1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PACKET - implements raw packet sockets.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
49 #include <linux/types.h>
51 #include <linux/capability.h>
52 #include <linux/fcntl.h>
53 #include <linux/socket.h>
55 #include <linux/inet.h>
56 #include <linux/netdevice.h>
57 #include <linux/if_packet.h>
58 #include <linux/wireless.h>
59 #include <linux/kernel.h>
60 #include <linux/kmod.h>
61 #include <linux/slab.h>
62 #include <linux/vmalloc.h>
63 #include <net/net_namespace.h>
65 #include <net/protocol.h>
66 #include <linux/skbuff.h>
68 #include <linux/errno.h>
69 #include <linux/timer.h>
70 #include <linux/uaccess.h>
71 #include <asm/ioctls.h>
73 #include <asm/cacheflush.h>
75 #include <linux/proc_fs.h>
76 #include <linux/seq_file.h>
77 #include <linux/poll.h>
78 #include <linux/module.h>
79 #include <linux/init.h>
80 #include <linux/mutex.h>
81 #include <linux/if_vlan.h>
82 #include <linux/virtio_net.h>
83 #include <linux/errqueue.h>
84 #include <linux/net_tstamp.h>
85 #include <linux/percpu.h>
87 #include <net/inet_common.h>
89 #include <linux/bpf.h>
90 #include <net/compat.h>
96 - If the device has no dev->header_ops->create, there is no LL header
97 visible above the device. In this case, its hard_header_len should be 0.
98 The device may prepend its own header internally. In this case, its
99 needed_headroom should be set to the space needed for it to add its
101 For example, a WiFi driver pretending to be an Ethernet driver should
102 set its hard_header_len to be the Ethernet header length, and set its
103 needed_headroom to be (the real WiFi header length - the fake Ethernet
105 - packet socket receives packets with pulled ll header,
106 so that SOCK_RAW should push it back.
111 Incoming, dev_has_header(dev) == true
112 mac_header -> ll header
115 Outgoing, dev_has_header(dev) == true
116 mac_header -> ll header
119 Incoming, dev_has_header(dev) == false
121 However drivers often make it point to the ll header.
122 This is incorrect because the ll header should be invisible to us.
125 Outgoing, dev_has_header(dev) == false
126 mac_header -> data. ll header is invisible to us.
130 If dev_has_header(dev) == false we are unable to restore the ll header,
131 because it is invisible to us.
137 dev->header_ops != NULL
138 mac_header -> ll header
141 dev->header_ops == NULL (ll header is invisible to us)
145 We should set network_header on output to the correct position,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
179 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
180 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
181 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
182 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
183 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
187 struct packet_type *pt, struct net_device *orig_dev);
189 static void *packet_previous_frame(struct packet_sock *po,
190 struct packet_ring_buffer *rb,
192 static void packet_increment_head(struct packet_ring_buffer *buff);
193 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
194 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
195 struct packet_sock *);
196 static void prb_retire_current_block(struct tpacket_kbdq_core *,
197 struct packet_sock *, unsigned int status);
198 static int prb_queue_frozen(struct tpacket_kbdq_core *);
199 static void prb_open_block(struct tpacket_kbdq_core *,
200 struct tpacket_block_desc *);
201 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
202 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
203 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
204 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
205 struct tpacket3_hdr *);
206 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
207 struct tpacket3_hdr *);
208 static void packet_flush_mclist(struct sock *sk);
209 static u16 packet_pick_tx_queue(struct sk_buff *skb);
211 struct packet_skb_cb {
213 struct sockaddr_pkt pkt;
215 /* Trick: alias skb original length with
216 * ll.sll_family and ll.protocol in order
219 unsigned int origlen;
220 struct sockaddr_ll ll;
225 #define vio_le() virtio_legacy_is_little_endian()
227 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
229 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
230 #define GET_PBLOCK_DESC(x, bid) \
231 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
232 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
233 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
234 #define GET_NEXT_PRB_BLK_NUM(x) \
235 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
236 ((x)->kactive_blk_num+1) : 0)
238 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
239 static void __fanout_link(struct sock *sk, struct packet_sock *po);
241 static int packet_direct_xmit(struct sk_buff *skb)
243 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
246 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
248 struct net_device *dev;
251 dev = rcu_dereference(po->cached_dev);
259 static void packet_cached_dev_assign(struct packet_sock *po,
260 struct net_device *dev)
262 rcu_assign_pointer(po->cached_dev, dev);
265 static void packet_cached_dev_reset(struct packet_sock *po)
267 RCU_INIT_POINTER(po->cached_dev, NULL);
270 static bool packet_use_direct_xmit(const struct packet_sock *po)
272 return po->xmit == packet_direct_xmit;
275 static u16 packet_pick_tx_queue(struct sk_buff *skb)
277 struct net_device *dev = skb->dev;
278 const struct net_device_ops *ops = dev->netdev_ops;
279 int cpu = raw_smp_processor_id();
283 skb->sender_cpu = cpu + 1;
285 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
286 if (ops->ndo_select_queue) {
287 queue_index = ops->ndo_select_queue(dev, skb, NULL);
288 queue_index = netdev_cap_txqueue(dev, queue_index);
290 queue_index = netdev_pick_tx(dev, skb, NULL);
296 /* __register_prot_hook must be invoked through register_prot_hook
297 * or from a context in which asynchronous accesses to the packet
298 * socket is not possible (packet_create()).
300 static void __register_prot_hook(struct sock *sk)
302 struct packet_sock *po = pkt_sk(sk);
306 __fanout_link(sk, po);
308 dev_add_pack(&po->prot_hook);
315 static void register_prot_hook(struct sock *sk)
317 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
318 __register_prot_hook(sk);
321 /* If the sync parameter is true, we will temporarily drop
322 * the po->bind_lock and do a synchronize_net to make sure no
323 * asynchronous packet processing paths still refer to the elements
324 * of po->prot_hook. If the sync parameter is false, it is the
325 * callers responsibility to take care of this.
327 static void __unregister_prot_hook(struct sock *sk, bool sync)
329 struct packet_sock *po = pkt_sk(sk);
331 lockdep_assert_held_once(&po->bind_lock);
336 __fanout_unlink(sk, po);
338 __dev_remove_pack(&po->prot_hook);
343 spin_unlock(&po->bind_lock);
345 spin_lock(&po->bind_lock);
349 static void unregister_prot_hook(struct sock *sk, bool sync)
351 struct packet_sock *po = pkt_sk(sk);
354 __unregister_prot_hook(sk, sync);
357 static inline struct page * __pure pgv_to_page(void *addr)
359 if (is_vmalloc_addr(addr))
360 return vmalloc_to_page(addr);
361 return virt_to_page(addr);
364 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
366 union tpacket_uhdr h;
369 switch (po->tp_version) {
371 h.h1->tp_status = status;
372 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
375 h.h2->tp_status = status;
376 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
379 h.h3->tp_status = status;
380 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
383 WARN(1, "TPACKET version not supported.\n");
390 static int __packet_get_status(const struct packet_sock *po, void *frame)
392 union tpacket_uhdr h;
397 switch (po->tp_version) {
399 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
400 return h.h1->tp_status;
402 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
403 return h.h2->tp_status;
405 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
406 return h.h3->tp_status;
408 WARN(1, "TPACKET version not supported.\n");
414 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
417 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
420 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
421 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
422 return TP_STATUS_TS_RAW_HARDWARE;
424 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
425 ktime_to_timespec64_cond(skb->tstamp, ts))
426 return TP_STATUS_TS_SOFTWARE;
431 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
434 union tpacket_uhdr h;
435 struct timespec64 ts;
438 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
443 * versions 1 through 3 overflow the timestamps in y2106, since they
444 * all store the seconds in a 32-bit unsigned integer.
445 * If we create a version 4, that should have a 64-bit timestamp,
446 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
449 switch (po->tp_version) {
451 h.h1->tp_sec = ts.tv_sec;
452 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
455 h.h2->tp_sec = ts.tv_sec;
456 h.h2->tp_nsec = ts.tv_nsec;
459 h.h3->tp_sec = ts.tv_sec;
460 h.h3->tp_nsec = ts.tv_nsec;
463 WARN(1, "TPACKET version not supported.\n");
467 /* one flush is safe, as both fields always lie on the same cacheline */
468 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
474 static void *packet_lookup_frame(const struct packet_sock *po,
475 const struct packet_ring_buffer *rb,
476 unsigned int position,
479 unsigned int pg_vec_pos, frame_offset;
480 union tpacket_uhdr h;
482 pg_vec_pos = position / rb->frames_per_block;
483 frame_offset = position % rb->frames_per_block;
485 h.raw = rb->pg_vec[pg_vec_pos].buffer +
486 (frame_offset * rb->frame_size);
488 if (status != __packet_get_status(po, h.raw))
494 static void *packet_current_frame(struct packet_sock *po,
495 struct packet_ring_buffer *rb,
498 return packet_lookup_frame(po, rb, rb->head, status);
501 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
503 del_timer_sync(&pkc->retire_blk_timer);
506 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
507 struct sk_buff_head *rb_queue)
509 struct tpacket_kbdq_core *pkc;
511 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
513 spin_lock_bh(&rb_queue->lock);
514 pkc->delete_blk_timer = 1;
515 spin_unlock_bh(&rb_queue->lock);
517 prb_del_retire_blk_timer(pkc);
520 static void prb_setup_retire_blk_timer(struct packet_sock *po)
522 struct tpacket_kbdq_core *pkc;
524 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
525 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
527 pkc->retire_blk_timer.expires = jiffies;
530 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
531 int blk_size_in_bytes)
533 struct net_device *dev;
534 unsigned int mbits, div;
535 struct ethtool_link_ksettings ecmd;
539 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
540 if (unlikely(!dev)) {
542 return DEFAULT_PRB_RETIRE_TOV;
544 err = __ethtool_get_link_ksettings(dev, &ecmd);
547 return DEFAULT_PRB_RETIRE_TOV;
549 /* If the link speed is so slow you don't really
550 * need to worry about perf anyways
552 if (ecmd.base.speed < SPEED_1000 ||
553 ecmd.base.speed == SPEED_UNKNOWN)
554 return DEFAULT_PRB_RETIRE_TOV;
556 div = ecmd.base.speed / 1000;
557 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
567 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
568 union tpacket_req_u *req_u)
570 p1->feature_req_word = req_u->req3.tp_feature_req_word;
573 static void init_prb_bdqc(struct packet_sock *po,
574 struct packet_ring_buffer *rb,
576 union tpacket_req_u *req_u)
578 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
579 struct tpacket_block_desc *pbd;
581 memset(p1, 0x0, sizeof(*p1));
583 p1->knxt_seq_num = 1;
585 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
586 p1->pkblk_start = pg_vec[0].buffer;
587 p1->kblk_size = req_u->req3.tp_block_size;
588 p1->knum_blocks = req_u->req3.tp_block_nr;
589 p1->hdrlen = po->tp_hdrlen;
590 p1->version = po->tp_version;
591 p1->last_kactive_blk_num = 0;
592 po->stats.stats3.tp_freeze_q_cnt = 0;
593 if (req_u->req3.tp_retire_blk_tov)
594 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
596 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
597 req_u->req3.tp_block_size);
598 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
599 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
600 rwlock_init(&p1->blk_fill_in_prog_lock);
602 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
603 prb_init_ft_ops(p1, req_u);
604 prb_setup_retire_blk_timer(po);
605 prb_open_block(p1, pbd);
608 /* Do NOT update the last_blk_num first.
609 * Assumes sk_buff_head lock is held.
611 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
613 mod_timer(&pkc->retire_blk_timer,
614 jiffies + pkc->tov_in_jiffies);
615 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
620 * 1) We refresh the timer only when we open a block.
621 * By doing this we don't waste cycles refreshing the timer
622 * on packet-by-packet basis.
624 * With a 1MB block-size, on a 1Gbps line, it will take
625 * i) ~8 ms to fill a block + ii) memcpy etc.
626 * In this cut we are not accounting for the memcpy time.
628 * So, if the user sets the 'tmo' to 10ms then the timer
629 * will never fire while the block is still getting filled
630 * (which is what we want). However, the user could choose
631 * to close a block early and that's fine.
633 * But when the timer does fire, we check whether or not to refresh it.
634 * Since the tmo granularity is in msecs, it is not too expensive
635 * to refresh the timer, lets say every '8' msecs.
636 * Either the user can set the 'tmo' or we can derive it based on
637 * a) line-speed and b) block-size.
638 * prb_calc_retire_blk_tmo() calculates the tmo.
641 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
643 struct packet_sock *po =
644 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
645 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
647 struct tpacket_block_desc *pbd;
649 spin_lock(&po->sk.sk_receive_queue.lock);
651 frozen = prb_queue_frozen(pkc);
652 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
654 if (unlikely(pkc->delete_blk_timer))
657 /* We only need to plug the race when the block is partially filled.
659 * lock(); increment BLOCK_NUM_PKTS; unlock()
660 * copy_bits() is in progress ...
661 * timer fires on other cpu:
662 * we can't retire the current block because copy_bits
666 if (BLOCK_NUM_PKTS(pbd)) {
667 /* Waiting for skb_copy_bits to finish... */
668 write_lock(&pkc->blk_fill_in_prog_lock);
669 write_unlock(&pkc->blk_fill_in_prog_lock);
672 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
674 if (!BLOCK_NUM_PKTS(pbd)) {
675 /* An empty block. Just refresh the timer. */
678 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
679 if (!prb_dispatch_next_block(pkc, po))
684 /* Case 1. Queue was frozen because user-space was
687 if (prb_curr_blk_in_use(pbd)) {
689 * Ok, user-space is still behind.
690 * So just refresh the timer.
694 /* Case 2. queue was frozen,user-space caught up,
695 * now the link went idle && the timer fired.
696 * We don't have a block to close.So we open this
697 * block and restart the timer.
698 * opening a block thaws the queue,restarts timer
699 * Thawing/timer-refresh is a side effect.
701 prb_open_block(pkc, pbd);
708 _prb_refresh_rx_retire_blk_timer(pkc);
711 spin_unlock(&po->sk.sk_receive_queue.lock);
714 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
715 struct tpacket_block_desc *pbd1, __u32 status)
717 /* Flush everything minus the block header */
719 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
724 /* Skip the block header(we know header WILL fit in 4K) */
727 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
728 for (; start < end; start += PAGE_SIZE)
729 flush_dcache_page(pgv_to_page(start));
734 /* Now update the block status. */
736 BLOCK_STATUS(pbd1) = status;
738 /* Flush the block header */
740 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
742 flush_dcache_page(pgv_to_page(start));
752 * 2) Increment active_blk_num
754 * Note:We DONT refresh the timer on purpose.
755 * Because almost always the next block will be opened.
757 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
758 struct tpacket_block_desc *pbd1,
759 struct packet_sock *po, unsigned int stat)
761 __u32 status = TP_STATUS_USER | stat;
763 struct tpacket3_hdr *last_pkt;
764 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
765 struct sock *sk = &po->sk;
767 if (atomic_read(&po->tp_drops))
768 status |= TP_STATUS_LOSING;
770 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
771 last_pkt->tp_next_offset = 0;
773 /* Get the ts of the last pkt */
774 if (BLOCK_NUM_PKTS(pbd1)) {
775 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
776 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
778 /* Ok, we tmo'd - so get the current time.
780 * It shouldn't really happen as we don't close empty
781 * blocks. See prb_retire_rx_blk_timer_expired().
783 struct timespec64 ts;
784 ktime_get_real_ts64(&ts);
785 h1->ts_last_pkt.ts_sec = ts.tv_sec;
786 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
791 /* Flush the block */
792 prb_flush_block(pkc1, pbd1, status);
794 sk->sk_data_ready(sk);
796 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
799 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
801 pkc->reset_pending_on_curr_blk = 0;
805 * Side effect of opening a block:
807 * 1) prb_queue is thawed.
808 * 2) retire_blk_timer is refreshed.
811 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
812 struct tpacket_block_desc *pbd1)
814 struct timespec64 ts;
815 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
819 /* We could have just memset this but we will lose the
820 * flexibility of making the priv area sticky
823 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
824 BLOCK_NUM_PKTS(pbd1) = 0;
825 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
827 ktime_get_real_ts64(&ts);
829 h1->ts_first_pkt.ts_sec = ts.tv_sec;
830 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
832 pkc1->pkblk_start = (char *)pbd1;
833 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
835 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
836 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
838 pbd1->version = pkc1->version;
839 pkc1->prev = pkc1->nxt_offset;
840 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
842 prb_thaw_queue(pkc1);
843 _prb_refresh_rx_retire_blk_timer(pkc1);
849 * Queue freeze logic:
850 * 1) Assume tp_block_nr = 8 blocks.
851 * 2) At time 't0', user opens Rx ring.
852 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
853 * 4) user-space is either sleeping or processing block '0'.
854 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
855 * it will close block-7,loop around and try to fill block '0'.
857 * __packet_lookup_frame_in_block
858 * prb_retire_current_block()
859 * prb_dispatch_next_block()
860 * |->(BLOCK_STATUS == USER) evaluates to true
861 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
862 * 6) Now there are two cases:
863 * 6.1) Link goes idle right after the queue is frozen.
864 * But remember, the last open_block() refreshed the timer.
865 * When this timer expires,it will refresh itself so that we can
866 * re-open block-0 in near future.
867 * 6.2) Link is busy and keeps on receiving packets. This is a simple
868 * case and __packet_lookup_frame_in_block will check if block-0
869 * is free and can now be re-used.
871 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
872 struct packet_sock *po)
874 pkc->reset_pending_on_curr_blk = 1;
875 po->stats.stats3.tp_freeze_q_cnt++;
878 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
881 * If the next block is free then we will dispatch it
882 * and return a good offset.
883 * Else, we will freeze the queue.
884 * So, caller must check the return value.
886 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
887 struct packet_sock *po)
889 struct tpacket_block_desc *pbd;
893 /* 1. Get current block num */
894 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
896 /* 2. If this block is currently in_use then freeze the queue */
897 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
898 prb_freeze_queue(pkc, po);
904 * open this block and return the offset where the first packet
905 * needs to get stored.
907 prb_open_block(pkc, pbd);
908 return (void *)pkc->nxt_offset;
911 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
912 struct packet_sock *po, unsigned int status)
914 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
916 /* retire/close the current block */
917 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
919 * Plug the case where copy_bits() is in progress on
920 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
921 * have space to copy the pkt in the current block and
922 * called prb_retire_current_block()
924 * We don't need to worry about the TMO case because
925 * the timer-handler already handled this case.
927 if (!(status & TP_STATUS_BLK_TMO)) {
928 /* Waiting for skb_copy_bits to finish... */
929 write_lock(&pkc->blk_fill_in_prog_lock);
930 write_unlock(&pkc->blk_fill_in_prog_lock);
932 prb_close_block(pkc, pbd, po, status);
937 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
939 return TP_STATUS_USER & BLOCK_STATUS(pbd);
942 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
944 return pkc->reset_pending_on_curr_blk;
947 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
948 __releases(&pkc->blk_fill_in_prog_lock)
950 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
952 read_unlock(&pkc->blk_fill_in_prog_lock);
955 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
956 struct tpacket3_hdr *ppd)
958 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
961 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
962 struct tpacket3_hdr *ppd)
964 ppd->hv1.tp_rxhash = 0;
967 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
968 struct tpacket3_hdr *ppd)
970 if (skb_vlan_tag_present(pkc->skb)) {
971 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
972 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
973 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
975 ppd->hv1.tp_vlan_tci = 0;
976 ppd->hv1.tp_vlan_tpid = 0;
977 ppd->tp_status = TP_STATUS_AVAILABLE;
981 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
982 struct tpacket3_hdr *ppd)
984 ppd->hv1.tp_padding = 0;
985 prb_fill_vlan_info(pkc, ppd);
987 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
988 prb_fill_rxhash(pkc, ppd);
990 prb_clear_rxhash(pkc, ppd);
993 static void prb_fill_curr_block(char *curr,
994 struct tpacket_kbdq_core *pkc,
995 struct tpacket_block_desc *pbd,
997 __acquires(&pkc->blk_fill_in_prog_lock)
999 struct tpacket3_hdr *ppd;
1001 ppd = (struct tpacket3_hdr *)curr;
1002 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1006 BLOCK_NUM_PKTS(pbd) += 1;
1007 read_lock(&pkc->blk_fill_in_prog_lock);
1008 prb_run_all_ft_ops(pkc, ppd);
1011 /* Assumes caller has the sk->rx_queue.lock */
1012 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1013 struct sk_buff *skb,
1017 struct tpacket_kbdq_core *pkc;
1018 struct tpacket_block_desc *pbd;
1021 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1022 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024 /* Queue is frozen when user space is lagging behind */
1025 if (prb_queue_frozen(pkc)) {
1027 * Check if that last block which caused the queue to freeze,
1028 * is still in_use by user-space.
1030 if (prb_curr_blk_in_use(pbd)) {
1031 /* Can't record this packet */
1035 * Ok, the block was released by user-space.
1036 * Now let's open that block.
1037 * opening a block also thaws the queue.
1038 * Thawing is a side effect.
1040 prb_open_block(pkc, pbd);
1045 curr = pkc->nxt_offset;
1047 end = (char *)pbd + pkc->kblk_size;
1049 /* first try the current block */
1050 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1051 prb_fill_curr_block(curr, pkc, pbd, len);
1052 return (void *)curr;
1055 /* Ok, close the current block */
1056 prb_retire_current_block(pkc, po, 0);
1058 /* Now, try to dispatch the next block */
1059 curr = (char *)prb_dispatch_next_block(pkc, po);
1061 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1062 prb_fill_curr_block(curr, pkc, pbd, len);
1063 return (void *)curr;
1067 * No free blocks are available.user_space hasn't caught up yet.
1068 * Queue was just frozen and now this packet will get dropped.
1073 static void *packet_current_rx_frame(struct packet_sock *po,
1074 struct sk_buff *skb,
1075 int status, unsigned int len)
1078 switch (po->tp_version) {
1081 curr = packet_lookup_frame(po, &po->rx_ring,
1082 po->rx_ring.head, status);
1085 return __packet_lookup_frame_in_block(po, skb, len);
1087 WARN(1, "TPACKET version not supported\n");
1093 static void *prb_lookup_block(const struct packet_sock *po,
1094 const struct packet_ring_buffer *rb,
1098 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1099 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1101 if (status != BLOCK_STATUS(pbd))
1106 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1109 if (rb->prb_bdqc.kactive_blk_num)
1110 prev = rb->prb_bdqc.kactive_blk_num-1;
1112 prev = rb->prb_bdqc.knum_blocks-1;
1116 /* Assumes caller has held the rx_queue.lock */
1117 static void *__prb_previous_block(struct packet_sock *po,
1118 struct packet_ring_buffer *rb,
1121 unsigned int previous = prb_previous_blk_num(rb);
1122 return prb_lookup_block(po, rb, previous, status);
1125 static void *packet_previous_rx_frame(struct packet_sock *po,
1126 struct packet_ring_buffer *rb,
1129 if (po->tp_version <= TPACKET_V2)
1130 return packet_previous_frame(po, rb, status);
1132 return __prb_previous_block(po, rb, status);
1135 static void packet_increment_rx_head(struct packet_sock *po,
1136 struct packet_ring_buffer *rb)
1138 switch (po->tp_version) {
1141 return packet_increment_head(rb);
1144 WARN(1, "TPACKET version not supported.\n");
1150 static void *packet_previous_frame(struct packet_sock *po,
1151 struct packet_ring_buffer *rb,
1154 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1155 return packet_lookup_frame(po, rb, previous, status);
1158 static void packet_increment_head(struct packet_ring_buffer *buff)
1160 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1163 static void packet_inc_pending(struct packet_ring_buffer *rb)
1165 this_cpu_inc(*rb->pending_refcnt);
1168 static void packet_dec_pending(struct packet_ring_buffer *rb)
1170 this_cpu_dec(*rb->pending_refcnt);
1173 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1175 unsigned int refcnt = 0;
1178 /* We don't use pending refcount in rx_ring. */
1179 if (rb->pending_refcnt == NULL)
1182 for_each_possible_cpu(cpu)
1183 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1188 static int packet_alloc_pending(struct packet_sock *po)
1190 po->rx_ring.pending_refcnt = NULL;
1192 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1193 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1199 static void packet_free_pending(struct packet_sock *po)
1201 free_percpu(po->tx_ring.pending_refcnt);
1204 #define ROOM_POW_OFF 2
1205 #define ROOM_NONE 0x0
1206 #define ROOM_LOW 0x1
1207 #define ROOM_NORMAL 0x2
1209 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1213 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1214 idx = READ_ONCE(po->rx_ring.head);
1216 idx += len >> pow_off;
1219 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1222 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1226 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1227 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1229 idx += len >> pow_off;
1232 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1235 static int __packet_rcv_has_room(const struct packet_sock *po,
1236 const struct sk_buff *skb)
1238 const struct sock *sk = &po->sk;
1239 int ret = ROOM_NONE;
1241 if (po->prot_hook.func != tpacket_rcv) {
1242 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1243 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1244 - (skb ? skb->truesize : 0);
1246 if (avail > (rcvbuf >> ROOM_POW_OFF))
1254 if (po->tp_version == TPACKET_V3) {
1255 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1257 else if (__tpacket_v3_has_room(po, 0))
1260 if (__tpacket_has_room(po, ROOM_POW_OFF))
1262 else if (__tpacket_has_room(po, 0))
1269 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1273 ret = __packet_rcv_has_room(po, skb);
1274 pressure = ret != ROOM_NORMAL;
1276 if (READ_ONCE(po->pressure) != pressure)
1277 WRITE_ONCE(po->pressure, pressure);
1282 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1284 if (READ_ONCE(po->pressure) &&
1285 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1286 WRITE_ONCE(po->pressure, 0);
1289 static void packet_sock_destruct(struct sock *sk)
1291 skb_queue_purge(&sk->sk_error_queue);
1293 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1294 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1296 if (!sock_flag(sk, SOCK_DEAD)) {
1297 pr_err("Attempt to release alive packet socket: %p\n", sk);
1301 sk_refcnt_debug_dec(sk);
1304 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1306 u32 *history = po->rollover->history;
1310 rxhash = skb_get_hash(skb);
1311 for (i = 0; i < ROLLOVER_HLEN; i++)
1312 if (READ_ONCE(history[i]) == rxhash)
1315 victim = prandom_u32() % ROLLOVER_HLEN;
1317 /* Avoid dirtying the cache line if possible */
1318 if (READ_ONCE(history[victim]) != rxhash)
1319 WRITE_ONCE(history[victim], rxhash);
1321 return count > (ROLLOVER_HLEN >> 1);
1324 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1325 struct sk_buff *skb,
1328 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1331 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1332 struct sk_buff *skb,
1335 unsigned int val = atomic_inc_return(&f->rr_cur);
1340 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1341 struct sk_buff *skb,
1344 return smp_processor_id() % num;
1347 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1348 struct sk_buff *skb,
1351 return prandom_u32_max(num);
1354 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1355 struct sk_buff *skb,
1356 unsigned int idx, bool try_self,
1359 struct packet_sock *po, *po_next, *po_skip = NULL;
1360 unsigned int i, j, room = ROOM_NONE;
1362 po = pkt_sk(rcu_dereference(f->arr[idx]));
1365 room = packet_rcv_has_room(po, skb);
1366 if (room == ROOM_NORMAL ||
1367 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1372 i = j = min_t(int, po->rollover->sock, num - 1);
1374 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1375 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1376 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1378 po->rollover->sock = i;
1379 atomic_long_inc(&po->rollover->num);
1380 if (room == ROOM_LOW)
1381 atomic_long_inc(&po->rollover->num_huge);
1389 atomic_long_inc(&po->rollover->num_failed);
1393 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1394 struct sk_buff *skb,
1397 return skb_get_queue_mapping(skb) % num;
1400 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1401 struct sk_buff *skb,
1404 struct bpf_prog *prog;
1405 unsigned int ret = 0;
1408 prog = rcu_dereference(f->bpf_prog);
1410 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1416 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1418 return f->flags & (flag >> 8);
1421 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1422 struct packet_type *pt, struct net_device *orig_dev)
1424 struct packet_fanout *f = pt->af_packet_priv;
1425 unsigned int num = READ_ONCE(f->num_members);
1426 struct net *net = read_pnet(&f->net);
1427 struct packet_sock *po;
1430 if (!net_eq(dev_net(dev), net) || !num) {
1435 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1436 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1441 case PACKET_FANOUT_HASH:
1443 idx = fanout_demux_hash(f, skb, num);
1445 case PACKET_FANOUT_LB:
1446 idx = fanout_demux_lb(f, skb, num);
1448 case PACKET_FANOUT_CPU:
1449 idx = fanout_demux_cpu(f, skb, num);
1451 case PACKET_FANOUT_RND:
1452 idx = fanout_demux_rnd(f, skb, num);
1454 case PACKET_FANOUT_QM:
1455 idx = fanout_demux_qm(f, skb, num);
1457 case PACKET_FANOUT_ROLLOVER:
1458 idx = fanout_demux_rollover(f, skb, 0, false, num);
1460 case PACKET_FANOUT_CBPF:
1461 case PACKET_FANOUT_EBPF:
1462 idx = fanout_demux_bpf(f, skb, num);
1466 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1467 idx = fanout_demux_rollover(f, skb, idx, true, num);
1469 po = pkt_sk(rcu_dereference(f->arr[idx]));
1470 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1473 DEFINE_MUTEX(fanout_mutex);
1474 EXPORT_SYMBOL_GPL(fanout_mutex);
1475 static LIST_HEAD(fanout_list);
1476 static u16 fanout_next_id;
1478 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1480 struct packet_fanout *f = po->fanout;
1482 spin_lock(&f->lock);
1483 rcu_assign_pointer(f->arr[f->num_members], sk);
1486 if (f->num_members == 1)
1487 dev_add_pack(&f->prot_hook);
1488 spin_unlock(&f->lock);
1491 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1493 struct packet_fanout *f = po->fanout;
1496 spin_lock(&f->lock);
1497 for (i = 0; i < f->num_members; i++) {
1498 if (rcu_dereference_protected(f->arr[i],
1499 lockdep_is_held(&f->lock)) == sk)
1502 BUG_ON(i >= f->num_members);
1503 rcu_assign_pointer(f->arr[i],
1504 rcu_dereference_protected(f->arr[f->num_members - 1],
1505 lockdep_is_held(&f->lock)));
1507 if (f->num_members == 0)
1508 __dev_remove_pack(&f->prot_hook);
1509 spin_unlock(&f->lock);
1512 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1514 if (sk->sk_family != PF_PACKET)
1517 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1520 static void fanout_init_data(struct packet_fanout *f)
1523 case PACKET_FANOUT_LB:
1524 atomic_set(&f->rr_cur, 0);
1526 case PACKET_FANOUT_CBPF:
1527 case PACKET_FANOUT_EBPF:
1528 RCU_INIT_POINTER(f->bpf_prog, NULL);
1533 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1535 struct bpf_prog *old;
1537 spin_lock(&f->lock);
1538 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1539 rcu_assign_pointer(f->bpf_prog, new);
1540 spin_unlock(&f->lock);
1544 bpf_prog_destroy(old);
1548 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1551 struct bpf_prog *new;
1552 struct sock_fprog fprog;
1555 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1558 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1562 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1566 __fanout_set_data_bpf(po->fanout, new);
1570 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1573 struct bpf_prog *new;
1576 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1578 if (len != sizeof(fd))
1580 if (copy_from_sockptr(&fd, data, len))
1583 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1585 return PTR_ERR(new);
1587 __fanout_set_data_bpf(po->fanout, new);
1591 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1594 switch (po->fanout->type) {
1595 case PACKET_FANOUT_CBPF:
1596 return fanout_set_data_cbpf(po, data, len);
1597 case PACKET_FANOUT_EBPF:
1598 return fanout_set_data_ebpf(po, data, len);
1604 static void fanout_release_data(struct packet_fanout *f)
1607 case PACKET_FANOUT_CBPF:
1608 case PACKET_FANOUT_EBPF:
1609 __fanout_set_data_bpf(f, NULL);
1613 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1615 struct packet_fanout *f;
1617 list_for_each_entry(f, &fanout_list, list) {
1618 if (f->id == candidate_id &&
1619 read_pnet(&f->net) == sock_net(sk)) {
1626 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1628 u16 id = fanout_next_id;
1631 if (__fanout_id_is_free(sk, id)) {
1633 fanout_next_id = id + 1;
1638 } while (id != fanout_next_id);
1643 static int fanout_add(struct sock *sk, struct fanout_args *args)
1645 struct packet_rollover *rollover = NULL;
1646 struct packet_sock *po = pkt_sk(sk);
1647 u16 type_flags = args->type_flags;
1648 struct packet_fanout *f, *match;
1649 u8 type = type_flags & 0xff;
1650 u8 flags = type_flags >> 8;
1655 case PACKET_FANOUT_ROLLOVER:
1656 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1658 case PACKET_FANOUT_HASH:
1659 case PACKET_FANOUT_LB:
1660 case PACKET_FANOUT_CPU:
1661 case PACKET_FANOUT_RND:
1662 case PACKET_FANOUT_QM:
1663 case PACKET_FANOUT_CBPF:
1664 case PACKET_FANOUT_EBPF:
1670 mutex_lock(&fanout_mutex);
1676 if (type == PACKET_FANOUT_ROLLOVER ||
1677 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1679 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1682 atomic_long_set(&rollover->num, 0);
1683 atomic_long_set(&rollover->num_huge, 0);
1684 atomic_long_set(&rollover->num_failed, 0);
1687 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1692 if (!fanout_find_new_id(sk, &id)) {
1696 /* ephemeral flag for the first socket in the group: drop it */
1697 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1701 list_for_each_entry(f, &fanout_list, list) {
1703 read_pnet(&f->net) == sock_net(sk)) {
1710 if (match->flags != flags)
1712 if (args->max_num_members &&
1713 args->max_num_members != match->max_num_members)
1716 if (args->max_num_members > PACKET_FANOUT_MAX)
1718 if (!args->max_num_members)
1719 /* legacy PACKET_FANOUT_MAX */
1720 args->max_num_members = 256;
1722 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1726 write_pnet(&match->net, sock_net(sk));
1729 match->flags = flags;
1730 INIT_LIST_HEAD(&match->list);
1731 spin_lock_init(&match->lock);
1732 refcount_set(&match->sk_ref, 0);
1733 fanout_init_data(match);
1734 match->prot_hook.type = po->prot_hook.type;
1735 match->prot_hook.dev = po->prot_hook.dev;
1736 match->prot_hook.func = packet_rcv_fanout;
1737 match->prot_hook.af_packet_priv = match;
1738 match->prot_hook.af_packet_net = read_pnet(&match->net);
1739 match->prot_hook.id_match = match_fanout_group;
1740 match->max_num_members = args->max_num_members;
1741 list_add(&match->list, &fanout_list);
1745 spin_lock(&po->bind_lock);
1747 match->type == type &&
1748 match->prot_hook.type == po->prot_hook.type &&
1749 match->prot_hook.dev == po->prot_hook.dev) {
1751 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1752 __dev_remove_pack(&po->prot_hook);
1754 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1755 WRITE_ONCE(po->fanout, match);
1757 po->rollover = rollover;
1759 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1760 __fanout_link(sk, po);
1764 spin_unlock(&po->bind_lock);
1766 if (err && !refcount_read(&match->sk_ref)) {
1767 list_del(&match->list);
1773 mutex_unlock(&fanout_mutex);
1777 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1778 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1779 * It is the responsibility of the caller to call fanout_release_data() and
1780 * free the returned packet_fanout (after synchronize_net())
1782 static struct packet_fanout *fanout_release(struct sock *sk)
1784 struct packet_sock *po = pkt_sk(sk);
1785 struct packet_fanout *f;
1787 mutex_lock(&fanout_mutex);
1792 if (refcount_dec_and_test(&f->sk_ref))
1797 mutex_unlock(&fanout_mutex);
1802 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1803 struct sk_buff *skb)
1805 /* Earlier code assumed this would be a VLAN pkt, double-check
1806 * this now that we have the actual packet in hand. We can only
1807 * do this check on Ethernet devices.
1809 if (unlikely(dev->type != ARPHRD_ETHER))
1812 skb_reset_mac_header(skb);
1813 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1816 static const struct proto_ops packet_ops;
1818 static const struct proto_ops packet_ops_spkt;
1820 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1821 struct packet_type *pt, struct net_device *orig_dev)
1824 struct sockaddr_pkt *spkt;
1827 * When we registered the protocol we saved the socket in the data
1828 * field for just this event.
1831 sk = pt->af_packet_priv;
1834 * Yank back the headers [hope the device set this
1835 * right or kerboom...]
1837 * Incoming packets have ll header pulled,
1840 * For outgoing ones skb->data == skb_mac_header(skb)
1841 * so that this procedure is noop.
1844 if (skb->pkt_type == PACKET_LOOPBACK)
1847 if (!net_eq(dev_net(dev), sock_net(sk)))
1850 skb = skb_share_check(skb, GFP_ATOMIC);
1854 /* drop any routing info */
1857 /* drop conntrack reference */
1860 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1862 skb_push(skb, skb->data - skb_mac_header(skb));
1865 * The SOCK_PACKET socket receives _all_ frames.
1868 spkt->spkt_family = dev->type;
1869 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1870 spkt->spkt_protocol = skb->protocol;
1873 * Charge the memory to the socket. This is done specifically
1874 * to prevent sockets using all the memory up.
1877 if (sock_queue_rcv_skb(sk, skb) == 0)
1886 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1888 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1889 sock->type == SOCK_RAW) {
1890 skb_reset_mac_header(skb);
1891 skb->protocol = dev_parse_header_protocol(skb);
1894 skb_probe_transport_header(skb);
1898 * Output a raw packet to a device layer. This bypasses all the other
1899 * protocol layers and you must therefore supply it with a complete frame
1902 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1905 struct sock *sk = sock->sk;
1906 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1907 struct sk_buff *skb = NULL;
1908 struct net_device *dev;
1909 struct sockcm_cookie sockc;
1915 * Get and verify the address.
1919 if (msg->msg_namelen < sizeof(struct sockaddr))
1921 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1922 proto = saddr->spkt_protocol;
1924 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1927 * Find the device first to size check it
1930 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1933 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1939 if (!(dev->flags & IFF_UP))
1943 * You may not queue a frame bigger than the mtu. This is the lowest level
1944 * raw protocol and you must do your own fragmentation at this level.
1947 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1948 if (!netif_supports_nofcs(dev)) {
1949 err = -EPROTONOSUPPORT;
1952 extra_len = 4; /* We're doing our own CRC */
1956 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1960 size_t reserved = LL_RESERVED_SPACE(dev);
1961 int tlen = dev->needed_tailroom;
1962 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1965 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1968 /* FIXME: Save some space for broken drivers that write a hard
1969 * header at transmission time by themselves. PPP is the notable
1970 * one here. This should really be fixed at the driver level.
1972 skb_reserve(skb, reserved);
1973 skb_reset_network_header(skb);
1975 /* Try to align data part correctly */
1980 skb_reset_network_header(skb);
1982 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1988 if (!dev_validate_header(dev, skb->data, len)) {
1992 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1993 !packet_extra_vlan_len_allowed(dev, skb)) {
1998 sockcm_init(&sockc, sk);
1999 if (msg->msg_controllen) {
2000 err = sock_cmsg_send(sk, msg, &sockc);
2005 skb->protocol = proto;
2007 skb->priority = sk->sk_priority;
2008 skb->mark = sk->sk_mark;
2009 skb->tstamp = sockc.transmit_time;
2011 skb_setup_tx_timestamp(skb, sockc.tsflags);
2013 if (unlikely(extra_len == 4))
2016 packet_parse_headers(skb, sock);
2018 dev_queue_xmit(skb);
2029 static unsigned int run_filter(struct sk_buff *skb,
2030 const struct sock *sk,
2033 struct sk_filter *filter;
2036 filter = rcu_dereference(sk->sk_filter);
2038 res = bpf_prog_run_clear_cb(filter->prog, skb);
2044 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2047 struct virtio_net_hdr vnet_hdr;
2049 if (*len < sizeof(vnet_hdr))
2051 *len -= sizeof(vnet_hdr);
2053 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2056 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2060 * This function makes lazy skb cloning in hope that most of packets
2061 * are discarded by BPF.
2063 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2064 * and skb->cb are mangled. It works because (and until) packets
2065 * falling here are owned by current CPU. Output packets are cloned
2066 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2067 * sequencially, so that if we return skb to original state on exit,
2068 * we will not harm anyone.
2071 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2072 struct packet_type *pt, struct net_device *orig_dev)
2075 struct sockaddr_ll *sll;
2076 struct packet_sock *po;
2077 u8 *skb_head = skb->data;
2078 int skb_len = skb->len;
2079 unsigned int snaplen, res;
2080 bool is_drop_n_account = false;
2082 if (skb->pkt_type == PACKET_LOOPBACK)
2085 sk = pt->af_packet_priv;
2088 if (!net_eq(dev_net(dev), sock_net(sk)))
2093 if (dev_has_header(dev)) {
2094 /* The device has an explicit notion of ll header,
2095 * exported to higher levels.
2097 * Otherwise, the device hides details of its frame
2098 * structure, so that corresponding packet head is
2099 * never delivered to user.
2101 if (sk->sk_type != SOCK_DGRAM)
2102 skb_push(skb, skb->data - skb_mac_header(skb));
2103 else if (skb->pkt_type == PACKET_OUTGOING) {
2104 /* Special case: outgoing packets have ll header at head */
2105 skb_pull(skb, skb_network_offset(skb));
2111 res = run_filter(skb, sk, snaplen);
2113 goto drop_n_restore;
2117 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2120 if (skb_shared(skb)) {
2121 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2125 if (skb_head != skb->data) {
2126 skb->data = skb_head;
2133 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2135 sll = &PACKET_SKB_CB(skb)->sa.ll;
2136 sll->sll_hatype = dev->type;
2137 sll->sll_pkttype = skb->pkt_type;
2138 if (unlikely(po->origdev))
2139 sll->sll_ifindex = orig_dev->ifindex;
2141 sll->sll_ifindex = dev->ifindex;
2143 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2145 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2146 * Use their space for storing the original skb length.
2148 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2150 if (pskb_trim(skb, snaplen))
2153 skb_set_owner_r(skb, sk);
2157 /* drop conntrack reference */
2160 spin_lock(&sk->sk_receive_queue.lock);
2161 po->stats.stats1.tp_packets++;
2162 sock_skb_set_dropcount(sk, skb);
2163 __skb_queue_tail(&sk->sk_receive_queue, skb);
2164 spin_unlock(&sk->sk_receive_queue.lock);
2165 sk->sk_data_ready(sk);
2169 is_drop_n_account = true;
2170 atomic_inc(&po->tp_drops);
2171 atomic_inc(&sk->sk_drops);
2174 if (skb_head != skb->data && skb_shared(skb)) {
2175 skb->data = skb_head;
2179 if (!is_drop_n_account)
2186 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2187 struct packet_type *pt, struct net_device *orig_dev)
2190 struct packet_sock *po;
2191 struct sockaddr_ll *sll;
2192 union tpacket_uhdr h;
2193 u8 *skb_head = skb->data;
2194 int skb_len = skb->len;
2195 unsigned int snaplen, res;
2196 unsigned long status = TP_STATUS_USER;
2197 unsigned short macoff, hdrlen;
2198 unsigned int netoff;
2199 struct sk_buff *copy_skb = NULL;
2200 struct timespec64 ts;
2202 bool is_drop_n_account = false;
2203 unsigned int slot_id = 0;
2204 bool do_vnet = false;
2206 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2207 * We may add members to them until current aligned size without forcing
2208 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2210 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2211 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2213 if (skb->pkt_type == PACKET_LOOPBACK)
2216 sk = pt->af_packet_priv;
2219 if (!net_eq(dev_net(dev), sock_net(sk)))
2222 if (dev_has_header(dev)) {
2223 if (sk->sk_type != SOCK_DGRAM)
2224 skb_push(skb, skb->data - skb_mac_header(skb));
2225 else if (skb->pkt_type == PACKET_OUTGOING) {
2226 /* Special case: outgoing packets have ll header at head */
2227 skb_pull(skb, skb_network_offset(skb));
2233 res = run_filter(skb, sk, snaplen);
2235 goto drop_n_restore;
2237 /* If we are flooded, just give up */
2238 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2239 atomic_inc(&po->tp_drops);
2240 goto drop_n_restore;
2243 if (skb->ip_summed == CHECKSUM_PARTIAL)
2244 status |= TP_STATUS_CSUMNOTREADY;
2245 else if (skb->pkt_type != PACKET_OUTGOING &&
2246 (skb->ip_summed == CHECKSUM_COMPLETE ||
2247 skb_csum_unnecessary(skb)))
2248 status |= TP_STATUS_CSUM_VALID;
2253 if (sk->sk_type == SOCK_DGRAM) {
2254 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2257 unsigned int maclen = skb_network_offset(skb);
2258 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2259 (maclen < 16 ? 16 : maclen)) +
2261 if (po->has_vnet_hdr) {
2262 netoff += sizeof(struct virtio_net_hdr);
2265 macoff = netoff - maclen;
2267 if (netoff > USHRT_MAX) {
2268 atomic_inc(&po->tp_drops);
2269 goto drop_n_restore;
2271 if (po->tp_version <= TPACKET_V2) {
2272 if (macoff + snaplen > po->rx_ring.frame_size) {
2273 if (po->copy_thresh &&
2274 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2275 if (skb_shared(skb)) {
2276 copy_skb = skb_clone(skb, GFP_ATOMIC);
2278 copy_skb = skb_get(skb);
2279 skb_head = skb->data;
2282 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2283 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2284 skb_set_owner_r(copy_skb, sk);
2287 snaplen = po->rx_ring.frame_size - macoff;
2288 if ((int)snaplen < 0) {
2293 } else if (unlikely(macoff + snaplen >
2294 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2297 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2298 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2299 snaplen, nval, macoff);
2301 if (unlikely((int)snaplen < 0)) {
2303 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2307 spin_lock(&sk->sk_receive_queue.lock);
2308 h.raw = packet_current_rx_frame(po, skb,
2309 TP_STATUS_KERNEL, (macoff+snaplen));
2311 goto drop_n_account;
2313 if (po->tp_version <= TPACKET_V2) {
2314 slot_id = po->rx_ring.head;
2315 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2316 goto drop_n_account;
2317 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2321 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2322 sizeof(struct virtio_net_hdr),
2323 vio_le(), true, 0)) {
2324 if (po->tp_version == TPACKET_V3)
2325 prb_clear_blk_fill_status(&po->rx_ring);
2326 goto drop_n_account;
2329 if (po->tp_version <= TPACKET_V2) {
2330 packet_increment_rx_head(po, &po->rx_ring);
2332 * LOSING will be reported till you read the stats,
2333 * because it's COR - Clear On Read.
2334 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2337 if (atomic_read(&po->tp_drops))
2338 status |= TP_STATUS_LOSING;
2341 po->stats.stats1.tp_packets++;
2343 status |= TP_STATUS_COPY;
2344 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2346 spin_unlock(&sk->sk_receive_queue.lock);
2348 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2350 /* Always timestamp; prefer an existing software timestamp taken
2351 * closer to the time of capture.
2353 ts_status = tpacket_get_timestamp(skb, &ts,
2354 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2356 ktime_get_real_ts64(&ts);
2358 status |= ts_status;
2360 switch (po->tp_version) {
2362 h.h1->tp_len = skb->len;
2363 h.h1->tp_snaplen = snaplen;
2364 h.h1->tp_mac = macoff;
2365 h.h1->tp_net = netoff;
2366 h.h1->tp_sec = ts.tv_sec;
2367 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2368 hdrlen = sizeof(*h.h1);
2371 h.h2->tp_len = skb->len;
2372 h.h2->tp_snaplen = snaplen;
2373 h.h2->tp_mac = macoff;
2374 h.h2->tp_net = netoff;
2375 h.h2->tp_sec = ts.tv_sec;
2376 h.h2->tp_nsec = ts.tv_nsec;
2377 if (skb_vlan_tag_present(skb)) {
2378 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2379 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2380 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2382 h.h2->tp_vlan_tci = 0;
2383 h.h2->tp_vlan_tpid = 0;
2385 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2386 hdrlen = sizeof(*h.h2);
2389 /* tp_nxt_offset,vlan are already populated above.
2390 * So DONT clear those fields here
2392 h.h3->tp_status |= status;
2393 h.h3->tp_len = skb->len;
2394 h.h3->tp_snaplen = snaplen;
2395 h.h3->tp_mac = macoff;
2396 h.h3->tp_net = netoff;
2397 h.h3->tp_sec = ts.tv_sec;
2398 h.h3->tp_nsec = ts.tv_nsec;
2399 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2400 hdrlen = sizeof(*h.h3);
2406 sll = h.raw + TPACKET_ALIGN(hdrlen);
2407 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2408 sll->sll_family = AF_PACKET;
2409 sll->sll_hatype = dev->type;
2410 sll->sll_protocol = skb->protocol;
2411 sll->sll_pkttype = skb->pkt_type;
2412 if (unlikely(po->origdev))
2413 sll->sll_ifindex = orig_dev->ifindex;
2415 sll->sll_ifindex = dev->ifindex;
2419 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2420 if (po->tp_version <= TPACKET_V2) {
2423 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2426 for (start = h.raw; start < end; start += PAGE_SIZE)
2427 flush_dcache_page(pgv_to_page(start));
2432 if (po->tp_version <= TPACKET_V2) {
2433 spin_lock(&sk->sk_receive_queue.lock);
2434 __packet_set_status(po, h.raw, status);
2435 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2436 spin_unlock(&sk->sk_receive_queue.lock);
2437 sk->sk_data_ready(sk);
2438 } else if (po->tp_version == TPACKET_V3) {
2439 prb_clear_blk_fill_status(&po->rx_ring);
2443 if (skb_head != skb->data && skb_shared(skb)) {
2444 skb->data = skb_head;
2448 if (!is_drop_n_account)
2455 spin_unlock(&sk->sk_receive_queue.lock);
2456 atomic_inc(&po->tp_drops);
2457 is_drop_n_account = true;
2459 sk->sk_data_ready(sk);
2460 kfree_skb(copy_skb);
2461 goto drop_n_restore;
2464 static void tpacket_destruct_skb(struct sk_buff *skb)
2466 struct packet_sock *po = pkt_sk(skb->sk);
2468 if (likely(po->tx_ring.pg_vec)) {
2472 ph = skb_zcopy_get_nouarg(skb);
2473 packet_dec_pending(&po->tx_ring);
2475 ts = __packet_set_timestamp(po, ph, skb);
2476 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2478 if (!packet_read_pending(&po->tx_ring))
2479 complete(&po->skb_completion);
2485 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2487 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2488 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2489 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2490 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2491 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2492 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2493 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2495 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2501 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2502 struct virtio_net_hdr *vnet_hdr)
2504 if (*len < sizeof(*vnet_hdr))
2506 *len -= sizeof(*vnet_hdr);
2508 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2511 return __packet_snd_vnet_parse(vnet_hdr, *len);
2514 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2515 void *frame, struct net_device *dev, void *data, int tp_len,
2516 __be16 proto, unsigned char *addr, int hlen, int copylen,
2517 const struct sockcm_cookie *sockc)
2519 union tpacket_uhdr ph;
2520 int to_write, offset, len, nr_frags, len_max;
2521 struct socket *sock = po->sk.sk_socket;
2527 skb->protocol = proto;
2529 skb->priority = po->sk.sk_priority;
2530 skb->mark = po->sk.sk_mark;
2531 skb->tstamp = sockc->transmit_time;
2532 skb_setup_tx_timestamp(skb, sockc->tsflags);
2533 skb_zcopy_set_nouarg(skb, ph.raw);
2535 skb_reserve(skb, hlen);
2536 skb_reset_network_header(skb);
2540 if (sock->type == SOCK_DGRAM) {
2541 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2543 if (unlikely(err < 0))
2545 } else if (copylen) {
2546 int hdrlen = min_t(int, copylen, tp_len);
2548 skb_push(skb, dev->hard_header_len);
2549 skb_put(skb, copylen - dev->hard_header_len);
2550 err = skb_store_bits(skb, 0, data, hdrlen);
2553 if (!dev_validate_header(dev, skb->data, hdrlen))
2560 offset = offset_in_page(data);
2561 len_max = PAGE_SIZE - offset;
2562 len = ((to_write > len_max) ? len_max : to_write);
2564 skb->data_len = to_write;
2565 skb->len += to_write;
2566 skb->truesize += to_write;
2567 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2569 while (likely(to_write)) {
2570 nr_frags = skb_shinfo(skb)->nr_frags;
2572 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2573 pr_err("Packet exceed the number of skb frags(%lu)\n",
2578 page = pgv_to_page(data);
2580 flush_dcache_page(page);
2582 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2585 len_max = PAGE_SIZE;
2586 len = ((to_write > len_max) ? len_max : to_write);
2589 packet_parse_headers(skb, sock);
2594 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2595 int size_max, void **data)
2597 union tpacket_uhdr ph;
2602 switch (po->tp_version) {
2604 if (ph.h3->tp_next_offset != 0) {
2605 pr_warn_once("variable sized slot not supported");
2608 tp_len = ph.h3->tp_len;
2611 tp_len = ph.h2->tp_len;
2614 tp_len = ph.h1->tp_len;
2617 if (unlikely(tp_len > size_max)) {
2618 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2622 if (unlikely(po->tp_tx_has_off)) {
2623 int off_min, off_max;
2625 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2626 off_max = po->tx_ring.frame_size - tp_len;
2627 if (po->sk.sk_type == SOCK_DGRAM) {
2628 switch (po->tp_version) {
2630 off = ph.h3->tp_net;
2633 off = ph.h2->tp_net;
2636 off = ph.h1->tp_net;
2640 switch (po->tp_version) {
2642 off = ph.h3->tp_mac;
2645 off = ph.h2->tp_mac;
2648 off = ph.h1->tp_mac;
2652 if (unlikely((off < off_min) || (off_max < off)))
2655 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2658 *data = frame + off;
2662 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2664 struct sk_buff *skb = NULL;
2665 struct net_device *dev;
2666 struct virtio_net_hdr *vnet_hdr = NULL;
2667 struct sockcm_cookie sockc;
2669 int err, reserve = 0;
2671 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2672 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2673 unsigned char *addr = NULL;
2674 int tp_len, size_max;
2677 int status = TP_STATUS_AVAILABLE;
2678 int hlen, tlen, copylen = 0;
2681 mutex_lock(&po->pg_vec_lock);
2683 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2684 * we need to confirm it under protection of pg_vec_lock.
2686 if (unlikely(!po->tx_ring.pg_vec)) {
2690 if (likely(saddr == NULL)) {
2691 dev = packet_cached_dev_get(po);
2692 proto = READ_ONCE(po->num);
2695 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2697 if (msg->msg_namelen < (saddr->sll_halen
2698 + offsetof(struct sockaddr_ll,
2701 proto = saddr->sll_protocol;
2702 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2703 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2704 if (dev && msg->msg_namelen < dev->addr_len +
2705 offsetof(struct sockaddr_ll, sll_addr))
2707 addr = saddr->sll_addr;
2712 if (unlikely(dev == NULL))
2715 if (unlikely(!(dev->flags & IFF_UP)))
2718 sockcm_init(&sockc, &po->sk);
2719 if (msg->msg_controllen) {
2720 err = sock_cmsg_send(&po->sk, msg, &sockc);
2725 if (po->sk.sk_socket->type == SOCK_RAW)
2726 reserve = dev->hard_header_len;
2727 size_max = po->tx_ring.frame_size
2728 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2730 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2731 size_max = dev->mtu + reserve + VLAN_HLEN;
2733 reinit_completion(&po->skb_completion);
2736 ph = packet_current_frame(po, &po->tx_ring,
2737 TP_STATUS_SEND_REQUEST);
2738 if (unlikely(ph == NULL)) {
2739 if (need_wait && skb) {
2740 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2741 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2743 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2747 /* check for additional frames */
2752 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2756 status = TP_STATUS_SEND_REQUEST;
2757 hlen = LL_RESERVED_SPACE(dev);
2758 tlen = dev->needed_tailroom;
2759 if (po->has_vnet_hdr) {
2761 data += sizeof(*vnet_hdr);
2762 tp_len -= sizeof(*vnet_hdr);
2764 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2768 copylen = __virtio16_to_cpu(vio_le(),
2771 copylen = max_t(int, copylen, dev->hard_header_len);
2772 skb = sock_alloc_send_skb(&po->sk,
2773 hlen + tlen + sizeof(struct sockaddr_ll) +
2774 (copylen - dev->hard_header_len),
2777 if (unlikely(skb == NULL)) {
2778 /* we assume the socket was initially writeable ... */
2779 if (likely(len_sum > 0))
2783 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2784 addr, hlen, copylen, &sockc);
2785 if (likely(tp_len >= 0) &&
2786 tp_len > dev->mtu + reserve &&
2787 !po->has_vnet_hdr &&
2788 !packet_extra_vlan_len_allowed(dev, skb))
2791 if (unlikely(tp_len < 0)) {
2794 __packet_set_status(po, ph,
2795 TP_STATUS_AVAILABLE);
2796 packet_increment_head(&po->tx_ring);
2800 status = TP_STATUS_WRONG_FORMAT;
2806 if (po->has_vnet_hdr) {
2807 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2811 virtio_net_hdr_set_proto(skb, vnet_hdr);
2814 skb->destructor = tpacket_destruct_skb;
2815 __packet_set_status(po, ph, TP_STATUS_SENDING);
2816 packet_inc_pending(&po->tx_ring);
2818 status = TP_STATUS_SEND_REQUEST;
2819 err = po->xmit(skb);
2820 if (unlikely(err != 0)) {
2822 err = net_xmit_errno(err);
2823 if (err && __packet_get_status(po, ph) ==
2824 TP_STATUS_AVAILABLE) {
2825 /* skb was destructed already */
2830 * skb was dropped but not destructed yet;
2831 * let's treat it like congestion or err < 0
2835 packet_increment_head(&po->tx_ring);
2837 } while (likely((ph != NULL) ||
2838 /* Note: packet_read_pending() might be slow if we have
2839 * to call it as it's per_cpu variable, but in fast-path
2840 * we already short-circuit the loop with the first
2841 * condition, and luckily don't have to go that path
2844 (need_wait && packet_read_pending(&po->tx_ring))));
2850 __packet_set_status(po, ph, status);
2855 mutex_unlock(&po->pg_vec_lock);
2859 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2860 size_t reserve, size_t len,
2861 size_t linear, int noblock,
2864 struct sk_buff *skb;
2866 /* Under a page? Don't bother with paged skb. */
2867 if (prepad + len < PAGE_SIZE || !linear)
2870 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2875 skb_reserve(skb, reserve);
2876 skb_put(skb, linear);
2877 skb->data_len = len - linear;
2878 skb->len += len - linear;
2883 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2885 struct sock *sk = sock->sk;
2886 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2887 struct sk_buff *skb;
2888 struct net_device *dev;
2890 unsigned char *addr = NULL;
2891 int err, reserve = 0;
2892 struct sockcm_cookie sockc;
2893 struct virtio_net_hdr vnet_hdr = { 0 };
2895 struct packet_sock *po = pkt_sk(sk);
2896 bool has_vnet_hdr = false;
2897 int hlen, tlen, linear;
2901 * Get and verify the address.
2904 if (likely(saddr == NULL)) {
2905 dev = packet_cached_dev_get(po);
2906 proto = READ_ONCE(po->num);
2909 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2911 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2913 proto = saddr->sll_protocol;
2914 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2915 if (sock->type == SOCK_DGRAM) {
2916 if (dev && msg->msg_namelen < dev->addr_len +
2917 offsetof(struct sockaddr_ll, sll_addr))
2919 addr = saddr->sll_addr;
2924 if (unlikely(dev == NULL))
2927 if (unlikely(!(dev->flags & IFF_UP)))
2930 sockcm_init(&sockc, sk);
2931 sockc.mark = sk->sk_mark;
2932 if (msg->msg_controllen) {
2933 err = sock_cmsg_send(sk, msg, &sockc);
2938 if (sock->type == SOCK_RAW)
2939 reserve = dev->hard_header_len;
2940 if (po->has_vnet_hdr) {
2941 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2944 has_vnet_hdr = true;
2947 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2948 if (!netif_supports_nofcs(dev)) {
2949 err = -EPROTONOSUPPORT;
2952 extra_len = 4; /* We're doing our own CRC */
2956 if (!vnet_hdr.gso_type &&
2957 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2961 hlen = LL_RESERVED_SPACE(dev);
2962 tlen = dev->needed_tailroom;
2963 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2964 linear = max(linear, min_t(int, len, dev->hard_header_len));
2965 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2966 msg->msg_flags & MSG_DONTWAIT, &err);
2970 skb_reset_network_header(skb);
2973 if (sock->type == SOCK_DGRAM) {
2974 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2975 if (unlikely(offset < 0))
2977 } else if (reserve) {
2978 skb_reserve(skb, -reserve);
2979 if (len < reserve + sizeof(struct ipv6hdr) &&
2980 dev->min_header_len != dev->hard_header_len)
2981 skb_reset_network_header(skb);
2984 /* Returns -EFAULT on error */
2985 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2989 if ((sock->type == SOCK_RAW &&
2990 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
2995 skb_setup_tx_timestamp(skb, sockc.tsflags);
2997 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2998 !packet_extra_vlan_len_allowed(dev, skb)) {
3003 skb->protocol = proto;
3005 skb->priority = sk->sk_priority;
3006 skb->mark = sockc.mark;
3007 skb->tstamp = sockc.transmit_time;
3010 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3013 len += sizeof(vnet_hdr);
3014 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3017 packet_parse_headers(skb, sock);
3019 if (unlikely(extra_len == 4))
3022 err = po->xmit(skb);
3023 if (unlikely(err != 0)) {
3025 err = net_xmit_errno(err);
3043 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3045 struct sock *sk = sock->sk;
3046 struct packet_sock *po = pkt_sk(sk);
3048 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3049 * tpacket_snd() will redo the check safely.
3051 if (data_race(po->tx_ring.pg_vec))
3052 return tpacket_snd(po, msg);
3054 return packet_snd(sock, msg, len);
3058 * Close a PACKET socket. This is fairly simple. We immediately go
3059 * to 'closed' state and remove our protocol entry in the device list.
3062 static int packet_release(struct socket *sock)
3064 struct sock *sk = sock->sk;
3065 struct packet_sock *po;
3066 struct packet_fanout *f;
3068 union tpacket_req_u req_u;
3076 mutex_lock(&net->packet.sklist_lock);
3077 sk_del_node_init_rcu(sk);
3078 mutex_unlock(&net->packet.sklist_lock);
3081 sock_prot_inuse_add(net, sk->sk_prot, -1);
3084 spin_lock(&po->bind_lock);
3085 unregister_prot_hook(sk, false);
3086 packet_cached_dev_reset(po);
3088 if (po->prot_hook.dev) {
3089 dev_put(po->prot_hook.dev);
3090 po->prot_hook.dev = NULL;
3092 spin_unlock(&po->bind_lock);
3094 packet_flush_mclist(sk);
3097 if (po->rx_ring.pg_vec) {
3098 memset(&req_u, 0, sizeof(req_u));
3099 packet_set_ring(sk, &req_u, 1, 0);
3102 if (po->tx_ring.pg_vec) {
3103 memset(&req_u, 0, sizeof(req_u));
3104 packet_set_ring(sk, &req_u, 1, 1);
3108 f = fanout_release(sk);
3112 kfree(po->rollover);
3114 fanout_release_data(f);
3118 * Now the socket is dead. No more input will appear.
3125 skb_queue_purge(&sk->sk_receive_queue);
3126 packet_free_pending(po);
3127 sk_refcnt_debug_release(sk);
3134 * Attach a packet hook.
3137 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3140 struct packet_sock *po = pkt_sk(sk);
3141 struct net_device *dev_curr;
3144 struct net_device *dev = NULL;
3146 bool unlisted = false;
3149 spin_lock(&po->bind_lock);
3158 dev = dev_get_by_name_rcu(sock_net(sk), name);
3163 } else if (ifindex) {
3164 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3174 proto_curr = po->prot_hook.type;
3175 dev_curr = po->prot_hook.dev;
3177 need_rehook = proto_curr != proto || dev_curr != dev;
3182 /* prevents packet_notifier() from calling
3183 * register_prot_hook()
3185 WRITE_ONCE(po->num, 0);
3186 __unregister_prot_hook(sk, true);
3188 dev_curr = po->prot_hook.dev;
3190 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3194 BUG_ON(po->running);
3195 WRITE_ONCE(po->num, proto);
3196 po->prot_hook.type = proto;
3198 if (unlikely(unlisted)) {
3200 po->prot_hook.dev = NULL;
3201 WRITE_ONCE(po->ifindex, -1);
3202 packet_cached_dev_reset(po);
3204 po->prot_hook.dev = dev;
3205 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3206 packet_cached_dev_assign(po, dev);
3212 if (proto == 0 || !need_rehook)
3215 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3216 register_prot_hook(sk);
3218 sk->sk_err = ENETDOWN;
3219 if (!sock_flag(sk, SOCK_DEAD))
3220 sk->sk_error_report(sk);
3225 spin_unlock(&po->bind_lock);
3231 * Bind a packet socket to a device
3234 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3237 struct sock *sk = sock->sk;
3238 char name[sizeof(uaddr->sa_data) + 1];
3244 if (addr_len != sizeof(struct sockaddr))
3246 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3249 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3250 name[sizeof(uaddr->sa_data)] = 0;
3252 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3255 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3257 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3258 struct sock *sk = sock->sk;
3264 if (addr_len < sizeof(struct sockaddr_ll))
3266 if (sll->sll_family != AF_PACKET)
3269 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3270 sll->sll_protocol ? : pkt_sk(sk)->num);
3273 static struct proto packet_proto = {
3275 .owner = THIS_MODULE,
3276 .obj_size = sizeof(struct packet_sock),
3280 * Create a packet of type SOCK_PACKET.
3283 static int packet_create(struct net *net, struct socket *sock, int protocol,
3287 struct packet_sock *po;
3288 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3291 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3293 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3294 sock->type != SOCK_PACKET)
3295 return -ESOCKTNOSUPPORT;
3297 sock->state = SS_UNCONNECTED;
3300 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3304 sock->ops = &packet_ops;
3305 if (sock->type == SOCK_PACKET)
3306 sock->ops = &packet_ops_spkt;
3308 sock_init_data(sock, sk);
3311 init_completion(&po->skb_completion);
3312 sk->sk_family = PF_PACKET;
3314 po->xmit = dev_queue_xmit;
3316 err = packet_alloc_pending(po);
3320 packet_cached_dev_reset(po);
3322 sk->sk_destruct = packet_sock_destruct;
3323 sk_refcnt_debug_inc(sk);
3326 * Attach a protocol block
3329 spin_lock_init(&po->bind_lock);
3330 mutex_init(&po->pg_vec_lock);
3331 po->rollover = NULL;
3332 po->prot_hook.func = packet_rcv;
3334 if (sock->type == SOCK_PACKET)
3335 po->prot_hook.func = packet_rcv_spkt;
3337 po->prot_hook.af_packet_priv = sk;
3338 po->prot_hook.af_packet_net = sock_net(sk);
3341 po->prot_hook.type = proto;
3342 __register_prot_hook(sk);
3345 mutex_lock(&net->packet.sklist_lock);
3346 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3347 mutex_unlock(&net->packet.sklist_lock);
3350 sock_prot_inuse_add(net, &packet_proto, 1);
3361 * Pull a packet from our receive queue and hand it to the user.
3362 * If necessary we block.
3365 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3368 struct sock *sk = sock->sk;
3369 struct sk_buff *skb;
3371 int vnet_hdr_len = 0;
3372 unsigned int origlen = 0;
3375 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3379 /* What error should we return now? EUNATTACH? */
3380 if (pkt_sk(sk)->ifindex < 0)
3384 if (flags & MSG_ERRQUEUE) {
3385 err = sock_recv_errqueue(sk, msg, len,
3386 SOL_PACKET, PACKET_TX_TIMESTAMP);
3391 * Call the generic datagram receiver. This handles all sorts
3392 * of horrible races and re-entrancy so we can forget about it
3393 * in the protocol layers.
3395 * Now it will return ENETDOWN, if device have just gone down,
3396 * but then it will block.
3399 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3402 * An error occurred so return it. Because skb_recv_datagram()
3403 * handles the blocking we don't see and worry about blocking
3410 packet_rcv_try_clear_pressure(pkt_sk(sk));
3412 if (pkt_sk(sk)->has_vnet_hdr) {
3413 err = packet_rcv_vnet(msg, skb, &len);
3416 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3419 /* You lose any data beyond the buffer you gave. If it worries
3420 * a user program they can ask the device for its MTU
3426 msg->msg_flags |= MSG_TRUNC;
3429 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3433 if (sock->type != SOCK_PACKET) {
3434 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3436 /* Original length was stored in sockaddr_ll fields */
3437 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3438 sll->sll_family = AF_PACKET;
3439 sll->sll_protocol = skb->protocol;
3442 sock_recv_ts_and_drops(msg, sk, skb);
3444 if (msg->msg_name) {
3445 const size_t max_len = min(sizeof(skb->cb),
3446 sizeof(struct sockaddr_storage));
3449 /* If the address length field is there to be filled
3450 * in, we fill it in now.
3452 if (sock->type == SOCK_PACKET) {
3453 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3454 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3455 copy_len = msg->msg_namelen;
3457 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3459 msg->msg_namelen = sll->sll_halen +
3460 offsetof(struct sockaddr_ll, sll_addr);
3461 copy_len = msg->msg_namelen;
3462 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3463 memset(msg->msg_name +
3464 offsetof(struct sockaddr_ll, sll_addr),
3465 0, sizeof(sll->sll_addr));
3466 msg->msg_namelen = sizeof(struct sockaddr_ll);
3469 if (WARN_ON_ONCE(copy_len > max_len)) {
3471 msg->msg_namelen = copy_len;
3473 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3476 if (pkt_sk(sk)->auxdata) {
3477 struct tpacket_auxdata aux;
3479 aux.tp_status = TP_STATUS_USER;
3480 if (skb->ip_summed == CHECKSUM_PARTIAL)
3481 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3482 else if (skb->pkt_type != PACKET_OUTGOING &&
3483 (skb->ip_summed == CHECKSUM_COMPLETE ||
3484 skb_csum_unnecessary(skb)))
3485 aux.tp_status |= TP_STATUS_CSUM_VALID;
3487 aux.tp_len = origlen;
3488 aux.tp_snaplen = skb->len;
3490 aux.tp_net = skb_network_offset(skb);
3491 if (skb_vlan_tag_present(skb)) {
3492 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3493 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3494 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3496 aux.tp_vlan_tci = 0;
3497 aux.tp_vlan_tpid = 0;
3499 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3503 * Free or return the buffer as appropriate. Again this
3504 * hides all the races and re-entrancy issues from us.
3506 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3509 skb_free_datagram(sk, skb);
3514 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3517 struct net_device *dev;
3518 struct sock *sk = sock->sk;
3523 uaddr->sa_family = AF_PACKET;
3524 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3526 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3528 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3531 return sizeof(*uaddr);
3534 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3537 struct net_device *dev;
3538 struct sock *sk = sock->sk;
3539 struct packet_sock *po = pkt_sk(sk);
3540 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3546 ifindex = READ_ONCE(po->ifindex);
3547 sll->sll_family = AF_PACKET;
3548 sll->sll_ifindex = ifindex;
3549 sll->sll_protocol = READ_ONCE(po->num);
3550 sll->sll_pkttype = 0;
3552 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3554 sll->sll_hatype = dev->type;
3555 sll->sll_halen = dev->addr_len;
3556 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3558 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3563 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3566 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3570 case PACKET_MR_MULTICAST:
3571 if (i->alen != dev->addr_len)
3574 return dev_mc_add(dev, i->addr);
3576 return dev_mc_del(dev, i->addr);
3578 case PACKET_MR_PROMISC:
3579 return dev_set_promiscuity(dev, what);
3580 case PACKET_MR_ALLMULTI:
3581 return dev_set_allmulti(dev, what);
3582 case PACKET_MR_UNICAST:
3583 if (i->alen != dev->addr_len)
3586 return dev_uc_add(dev, i->addr);
3588 return dev_uc_del(dev, i->addr);
3596 static void packet_dev_mclist_delete(struct net_device *dev,
3597 struct packet_mclist **mlp)
3599 struct packet_mclist *ml;
3601 while ((ml = *mlp) != NULL) {
3602 if (ml->ifindex == dev->ifindex) {
3603 packet_dev_mc(dev, ml, -1);
3611 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3613 struct packet_sock *po = pkt_sk(sk);
3614 struct packet_mclist *ml, *i;
3615 struct net_device *dev;
3621 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3626 if (mreq->mr_alen > dev->addr_len)
3630 i = kmalloc(sizeof(*i), GFP_KERNEL);
3635 for (ml = po->mclist; ml; ml = ml->next) {
3636 if (ml->ifindex == mreq->mr_ifindex &&
3637 ml->type == mreq->mr_type &&
3638 ml->alen == mreq->mr_alen &&
3639 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3641 /* Free the new element ... */
3647 i->type = mreq->mr_type;
3648 i->ifindex = mreq->mr_ifindex;
3649 i->alen = mreq->mr_alen;
3650 memcpy(i->addr, mreq->mr_address, i->alen);
3651 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3653 i->next = po->mclist;
3655 err = packet_dev_mc(dev, i, 1);
3657 po->mclist = i->next;
3666 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3668 struct packet_mclist *ml, **mlp;
3672 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3673 if (ml->ifindex == mreq->mr_ifindex &&
3674 ml->type == mreq->mr_type &&
3675 ml->alen == mreq->mr_alen &&
3676 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3677 if (--ml->count == 0) {
3678 struct net_device *dev;
3680 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3682 packet_dev_mc(dev, ml, -1);
3692 static void packet_flush_mclist(struct sock *sk)
3694 struct packet_sock *po = pkt_sk(sk);
3695 struct packet_mclist *ml;
3701 while ((ml = po->mclist) != NULL) {
3702 struct net_device *dev;
3704 po->mclist = ml->next;
3705 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3707 packet_dev_mc(dev, ml, -1);
3714 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3715 unsigned int optlen)
3717 struct sock *sk = sock->sk;
3718 struct packet_sock *po = pkt_sk(sk);
3721 if (level != SOL_PACKET)
3722 return -ENOPROTOOPT;
3725 case PACKET_ADD_MEMBERSHIP:
3726 case PACKET_DROP_MEMBERSHIP:
3728 struct packet_mreq_max mreq;
3730 memset(&mreq, 0, sizeof(mreq));
3731 if (len < sizeof(struct packet_mreq))
3733 if (len > sizeof(mreq))
3735 if (copy_from_sockptr(&mreq, optval, len))
3737 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3739 if (optname == PACKET_ADD_MEMBERSHIP)
3740 ret = packet_mc_add(sk, &mreq);
3742 ret = packet_mc_drop(sk, &mreq);
3746 case PACKET_RX_RING:
3747 case PACKET_TX_RING:
3749 union tpacket_req_u req_u;
3753 switch (po->tp_version) {
3756 len = sizeof(req_u.req);
3760 len = sizeof(req_u.req3);
3766 if (copy_from_sockptr(&req_u.req, optval, len))
3769 ret = packet_set_ring(sk, &req_u, 0,
3770 optname == PACKET_TX_RING);
3775 case PACKET_COPY_THRESH:
3779 if (optlen != sizeof(val))
3781 if (copy_from_sockptr(&val, optval, sizeof(val)))
3784 pkt_sk(sk)->copy_thresh = val;
3787 case PACKET_VERSION:
3791 if (optlen != sizeof(val))
3793 if (copy_from_sockptr(&val, optval, sizeof(val)))
3804 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3807 po->tp_version = val;
3813 case PACKET_RESERVE:
3817 if (optlen != sizeof(val))
3819 if (copy_from_sockptr(&val, optval, sizeof(val)))
3824 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3827 po->tp_reserve = val;
3837 if (optlen != sizeof(val))
3839 if (copy_from_sockptr(&val, optval, sizeof(val)))
3843 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3846 po->tp_loss = !!val;
3852 case PACKET_AUXDATA:
3856 if (optlen < sizeof(val))
3858 if (copy_from_sockptr(&val, optval, sizeof(val)))
3862 po->auxdata = !!val;
3866 case PACKET_ORIGDEV:
3870 if (optlen < sizeof(val))
3872 if (copy_from_sockptr(&val, optval, sizeof(val)))
3876 po->origdev = !!val;
3880 case PACKET_VNET_HDR:
3884 if (sock->type != SOCK_RAW)
3886 if (optlen < sizeof(val))
3888 if (copy_from_sockptr(&val, optval, sizeof(val)))
3892 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3895 po->has_vnet_hdr = !!val;
3901 case PACKET_TIMESTAMP:
3905 if (optlen != sizeof(val))
3907 if (copy_from_sockptr(&val, optval, sizeof(val)))
3910 po->tp_tstamp = val;
3915 struct fanout_args args = { 0 };
3917 if (optlen != sizeof(int) && optlen != sizeof(args))
3919 if (copy_from_sockptr(&args, optval, optlen))
3922 return fanout_add(sk, &args);
3924 case PACKET_FANOUT_DATA:
3926 /* Paired with the WRITE_ONCE() in fanout_add() */
3927 if (!READ_ONCE(po->fanout))
3930 return fanout_set_data(po, optval, optlen);
3932 case PACKET_IGNORE_OUTGOING:
3936 if (optlen != sizeof(val))
3938 if (copy_from_sockptr(&val, optval, sizeof(val)))
3940 if (val < 0 || val > 1)
3943 po->prot_hook.ignore_outgoing = !!val;
3946 case PACKET_TX_HAS_OFF:
3950 if (optlen != sizeof(val))
3952 if (copy_from_sockptr(&val, optval, sizeof(val)))
3956 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3959 po->tp_tx_has_off = !!val;
3965 case PACKET_QDISC_BYPASS:
3969 if (optlen != sizeof(val))
3971 if (copy_from_sockptr(&val, optval, sizeof(val)))
3974 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3978 return -ENOPROTOOPT;
3982 static int packet_getsockopt(struct socket *sock, int level, int optname,
3983 char __user *optval, int __user *optlen)
3986 int val, lv = sizeof(val);
3987 struct sock *sk = sock->sk;
3988 struct packet_sock *po = pkt_sk(sk);
3990 union tpacket_stats_u st;
3991 struct tpacket_rollover_stats rstats;
3994 if (level != SOL_PACKET)
3995 return -ENOPROTOOPT;
3997 if (get_user(len, optlen))
4004 case PACKET_STATISTICS:
4005 spin_lock_bh(&sk->sk_receive_queue.lock);
4006 memcpy(&st, &po->stats, sizeof(st));
4007 memset(&po->stats, 0, sizeof(po->stats));
4008 spin_unlock_bh(&sk->sk_receive_queue.lock);
4009 drops = atomic_xchg(&po->tp_drops, 0);
4011 if (po->tp_version == TPACKET_V3) {
4012 lv = sizeof(struct tpacket_stats_v3);
4013 st.stats3.tp_drops = drops;
4014 st.stats3.tp_packets += drops;
4017 lv = sizeof(struct tpacket_stats);
4018 st.stats1.tp_drops = drops;
4019 st.stats1.tp_packets += drops;
4024 case PACKET_AUXDATA:
4027 case PACKET_ORIGDEV:
4030 case PACKET_VNET_HDR:
4031 val = po->has_vnet_hdr;
4033 case PACKET_VERSION:
4034 val = po->tp_version;
4037 if (len > sizeof(int))
4039 if (len < sizeof(int))
4041 if (copy_from_user(&val, optval, len))
4045 val = sizeof(struct tpacket_hdr);
4048 val = sizeof(struct tpacket2_hdr);
4051 val = sizeof(struct tpacket3_hdr);
4057 case PACKET_RESERVE:
4058 val = po->tp_reserve;
4063 case PACKET_TIMESTAMP:
4064 val = po->tp_tstamp;
4068 ((u32)po->fanout->id |
4069 ((u32)po->fanout->type << 16) |
4070 ((u32)po->fanout->flags << 24)) :
4073 case PACKET_IGNORE_OUTGOING:
4074 val = po->prot_hook.ignore_outgoing;
4076 case PACKET_ROLLOVER_STATS:
4079 rstats.tp_all = atomic_long_read(&po->rollover->num);
4080 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4081 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4083 lv = sizeof(rstats);
4085 case PACKET_TX_HAS_OFF:
4086 val = po->tp_tx_has_off;
4088 case PACKET_QDISC_BYPASS:
4089 val = packet_use_direct_xmit(po);
4092 return -ENOPROTOOPT;
4097 if (put_user(len, optlen))
4099 if (copy_to_user(optval, data, len))
4104 static int packet_notifier(struct notifier_block *this,
4105 unsigned long msg, void *ptr)
4108 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4109 struct net *net = dev_net(dev);
4112 sk_for_each_rcu(sk, &net->packet.sklist) {
4113 struct packet_sock *po = pkt_sk(sk);
4116 case NETDEV_UNREGISTER:
4118 packet_dev_mclist_delete(dev, &po->mclist);
4122 if (dev->ifindex == po->ifindex) {
4123 spin_lock(&po->bind_lock);
4125 __unregister_prot_hook(sk, false);
4126 sk->sk_err = ENETDOWN;
4127 if (!sock_flag(sk, SOCK_DEAD))
4128 sk->sk_error_report(sk);
4130 if (msg == NETDEV_UNREGISTER) {
4131 packet_cached_dev_reset(po);
4132 WRITE_ONCE(po->ifindex, -1);
4133 if (po->prot_hook.dev)
4134 dev_put(po->prot_hook.dev);
4135 po->prot_hook.dev = NULL;
4137 spin_unlock(&po->bind_lock);
4141 if (dev->ifindex == po->ifindex) {
4142 spin_lock(&po->bind_lock);
4144 register_prot_hook(sk);
4145 spin_unlock(&po->bind_lock);
4155 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4158 struct sock *sk = sock->sk;
4163 int amount = sk_wmem_alloc_get(sk);
4165 return put_user(amount, (int __user *)arg);
4169 struct sk_buff *skb;
4172 spin_lock_bh(&sk->sk_receive_queue.lock);
4173 skb = skb_peek(&sk->sk_receive_queue);
4176 spin_unlock_bh(&sk->sk_receive_queue.lock);
4177 return put_user(amount, (int __user *)arg);
4187 case SIOCGIFBRDADDR:
4188 case SIOCSIFBRDADDR:
4189 case SIOCGIFNETMASK:
4190 case SIOCSIFNETMASK:
4191 case SIOCGIFDSTADDR:
4192 case SIOCSIFDSTADDR:
4194 return inet_dgram_ops.ioctl(sock, cmd, arg);
4198 return -ENOIOCTLCMD;
4203 static __poll_t packet_poll(struct file *file, struct socket *sock,
4206 struct sock *sk = sock->sk;
4207 struct packet_sock *po = pkt_sk(sk);
4208 __poll_t mask = datagram_poll(file, sock, wait);
4210 spin_lock_bh(&sk->sk_receive_queue.lock);
4211 if (po->rx_ring.pg_vec) {
4212 if (!packet_previous_rx_frame(po, &po->rx_ring,
4214 mask |= EPOLLIN | EPOLLRDNORM;
4216 packet_rcv_try_clear_pressure(po);
4217 spin_unlock_bh(&sk->sk_receive_queue.lock);
4218 spin_lock_bh(&sk->sk_write_queue.lock);
4219 if (po->tx_ring.pg_vec) {
4220 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4221 mask |= EPOLLOUT | EPOLLWRNORM;
4223 spin_unlock_bh(&sk->sk_write_queue.lock);
4228 /* Dirty? Well, I still did not learn better way to account
4232 static void packet_mm_open(struct vm_area_struct *vma)
4234 struct file *file = vma->vm_file;
4235 struct socket *sock = file->private_data;
4236 struct sock *sk = sock->sk;
4239 atomic_inc(&pkt_sk(sk)->mapped);
4242 static void packet_mm_close(struct vm_area_struct *vma)
4244 struct file *file = vma->vm_file;
4245 struct socket *sock = file->private_data;
4246 struct sock *sk = sock->sk;
4249 atomic_dec(&pkt_sk(sk)->mapped);
4252 static const struct vm_operations_struct packet_mmap_ops = {
4253 .open = packet_mm_open,
4254 .close = packet_mm_close,
4257 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4262 for (i = 0; i < len; i++) {
4263 if (likely(pg_vec[i].buffer)) {
4264 if (is_vmalloc_addr(pg_vec[i].buffer))
4265 vfree(pg_vec[i].buffer);
4267 free_pages((unsigned long)pg_vec[i].buffer,
4269 pg_vec[i].buffer = NULL;
4275 static char *alloc_one_pg_vec_page(unsigned long order)
4278 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4279 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4281 buffer = (char *) __get_free_pages(gfp_flags, order);
4285 /* __get_free_pages failed, fall back to vmalloc */
4286 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4290 /* vmalloc failed, lets dig into swap here */
4291 gfp_flags &= ~__GFP_NORETRY;
4292 buffer = (char *) __get_free_pages(gfp_flags, order);
4296 /* complete and utter failure */
4300 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4302 unsigned int block_nr = req->tp_block_nr;
4306 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4307 if (unlikely(!pg_vec))
4310 for (i = 0; i < block_nr; i++) {
4311 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4312 if (unlikely(!pg_vec[i].buffer))
4313 goto out_free_pgvec;
4320 free_pg_vec(pg_vec, order, block_nr);
4325 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4326 int closing, int tx_ring)
4328 struct pgv *pg_vec = NULL;
4329 struct packet_sock *po = pkt_sk(sk);
4330 unsigned long *rx_owner_map = NULL;
4331 int was_running, order = 0;
4332 struct packet_ring_buffer *rb;
4333 struct sk_buff_head *rb_queue;
4336 /* Added to avoid minimal code churn */
4337 struct tpacket_req *req = &req_u->req;
4339 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4340 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4344 if (atomic_read(&po->mapped))
4346 if (packet_read_pending(rb))
4350 if (req->tp_block_nr) {
4351 unsigned int min_frame_size;
4353 /* Sanity tests and some calculations */
4355 if (unlikely(rb->pg_vec))
4358 switch (po->tp_version) {
4360 po->tp_hdrlen = TPACKET_HDRLEN;
4363 po->tp_hdrlen = TPACKET2_HDRLEN;
4366 po->tp_hdrlen = TPACKET3_HDRLEN;
4371 if (unlikely((int)req->tp_block_size <= 0))
4373 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4375 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4376 if (po->tp_version >= TPACKET_V3 &&
4377 req->tp_block_size <
4378 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4380 if (unlikely(req->tp_frame_size < min_frame_size))
4382 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4385 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4386 if (unlikely(rb->frames_per_block == 0))
4388 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4390 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4395 order = get_order(req->tp_block_size);
4396 pg_vec = alloc_pg_vec(req, order);
4397 if (unlikely(!pg_vec))
4399 switch (po->tp_version) {
4401 /* Block transmit is not supported yet */
4403 init_prb_bdqc(po, rb, pg_vec, req_u);
4405 struct tpacket_req3 *req3 = &req_u->req3;
4407 if (req3->tp_retire_blk_tov ||
4408 req3->tp_sizeof_priv ||
4409 req3->tp_feature_req_word) {
4411 goto out_free_pg_vec;
4417 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4418 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4420 goto out_free_pg_vec;
4428 if (unlikely(req->tp_frame_nr))
4433 /* Detach socket from network */
4434 spin_lock(&po->bind_lock);
4435 was_running = po->running;
4438 WRITE_ONCE(po->num, 0);
4439 __unregister_prot_hook(sk, false);
4441 spin_unlock(&po->bind_lock);
4446 mutex_lock(&po->pg_vec_lock);
4447 if (closing || atomic_read(&po->mapped) == 0) {
4449 spin_lock_bh(&rb_queue->lock);
4450 swap(rb->pg_vec, pg_vec);
4451 if (po->tp_version <= TPACKET_V2)
4452 swap(rb->rx_owner_map, rx_owner_map);
4453 rb->frame_max = (req->tp_frame_nr - 1);
4455 rb->frame_size = req->tp_frame_size;
4456 spin_unlock_bh(&rb_queue->lock);
4458 swap(rb->pg_vec_order, order);
4459 swap(rb->pg_vec_len, req->tp_block_nr);
4461 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4462 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4463 tpacket_rcv : packet_rcv;
4464 skb_queue_purge(rb_queue);
4465 if (atomic_read(&po->mapped))
4466 pr_err("packet_mmap: vma is busy: %d\n",
4467 atomic_read(&po->mapped));
4469 mutex_unlock(&po->pg_vec_lock);
4471 spin_lock(&po->bind_lock);
4473 WRITE_ONCE(po->num, num);
4474 register_prot_hook(sk);
4476 spin_unlock(&po->bind_lock);
4477 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4478 /* Because we don't support block-based V3 on tx-ring */
4480 prb_shutdown_retire_blk_timer(po, rb_queue);
4485 bitmap_free(rx_owner_map);
4486 free_pg_vec(pg_vec, order, req->tp_block_nr);
4492 static int packet_mmap(struct file *file, struct socket *sock,
4493 struct vm_area_struct *vma)
4495 struct sock *sk = sock->sk;
4496 struct packet_sock *po = pkt_sk(sk);
4497 unsigned long size, expected_size;
4498 struct packet_ring_buffer *rb;
4499 unsigned long start;
4506 mutex_lock(&po->pg_vec_lock);
4509 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4511 expected_size += rb->pg_vec_len
4517 if (expected_size == 0)
4520 size = vma->vm_end - vma->vm_start;
4521 if (size != expected_size)
4524 start = vma->vm_start;
4525 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4526 if (rb->pg_vec == NULL)
4529 for (i = 0; i < rb->pg_vec_len; i++) {
4531 void *kaddr = rb->pg_vec[i].buffer;
4534 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4535 page = pgv_to_page(kaddr);
4536 err = vm_insert_page(vma, start, page);
4545 atomic_inc(&po->mapped);
4546 vma->vm_ops = &packet_mmap_ops;
4550 mutex_unlock(&po->pg_vec_lock);
4554 static const struct proto_ops packet_ops_spkt = {
4555 .family = PF_PACKET,
4556 .owner = THIS_MODULE,
4557 .release = packet_release,
4558 .bind = packet_bind_spkt,
4559 .connect = sock_no_connect,
4560 .socketpair = sock_no_socketpair,
4561 .accept = sock_no_accept,
4562 .getname = packet_getname_spkt,
4563 .poll = datagram_poll,
4564 .ioctl = packet_ioctl,
4565 .gettstamp = sock_gettstamp,
4566 .listen = sock_no_listen,
4567 .shutdown = sock_no_shutdown,
4568 .sendmsg = packet_sendmsg_spkt,
4569 .recvmsg = packet_recvmsg,
4570 .mmap = sock_no_mmap,
4571 .sendpage = sock_no_sendpage,
4574 static const struct proto_ops packet_ops = {
4575 .family = PF_PACKET,
4576 .owner = THIS_MODULE,
4577 .release = packet_release,
4578 .bind = packet_bind,
4579 .connect = sock_no_connect,
4580 .socketpair = sock_no_socketpair,
4581 .accept = sock_no_accept,
4582 .getname = packet_getname,
4583 .poll = packet_poll,
4584 .ioctl = packet_ioctl,
4585 .gettstamp = sock_gettstamp,
4586 .listen = sock_no_listen,
4587 .shutdown = sock_no_shutdown,
4588 .setsockopt = packet_setsockopt,
4589 .getsockopt = packet_getsockopt,
4590 .sendmsg = packet_sendmsg,
4591 .recvmsg = packet_recvmsg,
4592 .mmap = packet_mmap,
4593 .sendpage = sock_no_sendpage,
4596 static const struct net_proto_family packet_family_ops = {
4597 .family = PF_PACKET,
4598 .create = packet_create,
4599 .owner = THIS_MODULE,
4602 static struct notifier_block packet_netdev_notifier = {
4603 .notifier_call = packet_notifier,
4606 #ifdef CONFIG_PROC_FS
4608 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4611 struct net *net = seq_file_net(seq);
4614 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4617 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4619 struct net *net = seq_file_net(seq);
4620 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4623 static void packet_seq_stop(struct seq_file *seq, void *v)
4629 static int packet_seq_show(struct seq_file *seq, void *v)
4631 if (v == SEQ_START_TOKEN)
4632 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4634 struct sock *s = sk_entry(v);
4635 const struct packet_sock *po = pkt_sk(s);
4638 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4640 refcount_read(&s->sk_refcnt),
4642 ntohs(READ_ONCE(po->num)),
4643 READ_ONCE(po->ifindex),
4645 atomic_read(&s->sk_rmem_alloc),
4646 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4653 static const struct seq_operations packet_seq_ops = {
4654 .start = packet_seq_start,
4655 .next = packet_seq_next,
4656 .stop = packet_seq_stop,
4657 .show = packet_seq_show,
4661 static int __net_init packet_net_init(struct net *net)
4663 mutex_init(&net->packet.sklist_lock);
4664 INIT_HLIST_HEAD(&net->packet.sklist);
4666 #ifdef CONFIG_PROC_FS
4667 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4668 sizeof(struct seq_net_private)))
4670 #endif /* CONFIG_PROC_FS */
4675 static void __net_exit packet_net_exit(struct net *net)
4677 remove_proc_entry("packet", net->proc_net);
4678 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4681 static struct pernet_operations packet_net_ops = {
4682 .init = packet_net_init,
4683 .exit = packet_net_exit,
4687 static void __exit packet_exit(void)
4689 unregister_netdevice_notifier(&packet_netdev_notifier);
4690 unregister_pernet_subsys(&packet_net_ops);
4691 sock_unregister(PF_PACKET);
4692 proto_unregister(&packet_proto);
4695 static int __init packet_init(void)
4699 rc = proto_register(&packet_proto, 0);
4702 rc = sock_register(&packet_family_ops);
4705 rc = register_pernet_subsys(&packet_net_ops);
4708 rc = register_netdevice_notifier(&packet_netdev_notifier);
4715 unregister_pernet_subsys(&packet_net_ops);
4717 sock_unregister(PF_PACKET);
4719 proto_unregister(&packet_proto);
4724 module_init(packet_init);
4725 module_exit(packet_exit);
4726 MODULE_LICENSE("GPL");
4727 MODULE_ALIAS_NETPROTO(PF_PACKET);