2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PACKET - implements raw packet sockets.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Alan Cox, <gw4pts@gw4pts.ampr.org>
13 * Alan Cox : verify_area() now used correctly
14 * Alan Cox : new skbuff lists, look ma no backlogs!
15 * Alan Cox : tidied skbuff lists.
16 * Alan Cox : Now uses generic datagram routines I
17 * added. Also fixed the peek/read crash
18 * from all old Linux datagram code.
19 * Alan Cox : Uses the improved datagram code.
20 * Alan Cox : Added NULL's for socket options.
21 * Alan Cox : Re-commented the code.
22 * Alan Cox : Use new kernel side addressing
23 * Rob Janssen : Correct MTU usage.
24 * Dave Platt : Counter leaks caused by incorrect
25 * interrupt locking and some slightly
26 * dubious gcc output. Can you read
27 * compiler: it said _VOLATILE_
28 * Richard Kooijman : Timestamp fixes.
29 * Alan Cox : New buffers. Use sk->mac.raw.
30 * Alan Cox : sendmsg/recvmsg support.
31 * Alan Cox : Protocol setting support
32 * Alexey Kuznetsov : Untied from IPv4 stack.
33 * Cyrus Durgin : Fixed kerneld for kmod.
34 * Michal Ostrowski : Module initialization cleanup.
35 * Ulises Alonso : Frame number limit removal and
36 * packet_set_ring memory leak.
37 * Eric Biederman : Allow for > 8 byte hardware addresses.
38 * The convention is that longer addresses
39 * will simply extend the hardware address
40 * byte arrays at the end of sockaddr_ll
42 * Johann Baudy : Added TX RING.
43 * Chetan Loke : Implemented TPACKET_V3 block abstraction
45 * Copyright (C) 2011, <lokec@ccs.neu.edu>
48 * This program is free software; you can redistribute it and/or
49 * modify it under the terms of the GNU General Public License
50 * as published by the Free Software Foundation; either version
51 * 2 of the License, or (at your option) any later version.
55 #include <linux/types.h>
57 #include <linux/capability.h>
58 #include <linux/fcntl.h>
59 #include <linux/socket.h>
61 #include <linux/inet.h>
62 #include <linux/netdevice.h>
63 #include <linux/if_packet.h>
64 #include <linux/wireless.h>
65 #include <linux/kernel.h>
66 #include <linux/kmod.h>
67 #include <linux/slab.h>
68 #include <linux/vmalloc.h>
69 #include <net/net_namespace.h>
71 #include <net/protocol.h>
72 #include <linux/skbuff.h>
74 #include <linux/errno.h>
75 #include <linux/timer.h>
76 #include <linux/uaccess.h>
77 #include <asm/ioctls.h>
79 #include <asm/cacheflush.h>
81 #include <linux/proc_fs.h>
82 #include <linux/seq_file.h>
83 #include <linux/poll.h>
84 #include <linux/module.h>
85 #include <linux/init.h>
86 #include <linux/mutex.h>
87 #include <linux/if_vlan.h>
88 #include <linux/virtio_net.h>
89 #include <linux/errqueue.h>
90 #include <linux/net_tstamp.h>
91 #include <linux/percpu.h>
93 #include <net/inet_common.h>
95 #include <linux/bpf.h>
96 #include <net/compat.h>
102 - if device has no dev->hard_header routine, it adds and removes ll header
103 inside itself. In this case ll header is invisible outside of device,
104 but higher levels still should reserve dev->hard_header_len.
105 Some devices are enough clever to reallocate skb, when header
106 will not fit to reserved space (tunnel), another ones are silly
108 - packet socket receives packets with pulled ll header,
109 so that SOCK_RAW should push it back.
114 Incoming, dev->hard_header!=NULL
115 mac_header -> ll header
118 Outgoing, dev->hard_header!=NULL
119 mac_header -> ll header
122 Incoming, dev->hard_header==NULL
123 mac_header -> UNKNOWN position. It is very likely, that it points to ll
124 header. PPP makes it, that is wrong, because introduce
125 assymetry between rx and tx paths.
128 Outgoing, dev->hard_header==NULL
129 mac_header -> data. ll header is still not built!
133 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
139 dev->hard_header != NULL
140 mac_header -> ll header
143 dev->hard_header == NULL (ll header is added by device, we cannot control it)
147 We should set nh.raw on output to correct posistion,
148 packet classifier depends on it.
151 /* Private packet socket structures. */
153 /* identical to struct packet_mreq except it has
154 * a longer address field.
156 struct packet_mreq_max {
158 unsigned short mr_type;
159 unsigned short mr_alen;
160 unsigned char mr_address[MAX_ADDR_LEN];
164 struct tpacket_hdr *h1;
165 struct tpacket2_hdr *h2;
166 struct tpacket3_hdr *h3;
170 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
171 int closing, int tx_ring);
173 #define V3_ALIGNMENT (8)
175 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
177 #define BLK_PLUS_PRIV(sz_of_priv) \
178 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
190 struct packet_type *pt, struct net_device *orig_dev);
192 static void *packet_previous_frame(struct packet_sock *po,
193 struct packet_ring_buffer *rb,
195 static void packet_increment_head(struct packet_ring_buffer *buff);
196 static int prb_curr_blk_in_use(struct tpacket_block_desc *);
197 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
198 struct packet_sock *);
199 static void prb_retire_current_block(struct tpacket_kbdq_core *,
200 struct packet_sock *, unsigned int status);
201 static int prb_queue_frozen(struct tpacket_kbdq_core *);
202 static void prb_open_block(struct tpacket_kbdq_core *,
203 struct tpacket_block_desc *);
204 static void prb_retire_rx_blk_timer_expired(struct timer_list *);
205 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
206 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
207 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
208 struct tpacket3_hdr *);
209 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
210 struct tpacket3_hdr *);
211 static void packet_flush_mclist(struct sock *sk);
212 static u16 packet_pick_tx_queue(struct sk_buff *skb);
214 struct packet_skb_cb {
216 struct sockaddr_pkt pkt;
218 /* Trick: alias skb original length with
219 * ll.sll_family and ll.protocol in order
222 unsigned int origlen;
223 struct sockaddr_ll ll;
228 #define vio_le() virtio_legacy_is_little_endian()
230 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
232 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
233 #define GET_PBLOCK_DESC(x, bid) \
234 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
235 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
237 #define GET_NEXT_PRB_BLK_NUM(x) \
238 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
239 ((x)->kactive_blk_num+1) : 0)
241 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
242 static void __fanout_link(struct sock *sk, struct packet_sock *po);
244 static int packet_direct_xmit(struct sk_buff *skb)
246 return dev_direct_xmit(skb, packet_pick_tx_queue(skb));
249 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
251 struct net_device *dev;
254 dev = rcu_dereference(po->cached_dev);
262 static void packet_cached_dev_assign(struct packet_sock *po,
263 struct net_device *dev)
265 rcu_assign_pointer(po->cached_dev, dev);
268 static void packet_cached_dev_reset(struct packet_sock *po)
270 RCU_INIT_POINTER(po->cached_dev, NULL);
273 static bool packet_use_direct_xmit(const struct packet_sock *po)
275 return po->xmit == packet_direct_xmit;
278 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
280 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
283 static u16 packet_pick_tx_queue(struct sk_buff *skb)
285 struct net_device *dev = skb->dev;
286 const struct net_device_ops *ops = dev->netdev_ops;
289 if (ops->ndo_select_queue) {
290 queue_index = ops->ndo_select_queue(dev, skb, NULL,
291 __packet_pick_tx_queue);
292 queue_index = netdev_cap_txqueue(dev, queue_index);
294 queue_index = __packet_pick_tx_queue(dev, skb);
300 /* __register_prot_hook must be invoked through register_prot_hook
301 * or from a context in which asynchronous accesses to the packet
302 * socket is not possible (packet_create()).
304 static void __register_prot_hook(struct sock *sk)
306 struct packet_sock *po = pkt_sk(sk);
310 __fanout_link(sk, po);
312 dev_add_pack(&po->prot_hook);
319 static void register_prot_hook(struct sock *sk)
321 lockdep_assert_held_once(&pkt_sk(sk)->bind_lock);
322 __register_prot_hook(sk);
325 /* If the sync parameter is true, we will temporarily drop
326 * the po->bind_lock and do a synchronize_net to make sure no
327 * asynchronous packet processing paths still refer to the elements
328 * of po->prot_hook. If the sync parameter is false, it is the
329 * callers responsibility to take care of this.
331 static void __unregister_prot_hook(struct sock *sk, bool sync)
333 struct packet_sock *po = pkt_sk(sk);
335 lockdep_assert_held_once(&po->bind_lock);
340 __fanout_unlink(sk, po);
342 __dev_remove_pack(&po->prot_hook);
347 spin_unlock(&po->bind_lock);
349 spin_lock(&po->bind_lock);
353 static void unregister_prot_hook(struct sock *sk, bool sync)
355 struct packet_sock *po = pkt_sk(sk);
358 __unregister_prot_hook(sk, sync);
361 static inline struct page * __pure pgv_to_page(void *addr)
363 if (is_vmalloc_addr(addr))
364 return vmalloc_to_page(addr);
365 return virt_to_page(addr);
368 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
370 union tpacket_uhdr h;
373 switch (po->tp_version) {
375 h.h1->tp_status = status;
376 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
379 h.h2->tp_status = status;
380 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
383 h.h3->tp_status = status;
384 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
387 WARN(1, "TPACKET version not supported.\n");
394 static int __packet_get_status(struct packet_sock *po, void *frame)
396 union tpacket_uhdr h;
401 switch (po->tp_version) {
403 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
404 return h.h1->tp_status;
406 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
407 return h.h2->tp_status;
409 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
410 return h.h3->tp_status;
412 WARN(1, "TPACKET version not supported.\n");
418 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
421 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
424 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
425 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
426 return TP_STATUS_TS_RAW_HARDWARE;
428 if (ktime_to_timespec_cond(skb->tstamp, ts))
429 return TP_STATUS_TS_SOFTWARE;
434 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
437 union tpacket_uhdr h;
441 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
445 switch (po->tp_version) {
447 h.h1->tp_sec = ts.tv_sec;
448 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
451 h.h2->tp_sec = ts.tv_sec;
452 h.h2->tp_nsec = ts.tv_nsec;
455 h.h3->tp_sec = ts.tv_sec;
456 h.h3->tp_nsec = ts.tv_nsec;
459 WARN(1, "TPACKET version not supported.\n");
463 /* one flush is safe, as both fields always lie on the same cacheline */
464 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
470 static void *packet_lookup_frame(struct packet_sock *po,
471 struct packet_ring_buffer *rb,
472 unsigned int position,
475 unsigned int pg_vec_pos, frame_offset;
476 union tpacket_uhdr h;
478 pg_vec_pos = position / rb->frames_per_block;
479 frame_offset = position % rb->frames_per_block;
481 h.raw = rb->pg_vec[pg_vec_pos].buffer +
482 (frame_offset * rb->frame_size);
484 if (status != __packet_get_status(po, h.raw))
490 static void *packet_current_frame(struct packet_sock *po,
491 struct packet_ring_buffer *rb,
494 return packet_lookup_frame(po, rb, rb->head, status);
497 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
499 del_timer_sync(&pkc->retire_blk_timer);
502 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
503 struct sk_buff_head *rb_queue)
505 struct tpacket_kbdq_core *pkc;
507 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
509 spin_lock_bh(&rb_queue->lock);
510 pkc->delete_blk_timer = 1;
511 spin_unlock_bh(&rb_queue->lock);
513 prb_del_retire_blk_timer(pkc);
516 static void prb_setup_retire_blk_timer(struct packet_sock *po)
518 struct tpacket_kbdq_core *pkc;
520 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
521 timer_setup(&pkc->retire_blk_timer, prb_retire_rx_blk_timer_expired,
523 pkc->retire_blk_timer.expires = jiffies;
526 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
527 int blk_size_in_bytes)
529 struct net_device *dev;
530 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
531 struct ethtool_link_ksettings ecmd;
535 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
536 if (unlikely(!dev)) {
538 return DEFAULT_PRB_RETIRE_TOV;
540 err = __ethtool_get_link_ksettings(dev, &ecmd);
544 * If the link speed is so slow you don't really
545 * need to worry about perf anyways
547 if (ecmd.base.speed < SPEED_1000 ||
548 ecmd.base.speed == SPEED_UNKNOWN) {
549 return DEFAULT_PRB_RETIRE_TOV;
552 div = ecmd.base.speed / 1000;
556 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
568 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
569 union tpacket_req_u *req_u)
571 p1->feature_req_word = req_u->req3.tp_feature_req_word;
574 static void init_prb_bdqc(struct packet_sock *po,
575 struct packet_ring_buffer *rb,
577 union tpacket_req_u *req_u)
579 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
580 struct tpacket_block_desc *pbd;
582 memset(p1, 0x0, sizeof(*p1));
584 p1->knxt_seq_num = 1;
586 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
587 p1->pkblk_start = pg_vec[0].buffer;
588 p1->kblk_size = req_u->req3.tp_block_size;
589 p1->knum_blocks = req_u->req3.tp_block_nr;
590 p1->hdrlen = po->tp_hdrlen;
591 p1->version = po->tp_version;
592 p1->last_kactive_blk_num = 0;
593 po->stats.stats3.tp_freeze_q_cnt = 0;
594 if (req_u->req3.tp_retire_blk_tov)
595 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
597 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
598 req_u->req3.tp_block_size);
599 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
600 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
602 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
603 prb_init_ft_ops(p1, req_u);
604 prb_setup_retire_blk_timer(po);
605 prb_open_block(p1, pbd);
608 /* Do NOT update the last_blk_num first.
609 * Assumes sk_buff_head lock is held.
611 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
613 mod_timer(&pkc->retire_blk_timer,
614 jiffies + pkc->tov_in_jiffies);
615 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
620 * 1) We refresh the timer only when we open a block.
621 * By doing this we don't waste cycles refreshing the timer
622 * on packet-by-packet basis.
624 * With a 1MB block-size, on a 1Gbps line, it will take
625 * i) ~8 ms to fill a block + ii) memcpy etc.
626 * In this cut we are not accounting for the memcpy time.
628 * So, if the user sets the 'tmo' to 10ms then the timer
629 * will never fire while the block is still getting filled
630 * (which is what we want). However, the user could choose
631 * to close a block early and that's fine.
633 * But when the timer does fire, we check whether or not to refresh it.
634 * Since the tmo granularity is in msecs, it is not too expensive
635 * to refresh the timer, lets say every '8' msecs.
636 * Either the user can set the 'tmo' or we can derive it based on
637 * a) line-speed and b) block-size.
638 * prb_calc_retire_blk_tmo() calculates the tmo.
641 static void prb_retire_rx_blk_timer_expired(struct timer_list *t)
643 struct packet_sock *po =
644 from_timer(po, t, rx_ring.prb_bdqc.retire_blk_timer);
645 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
647 struct tpacket_block_desc *pbd;
649 spin_lock(&po->sk.sk_receive_queue.lock);
651 frozen = prb_queue_frozen(pkc);
652 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
654 if (unlikely(pkc->delete_blk_timer))
657 /* We only need to plug the race when the block is partially filled.
659 * lock(); increment BLOCK_NUM_PKTS; unlock()
660 * copy_bits() is in progress ...
661 * timer fires on other cpu:
662 * we can't retire the current block because copy_bits
666 if (BLOCK_NUM_PKTS(pbd)) {
667 while (atomic_read(&pkc->blk_fill_in_prog)) {
668 /* Waiting for skb_copy_bits to finish... */
673 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
675 if (!BLOCK_NUM_PKTS(pbd)) {
676 /* An empty block. Just refresh the timer. */
679 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
680 if (!prb_dispatch_next_block(pkc, po))
685 /* Case 1. Queue was frozen because user-space was
688 if (prb_curr_blk_in_use(pbd)) {
690 * Ok, user-space is still behind.
691 * So just refresh the timer.
695 /* Case 2. queue was frozen,user-space caught up,
696 * now the link went idle && the timer fired.
697 * We don't have a block to close.So we open this
698 * block and restart the timer.
699 * opening a block thaws the queue,restarts timer
700 * Thawing/timer-refresh is a side effect.
702 prb_open_block(pkc, pbd);
709 _prb_refresh_rx_retire_blk_timer(pkc);
712 spin_unlock(&po->sk.sk_receive_queue.lock);
715 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
716 struct tpacket_block_desc *pbd1, __u32 status)
718 /* Flush everything minus the block header */
720 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
725 /* Skip the block header(we know header WILL fit in 4K) */
728 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
729 for (; start < end; start += PAGE_SIZE)
730 flush_dcache_page(pgv_to_page(start));
735 /* Now update the block status. */
737 BLOCK_STATUS(pbd1) = status;
739 /* Flush the block header */
741 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
743 flush_dcache_page(pgv_to_page(start));
753 * 2) Increment active_blk_num
755 * Note:We DONT refresh the timer on purpose.
756 * Because almost always the next block will be opened.
758 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
759 struct tpacket_block_desc *pbd1,
760 struct packet_sock *po, unsigned int stat)
762 __u32 status = TP_STATUS_USER | stat;
764 struct tpacket3_hdr *last_pkt;
765 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
766 struct sock *sk = &po->sk;
768 if (po->stats.stats3.tp_drops)
769 status |= TP_STATUS_LOSING;
771 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
772 last_pkt->tp_next_offset = 0;
774 /* Get the ts of the last pkt */
775 if (BLOCK_NUM_PKTS(pbd1)) {
776 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
777 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
779 /* Ok, we tmo'd - so get the current time.
781 * It shouldn't really happen as we don't close empty
782 * blocks. See prb_retire_rx_blk_timer_expired().
786 h1->ts_last_pkt.ts_sec = ts.tv_sec;
787 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
792 /* Flush the block */
793 prb_flush_block(pkc1, pbd1, status);
795 sk->sk_data_ready(sk);
797 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
800 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
802 pkc->reset_pending_on_curr_blk = 0;
806 * Side effect of opening a block:
808 * 1) prb_queue is thawed.
809 * 2) retire_blk_timer is refreshed.
812 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
813 struct tpacket_block_desc *pbd1)
816 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
820 /* We could have just memset this but we will lose the
821 * flexibility of making the priv area sticky
824 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
825 BLOCK_NUM_PKTS(pbd1) = 0;
826 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
830 h1->ts_first_pkt.ts_sec = ts.tv_sec;
831 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
833 pkc1->pkblk_start = (char *)pbd1;
834 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
836 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
837 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
839 pbd1->version = pkc1->version;
840 pkc1->prev = pkc1->nxt_offset;
841 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
843 prb_thaw_queue(pkc1);
844 _prb_refresh_rx_retire_blk_timer(pkc1);
850 * Queue freeze logic:
851 * 1) Assume tp_block_nr = 8 blocks.
852 * 2) At time 't0', user opens Rx ring.
853 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
854 * 4) user-space is either sleeping or processing block '0'.
855 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
856 * it will close block-7,loop around and try to fill block '0'.
858 * __packet_lookup_frame_in_block
859 * prb_retire_current_block()
860 * prb_dispatch_next_block()
861 * |->(BLOCK_STATUS == USER) evaluates to true
862 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
863 * 6) Now there are two cases:
864 * 6.1) Link goes idle right after the queue is frozen.
865 * But remember, the last open_block() refreshed the timer.
866 * When this timer expires,it will refresh itself so that we can
867 * re-open block-0 in near future.
868 * 6.2) Link is busy and keeps on receiving packets. This is a simple
869 * case and __packet_lookup_frame_in_block will check if block-0
870 * is free and can now be re-used.
872 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
873 struct packet_sock *po)
875 pkc->reset_pending_on_curr_blk = 1;
876 po->stats.stats3.tp_freeze_q_cnt++;
879 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
882 * If the next block is free then we will dispatch it
883 * and return a good offset.
884 * Else, we will freeze the queue.
885 * So, caller must check the return value.
887 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
888 struct packet_sock *po)
890 struct tpacket_block_desc *pbd;
894 /* 1. Get current block num */
895 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
897 /* 2. If this block is currently in_use then freeze the queue */
898 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
899 prb_freeze_queue(pkc, po);
905 * open this block and return the offset where the first packet
906 * needs to get stored.
908 prb_open_block(pkc, pbd);
909 return (void *)pkc->nxt_offset;
912 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
913 struct packet_sock *po, unsigned int status)
915 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
917 /* retire/close the current block */
918 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
920 * Plug the case where copy_bits() is in progress on
921 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
922 * have space to copy the pkt in the current block and
923 * called prb_retire_current_block()
925 * We don't need to worry about the TMO case because
926 * the timer-handler already handled this case.
928 if (!(status & TP_STATUS_BLK_TMO)) {
929 while (atomic_read(&pkc->blk_fill_in_prog)) {
930 /* Waiting for skb_copy_bits to finish... */
934 prb_close_block(pkc, pbd, po, status);
939 static int prb_curr_blk_in_use(struct tpacket_block_desc *pbd)
941 return TP_STATUS_USER & BLOCK_STATUS(pbd);
944 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
946 return pkc->reset_pending_on_curr_blk;
949 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
951 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
952 atomic_dec(&pkc->blk_fill_in_prog);
955 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
956 struct tpacket3_hdr *ppd)
958 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
961 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
962 struct tpacket3_hdr *ppd)
964 ppd->hv1.tp_rxhash = 0;
967 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
968 struct tpacket3_hdr *ppd)
970 if (skb_vlan_tag_present(pkc->skb)) {
971 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
972 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
973 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
975 ppd->hv1.tp_vlan_tci = 0;
976 ppd->hv1.tp_vlan_tpid = 0;
977 ppd->tp_status = TP_STATUS_AVAILABLE;
981 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
982 struct tpacket3_hdr *ppd)
984 ppd->hv1.tp_padding = 0;
985 prb_fill_vlan_info(pkc, ppd);
987 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
988 prb_fill_rxhash(pkc, ppd);
990 prb_clear_rxhash(pkc, ppd);
993 static void prb_fill_curr_block(char *curr,
994 struct tpacket_kbdq_core *pkc,
995 struct tpacket_block_desc *pbd,
998 struct tpacket3_hdr *ppd;
1000 ppd = (struct tpacket3_hdr *)curr;
1001 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1003 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1004 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1005 BLOCK_NUM_PKTS(pbd) += 1;
1006 atomic_inc(&pkc->blk_fill_in_prog);
1007 prb_run_all_ft_ops(pkc, ppd);
1010 /* Assumes caller has the sk->rx_queue.lock */
1011 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1012 struct sk_buff *skb,
1017 struct tpacket_kbdq_core *pkc;
1018 struct tpacket_block_desc *pbd;
1021 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1022 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1024 /* Queue is frozen when user space is lagging behind */
1025 if (prb_queue_frozen(pkc)) {
1027 * Check if that last block which caused the queue to freeze,
1028 * is still in_use by user-space.
1030 if (prb_curr_blk_in_use(pbd)) {
1031 /* Can't record this packet */
1035 * Ok, the block was released by user-space.
1036 * Now let's open that block.
1037 * opening a block also thaws the queue.
1038 * Thawing is a side effect.
1040 prb_open_block(pkc, pbd);
1045 curr = pkc->nxt_offset;
1047 end = (char *)pbd + pkc->kblk_size;
1049 /* first try the current block */
1050 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1051 prb_fill_curr_block(curr, pkc, pbd, len);
1052 return (void *)curr;
1055 /* Ok, close the current block */
1056 prb_retire_current_block(pkc, po, 0);
1058 /* Now, try to dispatch the next block */
1059 curr = (char *)prb_dispatch_next_block(pkc, po);
1061 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1062 prb_fill_curr_block(curr, pkc, pbd, len);
1063 return (void *)curr;
1067 * No free blocks are available.user_space hasn't caught up yet.
1068 * Queue was just frozen and now this packet will get dropped.
1073 static void *packet_current_rx_frame(struct packet_sock *po,
1074 struct sk_buff *skb,
1075 int status, unsigned int len)
1078 switch (po->tp_version) {
1081 curr = packet_lookup_frame(po, &po->rx_ring,
1082 po->rx_ring.head, status);
1085 return __packet_lookup_frame_in_block(po, skb, status, len);
1087 WARN(1, "TPACKET version not supported\n");
1093 static void *prb_lookup_block(struct packet_sock *po,
1094 struct packet_ring_buffer *rb,
1098 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1099 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1101 if (status != BLOCK_STATUS(pbd))
1106 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1109 if (rb->prb_bdqc.kactive_blk_num)
1110 prev = rb->prb_bdqc.kactive_blk_num-1;
1112 prev = rb->prb_bdqc.knum_blocks-1;
1116 /* Assumes caller has held the rx_queue.lock */
1117 static void *__prb_previous_block(struct packet_sock *po,
1118 struct packet_ring_buffer *rb,
1121 unsigned int previous = prb_previous_blk_num(rb);
1122 return prb_lookup_block(po, rb, previous, status);
1125 static void *packet_previous_rx_frame(struct packet_sock *po,
1126 struct packet_ring_buffer *rb,
1129 if (po->tp_version <= TPACKET_V2)
1130 return packet_previous_frame(po, rb, status);
1132 return __prb_previous_block(po, rb, status);
1135 static void packet_increment_rx_head(struct packet_sock *po,
1136 struct packet_ring_buffer *rb)
1138 switch (po->tp_version) {
1141 return packet_increment_head(rb);
1144 WARN(1, "TPACKET version not supported.\n");
1150 static void *packet_previous_frame(struct packet_sock *po,
1151 struct packet_ring_buffer *rb,
1154 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1155 return packet_lookup_frame(po, rb, previous, status);
1158 static void packet_increment_head(struct packet_ring_buffer *buff)
1160 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1163 static void packet_inc_pending(struct packet_ring_buffer *rb)
1165 this_cpu_inc(*rb->pending_refcnt);
1168 static void packet_dec_pending(struct packet_ring_buffer *rb)
1170 this_cpu_dec(*rb->pending_refcnt);
1173 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1175 unsigned int refcnt = 0;
1178 /* We don't use pending refcount in rx_ring. */
1179 if (rb->pending_refcnt == NULL)
1182 for_each_possible_cpu(cpu)
1183 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1188 static int packet_alloc_pending(struct packet_sock *po)
1190 po->rx_ring.pending_refcnt = NULL;
1192 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1193 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1199 static void packet_free_pending(struct packet_sock *po)
1201 free_percpu(po->tx_ring.pending_refcnt);
1204 #define ROOM_POW_OFF 2
1205 #define ROOM_NONE 0x0
1206 #define ROOM_LOW 0x1
1207 #define ROOM_NORMAL 0x2
1209 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1213 len = po->rx_ring.frame_max + 1;
1214 idx = po->rx_ring.head;
1216 idx += len >> pow_off;
1219 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1222 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1226 len = po->rx_ring.prb_bdqc.knum_blocks;
1227 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1229 idx += len >> pow_off;
1232 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1235 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1237 struct sock *sk = &po->sk;
1238 int ret = ROOM_NONE;
1240 if (po->prot_hook.func != tpacket_rcv) {
1241 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1242 - (skb ? skb->truesize : 0);
1243 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1251 if (po->tp_version == TPACKET_V3) {
1252 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1254 else if (__tpacket_v3_has_room(po, 0))
1257 if (__tpacket_has_room(po, ROOM_POW_OFF))
1259 else if (__tpacket_has_room(po, 0))
1266 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1271 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1272 ret = __packet_rcv_has_room(po, skb);
1273 has_room = ret == ROOM_NORMAL;
1274 if (po->pressure == has_room)
1275 po->pressure = !has_room;
1276 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1281 static void packet_sock_destruct(struct sock *sk)
1283 skb_queue_purge(&sk->sk_error_queue);
1285 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1286 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
1288 if (!sock_flag(sk, SOCK_DEAD)) {
1289 pr_err("Attempt to release alive packet socket: %p\n", sk);
1293 sk_refcnt_debug_dec(sk);
1296 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1301 rxhash = skb_get_hash(skb);
1302 for (i = 0; i < ROLLOVER_HLEN; i++)
1303 if (po->rollover->history[i] == rxhash)
1306 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1307 return count > (ROLLOVER_HLEN >> 1);
1310 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1311 struct sk_buff *skb,
1314 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1317 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1318 struct sk_buff *skb,
1321 unsigned int val = atomic_inc_return(&f->rr_cur);
1326 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1327 struct sk_buff *skb,
1330 return smp_processor_id() % num;
1333 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1334 struct sk_buff *skb,
1337 return prandom_u32_max(num);
1340 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1341 struct sk_buff *skb,
1342 unsigned int idx, bool try_self,
1345 struct packet_sock *po, *po_next, *po_skip = NULL;
1346 unsigned int i, j, room = ROOM_NONE;
1348 po = pkt_sk(f->arr[idx]);
1351 room = packet_rcv_has_room(po, skb);
1352 if (room == ROOM_NORMAL ||
1353 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1358 i = j = min_t(int, po->rollover->sock, num - 1);
1360 po_next = pkt_sk(f->arr[i]);
1361 if (po_next != po_skip && !po_next->pressure &&
1362 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1364 po->rollover->sock = i;
1365 atomic_long_inc(&po->rollover->num);
1366 if (room == ROOM_LOW)
1367 atomic_long_inc(&po->rollover->num_huge);
1375 atomic_long_inc(&po->rollover->num_failed);
1379 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1380 struct sk_buff *skb,
1383 return skb_get_queue_mapping(skb) % num;
1386 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1387 struct sk_buff *skb,
1390 struct bpf_prog *prog;
1391 unsigned int ret = 0;
1394 prog = rcu_dereference(f->bpf_prog);
1396 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1402 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1404 return f->flags & (flag >> 8);
1407 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1408 struct packet_type *pt, struct net_device *orig_dev)
1410 struct packet_fanout *f = pt->af_packet_priv;
1411 unsigned int num = READ_ONCE(f->num_members);
1412 struct net *net = read_pnet(&f->net);
1413 struct packet_sock *po;
1416 if (!net_eq(dev_net(dev), net) || !num) {
1421 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1422 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1427 case PACKET_FANOUT_HASH:
1429 idx = fanout_demux_hash(f, skb, num);
1431 case PACKET_FANOUT_LB:
1432 idx = fanout_demux_lb(f, skb, num);
1434 case PACKET_FANOUT_CPU:
1435 idx = fanout_demux_cpu(f, skb, num);
1437 case PACKET_FANOUT_RND:
1438 idx = fanout_demux_rnd(f, skb, num);
1440 case PACKET_FANOUT_QM:
1441 idx = fanout_demux_qm(f, skb, num);
1443 case PACKET_FANOUT_ROLLOVER:
1444 idx = fanout_demux_rollover(f, skb, 0, false, num);
1446 case PACKET_FANOUT_CBPF:
1447 case PACKET_FANOUT_EBPF:
1448 idx = fanout_demux_bpf(f, skb, num);
1452 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1453 idx = fanout_demux_rollover(f, skb, idx, true, num);
1455 po = pkt_sk(f->arr[idx]);
1456 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1459 DEFINE_MUTEX(fanout_mutex);
1460 EXPORT_SYMBOL_GPL(fanout_mutex);
1461 static LIST_HEAD(fanout_list);
1462 static u16 fanout_next_id;
1464 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1466 struct packet_fanout *f = po->fanout;
1468 spin_lock(&f->lock);
1469 f->arr[f->num_members] = sk;
1472 if (f->num_members == 1)
1473 dev_add_pack(&f->prot_hook);
1474 spin_unlock(&f->lock);
1477 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1479 struct packet_fanout *f = po->fanout;
1482 spin_lock(&f->lock);
1483 for (i = 0; i < f->num_members; i++) {
1484 if (f->arr[i] == sk)
1487 BUG_ON(i >= f->num_members);
1488 f->arr[i] = f->arr[f->num_members - 1];
1490 if (f->num_members == 0)
1491 __dev_remove_pack(&f->prot_hook);
1492 spin_unlock(&f->lock);
1495 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1497 if (sk->sk_family != PF_PACKET)
1500 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1503 static void fanout_init_data(struct packet_fanout *f)
1506 case PACKET_FANOUT_LB:
1507 atomic_set(&f->rr_cur, 0);
1509 case PACKET_FANOUT_CBPF:
1510 case PACKET_FANOUT_EBPF:
1511 RCU_INIT_POINTER(f->bpf_prog, NULL);
1516 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1518 struct bpf_prog *old;
1520 spin_lock(&f->lock);
1521 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1522 rcu_assign_pointer(f->bpf_prog, new);
1523 spin_unlock(&f->lock);
1527 bpf_prog_destroy(old);
1531 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1534 struct bpf_prog *new;
1535 struct sock_fprog fprog;
1538 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1540 if (len != sizeof(fprog))
1542 if (copy_from_user(&fprog, data, len))
1545 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1549 __fanout_set_data_bpf(po->fanout, new);
1553 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1556 struct bpf_prog *new;
1559 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1561 if (len != sizeof(fd))
1563 if (copy_from_user(&fd, data, len))
1566 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1568 return PTR_ERR(new);
1570 __fanout_set_data_bpf(po->fanout, new);
1574 static int fanout_set_data(struct packet_sock *po, char __user *data,
1577 switch (po->fanout->type) {
1578 case PACKET_FANOUT_CBPF:
1579 return fanout_set_data_cbpf(po, data, len);
1580 case PACKET_FANOUT_EBPF:
1581 return fanout_set_data_ebpf(po, data, len);
1587 static void fanout_release_data(struct packet_fanout *f)
1590 case PACKET_FANOUT_CBPF:
1591 case PACKET_FANOUT_EBPF:
1592 __fanout_set_data_bpf(f, NULL);
1596 static bool __fanout_id_is_free(struct sock *sk, u16 candidate_id)
1598 struct packet_fanout *f;
1600 list_for_each_entry(f, &fanout_list, list) {
1601 if (f->id == candidate_id &&
1602 read_pnet(&f->net) == sock_net(sk)) {
1609 static bool fanout_find_new_id(struct sock *sk, u16 *new_id)
1611 u16 id = fanout_next_id;
1614 if (__fanout_id_is_free(sk, id)) {
1616 fanout_next_id = id + 1;
1621 } while (id != fanout_next_id);
1626 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1628 struct packet_rollover *rollover = NULL;
1629 struct packet_sock *po = pkt_sk(sk);
1630 struct packet_fanout *f, *match;
1631 u8 type = type_flags & 0xff;
1632 u8 flags = type_flags >> 8;
1636 case PACKET_FANOUT_ROLLOVER:
1637 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1639 case PACKET_FANOUT_HASH:
1640 case PACKET_FANOUT_LB:
1641 case PACKET_FANOUT_CPU:
1642 case PACKET_FANOUT_RND:
1643 case PACKET_FANOUT_QM:
1644 case PACKET_FANOUT_CBPF:
1645 case PACKET_FANOUT_EBPF:
1651 mutex_lock(&fanout_mutex);
1657 if (type == PACKET_FANOUT_ROLLOVER ||
1658 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1660 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1663 atomic_long_set(&rollover->num, 0);
1664 atomic_long_set(&rollover->num_huge, 0);
1665 atomic_long_set(&rollover->num_failed, 0);
1668 if (type_flags & PACKET_FANOUT_FLAG_UNIQUEID) {
1673 if (!fanout_find_new_id(sk, &id)) {
1677 /* ephemeral flag for the first socket in the group: drop it */
1678 flags &= ~(PACKET_FANOUT_FLAG_UNIQUEID >> 8);
1682 list_for_each_entry(f, &fanout_list, list) {
1684 read_pnet(&f->net) == sock_net(sk)) {
1690 if (match && match->flags != flags)
1694 match = kzalloc(sizeof(*match), GFP_KERNEL);
1697 write_pnet(&match->net, sock_net(sk));
1700 match->flags = flags;
1701 INIT_LIST_HEAD(&match->list);
1702 spin_lock_init(&match->lock);
1703 refcount_set(&match->sk_ref, 0);
1704 fanout_init_data(match);
1705 match->prot_hook.type = po->prot_hook.type;
1706 match->prot_hook.dev = po->prot_hook.dev;
1707 match->prot_hook.func = packet_rcv_fanout;
1708 match->prot_hook.af_packet_priv = match;
1709 match->prot_hook.id_match = match_fanout_group;
1710 list_add(&match->list, &fanout_list);
1714 spin_lock(&po->bind_lock);
1716 match->type == type &&
1717 match->prot_hook.type == po->prot_hook.type &&
1718 match->prot_hook.dev == po->prot_hook.dev) {
1720 if (refcount_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1721 __dev_remove_pack(&po->prot_hook);
1723 po->rollover = rollover;
1725 refcount_set(&match->sk_ref, refcount_read(&match->sk_ref) + 1);
1726 __fanout_link(sk, po);
1730 spin_unlock(&po->bind_lock);
1732 if (err && !refcount_read(&match->sk_ref)) {
1733 list_del(&match->list);
1739 mutex_unlock(&fanout_mutex);
1743 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1744 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1745 * It is the responsibility of the caller to call fanout_release_data() and
1746 * free the returned packet_fanout (after synchronize_net())
1748 static struct packet_fanout *fanout_release(struct sock *sk)
1750 struct packet_sock *po = pkt_sk(sk);
1751 struct packet_fanout *f;
1753 mutex_lock(&fanout_mutex);
1758 if (refcount_dec_and_test(&f->sk_ref))
1763 mutex_unlock(&fanout_mutex);
1768 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1769 struct sk_buff *skb)
1771 /* Earlier code assumed this would be a VLAN pkt, double-check
1772 * this now that we have the actual packet in hand. We can only
1773 * do this check on Ethernet devices.
1775 if (unlikely(dev->type != ARPHRD_ETHER))
1778 skb_reset_mac_header(skb);
1779 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1782 static const struct proto_ops packet_ops;
1784 static const struct proto_ops packet_ops_spkt;
1786 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1787 struct packet_type *pt, struct net_device *orig_dev)
1790 struct sockaddr_pkt *spkt;
1793 * When we registered the protocol we saved the socket in the data
1794 * field for just this event.
1797 sk = pt->af_packet_priv;
1800 * Yank back the headers [hope the device set this
1801 * right or kerboom...]
1803 * Incoming packets have ll header pulled,
1806 * For outgoing ones skb->data == skb_mac_header(skb)
1807 * so that this procedure is noop.
1810 if (skb->pkt_type == PACKET_LOOPBACK)
1813 if (!net_eq(dev_net(dev), sock_net(sk)))
1816 skb = skb_share_check(skb, GFP_ATOMIC);
1820 /* drop any routing info */
1823 /* drop conntrack reference */
1826 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1828 skb_push(skb, skb->data - skb_mac_header(skb));
1831 * The SOCK_PACKET socket receives _all_ frames.
1834 spkt->spkt_family = dev->type;
1835 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1836 spkt->spkt_protocol = skb->protocol;
1839 * Charge the memory to the socket. This is done specifically
1840 * to prevent sockets using all the memory up.
1843 if (sock_queue_rcv_skb(sk, skb) == 0)
1854 * Output a raw packet to a device layer. This bypasses all the other
1855 * protocol layers and you must therefore supply it with a complete frame
1858 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1861 struct sock *sk = sock->sk;
1862 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1863 struct sk_buff *skb = NULL;
1864 struct net_device *dev;
1865 struct sockcm_cookie sockc;
1871 * Get and verify the address.
1875 if (msg->msg_namelen < sizeof(struct sockaddr))
1877 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1878 proto = saddr->spkt_protocol;
1880 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1883 * Find the device first to size check it
1886 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1889 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1895 if (!(dev->flags & IFF_UP))
1899 * You may not queue a frame bigger than the mtu. This is the lowest level
1900 * raw protocol and you must do your own fragmentation at this level.
1903 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1904 if (!netif_supports_nofcs(dev)) {
1905 err = -EPROTONOSUPPORT;
1908 extra_len = 4; /* We're doing our own CRC */
1912 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1916 size_t reserved = LL_RESERVED_SPACE(dev);
1917 int tlen = dev->needed_tailroom;
1918 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1921 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1924 /* FIXME: Save some space for broken drivers that write a hard
1925 * header at transmission time by themselves. PPP is the notable
1926 * one here. This should really be fixed at the driver level.
1928 skb_reserve(skb, reserved);
1929 skb_reset_network_header(skb);
1931 /* Try to align data part correctly */
1936 skb_reset_network_header(skb);
1938 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1944 if (!dev_validate_header(dev, skb->data, len)) {
1948 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1949 !packet_extra_vlan_len_allowed(dev, skb)) {
1954 sockcm_init(&sockc, sk);
1955 if (msg->msg_controllen) {
1956 err = sock_cmsg_send(sk, msg, &sockc);
1961 skb->protocol = proto;
1963 skb->priority = sk->sk_priority;
1964 skb->mark = sk->sk_mark;
1965 skb->tstamp = sockc.transmit_time;
1967 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1969 if (unlikely(extra_len == 4))
1972 skb_probe_transport_header(skb, 0);
1974 dev_queue_xmit(skb);
1985 static unsigned int run_filter(struct sk_buff *skb,
1986 const struct sock *sk,
1989 struct sk_filter *filter;
1992 filter = rcu_dereference(sk->sk_filter);
1994 res = bpf_prog_run_clear_cb(filter->prog, skb);
2000 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
2003 struct virtio_net_hdr vnet_hdr;
2005 if (*len < sizeof(vnet_hdr))
2007 *len -= sizeof(vnet_hdr);
2009 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true, 0))
2012 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2016 * This function makes lazy skb cloning in hope that most of packets
2017 * are discarded by BPF.
2019 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2020 * and skb->cb are mangled. It works because (and until) packets
2021 * falling here are owned by current CPU. Output packets are cloned
2022 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2023 * sequencially, so that if we return skb to original state on exit,
2024 * we will not harm anyone.
2027 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2028 struct packet_type *pt, struct net_device *orig_dev)
2031 struct sockaddr_ll *sll;
2032 struct packet_sock *po;
2033 u8 *skb_head = skb->data;
2034 int skb_len = skb->len;
2035 unsigned int snaplen, res;
2036 bool is_drop_n_account = false;
2038 if (skb->pkt_type == PACKET_LOOPBACK)
2041 sk = pt->af_packet_priv;
2044 if (!net_eq(dev_net(dev), sock_net(sk)))
2049 if (dev->header_ops) {
2050 /* The device has an explicit notion of ll header,
2051 * exported to higher levels.
2053 * Otherwise, the device hides details of its frame
2054 * structure, so that corresponding packet head is
2055 * never delivered to user.
2057 if (sk->sk_type != SOCK_DGRAM)
2058 skb_push(skb, skb->data - skb_mac_header(skb));
2059 else if (skb->pkt_type == PACKET_OUTGOING) {
2060 /* Special case: outgoing packets have ll header at head */
2061 skb_pull(skb, skb_network_offset(skb));
2067 res = run_filter(skb, sk, snaplen);
2069 goto drop_n_restore;
2073 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2076 if (skb_shared(skb)) {
2077 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2081 if (skb_head != skb->data) {
2082 skb->data = skb_head;
2089 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2091 sll = &PACKET_SKB_CB(skb)->sa.ll;
2092 sll->sll_hatype = dev->type;
2093 sll->sll_pkttype = skb->pkt_type;
2094 if (unlikely(po->origdev))
2095 sll->sll_ifindex = orig_dev->ifindex;
2097 sll->sll_ifindex = dev->ifindex;
2099 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2101 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2102 * Use their space for storing the original skb length.
2104 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2106 if (pskb_trim(skb, snaplen))
2109 skb_set_owner_r(skb, sk);
2113 /* drop conntrack reference */
2116 spin_lock(&sk->sk_receive_queue.lock);
2117 po->stats.stats1.tp_packets++;
2118 sock_skb_set_dropcount(sk, skb);
2119 __skb_queue_tail(&sk->sk_receive_queue, skb);
2120 spin_unlock(&sk->sk_receive_queue.lock);
2121 sk->sk_data_ready(sk);
2125 is_drop_n_account = true;
2126 spin_lock(&sk->sk_receive_queue.lock);
2127 po->stats.stats1.tp_drops++;
2128 atomic_inc(&sk->sk_drops);
2129 spin_unlock(&sk->sk_receive_queue.lock);
2132 if (skb_head != skb->data && skb_shared(skb)) {
2133 skb->data = skb_head;
2137 if (!is_drop_n_account)
2144 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2145 struct packet_type *pt, struct net_device *orig_dev)
2148 struct packet_sock *po;
2149 struct sockaddr_ll *sll;
2150 union tpacket_uhdr h;
2151 u8 *skb_head = skb->data;
2152 int skb_len = skb->len;
2153 unsigned int snaplen, res;
2154 unsigned long status = TP_STATUS_USER;
2155 unsigned short macoff, netoff, hdrlen;
2156 struct sk_buff *copy_skb = NULL;
2159 bool is_drop_n_account = false;
2160 bool do_vnet = false;
2162 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2163 * We may add members to them until current aligned size without forcing
2164 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2166 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2167 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2169 if (skb->pkt_type == PACKET_LOOPBACK)
2172 sk = pt->af_packet_priv;
2175 if (!net_eq(dev_net(dev), sock_net(sk)))
2178 if (dev->header_ops) {
2179 if (sk->sk_type != SOCK_DGRAM)
2180 skb_push(skb, skb->data - skb_mac_header(skb));
2181 else if (skb->pkt_type == PACKET_OUTGOING) {
2182 /* Special case: outgoing packets have ll header at head */
2183 skb_pull(skb, skb_network_offset(skb));
2189 res = run_filter(skb, sk, snaplen);
2191 goto drop_n_restore;
2193 if (skb->ip_summed == CHECKSUM_PARTIAL)
2194 status |= TP_STATUS_CSUMNOTREADY;
2195 else if (skb->pkt_type != PACKET_OUTGOING &&
2196 (skb->ip_summed == CHECKSUM_COMPLETE ||
2197 skb_csum_unnecessary(skb)))
2198 status |= TP_STATUS_CSUM_VALID;
2203 if (sk->sk_type == SOCK_DGRAM) {
2204 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2207 unsigned int maclen = skb_network_offset(skb);
2208 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2209 (maclen < 16 ? 16 : maclen)) +
2211 if (po->has_vnet_hdr) {
2212 netoff += sizeof(struct virtio_net_hdr);
2215 macoff = netoff - maclen;
2217 if (po->tp_version <= TPACKET_V2) {
2218 if (macoff + snaplen > po->rx_ring.frame_size) {
2219 if (po->copy_thresh &&
2220 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2221 if (skb_shared(skb)) {
2222 copy_skb = skb_clone(skb, GFP_ATOMIC);
2224 copy_skb = skb_get(skb);
2225 skb_head = skb->data;
2228 skb_set_owner_r(copy_skb, sk);
2230 snaplen = po->rx_ring.frame_size - macoff;
2231 if ((int)snaplen < 0) {
2236 } else if (unlikely(macoff + snaplen >
2237 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2240 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2241 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2242 snaplen, nval, macoff);
2244 if (unlikely((int)snaplen < 0)) {
2246 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2250 spin_lock(&sk->sk_receive_queue.lock);
2251 h.raw = packet_current_rx_frame(po, skb,
2252 TP_STATUS_KERNEL, (macoff+snaplen));
2254 goto drop_n_account;
2255 if (po->tp_version <= TPACKET_V2) {
2256 packet_increment_rx_head(po, &po->rx_ring);
2258 * LOSING will be reported till you read the stats,
2259 * because it's COR - Clear On Read.
2260 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2263 if (po->stats.stats1.tp_drops)
2264 status |= TP_STATUS_LOSING;
2268 virtio_net_hdr_from_skb(skb, h.raw + macoff -
2269 sizeof(struct virtio_net_hdr),
2271 goto drop_n_account;
2273 po->stats.stats1.tp_packets++;
2275 status |= TP_STATUS_COPY;
2276 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2278 spin_unlock(&sk->sk_receive_queue.lock);
2280 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2282 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2283 getnstimeofday(&ts);
2285 status |= ts_status;
2287 switch (po->tp_version) {
2289 h.h1->tp_len = skb->len;
2290 h.h1->tp_snaplen = snaplen;
2291 h.h1->tp_mac = macoff;
2292 h.h1->tp_net = netoff;
2293 h.h1->tp_sec = ts.tv_sec;
2294 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2295 hdrlen = sizeof(*h.h1);
2298 h.h2->tp_len = skb->len;
2299 h.h2->tp_snaplen = snaplen;
2300 h.h2->tp_mac = macoff;
2301 h.h2->tp_net = netoff;
2302 h.h2->tp_sec = ts.tv_sec;
2303 h.h2->tp_nsec = ts.tv_nsec;
2304 if (skb_vlan_tag_present(skb)) {
2305 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2306 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2307 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2309 h.h2->tp_vlan_tci = 0;
2310 h.h2->tp_vlan_tpid = 0;
2312 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2313 hdrlen = sizeof(*h.h2);
2316 /* tp_nxt_offset,vlan are already populated above.
2317 * So DONT clear those fields here
2319 h.h3->tp_status |= status;
2320 h.h3->tp_len = skb->len;
2321 h.h3->tp_snaplen = snaplen;
2322 h.h3->tp_mac = macoff;
2323 h.h3->tp_net = netoff;
2324 h.h3->tp_sec = ts.tv_sec;
2325 h.h3->tp_nsec = ts.tv_nsec;
2326 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2327 hdrlen = sizeof(*h.h3);
2333 sll = h.raw + TPACKET_ALIGN(hdrlen);
2334 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2335 sll->sll_family = AF_PACKET;
2336 sll->sll_hatype = dev->type;
2337 sll->sll_protocol = skb->protocol;
2338 sll->sll_pkttype = skb->pkt_type;
2339 if (unlikely(po->origdev))
2340 sll->sll_ifindex = orig_dev->ifindex;
2342 sll->sll_ifindex = dev->ifindex;
2346 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2347 if (po->tp_version <= TPACKET_V2) {
2350 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2353 for (start = h.raw; start < end; start += PAGE_SIZE)
2354 flush_dcache_page(pgv_to_page(start));
2359 if (po->tp_version <= TPACKET_V2) {
2360 __packet_set_status(po, h.raw, status);
2361 sk->sk_data_ready(sk);
2363 prb_clear_blk_fill_status(&po->rx_ring);
2367 if (skb_head != skb->data && skb_shared(skb)) {
2368 skb->data = skb_head;
2372 if (!is_drop_n_account)
2379 is_drop_n_account = true;
2380 po->stats.stats1.tp_drops++;
2381 spin_unlock(&sk->sk_receive_queue.lock);
2383 sk->sk_data_ready(sk);
2384 kfree_skb(copy_skb);
2385 goto drop_n_restore;
2388 static void tpacket_destruct_skb(struct sk_buff *skb)
2390 struct packet_sock *po = pkt_sk(skb->sk);
2392 if (likely(po->tx_ring.pg_vec)) {
2396 ph = skb_shinfo(skb)->destructor_arg;
2397 packet_dec_pending(&po->tx_ring);
2399 ts = __packet_set_timestamp(po, ph, skb);
2400 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2406 static void tpacket_set_protocol(const struct net_device *dev,
2407 struct sk_buff *skb)
2409 if (dev->type == ARPHRD_ETHER) {
2410 skb_reset_mac_header(skb);
2411 skb->protocol = eth_hdr(skb)->h_proto;
2415 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2417 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2418 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2419 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2420 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2421 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2422 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2423 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2425 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2431 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2432 struct virtio_net_hdr *vnet_hdr)
2434 if (*len < sizeof(*vnet_hdr))
2436 *len -= sizeof(*vnet_hdr);
2438 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2441 return __packet_snd_vnet_parse(vnet_hdr, *len);
2444 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2445 void *frame, struct net_device *dev, void *data, int tp_len,
2446 __be16 proto, unsigned char *addr, int hlen, int copylen,
2447 const struct sockcm_cookie *sockc)
2449 union tpacket_uhdr ph;
2450 int to_write, offset, len, nr_frags, len_max;
2451 struct socket *sock = po->sk.sk_socket;
2457 skb->protocol = proto;
2459 skb->priority = po->sk.sk_priority;
2460 skb->mark = po->sk.sk_mark;
2461 skb->tstamp = sockc->transmit_time;
2462 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2463 skb_shinfo(skb)->destructor_arg = ph.raw;
2465 skb_reserve(skb, hlen);
2466 skb_reset_network_header(skb);
2470 if (sock->type == SOCK_DGRAM) {
2471 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2473 if (unlikely(err < 0))
2475 } else if (copylen) {
2476 int hdrlen = min_t(int, copylen, tp_len);
2478 skb_push(skb, dev->hard_header_len);
2479 skb_put(skb, copylen - dev->hard_header_len);
2480 err = skb_store_bits(skb, 0, data, hdrlen);
2483 if (!dev_validate_header(dev, skb->data, hdrlen))
2486 tpacket_set_protocol(dev, skb);
2492 offset = offset_in_page(data);
2493 len_max = PAGE_SIZE - offset;
2494 len = ((to_write > len_max) ? len_max : to_write);
2496 skb->data_len = to_write;
2497 skb->len += to_write;
2498 skb->truesize += to_write;
2499 refcount_add(to_write, &po->sk.sk_wmem_alloc);
2501 while (likely(to_write)) {
2502 nr_frags = skb_shinfo(skb)->nr_frags;
2504 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2505 pr_err("Packet exceed the number of skb frags(%lu)\n",
2510 page = pgv_to_page(data);
2512 flush_dcache_page(page);
2514 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2517 len_max = PAGE_SIZE;
2518 len = ((to_write > len_max) ? len_max : to_write);
2521 skb_probe_transport_header(skb, 0);
2526 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2527 int size_max, void **data)
2529 union tpacket_uhdr ph;
2534 switch (po->tp_version) {
2536 if (ph.h3->tp_next_offset != 0) {
2537 pr_warn_once("variable sized slot not supported");
2540 tp_len = ph.h3->tp_len;
2543 tp_len = ph.h2->tp_len;
2546 tp_len = ph.h1->tp_len;
2549 if (unlikely(tp_len > size_max)) {
2550 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2554 if (unlikely(po->tp_tx_has_off)) {
2555 int off_min, off_max;
2557 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2558 off_max = po->tx_ring.frame_size - tp_len;
2559 if (po->sk.sk_type == SOCK_DGRAM) {
2560 switch (po->tp_version) {
2562 off = ph.h3->tp_net;
2565 off = ph.h2->tp_net;
2568 off = ph.h1->tp_net;
2572 switch (po->tp_version) {
2574 off = ph.h3->tp_mac;
2577 off = ph.h2->tp_mac;
2580 off = ph.h1->tp_mac;
2584 if (unlikely((off < off_min) || (off_max < off)))
2587 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2590 *data = frame + off;
2594 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2596 struct sk_buff *skb;
2597 struct net_device *dev;
2598 struct virtio_net_hdr *vnet_hdr = NULL;
2599 struct sockcm_cookie sockc;
2601 int err, reserve = 0;
2603 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2604 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2605 int tp_len, size_max;
2606 unsigned char *addr;
2609 int status = TP_STATUS_AVAILABLE;
2610 int hlen, tlen, copylen = 0;
2612 mutex_lock(&po->pg_vec_lock);
2614 if (likely(saddr == NULL)) {
2615 dev = packet_cached_dev_get(po);
2620 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2622 if (msg->msg_namelen < (saddr->sll_halen
2623 + offsetof(struct sockaddr_ll,
2626 proto = saddr->sll_protocol;
2627 addr = saddr->sll_addr;
2628 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2632 if (unlikely(dev == NULL))
2635 if (unlikely(!(dev->flags & IFF_UP)))
2638 sockcm_init(&sockc, &po->sk);
2639 if (msg->msg_controllen) {
2640 err = sock_cmsg_send(&po->sk, msg, &sockc);
2645 if (po->sk.sk_socket->type == SOCK_RAW)
2646 reserve = dev->hard_header_len;
2647 size_max = po->tx_ring.frame_size
2648 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2650 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2651 size_max = dev->mtu + reserve + VLAN_HLEN;
2654 ph = packet_current_frame(po, &po->tx_ring,
2655 TP_STATUS_SEND_REQUEST);
2656 if (unlikely(ph == NULL)) {
2657 if (need_wait && need_resched())
2663 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2667 status = TP_STATUS_SEND_REQUEST;
2668 hlen = LL_RESERVED_SPACE(dev);
2669 tlen = dev->needed_tailroom;
2670 if (po->has_vnet_hdr) {
2672 data += sizeof(*vnet_hdr);
2673 tp_len -= sizeof(*vnet_hdr);
2675 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2679 copylen = __virtio16_to_cpu(vio_le(),
2682 copylen = max_t(int, copylen, dev->hard_header_len);
2683 skb = sock_alloc_send_skb(&po->sk,
2684 hlen + tlen + sizeof(struct sockaddr_ll) +
2685 (copylen - dev->hard_header_len),
2688 if (unlikely(skb == NULL)) {
2689 /* we assume the socket was initially writeable ... */
2690 if (likely(len_sum > 0))
2694 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2695 addr, hlen, copylen, &sockc);
2696 if (likely(tp_len >= 0) &&
2697 tp_len > dev->mtu + reserve &&
2698 !po->has_vnet_hdr &&
2699 !packet_extra_vlan_len_allowed(dev, skb))
2702 if (unlikely(tp_len < 0)) {
2705 __packet_set_status(po, ph,
2706 TP_STATUS_AVAILABLE);
2707 packet_increment_head(&po->tx_ring);
2711 status = TP_STATUS_WRONG_FORMAT;
2717 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2723 skb->destructor = tpacket_destruct_skb;
2724 __packet_set_status(po, ph, TP_STATUS_SENDING);
2725 packet_inc_pending(&po->tx_ring);
2727 status = TP_STATUS_SEND_REQUEST;
2728 err = po->xmit(skb);
2729 if (unlikely(err > 0)) {
2730 err = net_xmit_errno(err);
2731 if (err && __packet_get_status(po, ph) ==
2732 TP_STATUS_AVAILABLE) {
2733 /* skb was destructed already */
2738 * skb was dropped but not destructed yet;
2739 * let's treat it like congestion or err < 0
2743 packet_increment_head(&po->tx_ring);
2745 } while (likely((ph != NULL) ||
2746 /* Note: packet_read_pending() might be slow if we have
2747 * to call it as it's per_cpu variable, but in fast-path
2748 * we already short-circuit the loop with the first
2749 * condition, and luckily don't have to go that path
2752 (need_wait && packet_read_pending(&po->tx_ring))));
2758 __packet_set_status(po, ph, status);
2763 mutex_unlock(&po->pg_vec_lock);
2767 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2768 size_t reserve, size_t len,
2769 size_t linear, int noblock,
2772 struct sk_buff *skb;
2774 /* Under a page? Don't bother with paged skb. */
2775 if (prepad + len < PAGE_SIZE || !linear)
2778 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2783 skb_reserve(skb, reserve);
2784 skb_put(skb, linear);
2785 skb->data_len = len - linear;
2786 skb->len += len - linear;
2791 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2793 struct sock *sk = sock->sk;
2794 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2795 struct sk_buff *skb;
2796 struct net_device *dev;
2798 unsigned char *addr;
2799 int err, reserve = 0;
2800 struct sockcm_cookie sockc;
2801 struct virtio_net_hdr vnet_hdr = { 0 };
2803 struct packet_sock *po = pkt_sk(sk);
2804 bool has_vnet_hdr = false;
2805 int hlen, tlen, linear;
2809 * Get and verify the address.
2812 if (likely(saddr == NULL)) {
2813 dev = packet_cached_dev_get(po);
2818 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2820 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2822 proto = saddr->sll_protocol;
2823 addr = saddr->sll_addr;
2824 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2828 if (unlikely(dev == NULL))
2831 if (unlikely(!(dev->flags & IFF_UP)))
2834 sockcm_init(&sockc, sk);
2835 sockc.mark = sk->sk_mark;
2836 if (msg->msg_controllen) {
2837 err = sock_cmsg_send(sk, msg, &sockc);
2842 if (sock->type == SOCK_RAW)
2843 reserve = dev->hard_header_len;
2844 if (po->has_vnet_hdr) {
2845 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2848 has_vnet_hdr = true;
2851 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2852 if (!netif_supports_nofcs(dev)) {
2853 err = -EPROTONOSUPPORT;
2856 extra_len = 4; /* We're doing our own CRC */
2860 if (!vnet_hdr.gso_type &&
2861 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2865 hlen = LL_RESERVED_SPACE(dev);
2866 tlen = dev->needed_tailroom;
2867 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2868 linear = max(linear, min_t(int, len, dev->hard_header_len));
2869 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2870 msg->msg_flags & MSG_DONTWAIT, &err);
2874 skb_reset_network_header(skb);
2877 if (sock->type == SOCK_DGRAM) {
2878 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2879 if (unlikely(offset < 0))
2881 } else if (reserve) {
2882 skb_reserve(skb, -reserve);
2885 /* Returns -EFAULT on error */
2886 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2890 if (sock->type == SOCK_RAW &&
2891 !dev_validate_header(dev, skb->data, len)) {
2896 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2898 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2899 !packet_extra_vlan_len_allowed(dev, skb)) {
2904 skb->protocol = proto;
2906 skb->priority = sk->sk_priority;
2907 skb->mark = sockc.mark;
2908 skb->tstamp = sockc.transmit_time;
2911 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2914 len += sizeof(vnet_hdr);
2917 skb_probe_transport_header(skb, reserve);
2919 if (unlikely(extra_len == 4))
2922 err = po->xmit(skb);
2923 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2939 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2941 struct sock *sk = sock->sk;
2942 struct packet_sock *po = pkt_sk(sk);
2944 if (po->tx_ring.pg_vec)
2945 return tpacket_snd(po, msg);
2947 return packet_snd(sock, msg, len);
2951 * Close a PACKET socket. This is fairly simple. We immediately go
2952 * to 'closed' state and remove our protocol entry in the device list.
2955 static int packet_release(struct socket *sock)
2957 struct sock *sk = sock->sk;
2958 struct packet_sock *po;
2959 struct packet_fanout *f;
2961 union tpacket_req_u req_u;
2969 mutex_lock(&net->packet.sklist_lock);
2970 sk_del_node_init_rcu(sk);
2971 mutex_unlock(&net->packet.sklist_lock);
2974 sock_prot_inuse_add(net, sk->sk_prot, -1);
2977 spin_lock(&po->bind_lock);
2978 unregister_prot_hook(sk, false);
2979 packet_cached_dev_reset(po);
2981 if (po->prot_hook.dev) {
2982 dev_put(po->prot_hook.dev);
2983 po->prot_hook.dev = NULL;
2985 spin_unlock(&po->bind_lock);
2987 packet_flush_mclist(sk);
2990 if (po->rx_ring.pg_vec) {
2991 memset(&req_u, 0, sizeof(req_u));
2992 packet_set_ring(sk, &req_u, 1, 0);
2995 if (po->tx_ring.pg_vec) {
2996 memset(&req_u, 0, sizeof(req_u));
2997 packet_set_ring(sk, &req_u, 1, 1);
3001 f = fanout_release(sk);
3006 kfree(po->rollover);
3007 fanout_release_data(f);
3011 * Now the socket is dead. No more input will appear.
3018 skb_queue_purge(&sk->sk_receive_queue);
3019 packet_free_pending(po);
3020 sk_refcnt_debug_release(sk);
3027 * Attach a packet hook.
3030 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3033 struct packet_sock *po = pkt_sk(sk);
3034 struct net_device *dev_curr;
3037 struct net_device *dev = NULL;
3039 bool unlisted = false;
3042 spin_lock(&po->bind_lock);
3051 dev = dev_get_by_name_rcu(sock_net(sk), name);
3056 } else if (ifindex) {
3057 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3067 proto_curr = po->prot_hook.type;
3068 dev_curr = po->prot_hook.dev;
3070 need_rehook = proto_curr != proto || dev_curr != dev;
3075 /* prevents packet_notifier() from calling
3076 * register_prot_hook()
3079 __unregister_prot_hook(sk, true);
3081 dev_curr = po->prot_hook.dev;
3083 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3087 BUG_ON(po->running);
3089 po->prot_hook.type = proto;
3091 if (unlikely(unlisted)) {
3093 po->prot_hook.dev = NULL;
3095 packet_cached_dev_reset(po);
3097 po->prot_hook.dev = dev;
3098 po->ifindex = dev ? dev->ifindex : 0;
3099 packet_cached_dev_assign(po, dev);
3105 if (proto == 0 || !need_rehook)
3108 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3109 register_prot_hook(sk);
3111 sk->sk_err = ENETDOWN;
3112 if (!sock_flag(sk, SOCK_DEAD))
3113 sk->sk_error_report(sk);
3118 spin_unlock(&po->bind_lock);
3124 * Bind a packet socket to a device
3127 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3130 struct sock *sk = sock->sk;
3131 char name[sizeof(uaddr->sa_data) + 1];
3137 if (addr_len != sizeof(struct sockaddr))
3139 /* uaddr->sa_data comes from the userspace, it's not guaranteed to be
3142 memcpy(name, uaddr->sa_data, sizeof(uaddr->sa_data));
3143 name[sizeof(uaddr->sa_data)] = 0;
3145 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3148 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3150 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3151 struct sock *sk = sock->sk;
3157 if (addr_len < sizeof(struct sockaddr_ll))
3159 if (sll->sll_family != AF_PACKET)
3162 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3163 sll->sll_protocol ? : pkt_sk(sk)->num);
3166 static struct proto packet_proto = {
3168 .owner = THIS_MODULE,
3169 .obj_size = sizeof(struct packet_sock),
3173 * Create a packet of type SOCK_PACKET.
3176 static int packet_create(struct net *net, struct socket *sock, int protocol,
3180 struct packet_sock *po;
3181 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3184 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3186 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3187 sock->type != SOCK_PACKET)
3188 return -ESOCKTNOSUPPORT;
3190 sock->state = SS_UNCONNECTED;
3193 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3197 sock->ops = &packet_ops;
3198 if (sock->type == SOCK_PACKET)
3199 sock->ops = &packet_ops_spkt;
3201 sock_init_data(sock, sk);
3204 sk->sk_family = PF_PACKET;
3206 po->xmit = dev_queue_xmit;
3208 err = packet_alloc_pending(po);
3212 packet_cached_dev_reset(po);
3214 sk->sk_destruct = packet_sock_destruct;
3215 sk_refcnt_debug_inc(sk);
3218 * Attach a protocol block
3221 spin_lock_init(&po->bind_lock);
3222 mutex_init(&po->pg_vec_lock);
3223 po->rollover = NULL;
3224 po->prot_hook.func = packet_rcv;
3226 if (sock->type == SOCK_PACKET)
3227 po->prot_hook.func = packet_rcv_spkt;
3229 po->prot_hook.af_packet_priv = sk;
3232 po->prot_hook.type = proto;
3233 __register_prot_hook(sk);
3236 mutex_lock(&net->packet.sklist_lock);
3237 sk_add_node_rcu(sk, &net->packet.sklist);
3238 mutex_unlock(&net->packet.sklist_lock);
3241 sock_prot_inuse_add(net, &packet_proto, 1);
3252 * Pull a packet from our receive queue and hand it to the user.
3253 * If necessary we block.
3256 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3259 struct sock *sk = sock->sk;
3260 struct sk_buff *skb;
3262 int vnet_hdr_len = 0;
3263 unsigned int origlen = 0;
3266 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3270 /* What error should we return now? EUNATTACH? */
3271 if (pkt_sk(sk)->ifindex < 0)
3275 if (flags & MSG_ERRQUEUE) {
3276 err = sock_recv_errqueue(sk, msg, len,
3277 SOL_PACKET, PACKET_TX_TIMESTAMP);
3282 * Call the generic datagram receiver. This handles all sorts
3283 * of horrible races and re-entrancy so we can forget about it
3284 * in the protocol layers.
3286 * Now it will return ENETDOWN, if device have just gone down,
3287 * but then it will block.
3290 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3293 * An error occurred so return it. Because skb_recv_datagram()
3294 * handles the blocking we don't see and worry about blocking
3301 if (pkt_sk(sk)->pressure)
3302 packet_rcv_has_room(pkt_sk(sk), NULL);
3304 if (pkt_sk(sk)->has_vnet_hdr) {
3305 err = packet_rcv_vnet(msg, skb, &len);
3308 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3311 /* You lose any data beyond the buffer you gave. If it worries
3312 * a user program they can ask the device for its MTU
3318 msg->msg_flags |= MSG_TRUNC;
3321 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3325 if (sock->type != SOCK_PACKET) {
3326 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3328 /* Original length was stored in sockaddr_ll fields */
3329 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3330 sll->sll_family = AF_PACKET;
3331 sll->sll_protocol = skb->protocol;
3334 sock_recv_ts_and_drops(msg, sk, skb);
3336 if (msg->msg_name) {
3337 /* If the address length field is there to be filled
3338 * in, we fill it in now.
3340 if (sock->type == SOCK_PACKET) {
3341 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3342 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3344 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3346 msg->msg_namelen = sll->sll_halen +
3347 offsetof(struct sockaddr_ll, sll_addr);
3349 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3353 if (pkt_sk(sk)->auxdata) {
3354 struct tpacket_auxdata aux;
3356 aux.tp_status = TP_STATUS_USER;
3357 if (skb->ip_summed == CHECKSUM_PARTIAL)
3358 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3359 else if (skb->pkt_type != PACKET_OUTGOING &&
3360 (skb->ip_summed == CHECKSUM_COMPLETE ||
3361 skb_csum_unnecessary(skb)))
3362 aux.tp_status |= TP_STATUS_CSUM_VALID;
3364 aux.tp_len = origlen;
3365 aux.tp_snaplen = skb->len;
3367 aux.tp_net = skb_network_offset(skb);
3368 if (skb_vlan_tag_present(skb)) {
3369 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3370 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3371 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3373 aux.tp_vlan_tci = 0;
3374 aux.tp_vlan_tpid = 0;
3376 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3380 * Free or return the buffer as appropriate. Again this
3381 * hides all the races and re-entrancy issues from us.
3383 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3386 skb_free_datagram(sk, skb);
3391 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3394 struct net_device *dev;
3395 struct sock *sk = sock->sk;
3400 uaddr->sa_family = AF_PACKET;
3401 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3403 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3405 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3408 return sizeof(*uaddr);
3411 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3414 struct net_device *dev;
3415 struct sock *sk = sock->sk;
3416 struct packet_sock *po = pkt_sk(sk);
3417 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3422 sll->sll_family = AF_PACKET;
3423 sll->sll_ifindex = po->ifindex;
3424 sll->sll_protocol = po->num;
3425 sll->sll_pkttype = 0;
3427 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3429 sll->sll_hatype = dev->type;
3430 sll->sll_halen = dev->addr_len;
3431 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3433 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3438 return offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3441 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3445 case PACKET_MR_MULTICAST:
3446 if (i->alen != dev->addr_len)
3449 return dev_mc_add(dev, i->addr);
3451 return dev_mc_del(dev, i->addr);
3453 case PACKET_MR_PROMISC:
3454 return dev_set_promiscuity(dev, what);
3455 case PACKET_MR_ALLMULTI:
3456 return dev_set_allmulti(dev, what);
3457 case PACKET_MR_UNICAST:
3458 if (i->alen != dev->addr_len)
3461 return dev_uc_add(dev, i->addr);
3463 return dev_uc_del(dev, i->addr);
3471 static void packet_dev_mclist_delete(struct net_device *dev,
3472 struct packet_mclist **mlp)
3474 struct packet_mclist *ml;
3476 while ((ml = *mlp) != NULL) {
3477 if (ml->ifindex == dev->ifindex) {
3478 packet_dev_mc(dev, ml, -1);
3486 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3488 struct packet_sock *po = pkt_sk(sk);
3489 struct packet_mclist *ml, *i;
3490 struct net_device *dev;
3496 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3501 if (mreq->mr_alen > dev->addr_len)
3505 i = kmalloc(sizeof(*i), GFP_KERNEL);
3510 for (ml = po->mclist; ml; ml = ml->next) {
3511 if (ml->ifindex == mreq->mr_ifindex &&
3512 ml->type == mreq->mr_type &&
3513 ml->alen == mreq->mr_alen &&
3514 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3516 /* Free the new element ... */
3522 i->type = mreq->mr_type;
3523 i->ifindex = mreq->mr_ifindex;
3524 i->alen = mreq->mr_alen;
3525 memcpy(i->addr, mreq->mr_address, i->alen);
3526 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3528 i->next = po->mclist;
3530 err = packet_dev_mc(dev, i, 1);
3532 po->mclist = i->next;
3541 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3543 struct packet_mclist *ml, **mlp;
3547 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3548 if (ml->ifindex == mreq->mr_ifindex &&
3549 ml->type == mreq->mr_type &&
3550 ml->alen == mreq->mr_alen &&
3551 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3552 if (--ml->count == 0) {
3553 struct net_device *dev;
3555 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3557 packet_dev_mc(dev, ml, -1);
3567 static void packet_flush_mclist(struct sock *sk)
3569 struct packet_sock *po = pkt_sk(sk);
3570 struct packet_mclist *ml;
3576 while ((ml = po->mclist) != NULL) {
3577 struct net_device *dev;
3579 po->mclist = ml->next;
3580 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3582 packet_dev_mc(dev, ml, -1);
3589 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3591 struct sock *sk = sock->sk;
3592 struct packet_sock *po = pkt_sk(sk);
3595 if (level != SOL_PACKET)
3596 return -ENOPROTOOPT;
3599 case PACKET_ADD_MEMBERSHIP:
3600 case PACKET_DROP_MEMBERSHIP:
3602 struct packet_mreq_max mreq;
3604 memset(&mreq, 0, sizeof(mreq));
3605 if (len < sizeof(struct packet_mreq))
3607 if (len > sizeof(mreq))
3609 if (copy_from_user(&mreq, optval, len))
3611 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3613 if (optname == PACKET_ADD_MEMBERSHIP)
3614 ret = packet_mc_add(sk, &mreq);
3616 ret = packet_mc_drop(sk, &mreq);
3620 case PACKET_RX_RING:
3621 case PACKET_TX_RING:
3623 union tpacket_req_u req_u;
3627 switch (po->tp_version) {
3630 len = sizeof(req_u.req);
3634 len = sizeof(req_u.req3);
3640 if (copy_from_user(&req_u.req, optval, len))
3643 ret = packet_set_ring(sk, &req_u, 0,
3644 optname == PACKET_TX_RING);
3649 case PACKET_COPY_THRESH:
3653 if (optlen != sizeof(val))
3655 if (copy_from_user(&val, optval, sizeof(val)))
3658 pkt_sk(sk)->copy_thresh = val;
3661 case PACKET_VERSION:
3665 if (optlen != sizeof(val))
3667 if (copy_from_user(&val, optval, sizeof(val)))
3678 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3681 po->tp_version = val;
3687 case PACKET_RESERVE:
3691 if (optlen != sizeof(val))
3693 if (copy_from_user(&val, optval, sizeof(val)))
3698 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3701 po->tp_reserve = val;
3711 if (optlen != sizeof(val))
3713 if (copy_from_user(&val, optval, sizeof(val)))
3717 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3720 po->tp_loss = !!val;
3726 case PACKET_AUXDATA:
3730 if (optlen < sizeof(val))
3732 if (copy_from_user(&val, optval, sizeof(val)))
3736 po->auxdata = !!val;
3740 case PACKET_ORIGDEV:
3744 if (optlen < sizeof(val))
3746 if (copy_from_user(&val, optval, sizeof(val)))
3750 po->origdev = !!val;
3754 case PACKET_VNET_HDR:
3758 if (sock->type != SOCK_RAW)
3760 if (optlen < sizeof(val))
3762 if (copy_from_user(&val, optval, sizeof(val)))
3766 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3769 po->has_vnet_hdr = !!val;
3775 case PACKET_TIMESTAMP:
3779 if (optlen != sizeof(val))
3781 if (copy_from_user(&val, optval, sizeof(val)))
3784 po->tp_tstamp = val;
3791 if (optlen != sizeof(val))
3793 if (copy_from_user(&val, optval, sizeof(val)))
3796 return fanout_add(sk, val & 0xffff, val >> 16);
3798 case PACKET_FANOUT_DATA:
3803 return fanout_set_data(po, optval, optlen);
3805 case PACKET_TX_HAS_OFF:
3809 if (optlen != sizeof(val))
3811 if (copy_from_user(&val, optval, sizeof(val)))
3815 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3818 po->tp_tx_has_off = !!val;
3824 case PACKET_QDISC_BYPASS:
3828 if (optlen != sizeof(val))
3830 if (copy_from_user(&val, optval, sizeof(val)))
3833 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3837 return -ENOPROTOOPT;
3841 static int packet_getsockopt(struct socket *sock, int level, int optname,
3842 char __user *optval, int __user *optlen)
3845 int val, lv = sizeof(val);
3846 struct sock *sk = sock->sk;
3847 struct packet_sock *po = pkt_sk(sk);
3849 union tpacket_stats_u st;
3850 struct tpacket_rollover_stats rstats;
3852 if (level != SOL_PACKET)
3853 return -ENOPROTOOPT;
3855 if (get_user(len, optlen))
3862 case PACKET_STATISTICS:
3863 spin_lock_bh(&sk->sk_receive_queue.lock);
3864 memcpy(&st, &po->stats, sizeof(st));
3865 memset(&po->stats, 0, sizeof(po->stats));
3866 spin_unlock_bh(&sk->sk_receive_queue.lock);
3868 if (po->tp_version == TPACKET_V3) {
3869 lv = sizeof(struct tpacket_stats_v3);
3870 st.stats3.tp_packets += st.stats3.tp_drops;
3873 lv = sizeof(struct tpacket_stats);
3874 st.stats1.tp_packets += st.stats1.tp_drops;
3879 case PACKET_AUXDATA:
3882 case PACKET_ORIGDEV:
3885 case PACKET_VNET_HDR:
3886 val = po->has_vnet_hdr;
3888 case PACKET_VERSION:
3889 val = po->tp_version;
3892 if (len > sizeof(int))
3894 if (len < sizeof(int))
3896 if (copy_from_user(&val, optval, len))
3900 val = sizeof(struct tpacket_hdr);
3903 val = sizeof(struct tpacket2_hdr);
3906 val = sizeof(struct tpacket3_hdr);
3912 case PACKET_RESERVE:
3913 val = po->tp_reserve;
3918 case PACKET_TIMESTAMP:
3919 val = po->tp_tstamp;
3923 ((u32)po->fanout->id |
3924 ((u32)po->fanout->type << 16) |
3925 ((u32)po->fanout->flags << 24)) :
3928 case PACKET_ROLLOVER_STATS:
3931 rstats.tp_all = atomic_long_read(&po->rollover->num);
3932 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3933 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3935 lv = sizeof(rstats);
3937 case PACKET_TX_HAS_OFF:
3938 val = po->tp_tx_has_off;
3940 case PACKET_QDISC_BYPASS:
3941 val = packet_use_direct_xmit(po);
3944 return -ENOPROTOOPT;
3949 if (put_user(len, optlen))
3951 if (copy_to_user(optval, data, len))
3957 #ifdef CONFIG_COMPAT
3958 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3959 char __user *optval, unsigned int optlen)
3961 struct packet_sock *po = pkt_sk(sock->sk);
3963 if (level != SOL_PACKET)
3964 return -ENOPROTOOPT;
3966 if (optname == PACKET_FANOUT_DATA &&
3967 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3968 optval = (char __user *)get_compat_bpf_fprog(optval);
3971 optlen = sizeof(struct sock_fprog);
3974 return packet_setsockopt(sock, level, optname, optval, optlen);
3978 static int packet_notifier(struct notifier_block *this,
3979 unsigned long msg, void *ptr)
3982 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3983 struct net *net = dev_net(dev);
3986 sk_for_each_rcu(sk, &net->packet.sklist) {
3987 struct packet_sock *po = pkt_sk(sk);
3990 case NETDEV_UNREGISTER:
3992 packet_dev_mclist_delete(dev, &po->mclist);
3996 if (dev->ifindex == po->ifindex) {
3997 spin_lock(&po->bind_lock);
3999 __unregister_prot_hook(sk, false);
4000 sk->sk_err = ENETDOWN;
4001 if (!sock_flag(sk, SOCK_DEAD))
4002 sk->sk_error_report(sk);
4004 if (msg == NETDEV_UNREGISTER) {
4005 packet_cached_dev_reset(po);
4007 if (po->prot_hook.dev)
4008 dev_put(po->prot_hook.dev);
4009 po->prot_hook.dev = NULL;
4011 spin_unlock(&po->bind_lock);
4015 if (dev->ifindex == po->ifindex) {
4016 spin_lock(&po->bind_lock);
4018 register_prot_hook(sk);
4019 spin_unlock(&po->bind_lock);
4029 static int packet_ioctl(struct socket *sock, unsigned int cmd,
4032 struct sock *sk = sock->sk;
4037 int amount = sk_wmem_alloc_get(sk);
4039 return put_user(amount, (int __user *)arg);
4043 struct sk_buff *skb;
4046 spin_lock_bh(&sk->sk_receive_queue.lock);
4047 skb = skb_peek(&sk->sk_receive_queue);
4050 spin_unlock_bh(&sk->sk_receive_queue.lock);
4051 return put_user(amount, (int __user *)arg);
4054 return sock_get_timestamp(sk, (struct timeval __user *)arg);
4056 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4066 case SIOCGIFBRDADDR:
4067 case SIOCSIFBRDADDR:
4068 case SIOCGIFNETMASK:
4069 case SIOCSIFNETMASK:
4070 case SIOCGIFDSTADDR:
4071 case SIOCSIFDSTADDR:
4073 return inet_dgram_ops.ioctl(sock, cmd, arg);
4077 return -ENOIOCTLCMD;
4082 static __poll_t packet_poll(struct file *file, struct socket *sock,
4085 struct sock *sk = sock->sk;
4086 struct packet_sock *po = pkt_sk(sk);
4087 __poll_t mask = datagram_poll(file, sock, wait);
4089 spin_lock_bh(&sk->sk_receive_queue.lock);
4090 if (po->rx_ring.pg_vec) {
4091 if (!packet_previous_rx_frame(po, &po->rx_ring,
4093 mask |= EPOLLIN | EPOLLRDNORM;
4095 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4097 spin_unlock_bh(&sk->sk_receive_queue.lock);
4098 spin_lock_bh(&sk->sk_write_queue.lock);
4099 if (po->tx_ring.pg_vec) {
4100 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4101 mask |= EPOLLOUT | EPOLLWRNORM;
4103 spin_unlock_bh(&sk->sk_write_queue.lock);
4108 /* Dirty? Well, I still did not learn better way to account
4112 static void packet_mm_open(struct vm_area_struct *vma)
4114 struct file *file = vma->vm_file;
4115 struct socket *sock = file->private_data;
4116 struct sock *sk = sock->sk;
4119 atomic_inc(&pkt_sk(sk)->mapped);
4122 static void packet_mm_close(struct vm_area_struct *vma)
4124 struct file *file = vma->vm_file;
4125 struct socket *sock = file->private_data;
4126 struct sock *sk = sock->sk;
4129 atomic_dec(&pkt_sk(sk)->mapped);
4132 static const struct vm_operations_struct packet_mmap_ops = {
4133 .open = packet_mm_open,
4134 .close = packet_mm_close,
4137 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4142 for (i = 0; i < len; i++) {
4143 if (likely(pg_vec[i].buffer)) {
4144 if (is_vmalloc_addr(pg_vec[i].buffer))
4145 vfree(pg_vec[i].buffer);
4147 free_pages((unsigned long)pg_vec[i].buffer,
4149 pg_vec[i].buffer = NULL;
4155 static char *alloc_one_pg_vec_page(unsigned long order)
4158 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4159 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4161 buffer = (char *) __get_free_pages(gfp_flags, order);
4165 /* __get_free_pages failed, fall back to vmalloc */
4166 buffer = vzalloc(array_size((1 << order), PAGE_SIZE));
4170 /* vmalloc failed, lets dig into swap here */
4171 gfp_flags &= ~__GFP_NORETRY;
4172 buffer = (char *) __get_free_pages(gfp_flags, order);
4176 /* complete and utter failure */
4180 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4182 unsigned int block_nr = req->tp_block_nr;
4186 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4187 if (unlikely(!pg_vec))
4190 for (i = 0; i < block_nr; i++) {
4191 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4192 if (unlikely(!pg_vec[i].buffer))
4193 goto out_free_pgvec;
4200 free_pg_vec(pg_vec, order, block_nr);
4205 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4206 int closing, int tx_ring)
4208 struct pgv *pg_vec = NULL;
4209 struct packet_sock *po = pkt_sk(sk);
4210 int was_running, order = 0;
4211 struct packet_ring_buffer *rb;
4212 struct sk_buff_head *rb_queue;
4215 /* Added to avoid minimal code churn */
4216 struct tpacket_req *req = &req_u->req;
4218 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4219 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4223 if (atomic_read(&po->mapped))
4225 if (packet_read_pending(rb))
4229 if (req->tp_block_nr) {
4230 /* Sanity tests and some calculations */
4232 if (unlikely(rb->pg_vec))
4235 switch (po->tp_version) {
4237 po->tp_hdrlen = TPACKET_HDRLEN;
4240 po->tp_hdrlen = TPACKET2_HDRLEN;
4243 po->tp_hdrlen = TPACKET3_HDRLEN;
4248 if (unlikely((int)req->tp_block_size <= 0))
4250 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4252 if (po->tp_version >= TPACKET_V3 &&
4253 req->tp_block_size <=
4254 BLK_PLUS_PRIV((u64)req_u->req3.tp_sizeof_priv) + sizeof(struct tpacket3_hdr))
4256 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4259 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4262 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4263 if (unlikely(rb->frames_per_block == 0))
4265 if (unlikely(req->tp_block_size > UINT_MAX / req->tp_block_nr))
4267 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4272 order = get_order(req->tp_block_size);
4273 pg_vec = alloc_pg_vec(req, order);
4274 if (unlikely(!pg_vec))
4276 switch (po->tp_version) {
4278 /* Block transmit is not supported yet */
4280 init_prb_bdqc(po, rb, pg_vec, req_u);
4282 struct tpacket_req3 *req3 = &req_u->req3;
4284 if (req3->tp_retire_blk_tov ||
4285 req3->tp_sizeof_priv ||
4286 req3->tp_feature_req_word) {
4299 if (unlikely(req->tp_frame_nr))
4304 /* Detach socket from network */
4305 spin_lock(&po->bind_lock);
4306 was_running = po->running;
4310 __unregister_prot_hook(sk, false);
4312 spin_unlock(&po->bind_lock);
4317 mutex_lock(&po->pg_vec_lock);
4318 if (closing || atomic_read(&po->mapped) == 0) {
4320 spin_lock_bh(&rb_queue->lock);
4321 swap(rb->pg_vec, pg_vec);
4322 rb->frame_max = (req->tp_frame_nr - 1);
4324 rb->frame_size = req->tp_frame_size;
4325 spin_unlock_bh(&rb_queue->lock);
4327 swap(rb->pg_vec_order, order);
4328 swap(rb->pg_vec_len, req->tp_block_nr);
4330 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4331 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4332 tpacket_rcv : packet_rcv;
4333 skb_queue_purge(rb_queue);
4334 if (atomic_read(&po->mapped))
4335 pr_err("packet_mmap: vma is busy: %d\n",
4336 atomic_read(&po->mapped));
4338 mutex_unlock(&po->pg_vec_lock);
4340 spin_lock(&po->bind_lock);
4343 register_prot_hook(sk);
4345 spin_unlock(&po->bind_lock);
4346 if (pg_vec && (po->tp_version > TPACKET_V2)) {
4347 /* Because we don't support block-based V3 on tx-ring */
4349 prb_shutdown_retire_blk_timer(po, rb_queue);
4353 free_pg_vec(pg_vec, order, req->tp_block_nr);
4358 static int packet_mmap(struct file *file, struct socket *sock,
4359 struct vm_area_struct *vma)
4361 struct sock *sk = sock->sk;
4362 struct packet_sock *po = pkt_sk(sk);
4363 unsigned long size, expected_size;
4364 struct packet_ring_buffer *rb;
4365 unsigned long start;
4372 mutex_lock(&po->pg_vec_lock);
4375 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4377 expected_size += rb->pg_vec_len
4383 if (expected_size == 0)
4386 size = vma->vm_end - vma->vm_start;
4387 if (size != expected_size)
4390 start = vma->vm_start;
4391 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4392 if (rb->pg_vec == NULL)
4395 for (i = 0; i < rb->pg_vec_len; i++) {
4397 void *kaddr = rb->pg_vec[i].buffer;
4400 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4401 page = pgv_to_page(kaddr);
4402 err = vm_insert_page(vma, start, page);
4411 atomic_inc(&po->mapped);
4412 vma->vm_ops = &packet_mmap_ops;
4416 mutex_unlock(&po->pg_vec_lock);
4420 static const struct proto_ops packet_ops_spkt = {
4421 .family = PF_PACKET,
4422 .owner = THIS_MODULE,
4423 .release = packet_release,
4424 .bind = packet_bind_spkt,
4425 .connect = sock_no_connect,
4426 .socketpair = sock_no_socketpair,
4427 .accept = sock_no_accept,
4428 .getname = packet_getname_spkt,
4429 .poll = datagram_poll,
4430 .ioctl = packet_ioctl,
4431 .listen = sock_no_listen,
4432 .shutdown = sock_no_shutdown,
4433 .setsockopt = sock_no_setsockopt,
4434 .getsockopt = sock_no_getsockopt,
4435 .sendmsg = packet_sendmsg_spkt,
4436 .recvmsg = packet_recvmsg,
4437 .mmap = sock_no_mmap,
4438 .sendpage = sock_no_sendpage,
4441 static const struct proto_ops packet_ops = {
4442 .family = PF_PACKET,
4443 .owner = THIS_MODULE,
4444 .release = packet_release,
4445 .bind = packet_bind,
4446 .connect = sock_no_connect,
4447 .socketpair = sock_no_socketpair,
4448 .accept = sock_no_accept,
4449 .getname = packet_getname,
4450 .poll = packet_poll,
4451 .ioctl = packet_ioctl,
4452 .listen = sock_no_listen,
4453 .shutdown = sock_no_shutdown,
4454 .setsockopt = packet_setsockopt,
4455 .getsockopt = packet_getsockopt,
4456 #ifdef CONFIG_COMPAT
4457 .compat_setsockopt = compat_packet_setsockopt,
4459 .sendmsg = packet_sendmsg,
4460 .recvmsg = packet_recvmsg,
4461 .mmap = packet_mmap,
4462 .sendpage = sock_no_sendpage,
4465 static const struct net_proto_family packet_family_ops = {
4466 .family = PF_PACKET,
4467 .create = packet_create,
4468 .owner = THIS_MODULE,
4471 static struct notifier_block packet_netdev_notifier = {
4472 .notifier_call = packet_notifier,
4475 #ifdef CONFIG_PROC_FS
4477 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4480 struct net *net = seq_file_net(seq);
4483 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4486 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4488 struct net *net = seq_file_net(seq);
4489 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4492 static void packet_seq_stop(struct seq_file *seq, void *v)
4498 static int packet_seq_show(struct seq_file *seq, void *v)
4500 if (v == SEQ_START_TOKEN)
4501 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4503 struct sock *s = sk_entry(v);
4504 const struct packet_sock *po = pkt_sk(s);
4507 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4509 refcount_read(&s->sk_refcnt),
4514 atomic_read(&s->sk_rmem_alloc),
4515 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4522 static const struct seq_operations packet_seq_ops = {
4523 .start = packet_seq_start,
4524 .next = packet_seq_next,
4525 .stop = packet_seq_stop,
4526 .show = packet_seq_show,
4530 static int __net_init packet_net_init(struct net *net)
4532 mutex_init(&net->packet.sklist_lock);
4533 INIT_HLIST_HEAD(&net->packet.sklist);
4535 if (!proc_create_net("packet", 0, net->proc_net, &packet_seq_ops,
4536 sizeof(struct seq_net_private)))
4542 static void __net_exit packet_net_exit(struct net *net)
4544 remove_proc_entry("packet", net->proc_net);
4545 WARN_ON_ONCE(!hlist_empty(&net->packet.sklist));
4548 static struct pernet_operations packet_net_ops = {
4549 .init = packet_net_init,
4550 .exit = packet_net_exit,
4554 static void __exit packet_exit(void)
4556 unregister_netdevice_notifier(&packet_netdev_notifier);
4557 unregister_pernet_subsys(&packet_net_ops);
4558 sock_unregister(PF_PACKET);
4559 proto_unregister(&packet_proto);
4562 static int __init packet_init(void)
4564 int rc = proto_register(&packet_proto, 0);
4569 sock_register(&packet_family_ops);
4570 register_pernet_subsys(&packet_net_ops);
4571 register_netdevice_notifier(&packet_netdev_notifier);
4576 module_init(packet_init);
4577 module_exit(packet_exit);
4578 MODULE_LICENSE("GPL");
4579 MODULE_ALIAS_NETPROTO(PF_PACKET);