1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * PACKET - implements raw packet sockets.
10 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 * Alan Cox, <gw4pts@gw4pts.ampr.org>
14 * Alan Cox : verify_area() now used correctly
15 * Alan Cox : new skbuff lists, look ma no backlogs!
16 * Alan Cox : tidied skbuff lists.
17 * Alan Cox : Now uses generic datagram routines I
18 * added. Also fixed the peek/read crash
19 * from all old Linux datagram code.
20 * Alan Cox : Uses the improved datagram code.
21 * Alan Cox : Added NULL's for socket options.
22 * Alan Cox : Re-commented the code.
23 * Alan Cox : Use new kernel side addressing
24 * Rob Janssen : Correct MTU usage.
25 * Dave Platt : Counter leaks caused by incorrect
26 * interrupt locking and some slightly
27 * dubious gcc output. Can you read
28 * compiler: it said _VOLATILE_
29 * Richard Kooijman : Timestamp fixes.
30 * Alan Cox : New buffers. Use sk->mac.raw.
31 * Alan Cox : sendmsg/recvmsg support.
32 * Alan Cox : Protocol setting support
33 * Alexey Kuznetsov : Untied from IPv4 stack.
34 * Cyrus Durgin : Fixed kerneld for kmod.
35 * Michal Ostrowski : Module initialization cleanup.
36 * Ulises Alonso : Frame number limit removal and
37 * packet_set_ring memory leak.
38 * Eric Biederman : Allow for > 8 byte hardware addresses.
39 * The convention is that longer addresses
40 * will simply extend the hardware address
41 * byte arrays at the end of sockaddr_ll
43 * Johann Baudy : Added TX RING.
44 * Chetan Loke : Implemented TPACKET_V3 block abstraction
46 * Copyright (C) 2011, <lokec@ccs.neu.edu>
49 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
51 #include <linux/ethtool.h>
52 #include <linux/filter.h>
53 #include <linux/types.h>
55 #include <linux/capability.h>
56 #include <linux/fcntl.h>
57 #include <linux/socket.h>
59 #include <linux/inet.h>
60 #include <linux/netdevice.h>
61 #include <linux/if_packet.h>
62 #include <linux/wireless.h>
63 #include <linux/kernel.h>
64 #include <linux/kmod.h>
65 #include <linux/slab.h>
66 #include <linux/vmalloc.h>
67 #include <net/net_namespace.h>
69 #include <net/protocol.h>
70 #include <linux/skbuff.h>
72 #include <linux/errno.h>
73 #include <linux/timer.h>
74 #include <linux/uaccess.h>
75 #include <asm/ioctls.h>
77 #include <asm/cacheflush.h>
79 #include <linux/proc_fs.h>
80 #include <linux/seq_file.h>
81 #include <linux/poll.h>
82 #include <linux/module.h>
83 #include <linux/init.h>
84 #include <linux/mutex.h>
85 #include <linux/if_vlan.h>
86 #include <linux/virtio_net.h>
87 #include <linux/errqueue.h>
88 #include <linux/net_tstamp.h>
89 #include <linux/percpu.h>
91 #include <net/inet_common.h>
93 #include <linux/bpf.h>
94 #include <net/compat.h>
95 #include <linux/netfilter_netdev.h>
101 - If the device has no dev->header_ops->create, there is no LL header
102 visible above the device. In this case, its hard_header_len should be 0.
103 The device may prepend its own header internally. In this case, its
104 needed_headroom should be set to the space needed for it to add its
106 For example, a WiFi driver pretending to be an Ethernet driver should
107 set its hard_header_len to be the Ethernet header length, and set its
108 needed_headroom to be (the real WiFi header length - the fake Ethernet
110 - packet socket receives packets with pulled ll header,
111 so that SOCK_RAW should push it back.
116 Incoming, dev_has_header(dev) == true
117 mac_header -> ll header
120 Outgoing, dev_has_header(dev) == true
121 mac_header -> ll header
124 Incoming, dev_has_header(dev) == false
126 However drivers often make it point to the ll header.
127 This is incorrect because the ll header should be invisible to us.
130 Outgoing, dev_has_header(dev) == false
131 mac_header -> data. ll header is invisible to us.
135 If dev_has_header(dev) == false we are unable to restore the ll header,
136 because it is invisible to us.
142 dev_has_header(dev) == true
143 mac_header -> ll header
146 dev_has_header(dev) == false (ll header is invisible to us)
150 We should set network_header on output to the correct position,
151 packet classifier depends on it.
154 /* Private packet socket structures. */
156 /* identical to struct packet_mreq except it has
157 * a longer address field.
159 struct packet_mreq_max {
161 unsigned short mr_type;
162 unsigned short mr_alen;
163 unsigned char mr_address[MAX_ADDR_LEN];
167 struct tpacket_hdr *h1;
168 struct tpacket2_hdr *h2;
169 struct tpacket3_hdr *h3;
173 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
174 int closing, int tx_ring);
176 #define V3_ALIGNMENT (8)
178 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
180 #define BLK_PLUS_PRIV(sz_of_priv) \
181 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
183 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
184 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
185 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
186 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
187 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
188 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
191 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
192 struct packet_type *pt, struct net_device *orig_dev);
194 static void *packet_previous_frame(struct packet_sock *po,
195 struct packet_ring_buffer *rb,
197 static void packet_increment_head(struct packet_ring_buffer *buff);
198 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
209 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
210 struct tpacket3_hdr *);
211 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
212 struct tpacket3_hdr *);
213 static void packet_flush_mclist(struct sock *sk);
214 static u16 packet_pick_tx_queue(struct sk_buff *skb);
216 struct packet_skb_cb {
218 struct sockaddr_pkt pkt;
220 /* Trick: alias skb original length with
221 * ll.sll_family and ll.protocol in order
224 unsigned int origlen;
225 struct sockaddr_ll ll;
230 #define vio_le() virtio_legacy_is_little_endian()
232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246 #ifdef CONFIG_NETFILTER_EGRESS
247 static noinline struct sk_buff *nf_hook_direct_egress(struct sk_buff *skb)
249 struct sk_buff *next, *head = NULL, *tail;
253 for (; skb != NULL; skb = next) {
255 skb_mark_not_on_list(skb);
257 if (!nf_hook_egress(skb, &rc, skb->dev))
273 static int packet_direct_xmit(struct sk_buff *skb)
275 #ifdef CONFIG_NETFILTER_EGRESS
276 if (nf_hook_egress_active()) {
277 skb = nf_hook_direct_egress(skb);
279 return NET_XMIT_DROP;
282 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
285 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
287 struct net_device *dev;
290 dev = rcu_dereference(po->cached_dev);
297 static void packet_cached_dev_assign(struct packet_sock *po,
298 struct net_device *dev)
300 rcu_assign_pointer(po->cached_dev, dev);
303 static void packet_cached_dev_reset(struct packet_sock *po)
305 RCU_INIT_POINTER(po->cached_dev, NULL);
308 static bool packet_use_direct_xmit(const struct packet_sock *po)
310 /* Paired with WRITE_ONCE() in packet_setsockopt() */
311 return READ_ONCE(po->xmit) == packet_direct_xmit;
314 static u16 packet_pick_tx_queue(struct sk_buff *skb)
316 struct net_device *dev = skb->dev;
317 const struct net_device_ops *ops = dev->netdev_ops;
318 int cpu = raw_smp_processor_id();
322 skb->sender_cpu = cpu + 1;
324 skb_record_rx_queue(skb, cpu % dev->real_num_tx_queues);
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL);
327 queue_index = netdev_cap_txqueue(dev, queue_index);
329 queue_index = netdev_pick_tx(dev, skb, NULL);
335 /* __register_prot_hook must be invoked through register_prot_hook
336 * or from a context in which asynchronous accesses to the packet
337 * socket is not possible (packet_create()).
339 static void __register_prot_hook(struct sock *sk)
341 struct packet_sock *po = pkt_sk(sk);
345 __fanout_link(sk, po);
347 dev_add_pack(&po->prot_hook);
354 static void register_prot_hook(struct sock *sk)
356 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
357 __register_prot_hook(sk);
360 /* If the sync parameter is true, we will temporarily drop
361 * the po->bind_lock and do a synchronize_net to make sure no
362 * asynchronous packet processing paths still refer to the elements
363 * of po->prot_hook. If the sync parameter is false, it is the
364 * callers responsibility to take care of this.
366 static void __unregister_prot_hook(struct sock *sk, bool sync)
368 struct packet_sock *po = pkt_sk(sk);
370 lockdep_assert_held_once(&po->bind_lock);
375 __fanout_unlink(sk, po);
377 __dev_remove_pack(&po->prot_hook);
382 spin_unlock(&po->bind_lock);
384 spin_lock(&po->bind_lock);
388 static void unregister_prot_hook(struct sock *sk, bool sync)
390 struct packet_sock *po = pkt_sk(sk);
393 __unregister_prot_hook(sk, sync);
396 static inline struct page * __pure pgv_to_page(void *addr)
398 if (is_vmalloc_addr(addr))
399 return vmalloc_to_page(addr);
400 return virt_to_page(addr);
403 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
405 union tpacket_uhdr h;
408 switch (po->tp_version) {
410 h.h1->tp_status = status;
411 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
414 h.h2->tp_status = status;
415 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
418 h.h3->tp_status = status;
419 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
422 WARN(1, "TPACKET version not supported.\n");
429 static int __packet_get_status(const struct packet_sock *po, void *frame)
431 union tpacket_uhdr h;
436 switch (po->tp_version) {
438 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
439 return h.h1->tp_status;
441 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
442 return h.h2->tp_status;
444 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
445 return h.h3->tp_status;
447 WARN(1, "TPACKET version not supported.\n");
453 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec64 *ts,
456 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
459 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
460 ktime_to_timespec64_cond(shhwtstamps->hwtstamp, ts))
461 return TP_STATUS_TS_RAW_HARDWARE;
463 if ((flags & SOF_TIMESTAMPING_SOFTWARE) &&
464 ktime_to_timespec64_cond(skb_tstamp(skb), ts))
465 return TP_STATUS_TS_SOFTWARE;
470 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
473 union tpacket_uhdr h;
474 struct timespec64 ts;
477 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
482 * versions 1 through 3 overflow the timestamps in y2106, since they
483 * all store the seconds in a 32-bit unsigned integer.
484 * If we create a version 4, that should have a 64-bit timestamp,
485 * either 64-bit seconds + 32-bit nanoseconds, or just 64-bit
488 switch (po->tp_version) {
490 h.h1->tp_sec = ts.tv_sec;
491 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
494 h.h2->tp_sec = ts.tv_sec;
495 h.h2->tp_nsec = ts.tv_nsec;
498 h.h3->tp_sec = ts.tv_sec;
499 h.h3->tp_nsec = ts.tv_nsec;
502 WARN(1, "TPACKET version not supported.\n");
506 /* one flush is safe, as both fields always lie on the same cacheline */
507 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
513 static void *packet_lookup_frame(const struct packet_sock *po,
514 const struct packet_ring_buffer *rb,
515 unsigned int position,
518 unsigned int pg_vec_pos, frame_offset;
519 union tpacket_uhdr h;
521 pg_vec_pos = position / rb->frames_per_block;
522 frame_offset = position % rb->frames_per_block;
524 h.raw = rb->pg_vec[pg_vec_pos].buffer +
525 (frame_offset * rb->frame_size);
527 if (status != __packet_get_status(po, h.raw))
533 static void *packet_current_frame(struct packet_sock *po,
534 struct packet_ring_buffer *rb,
537 return packet_lookup_frame(po, rb, rb->head, status);
540 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
542 del_timer_sync(&pkc->retire_blk_timer);
545 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
546 struct sk_buff_head *rb_queue)
548 struct tpacket_kbdq_core *pkc;
550 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552 spin_lock_bh(&rb_queue->lock);
553 pkc->delete_blk_timer = 1;
554 spin_unlock_bh(&rb_queue->lock);
556 prb_del_retire_blk_timer(pkc);
559 static void prb_setup_retire_blk_timer(struct packet_sock *po)
561 struct tpacket_kbdq_core *pkc;
563 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
564 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
566 pkc->retire_blk_timer.expires = jiffies;
569 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
570 int blk_size_in_bytes)
572 struct net_device *dev;
573 unsigned int mbits, div;
574 struct ethtool_link_ksettings ecmd;
578 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
579 if (unlikely(!dev)) {
581 return DEFAULT_PRB_RETIRE_TOV;
583 err = __ethtool_get_link_ksettings(dev, &ecmd);
586 return DEFAULT_PRB_RETIRE_TOV;
588 /* If the link speed is so slow you don't really
589 * need to worry about perf anyways
591 if (ecmd.base.speed < SPEED_1000 ||
592 ecmd.base.speed == SPEED_UNKNOWN)
593 return DEFAULT_PRB_RETIRE_TOV;
595 div = ecmd.base.speed / 1000;
596 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
606 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
607 union tpacket_req_u *req_u)
609 p1->feature_req_word = req_u->req3.tp_feature_req_word;
612 static void init_prb_bdqc(struct packet_sock *po,
613 struct packet_ring_buffer *rb,
615 union tpacket_req_u *req_u)
617 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
618 struct tpacket_block_desc *pbd;
620 memset(p1, 0x0, sizeof(*p1));
622 p1->knxt_seq_num = 1;
624 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
625 p1->pkblk_start = pg_vec[0].buffer;
626 p1->kblk_size = req_u->req3.tp_block_size;
627 p1->knum_blocks = req_u->req3.tp_block_nr;
628 p1->hdrlen = po->tp_hdrlen;
629 p1->version = po->tp_version;
630 p1->last_kactive_blk_num = 0;
631 po->stats.stats3.tp_freeze_q_cnt = 0;
632 if (req_u->req3.tp_retire_blk_tov)
633 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
635 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
636 req_u->req3.tp_block_size);
637 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
638 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
639 rwlock_init(&p1->blk_fill_in_prog_lock);
641 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
642 prb_init_ft_ops(p1, req_u);
643 prb_setup_retire_blk_timer(po);
644 prb_open_block(p1, pbd);
647 /* Do NOT update the last_blk_num first.
648 * Assumes sk_buff_head lock is held.
650 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
652 mod_timer(&pkc->retire_blk_timer,
653 jiffies + pkc->tov_in_jiffies);
654 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
659 * 1) We refresh the timer only when we open a block.
660 * By doing this we don't waste cycles refreshing the timer
661 * on packet-by-packet basis.
663 * With a 1MB block-size, on a 1Gbps line, it will take
664 * i) ~8 ms to fill a block + ii) memcpy etc.
665 * In this cut we are not accounting for the memcpy time.
667 * So, if the user sets the 'tmo' to 10ms then the timer
668 * will never fire while the block is still getting filled
669 * (which is what we want). However, the user could choose
670 * to close a block early and that's fine.
672 * But when the timer does fire, we check whether or not to refresh it.
673 * Since the tmo granularity is in msecs, it is not too expensive
674 * to refresh the timer, lets say every '8' msecs.
675 * Either the user can set the 'tmo' or we can derive it based on
676 * a) line-speed and b) block-size.
677 * prb_calc_retire_blk_tmo() calculates the tmo.
680 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
682 struct packet_sock *po =
683 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
684 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
686 struct tpacket_block_desc *pbd;
688 spin_lock(&po->sk.sk_receive_queue.lock);
690 frozen = prb_queue_frozen(pkc);
691 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
693 if (unlikely(pkc->delete_blk_timer))
696 /* We only need to plug the race when the block is partially filled.
698 * lock(); increment BLOCK_NUM_PKTS; unlock()
699 * copy_bits() is in progress ...
700 * timer fires on other cpu:
701 * we can't retire the current block because copy_bits
705 if (BLOCK_NUM_PKTS(pbd)) {
706 /* Waiting for skb_copy_bits to finish... */
707 write_lock(&pkc->blk_fill_in_prog_lock);
708 write_unlock(&pkc->blk_fill_in_prog_lock);
711 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
713 if (!BLOCK_NUM_PKTS(pbd)) {
714 /* An empty block. Just refresh the timer. */
717 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
718 if (!prb_dispatch_next_block(pkc, po))
723 /* Case 1. Queue was frozen because user-space was
726 if (prb_curr_blk_in_use(pbd)) {
728 * Ok, user-space is still behind.
729 * So just refresh the timer.
733 /* Case 2. queue was frozen,user-space caught up,
734 * now the link went idle && the timer fired.
735 * We don't have a block to close.So we open this
736 * block and restart the timer.
737 * opening a block thaws the queue,restarts timer
738 * Thawing/timer-refresh is a side effect.
740 prb_open_block(pkc, pbd);
747 _prb_refresh_rx_retire_blk_timer(pkc);
750 spin_unlock(&po->sk.sk_receive_queue.lock);
753 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
754 struct tpacket_block_desc *pbd1, __u32 status)
756 /* Flush everything minus the block header */
758 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
763 /* Skip the block header(we know header WILL fit in 4K) */
766 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
767 for (; start < end; start += PAGE_SIZE)
768 flush_dcache_page(pgv_to_page(start));
773 /* Now update the block status. */
775 BLOCK_STATUS(pbd1) = status;
777 /* Flush the block header */
779 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
781 flush_dcache_page(pgv_to_page(start));
791 * 2) Increment active_blk_num
793 * Note:We DONT refresh the timer on purpose.
794 * Because almost always the next block will be opened.
796 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
797 struct tpacket_block_desc *pbd1,
798 struct packet_sock *po, unsigned int stat)
800 __u32 status = TP_STATUS_USER | stat;
802 struct tpacket3_hdr *last_pkt;
803 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
804 struct sock *sk = &po->sk;
806 if (atomic_read(&po->tp_drops))
807 status |= TP_STATUS_LOSING;
809 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
810 last_pkt->tp_next_offset = 0;
812 /* Get the ts of the last pkt */
813 if (BLOCK_NUM_PKTS(pbd1)) {
814 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
815 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
817 /* Ok, we tmo'd - so get the current time.
819 * It shouldn't really happen as we don't close empty
820 * blocks. See prb_retire_rx_blk_timer_expired().
822 struct timespec64 ts;
823 ktime_get_real_ts64(&ts);
824 h1->ts_last_pkt.ts_sec = ts.tv_sec;
825 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
830 /* Flush the block */
831 prb_flush_block(pkc1, pbd1, status);
833 sk->sk_data_ready(sk);
835 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
838 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
840 pkc->reset_pending_on_curr_blk = 0;
844 * Side effect of opening a block:
846 * 1) prb_queue is thawed.
847 * 2) retire_blk_timer is refreshed.
850 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
851 struct tpacket_block_desc *pbd1)
853 struct timespec64 ts;
854 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
858 /* We could have just memset this but we will lose the
859 * flexibility of making the priv area sticky
862 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
863 BLOCK_NUM_PKTS(pbd1) = 0;
864 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 ktime_get_real_ts64(&ts);
868 h1->ts_first_pkt.ts_sec = ts.tv_sec;
869 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
871 pkc1->pkblk_start = (char *)pbd1;
872 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
874 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
875 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
877 pbd1->version = pkc1->version;
878 pkc1->prev = pkc1->nxt_offset;
879 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
881 prb_thaw_queue(pkc1);
882 _prb_refresh_rx_retire_blk_timer(pkc1);
888 * Queue freeze logic:
889 * 1) Assume tp_block_nr = 8 blocks.
890 * 2) At time 't0', user opens Rx ring.
891 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
892 * 4) user-space is either sleeping or processing block '0'.
893 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
894 * it will close block-7,loop around and try to fill block '0'.
896 * __packet_lookup_frame_in_block
897 * prb_retire_current_block()
898 * prb_dispatch_next_block()
899 * |->(BLOCK_STATUS == USER) evaluates to true
900 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
901 * 6) Now there are two cases:
902 * 6.1) Link goes idle right after the queue is frozen.
903 * But remember, the last open_block() refreshed the timer.
904 * When this timer expires,it will refresh itself so that we can
905 * re-open block-0 in near future.
906 * 6.2) Link is busy and keeps on receiving packets. This is a simple
907 * case and __packet_lookup_frame_in_block will check if block-0
908 * is free and can now be re-used.
910 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
911 struct packet_sock *po)
913 pkc->reset_pending_on_curr_blk = 1;
914 po->stats.stats3.tp_freeze_q_cnt++;
917 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
920 * If the next block is free then we will dispatch it
921 * and return a good offset.
922 * Else, we will freeze the queue.
923 * So, caller must check the return value.
925 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
926 struct packet_sock *po)
928 struct tpacket_block_desc *pbd;
932 /* 1. Get current block num */
933 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
935 /* 2. If this block is currently in_use then freeze the queue */
936 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
937 prb_freeze_queue(pkc, po);
943 * open this block and return the offset where the first packet
944 * needs to get stored.
946 prb_open_block(pkc, pbd);
947 return (void *)pkc->nxt_offset;
950 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
951 struct packet_sock *po, unsigned int status)
953 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
955 /* retire/close the current block */
956 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
958 * Plug the case where copy_bits() is in progress on
959 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
960 * have space to copy the pkt in the current block and
961 * called prb_retire_current_block()
963 * We don't need to worry about the TMO case because
964 * the timer-handler already handled this case.
966 if (!(status & TP_STATUS_BLK_TMO)) {
967 /* Waiting for skb_copy_bits to finish... */
968 write_lock(&pkc->blk_fill_in_prog_lock);
969 write_unlock(&pkc->blk_fill_in_prog_lock);
971 prb_close_block(pkc, pbd, po, status);
976 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
978 return TP_STATUS_USER & BLOCK_STATUS(pbd);
981 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
983 return pkc->reset_pending_on_curr_blk;
986 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
987 __releases(&pkc->blk_fill_in_prog_lock)
989 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
991 read_unlock(&pkc->blk_fill_in_prog_lock);
994 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
995 struct tpacket3_hdr *ppd)
997 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
1000 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
1001 struct tpacket3_hdr *ppd)
1003 ppd->hv1.tp_rxhash = 0;
1006 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1007 struct tpacket3_hdr *ppd)
1009 if (skb_vlan_tag_present(pkc->skb)) {
1010 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1011 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1012 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1014 ppd->hv1.tp_vlan_tci = 0;
1015 ppd->hv1.tp_vlan_tpid = 0;
1016 ppd->tp_status = TP_STATUS_AVAILABLE;
1020 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1021 struct tpacket3_hdr *ppd)
1023 ppd->hv1.tp_padding = 0;
1024 prb_fill_vlan_info(pkc, ppd);
1026 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1027 prb_fill_rxhash(pkc, ppd);
1029 prb_clear_rxhash(pkc, ppd);
1032 static void prb_fill_curr_block(char *curr,
1033 struct tpacket_kbdq_core *pkc,
1034 struct tpacket_block_desc *pbd,
1036 __acquires(&pkc->blk_fill_in_prog_lock)
1038 struct tpacket3_hdr *ppd;
1040 ppd = (struct tpacket3_hdr *)curr;
1041 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1043 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1044 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1045 BLOCK_NUM_PKTS(pbd) += 1;
1046 read_lock(&pkc->blk_fill_in_prog_lock);
1047 prb_run_all_ft_ops(pkc, ppd);
1050 /* Assumes caller has the sk->rx_queue.lock */
1051 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1052 struct sk_buff *skb,
1056 struct tpacket_kbdq_core *pkc;
1057 struct tpacket_block_desc *pbd;
1060 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1061 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1063 /* Queue is frozen when user space is lagging behind */
1064 if (prb_queue_frozen(pkc)) {
1066 * Check if that last block which caused the queue to freeze,
1067 * is still in_use by user-space.
1069 if (prb_curr_blk_in_use(pbd)) {
1070 /* Can't record this packet */
1074 * Ok, the block was released by user-space.
1075 * Now let's open that block.
1076 * opening a block also thaws the queue.
1077 * Thawing is a side effect.
1079 prb_open_block(pkc, pbd);
1084 curr = pkc->nxt_offset;
1086 end = (char *)pbd + pkc->kblk_size;
1088 /* first try the current block */
1089 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1090 prb_fill_curr_block(curr, pkc, pbd, len);
1091 return (void *)curr;
1094 /* Ok, close the current block */
1095 prb_retire_current_block(pkc, po, 0);
1097 /* Now, try to dispatch the next block */
1098 curr = (char *)prb_dispatch_next_block(pkc, po);
1100 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1101 prb_fill_curr_block(curr, pkc, pbd, len);
1102 return (void *)curr;
1106 * No free blocks are available.user_space hasn't caught up yet.
1107 * Queue was just frozen and now this packet will get dropped.
1112 static void *packet_current_rx_frame(struct packet_sock *po,
1113 struct sk_buff *skb,
1114 int status, unsigned int len)
1117 switch (po->tp_version) {
1120 curr = packet_lookup_frame(po, &po->rx_ring,
1121 po->rx_ring.head, status);
1124 return __packet_lookup_frame_in_block(po, skb, len);
1126 WARN(1, "TPACKET version not supported\n");
1132 static void *prb_lookup_block(const struct packet_sock *po,
1133 const struct packet_ring_buffer *rb,
1137 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1138 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1140 if (status != BLOCK_STATUS(pbd))
1145 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1148 if (rb->prb_bdqc.kactive_blk_num)
1149 prev = rb->prb_bdqc.kactive_blk_num-1;
1151 prev = rb->prb_bdqc.knum_blocks-1;
1155 /* Assumes caller has held the rx_queue.lock */
1156 static void *__prb_previous_block(struct packet_sock *po,
1157 struct packet_ring_buffer *rb,
1160 unsigned int previous = prb_previous_blk_num(rb);
1161 return prb_lookup_block(po, rb, previous, status);
1164 static void *packet_previous_rx_frame(struct packet_sock *po,
1165 struct packet_ring_buffer *rb,
1168 if (po->tp_version <= TPACKET_V2)
1169 return packet_previous_frame(po, rb, status);
1171 return __prb_previous_block(po, rb, status);
1174 static void packet_increment_rx_head(struct packet_sock *po,
1175 struct packet_ring_buffer *rb)
1177 switch (po->tp_version) {
1180 return packet_increment_head(rb);
1183 WARN(1, "TPACKET version not supported.\n");
1189 static void *packet_previous_frame(struct packet_sock *po,
1190 struct packet_ring_buffer *rb,
1193 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1194 return packet_lookup_frame(po, rb, previous, status);
1197 static void packet_increment_head(struct packet_ring_buffer *buff)
1199 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1202 static void packet_inc_pending(struct packet_ring_buffer *rb)
1204 this_cpu_inc(*rb->pending_refcnt);
1207 static void packet_dec_pending(struct packet_ring_buffer *rb)
1209 this_cpu_dec(*rb->pending_refcnt);
1212 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1214 unsigned int refcnt = 0;
1217 /* We don't use pending refcount in rx_ring. */
1218 if (rb->pending_refcnt == NULL)
1221 for_each_possible_cpu(cpu)
1222 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1227 static int packet_alloc_pending(struct packet_sock *po)
1229 po->rx_ring.pending_refcnt = NULL;
1231 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1232 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1238 static void packet_free_pending(struct packet_sock *po)
1240 free_percpu(po->tx_ring.pending_refcnt);
1243 #define ROOM_POW_OFF 2
1244 #define ROOM_NONE 0x0
1245 #define ROOM_LOW 0x1
1246 #define ROOM_NORMAL 0x2
1248 static bool __tpacket_has_room(const struct packet_sock *po, int pow_off)
1252 len = READ_ONCE(po->rx_ring.frame_max) + 1;
1253 idx = READ_ONCE(po->rx_ring.head);
1255 idx += len >> pow_off;
1258 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1261 static bool __tpacket_v3_has_room(const struct packet_sock *po, int pow_off)
1265 len = READ_ONCE(po->rx_ring.prb_bdqc.knum_blocks);
1266 idx = READ_ONCE(po->rx_ring.prb_bdqc.kactive_blk_num);
1268 idx += len >> pow_off;
1271 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1274 static int __packet_rcv_has_room(const struct packet_sock *po,
1275 const struct sk_buff *skb)
1277 const struct sock *sk = &po->sk;
1278 int ret = ROOM_NONE;
1280 if (po->prot_hook.func != tpacket_rcv) {
1281 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1282 int avail = rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1283 - (skb ? skb->truesize : 0);
1285 if (avail > (rcvbuf >> ROOM_POW_OFF))
1293 if (po->tp_version == TPACKET_V3) {
1294 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1296 else if (__tpacket_v3_has_room(po, 0))
1299 if (__tpacket_has_room(po, ROOM_POW_OFF))
1301 else if (__tpacket_has_room(po, 0))
1308 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1312 ret = __packet_rcv_has_room(po, skb);
1313 pressure = ret != ROOM_NORMAL;
1315 if (READ_ONCE(po->pressure) != pressure)
1316 WRITE_ONCE(po->pressure, pressure);
1321 static void packet_rcv_try_clear_pressure(struct packet_sock *po)
1323 if (READ_ONCE(po->pressure) &&
1324 __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
1325 WRITE_ONCE(po->pressure, 0);
1328 static void packet_sock_destruct(struct sock *sk)
1330 skb_queue_purge(&sk->sk_error_queue);
1332 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1333 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1335 if (!sock_flag(sk, SOCK_DEAD)) {
1336 pr_err("Attempt to release alive packet socket: %p\n", sk);
1341 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1343 u32 *history = po->rollover->history;
1347 rxhash = skb_get_hash(skb);
1348 for (i = 0; i < ROLLOVER_HLEN; i++)
1349 if (READ_ONCE(history[i]) == rxhash)
1352 victim = get_random_u32_below(ROLLOVER_HLEN);
1354 /* Avoid dirtying the cache line if possible */
1355 if (READ_ONCE(history[victim]) != rxhash)
1356 WRITE_ONCE(history[victim], rxhash);
1358 return count > (ROLLOVER_HLEN >> 1);
1361 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1362 struct sk_buff *skb,
1365 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1368 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1369 struct sk_buff *skb,
1372 unsigned int val = atomic_inc_return(&f->rr_cur);
1377 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1378 struct sk_buff *skb,
1381 return smp_processor_id() % num;
1384 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1385 struct sk_buff *skb,
1388 return get_random_u32_below(num);
1391 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1392 struct sk_buff *skb,
1393 unsigned int idx, bool try_self,
1396 struct packet_sock *po, *po_next, *po_skip = NULL;
1397 unsigned int i, j, room = ROOM_NONE;
1399 po = pkt_sk(rcu_dereference(f->arr[idx]));
1402 room = packet_rcv_has_room(po, skb);
1403 if (room == ROOM_NORMAL ||
1404 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1409 i = j = min_t(int, po->rollover->sock, num - 1);
1411 po_next = pkt_sk(rcu_dereference(f->arr[i]));
1412 if (po_next != po_skip && !READ_ONCE(po_next->pressure) &&
1413 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1415 po->rollover->sock = i;
1416 atomic_long_inc(&po->rollover->num);
1417 if (room == ROOM_LOW)
1418 atomic_long_inc(&po->rollover->num_huge);
1426 atomic_long_inc(&po->rollover->num_failed);
1430 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1431 struct sk_buff *skb,
1434 return skb_get_queue_mapping(skb) % num;
1437 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1438 struct sk_buff *skb,
1441 struct bpf_prog *prog;
1442 unsigned int ret = 0;
1445 prog = rcu_dereference(f->bpf_prog);
1447 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1453 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1455 return f->flags & (flag >> 8);
1458 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1459 struct packet_type *pt, struct net_device *orig_dev)
1461 struct packet_fanout *f = pt->af_packet_priv;
1462 unsigned int num = READ_ONCE(f->num_members);
1463 struct net *net = read_pnet(&f->net);
1464 struct packet_sock *po;
1467 if (!net_eq(dev_net(dev), net) || !num) {
1472 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1473 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1478 case PACKET_FANOUT_HASH:
1480 idx = fanout_demux_hash(f, skb, num);
1482 case PACKET_FANOUT_LB:
1483 idx = fanout_demux_lb(f, skb, num);
1485 case PACKET_FANOUT_CPU:
1486 idx = fanout_demux_cpu(f, skb, num);
1488 case PACKET_FANOUT_RND:
1489 idx = fanout_demux_rnd(f, skb, num);
1491 case PACKET_FANOUT_QM:
1492 idx = fanout_demux_qm(f, skb, num);
1494 case PACKET_FANOUT_ROLLOVER:
1495 idx = fanout_demux_rollover(f, skb, 0, false, num);
1497 case PACKET_FANOUT_CBPF:
1498 case PACKET_FANOUT_EBPF:
1499 idx = fanout_demux_bpf(f, skb, num);
1503 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1504 idx = fanout_demux_rollover(f, skb, idx, true, num);
1506 po = pkt_sk(rcu_dereference(f->arr[idx]));
1507 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1510 DEFINE_MUTEX(fanout_mutex);
1511 EXPORT_SYMBOL_GPL(fanout_mutex);
1512 static LIST_HEAD(fanout_list);
1513 static u16 fanout_next_id;
1515 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1517 struct packet_fanout *f = po->fanout;
1519 spin_lock(&f->lock);
1520 rcu_assign_pointer(f->arr[f->num_members], sk);
1523 if (f->num_members == 1)
1524 dev_add_pack(&f->prot_hook);
1525 spin_unlock(&f->lock);
1528 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1530 struct packet_fanout *f = po->fanout;
1533 spin_lock(&f->lock);
1534 for (i = 0; i < f->num_members; i++) {
1535 if (rcu_dereference_protected(f->arr[i],
1536 lockdep_is_held(&f->lock)) == sk)
1539 BUG_ON(i >= f->num_members);
1540 rcu_assign_pointer(f->arr[i],
1541 rcu_dereference_protected(f->arr[f->num_members - 1],
1542 lockdep_is_held(&f->lock)));
1544 if (f->num_members == 0)
1545 __dev_remove_pack(&f->prot_hook);
1546 spin_unlock(&f->lock);
1549 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1551 if (sk->sk_family != PF_PACKET)
1554 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1557 static void fanout_init_data(struct packet_fanout *f)
1560 case PACKET_FANOUT_LB:
1561 atomic_set(&f->rr_cur, 0);
1563 case PACKET_FANOUT_CBPF:
1564 case PACKET_FANOUT_EBPF:
1565 RCU_INIT_POINTER(f->bpf_prog, NULL);
1570 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1572 struct bpf_prog *old;
1574 spin_lock(&f->lock);
1575 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1576 rcu_assign_pointer(f->bpf_prog, new);
1577 spin_unlock(&f->lock);
1581 bpf_prog_destroy(old);
1585 static int fanout_set_data_cbpf(struct packet_sock *po, sockptr_t data,
1588 struct bpf_prog *new;
1589 struct sock_fprog fprog;
1592 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1595 ret = copy_bpf_fprog_from_user(&fprog, data, len);
1599 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1603 __fanout_set_data_bpf(po->fanout, new);
1607 static int fanout_set_data_ebpf(struct packet_sock *po, sockptr_t data,
1610 struct bpf_prog *new;
1613 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1615 if (len != sizeof(fd))
1617 if (copy_from_sockptr(&fd, data, len))
1620 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1622 return PTR_ERR(new);
1624 __fanout_set_data_bpf(po->fanout, new);
1628 static int fanout_set_data(struct packet_sock *po, sockptr_t data,
1631 switch (po->fanout->type) {
1632 case PACKET_FANOUT_CBPF:
1633 return fanout_set_data_cbpf(po, data, len);
1634 case PACKET_FANOUT_EBPF:
1635 return fanout_set_data_ebpf(po, data, len);
1641 static void fanout_release_data(struct packet_fanout *f)
1644 case PACKET_FANOUT_CBPF:
1645 case PACKET_FANOUT_EBPF:
1646 __fanout_set_data_bpf(f, NULL);
1650 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1652 struct packet_fanout *f;
1654 list_for_each_entry(f, &fanout_list, list) {
1655 if (f->id == candidate_id &&
1656 read_pnet(&f->net) == sock_net(sk)) {
1663 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1665 u16 id = fanout_next_id;
1668 if (__fanout_id_is_free(sk, id)) {
1670 fanout_next_id = id + 1;
1675 } while (id != fanout_next_id);
1680 static int fanout_add(struct sock *sk, struct fanout_args *args)
1682 struct packet_rollover *rollover = NULL;
1683 struct packet_sock *po = pkt_sk(sk);
1684 u16 type_flags = args->type_flags;
1685 struct packet_fanout *f, *match;
1686 u8 type = type_flags & 0xff;
1687 u8 flags = type_flags >> 8;
1692 case PACKET_FANOUT_ROLLOVER:
1693 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1696 case PACKET_FANOUT_HASH:
1697 case PACKET_FANOUT_LB:
1698 case PACKET_FANOUT_CPU:
1699 case PACKET_FANOUT_RND:
1700 case PACKET_FANOUT_QM:
1701 case PACKET_FANOUT_CBPF:
1702 case PACKET_FANOUT_EBPF:
1708 mutex_lock(&fanout_mutex);
1714 if (type == PACKET_FANOUT_ROLLOVER ||
1715 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1717 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1720 atomic_long_set(&rollover->num, 0);
1721 atomic_long_set(&rollover->num_huge, 0);
1722 atomic_long_set(&rollover->num_failed, 0);
1725 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1730 if (!fanout_find_new_id(sk, &id)) {
1734 /* ephemeral flag for the first socket in the group: drop it */
1735 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1739 list_for_each_entry(f, &fanout_list, list) {
1741 read_pnet(&f->net) == sock_net(sk)) {
1748 if (match->flags != flags)
1750 if (args->max_num_members &&
1751 args->max_num_members != match->max_num_members)
1754 if (args->max_num_members > PACKET_FANOUT_MAX)
1756 if (!args->max_num_members)
1757 /* legacy PACKET_FANOUT_MAX */
1758 args->max_num_members = 256;
1760 match = kvzalloc(struct_size(match, arr, args->max_num_members),
1764 write_pnet(&match->net, sock_net(sk));
1767 match->flags = flags;
1768 INIT_LIST_HEAD(&match->list);
1769 spin_lock_init(&match->lock);
1770 refcount_set(&match->sk_ref, 0);
1771 fanout_init_data(match);
1772 match->prot_hook.type = po->prot_hook.type;
1773 match->prot_hook.dev = po->prot_hook.dev;
1774 match->prot_hook.func = packet_rcv_fanout;
1775 match->prot_hook.af_packet_priv = match;
1776 match->prot_hook.af_packet_net = read_pnet(&match->net);
1777 match->prot_hook.id_match = match_fanout_group;
1778 match->max_num_members = args->max_num_members;
1779 match->prot_hook.ignore_outgoing = type_flags & PACKET_FANOUT_FLAG_IGNORE_OUTGOING;
1780 list_add(&match->list, &fanout_list);
1784 spin_lock(&po->bind_lock);
1786 match->type == type &&
1787 match->prot_hook.type == po->prot_hook.type &&
1788 match->prot_hook.dev == po->prot_hook.dev) {
1790 if (refcount_read(&match->sk_ref) < match->max_num_members) {
1791 __dev_remove_pack(&po->prot_hook);
1793 /* Paired with packet_setsockopt(PACKET_FANOUT_DATA) */
1794 WRITE_ONCE(po->fanout, match);
1796 po->rollover = rollover;
1798 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1799 __fanout_link(sk, po);
1803 spin_unlock(&po->bind_lock);
1805 if (err && !refcount_read(&match->sk_ref)) {
1806 list_del(&match->list);
1812 mutex_unlock(&fanout_mutex);
1816 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1817 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1818 * It is the responsibility of the caller to call fanout_release_data() and
1819 * free the returned packet_fanout (after synchronize_net())
1821 static struct packet_fanout *fanout_release(struct sock *sk)
1823 struct packet_sock *po = pkt_sk(sk);
1824 struct packet_fanout *f;
1826 mutex_lock(&fanout_mutex);
1831 if (refcount_dec_and_test(&f->sk_ref))
1836 mutex_unlock(&fanout_mutex);
1841 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1842 struct sk_buff *skb)
1844 /* Earlier code assumed this would be a VLAN pkt, double-check
1845 * this now that we have the actual packet in hand. We can only
1846 * do this check on Ethernet devices.
1848 if (unlikely(dev->type != ARPHRD_ETHER))
1851 skb_reset_mac_header(skb);
1852 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1855 static const struct proto_ops packet_ops;
1857 static const struct proto_ops packet_ops_spkt;
1859 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1860 struct packet_type *pt, struct net_device *orig_dev)
1863 struct sockaddr_pkt *spkt;
1866 * When we registered the protocol we saved the socket in the data
1867 * field for just this event.
1870 sk = pt->af_packet_priv;
1873 * Yank back the headers [hope the device set this
1874 * right or kerboom...]
1876 * Incoming packets have ll header pulled,
1879 * For outgoing ones skb->data == skb_mac_header(skb)
1880 * so that this procedure is noop.
1883 if (skb->pkt_type == PACKET_LOOPBACK)
1886 if (!net_eq(dev_net(dev), sock_net(sk)))
1889 skb = skb_share_check(skb, GFP_ATOMIC);
1893 /* drop any routing info */
1896 /* drop conntrack reference */
1899 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1901 skb_push(skb, skb->data - skb_mac_header(skb));
1904 * The SOCK_PACKET socket receives _all_ frames.
1907 spkt->spkt_family = dev->type;
1908 strscpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1909 spkt->spkt_protocol = skb->protocol;
1912 * Charge the memory to the socket. This is done specifically
1913 * to prevent sockets using all the memory up.
1916 if (sock_queue_rcv_skb(sk, skb) == 0)
1925 static void packet_parse_headers(struct sk_buff *skb, struct socket *sock)
1929 if ((!skb->protocol || skb->protocol == htons(ETH_P_ALL)) &&
1930 sock->type == SOCK_RAW) {
1931 skb_reset_mac_header(skb);
1932 skb->protocol = dev_parse_header_protocol(skb);
1935 /* Move network header to the right position for VLAN tagged packets */
1936 if (likely(skb->dev->type == ARPHRD_ETHER) &&
1937 eth_type_vlan(skb->protocol) &&
1938 __vlan_get_protocol(skb, skb->protocol, &depth) != 0) {
1939 if (pskb_may_pull(skb, depth))
1940 skb_set_network_header(skb, depth);
1943 skb_probe_transport_header(skb);
1947 * Output a raw packet to a device layer. This bypasses all the other
1948 * protocol layers and you must therefore supply it with a complete frame
1951 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1954 struct sock *sk = sock->sk;
1955 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1956 struct sk_buff *skb = NULL;
1957 struct net_device *dev;
1958 struct sockcm_cookie sockc;
1964 * Get and verify the address.
1968 if (msg->msg_namelen < sizeof(struct sockaddr))
1970 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1971 proto = saddr->spkt_protocol;
1973 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1976 * Find the device first to size check it
1979 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1982 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1988 if (!(dev->flags & IFF_UP))
1992 * You may not queue a frame bigger than the mtu. This is the lowest level
1993 * raw protocol and you must do your own fragmentation at this level.
1996 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1997 if (!netif_supports_nofcs(dev)) {
1998 err = -EPROTONOSUPPORT;
2001 extra_len = 4; /* We're doing our own CRC */
2005 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
2009 size_t reserved = LL_RESERVED_SPACE(dev);
2010 int tlen = dev->needed_tailroom;
2011 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
2014 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
2017 /* FIXME: Save some space for broken drivers that write a hard
2018 * header at transmission time by themselves. PPP is the notable
2019 * one here. This should really be fixed at the driver level.
2021 skb_reserve(skb, reserved);
2022 skb_reset_network_header(skb);
2024 /* Try to align data part correctly */
2029 skb_reset_network_header(skb);
2031 err = memcpy_from_msg(skb_put(skb, len), msg, len);
2037 if (!dev_validate_header(dev, skb->data, len)) {
2041 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
2042 !packet_extra_vlan_len_allowed(dev, skb)) {
2047 sockcm_init(&sockc, sk);
2048 if (msg->msg_controllen) {
2049 err = sock_cmsg_send(sk, msg, &sockc);
2054 skb->protocol = proto;
2056 skb->priority = sk->sk_priority;
2057 skb->mark = sk->sk_mark;
2058 skb->tstamp = sockc.transmit_time;
2060 skb_setup_tx_timestamp(skb, sockc.tsflags);
2062 if (unlikely(extra_len == 4))
2065 packet_parse_headers(skb, sock);
2067 dev_queue_xmit(skb);
2078 static unsigned int run_filter(struct sk_buff *skb,
2079 const struct sock *sk,
2082 struct sk_filter *filter;
2085 filter = rcu_dereference(sk->sk_filter);
2087 res = bpf_prog_run_clear_cb(filter->prog, skb);
2093 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2096 struct virtio_net_hdr vnet_hdr;
2098 if (*len < sizeof(vnet_hdr))
2100 *len -= sizeof(vnet_hdr);
2102 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2105 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2109 * This function makes lazy skb cloning in hope that most of packets
2110 * are discarded by BPF.
2112 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2113 * and skb->cb are mangled. It works because (and until) packets
2114 * falling here are owned by current CPU. Output packets are cloned
2115 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2116 * sequentially, so that if we return skb to original state on exit,
2117 * we will not harm anyone.
2120 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2121 struct packet_type *pt, struct net_device *orig_dev)
2124 struct sockaddr_ll *sll;
2125 struct packet_sock *po;
2126 u8 *skb_head = skb->data;
2127 int skb_len = skb->len;
2128 unsigned int snaplen, res;
2129 bool is_drop_n_account = false;
2131 if (skb->pkt_type == PACKET_LOOPBACK)
2134 sk = pt->af_packet_priv;
2137 if (!net_eq(dev_net(dev), sock_net(sk)))
2142 if (dev_has_header(dev)) {
2143 /* The device has an explicit notion of ll header,
2144 * exported to higher levels.
2146 * Otherwise, the device hides details of its frame
2147 * structure, so that corresponding packet head is
2148 * never delivered to user.
2150 if (sk->sk_type != SOCK_DGRAM)
2151 skb_push(skb, skb->data - skb_mac_header(skb));
2152 else if (skb->pkt_type == PACKET_OUTGOING) {
2153 /* Special case: outgoing packets have ll header at head */
2154 skb_pull(skb, skb_network_offset(skb));
2160 res = run_filter(skb, sk, snaplen);
2162 goto drop_n_restore;
2166 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2169 if (skb_shared(skb)) {
2170 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2174 if (skb_head != skb->data) {
2175 skb->data = skb_head;
2182 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2184 sll = &PACKET_SKB_CB(skb)->sa.ll;
2185 sll->sll_hatype = dev->type;
2186 sll->sll_pkttype = skb->pkt_type;
2187 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2188 sll->sll_ifindex = orig_dev->ifindex;
2190 sll->sll_ifindex = dev->ifindex;
2192 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2194 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2195 * Use their space for storing the original skb length.
2197 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2199 if (pskb_trim(skb, snaplen))
2202 skb_set_owner_r(skb, sk);
2206 /* drop conntrack reference */
2209 spin_lock(&sk->sk_receive_queue.lock);
2210 po->stats.stats1.tp_packets++;
2211 sock_skb_set_dropcount(sk, skb);
2212 skb_clear_delivery_time(skb);
2213 __skb_queue_tail(&sk->sk_receive_queue, skb);
2214 spin_unlock(&sk->sk_receive_queue.lock);
2215 sk->sk_data_ready(sk);
2219 is_drop_n_account = true;
2220 atomic_inc(&po->tp_drops);
2221 atomic_inc(&sk->sk_drops);
2224 if (skb_head != skb->data && skb_shared(skb)) {
2225 skb->data = skb_head;
2229 if (!is_drop_n_account)
2236 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2237 struct packet_type *pt, struct net_device *orig_dev)
2240 struct packet_sock *po;
2241 struct sockaddr_ll *sll;
2242 union tpacket_uhdr h;
2243 u8 *skb_head = skb->data;
2244 int skb_len = skb->len;
2245 unsigned int snaplen, res;
2246 unsigned long status = TP_STATUS_USER;
2247 unsigned short macoff, hdrlen;
2248 unsigned int netoff;
2249 struct sk_buff *copy_skb = NULL;
2250 struct timespec64 ts;
2252 bool is_drop_n_account = false;
2253 unsigned int slot_id = 0;
2254 bool do_vnet = false;
2256 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2257 * We may add members to them until current aligned size without forcing
2258 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2260 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2261 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2263 if (skb->pkt_type == PACKET_LOOPBACK)
2266 sk = pt->af_packet_priv;
2269 if (!net_eq(dev_net(dev), sock_net(sk)))
2272 if (dev_has_header(dev)) {
2273 if (sk->sk_type != SOCK_DGRAM)
2274 skb_push(skb, skb->data - skb_mac_header(skb));
2275 else if (skb->pkt_type == PACKET_OUTGOING) {
2276 /* Special case: outgoing packets have ll header at head */
2277 skb_pull(skb, skb_network_offset(skb));
2283 res = run_filter(skb, sk, snaplen);
2285 goto drop_n_restore;
2287 /* If we are flooded, just give up */
2288 if (__packet_rcv_has_room(po, skb) == ROOM_NONE) {
2289 atomic_inc(&po->tp_drops);
2290 goto drop_n_restore;
2293 if (skb->ip_summed == CHECKSUM_PARTIAL)
2294 status |= TP_STATUS_CSUMNOTREADY;
2295 else if (skb->pkt_type != PACKET_OUTGOING &&
2296 skb_csum_unnecessary(skb))
2297 status |= TP_STATUS_CSUM_VALID;
2298 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
2299 status |= TP_STATUS_GSO_TCP;
2304 if (sk->sk_type == SOCK_DGRAM) {
2305 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2308 unsigned int maclen = skb_network_offset(skb);
2309 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2310 (maclen < 16 ? 16 : maclen)) +
2312 if (po->has_vnet_hdr) {
2313 netoff += sizeof(struct virtio_net_hdr);
2316 macoff = netoff - maclen;
2318 if (netoff > USHRT_MAX) {
2319 atomic_inc(&po->tp_drops);
2320 goto drop_n_restore;
2322 if (po->tp_version <= TPACKET_V2) {
2323 if (macoff + snaplen > po->rx_ring.frame_size) {
2324 if (po->copy_thresh &&
2325 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2326 if (skb_shared(skb)) {
2327 copy_skb = skb_clone(skb, GFP_ATOMIC);
2329 copy_skb = skb_get(skb);
2330 skb_head = skb->data;
2333 memset(&PACKET_SKB_CB(copy_skb)->sa.ll, 0,
2334 sizeof(PACKET_SKB_CB(copy_skb)->sa.ll));
2335 skb_set_owner_r(copy_skb, sk);
2338 snaplen = po->rx_ring.frame_size - macoff;
2339 if ((int)snaplen < 0) {
2344 } else if (unlikely(macoff + snaplen >
2345 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2348 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2349 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2350 snaplen, nval, macoff);
2352 if (unlikely((int)snaplen < 0)) {
2354 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2358 spin_lock(&sk->sk_receive_queue.lock);
2359 h.raw = packet_current_rx_frame(po, skb,
2360 TP_STATUS_KERNEL, (macoff+snaplen));
2362 goto drop_n_account;
2364 if (po->tp_version <= TPACKET_V2) {
2365 slot_id = po->rx_ring.head;
2366 if (test_bit(slot_id, po->rx_ring.rx_owner_map))
2367 goto drop_n_account;
2368 __set_bit(slot_id, po->rx_ring.rx_owner_map);
2372 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2373 sizeof(struct virtio_net_hdr),
2374 vio_le(), true, 0)) {
2375 if (po->tp_version == TPACKET_V3)
2376 prb_clear_blk_fill_status(&po->rx_ring);
2377 goto drop_n_account;
2380 if (po->tp_version <= TPACKET_V2) {
2381 packet_increment_rx_head(po, &po->rx_ring);
2383 * LOSING will be reported till you read the stats,
2384 * because it's COR - Clear On Read.
2385 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2388 if (atomic_read(&po->tp_drops))
2389 status |= TP_STATUS_LOSING;
2392 po->stats.stats1.tp_packets++;
2394 status |= TP_STATUS_COPY;
2395 skb_clear_delivery_time(copy_skb);
2396 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2398 spin_unlock(&sk->sk_receive_queue.lock);
2400 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2402 /* Always timestamp; prefer an existing software timestamp taken
2403 * closer to the time of capture.
2405 ts_status = tpacket_get_timestamp(skb, &ts,
2406 po->tp_tstamp | SOF_TIMESTAMPING_SOFTWARE);
2408 ktime_get_real_ts64(&ts);
2410 status |= ts_status;
2412 switch (po->tp_version) {
2414 h.h1->tp_len = skb->len;
2415 h.h1->tp_snaplen = snaplen;
2416 h.h1->tp_mac = macoff;
2417 h.h1->tp_net = netoff;
2418 h.h1->tp_sec = ts.tv_sec;
2419 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2420 hdrlen = sizeof(*h.h1);
2423 h.h2->tp_len = skb->len;
2424 h.h2->tp_snaplen = snaplen;
2425 h.h2->tp_mac = macoff;
2426 h.h2->tp_net = netoff;
2427 h.h2->tp_sec = ts.tv_sec;
2428 h.h2->tp_nsec = ts.tv_nsec;
2429 if (skb_vlan_tag_present(skb)) {
2430 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2431 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2432 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2434 h.h2->tp_vlan_tci = 0;
2435 h.h2->tp_vlan_tpid = 0;
2437 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2438 hdrlen = sizeof(*h.h2);
2441 /* tp_nxt_offset,vlan are already populated above.
2442 * So DONT clear those fields here
2444 h.h3->tp_status |= status;
2445 h.h3->tp_len = skb->len;
2446 h.h3->tp_snaplen = snaplen;
2447 h.h3->tp_mac = macoff;
2448 h.h3->tp_net = netoff;
2449 h.h3->tp_sec = ts.tv_sec;
2450 h.h3->tp_nsec = ts.tv_nsec;
2451 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2452 hdrlen = sizeof(*h.h3);
2458 sll = h.raw + TPACKET_ALIGN(hdrlen);
2459 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2460 sll->sll_family = AF_PACKET;
2461 sll->sll_hatype = dev->type;
2462 sll->sll_protocol = skb->protocol;
2463 sll->sll_pkttype = skb->pkt_type;
2464 if (unlikely(packet_sock_flag(po, PACKET_SOCK_ORIGDEV)))
2465 sll->sll_ifindex = orig_dev->ifindex;
2467 sll->sll_ifindex = dev->ifindex;
2471 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2472 if (po->tp_version <= TPACKET_V2) {
2475 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2478 for (start = h.raw; start < end; start += PAGE_SIZE)
2479 flush_dcache_page(pgv_to_page(start));
2484 if (po->tp_version <= TPACKET_V2) {
2485 spin_lock(&sk->sk_receive_queue.lock);
2486 __packet_set_status(po, h.raw, status);
2487 __clear_bit(slot_id, po->rx_ring.rx_owner_map);
2488 spin_unlock(&sk->sk_receive_queue.lock);
2489 sk->sk_data_ready(sk);
2490 } else if (po->tp_version == TPACKET_V3) {
2491 prb_clear_blk_fill_status(&po->rx_ring);
2495 if (skb_head != skb->data && skb_shared(skb)) {
2496 skb->data = skb_head;
2500 if (!is_drop_n_account)
2507 spin_unlock(&sk->sk_receive_queue.lock);
2508 atomic_inc(&po->tp_drops);
2509 is_drop_n_account = true;
2511 sk->sk_data_ready(sk);
2512 kfree_skb(copy_skb);
2513 goto drop_n_restore;
2516 static void tpacket_destruct_skb(struct sk_buff *skb)
2518 struct packet_sock *po = pkt_sk(skb->sk);
2520 if (likely(po->tx_ring.pg_vec)) {
2524 ph = skb_zcopy_get_nouarg(skb);
2525 packet_dec_pending(&po->tx_ring);
2527 ts = __packet_set_timestamp(po, ph, skb);
2528 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2530 if (!packet_read_pending(&po->tx_ring))
2531 complete(&po->skb_completion);
2537 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2539 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2540 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2541 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2542 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2543 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2544 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2545 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2547 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2553 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2554 struct virtio_net_hdr *vnet_hdr)
2556 if (*len < sizeof(*vnet_hdr))
2558 *len -= sizeof(*vnet_hdr);
2560 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2563 return __packet_snd_vnet_parse(vnet_hdr, *len);
2566 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2567 void *frame, struct net_device *dev, void *data, int tp_len,
2568 __be16 proto, unsigned char *addr, int hlen, int copylen,
2569 const struct sockcm_cookie *sockc)
2571 union tpacket_uhdr ph;
2572 int to_write, offset, len, nr_frags, len_max;
2573 struct socket *sock = po->sk.sk_socket;
2579 skb->protocol = proto;
2581 skb->priority = po->sk.sk_priority;
2582 skb->mark = po->sk.sk_mark;
2583 skb->tstamp = sockc->transmit_time;
2584 skb_setup_tx_timestamp(skb, sockc->tsflags);
2585 skb_zcopy_set_nouarg(skb, ph.raw);
2587 skb_reserve(skb, hlen);
2588 skb_reset_network_header(skb);
2592 if (sock->type == SOCK_DGRAM) {
2593 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2595 if (unlikely(err < 0))
2597 } else if (copylen) {
2598 int hdrlen = min_t(int, copylen, tp_len);
2600 skb_push(skb, dev->hard_header_len);
2601 skb_put(skb, copylen - dev->hard_header_len);
2602 err = skb_store_bits(skb, 0, data, hdrlen);
2605 if (!dev_validate_header(dev, skb->data, hdrlen))
2612 offset = offset_in_page(data);
2613 len_max = PAGE_SIZE - offset;
2614 len = ((to_write > len_max) ? len_max : to_write);
2616 skb->data_len = to_write;
2617 skb->len += to_write;
2618 skb->truesize += to_write;
2619 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2621 while (likely(to_write)) {
2622 nr_frags = skb_shinfo(skb)->nr_frags;
2624 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2625 pr_err("Packet exceed the number of skb frags(%lu)\n",
2630 page = pgv_to_page(data);
2632 flush_dcache_page(page);
2634 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2637 len_max = PAGE_SIZE;
2638 len = ((to_write > len_max) ? len_max : to_write);
2641 packet_parse_headers(skb, sock);
2646 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2647 int size_max, void **data)
2649 union tpacket_uhdr ph;
2654 switch (po->tp_version) {
2656 if (ph.h3->tp_next_offset != 0) {
2657 pr_warn_once("variable sized slot not supported");
2660 tp_len = ph.h3->tp_len;
2663 tp_len = ph.h2->tp_len;
2666 tp_len = ph.h1->tp_len;
2669 if (unlikely(tp_len > size_max)) {
2670 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2674 if (unlikely(po->tp_tx_has_off)) {
2675 int off_min, off_max;
2677 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2678 off_max = po->tx_ring.frame_size - tp_len;
2679 if (po->sk.sk_type == SOCK_DGRAM) {
2680 switch (po->tp_version) {
2682 off = ph.h3->tp_net;
2685 off = ph.h2->tp_net;
2688 off = ph.h1->tp_net;
2692 switch (po->tp_version) {
2694 off = ph.h3->tp_mac;
2697 off = ph.h2->tp_mac;
2700 off = ph.h1->tp_mac;
2704 if (unlikely((off < off_min) || (off_max < off)))
2707 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2710 *data = frame + off;
2714 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2716 struct sk_buff *skb = NULL;
2717 struct net_device *dev;
2718 struct virtio_net_hdr *vnet_hdr = NULL;
2719 struct sockcm_cookie sockc;
2721 int err, reserve = 0;
2723 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2724 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2725 unsigned char *addr = NULL;
2726 int tp_len, size_max;
2729 int status = TP_STATUS_AVAILABLE;
2730 int hlen, tlen, copylen = 0;
2733 mutex_lock(&po->pg_vec_lock);
2735 /* packet_sendmsg() check on tx_ring.pg_vec was lockless,
2736 * we need to confirm it under protection of pg_vec_lock.
2738 if (unlikely(!po->tx_ring.pg_vec)) {
2742 if (likely(saddr == NULL)) {
2743 dev = packet_cached_dev_get(po);
2744 proto = READ_ONCE(po->num);
2747 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2749 if (msg->msg_namelen < (saddr->sll_halen
2750 + offsetof(struct sockaddr_ll,
2753 proto = saddr->sll_protocol;
2754 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2755 if (po->sk.sk_socket->type == SOCK_DGRAM) {
2756 if (dev && msg->msg_namelen < dev->addr_len +
2757 offsetof(struct sockaddr_ll, sll_addr))
2759 addr = saddr->sll_addr;
2764 if (unlikely(dev == NULL))
2767 if (unlikely(!(dev->flags & IFF_UP)))
2770 sockcm_init(&sockc, &po->sk);
2771 if (msg->msg_controllen) {
2772 err = sock_cmsg_send(&po->sk, msg, &sockc);
2777 if (po->sk.sk_socket->type == SOCK_RAW)
2778 reserve = dev->hard_header_len;
2779 size_max = po->tx_ring.frame_size
2780 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2782 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2783 size_max = dev->mtu + reserve + VLAN_HLEN;
2785 reinit_completion(&po->skb_completion);
2788 ph = packet_current_frame(po, &po->tx_ring,
2789 TP_STATUS_SEND_REQUEST);
2790 if (unlikely(ph == NULL)) {
2791 if (need_wait && skb) {
2792 timeo = sock_sndtimeo(&po->sk, msg->msg_flags & MSG_DONTWAIT);
2793 timeo = wait_for_completion_interruptible_timeout(&po->skb_completion, timeo);
2795 err = !timeo ? -ETIMEDOUT : -ERESTARTSYS;
2799 /* check for additional frames */
2804 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2808 status = TP_STATUS_SEND_REQUEST;
2809 hlen = LL_RESERVED_SPACE(dev);
2810 tlen = dev->needed_tailroom;
2811 if (po->has_vnet_hdr) {
2813 data += sizeof(*vnet_hdr);
2814 tp_len -= sizeof(*vnet_hdr);
2816 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2820 copylen = __virtio16_to_cpu(vio_le(),
2823 copylen = max_t(int, copylen, dev->hard_header_len);
2824 skb = sock_alloc_send_skb(&po->sk,
2825 hlen + tlen + sizeof(struct sockaddr_ll) +
2826 (copylen - dev->hard_header_len),
2829 if (unlikely(skb == NULL)) {
2830 /* we assume the socket was initially writeable ... */
2831 if (likely(len_sum > 0))
2835 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2836 addr, hlen, copylen, &sockc);
2837 if (likely(tp_len >= 0) &&
2838 tp_len > dev->mtu + reserve &&
2839 !po->has_vnet_hdr &&
2840 !packet_extra_vlan_len_allowed(dev, skb))
2843 if (unlikely(tp_len < 0)) {
2846 __packet_set_status(po, ph,
2847 TP_STATUS_AVAILABLE);
2848 packet_increment_head(&po->tx_ring);
2852 status = TP_STATUS_WRONG_FORMAT;
2858 if (po->has_vnet_hdr) {
2859 if (virtio_net_hdr_to_skb(skb, vnet_hdr, vio_le())) {
2863 virtio_net_hdr_set_proto(skb, vnet_hdr);
2866 skb->destructor = tpacket_destruct_skb;
2867 __packet_set_status(po, ph, TP_STATUS_SENDING);
2868 packet_inc_pending(&po->tx_ring);
2870 status = TP_STATUS_SEND_REQUEST;
2871 /* Paired with WRITE_ONCE() in packet_setsockopt() */
2872 err = READ_ONCE(po->xmit)(skb);
2873 if (unlikely(err != 0)) {
2875 err = net_xmit_errno(err);
2876 if (err && __packet_get_status(po, ph) ==
2877 TP_STATUS_AVAILABLE) {
2878 /* skb was destructed already */
2883 * skb was dropped but not destructed yet;
2884 * let's treat it like congestion or err < 0
2888 packet_increment_head(&po->tx_ring);
2890 } while (likely((ph != NULL) ||
2891 /* Note: packet_read_pending() might be slow if we have
2892 * to call it as it's per_cpu variable, but in fast-path
2893 * we already short-circuit the loop with the first
2894 * condition, and luckily don't have to go that path
2897 (need_wait && packet_read_pending(&po->tx_ring))));
2903 __packet_set_status(po, ph, status);
2908 mutex_unlock(&po->pg_vec_lock);
2912 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2913 size_t reserve, size_t len,
2914 size_t linear, int noblock,
2917 struct sk_buff *skb;
2919 /* Under a page? Don't bother with paged skb. */
2920 if (prepad + len < PAGE_SIZE || !linear)
2923 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2928 skb_reserve(skb, reserve);
2929 skb_put(skb, linear);
2930 skb->data_len = len - linear;
2931 skb->len += len - linear;
2936 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2938 struct sock *sk = sock->sk;
2939 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2940 struct sk_buff *skb;
2941 struct net_device *dev;
2943 unsigned char *addr = NULL;
2944 int err, reserve = 0;
2945 struct sockcm_cookie sockc;
2946 struct virtio_net_hdr vnet_hdr = { 0 };
2948 struct packet_sock *po = pkt_sk(sk);
2949 bool has_vnet_hdr = false;
2950 int hlen, tlen, linear;
2954 * Get and verify the address.
2957 if (likely(saddr == NULL)) {
2958 dev = packet_cached_dev_get(po);
2959 proto = READ_ONCE(po->num);
2962 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2964 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2966 proto = saddr->sll_protocol;
2967 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2968 if (sock->type == SOCK_DGRAM) {
2969 if (dev && msg->msg_namelen < dev->addr_len +
2970 offsetof(struct sockaddr_ll, sll_addr))
2972 addr = saddr->sll_addr;
2977 if (unlikely(dev == NULL))
2980 if (unlikely(!(dev->flags & IFF_UP)))
2983 sockcm_init(&sockc, sk);
2984 sockc.mark = sk->sk_mark;
2985 if (msg->msg_controllen) {
2986 err = sock_cmsg_send(sk, msg, &sockc);
2991 if (sock->type == SOCK_RAW)
2992 reserve = dev->hard_header_len;
2993 if (po->has_vnet_hdr) {
2994 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2997 has_vnet_hdr = true;
3000 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
3001 if (!netif_supports_nofcs(dev)) {
3002 err = -EPROTONOSUPPORT;
3005 extra_len = 4; /* We're doing our own CRC */
3009 if (!vnet_hdr.gso_type &&
3010 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
3014 hlen = LL_RESERVED_SPACE(dev);
3015 tlen = dev->needed_tailroom;
3016 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
3017 linear = max(linear, min_t(int, len, dev->hard_header_len));
3018 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
3019 msg->msg_flags & MSG_DONTWAIT, &err);
3023 skb_reset_network_header(skb);
3026 if (sock->type == SOCK_DGRAM) {
3027 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
3028 if (unlikely(offset < 0))
3030 } else if (reserve) {
3031 skb_reserve(skb, -reserve);
3032 if (len < reserve + sizeof(struct ipv6hdr) &&
3033 dev->min_header_len != dev->hard_header_len)
3034 skb_reset_network_header(skb);
3037 /* Returns -EFAULT on error */
3038 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
3042 if ((sock->type == SOCK_RAW &&
3043 !dev_validate_header(dev, skb->data, len)) || !skb->len) {
3048 skb_setup_tx_timestamp(skb, sockc.tsflags);
3050 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
3051 !packet_extra_vlan_len_allowed(dev, skb)) {
3056 skb->protocol = proto;
3058 skb->priority = sk->sk_priority;
3059 skb->mark = sockc.mark;
3060 skb->tstamp = sockc.transmit_time;
3062 if (unlikely(extra_len == 4))
3065 packet_parse_headers(skb, sock);
3068 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
3071 len += sizeof(vnet_hdr);
3072 virtio_net_hdr_set_proto(skb, &vnet_hdr);
3075 /* Paired with WRITE_ONCE() in packet_setsockopt() */
3076 err = READ_ONCE(po->xmit)(skb);
3077 if (unlikely(err != 0)) {
3079 err = net_xmit_errno(err);
3096 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
3098 struct sock *sk = sock->sk;
3099 struct packet_sock *po = pkt_sk(sk);
3101 /* Reading tx_ring.pg_vec without holding pg_vec_lock is racy.
3102 * tpacket_snd() will redo the check safely.
3104 if (data_race(po->tx_ring.pg_vec))
3105 return tpacket_snd(po, msg);
3107 return packet_snd(sock, msg, len);
3111 * Close a PACKET socket. This is fairly simple. We immediately go
3112 * to 'closed' state and remove our protocol entry in the device list.
3115 static int packet_release(struct socket *sock)
3117 struct sock *sk = sock->sk;
3118 struct packet_sock *po;
3119 struct packet_fanout *f;
3121 union tpacket_req_u req_u;
3129 mutex_lock(&net->packet.sklist_lock);
3130 sk_del_node_init_rcu(sk);
3131 mutex_unlock(&net->packet.sklist_lock);
3133 sock_prot_inuse_add(net, sk->sk_prot, -1);
3135 spin_lock(&po->bind_lock);
3136 unregister_prot_hook(sk, false);
3137 packet_cached_dev_reset(po);
3139 if (po->prot_hook.dev) {
3140 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3141 po->prot_hook.dev = NULL;
3143 spin_unlock(&po->bind_lock);
3145 packet_flush_mclist(sk);
3148 if (po->rx_ring.pg_vec) {
3149 memset(&req_u, 0, sizeof(req_u));
3150 packet_set_ring(sk, &req_u, 1, 0);
3153 if (po->tx_ring.pg_vec) {
3154 memset(&req_u, 0, sizeof(req_u));
3155 packet_set_ring(sk, &req_u, 1, 1);
3159 f = fanout_release(sk);
3163 kfree(po->rollover);
3165 fanout_release_data(f);
3169 * Now the socket is dead. No more input will appear.
3176 skb_queue_purge(&sk->sk_receive_queue);
3177 packet_free_pending(po);
3184 * Attach a packet hook.
3187 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3190 struct packet_sock *po = pkt_sk(sk);
3191 struct net_device *dev = NULL;
3192 bool unlisted = false;
3197 spin_lock(&po->bind_lock);
3206 dev = dev_get_by_name_rcu(sock_net(sk), name);
3211 } else if (ifindex) {
3212 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3219 need_rehook = po->prot_hook.type != proto || po->prot_hook.dev != dev;
3225 /* prevents packet_notifier() from calling
3226 * register_prot_hook()
3228 WRITE_ONCE(po->num, 0);
3229 __unregister_prot_hook(sk, true);
3232 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3236 BUG_ON(po->running);
3237 WRITE_ONCE(po->num, proto);
3238 po->prot_hook.type = proto;
3240 netdev_put(po->prot_hook.dev, &po->prot_hook.dev_tracker);
3242 if (unlikely(unlisted)) {
3243 po->prot_hook.dev = NULL;
3244 WRITE_ONCE(po->ifindex, -1);
3245 packet_cached_dev_reset(po);
3247 netdev_hold(dev, &po->prot_hook.dev_tracker,
3249 po->prot_hook.dev = dev;
3250 WRITE_ONCE(po->ifindex, dev ? dev->ifindex : 0);
3251 packet_cached_dev_assign(po, dev);
3256 if (proto == 0 || !need_rehook)
3259 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3260 register_prot_hook(sk);
3262 sk->sk_err = ENETDOWN;
3263 if (!sock_flag(sk, SOCK_DEAD))
3264 sk_error_report(sk);
3269 spin_unlock(&po->bind_lock);
3275 * Bind a packet socket to a device
3278 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3281 struct sock *sk = sock->sk;
3282 char name[sizeof(uaddr->sa_data_min) + 1];
3288 if (addr_len != sizeof(struct sockaddr))
3290 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3293 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data_min));
3294 name[sizeof(uaddr->sa_data_min)] = 0;
3296 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3299 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3301 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3302 struct sock *sk = sock->sk;
3308 if (addr_len < sizeof(struct sockaddr_ll))
3310 if (sll->sll_family != AF_PACKET)
3313 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3314 sll->sll_protocol ? : pkt_sk(sk)->num);
3317 static struct proto packet_proto = {
3319 .owner = THIS_MODULE,
3320 .obj_size = sizeof(struct packet_sock),
3324 * Create a packet of type SOCK_PACKET.
3327 static int packet_create(struct net *net, struct socket *sock, int protocol,
3331 struct packet_sock *po;
3332 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3335 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3337 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3338 sock->type != SOCK_PACKET)
3339 return -ESOCKTNOSUPPORT;
3341 sock->state = SS_UNCONNECTED;
3344 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3348 sock->ops = &packet_ops;
3349 if (sock->type == SOCK_PACKET)
3350 sock->ops = &packet_ops_spkt;
3352 sock_init_data(sock, sk);
3355 init_completion(&po->skb_completion);
3356 sk->sk_family = PF_PACKET;
3358 po->xmit = dev_queue_xmit;
3360 err = packet_alloc_pending(po);
3364 packet_cached_dev_reset(po);
3366 sk->sk_destruct = packet_sock_destruct;
3369 * Attach a protocol block
3372 spin_lock_init(&po->bind_lock);
3373 mutex_init(&po->pg_vec_lock);
3374 po->rollover = NULL;
3375 po->prot_hook.func = packet_rcv;
3377 if (sock->type == SOCK_PACKET)
3378 po->prot_hook.func = packet_rcv_spkt;
3380 po->prot_hook.af_packet_priv = sk;
3381 po->prot_hook.af_packet_net = sock_net(sk);
3384 po->prot_hook.type = proto;
3385 __register_prot_hook(sk);
3388 mutex_lock(&net->packet.sklist_lock);
3389 sk_add_node_tail_rcu(sk, &net->packet.sklist);
3390 mutex_unlock(&net->packet.sklist_lock);
3392 sock_prot_inuse_add(net, &packet_proto, 1);
3402 * Pull a packet from our receive queue and hand it to the user.
3403 * If necessary we block.
3406 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3409 struct sock *sk = sock->sk;
3410 struct sk_buff *skb;
3412 int vnet_hdr_len = 0;
3413 unsigned int origlen = 0;
3416 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3420 /* What error should we return now? EUNATTACH? */
3421 if (pkt_sk(sk)->ifindex < 0)
3425 if (flags & MSG_ERRQUEUE) {
3426 err = sock_recv_errqueue(sk, msg, len,
3427 SOL_PACKET, PACKET_TX_TIMESTAMP);
3432 * Call the generic datagram receiver. This handles all sorts
3433 * of horrible races and re-entrancy so we can forget about it
3434 * in the protocol layers.
3436 * Now it will return ENETDOWN, if device have just gone down,
3437 * but then it will block.
3440 skb = skb_recv_datagram(sk, flags, &err);
3443 * An error occurred so return it. Because skb_recv_datagram()
3444 * handles the blocking we don't see and worry about blocking
3451 packet_rcv_try_clear_pressure(pkt_sk(sk));
3453 if (pkt_sk(sk)->has_vnet_hdr) {
3454 err = packet_rcv_vnet(msg, skb, &len);
3457 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3460 /* You lose any data beyond the buffer you gave. If it worries
3461 * a user program they can ask the device for its MTU
3467 msg->msg_flags |= MSG_TRUNC;
3470 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3474 if (sock->type != SOCK_PACKET) {
3475 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3477 /* Original length was stored in sockaddr_ll fields */
3478 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3479 sll->sll_family = AF_PACKET;
3480 sll->sll_protocol = skb->protocol;
3483 sock_recv_cmsgs(msg, sk, skb);
3485 if (msg->msg_name) {
3486 const size_t max_len = min(sizeof(skb->cb),
3487 sizeof(struct sockaddr_storage));
3490 /* If the address length field is there to be filled
3491 * in, we fill it in now.
3493 if (sock->type == SOCK_PACKET) {
3494 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3495 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3496 copy_len = msg->msg_namelen;
3498 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3500 msg->msg_namelen = sll->sll_halen +
3501 offsetof(struct sockaddr_ll, sll_addr);
3502 copy_len = msg->msg_namelen;
3503 if (msg->msg_namelen < sizeof(struct sockaddr_ll)) {
3504 memset(msg->msg_name +
3505 offsetof(struct sockaddr_ll, sll_addr),
3506 0, sizeof(sll->sll_addr));
3507 msg->msg_namelen = sizeof(struct sockaddr_ll);
3510 if (WARN_ON_ONCE(copy_len > max_len)) {
3512 msg->msg_namelen = copy_len;
3514 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa, copy_len);
3517 if (pkt_sk(sk)->auxdata) {
3518 struct tpacket_auxdata aux;
3520 aux.tp_status = TP_STATUS_USER;
3521 if (skb->ip_summed == CHECKSUM_PARTIAL)
3522 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3523 else if (skb->pkt_type != PACKET_OUTGOING &&
3524 skb_csum_unnecessary(skb))
3525 aux.tp_status |= TP_STATUS_CSUM_VALID;
3526 if (skb_is_gso(skb) && skb_is_gso_tcp(skb))
3527 aux.tp_status |= TP_STATUS_GSO_TCP;
3529 aux.tp_len = origlen;
3530 aux.tp_snaplen = skb->len;
3532 aux.tp_net = skb_network_offset(skb);
3533 if (skb_vlan_tag_present(skb)) {
3534 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3535 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3536 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3538 aux.tp_vlan_tci = 0;
3539 aux.tp_vlan_tpid = 0;
3541 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3545 * Free or return the buffer as appropriate. Again this
3546 * hides all the races and re-entrancy issues from us.
3548 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3551 skb_free_datagram(sk, skb);
3556 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3559 struct net_device *dev;
3560 struct sock *sk = sock->sk;
3565 uaddr->sa_family = AF_PACKET;
3566 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data_min));
3568 dev = dev_get_by_index_rcu(sock_net(sk), READ_ONCE(pkt_sk(sk)->ifindex));
3570 strscpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data_min));
3573 return sizeof(*uaddr);
3576 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3579 struct net_device *dev;
3580 struct sock *sk = sock->sk;
3581 struct packet_sock *po = pkt_sk(sk);
3582 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3588 ifindex = READ_ONCE(po->ifindex);
3589 sll->sll_family = AF_PACKET;
3590 sll->sll_ifindex = ifindex;
3591 sll->sll_protocol = READ_ONCE(po->num);
3592 sll->sll_pkttype = 0;
3594 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3596 sll->sll_hatype = dev->type;
3597 sll->sll_halen = dev->addr_len;
3598 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3600 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3605 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3608 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3612 case PACKET_MR_MULTICAST:
3613 if (i->alen != dev->addr_len)
3616 return dev_mc_add(dev, i->addr);
3618 return dev_mc_del(dev, i->addr);
3620 case PACKET_MR_PROMISC:
3621 return dev_set_promiscuity(dev, what);
3622 case PACKET_MR_ALLMULTI:
3623 return dev_set_allmulti(dev, what);
3624 case PACKET_MR_UNICAST:
3625 if (i->alen != dev->addr_len)
3628 return dev_uc_add(dev, i->addr);
3630 return dev_uc_del(dev, i->addr);
3638 static void packet_dev_mclist_delete(struct net_device *dev,
3639 struct packet_mclist **mlp)
3641 struct packet_mclist *ml;
3643 while ((ml = *mlp) != NULL) {
3644 if (ml->ifindex == dev->ifindex) {
3645 packet_dev_mc(dev, ml, -1);
3653 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3655 struct packet_sock *po = pkt_sk(sk);
3656 struct packet_mclist *ml, *i;
3657 struct net_device *dev;
3663 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3668 if (mreq->mr_alen > dev->addr_len)
3672 i = kmalloc(sizeof(*i), GFP_KERNEL);
3677 for (ml = po->mclist; ml; ml = ml->next) {
3678 if (ml->ifindex == mreq->mr_ifindex &&
3679 ml->type == mreq->mr_type &&
3680 ml->alen == mreq->mr_alen &&
3681 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3683 /* Free the new element ... */
3689 i->type = mreq->mr_type;
3690 i->ifindex = mreq->mr_ifindex;
3691 i->alen = mreq->mr_alen;
3692 memcpy(i->addr, mreq->mr_address, i->alen);
3693 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3695 i->next = po->mclist;
3697 err = packet_dev_mc(dev, i, 1);
3699 po->mclist = i->next;
3708 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3710 struct packet_mclist *ml, **mlp;
3714 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3715 if (ml->ifindex == mreq->mr_ifindex &&
3716 ml->type == mreq->mr_type &&
3717 ml->alen == mreq->mr_alen &&
3718 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3719 if (--ml->count == 0) {
3720 struct net_device *dev;
3722 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3724 packet_dev_mc(dev, ml, -1);
3734 static void packet_flush_mclist(struct sock *sk)
3736 struct packet_sock *po = pkt_sk(sk);
3737 struct packet_mclist *ml;
3743 while ((ml = po->mclist) != NULL) {
3744 struct net_device *dev;
3746 po->mclist = ml->next;
3747 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3749 packet_dev_mc(dev, ml, -1);
3756 packet_setsockopt(struct socket *sock, int level, int optname, sockptr_t optval,
3757 unsigned int optlen)
3759 struct sock *sk = sock->sk;
3760 struct packet_sock *po = pkt_sk(sk);
3763 if (level != SOL_PACKET)
3764 return -ENOPROTOOPT;
3767 case PACKET_ADD_MEMBERSHIP:
3768 case PACKET_DROP_MEMBERSHIP:
3770 struct packet_mreq_max mreq;
3772 memset(&mreq, 0, sizeof(mreq));
3773 if (len < sizeof(struct packet_mreq))
3775 if (len > sizeof(mreq))
3777 if (copy_from_sockptr(&mreq, optval, len))
3779 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3781 if (optname == PACKET_ADD_MEMBERSHIP)
3782 ret = packet_mc_add(sk, &mreq);
3784 ret = packet_mc_drop(sk, &mreq);
3788 case PACKET_RX_RING:
3789 case PACKET_TX_RING:
3791 union tpacket_req_u req_u;
3795 switch (po->tp_version) {
3798 len = sizeof(req_u.req);
3802 len = sizeof(req_u.req3);
3808 if (copy_from_sockptr(&req_u.req, optval, len))
3811 ret = packet_set_ring(sk, &req_u, 0,
3812 optname == PACKET_TX_RING);
3817 case PACKET_COPY_THRESH:
3821 if (optlen != sizeof(val))
3823 if (copy_from_sockptr(&val, optval, sizeof(val)))
3826 pkt_sk(sk)->copy_thresh = val;
3829 case PACKET_VERSION:
3833 if (optlen != sizeof(val))
3835 if (copy_from_sockptr(&val, optval, sizeof(val)))
3846 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3849 po->tp_version = val;
3855 case PACKET_RESERVE:
3859 if (optlen != sizeof(val))
3861 if (copy_from_sockptr(&val, optval, sizeof(val)))
3866 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3869 po->tp_reserve = val;
3879 if (optlen != sizeof(val))
3881 if (copy_from_sockptr(&val, optval, sizeof(val)))
3885 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3888 po->tp_loss = !!val;
3894 case PACKET_AUXDATA:
3898 if (optlen < sizeof(val))
3900 if (copy_from_sockptr(&val, optval, sizeof(val)))
3904 po->auxdata = !!val;
3908 case PACKET_ORIGDEV:
3912 if (optlen < sizeof(val))
3914 if (copy_from_sockptr(&val, optval, sizeof(val)))
3917 packet_sock_flag_set(po, PACKET_SOCK_ORIGDEV, val);
3920 case PACKET_VNET_HDR:
3924 if (sock->type != SOCK_RAW)
3926 if (optlen < sizeof(val))
3928 if (copy_from_sockptr(&val, optval, sizeof(val)))
3932 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3935 po->has_vnet_hdr = !!val;
3941 case PACKET_TIMESTAMP:
3945 if (optlen != sizeof(val))
3947 if (copy_from_sockptr(&val, optval, sizeof(val)))
3950 po->tp_tstamp = val;
3955 struct fanout_args args = { 0 };
3957 if (optlen != sizeof(int) && optlen != sizeof(args))
3959 if (copy_from_sockptr(&args, optval, optlen))
3962 return fanout_add(sk, &args);
3964 case PACKET_FANOUT_DATA:
3966 /* Paired with the WRITE_ONCE() in fanout_add() */
3967 if (!READ_ONCE(po->fanout))
3970 return fanout_set_data(po, optval, optlen);
3972 case PACKET_IGNORE_OUTGOING:
3976 if (optlen != sizeof(val))
3978 if (copy_from_sockptr(&val, optval, sizeof(val)))
3980 if (val < 0 || val > 1)
3983 po->prot_hook.ignore_outgoing = !!val;
3986 case PACKET_TX_HAS_OFF:
3990 if (optlen != sizeof(val))
3992 if (copy_from_sockptr(&val, optval, sizeof(val)))
3996 if (!po->rx_ring.pg_vec && !po->tx_ring.pg_vec)
3997 po->tp_tx_has_off = !!val;
4002 case PACKET_QDISC_BYPASS:
4006 if (optlen != sizeof(val))
4008 if (copy_from_sockptr(&val, optval, sizeof(val)))
4011 /* Paired with all lockless reads of po->xmit */
4012 WRITE_ONCE(po->xmit, val ? packet_direct_xmit : dev_queue_xmit);
4016 return -ENOPROTOOPT;
4020 static int packet_getsockopt(struct socket *sock, int level, int optname,
4021 char __user *optval, int __user *optlen)
4024 int val, lv = sizeof(val);
4025 struct sock *sk = sock->sk;
4026 struct packet_sock *po = pkt_sk(sk);
4028 union tpacket_stats_u st;
4029 struct tpacket_rollover_stats rstats;
4032 if (level != SOL_PACKET)
4033 return -ENOPROTOOPT;
4035 if (get_user(len, optlen))
4042 case PACKET_STATISTICS:
4043 spin_lock_bh(&sk->sk_receive_queue.lock);
4044 memcpy(&st, &po->stats, sizeof(st));
4045 memset(&po->stats, 0, sizeof(po->stats));
4046 spin_unlock_bh(&sk->sk_receive_queue.lock);
4047 drops = atomic_xchg(&po->tp_drops, 0);
4049 if (po->tp_version == TPACKET_V3) {
4050 lv = sizeof(struct tpacket_stats_v3);
4051 st.stats3.tp_drops = drops;
4052 st.stats3.tp_packets += drops;
4055 lv = sizeof(struct tpacket_stats);
4056 st.stats1.tp_drops = drops;
4057 st.stats1.tp_packets += drops;
4062 case PACKET_AUXDATA:
4065 case PACKET_ORIGDEV:
4066 val = packet_sock_flag(po, PACKET_SOCK_ORIGDEV);
4068 case PACKET_VNET_HDR:
4069 val = po->has_vnet_hdr;
4071 case PACKET_VERSION:
4072 val = po->tp_version;
4075 if (len > sizeof(int))
4077 if (len < sizeof(int))
4079 if (copy_from_user(&val, optval, len))
4083 val = sizeof(struct tpacket_hdr);
4086 val = sizeof(struct tpacket2_hdr);
4089 val = sizeof(struct tpacket3_hdr);
4095 case PACKET_RESERVE:
4096 val = po->tp_reserve;
4101 case PACKET_TIMESTAMP:
4102 val = po->tp_tstamp;
4106 ((u32)po->fanout->id |
4107 ((u32)po->fanout->type << 16) |
4108 ((u32)po->fanout->flags << 24)) :
4111 case PACKET_IGNORE_OUTGOING:
4112 val = po->prot_hook.ignore_outgoing;
4114 case PACKET_ROLLOVER_STATS:
4117 rstats.tp_all = atomic_long_read(&po->rollover->num);
4118 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
4119 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
4121 lv = sizeof(rstats);
4123 case PACKET_TX_HAS_OFF:
4124 val = po->tp_tx_has_off;
4126 case PACKET_QDISC_BYPASS:
4127 val = packet_use_direct_xmit(po);
4130 return -ENOPROTOOPT;
4135 if (put_user(len, optlen))
4137 if (copy_to_user(optval, data, len))
4142 static int packet_notifier(struct notifier_block *this,
4143 unsigned long msg, void *ptr)
4146 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
4147 struct net *net = dev_net(dev);
4150 sk_for_each_rcu(sk, &net->packet.sklist) {
4151 struct packet_sock *po = pkt_sk(sk);
4154 case NETDEV_UNREGISTER:
4156 packet_dev_mclist_delete(dev, &po->mclist);
4160 if (dev->ifindex == po->ifindex) {
4161 spin_lock(&po->bind_lock);
4163 __unregister_prot_hook(sk, false);
4164 sk->sk_err = ENETDOWN;
4165 if (!sock_flag(sk, SOCK_DEAD))
4166 sk_error_report(sk);
4168 if (msg == NETDEV_UNREGISTER) {
4169 packet_cached_dev_reset(po);
4170 WRITE_ONCE(po->ifindex, -1);
4171 netdev_put(po->prot_hook.dev,
4172 &po->prot_hook.dev_tracker);
4173 po->prot_hook.dev = NULL;
4175 spin_unlock(&po->bind_lock);
4179 if (dev->ifindex == po->ifindex) {
4180 spin_lock(&po->bind_lock);
4182 register_prot_hook(sk);
4183 spin_unlock(&po->bind_lock);
4193 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4196 struct sock *sk = sock->sk;
4201 int amount = sk_wmem_alloc_get(sk);
4203 return put_user(amount, (int __user *)arg);
4207 struct sk_buff *skb;
4210 spin_lock_bh(&sk->sk_receive_queue.lock);
4211 skb = skb_peek(&sk->sk_receive_queue);
4214 spin_unlock_bh(&sk->sk_receive_queue.lock);
4215 return put_user(amount, (int __user *)arg);
4225 case SIOCGIFBRDADDR:
4226 case SIOCSIFBRDADDR:
4227 case SIOCGIFNETMASK:
4228 case SIOCSIFNETMASK:
4229 case SIOCGIFDSTADDR:
4230 case SIOCSIFDSTADDR:
4232 return inet_dgram_ops.ioctl(sock, cmd, arg);
4236 return -ENOIOCTLCMD;
4241 static __poll_t packet_poll(struct file *file, struct socket *sock,
4244 struct sock *sk = sock->sk;
4245 struct packet_sock *po = pkt_sk(sk);
4246 __poll_t mask = datagram_poll(file, sock, wait);
4248 spin_lock_bh(&sk->sk_receive_queue.lock);
4249 if (po->rx_ring.pg_vec) {
4250 if (!packet_previous_rx_frame(po, &po->rx_ring,
4252 mask |= EPOLLIN | EPOLLRDNORM;
4254 packet_rcv_try_clear_pressure(po);
4255 spin_unlock_bh(&sk->sk_receive_queue.lock);
4256 spin_lock_bh(&sk->sk_write_queue.lock);
4257 if (po->tx_ring.pg_vec) {
4258 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4259 mask |= EPOLLOUT | EPOLLWRNORM;
4261 spin_unlock_bh(&sk->sk_write_queue.lock);
4266 /* Dirty? Well, I still did not learn better way to account
4270 static void packet_mm_open(struct vm_area_struct *vma)
4272 struct file *file = vma->vm_file;
4273 struct socket *sock = file->private_data;
4274 struct sock *sk = sock->sk;
4277 atomic_inc(&pkt_sk(sk)->mapped);
4280 static void packet_mm_close(struct vm_area_struct *vma)
4282 struct file *file = vma->vm_file;
4283 struct socket *sock = file->private_data;
4284 struct sock *sk = sock->sk;
4287 atomic_dec(&pkt_sk(sk)->mapped);
4290 static const struct vm_operations_struct packet_mmap_ops = {
4291 .open = packet_mm_open,
4292 .close = packet_mm_close,
4295 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4300 for (i = 0; i < len; i++) {
4301 if (likely(pg_vec[i].buffer)) {
4302 if (is_vmalloc_addr(pg_vec[i].buffer))
4303 vfree(pg_vec[i].buffer);
4305 free_pages((unsigned long)pg_vec[i].buffer,
4307 pg_vec[i].buffer = NULL;
4313 static char *alloc_one_pg_vec_page(unsigned long order)
4316 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4317 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4319 buffer = (char *) __get_free_pages(gfp_flags, order);
4323 /* __get_free_pages failed, fall back to vmalloc */
4324 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4328 /* vmalloc failed, lets dig into swap here */
4329 gfp_flags &= ~__GFP_NORETRY;
4330 buffer = (char *) __get_free_pages(gfp_flags, order);
4334 /* complete and utter failure */
4338 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4340 unsigned int block_nr = req->tp_block_nr;
4344 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL | __GFP_NOWARN);
4345 if (unlikely(!pg_vec))
4348 for (i = 0; i < block_nr; i++) {
4349 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4350 if (unlikely(!pg_vec[i].buffer))
4351 goto out_free_pgvec;
4358 free_pg_vec(pg_vec, order, block_nr);
4363 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4364 int closing, int tx_ring)
4366 struct pgv *pg_vec = NULL;
4367 struct packet_sock *po = pkt_sk(sk);
4368 unsigned long *rx_owner_map = NULL;
4369 int was_running, order = 0;
4370 struct packet_ring_buffer *rb;
4371 struct sk_buff_head *rb_queue;
4374 /* Added to avoid minimal code churn */
4375 struct tpacket_req *req = &req_u->req;
4377 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4378 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4382 if (atomic_read(&po->mapped))
4384 if (packet_read_pending(rb))
4388 if (req->tp_block_nr) {
4389 unsigned int min_frame_size;
4391 /* Sanity tests and some calculations */
4393 if (unlikely(rb->pg_vec))
4396 switch (po->tp_version) {
4398 po->tp_hdrlen = TPACKET_HDRLEN;
4401 po->tp_hdrlen = TPACKET2_HDRLEN;
4404 po->tp_hdrlen = TPACKET3_HDRLEN;
4409 if (unlikely((int)req->tp_block_size <= 0))
4411 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4413 min_frame_size = po->tp_hdrlen + po->tp_reserve;
4414 if (po->tp_version >= TPACKET_V3 &&
4415 req->tp_block_size <
4416 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + min_frame_size)
4418 if (unlikely(req->tp_frame_size < min_frame_size))
4420 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4423 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4424 if (unlikely(rb->frames_per_block == 0))
4426 if (unlikely(rb->frames_per_block > UINT_MAX / req->tp_block_nr))
4428 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4433 order = get_order(req->tp_block_size);
4434 pg_vec = alloc_pg_vec(req, order);
4435 if (unlikely(!pg_vec))
4437 switch (po->tp_version) {
4439 /* Block transmit is not supported yet */
4441 init_prb_bdqc(po, rb, pg_vec, req_u);
4443 struct tpacket_req3 *req3 = &req_u->req3;
4445 if (req3->tp_retire_blk_tov ||
4446 req3->tp_sizeof_priv ||
4447 req3->tp_feature_req_word) {
4449 goto out_free_pg_vec;
4455 rx_owner_map = bitmap_alloc(req->tp_frame_nr,
4456 GFP_KERNEL | __GFP_NOWARN | __GFP_ZERO);
4458 goto out_free_pg_vec;
4466 if (unlikely(req->tp_frame_nr))
4471 /* Detach socket from network */
4472 spin_lock(&po->bind_lock);
4473 was_running = po->running;
4476 WRITE_ONCE(po->num, 0);
4477 __unregister_prot_hook(sk, false);
4479 spin_unlock(&po->bind_lock);
4484 mutex_lock(&po->pg_vec_lock);
4485 if (closing || atomic_read(&po->mapped) == 0) {
4487 spin_lock_bh(&rb_queue->lock);
4488 swap(rb->pg_vec, pg_vec);
4489 if (po->tp_version <= TPACKET_V2)
4490 swap(rb->rx_owner_map, rx_owner_map);
4491 rb->frame_max = (req->tp_frame_nr - 1);
4493 rb->frame_size = req->tp_frame_size;
4494 spin_unlock_bh(&rb_queue->lock);
4496 swap(rb->pg_vec_order, order);
4497 swap(rb->pg_vec_len, req->tp_block_nr);
4499 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4500 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4501 tpacket_rcv : packet_rcv;
4502 skb_queue_purge(rb_queue);
4503 if (atomic_read(&po->mapped))
4504 pr_err("packet_mmap: vma is busy: %d\n",
4505 atomic_read(&po->mapped));
4507 mutex_unlock(&po->pg_vec_lock);
4509 spin_lock(&po->bind_lock);
4511 WRITE_ONCE(po->num, num);
4512 register_prot_hook(sk);
4514 spin_unlock(&po->bind_lock);
4515 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4516 /* Because we don't support block-based V3 on tx-ring */
4518 prb_shutdown_retire_blk_timer(po, rb_queue);
4523 bitmap_free(rx_owner_map);
4524 free_pg_vec(pg_vec, order, req->tp_block_nr);
4530 static int packet_mmap(struct file *file, struct socket *sock,
4531 struct vm_area_struct *vma)
4533 struct sock *sk = sock->sk;
4534 struct packet_sock *po = pkt_sk(sk);
4535 unsigned long size, expected_size;
4536 struct packet_ring_buffer *rb;
4537 unsigned long start;
4544 mutex_lock(&po->pg_vec_lock);
4547 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4549 expected_size += rb->pg_vec_len
4555 if (expected_size == 0)
4558 size = vma->vm_end - vma->vm_start;
4559 if (size != expected_size)
4562 start = vma->vm_start;
4563 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4564 if (rb->pg_vec == NULL)
4567 for (i = 0; i < rb->pg_vec_len; i++) {
4569 void *kaddr = rb->pg_vec[i].buffer;
4572 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4573 page = pgv_to_page(kaddr);
4574 err = vm_insert_page(vma, start, page);
4583 atomic_inc(&po->mapped);
4584 vma->vm_ops = &packet_mmap_ops;
4588 mutex_unlock(&po->pg_vec_lock);
4592 static const struct proto_ops packet_ops_spkt = {
4593 .family = PF_PACKET,
4594 .owner = THIS_MODULE,
4595 .release = packet_release,
4596 .bind = packet_bind_spkt,
4597 .connect = sock_no_connect,
4598 .socketpair = sock_no_socketpair,
4599 .accept = sock_no_accept,
4600 .getname = packet_getname_spkt,
4601 .poll = datagram_poll,
4602 .ioctl = packet_ioctl,
4603 .gettstamp = sock_gettstamp,
4604 .listen = sock_no_listen,
4605 .shutdown = sock_no_shutdown,
4606 .sendmsg = packet_sendmsg_spkt,
4607 .recvmsg = packet_recvmsg,
4608 .mmap = sock_no_mmap,
4609 .sendpage = sock_no_sendpage,
4612 static const struct proto_ops packet_ops = {
4613 .family = PF_PACKET,
4614 .owner = THIS_MODULE,
4615 .release = packet_release,
4616 .bind = packet_bind,
4617 .connect = sock_no_connect,
4618 .socketpair = sock_no_socketpair,
4619 .accept = sock_no_accept,
4620 .getname = packet_getname,
4621 .poll = packet_poll,
4622 .ioctl = packet_ioctl,
4623 .gettstamp = sock_gettstamp,
4624 .listen = sock_no_listen,
4625 .shutdown = sock_no_shutdown,
4626 .setsockopt = packet_setsockopt,
4627 .getsockopt = packet_getsockopt,
4628 .sendmsg = packet_sendmsg,
4629 .recvmsg = packet_recvmsg,
4630 .mmap = packet_mmap,
4631 .sendpage = sock_no_sendpage,
4634 static const struct net_proto_family packet_family_ops = {
4635 .family = PF_PACKET,
4636 .create = packet_create,
4637 .owner = THIS_MODULE,
4640 static struct notifier_block packet_netdev_notifier = {
4641 .notifier_call = packet_notifier,
4644 #ifdef CONFIG_PROC_FS
4646 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4649 struct net *net = seq_file_net(seq);
4652 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4655 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4657 struct net *net = seq_file_net(seq);
4658 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4661 static void packet_seq_stop(struct seq_file *seq, void *v)
4667 static int packet_seq_show(struct seq_file *seq, void *v)
4669 if (v == SEQ_START_TOKEN)
4671 "%*sRefCnt Type Proto Iface R Rmem User Inode\n",
4672 IS_ENABLED(CONFIG_64BIT) ? -17 : -9, "sk");
4674 struct sock *s = sk_entry(v);
4675 const struct packet_sock *po = pkt_sk(s);
4678 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4680 refcount_read(&s->sk_refcnt),
4682 ntohs(READ_ONCE(po->num)),
4683 READ_ONCE(po->ifindex),
4685 atomic_read(&s->sk_rmem_alloc),
4686 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4693 static const struct seq_operations packet_seq_ops = {
4694 .start = packet_seq_start,
4695 .next = packet_seq_next,
4696 .stop = packet_seq_stop,
4697 .show = packet_seq_show,
4701 static int __net_init packet_net_init(struct net *net)
4703 mutex_init(&net->packet.sklist_lock);
4704 INIT_HLIST_HEAD(&net->packet.sklist);
4706 #ifdef CONFIG_PROC_FS
4707 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4708 sizeof(struct seq_net_private)))
4710 #endif /* CONFIG_PROC_FS */
4715 static void __net_exit packet_net_exit(struct net *net)
4717 remove_proc_entry("packet", net->proc_net);
4718 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4721 static struct pernet_operations packet_net_ops = {
4722 .init = packet_net_init,
4723 .exit = packet_net_exit,
4727 static void __exit packet_exit(void)
4729 sock_unregister(PF_PACKET);
4730 proto_unregister(&packet_proto);
4731 unregister_netdevice_notifier(&packet_netdev_notifier);
4732 unregister_pernet_subsys(&packet_net_ops);
4735 static int __init packet_init(void)
4739 rc = register_pernet_subsys(&packet_net_ops);
4742 rc = register_netdevice_notifier(&packet_netdev_notifier);
4745 rc = proto_register(&packet_proto, 0);
4748 rc = sock_register(&packet_family_ops);
4755 proto_unregister(&packet_proto);
4757 unregister_netdevice_notifier(&packet_netdev_notifier);
4759 unregister_pernet_subsys(&packet_net_ops);
4764 module_init(packet_init);
4765 module_exit(packet_exit);
4766 MODULE_LICENSE("GPL");
4767 MODULE_ALIAS_NETPROTO(PF_PACKET);