLLRDNORM;
4251 packet_rcv_try_clear_pressure(po);
4252 spin_unlock_bh(&sk->sk_receive_queue.lock);
4253 spin_lock_bh(&sk->sk_write_queue.lock);
4254 if (po->tx_ring.pg_vec) {
4255 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4256 mask |= EPOLLOUT | EPOLLWRNORM;
4258 spin_unlock_bh(&sk->sk_write_queue.lock);
4263 /* Dirty? Well, I still did not learn better way to account
4267 static void packet_mm_open(struct vm_area_struct *vma)
4269 struct file *file = vma->vm_file;
4270 struct socket *sock = file->private_data;
4271 struct sock *sk = sock->sk;
4274 atomic_inc(&pkt_sk(sk)->mapped);
4277 static void packet_mm_close(struct vm_area_struct *vma)
4279 struct file *file = vma->vm_file;
4280 struct socket *sock = file->private_data;
4281 struct sock *sk = sock->sk;
4284 atomic_dec(&pkt_sk(sk)->mapped);
4287 static const struct vm_operations_struct packet_mmap_ops = {
4288 .open = packet_mm_open,
4289 .close = packet_mm_close,
4292 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4297 for (i = 0; i < len; i++) {
4298 if (likely(pg_vec[i].buffer)) {
4299 if (is_vmalloc_addr(pg_vec[i].buffer))
4300 vfree(pg_vec[i].buffer);
4302 free_pages((unsigned long)pg_vec[i].buffer,
4304 pg_vec[i].buffer = NULL;
4310 static char *alloc_one_pg_vec_page(unsigned long order)
4313 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4314 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4316 buffer = (char *) __get_free_pages(gfp_flags, order);
4320 /* __get_free_pages failed, fall back to vmalloc */
4321 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4325 /* vmalloc failed, lets dig into swap here */
4326 gfp_flags &= ~__GFP_NORETRY;
4327 buffer = (char *) __get_free_pages(gfp_flags, order);
4331 /* complete and utter failure */
4335 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4337 unsigned int block_nr = req->tp_block_nr;
4341 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4342 if (unlikely(!pg_vec))
4345 for (i = 0; i < block_nr; i++) {
4346 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4347 if (unlikely(!pg_vec[i].buffer))
4348 goto out_free_pgvec;
4355 free_pg_vec(pg_vec, order, block_nr);
4360 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4361 int closing, int tx_ring)
4363 struct pgv *pg_vec = NULL;
4364 struct packet_sock *po = pkt_sk(sk);
4365 unsigned long *rx_owner_map = NULL;
4366 int was_running, order = 0;
4367 struct packet_ring_buffer *rb;
4368 struct sk_buff_head *rb_queue;
4371 /* Added to avoid minimal code churn */
4372 struct tpacket_req *req = &req_u->req;
4374 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4375 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4379 if (atomic_read(&po->mapped))
4381 if (packet_read_pending(rb))
4385 if (req->tp_block_nr) {
4386 unsigned int min_frame_size;
4388 /* Sanity tests and some calculations */
4390 if (unlikely(rb->pg_vec))
4393 switch (po->tp_version) {
4395 po->tp_hdrlen = TPACKET_HDRLEN;
4398 po->tp_hdrlen = TPACKET2_HDRLEN;
4401 po->tp_hdrlen = TPACKET3_HDRLEN;
4406 if (unlikely((int)req->tp_block_size <= 0))
4408 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4410 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4411 if (po->tp_version >= TPACKET_V3 &&
4412 req->tp_block_size <
4413 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4415 if (unlikely(req->tp_frame_size < min_frame_size))
4417 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4420 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4421 if (unlikely(rb->frames_per_block == 0))
4423 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4425 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4430 order = get_order(req->tp_block_size);
4431 pg_vec = alloc_pg_vec(req, order);
4432 if (unlikely(!pg_vec))
4434 switch (po->tp_version) {
4436 /* Block transmit is not supported yet */
4438 init_prb_bdqc(po, rb, pg_vec, req_u);
4440 struct tpacket_req3 *req3 = &req_u->req3;
4442 if (req3->tp_retire_blk_tov ||
4443 req3->tp_sizeof_priv ||
4444 req3->tp_feature_req_word) {
4446 goto out_free_pg_vec;
4452 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4453 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4455 goto out_free_pg_vec;
4463 if (unlikely(req->tp_frame_nr))
4468 /* Detach socket from network */
4469 spin_lock(&po->bind_lock);
4470 was_running = po->running;
4473 WRITE_ONCE(po->num, 0);
4474 __unregister_prot_hook(sk, false);
4476 spin_unlock(&po->bind_lock);
4481 mutex_lock(&po->pg_vec_lock);
4482 if (closing || atomic_read(&po->mapped) == 0) {
4484 spin_lock_bh(&rb_queue->lock);
4485 swap(rb->pg_vec, pg_vec);
4486 if (po->tp_version <= TPACKET_V2)
4487 swap(rb->rx_owner_map, rx_owner_map);
4488 rb->frame_max = (req->tp_frame_nr - 1);
4490 rb->frame_size = req->tp_frame_size;
4491 spin_unlock_bh(&rb_queue->lock);
4493 swap(rb->pg_vec_order, order);
4494 swap(rb->pg_vec_len, req->tp_block_nr);
4496 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4497 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4498 tpacket_rcv : packet_rcv;
4499 skb_queue_purge(rb_queue);
4500 if (atomic_read(&po->mapped))
4501 pr_err("packet_mmap: vma is busy: %d\n",
4502 atomic_read(&po->mapped));
4504 mutex_unlock(&po->pg_vec_lock);
4506 spin_lock(&po->bind_lock);
4508 WRITE_ONCE(po->num, num);
4509 register_prot_hook(sk);
4511 spin_unlock(&po->bind_lock);
4512 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4513 /* Because we don't support block-based V3 on tx-ring */
4515 prb_shutdown_retire_blk_timer(po, rb_queue);
4520 bitmap_free(rx_owner_map);
4521 free_pg_vec(pg_vec, order, req->tp_block_nr);
4527 static int packet_mmap(struct file *file, struct socket *sock,
4528 struct vm_area_struct *vma)
4530 struct sock *sk = sock->sk;
4531 struct packet_sock *po = pkt_sk(sk);
4532 unsigned long size, expected_size;
4533 struct packet_ring_buffer *rb;
4534 unsigned long start;
4541 mutex_lock(&po->pg_vec_lock);
4544 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4546 expected_size += rb->pg_vec_len
4552 if (expected_size == 0)
4555 size = vma->vm_end - vma->vm_start;
4556 if (size != expected_size)
4559 start = vma->vm_start;
4560 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4561 if (rb->pg_vec == NULL)
4564 for (i = 0; i < rb->pg_vec_len; i++) {
4566 void *kaddr = rb->pg_vec[i].buffer;
4569 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4570 page = pgv_to_page(kaddr);
4571 err = vm_insert_page(vma, start, page);
4580 atomic_inc(&po->mapped);
4581 vma->vm_ops = &packet_mmap_ops;
4585 mutex_unlock(&po->pg_vec_lock);
4589 static const struct proto_ops packet_ops_spkt = {
4590 .family = PF_PACKET,
4591 .owner = THIS_MODULE,
4592 .release = packet_release,
4593 .bind = packet_bind_spkt,
4594 .connect = sock_no_connect,
4595 .socketpair = sock_no_socketpair,
4596 .accept = sock_no_accept,
4597 .getname = packet_getname_spkt,
4598 .poll = datagram_poll,
4599 .ioctl = packet_ioctl,
4600 .gettstamp = sock_gettstamp,
4601 .listen = sock_no_listen,
4602 .shutdown = sock_no_shutdown,
4603 .sendmsg = packet_sendmsg_spkt,
4604 .recvmsg = packet_recvmsg,
4605 .mmap = sock_no_mmap,
4606 .sendpage = sock_no_sendpage,
4609 static const struct proto_ops packet_ops = {
4610 .family = PF_PACKET,
4611 .owner = THIS_MODULE,
4612 .release = packet_release,
4613 .bind = packet_bind,
4614 .connect = sock_no_connect,
4615 .socketpair = sock_no_socketpair,
4616 .accept = sock_no_accept,
4617 .getname = packet_getname,
4618 .poll = packet_poll,
4619 .ioctl = packet_ioctl,
4620 .gettstamp = sock_gettstamp,
4621 .listen = sock_no_listen,
4622 .shutdown = sock_no_shutdown,
4623 .setsockopt = packet_setsockopt,
4624 .getsockopt = packet_getsockopt,
4625 .sendmsg = packet_sendmsg,
4626 .recvmsg = packet_recvmsg,
4627 .mmap = packet_mmap,
4628 .sendpage = sock_no_sendpage,
4631 static const struct net_proto_family packet_family_ops = {
4632 .family = PF_PACKET,
4633 .create = packet_create,
4634 .owner = THIS_MODULE,
4637 static struct notifier_block packet_netdev_notifier = {
4638 .notifier_call = packet_notifier,
4641 #ifdef CONFIG_PROC_FS
4643 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4646 struct net *net = seq_file_net(seq);
4649 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4652 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4654 struct net *net = seq_file_net(seq);
4655 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4658 static void packet_seq_stop(struct seq_file *seq, void *v)
4664 static int packet_seq_show(struct seq_file *seq, void *v)
4666 if (v == SEQ_START_TOKEN)
4668 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4669 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4671 struct sock *s = sk_entry(v);
4672 const struct packet_sock *po = pkt_sk(s);
4675 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4677 refcount_read(&s->sk_refcnt),
4679 ntohs(READ_ONCE(po->num)),
4680 READ_ONCE(po->ifindex),
4682 atomic_read(&s->sk_rmem_alloc),
4683 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4690 static const struct seq_operations packet_seq_ops = {
4691 .start = packet_seq_start,
4692 .next = packet_seq_next,
4693 .stop = packet_seq_stop,
4694 .show = packet_seq_show,
4698 static int __net_init packet_net_init(struct net *net)
4700 mutex_init(&net->packet.sklist_lock);
4701 INIT_HLIST_HEAD(&net->packet.sklist);
4703 #ifdef CONFIG_PROC_FS
4704 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4705 sizeof(struct seq_net_private)))
4707 #endif /* CONFIG_PROC_FS */
4712 static void __net_exit packet_net_exit(struct net *net)
4714 remove_proc_entry("packet", net->proc_net);
4715 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4718 static struct pernet_operations packet_net_ops = {
4719 .init = packet_net_init,
4720 .exit = packet_net_exit,
4724 static void __exit packet_exit(void)
4726 sock_unregister(PF_PACKET);
4727 proto_unregister(&packet_proto);
4728 unregister_netdevice_notifier(&packet_netdev_notifier);
4729 unregister_pernet_subsys(&packet_net_ops);
4732 static int __init packet_init(void)
4736 rc = register_pernet_subsys(&packet_net_ops);
4739 rc = register_netdevice_notifier(&packet_netdev_notifier);
4742 rc = proto_register(&packet_proto, 0);
4745 rc = sock_register(&packet_family_ops);
4752 proto_unregister(&packet_proto);
4754 unregister_netdevice_notifier(&packet_netdev_notifier);
4756 unregister_pernet_subsys(&packet_net_ops);
4761 module_init(packet_init);
4762 module_exit(packet_exit);
4763 MODULE_LICENSE("GPL");
4764 MODULE_ALIAS_NETPROTO(PF_PACKET);