1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PACKET - implements raw packet sockets.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
51 #include <linux/ethtool.h>
52 #include <linux/filter.h>
53 #include <linux/types.h>
55 #include <linux/capability.h>
56 #include <linux/fcntl.h>
57 #include <linux/socket.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/if_packet.h>
62 #include <linux/wireless.h>
63 #include <linux/kernel.h>
64 #include <linux/kmod.h>
65 #include <linux/slab.h>
66 #include <linux/vmalloc.h>
67 #include <net/net_namespace.h>
69 #include <net/protocol.h>
70 #include <linux/skbuff.h>
72 #include <linux/errno.h>
73 #include <linux/timer.h>
74 #include <linux/uaccess.h>
75 #include <asm/ioctls.h>
77 #include <asm/cacheflush.h>
79 #include <linux/proc_fs.h>
80 #include <linux/seq_file.h>
81 #include <linux/poll.h>
82 #include <linux/module.h>
83 #include <linux/init.h>
84 #include <linux/mutex.h>
85 #include <linux/if_vlan.h>
86 #include <linux/virtio_net.h>
87 #include <linux/errqueue.h>
88 #include <linux/net_tstamp.h>
89 #include <linux/percpu.h>
91 #include <net/inet_common.h>
93 #include <linux/bpf.h>
94 #include <net/compat.h>
95 #include <linux/netfilter_netdev.h>
101 - If the device has no dev->header_ops->create, there is no LL header
102 visible above the device. In this case, its hard_header_len should be 0.
103 The device may prepend its own header internally. In this case, its
104 needed_headroom should be set to the space needed for it to add its
106 For example, a WiFi driver pretending to be an Ethernet driver should
107 set its hard_header_len to be the Ethernet header length, and set its
108 needed_headroom to be (the real WiFi header length - the fake Ethernet
110 - packet socket receives packets with pulled ll header,
111 so that SOCK_RAW should push it back.
116 Incoming, dev_has_header(dev) == true
117 mac_header -> ll header
120 Outgoing, dev_has_header(dev) == true
121 mac_header -> ll header
124 Incoming, dev_has_header(dev) == false
126 However drivers often make it point to the ll header.
127 This is incorrect because the ll header should be invisible to us.
130 Outgoing, dev_has_header(dev) == false
131 mac_header -> data. ll header is invisible to us.
135 If dev_has_header(dev) == false we are unable to restore the ll header,
136 because it is invisible to us.
142 dev_has_header(dev) == true
143 mac_header -> ll header
146 dev_has_header(dev) == false (ll header is invisible to us)
150 We should set network_header on output to the correct position,
151 packet classifier depends on it.
154 /* Private packet socket structures. */
156 /* identical to struct packet_mreq except it has
157 * a longer address field.
159 struct packet_mreq_max {
161 unsigned short mr_type;
162 unsigned short mr_alen;
163 unsigned char mr_address[MAX_ADDR_LEN];
167 struct tpacket_hdr *h1;
168 struct tpacket2_hdr *h2;
169 struct tpacket3_hdr *h3;
173 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
174 int closing, int tx_ring);
176 #define V3_ALIGNMENT (8)
178 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
180 #define BLK_PLUS_PRIV(sz_of_priv) \
181 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
183 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
184 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
185 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
186 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
187 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
188 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
194 static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
209 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
210 struct tpacket3_hdr *);
211 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
212 struct tpacket3_hdr *);
213 static void packet_flush_mclist(struct sock *sk);
214 static u16 packet_pick_tx_queue(struct sk_buff *skb);
216 struct packet_skb_cb {
218 struct sockaddr_pkt pkt;
220 /* Trick: alias skb original length with
221 * ll.sll_family and ll.protocol in order
224 unsigned int origlen;
225 struct sockaddr_ll ll;
230 #define vio_le() virtio_legacy_is_little_endian()
232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246 #ifdef CONFIG_NETFILTER_EGRESS
247 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
249 struct sk_buff *next, *head = NULL, *tail;
253 for (; skb != NULL; skb = next) {
255 skb_mark_not_on_list(skb);
257 if (!nf_hook_egress(skb, &rc, skb->dev))
273 static int packet_xmit(const struct packet_sock *po, struct sk_buff *skb)
275 if (!packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS))
276 return dev_queue_xmit(skb);
278 #ifdef CONFIG_NETFILTER_EGRESS
279 if (nf_hook_egress_active()) {
280 skb = nf_hook_direct_egress(skb);
282 return NET_XMIT_DROP;
285 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
288 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
290 struct net_device *dev;
293 dev = rcu_dereference(po->cached_dev);
300 static void packet_cached_dev_assign(struct packet_sock *po,
301 struct net_device *dev)
303 rcu_assign_pointer(po->cached_dev, dev);
306 static void packet_cached_dev_reset(struct packet_sock *po)
308 RCU_INIT_POINTER(po->cached_dev, NULL);
311 static u16 packet_pick_tx_queue(struct sk_buff *skb)
313 struct net_device *dev = skb->dev;
314 const struct net_device_ops *ops = dev->netdev_ops;
315 int cpu = raw_smp_processor_id();
319 skb->sender_cpu = cpu + 1;
321 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
322 if (ops->ndo_select_queue) {
323 queue_index = ops->ndo_select_queue(dev, skb, NULL);
324 queue_index = netdev_cap_txqueue(dev, queue_index);
326 queue_index = netdev_pick_tx(dev, skb, NULL);
332 /* __register_prot_hook must be invoked through register_prot_hook
333 * or from a context in which asynchronous accesses to the packet
334 * socket is not possible (packet_create()).
336 static void __register_prot_hook(struct sock *sk)
338 struct packet_sock *po = pkt_sk(sk);
340 if (!packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
342 __fanout_link(sk, po);
344 dev_add_pack(&po->prot_hook);
347 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 1);
351 static void register_prot_hook(struct sock *sk)
353 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
354 __register_prot_hook(sk);
357 /* If the sync parameter is true, we will temporarily drop
358 * the po->bind_lock and do a synchronize_net to make sure no
359 * asynchronous packet processing paths still refer to the elements
360 * of po->prot_hook. If the sync parameter is false, it is the
361 * callers responsibility to take care of this.
363 static void __unregister_prot_hook(struct sock *sk, bool sync)
365 struct packet_sock *po = pkt_sk(sk);
367 lockdep_assert_held_once(&po->bind_lock);
369 packet_sock_flag_set(po, PACKET_SOCK_RUNNING, 0);
372 __fanout_unlink(sk, po);
374 __dev_remove_pack(&po->prot_hook);
379 spin_unlock(&po->bind_lock);
381 spin_lock(&po->bind_lock);
385 static void unregister_prot_hook(struct sock *sk, bool sync)
387 struct packet_sock *po = pkt_sk(sk);
389 if (packet_sock_flag(po, PACKET_SOCK_RUNNING))
390 __unregister_prot_hook(sk, sync);
393 static inline struct page * __pure pgv_to_page(void *addr)
395 if (is_vmalloc_addr(addr))
396 return vmalloc_to_page(addr);
397 return virt_to_page(addr);
400 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
402 union tpacket_uhdr h;
405 switch (po->tp_version) {
407 h.h1->tp_status = status;
408 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
411 h.h2->tp_status = status;
412 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
415 h.h3->tp_status = status;
416 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
419 WARN(1, "TPACKET version not supported.\n");
426 static int __packet_get_status(const struct packet_sock *po, void *frame)
428 union tpacket_uhdr h;
433 switch (po->tp_version) {
435 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
436 return h.h1->tp_status;
438 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
439 return h.h2->tp_status;
441 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
442 return h.h3->tp_status;
444 WARN(1, "TPACKET version not supported.\n");
450 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
453 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
456 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
457 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
458 return TP_STATUS_TS_RAW_HARDWARE;
460 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
461 ktime_to_timespec64_cond(skb_tstamp(skb), ts))
462 return TP_STATUS_TS_SOFTWARE;
467 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
470 union tpacket_uhdr h;
471 struct timespec64 ts;
474 if (!(ts_status = tpacket_get_timestamp(skb, &ts, READ_ONCE(po->tp_tstamp))))
479 * versions 1 through 3 overflow the timestamps in y2106, since they
480 * all store the seconds in a 32-bit unsigned integer.
481 * If we create a version 4, that should have a 64-bit timestamp,
482 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
485 switch (po->tp_version) {
487 h.h1->tp_sec = ts.tv_sec;
488 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
491 h.h2->tp_sec = ts.tv_sec;
492 h.h2->tp_nsec = ts.tv_nsec;
495 h.h3->tp_sec = ts.tv_sec;
496 h.h3->tp_nsec = ts.tv_nsec;
499 WARN(1, "TPACKET version not supported.\n");
503 /* one flush is safe, as both fields always lie on the same cacheline */
504 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
510 static void *packet_lookup_frame(const struct packet_sock *po,
511 const struct packet_ring_buffer *rb,
512 unsigned int position,
515 unsigned int pg_vec_pos, frame_offset;
516 union tpacket_uhdr h;
518 pg_vec_pos = position / rb->frames_per_block;
519 frame_offset = position % rb->frames_per_block;
521 h.raw = rb->pg_vec[pg_vec_pos].buffer +
522 (frame_offset * rb->frame_size);
524 if (status != __packet_get_status(po, h.raw))
530 static void *packet_current_frame(struct packet_sock *po,
531 struct packet_ring_buffer *rb,
534 return packet_lookup_frame(po, rb, rb->head, status);
537 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
539 del_timer_sync(&pkc->retire_blk_timer);
542 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
543 struct sk_buff_head *rb_queue)
545 struct tpacket_kbdq_core *pkc;
547 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
549 spin_lock_bh(&rb_queue->lock);
550 pkc->delete_blk_timer = 1;
551 spin_unlock_bh(&rb_queue->lock);
553 prb_del_retire_blk_timer(pkc);
556 static void prb_setup_retire_blk_timer(struct packet_sock *po)
558 struct tpacket_kbdq_core *pkc;
560 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
561 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
563 pkc->retire_blk_timer.expires = jiffies;
566 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
567 int blk_size_in_bytes)
569 struct net_device *dev;
570 unsigned int mbits, div;
571 struct ethtool_link_ksettings ecmd;
575 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
576 if (unlikely(!dev)) {
578 return DEFAULT_PRB_RETIRE_TOV;
580 err = __ethtool_get_link_ksettings(dev, &ecmd);
583 return DEFAULT_PRB_RETIRE_TOV;
585 /* If the link speed is so slow you don't really
586 * need to worry about perf anyways
588 if (ecmd.base.speed < SPEED_1000 ||
589 ecmd.base.speed == SPEED_UNKNOWN)
590 return DEFAULT_PRB_RETIRE_TOV;
592 div = ecmd.base.speed / 1000;
593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
603 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
604 union tpacket_req_u *req_u)
606 p1->feature_req_word = req_u->req3.tp_feature_req_word;
609 static void init_prb_bdqc(struct packet_sock *po,
610 struct packet_ring_buffer *rb,
612 union tpacket_req_u *req_u)
614 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
615 struct tpacket_block_desc *pbd;
617 memset(p1, 0x0, sizeof(*p1));
619 p1->knxt_seq_num = 1;
621 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
622 p1->pkblk_start = pg_vec[0].buffer;
623 p1->kblk_size = req_u->req3.tp_block_size;
624 p1->knum_blocks = req_u->req3.tp_block_nr;
625 p1->hdrlen = po->tp_hdrlen;
626 p1->version = po->tp_version;
627 p1->last_kactive_blk_num = 0;
628 po->stats.stats3.tp_freeze_q_cnt = 0;
629 if (req_u->req3.tp_retire_blk_tov)
630 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
632 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
633 req_u->req3.tp_block_size);
634 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
635 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
636 rwlock_init(&p1->blk_fill_in_prog_lock);
638 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
639 prb_init_ft_ops(p1, req_u);
640 prb_setup_retire_blk_timer(po);
641 prb_open_block(p1, pbd);
644 /* Do NOT update the last_blk_num first.
645 * Assumes sk_buff_head lock is held.
647 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
649 mod_timer(&pkc->retire_blk_timer,
650 jiffies + pkc->tov_in_jiffies);
651 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
656 * 1) We refresh the timer only when we open a block.
657 * By doing this we don't waste cycles refreshing the timer
658 * on packet-by-packet basis.
660 * With a 1MB block-size, on a 1Gbps line, it will take
661 * i) ~8 ms to fill a block + ii) memcpy etc.
662 * In this cut we are not accounting for the memcpy time.
664 * So, if the user sets the 'tmo' to 10ms then the timer
665 * will never fire while the block is still getting filled
666 * (which is what we want). However, the user could choose
667 * to close a block early and that's fine.
669 * But when the timer does fire, we check whether or not to refresh it.
670 * Since the tmo granularity is in msecs, it is not too expensive
671 * to refresh the timer, lets say every '8' msecs.
672 * Either the user can set the 'tmo' or we can derive it based on
673 * a) line-speed and b) block-size.
674 * prb_calc_retire_blk_tmo() calculates the tmo.
677 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
679 struct packet_sock *po =
680 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
681 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
683 struct tpacket_block_desc *pbd;
685 spin_lock(&po->sk.sk_receive_queue.lock);
687 frozen = prb_queue_frozen(pkc);
688 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
690 if (unlikely(pkc->delete_blk_timer))
693 /* We only need to plug the race when the block is partially filled.
695 * lock(); increment BLOCK_NUM_PKTS; unlock()
696 * copy_bits() is in progress ...
697 * timer fires on other cpu:
698 * we can't retire the current block because copy_bits
702 if (BLOCK_NUM_PKTS(pbd)) {
703 /* Waiting for skb_copy_bits to finish... */
704 write_lock(&pkc->blk_fill_in_prog_lock);
705 write_unlock(&pkc->blk_fill_in_prog_lock);
708 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
710 if (!BLOCK_NUM_PKTS(pbd)) {
711 /* An empty block. Just refresh the timer. */
714 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
715 if (!prb_dispatch_next_block(pkc, po))
720 /* Case 1. Queue was frozen because user-space was
723 if (prb_curr_blk_in_use(pbd)) {
725 * Ok, user-space is still behind.
726 * So just refresh the timer.
730 /* Case 2. queue was frozen,user-space caught up,
731 * now the link went idle && the timer fired.
732 * We don't have a block to close.So we open this
733 * block and restart the timer.
734 * opening a block thaws the queue,restarts timer
735 * Thawing/timer-refresh is a side effect.
737 prb_open_block(pkc, pbd);
744 _prb_refresh_rx_retire_blk_timer(pkc);
747 spin_unlock(&po->sk.sk_receive_queue.lock);
750 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
751 struct tpacket_block_desc *pbd1, __u32 status)
753 /* Flush everything minus the block header */
755 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
760 /* Skip the block header(we know header WILL fit in 4K) */
763 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
764 for (; start < end; start += PAGE_SIZE)
765 flush_dcache_page(pgv_to_page(start));
770 /* Now update the block status. */
772 BLOCK_STATUS(pbd1) = status;
774 /* Flush the block header */
776 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
778 flush_dcache_page(pgv_to_page(start));
788 * 2) Increment active_blk_num
790 * Note:We DONT refresh the timer on purpose.
791 * Because almost always the next block will be opened.
793 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
794 struct tpacket_block_desc *pbd1,
795 struct packet_sock *po, unsigned int stat)
797 __u32 status = TP_STATUS_USER | stat;
799 struct tpacket3_hdr *last_pkt;
800 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
801 struct sock *sk = &po->sk;
803 if (atomic_read(&po->tp_drops))
804 status |= TP_STATUS_LOSING;
806 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
807 last_pkt->tp_next_offset = 0;
809 /* Get the ts of the last pkt */
810 if (BLOCK_NUM_PKTS(pbd1)) {
811 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
812 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
814 /* Ok, we tmo'd - so get the current time.
816 * It shouldn't really happen as we don't close empty
817 * blocks. See prb_retire_rx_blk_timer_expired().
819 struct timespec64 ts;
820 ktime_get_real_ts64(&ts);
821 h1->ts_last_pkt.ts_sec = ts.tv_sec;
822 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
827 /* Flush the block */
828 prb_flush_block(pkc1, pbd1, status);
830 sk->sk_data_ready(sk);
832 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
835 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
837 pkc->reset_pending_on_curr_blk = 0;
841 * Side effect of opening a block:
843 * 1) prb_queue is thawed.
844 * 2) retire_blk_timer is refreshed.
847 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
848 struct tpacket_block_desc *pbd1)
850 struct timespec64 ts;
851 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
855 /* We could have just memset this but we will lose the
856 * flexibility of making the priv area sticky
859 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
860 BLOCK_NUM_PKTS(pbd1) = 0;
861 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
863 ktime_get_real_ts64(&ts);
865 h1->ts_first_pkt.ts_sec = ts.tv_sec;
866 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
868 pkc1->pkblk_start = (char *)pbd1;
869 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
871 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
872 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
874 pbd1->version = pkc1->version;
875 pkc1->prev = pkc1->nxt_offset;
876 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
878 prb_thaw_queue(pkc1);
879 _prb_refresh_rx_retire_blk_timer(pkc1);
885 * Queue freeze logic:
886 * 1) Assume tp_block_nr = 8 blocks.
887 * 2) At time 't0', user opens Rx ring.
888 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
889 * 4) user-space is either sleeping or processing block '0'.
890 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
891 * it will close block-7,loop around and try to fill block '0'.
893 * __packet_lookup_frame_in_block
894 * prb_retire_current_block()
895 * prb_dispatch_next_block()
896 * |->(BLOCK_STATUS == USER) evaluates to true
897 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
898 * 6) Now there are two cases:
899 * 6.1) Link goes idle right after the queue is frozen.
900 * But remember, the last open_block() refreshed the timer.
901 * When this timer expires,it will refresh itself so that we can
902 * re-open block-0 in near future.
903 * 6.2) Link is busy and keeps on receiving packets. This is a simple
904 * case and __packet_lookup_frame_in_block will check if block-0
905 * is free and can now be re-used.
907 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
908 struct packet_sock *po)
910 pkc->reset_pending_on_curr_blk = 1;
911 po->stats.stats3.tp_freeze_q_cnt++;
914 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
917 * If the next block is free then we will dispatch it
918 * and return a good offset.
919 * Else, we will freeze the queue.
920 * So, caller must check the return value.
922 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
923 struct packet_sock *po)
925 struct tpacket_block_desc *pbd;
929 /* 1. Get current block num */
930 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
932 /* 2. If this block is currently in_use then freeze the queue */
933 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
934 prb_freeze_queue(pkc, po);
940 * open this block and return the offset where the first packet
941 * needs to get stored.
943 prb_open_block(pkc, pbd);
944 return (void *)pkc->nxt_offset;
947 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
948 struct packet_sock *po, unsigned int status)
950 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
952 /* retire/close the current block */
953 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
955 * Plug the case where copy_bits() is in progress on
956 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
957 * have space to copy the pkt in the current block and
958 * called prb_retire_current_block()
960 * We don't need to worry about the TMO case because
961 * the timer-handler already handled this case.
963 if (!(status & TP_STATUS_BLK_TMO)) {
964 /* Waiting for skb_copy_bits to finish... */
965 write_lock(&pkc->blk_fill_in_prog_lock);
966 write_unlock(&pkc->blk_fill_in_prog_lock);
968 prb_close_block(pkc, pbd, po, status);
973 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
975 return TP_STATUS_USER & BLOCK_STATUS(pbd);
978 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
980 return pkc->reset_pending_on_curr_blk;
983 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
984 __releases(&pkc->blk_fill_in_prog_lock)
986 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
988 read_unlock(&pkc->blk_fill_in_prog_lock);
991 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
992 struct tpacket3_hdr *ppd)
994 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
997 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
998 struct tpacket3_hdr *ppd)
1000 ppd->hv1.tp_rxhash = 0;
1003 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1004 struct tpacket3_hdr *ppd)
1006 if (skb_vlan_tag_present(pkc->skb)) {
1007 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1008 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1009 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1011 ppd->hv1.tp_vlan_tci = 0;
1012 ppd->hv1.tp_vlan_tpid = 0;
1013 ppd->tp_status = TP_STATUS_AVAILABLE;
1017 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1018 struct tpacket3_hdr *ppd)
1020 ppd->hv1.tp_padding = 0;
1021 prb_fill_vlan_info(pkc, ppd);
1023 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1024 prb_fill_rxhash(pkc, ppd);
1026 prb_clear_rxhash(pkc, ppd);
1029 static void prb_fill_curr_block(char *curr,
1030 struct tpacket_kbdq_core *pkc,
1031 struct tpacket_block_desc *pbd,
1033 __acquires(&pkc->blk_fill_in_prog_lock)
1035 struct tpacket3_hdr *ppd;
1037 ppd = (struct tpacket3_hdr *)curr;
1038 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1040 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1042 BLOCK_NUM_PKTS(pbd) += 1;
1043 read_lock(&pkc->blk_fill_in_prog_lock);
1044 prb_run_all_ft_ops(pkc, ppd);
1047 /* Assumes caller has the sk->rx_queue.lock */
1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1049 struct sk_buff *skb,
1053 struct tpacket_kbdq_core *pkc;
1054 struct tpacket_block_desc *pbd;
1057 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1058 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1060 /* Queue is frozen when user space is lagging behind */
1061 if (prb_queue_frozen(pkc)) {
1063 * Check if that last block which caused the queue to freeze,
1064 * is still in_use by user-space.
1066 if (prb_curr_blk_in_use(pbd)) {
1067 /* Can't record this packet */
1071 * Ok, the block was released by user-space.
1072 * Now let's open that block.
1073 * opening a block also thaws the queue.
1074 * Thawing is a side effect.
1076 prb_open_block(pkc, pbd);
1081 curr = pkc->nxt_offset;
1083 end = (char *)pbd + pkc->kblk_size;
1085 /* first try the current block */
1086 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1087 prb_fill_curr_block(curr, pkc, pbd, len);
1088 return (void *)curr;
1091 /* Ok, close the current block */
1092 prb_retire_current_block(pkc, po, 0);
1094 /* Now, try to dispatch the next block */
1095 curr = (char *)prb_dispatch_next_block(pkc, po);
1097 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1098 prb_fill_curr_block(curr, pkc, pbd, len);
1099 return (void *)curr;
1103 * No free blocks are available.user_space hasn't caught up yet.
1104 * Queue was just frozen and now this packet will get dropped.
1109 static void *packet_current_rx_frame(struct packet_sock *po,
1110 struct sk_buff *skb,
1111 int status, unsigned int len)
1114 switch (po->tp_version) {
1117 curr = packet_lookup_frame(po, &po->rx_ring,
1118 po->rx_ring.head, status);
1121 return __packet_lookup_frame_in_block(po, skb, len);
1123 WARN(1, "TPACKET version not supported\n");
1129 static void *prb_lookup_block(const struct packet_sock *po,
1130 const struct packet_ring_buffer *rb,
1134 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1135 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1137 if (status != BLOCK_STATUS(pbd))
1142 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1145 if (rb->prb_bdqc.kactive_blk_num)
1146 prev = rb->prb_bdqc.kactive_blk_num-1;
1148 prev = rb->prb_bdqc.knum_blocks-1;
1152 /* Assumes caller has held the rx_queue.lock */
1153 static void *__prb_previous_block(struct packet_sock *po,
1154 struct packet_ring_buffer *rb,
1157 unsigned int previous = prb_previous_blk_num(rb);
1158 return prb_lookup_block(po, rb, previous, status);
1161 static void *packet_previous_rx_frame(struct packet_sock *po,
1162 struct packet_ring_buffer *rb,
1165 if (po->tp_version <= TPACKET_V2)
1166 return packet_previous_frame(po, rb, status);
1168 return __prb_previous_block(po, rb, status);
1171 static void packet_increment_rx_head(struct packet_sock *po,
1172 struct packet_ring_buffer *rb)
1174 switch (po->tp_version) {
1177 return packet_increment_head(rb);
1180 WARN(1, "TPACKET version not supported.\n");
1186 static void *packet_previous_frame(struct packet_sock *po,
1187 struct packet_ring_buffer *rb,
1190 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1191 return packet_lookup_frame(po, rb, previous, status);
1194 static void packet_increment_head(struct packet_ring_buffer *buff)
1196 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1199 static void packet_inc_pending(struct packet_ring_buffer *rb)
1201 this_cpu_inc(*rb->pending_refcnt);
1204 static void packet_dec_pending(struct packet_ring_buffer *rb)
1206 this_cpu_dec(*rb->pending_refcnt);
1209 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1211 unsigned int refcnt = 0;
1214 /* We don't use pending refcount in rx_ring. */
1215 if (rb->pending_refcnt == NULL)
1218 for_each_possible_cpu(cpu)
1219 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1224 static int packet_alloc_pending(struct packet_sock *po)
1226 po->rx_ring.pending_refcnt = NULL;
1228 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1229 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1235 static void packet_free_pending(struct packet_sock *po)
1237 free_percpu(po->tx_ring.pending_refcnt);
1240 #define ROOM_POW_OFF 2
1241 #define ROOM_NONE 0x0
1242 #define ROOM_LOW 0x1
1243 #define ROOM_NORMAL 0x2
1245 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1249 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1250 idx = READ_ONCE(po->rx_ring.head);
1252 idx += len >> pow_off;
1255 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1258 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1262 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1263 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1265 idx += len >> pow_off;
1268 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1271 static int __packet_rcv_has_room(const struct packet_sock *po,
1272 const struct sk_buff *skb)
1274 const struct sock *sk = &po->sk;
1275 int ret = ROOM_NONE;
1277 if (po->prot_hook.func != tpacket_rcv) {
1278 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1279 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1280 - (skb ? skb->truesize : 0);
1282 if (avail > (rcvbuf >> ROOM_POW_OFF))
1290 if (po->tp_version == TPACKET_V3) {
1291 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1293 else if (__tpacket_v3_has_room(po, 0))
1296 if (__tpacket_has_room(po, ROOM_POW_OFF))
1298 else if (__tpacket_has_room(po, 0))
1305 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1310 ret = __packet_rcv_has_room(po, skb);
1311 pressure = ret != ROOM_NORMAL;
1313 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) != pressure)
1314 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, pressure);
1319 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1321 if (packet_sock_flag(po, PACKET_SOCK_PRESSURE) &&
1322 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1323 packet_sock_flag_set(po, PACKET_SOCK_PRESSURE, false);
1326 static void packet_sock_destruct(struct sock *sk)
1328 skb_queue_purge(&sk->sk_error_queue);
1330 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1331 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1333 if (!sock_flag(sk, SOCK_DEAD)) {
1334 pr_err("Attempt to release alive packet socket: %p\n", sk);
1339 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1341 u32 *history = po->rollover->history;
1345 rxhash = skb_get_hash(skb);
1346 for (i = 0; i < ROLLOVER_HLEN; i++)
1347 if (READ_ONCE(history[i]) == rxhash)
1350 victim = get_random_u32_below(ROLLOVER_HLEN);
1352 /* Avoid dirtying the cache line if possible */
1353 if (READ_ONCE(history[victim]) != rxhash)
1354 WRITE_ONCE(history[victim], rxhash);
1356 return count > (ROLLOVER_HLEN >> 1);
1359 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1360 struct sk_buff *skb,
1363 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1366 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1367 struct sk_buff *skb,
1370 unsigned int val = atomic_inc_return(&f->rr_cur);
1375 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1376 struct sk_buff *skb,
1379 return smp_processor_id() % num;
1382 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1383 struct sk_buff *skb,
1386 return get_random_u32_below(num);
1389 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1390 struct sk_buff *skb,
1391 unsigned int idx, bool try_self,
1394 struct packet_sock *po, *po_next, *po_skip = NULL;
1395 unsigned int i, j, room = ROOM_NONE;
1397 po = pkt_sk(rcu_dereference(f->arr[idx]));
1400 room = packet_rcv_has_room(po, skb);
1401 if (room == ROOM_NORMAL ||
1402 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1407 i = j = min_t(int, po->rollover->sock, num - 1);
1409 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1410 if (po_next != po_skip &&
1411 !packet_sock_flag(po_next, PACKET_SOCK_PRESSURE) &&
1412 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1414 po->rollover->sock = i;
1415 atomic_long_inc(&po->rollover->num);
1416 if (room == ROOM_LOW)
1417 atomic_long_inc(&po->rollover->num_huge);
1425 atomic_long_inc(&po->rollover->num_failed);
1429 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1430 struct sk_buff *skb,
1433 return skb_get_queue_mapping(skb) % num;
1436 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1437 struct sk_buff *skb,
1440 struct bpf_prog *prog;
1441 unsigned int ret = 0;
1444 prog = rcu_dereference(f->bpf_prog);
1446 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1452 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1454 return f->flags & (flag >> 8);
1457 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1458 struct packet_type *pt, struct net_device *orig_dev)
1460 struct packet_fanout *f = pt->af_packet_priv;
1461 unsigned int num = READ_ONCE(f->num_members);
1462 struct net *net = read_pnet(&f->net);
1463 struct packet_sock *po;
1466 if (!net_eq(dev_net(dev), net) || !num) {
1471 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1472 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1477 case PACKET_FANOUT_HASH:
1479 idx = fanout_demux_hash(f, skb, num);
1481 case PACKET_FANOUT_LB:
1482 idx = fanout_demux_lb(f, skb, num);
1484 case PACKET_FANOUT_CPU:
1485 idx = fanout_demux_cpu(f, skb, num);
1487 case PACKET_FANOUT_RND:
1488 idx = fanout_demux_rnd(f, skb, num);
1490 case PACKET_FANOUT_QM:
1491 idx = fanout_demux_qm(f, skb, num);
1493 case PACKET_FANOUT_ROLLOVER:
1494 idx = fanout_demux_rollover(f, skb, 0, false, num);
1496 case PACKET_FANOUT_CBPF:
1497 case PACKET_FANOUT_EBPF:
1498 idx = fanout_demux_bpf(f, skb, num);
1502 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1503 idx = fanout_demux_rollover(f, skb, idx, true, num);
1505 po = pkt_sk(rcu_dereference(f->arr[idx]));
1506 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1509 DEFINE_MUTEX(fanout_mutex);
1510 EXPORT_SYMBOL_GPL(fanout_mutex);
1511 static LIST_HEAD(fanout_list);
1512 static u16 fanout_next_id;
1514 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1516 struct packet_fanout *f = po->fanout;
1518 spin_lock(&f->lock);
1519 rcu_assign_pointer(f->arr[f->num_members], sk);
1522 if (f->num_members == 1)
1523 dev_add_pack(&f->prot_hook);
1524 spin_unlock(&f->lock);
1527 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1529 struct packet_fanout *f = po->fanout;
1532 spin_lock(&f->lock);
1533 for (i = 0; i < f->num_members; i++) {
1534 if (rcu_dereference_protected(f->arr[i],
1535 lockdep_is_held(&f->lock)) == sk)
1538 BUG_ON(i >= f->num_members);
1539 rcu_assign_pointer(f->arr[i],
1540 rcu_dereference_protected(f->arr[f->num_members - 1],
1541 lockdep_is_held(&f->lock)));
1543 if (f->num_members == 0)
1544 __dev_remove_pack(&f->prot_hook);
1545 spin_unlock(&f->lock);
1548 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1550 if (sk->sk_family != PF_PACKET)
1553 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1556 static void fanout_init_data(struct packet_fanout *f)
1559 case PACKET_FANOUT_LB:
1560 atomic_set(&f->rr_cur, 0);
1562 case PACKET_FANOUT_CBPF:
1563 case PACKET_FANOUT_EBPF:
1564 RCU_INIT_POINTER(f->bpf_prog, NULL);
1569 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1571 struct bpf_prog *old;
1573 spin_lock(&f->lock);
1574 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1575 rcu_assign_pointer(f->bpf_prog, new);
1576 spin_unlock(&f->lock);
1580 bpf_prog_destroy(old);
1584 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1587 struct bpf_prog *new;
1588 struct sock_fprog fprog;
1591 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1594 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1598 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1602 __fanout_set_data_bpf(po->fanout, new);
1606 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1609 struct bpf_prog *new;
1612 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1614 if (len != sizeof(fd))
1616 if (copy_from_sockptr(&fd, data, len))
1619 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1621 return PTR_ERR(new);
1623 __fanout_set_data_bpf(po->fanout, new);
1627 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1630 switch (po->fanout->type) {
1631 case PACKET_FANOUT_CBPF:
1632 return fanout_set_data_cbpf(po, data, len);
1633 case PACKET_FANOUT_EBPF:
1634 return fanout_set_data_ebpf(po, data, len);
1640 static void fanout_release_data(struct packet_fanout *f)
1643 case PACKET_FANOUT_CBPF:
1644 case PACKET_FANOUT_EBPF:
1645 __fanout_set_data_bpf(f, NULL);
1649 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1651 struct packet_fanout *f;
1653 list_for_each_entry(f, &fanout_list, list) {
1654 if (f->id == candidate_id &&
1655 read_pnet(&f->net) == sock_net(sk)) {
1662 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1664 u16 id = fanout_next_id;
1667 if (__fanout_id_is_free(sk, id)) {
1669 fanout_next_id = id + 1;
1674 } while (id != fanout_next_id);
1679 static int fanout_add(struct sock *sk, struct fanout_args *args)
1681 struct packet_rollover *rollover = NULL;
1682 struct packet_sock *po = pkt_sk(sk);
1683 u16 type_flags = args->type_flags;
1684 struct packet_fanout *f, *match;
1685 u8 type = type_flags & 0xff;
1686 u8 flags = type_flags >> 8;
1691 case PACKET_FANOUT_ROLLOVER:
1692 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1695 case PACKET_FANOUT_HASH:
1696 case PACKET_FANOUT_LB:
1697 case PACKET_FANOUT_CPU:
1698 case PACKET_FANOUT_RND:
1699 case PACKET_FANOUT_QM:
1700 case PACKET_FANOUT_CBPF:
1701 case PACKET_FANOUT_EBPF:
1707 mutex_lock(&fanout_mutex);
1713 if (type == PACKET_FANOUT_ROLLOVER ||
1714 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1716 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1719 atomic_long_set(&rollover->num, 0);
1720 atomic_long_set(&rollover->num_huge, 0);
1721 atomic_long_set(&rollover->num_failed, 0);
1724 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1729 if (!fanout_find_new_id(sk, &id)) {
1733 /* ephemeral flag for the first socket in the group: drop it */
1734 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1738 list_for_each_entry(f, &fanout_list, list) {
1740 read_pnet(&f->net) == sock_net(sk)) {
1747 if (match->flags != flags)
1749 if (args->max_num_members &&
1750 args->max_num_members != match->max_num_members)
1753 if (args->max_num_members > PACKET_FANOUT_MAX)
1755 if (!args->max_num_members)
1756 /* legacy PACKET_FANOUT_MAX */
1757 args->max_num_members = 256;
1759 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1763 write_pnet(&match->net, sock_net(sk));
1766 match->flags = flags;
1767 INIT_LIST_HEAD(&match->list);
1768 spin_lock_init(&match->lock);
1769 refcount_set(&match->sk_ref, 0);
1770 fanout_init_data(match);
1771 match->prot_hook.type = po->prot_hook.type;
1772 match->prot_hook.dev = po->prot_hook.dev;
1773 match->prot_hook.func = packet_rcv_fanout;
1774 match->prot_hook.af_packet_priv = match;
1775 match->prot_hook.af_packet_net = read_pnet(&match->net);
1776 match->prot_hook.id_match = match_fanout_group;
1777 match->max_num_members = args->max_num_members;
1778 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1779 list_add(&match->list, &fanout_list);
1783 spin_lock(&po->bind_lock);
1784 if (packet_sock_flag(po, PACKET_SOCK_RUNNING) &&
1785 match->type == type &&
1786 match->prot_hook.type == po->prot_hook.type &&
1787 match->prot_hook.dev == po->prot_hook.dev) {
1789 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1790 __dev_remove_pack(&po->prot_hook);
1792 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1793 WRITE_ONCE(po->fanout, match);
1795 po->rollover = rollover;
1797 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1798 __fanout_link(sk, po);
1802 spin_unlock(&po->bind_lock);
1804 if (err && !refcount_read(&match->sk_ref)) {
1805 list_del(&match->list);
1811 mutex_unlock(&fanout_mutex);
1815 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1816 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1817 * It is the responsibility of the caller to call fanout_release_data() and
1818 * free the returned packet_fanout (after synchronize_net())
1820 static struct packet_fanout *fanout_release(struct sock *sk)
1822 struct packet_sock *po = pkt_sk(sk);
1823 struct packet_fanout *f;
1825 mutex_lock(&fanout_mutex);
1830 if (refcount_dec_and_test(&f->sk_ref))
1835 mutex_unlock(&fanout_mutex);
1840 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1841 struct sk_buff *skb)
1843 /* Earlier code assumed this would be a VLAN pkt, double-check
1844 * this now that we have the actual packet in hand. We can only
1845 * do this check on Ethernet devices.
1847 if (unlikely(dev->type != ARPHRD_ETHER))
1850 skb_reset_mac_header(skb);
1851 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1854 static const struct proto_ops packet_ops;
1856 static const struct proto_ops packet_ops_spkt;
1858 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1859 struct packet_type *pt, struct net_device *orig_dev)
1862 struct sockaddr_pkt *spkt;
1865 * When we registered the protocol we saved the socket in the data
1866 * field for just this event.
1869 sk = pt->af_packet_priv;
1872 * Yank back the headers [hope the device set this
1873 * right or kerboom...]
1875 * Incoming packets have ll header pulled,
1878 * For outgoing ones skb->data == skb_mac_header(skb)
1879 * so that this procedure is noop.
1882 if (skb->pkt_type == PACKET_LOOPBACK)
1885 if (!net_eq(dev_net(dev), sock_net(sk)))
1888 skb = skb_share_check(skb, GFP_ATOMIC);
1892 /* drop any routing info */
1895 /* drop conntrack reference */
1898 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1900 skb_push(skb, skb->data - skb_mac_header(skb));
1903 * The SOCK_PACKET socket receives _all_ frames.
1906 spkt->spkt_family = dev->type;
1907 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1908 spkt->spkt_protocol = skb->protocol;
1911 * Charge the memory to the socket. This is done specifically
1912 * to prevent sockets using all the memory up.
1915 if (sock_queue_rcv_skb(sk, skb) == 0)
1924 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1928 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1929 sock->type == SOCK_RAW) {
1930 skb_reset_mac_header(skb);
1931 skb->protocol = dev_parse_header_protocol(skb);
1934 /* Move network header to the right position for VLAN tagged packets */
1935 if (likely(skb->dev->type == ARPHRD_ETHER) &&
1936 eth_type_vlan(skb->protocol) &&
1937 vlan_get_protocol_and_depth(skb, skb->protocol, &depth) != 0)
1938 skb_set_network_header(skb, depth);
1940 skb_probe_transport_header(skb);
1944 * Output a raw packet to a device layer. This bypasses all the other
1945 * protocol layers and you must therefore supply it with a complete frame
1948 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1951 struct sock *sk = sock->sk;
1952 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1953 struct sk_buff *skb = NULL;
1954 struct net_device *dev;
1955 struct sockcm_cookie sockc;
1961 * Get and verify the address.
1965 if (msg->msg_namelen < sizeof(struct sockaddr))
1967 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1968 proto = saddr->spkt_protocol;
1970 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1973 * Find the device first to size check it
1976 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1979 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1985 if (!(dev->flags & IFF_UP))
1989 * You may not queue a frame bigger than the mtu. This is the lowest level
1990 * raw protocol and you must do your own fragmentation at this level.
1993 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1994 if (!netif_supports_nofcs(dev)) {
1995 err = -EPROTONOSUPPORT;
1998 extra_len = 4; /* We're doing our own CRC */
2002 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2006 size_t reserved = LL_RESERVED_SPACE(dev);
2007 int tlen = dev->needed_tailroom;
2008 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2011 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2014 /* FIXME: Save some space for broken drivers that write a hard
2015 * header at transmission time by themselves. PPP is the notable
2016 * one here. This should really be fixed at the driver level.
2018 skb_reserve(skb, reserved);
2019 skb_reset_network_header(skb);
2021 /* Try to align data part correctly */
2026 skb_reset_network_header(skb);
2028 err = memcpy_from_msg(skb_put(skb, len), msg, len);
2034 if (!dev_validate_header(dev, skb->data, len) || !skb->len) {
2038 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2039 !packet_extra_vlan_len_allowed(dev, skb)) {
2044 sockcm_init(&sockc, sk);
2045 if (msg->msg_controllen) {
2046 err = sock_cmsg_send(sk, msg, &sockc);
2051 skb->protocol = proto;
2053 skb->priority = sk->sk_priority;
2054 skb->mark = sk->sk_mark;
2055 skb->tstamp = sockc.transmit_time;
2057 skb_setup_tx_timestamp(skb, sockc.tsflags);
2059 if (unlikely(extra_len == 4))
2062 packet_parse_headers(skb, sock);
2064 dev_queue_xmit(skb);
2075 static unsigned int run_filter(struct sk_buff *skb,
2076 const struct sock *sk,
2079 struct sk_filter *filter;
2082 filter = rcu_dereference(sk->sk_filter);
2084 res = bpf_prog_run_clear_cb(filter->prog, skb);
2090 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2091 size_t *len, int vnet_hdr_sz)
2093 struct virtio_net_hdr_mrg_rxbuf vnet_hdr = { .num_buffers = 0 };
2095 if (*len < vnet_hdr_sz)
2097 *len -= vnet_hdr_sz;
2099 if (virtio_net_hdr_from_skb(skb, (struct virtio_net_hdr *)&vnet_hdr, vio_le(), true, 0))
2102 return memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_sz);
2106 * This function makes lazy skb cloning in hope that most of packets
2107 * are discarded by BPF.
2109 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2110 * and skb->cb are mangled. It works because (and until) packets
2111 * falling here are owned by current CPU. Output packets are cloned
2112 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2113 * sequentially, so that if we return skb to original state on exit,
2114 * we will not harm anyone.
2117 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2118 struct packet_type *pt, struct net_device *orig_dev)
2121 struct sockaddr_ll *sll;
2122 struct packet_sock *po;
2123 u8 *skb_head = skb->data;
2124 int skb_len = skb->len;
2125 unsigned int snaplen, res;
2126 bool is_drop_n_account = false;
2128 if (skb->pkt_type == PACKET_LOOPBACK)
2131 sk = pt->af_packet_priv;
2134 if (!net_eq(dev_net(dev), sock_net(sk)))
2139 if (dev_has_header(dev)) {
2140 /* The device has an explicit notion of ll header,
2141 * exported to higher levels.
2143 * Otherwise, the device hides details of its frame
2144 * structure, so that corresponding packet head is
2145 * never delivered to user.
2147 if (sk->sk_type != SOCK_DGRAM)
2148 skb_push(skb, skb->data - skb_mac_header(skb));
2149 else if (skb->pkt_type == PACKET_OUTGOING) {
2150 /* Special case: outgoing packets have ll header at head */
2151 skb_pull(skb, skb_network_offset(skb));
2157 res = run_filter(skb, sk, snaplen);
2159 goto drop_n_restore;
2163 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2166 if (skb_shared(skb)) {
2167 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2171 if (skb_head != skb->data) {
2172 skb->data = skb_head;
2179 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2181 sll = &PACKET_SKB_CB(skb)->sa.ll;
2182 sll->sll_hatype = dev->type;
2183 sll->sll_pkttype = skb->pkt_type;
2184 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2185 sll->sll_ifindex = orig_dev->ifindex;
2187 sll->sll_ifindex = dev->ifindex;
2189 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2191 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2192 * Use their space for storing the original skb length.
2194 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2196 if (pskb_trim(skb, snaplen))
2199 skb_set_owner_r(skb, sk);
2203 /* drop conntrack reference */
2206 spin_lock(&sk->sk_receive_queue.lock);
2207 po->stats.stats1.tp_packets++;
2208 sock_skb_set_dropcount(sk, skb);
2209 skb_clear_delivery_time(skb);
2210 __skb_queue_tail(&sk->sk_receive_queue, skb);
2211 spin_unlock(&sk->sk_receive_queue.lock);
2212 sk->sk_data_ready(sk);
2216 is_drop_n_account = true;
2217 atomic_inc(&po->tp_drops);
2218 atomic_inc(&sk->sk_drops);
2221 if (skb_head != skb->data && skb_shared(skb)) {
2222 skb->data = skb_head;
2226 if (!is_drop_n_account)
2233 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2234 struct packet_type *pt, struct net_device *orig_dev)
2237 struct packet_sock *po;
2238 struct sockaddr_ll *sll;
2239 union tpacket_uhdr h;
2240 u8 *skb_head = skb->data;
2241 int skb_len = skb->len;
2242 unsigned int snaplen, res;
2243 unsigned long status = TP_STATUS_USER;
2244 unsigned short macoff, hdrlen;
2245 unsigned int netoff;
2246 struct sk_buff *copy_skb = NULL;
2247 struct timespec64 ts;
2249 bool is_drop_n_account = false;
2250 unsigned int slot_id = 0;
2251 int vnet_hdr_sz = 0;
2253 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2254 * We may add members to them until current aligned size without forcing
2255 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2257 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2258 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2260 if (skb->pkt_type == PACKET_LOOPBACK)
2263 sk = pt->af_packet_priv;
2266 if (!net_eq(dev_net(dev), sock_net(sk)))
2269 if (dev_has_header(dev)) {
2270 if (sk->sk_type != SOCK_DGRAM)
2271 skb_push(skb, skb->data - skb_mac_header(skb));
2272 else if (skb->pkt_type == PACKET_OUTGOING) {
2273 /* Special case: outgoing packets have ll header at head */
2274 skb_pull(skb, skb_network_offset(skb));
2280 res = run_filter(skb, sk, snaplen);
2282 goto drop_n_restore;
2284 /* If we are flooded, just give up */
2285 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2286 atomic_inc(&po->tp_drops);
2287 goto drop_n_restore;
2290 if (skb->ip_summed == CHECKSUM_PARTIAL)
2291 status |= TP_STATUS_CSUMNOTREADY;
2292 else if (skb->pkt_type != PACKET_OUTGOING &&
2293 skb_csum_unnecessary(skb))
2294 status |= TP_STATUS_CSUM_VALID;
2295 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2296 status |= TP_STATUS_GSO_TCP;
2301 if (sk->sk_type == SOCK_DGRAM) {
2302 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2305 unsigned int maclen = skb_network_offset(skb);
2306 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2307 (maclen < 16 ? 16 : maclen)) +
2309 vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2311 netoff += vnet_hdr_sz;
2312 macoff = netoff - maclen;
2314 if (netoff > USHRT_MAX) {
2315 atomic_inc(&po->tp_drops);
2316 goto drop_n_restore;
2318 if (po->tp_version <= TPACKET_V2) {
2319 if (macoff + snaplen > po->rx_ring.frame_size) {
2320 if (po->copy_thresh &&
2321 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2322 if (skb_shared(skb)) {
2323 copy_skb = skb_clone(skb, GFP_ATOMIC);
2325 copy_skb = skb_get(skb);
2326 skb_head = skb->data;
2329 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2330 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2331 skb_set_owner_r(copy_skb, sk);
2334 snaplen = po->rx_ring.frame_size - macoff;
2335 if ((int)snaplen < 0) {
2340 } else if (unlikely(macoff + snaplen >
2341 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2344 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2345 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2346 snaplen, nval, macoff);
2348 if (unlikely((int)snaplen < 0)) {
2350 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2354 spin_lock(&sk->sk_receive_queue.lock);
2355 h.raw = packet_current_rx_frame(po, skb,
2356 TP_STATUS_KERNEL, (macoff+snaplen));
2358 goto drop_n_account;
2360 if (po->tp_version <= TPACKET_V2) {
2361 slot_id = po->rx_ring.head;
2362 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2363 goto drop_n_account;
2364 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2368 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2369 sizeof(struct virtio_net_hdr),
2370 vio_le(), true, 0)) {
2371 if (po->tp_version == TPACKET_V3)
2372 prb_clear_blk_fill_status(&po->rx_ring);
2373 goto drop_n_account;
2376 if (po->tp_version <= TPACKET_V2) {
2377 packet_increment_rx_head(po, &po->rx_ring);
2379 * LOSING will be reported till you read the stats,
2380 * because it's COR - Clear On Read.
2381 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2384 if (atomic_read(&po->tp_drops))
2385 status |= TP_STATUS_LOSING;
2388 po->stats.stats1.tp_packets++;
2390 status |= TP_STATUS_COPY;
2391 skb_clear_delivery_time(copy_skb);
2392 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2394 spin_unlock(&sk->sk_receive_queue.lock);
2396 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2398 /* Always timestamp; prefer an existing software timestamp taken
2399 * closer to the time of capture.
2401 ts_status = tpacket_get_timestamp(skb, &ts,
2402 READ_ONCE(po->tp_tstamp) |
2403 SOF_TIMESTAMPING_SOFTWARE);
2405 ktime_get_real_ts64(&ts);
2407 status |= ts_status;
2409 switch (po->tp_version) {
2411 h.h1->tp_len = skb->len;
2412 h.h1->tp_snaplen = snaplen;
2413 h.h1->tp_mac = macoff;
2414 h.h1->tp_net = netoff;
2415 h.h1->tp_sec = ts.tv_sec;
2416 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2417 hdrlen = sizeof(*h.h1);
2420 h.h2->tp_len = skb->len;
2421 h.h2->tp_snaplen = snaplen;
2422 h.h2->tp_mac = macoff;
2423 h.h2->tp_net = netoff;
2424 h.h2->tp_sec = ts.tv_sec;
2425 h.h2->tp_nsec = ts.tv_nsec;
2426 if (skb_vlan_tag_present(skb)) {
2427 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2428 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2429 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2431 h.h2->tp_vlan_tci = 0;
2432 h.h2->tp_vlan_tpid = 0;
2434 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2435 hdrlen = sizeof(*h.h2);
2438 /* tp_nxt_offset,vlan are already populated above.
2439 * So DONT clear those fields here
2441 h.h3->tp_status |= status;
2442 h.h3->tp_len = skb->len;
2443 h.h3->tp_snaplen = snaplen;
2444 h.h3->tp_mac = macoff;
2445 h.h3->tp_net = netoff;
2446 h.h3->tp_sec = ts.tv_sec;
2447 h.h3->tp_nsec = ts.tv_nsec;
2448 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2449 hdrlen = sizeof(*h.h3);
2455 sll = h.raw + TPACKET_ALIGN(hdrlen);
2456 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2457 sll->sll_family = AF_PACKET;
2458 sll->sll_hatype = dev->type;
2459 sll->sll_protocol = skb->protocol;
2460 sll->sll_pkttype = skb->pkt_type;
2461 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2462 sll->sll_ifindex = orig_dev->ifindex;
2464 sll->sll_ifindex = dev->ifindex;
2468 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2469 if (po->tp_version <= TPACKET_V2) {
2472 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2475 for (start = h.raw; start < end; start += PAGE_SIZE)
2476 flush_dcache_page(pgv_to_page(start));
2481 if (po->tp_version <= TPACKET_V2) {
2482 spin_lock(&sk->sk_receive_queue.lock);
2483 __packet_set_status(po, h.raw, status);
2484 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2485 spin_unlock(&sk->sk_receive_queue.lock);
2486 sk->sk_data_ready(sk);
2487 } else if (po->tp_version == TPACKET_V3) {
2488 prb_clear_blk_fill_status(&po->rx_ring);
2492 if (skb_head != skb->data && skb_shared(skb)) {
2493 skb->data = skb_head;
2497 if (!is_drop_n_account)
2504 spin_unlock(&sk->sk_receive_queue.lock);
2505 atomic_inc(&po->tp_drops);
2506 is_drop_n_account = true;
2508 sk->sk_data_ready(sk);
2509 kfree_skb(copy_skb);
2510 goto drop_n_restore;
2513 static void tpacket_destruct_skb(struct sk_buff *skb)
2515 struct packet_sock *po = pkt_sk(skb->sk);
2517 if (likely(po->tx_ring.pg_vec)) {
2521 ph = skb_zcopy_get_nouarg(skb);
2522 packet_dec_pending(&po->tx_ring);
2524 ts = __packet_set_timestamp(po, ph, skb);
2525 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2527 if (!packet_read_pending(&po->tx_ring))
2528 complete(&po->skb_completion);
2534 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2536 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2537 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2538 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2539 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2540 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2541 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2542 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2544 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2550 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2551 struct virtio_net_hdr *vnet_hdr, int vnet_hdr_sz)
2555 if (*len < vnet_hdr_sz)
2557 *len -= vnet_hdr_sz;
2559 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2562 ret = __packet_snd_vnet_parse(vnet_hdr, *len);
2566 /* move iter to point to the start of mac header */
2567 if (vnet_hdr_sz != sizeof(struct virtio_net_hdr))
2568 iov_iter_advance(&msg->msg_iter, vnet_hdr_sz - sizeof(struct virtio_net_hdr));
2573 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2574 void *frame, struct net_device *dev, void *data, int tp_len,
2575 __be16 proto, unsigned char *addr, int hlen, int copylen,
2576 const struct sockcm_cookie *sockc)
2578 union tpacket_uhdr ph;
2579 int to_write, offset, len, nr_frags, len_max;
2580 struct socket *sock = po->sk.sk_socket;
2586 skb->protocol = proto;
2588 skb->priority = po->sk.sk_priority;
2589 skb->mark = po->sk.sk_mark;
2590 skb->tstamp = sockc->transmit_time;
2591 skb_setup_tx_timestamp(skb, sockc->tsflags);
2592 skb_zcopy_set_nouarg(skb, ph.raw);
2594 skb_reserve(skb, hlen);
2595 skb_reset_network_header(skb);
2599 if (sock->type == SOCK_DGRAM) {
2600 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2602 if (unlikely(err < 0))
2604 } else if (copylen) {
2605 int hdrlen = min_t(int, copylen, tp_len);
2607 skb_push(skb, dev->hard_header_len);
2608 skb_put(skb, copylen - dev->hard_header_len);
2609 err = skb_store_bits(skb, 0, data, hdrlen);
2612 if (!dev_validate_header(dev, skb->data, hdrlen))
2619 offset = offset_in_page(data);
2620 len_max = PAGE_SIZE - offset;
2621 len = ((to_write > len_max) ? len_max : to_write);
2623 skb->data_len = to_write;
2624 skb->len += to_write;
2625 skb->truesize += to_write;
2626 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2628 while (likely(to_write)) {
2629 nr_frags = skb_shinfo(skb)->nr_frags;
2631 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2632 pr_err("Packet exceed the number of skb frags(%u)\n",
2633 (unsigned int)MAX_SKB_FRAGS);
2637 page = pgv_to_page(data);
2639 flush_dcache_page(page);
2641 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2644 len_max = PAGE_SIZE;
2645 len = ((to_write > len_max) ? len_max : to_write);
2648 packet_parse_headers(skb, sock);
2653 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2654 int size_max, void **data)
2656 union tpacket_uhdr ph;
2661 switch (po->tp_version) {
2663 if (ph.h3->tp_next_offset != 0) {
2664 pr_warn_once("variable sized slot not supported");
2667 tp_len = ph.h3->tp_len;
2670 tp_len = ph.h2->tp_len;
2673 tp_len = ph.h1->tp_len;
2676 if (unlikely(tp_len > size_max)) {
2677 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2681 if (unlikely(packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF))) {
2682 int off_min, off_max;
2684 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2685 off_max = po->tx_ring.frame_size - tp_len;
2686 if (po->sk.sk_type == SOCK_DGRAM) {
2687 switch (po->tp_version) {
2689 off = ph.h3->tp_net;
2692 off = ph.h2->tp_net;
2695 off = ph.h1->tp_net;
2699 switch (po->tp_version) {
2701 off = ph.h3->tp_mac;
2704 off = ph.h2->tp_mac;
2707 off = ph.h1->tp_mac;
2711 if (unlikely((off < off_min) || (off_max < off)))
2714 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2717 *data = frame + off;
2721 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2723 struct sk_buff *skb = NULL;
2724 struct net_device *dev;
2725 struct virtio_net_hdr *vnet_hdr = NULL;
2726 struct sockcm_cookie sockc;
2728 int err, reserve = 0;
2730 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2731 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2732 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2733 unsigned char *addr = NULL;
2734 int tp_len, size_max;
2737 int status = TP_STATUS_AVAILABLE;
2738 int hlen, tlen, copylen = 0;
2741 mutex_lock(&po->pg_vec_lock);
2743 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2744 * we need to confirm it under protection of pg_vec_lock.
2746 if (unlikely(!po->tx_ring.pg_vec)) {
2750 if (likely(saddr == NULL)) {
2751 dev = packet_cached_dev_get(po);
2752 proto = READ_ONCE(po->num);
2755 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2757 if (msg->msg_namelen < (saddr->sll_halen
2758 + offsetof(struct sockaddr_ll,
2761 proto = saddr->sll_protocol;
2762 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2763 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2764 if (dev && msg->msg_namelen < dev->addr_len +
2765 offsetof(struct sockaddr_ll, sll_addr))
2767 addr = saddr->sll_addr;
2772 if (unlikely(dev == NULL))
2775 if (unlikely(!(dev->flags & IFF_UP)))
2778 sockcm_init(&sockc, &po->sk);
2779 if (msg->msg_controllen) {
2780 err = sock_cmsg_send(&po->sk, msg, &sockc);
2785 if (po->sk.sk_socket->type == SOCK_RAW)
2786 reserve = dev->hard_header_len;
2787 size_max = po->tx_ring.frame_size
2788 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2790 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !vnet_hdr_sz)
2791 size_max = dev->mtu + reserve + VLAN_HLEN;
2793 reinit_completion(&po->skb_completion);
2796 ph = packet_current_frame(po, &po->tx_ring,
2797 TP_STATUS_SEND_REQUEST);
2798 if (unlikely(ph == NULL)) {
2799 if (need_wait && skb) {
2800 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2801 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2803 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2807 /* check for additional frames */
2812 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2816 status = TP_STATUS_SEND_REQUEST;
2817 hlen = LL_RESERVED_SPACE(dev);
2818 tlen = dev->needed_tailroom;
2821 data += vnet_hdr_sz;
2822 tp_len -= vnet_hdr_sz;
2824 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2828 copylen = __virtio16_to_cpu(vio_le(),
2831 copylen = max_t(int, copylen, dev->hard_header_len);
2832 skb = sock_alloc_send_skb(&po->sk,
2833 hlen + tlen + sizeof(struct sockaddr_ll) +
2834 (copylen - dev->hard_header_len),
2837 if (unlikely(skb == NULL)) {
2838 /* we assume the socket was initially writeable ... */
2839 if (likely(len_sum > 0))
2843 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2844 addr, hlen, copylen, &sockc);
2845 if (likely(tp_len >= 0) &&
2846 tp_len > dev->mtu + reserve &&
2848 !packet_extra_vlan_len_allowed(dev, skb))
2851 if (unlikely(tp_len < 0)) {
2853 if (packet_sock_flag(po, PACKET_SOCK_TP_LOSS)) {
2854 __packet_set_status(po, ph,
2855 TP_STATUS_AVAILABLE);
2856 packet_increment_head(&po->tx_ring);
2860 status = TP_STATUS_WRONG_FORMAT;
2867 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2871 virtio_net_hdr_set_proto(skb, vnet_hdr);
2874 skb->destructor = tpacket_destruct_skb;
2875 __packet_set_status(po, ph, TP_STATUS_SENDING);
2876 packet_inc_pending(&po->tx_ring);
2878 status = TP_STATUS_SEND_REQUEST;
2879 err = packet_xmit(po, skb);
2880 if (unlikely(err != 0)) {
2882 err = net_xmit_errno(err);
2883 if (err && __packet_get_status(po, ph) ==
2884 TP_STATUS_AVAILABLE) {
2885 /* skb was destructed already */
2890 * skb was dropped but not destructed yet;
2891 * let's treat it like congestion or err < 0
2895 packet_increment_head(&po->tx_ring);
2897 } while (likely((ph != NULL) ||
2898 /* Note: packet_read_pending() might be slow if we have
2899 * to call it as it's per_cpu variable, but in fast-path
2900 * we already short-circuit the loop with the first
2901 * condition, and luckily don't have to go that path
2904 (need_wait && packet_read_pending(&po->tx_ring))));
2910 __packet_set_status(po, ph, status);
2915 mutex_unlock(&po->pg_vec_lock);
2919 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2920 size_t reserve, size_t len,
2921 size_t linear, int noblock,
2924 struct sk_buff *skb;
2926 /* Under a page? Don't bother with paged skb. */
2927 if (prepad + len < PAGE_SIZE || !linear)
2930 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2935 skb_reserve(skb, reserve);
2936 skb_put(skb, linear);
2937 skb->data_len = len - linear;
2938 skb->len += len - linear;
2943 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2945 struct sock *sk = sock->sk;
2946 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2947 struct sk_buff *skb;
2948 struct net_device *dev;
2950 unsigned char *addr = NULL;
2951 int err, reserve = 0;
2952 struct sockcm_cookie sockc;
2953 struct virtio_net_hdr vnet_hdr = { 0 };
2955 struct packet_sock *po = pkt_sk(sk);
2956 int vnet_hdr_sz = READ_ONCE(po->vnet_hdr_sz);
2957 int hlen, tlen, linear;
2961 * Get and verify the address.
2964 if (likely(saddr == NULL)) {
2965 dev = packet_cached_dev_get(po);
2966 proto = READ_ONCE(po->num);
2969 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2971 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2973 proto = saddr->sll_protocol;
2974 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2975 if (sock->type == SOCK_DGRAM) {
2976 if (dev && msg->msg_namelen < dev->addr_len +
2977 offsetof(struct sockaddr_ll, sll_addr))
2979 addr = saddr->sll_addr;
2984 if (unlikely(dev == NULL))
2987 if (unlikely(!(dev->flags & IFF_UP)))
2990 sockcm_init(&sockc, sk);
2991 sockc.mark = sk->sk_mark;
2992 if (msg->msg_controllen) {
2993 err = sock_cmsg_send(sk, msg, &sockc);
2998 if (sock->type == SOCK_RAW)
2999 reserve = dev->hard_header_len;
3001 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr, vnet_hdr_sz);
3006 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3007 if (!netif_supports_nofcs(dev)) {
3008 err = -EPROTONOSUPPORT;
3011 extra_len = 4; /* We're doing our own CRC */
3015 if (!vnet_hdr.gso_type &&
3016 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3020 hlen = LL_RESERVED_SPACE(dev);
3021 tlen = dev->needed_tailroom;
3022 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3023 linear = max(linear, min_t(int, len, dev->hard_header_len));
3024 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3025 msg->msg_flags & MSG_DONTWAIT, &err);
3029 skb_reset_network_header(skb);
3032 if (sock->type == SOCK_DGRAM) {
3033 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3034 if (unlikely(offset < 0))
3036 } else if (reserve) {
3037 skb_reserve(skb, -reserve);
3038 if (len < reserve + sizeof(struct ipv6hdr) &&
3039 dev->min_header_len != dev->hard_header_len)
3040 skb_reset_network_header(skb);
3043 /* Returns -EFAULT on error */
3044 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3048 if ((sock->type == SOCK_RAW &&
3049 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3054 skb_setup_tx_timestamp(skb, sockc.tsflags);
3056 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3057 !packet_extra_vlan_len_allowed(dev, skb)) {
3062 skb->protocol = proto;
3064 skb->priority = sk->sk_priority;
3065 skb->mark = sockc.mark;
3066 skb->tstamp = sockc.transmit_time;
3068 if (unlikely(extra_len == 4))
3071 packet_parse_headers(skb, sock);
3074 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3078 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3081 err = packet_xmit(po, skb);
3083 if (unlikely(err != 0)) {
3085 err = net_xmit_errno(err);
3102 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3104 struct sock *sk = sock->sk;
3105 struct packet_sock *po = pkt_sk(sk);
3107 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3108 * tpacket_snd() will redo the check safely.
3110 if (data_race(po->tx_ring.pg_vec))
3111 return tpacket_snd(po, msg);
3113 return packet_snd(sock, msg, len);
3117 * Close a PACKET socket. This is fairly simple. We immediately go
3118 * to 'closed' state and remove our protocol entry in the device list.
3121 static int packet_release(struct socket *sock)
3123 struct sock *sk = sock->sk;
3124 struct packet_sock *po;
3125 struct packet_fanout *f;
3127 union tpacket_req_u req_u;
3135 mutex_lock(&net->packet.sklist_lock);
3136 sk_del_node_init_rcu(sk);
3137 mutex_unlock(&net->packet.sklist_lock);
3139 sock_prot_inuse_add(net, sk->sk_prot, -1);
3141 spin_lock(&po->bind_lock);
3142 unregister_prot_hook(sk, false);
3143 packet_cached_dev_reset(po);
3145 if (po->prot_hook.dev) {
3146 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3147 po->prot_hook.dev = NULL;
3149 spin_unlock(&po->bind_lock);
3151 packet_flush_mclist(sk);
3154 if (po->rx_ring.pg_vec) {
3155 memset(&req_u, 0, sizeof(req_u));
3156 packet_set_ring(sk, &req_u, 1, 0);
3159 if (po->tx_ring.pg_vec) {
3160 memset(&req_u, 0, sizeof(req_u));
3161 packet_set_ring(sk, &req_u, 1, 1);
3165 f = fanout_release(sk);
3169 kfree(po->rollover);
3171 fanout_release_data(f);
3175 * Now the socket is dead. No more input will appear.
3182 skb_queue_purge(&sk->sk_receive_queue);
3183 packet_free_pending(po);
3190 * Attach a packet hook.
3193 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3196 struct packet_sock *po = pkt_sk(sk);
3197 struct net_device *dev = NULL;
3198 bool unlisted = false;
3203 spin_lock(&po->bind_lock);
3215 dev = dev_get_by_name_rcu(sock_net(sk), name);
3220 } else if (ifindex) {
3221 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3228 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3232 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
3234 /* prevents packet_notifier() from calling
3235 * register_prot_hook()
3237 WRITE_ONCE(po->num, 0);
3238 __unregister_prot_hook(sk, true);
3241 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3245 BUG_ON(packet_sock_flag(po, PACKET_SOCK_RUNNING));
3246 WRITE_ONCE(po->num, proto);
3247 po->prot_hook.type = proto;
3249 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3251 if (unlikely(unlisted)) {
3252 po->prot_hook.dev = NULL;
3253 WRITE_ONCE(po->ifindex, -1);
3254 packet_cached_dev_reset(po);
3256 netdev_hold(dev, &po->prot_hook.dev_tracker,
3258 po->prot_hook.dev = dev;
3259 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3260 packet_cached_dev_assign(po, dev);
3265 if (proto == 0 || !need_rehook)
3268 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3269 register_prot_hook(sk);
3271 sk->sk_err = ENETDOWN;
3272 if (!sock_flag(sk, SOCK_DEAD))
3273 sk_error_report(sk);
3278 spin_unlock(&po->bind_lock);
3284 * Bind a packet socket to a device
3287 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3290 struct sock *sk = sock->sk;
3291 char name[sizeof(uaddr->sa_data_min) + 1];
3297 if (addr_len != sizeof(struct sockaddr))
3299 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3302 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3303 name[sizeof(uaddr->sa_data_min)] = 0;
3305 return packet_do_bind(sk, name, 0, 0);
3308 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3310 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3311 struct sock *sk = sock->sk;
3317 if (addr_len < sizeof(struct sockaddr_ll))
3319 if (sll->sll_family != AF_PACKET)
3322 return packet_do_bind(sk, NULL, sll->sll_ifindex, sll->sll_protocol);
3325 static struct proto packet_proto = {
3327 .owner = THIS_MODULE,
3328 .obj_size = sizeof(struct packet_sock),
3332 * Create a packet of type SOCK_PACKET.
3335 static int packet_create(struct net *net, struct socket *sock, int protocol,
3339 struct packet_sock *po;
3340 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3343 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3345 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3346 sock->type != SOCK_PACKET)
3347 return -ESOCKTNOSUPPORT;
3349 sock->state = SS_UNCONNECTED;
3352 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3356 sock->ops = &packet_ops;
3357 if (sock->type == SOCK_PACKET)
3358 sock->ops = &packet_ops_spkt;
3360 sock_init_data(sock, sk);
3363 init_completion(&po->skb_completion);
3364 sk->sk_family = PF_PACKET;
3367 err = packet_alloc_pending(po);
3371 packet_cached_dev_reset(po);
3373 sk->sk_destruct = packet_sock_destruct;
3376 * Attach a protocol block
3379 spin_lock_init(&po->bind_lock);
3380 mutex_init(&po->pg_vec_lock);
3381 po->rollover = NULL;
3382 po->prot_hook.func = packet_rcv;
3384 if (sock->type == SOCK_PACKET)
3385 po->prot_hook.func = packet_rcv_spkt;
3387 po->prot_hook.af_packet_priv = sk;
3388 po->prot_hook.af_packet_net = sock_net(sk);
3391 po->prot_hook.type = proto;
3392 __register_prot_hook(sk);
3395 mutex_lock(&net->packet.sklist_lock);
3396 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3397 mutex_unlock(&net->packet.sklist_lock);
3399 sock_prot_inuse_add(net, &packet_proto, 1);
3409 * Pull a packet from our receive queue and hand it to the user.
3410 * If necessary we block.
3413 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3416 struct sock *sk = sock->sk;
3417 struct sk_buff *skb;
3419 int vnet_hdr_len = READ_ONCE(pkt_sk(sk)->vnet_hdr_sz);
3420 unsigned int origlen = 0;
3423 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3427 /* What error should we return now? EUNATTACH? */
3428 if (pkt_sk(sk)->ifindex < 0)
3432 if (flags & MSG_ERRQUEUE) {
3433 err = sock_recv_errqueue(sk, msg, len,
3434 SOL_PACKET, PACKET_TX_TIMESTAMP);
3439 * Call the generic datagram receiver. This handles all sorts
3440 * of horrible races and re-entrancy so we can forget about it
3441 * in the protocol layers.
3443 * Now it will return ENETDOWN, if device have just gone down,
3444 * but then it will block.
3447 skb = skb_recv_datagram(sk, flags, &err);
3450 * An error occurred so return it. Because skb_recv_datagram()
3451 * handles the blocking we don't see and worry about blocking
3458 packet_rcv_try_clear_pressure(pkt_sk(sk));
3461 err = packet_rcv_vnet(msg, skb, &len, vnet_hdr_len);
3466 /* You lose any data beyond the buffer you gave. If it worries
3467 * a user program they can ask the device for its MTU
3473 msg->msg_flags |= MSG_TRUNC;
3476 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3480 if (sock->type != SOCK_PACKET) {
3481 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3483 /* Original length was stored in sockaddr_ll fields */
3484 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3485 sll->sll_family = AF_PACKET;
3486 sll->sll_protocol = skb->protocol;
3489 sock_recv_cmsgs(msg, sk, skb);
3491 if (msg->msg_name) {
3492 const size_t max_len = min(sizeof(skb->cb),
3493 sizeof(struct sockaddr_storage));
3496 /* If the address length field is there to be filled
3497 * in, we fill it in now.
3499 if (sock->type == SOCK_PACKET) {
3500 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3501 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3502 copy_len = msg->msg_namelen;
3504 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3506 msg->msg_namelen = sll->sll_halen +
3507 offsetof(struct sockaddr_ll, sll_addr);
3508 copy_len = msg->msg_namelen;
3509 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3510 memset(msg->msg_name +
3511 offsetof(struct sockaddr_ll, sll_addr),
3512 0, sizeof(sll->sll_addr));
3513 msg->msg_namelen = sizeof(struct sockaddr_ll);
3516 if (WARN_ON_ONCE(copy_len > max_len)) {
3518 msg->msg_namelen = copy_len;
3520 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3523 if (packet_sock_flag(pkt_sk(sk), PACKET_SOCK_AUXDATA)) {
3524 struct tpacket_auxdata aux;
3526 aux.tp_status = TP_STATUS_USER;
3527 if (skb->ip_summed == CHECKSUM_PARTIAL)
3528 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3529 else if (skb->pkt_type != PACKET_OUTGOING &&
3530 skb_csum_unnecessary(skb))
3531 aux.tp_status |= TP_STATUS_CSUM_VALID;
3532 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3533 aux.tp_status |= TP_STATUS_GSO_TCP;
3535 aux.tp_len = origlen;
3536 aux.tp_snaplen = skb->len;
3538 aux.tp_net = skb_network_offset(skb);
3539 if (skb_vlan_tag_present(skb)) {
3540 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3541 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3542 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3544 aux.tp_vlan_tci = 0;
3545 aux.tp_vlan_tpid = 0;
3547 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3551 * Free or return the buffer as appropriate. Again this
3552 * hides all the races and re-entrancy issues from us.
3554 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3557 skb_free_datagram(sk, skb);
3562 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3565 struct net_device *dev;
3566 struct sock *sk = sock->sk;
3571 uaddr->sa_family = AF_PACKET;
3572 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3574 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3576 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3579 return sizeof(*uaddr);
3582 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3585 struct net_device *dev;
3586 struct sock *sk = sock->sk;
3587 struct packet_sock *po = pkt_sk(sk);
3588 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3594 ifindex = READ_ONCE(po->ifindex);
3595 sll->sll_family = AF_PACKET;
3596 sll->sll_ifindex = ifindex;
3597 sll->sll_protocol = READ_ONCE(po->num);
3598 sll->sll_pkttype = 0;
3600 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3602 sll->sll_hatype = dev->type;
3603 sll->sll_halen = dev->addr_len;
3604 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3606 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3611 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3614 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3618 case PACKET_MR_MULTICAST:
3619 if (i->alen != dev->addr_len)
3622 return dev_mc_add(dev, i->addr);
3624 return dev_mc_del(dev, i->addr);
3626 case PACKET_MR_PROMISC:
3627 return dev_set_promiscuity(dev, what);
3628 case PACKET_MR_ALLMULTI:
3629 return dev_set_allmulti(dev, what);
3630 case PACKET_MR_UNICAST:
3631 if (i->alen != dev->addr_len)
3634 return dev_uc_add(dev, i->addr);
3636 return dev_uc_del(dev, i->addr);
3644 static void packet_dev_mclist_delete(struct net_device *dev,
3645 struct packet_mclist **mlp)
3647 struct packet_mclist *ml;
3649 while ((ml = *mlp) != NULL) {
3650 if (ml->ifindex == dev->ifindex) {
3651 packet_dev_mc(dev, ml, -1);
3659 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3661 struct packet_sock *po = pkt_sk(sk);
3662 struct packet_mclist *ml, *i;
3663 struct net_device *dev;
3669 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3674 if (mreq->mr_alen > dev->addr_len)
3678 i = kmalloc(sizeof(*i), GFP_KERNEL);
3683 for (ml = po->mclist; ml; ml = ml->next) {
3684 if (ml->ifindex == mreq->mr_ifindex &&
3685 ml->type == mreq->mr_type &&
3686 ml->alen == mreq->mr_alen &&
3687 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3689 /* Free the new element ... */
3695 i->type = mreq->mr_type;
3696 i->ifindex = mreq->mr_ifindex;
3697 i->alen = mreq->mr_alen;
3698 memcpy(i->addr, mreq->mr_address, i->alen);
3699 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3701 i->next = po->mclist;
3703 err = packet_dev_mc(dev, i, 1);
3705 po->mclist = i->next;
3714 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3716 struct packet_mclist *ml, **mlp;
3720 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3721 if (ml->ifindex == mreq->mr_ifindex &&
3722 ml->type == mreq->mr_type &&
3723 ml->alen == mreq->mr_alen &&
3724 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3725 if (--ml->count == 0) {
3726 struct net_device *dev;
3728 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3730 packet_dev_mc(dev, ml, -1);
3740 static void packet_flush_mclist(struct sock *sk)
3742 struct packet_sock *po = pkt_sk(sk);
3743 struct packet_mclist *ml;
3749 while ((ml = po->mclist) != NULL) {
3750 struct net_device *dev;
3752 po->mclist = ml->next;
3753 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3755 packet_dev_mc(dev, ml, -1);
3762 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3763 unsigned int optlen)
3765 struct sock *sk = sock->sk;
3766 struct packet_sock *po = pkt_sk(sk);
3769 if (level != SOL_PACKET)
3770 return -ENOPROTOOPT;
3773 case PACKET_ADD_MEMBERSHIP:
3774 case PACKET_DROP_MEMBERSHIP:
3776 struct packet_mreq_max mreq;
3778 memset(&mreq, 0, sizeof(mreq));
3779 if (len < sizeof(struct packet_mreq))
3781 if (len > sizeof(mreq))
3783 if (copy_from_sockptr(&mreq, optval, len))
3785 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3787 if (optname == PACKET_ADD_MEMBERSHIP)
3788 ret = packet_mc_add(sk, &mreq);
3790 ret = packet_mc_drop(sk, &mreq);
3794 case PACKET_RX_RING:
3795 case PACKET_TX_RING:
3797 union tpacket_req_u req_u;
3801 switch (po->tp_version) {
3804 len = sizeof(req_u.req);
3808 len = sizeof(req_u.req3);
3814 if (copy_from_sockptr(&req_u.req, optval, len))
3817 ret = packet_set_ring(sk, &req_u, 0,
3818 optname == PACKET_TX_RING);
3823 case PACKET_COPY_THRESH:
3827 if (optlen != sizeof(val))
3829 if (copy_from_sockptr(&val, optval, sizeof(val)))
3832 pkt_sk(sk)->copy_thresh = val;
3835 case PACKET_VERSION:
3839 if (optlen != sizeof(val))
3841 if (copy_from_sockptr(&val, optval, sizeof(val)))
3852 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3855 po->tp_version = val;
3861 case PACKET_RESERVE:
3865 if (optlen != sizeof(val))
3867 if (copy_from_sockptr(&val, optval, sizeof(val)))
3872 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3875 po->tp_reserve = val;
3885 if (optlen != sizeof(val))
3887 if (copy_from_sockptr(&val, optval, sizeof(val)))
3891 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3894 packet_sock_flag_set(po, PACKET_SOCK_TP_LOSS, val);
3900 case PACKET_AUXDATA:
3904 if (optlen < sizeof(val))
3906 if (copy_from_sockptr(&val, optval, sizeof(val)))
3909 packet_sock_flag_set(po, PACKET_SOCK_AUXDATA, val);
3912 case PACKET_ORIGDEV:
3916 if (optlen < sizeof(val))
3918 if (copy_from_sockptr(&val, optval, sizeof(val)))
3921 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3924 case PACKET_VNET_HDR:
3925 case PACKET_VNET_HDR_SZ:
3929 if (sock->type != SOCK_RAW)
3931 if (optlen < sizeof(val))
3933 if (copy_from_sockptr(&val, optval, sizeof(val)))
3936 if (optname == PACKET_VNET_HDR_SZ) {
3937 if (val && val != sizeof(struct virtio_net_hdr) &&
3938 val != sizeof(struct virtio_net_hdr_mrg_rxbuf))
3942 hdr_len = val ? sizeof(struct virtio_net_hdr) : 0;
3945 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3948 WRITE_ONCE(po->vnet_hdr_sz, hdr_len);
3954 case PACKET_TIMESTAMP:
3958 if (optlen != sizeof(val))
3960 if (copy_from_sockptr(&val, optval, sizeof(val)))
3963 WRITE_ONCE(po->tp_tstamp, val);
3968 struct fanout_args args = { 0 };
3970 if (optlen != sizeof(int) && optlen != sizeof(args))
3972 if (copy_from_sockptr(&args, optval, optlen))
3975 return fanout_add(sk, &args);
3977 case PACKET_FANOUT_DATA:
3979 /* Paired with the WRITE_ONCE() in fanout_add() */
3980 if (!READ_ONCE(po->fanout))
3983 return fanout_set_data(po, optval, optlen);
3985 case PACKET_IGNORE_OUTGOING:
3989 if (optlen != sizeof(val))
3991 if (copy_from_sockptr(&val, optval, sizeof(val)))
3993 if (val < 0 || val > 1)
3996 po->prot_hook.ignore_outgoing = !!val;
3999 case PACKET_TX_HAS_OFF:
4003 if (optlen != sizeof(val))
4005 if (copy_from_sockptr(&val, optval, sizeof(val)))
4009 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
4010 packet_sock_flag_set(po, PACKET_SOCK_TX_HAS_OFF, val);
4015 case PACKET_QDISC_BYPASS:
4019 if (optlen != sizeof(val))
4021 if (copy_from_sockptr(&val, optval, sizeof(val)))
4024 packet_sock_flag_set(po, PACKET_SOCK_QDISC_BYPASS, val);
4028 return -ENOPROTOOPT;
4032 static int packet_getsockopt(struct socket *sock, int level, int optname,
4033 char __user *optval, int __user *optlen)
4036 int val, lv = sizeof(val);
4037 struct sock *sk = sock->sk;
4038 struct packet_sock *po = pkt_sk(sk);
4040 union tpacket_stats_u st;
4041 struct tpacket_rollover_stats rstats;
4044 if (level != SOL_PACKET)
4045 return -ENOPROTOOPT;
4047 if (get_user(len, optlen))
4054 case PACKET_STATISTICS:
4055 spin_lock_bh(&sk->sk_receive_queue.lock);
4056 memcpy(&st, &po->stats, sizeof(st));
4057 memset(&po->stats, 0, sizeof(po->stats));
4058 spin_unlock_bh(&sk->sk_receive_queue.lock);
4059 drops = atomic_xchg(&po->tp_drops, 0);
4061 if (po->tp_version == TPACKET_V3) {
4062 lv = sizeof(struct tpacket_stats_v3);
4063 st.stats3.tp_drops = drops;
4064 st.stats3.tp_packets += drops;
4067 lv = sizeof(struct tpacket_stats);
4068 st.stats1.tp_drops = drops;
4069 st.stats1.tp_packets += drops;
4074 case PACKET_AUXDATA:
4075 val = packet_sock_flag(po, PACKET_SOCK_AUXDATA);
4077 case PACKET_ORIGDEV:
4078 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4080 case PACKET_VNET_HDR:
4081 val = !!READ_ONCE(po->vnet_hdr_sz);
4083 case PACKET_VNET_HDR_SZ:
4084 val = READ_ONCE(po->vnet_hdr_sz);
4086 case PACKET_VERSION:
4087 val = po->tp_version;
4090 if (len > sizeof(int))
4092 if (len < sizeof(int))
4094 if (copy_from_user(&val, optval, len))
4098 val = sizeof(struct tpacket_hdr);
4101 val = sizeof(struct tpacket2_hdr);
4104 val = sizeof(struct tpacket3_hdr);
4110 case PACKET_RESERVE:
4111 val = po->tp_reserve;
4114 val = packet_sock_flag(po, PACKET_SOCK_TP_LOSS);
4116 case PACKET_TIMESTAMP:
4117 val = READ_ONCE(po->tp_tstamp);
4121 ((u32)po->fanout->id |
4122 ((u32)po->fanout->type << 16) |
4123 ((u32)po->fanout->flags << 24)) :
4126 case PACKET_IGNORE_OUTGOING:
4127 val = po->prot_hook.ignore_outgoing;
4129 case PACKET_ROLLOVER_STATS:
4132 rstats.tp_all = atomic_long_read(&po->rollover->num);
4133 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4134 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4136 lv = sizeof(rstats);
4138 case PACKET_TX_HAS_OFF:
4139 val = packet_sock_flag(po, PACKET_SOCK_TX_HAS_OFF);
4141 case PACKET_QDISC_BYPASS:
4142 val = packet_sock_flag(po, PACKET_SOCK_QDISC_BYPASS);
4145 return -ENOPROTOOPT;
4150 if (put_user(len, optlen))
4152 if (copy_to_user(optval, data, len))
4157 static int packet_notifier(struct notifier_block *this,
4158 unsigned long msg, void *ptr)
4161 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4162 struct net *net = dev_net(dev);
4165 sk_for_each_rcu(sk, &net->packet.sklist) {
4166 struct packet_sock *po = pkt_sk(sk);
4169 case NETDEV_UNREGISTER:
4171 packet_dev_mclist_delete(dev, &po->mclist);
4175 if (dev->ifindex == po->ifindex) {
4176 spin_lock(&po->bind_lock);
4177 if (packet_sock_flag(po, PACKET_SOCK_RUNNING)) {
4178 __unregister_prot_hook(sk, false);
4179 sk->sk_err = ENETDOWN;
4180 if (!sock_flag(sk, SOCK_DEAD))
4181 sk_error_report(sk);
4183 if (msg == NETDEV_UNREGISTER) {
4184 packet_cached_dev_reset(po);
4185 WRITE_ONCE(po->ifindex, -1);
4186 netdev_put(po->prot_hook.dev,
4187 &po->prot_hook.dev_tracker);
4188 po->prot_hook.dev = NULL;
4190 spin_unlock(&po->bind_lock);
4194 if (dev->ifindex == po->ifindex) {
4195 spin_lock(&po->bind_lock);
4197 register_prot_hook(sk);
4198 spin_unlock(&po->bind_lock);
4208 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4211 struct sock *sk = sock->sk;
4216 int amount = sk_wmem_alloc_get(sk);
4218 return put_user(amount, (int __user *)arg);
4222 struct sk_buff *skb;
4225 spin_lock_bh(&sk->sk_receive_queue.lock);
4226 skb = skb_peek(&sk->sk_receive_queue);
4229 spin_unlock_bh(&sk->sk_receive_queue.lock);
4230 return put_user(amount, (int __user *)arg);
4240 case SIOCGIFBRDADDR:
4241 case SIOCSIFBRDADDR:
4242 case SIOCGIFNETMASK:
4243 case SIOCSIFNETMASK:
4244 case SIOCGIFDSTADDR:
4245 case SIOCSIFDSTADDR:
4247 return inet_dgram_ops.ioctl(sock, cmd, arg);
4251 return -ENOIOCTLCMD;
4256 static __poll_t packet_poll(struct file *file, struct socket *sock,
4259 struct sock *sk = sock->sk;
4260 struct packet_sock *po = pkt_sk(sk);
4261 __poll_t mask = datagram_poll(file, sock, wait);
4263 spin_lock_bh(&sk->sk_receive_queue.lock);
4264 if (po->rx_ring.pg_vec) {
4265 if (!packet_previous_rx_frame(po, &po->rx_ring,
4267 mask |= EPOLLIN | EPOLLRDNORM;
4269 packet_rcv_try_clear_pressure(po);
4270 spin_unlock_bh(&sk->sk_receive_queue.lock);
4271 spin_lock_bh(&sk->sk_write_queue.lock);
4272 if (po->tx_ring.pg_vec) {
4273 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4274 mask |= EPOLLOUT | EPOLLWRNORM;
4276 spin_unlock_bh(&sk->sk_write_queue.lock);
4281 /* Dirty? Well, I still did not learn better way to account
4285 static void packet_mm_open(struct vm_area_struct *vma)
4287 struct file *file = vma->vm_file;
4288 struct socket *sock = file->private_data;
4289 struct sock *sk = sock->sk;
4292 atomic_inc(&pkt_sk(sk)->mapped);
4295 static void packet_mm_close(struct vm_area_struct *vma)
4297 struct file *file = vma->vm_file;
4298 struct socket *sock = file->private_data;
4299 struct sock *sk = sock->sk;
4302 atomic_dec(&pkt_sk(sk)->mapped);
4305 static const struct vm_operations_struct packet_mmap_ops = {
4306 .open = packet_mm_open,
4307 .close = packet_mm_close,
4310 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4315 for (i = 0; i < len; i++) {
4316 if (likely(pg_vec[i].buffer)) {
4317 if (is_vmalloc_addr(pg_vec[i].buffer))
4318 vfree(pg_vec[i].buffer);
4320 free_pages((unsigned long)pg_vec[i].buffer,
4322 pg_vec[i].buffer = NULL;
4328 static char *alloc_one_pg_vec_page(unsigned long order)
4331 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4332 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4334 buffer = (char *) __get_free_pages(gfp_flags, order);
4338 /* __get_free_pages failed, fall back to vmalloc */
4339 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4343 /* vmalloc failed, lets dig into swap here */
4344 gfp_flags &= ~__GFP_NORETRY;
4345 buffer = (char *) __get_free_pages(gfp_flags, order);
4349 /* complete and utter failure */
4353 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4355 unsigned int block_nr = req->tp_block_nr;
4359 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4360 if (unlikely(!pg_vec))
4363 for (i = 0; i < block_nr; i++) {
4364 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4365 if (unlikely(!pg_vec[i].buffer))
4366 goto out_free_pgvec;
4373 free_pg_vec(pg_vec, order, block_nr);
4378 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4379 int closing, int tx_ring)
4381 struct pgv *pg_vec = NULL;
4382 struct packet_sock *po = pkt_sk(sk);
4383 unsigned long *rx_owner_map = NULL;
4384 int was_running, order = 0;
4385 struct packet_ring_buffer *rb;
4386 struct sk_buff_head *rb_queue;
4389 /* Added to avoid minimal code churn */
4390 struct tpacket_req *req = &req_u->req;
4392 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4393 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4397 if (atomic_read(&po->mapped))
4399 if (packet_read_pending(rb))
4403 if (req->tp_block_nr) {
4404 unsigned int min_frame_size;
4406 /* Sanity tests and some calculations */
4408 if (unlikely(rb->pg_vec))
4411 switch (po->tp_version) {
4413 po->tp_hdrlen = TPACKET_HDRLEN;
4416 po->tp_hdrlen = TPACKET2_HDRLEN;
4419 po->tp_hdrlen = TPACKET3_HDRLEN;
4424 if (unlikely((int)req->tp_block_size <= 0))
4426 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4428 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4429 if (po->tp_version >= TPACKET_V3 &&
4430 req->tp_block_size <
4431 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4433 if (unlikely(req->tp_frame_size < min_frame_size))
4435 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4438 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4439 if (unlikely(rb->frames_per_block == 0))
4441 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4443 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4448 order = get_order(req->tp_block_size);
4449 pg_vec = alloc_pg_vec(req, order);
4450 if (unlikely(!pg_vec))
4452 switch (po->tp_version) {
4454 /* Block transmit is not supported yet */
4456 init_prb_bdqc(po, rb, pg_vec, req_u);
4458 struct tpacket_req3 *req3 = &req_u->req3;
4460 if (req3->tp_retire_blk_tov ||
4461 req3->tp_sizeof_priv ||
4462 req3->tp_feature_req_word) {
4464 goto out_free_pg_vec;
4470 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4471 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4473 goto out_free_pg_vec;
4481 if (unlikely(req->tp_frame_nr))
4486 /* Detach socket from network */
4487 spin_lock(&po->bind_lock);
4488 was_running = packet_sock_flag(po, PACKET_SOCK_RUNNING);
4491 WRITE_ONCE(po->num, 0);
4492 __unregister_prot_hook(sk, false);
4494 spin_unlock(&po->bind_lock);
4499 mutex_lock(&po->pg_vec_lock);
4500 if (closing || atomic_read(&po->mapped) == 0) {
4502 spin_lock_bh(&rb_queue->lock);
4503 swap(rb->pg_vec, pg_vec);
4504 if (po->tp_version <= TPACKET_V2)
4505 swap(rb->rx_owner_map, rx_owner_map);
4506 rb->frame_max = (req->tp_frame_nr - 1);
4508 rb->frame_size = req->tp_frame_size;
4509 spin_unlock_bh(&rb_queue->lock);
4511 swap(rb->pg_vec_order, order);
4512 swap(rb->pg_vec_len, req->tp_block_nr);
4514 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4515 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4516 tpacket_rcv : packet_rcv;
4517 skb_queue_purge(rb_queue);
4518 if (atomic_read(&po->mapped))
4519 pr_err("packet_mmap: vma is busy: %d\n",
4520 atomic_read(&po->mapped));
4522 mutex_unlock(&po->pg_vec_lock);
4524 spin_lock(&po->bind_lock);
4526 WRITE_ONCE(po->num, num);
4527 register_prot_hook(sk);
4529 spin_unlock(&po->bind_lock);
4530 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4531 /* Because we don't support block-based V3 on tx-ring */
4533 prb_shutdown_retire_blk_timer(po, rb_queue);
4538 bitmap_free(rx_owner_map);
4539 free_pg_vec(pg_vec, order, req->tp_block_nr);
4545 static int packet_mmap(struct file *file, struct socket *sock,
4546 struct vm_area_struct *vma)
4548 struct sock *sk = sock->sk;
4549 struct packet_sock *po = pkt_sk(sk);
4550 unsigned long size, expected_size;
4551 struct packet_ring_buffer *rb;
4552 unsigned long start;
4559 mutex_lock(&po->pg_vec_lock);
4562 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4564 expected_size += rb->pg_vec_len
4570 if (expected_size == 0)
4573 size = vma->vm_end - vma->vm_start;
4574 if (size != expected_size)
4577 start = vma->vm_start;
4578 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4579 if (rb->pg_vec == NULL)
4582 for (i = 0; i < rb->pg_vec_len; i++) {
4584 void *kaddr = rb->pg_vec[i].buffer;
4587 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4588 page = pgv_to_page(kaddr);
4589 err = vm_insert_page(vma, start, page);
4598 atomic_inc(&po->mapped);
4599 vma->vm_ops = &packet_mmap_ops;
4603 mutex_unlock(&po->pg_vec_lock);
4607 static const struct proto_ops packet_ops_spkt = {
4608 .family = PF_PACKET,
4609 .owner = THIS_MODULE,
4610 .release = packet_release,
4611 .bind = packet_bind_spkt,
4612 .connect = sock_no_connect,
4613 .socketpair = sock_no_socketpair,
4614 .accept = sock_no_accept,
4615 .getname = packet_getname_spkt,
4616 .poll = datagram_poll,
4617 .ioctl = packet_ioctl,
4618 .gettstamp = sock_gettstamp,
4619 .listen = sock_no_listen,
4620 .shutdown = sock_no_shutdown,
4621 .sendmsg = packet_sendmsg_spkt,
4622 .recvmsg = packet_recvmsg,
4623 .mmap = sock_no_mmap,
4624 .sendpage = sock_no_sendpage,
4627 static const struct proto_ops packet_ops = {
4628 .family = PF_PACKET,
4629 .owner = THIS_MODULE,
4630 .release = packet_release,
4631 .bind = packet_bind,
4632 .connect = sock_no_connect,
4633 .socketpair = sock_no_socketpair,
4634 .accept = sock_no_accept,
4635 .getname = packet_getname,
4636 .poll = packet_poll,
4637 .ioctl = packet_ioctl,
4638 .gettstamp = sock_gettstamp,
4639 .listen = sock_no_listen,
4640 .shutdown = sock_no_shutdown,
4641 .setsockopt = packet_setsockopt,
4642 .getsockopt = packet_getsockopt,
4643 .sendmsg = packet_sendmsg,
4644 .recvmsg = packet_recvmsg,
4645 .mmap = packet_mmap,
4646 .sendpage = sock_no_sendpage,
4649 static const struct net_proto_family packet_family_ops = {
4650 .family = PF_PACKET,
4651 .create = packet_create,
4652 .owner = THIS_MODULE,
4655 static struct notifier_block packet_netdev_notifier = {
4656 .notifier_call = packet_notifier,
4659 #ifdef CONFIG_PROC_FS
4661 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4664 struct net *net = seq_file_net(seq);
4667 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4670 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4672 struct net *net = seq_file_net(seq);
4673 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4676 static void packet_seq_stop(struct seq_file *seq, void *v)
4682 static int packet_seq_show(struct seq_file *seq, void *v)
4684 if (v == SEQ_START_TOKEN)
4686 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4687 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4689 struct sock *s = sk_entry(v);
4690 const struct packet_sock *po = pkt_sk(s);
4693 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4695 refcount_read(&s->sk_refcnt),
4697 ntohs(READ_ONCE(po->num)),
4698 READ_ONCE(po->ifindex),
4699 packet_sock_flag(po, PACKET_SOCK_RUNNING),
4700 atomic_read(&s->sk_rmem_alloc),
4701 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4708 static const struct seq_operations packet_seq_ops = {
4709 .start = packet_seq_start,
4710 .next = packet_seq_next,
4711 .stop = packet_seq_stop,
4712 .show = packet_seq_show,
4716 static int __net_init packet_net_init(struct net *net)
4718 mutex_init(&net->packet.sklist_lock);
4719 INIT_HLIST_HEAD(&net->packet.sklist);
4721 #ifdef CONFIG_PROC_FS
4722 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4723 sizeof(struct seq_net_private)))
4725 #endif /* CONFIG_PROC_FS */
4730 static void __net_exit packet_net_exit(struct net *net)
4732 remove_proc_entry("packet", net->proc_net);
4733 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4736 static struct pernet_operations packet_net_ops = {
4737 .init = packet_net_init,
4738 .exit = packet_net_exit,
4742 static void __exit packet_exit(void)
4744 sock_unregister(PF_PACKET);
4745 proto_unregister(&packet_proto);
4746 unregister_netdevice_notifier(&packet_netdev_notifier);
4747 unregister_pernet_subsys(&packet_net_ops);
4750 static int __init packet_init(void)
4754 rc = register_pernet_subsys(&packet_net_ops);
4757 rc = register_netdevice_notifier(&packet_netdev_notifier);
4760 rc = proto_register(&packet_proto, 0);
4763 rc = sock_register(&packet_family_ops);
4770 proto_unregister(&packet_proto);
4772 unregister_netdevice_notifier(&packet_netdev_notifier);
4774 unregister_pernet_subsys(&packet_net_ops);
4779 module_init(packet_init);
4780 module_exit(packet_exit);
4781 MODULE_LICENSE("GPL");
4782 MODULE_ALIAS_NETPROTO(PF_PACKET);