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 PGV_FROM_VMALLOC 1
182 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
183 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
184 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
185 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
186 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
187 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
188 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
191 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
192 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
193 struct packet_type *pt, struct net_device *orig_dev);
195 static void *packet_previous_frame(struct packet_sock *po,
196 struct packet_ring_buffer *rb,
198 static void packet_increment_head(struct packet_ring_buffer *buff);
199 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
200 struct tpacket_block_desc *);
201 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
202 struct packet_sock *);
203 static void prb_retire_current_block(struct tpacket_kbdq_core *,
204 struct packet_sock *, unsigned int status);
205 static int prb_queue_frozen(struct tpacket_kbdq_core *);
206 static void prb_open_block(struct tpacket_kbdq_core *,
207 struct tpacket_block_desc *);
208 static void prb_retire_rx_blk_timer_expired(unsigned long);
209 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
210 static void prb_init_blk_timer(struct packet_sock *,
211 struct tpacket_kbdq_core *,
212 void (*func) (unsigned long));
213 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
214 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
217 struct tpacket3_hdr *);
218 static void packet_flush_mclist(struct sock *sk);
220 struct packet_skb_cb {
222 struct sockaddr_pkt pkt;
224 /* Trick: alias skb original length with
225 * ll.sll_family and ll.protocol in order
228 unsigned int origlen;
229 struct sockaddr_ll ll;
234 #define vio_le() virtio_legacy_is_little_endian()
236 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
238 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
239 #define GET_PBLOCK_DESC(x, bid) \
240 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
241 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
242 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
243 #define GET_NEXT_PRB_BLK_NUM(x) \
244 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
245 ((x)->kactive_blk_num+1) : 0)
247 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
248 static void __fanout_link(struct sock *sk, struct packet_sock *po);
250 static int packet_direct_xmit(struct sk_buff *skb)
252 struct net_device *dev = skb->dev;
253 struct sk_buff *orig_skb = skb;
254 struct netdev_queue *txq;
255 int ret = NETDEV_TX_BUSY;
257 if (unlikely(!netif_running(dev) ||
258 !netif_carrier_ok(dev)))
261 skb = validate_xmit_skb_list(skb, dev);
265 txq = skb_get_tx_queue(dev, skb);
269 HARD_TX_LOCK(dev, txq, smp_processor_id());
270 if (!netif_xmit_frozen_or_drv_stopped(txq))
271 ret = netdev_start_xmit(skb, dev, txq, false);
272 HARD_TX_UNLOCK(dev, txq);
276 if (!dev_xmit_complete(ret))
281 atomic_long_inc(&dev->tx_dropped);
283 return NET_XMIT_DROP;
286 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
288 struct net_device *dev;
291 dev = rcu_dereference(po->cached_dev);
299 static void packet_cached_dev_assign(struct packet_sock *po,
300 struct net_device *dev)
302 rcu_assign_pointer(po->cached_dev, dev);
305 static void packet_cached_dev_reset(struct packet_sock *po)
307 RCU_INIT_POINTER(po->cached_dev, NULL);
310 static bool packet_use_direct_xmit(const struct packet_sock *po)
312 return po->xmit == packet_direct_xmit;
315 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
317 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
320 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
322 const struct net_device_ops *ops = dev->netdev_ops;
325 if (ops->ndo_select_queue) {
326 queue_index = ops->ndo_select_queue(dev, skb, NULL,
327 __packet_pick_tx_queue);
328 queue_index = netdev_cap_txqueue(dev, queue_index);
330 queue_index = __packet_pick_tx_queue(dev, skb);
333 skb_set_queue_mapping(skb, queue_index);
336 /* register_prot_hook must be invoked with the po->bind_lock held,
337 * or from a context in which asynchronous accesses to the packet
338 * socket is not possible (packet_create()).
340 static void register_prot_hook(struct sock *sk)
342 struct packet_sock *po = pkt_sk(sk);
346 __fanout_link(sk, po);
348 dev_add_pack(&po->prot_hook);
355 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
356 * held. If the sync parameter is true, we will temporarily drop
357 * the po->bind_lock and do a synchronize_net to make sure no
358 * asynchronous packet processing paths still refer to the elements
359 * of po->prot_hook. If the sync parameter is false, it is the
360 * callers responsibility to take care of this.
362 static void __unregister_prot_hook(struct sock *sk, bool sync)
364 struct packet_sock *po = pkt_sk(sk);
369 __fanout_unlink(sk, po);
371 __dev_remove_pack(&po->prot_hook);
376 spin_unlock(&po->bind_lock);
378 spin_lock(&po->bind_lock);
382 static void unregister_prot_hook(struct sock *sk, bool sync)
384 struct packet_sock *po = pkt_sk(sk);
387 __unregister_prot_hook(sk, sync);
390 static inline struct page * __pure pgv_to_page(void *addr)
392 if (is_vmalloc_addr(addr))
393 return vmalloc_to_page(addr);
394 return virt_to_page(addr);
397 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
399 union tpacket_uhdr h;
402 switch (po->tp_version) {
404 h.h1->tp_status = status;
405 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
408 h.h2->tp_status = status;
409 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
412 h.h3->tp_status = status;
413 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
416 WARN(1, "TPACKET version not supported.\n");
423 static int __packet_get_status(struct packet_sock *po, void *frame)
425 union tpacket_uhdr h;
430 switch (po->tp_version) {
432 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
433 return h.h1->tp_status;
435 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
436 return h.h2->tp_status;
438 flush_dcache_page(pgv_to_page(&h.h3->tp_status));
439 return h.h3->tp_status;
441 WARN(1, "TPACKET version not supported.\n");
447 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
450 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
453 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
454 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
455 return TP_STATUS_TS_RAW_HARDWARE;
457 if (ktime_to_timespec_cond(skb->tstamp, ts))
458 return TP_STATUS_TS_SOFTWARE;
463 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
466 union tpacket_uhdr h;
470 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
474 switch (po->tp_version) {
476 h.h1->tp_sec = ts.tv_sec;
477 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
480 h.h2->tp_sec = ts.tv_sec;
481 h.h2->tp_nsec = ts.tv_nsec;
484 h.h3->tp_sec = ts.tv_sec;
485 h.h3->tp_nsec = ts.tv_nsec;
488 WARN(1, "TPACKET version not supported.\n");
492 /* one flush is safe, as both fields always lie on the same cacheline */
493 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
499 static void *packet_lookup_frame(struct packet_sock *po,
500 struct packet_ring_buffer *rb,
501 unsigned int position,
504 unsigned int pg_vec_pos, frame_offset;
505 union tpacket_uhdr h;
507 pg_vec_pos = position / rb->frames_per_block;
508 frame_offset = position % rb->frames_per_block;
510 h.raw = rb->pg_vec[pg_vec_pos].buffer +
511 (frame_offset * rb->frame_size);
513 if (status != __packet_get_status(po, h.raw))
519 static void *packet_current_frame(struct packet_sock *po,
520 struct packet_ring_buffer *rb,
523 return packet_lookup_frame(po, rb, rb->head, status);
526 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
528 del_timer_sync(&pkc->retire_blk_timer);
531 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
532 struct sk_buff_head *rb_queue)
534 struct tpacket_kbdq_core *pkc;
536 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
538 spin_lock_bh(&rb_queue->lock);
539 pkc->delete_blk_timer = 1;
540 spin_unlock_bh(&rb_queue->lock);
542 prb_del_retire_blk_timer(pkc);
545 static void prb_init_blk_timer(struct packet_sock *po,
546 struct tpacket_kbdq_core *pkc,
547 void (*func) (unsigned long))
549 init_timer(&pkc->retire_blk_timer);
550 pkc->retire_blk_timer.data = (long)po;
551 pkc->retire_blk_timer.function = func;
552 pkc->retire_blk_timer.expires = jiffies;
555 static void prb_setup_retire_blk_timer(struct packet_sock *po)
557 struct tpacket_kbdq_core *pkc;
559 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
560 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
563 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
564 int blk_size_in_bytes)
566 struct net_device *dev;
567 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
568 struct ethtool_link_ksettings ecmd;
572 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
573 if (unlikely(!dev)) {
575 return DEFAULT_PRB_RETIRE_TOV;
577 err = __ethtool_get_link_ksettings(dev, &ecmd);
581 * If the link speed is so slow you don't really
582 * need to worry about perf anyways
584 if (ecmd.base.speed < SPEED_1000 ||
585 ecmd.base.speed == SPEED_UNKNOWN) {
586 return DEFAULT_PRB_RETIRE_TOV;
589 div = ecmd.base.speed / 1000;
593 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
605 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
606 union tpacket_req_u *req_u)
608 p1->feature_req_word = req_u->req3.tp_feature_req_word;
611 static void init_prb_bdqc(struct packet_sock *po,
612 struct packet_ring_buffer *rb,
614 union tpacket_req_u *req_u)
616 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
617 struct tpacket_block_desc *pbd;
619 memset(p1, 0x0, sizeof(*p1));
621 p1->knxt_seq_num = 1;
623 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
624 p1->pkblk_start = pg_vec[0].buffer;
625 p1->kblk_size = req_u->req3.tp_block_size;
626 p1->knum_blocks = req_u->req3.tp_block_nr;
627 p1->hdrlen = po->tp_hdrlen;
628 p1->version = po->tp_version;
629 p1->last_kactive_blk_num = 0;
630 po->stats.stats3.tp_freeze_q_cnt = 0;
631 if (req_u->req3.tp_retire_blk_tov)
632 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
634 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
635 req_u->req3.tp_block_size);
636 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
637 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
639 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
640 prb_init_ft_ops(p1, req_u);
641 prb_setup_retire_blk_timer(po);
642 prb_open_block(p1, pbd);
645 /* Do NOT update the last_blk_num first.
646 * Assumes sk_buff_head lock is held.
648 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
650 mod_timer(&pkc->retire_blk_timer,
651 jiffies + pkc->tov_in_jiffies);
652 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
657 * 1) We refresh the timer only when we open a block.
658 * By doing this we don't waste cycles refreshing the timer
659 * on packet-by-packet basis.
661 * With a 1MB block-size, on a 1Gbps line, it will take
662 * i) ~8 ms to fill a block + ii) memcpy etc.
663 * In this cut we are not accounting for the memcpy time.
665 * So, if the user sets the 'tmo' to 10ms then the timer
666 * will never fire while the block is still getting filled
667 * (which is what we want). However, the user could choose
668 * to close a block early and that's fine.
670 * But when the timer does fire, we check whether or not to refresh it.
671 * Since the tmo granularity is in msecs, it is not too expensive
672 * to refresh the timer, lets say every '8' msecs.
673 * Either the user can set the 'tmo' or we can derive it based on
674 * a) line-speed and b) block-size.
675 * prb_calc_retire_blk_tmo() calculates the tmo.
678 static void prb_retire_rx_blk_timer_expired(unsigned long data)
680 struct packet_sock *po = (struct packet_sock *)data;
681 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
683 struct tpacket_block_desc *pbd;
685 spin_lock(&po->sk.sk_receive_queue.lock);
687 frozen = prb_queue_frozen(pkc);
688 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
690 if (unlikely(pkc->delete_blk_timer))
693 /* We only need to plug the race when the block is partially filled.
695 * lock(); increment BLOCK_NUM_PKTS; unlock()
696 * copy_bits() is in progress ...
697 * timer fires on other cpu:
698 * we can't retire the current block because copy_bits
702 if (BLOCK_NUM_PKTS(pbd)) {
703 while (atomic_read(&pkc->blk_fill_in_prog)) {
704 /* Waiting for skb_copy_bits to finish... */
709 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
711 if (!BLOCK_NUM_PKTS(pbd)) {
712 /* An empty block. Just refresh the timer. */
715 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
716 if (!prb_dispatch_next_block(pkc, po))
721 /* Case 1. Queue was frozen because user-space was
724 if (prb_curr_blk_in_use(pkc, pbd)) {
726 * Ok, user-space is still behind.
727 * So just refresh the timer.
731 /* Case 2. queue was frozen,user-space caught up,
732 * now the link went idle && the timer fired.
733 * We don't have a block to close.So we open this
734 * block and restart the timer.
735 * opening a block thaws the queue,restarts timer
736 * Thawing/timer-refresh is a side effect.
738 prb_open_block(pkc, pbd);
745 _prb_refresh_rx_retire_blk_timer(pkc);
748 spin_unlock(&po->sk.sk_receive_queue.lock);
751 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
752 struct tpacket_block_desc *pbd1, __u32 status)
754 /* Flush everything minus the block header */
756 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
761 /* Skip the block header(we know header WILL fit in 4K) */
764 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
765 for (; start < end; start += PAGE_SIZE)
766 flush_dcache_page(pgv_to_page(start));
771 /* Now update the block status. */
773 BLOCK_STATUS(pbd1) = status;
775 /* Flush the block header */
777 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
779 flush_dcache_page(pgv_to_page(start));
789 * 2) Increment active_blk_num
791 * Note:We DONT refresh the timer on purpose.
792 * Because almost always the next block will be opened.
794 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
795 struct tpacket_block_desc *pbd1,
796 struct packet_sock *po, unsigned int stat)
798 __u32 status = TP_STATUS_USER | stat;
800 struct tpacket3_hdr *last_pkt;
801 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
802 struct sock *sk = &po->sk;
804 if (po->stats.stats3.tp_drops)
805 status |= TP_STATUS_LOSING;
807 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
808 last_pkt->tp_next_offset = 0;
810 /* Get the ts of the last pkt */
811 if (BLOCK_NUM_PKTS(pbd1)) {
812 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
813 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
815 /* Ok, we tmo'd - so get the current time.
817 * It shouldn't really happen as we don't close empty
818 * blocks. See prb_retire_rx_blk_timer_expired().
822 h1->ts_last_pkt.ts_sec = ts.tv_sec;
823 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
828 /* Flush the block */
829 prb_flush_block(pkc1, pbd1, status);
831 sk->sk_data_ready(sk);
833 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
836 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
838 pkc->reset_pending_on_curr_blk = 0;
842 * Side effect of opening a block:
844 * 1) prb_queue is thawed.
845 * 2) retire_blk_timer is refreshed.
848 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
849 struct tpacket_block_desc *pbd1)
852 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
856 /* We could have just memset this but we will lose the
857 * flexibility of making the priv area sticky
860 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
861 BLOCK_NUM_PKTS(pbd1) = 0;
862 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
866 h1->ts_first_pkt.ts_sec = ts.tv_sec;
867 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
869 pkc1->pkblk_start = (char *)pbd1;
870 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
872 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
873 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
875 pbd1->version = pkc1->version;
876 pkc1->prev = pkc1->nxt_offset;
877 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
879 prb_thaw_queue(pkc1);
880 _prb_refresh_rx_retire_blk_timer(pkc1);
886 * Queue freeze logic:
887 * 1) Assume tp_block_nr = 8 blocks.
888 * 2) At time 't0', user opens Rx ring.
889 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
890 * 4) user-space is either sleeping or processing block '0'.
891 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
892 * it will close block-7,loop around and try to fill block '0'.
894 * __packet_lookup_frame_in_block
895 * prb_retire_current_block()
896 * prb_dispatch_next_block()
897 * |->(BLOCK_STATUS == USER) evaluates to true
898 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
899 * 6) Now there are two cases:
900 * 6.1) Link goes idle right after the queue is frozen.
901 * But remember, the last open_block() refreshed the timer.
902 * When this timer expires,it will refresh itself so that we can
903 * re-open block-0 in near future.
904 * 6.2) Link is busy and keeps on receiving packets. This is a simple
905 * case and __packet_lookup_frame_in_block will check if block-0
906 * is free and can now be re-used.
908 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
909 struct packet_sock *po)
911 pkc->reset_pending_on_curr_blk = 1;
912 po->stats.stats3.tp_freeze_q_cnt++;
915 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
918 * If the next block is free then we will dispatch it
919 * and return a good offset.
920 * Else, we will freeze the queue.
921 * So, caller must check the return value.
923 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
924 struct packet_sock *po)
926 struct tpacket_block_desc *pbd;
930 /* 1. Get current block num */
931 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
933 /* 2. If this block is currently in_use then freeze the queue */
934 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
935 prb_freeze_queue(pkc, po);
941 * open this block and return the offset where the first packet
942 * needs to get stored.
944 prb_open_block(pkc, pbd);
945 return (void *)pkc->nxt_offset;
948 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
949 struct packet_sock *po, unsigned int status)
951 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
953 /* retire/close the current block */
954 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
956 * Plug the case where copy_bits() is in progress on
957 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
958 * have space to copy the pkt in the current block and
959 * called prb_retire_current_block()
961 * We don't need to worry about the TMO case because
962 * the timer-handler already handled this case.
964 if (!(status & TP_STATUS_BLK_TMO)) {
965 while (atomic_read(&pkc->blk_fill_in_prog)) {
966 /* Waiting for skb_copy_bits to finish... */
970 prb_close_block(pkc, pbd, po, status);
975 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
976 struct tpacket_block_desc *pbd)
978 return TP_STATUS_USER & BLOCK_STATUS(pbd);
981 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
983 return pkc->reset_pending_on_curr_blk;
986 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
988 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
989 atomic_dec(&pkc->blk_fill_in_prog);
992 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
993 struct tpacket3_hdr *ppd)
995 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
998 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
999 struct tpacket3_hdr *ppd)
1001 ppd->hv1.tp_rxhash = 0;
1004 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1005 struct tpacket3_hdr *ppd)
1007 if (skb_vlan_tag_present(pkc->skb)) {
1008 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1009 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1010 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1012 ppd->hv1.tp_vlan_tci = 0;
1013 ppd->hv1.tp_vlan_tpid = 0;
1014 ppd->tp_status = TP_STATUS_AVAILABLE;
1018 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1019 struct tpacket3_hdr *ppd)
1021 ppd->hv1.tp_padding = 0;
1022 prb_fill_vlan_info(pkc, ppd);
1024 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1025 prb_fill_rxhash(pkc, ppd);
1027 prb_clear_rxhash(pkc, ppd);
1030 static void prb_fill_curr_block(char *curr,
1031 struct tpacket_kbdq_core *pkc,
1032 struct tpacket_block_desc *pbd,
1035 struct tpacket3_hdr *ppd;
1037 ppd = (struct tpacket3_hdr *)curr;
1038 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1040 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1041 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1042 BLOCK_NUM_PKTS(pbd) += 1;
1043 atomic_inc(&pkc->blk_fill_in_prog);
1044 prb_run_all_ft_ops(pkc, ppd);
1047 /* Assumes caller has the sk->rx_queue.lock */
1048 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1049 struct sk_buff *skb,
1054 struct tpacket_kbdq_core *pkc;
1055 struct tpacket_block_desc *pbd;
1058 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1059 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1061 /* Queue is frozen when user space is lagging behind */
1062 if (prb_queue_frozen(pkc)) {
1064 * Check if that last block which caused the queue to freeze,
1065 * is still in_use by user-space.
1067 if (prb_curr_blk_in_use(pkc, pbd)) {
1068 /* Can't record this packet */
1072 * Ok, the block was released by user-space.
1073 * Now let's open that block.
1074 * opening a block also thaws the queue.
1075 * Thawing is a side effect.
1077 prb_open_block(pkc, pbd);
1082 curr = pkc->nxt_offset;
1084 end = (char *)pbd + pkc->kblk_size;
1086 /* first try the current block */
1087 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1088 prb_fill_curr_block(curr, pkc, pbd, len);
1089 return (void *)curr;
1092 /* Ok, close the current block */
1093 prb_retire_current_block(pkc, po, 0);
1095 /* Now, try to dispatch the next block */
1096 curr = (char *)prb_dispatch_next_block(pkc, po);
1098 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1099 prb_fill_curr_block(curr, pkc, pbd, len);
1100 return (void *)curr;
1104 * No free blocks are available.user_space hasn't caught up yet.
1105 * Queue was just frozen and now this packet will get dropped.
1110 static void *packet_current_rx_frame(struct packet_sock *po,
1111 struct sk_buff *skb,
1112 int status, unsigned int len)
1115 switch (po->tp_version) {
1118 curr = packet_lookup_frame(po, &po->rx_ring,
1119 po->rx_ring.head, status);
1122 return __packet_lookup_frame_in_block(po, skb, status, len);
1124 WARN(1, "TPACKET version not supported\n");
1130 static void *prb_lookup_block(struct packet_sock *po,
1131 struct packet_ring_buffer *rb,
1135 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1136 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1138 if (status != BLOCK_STATUS(pbd))
1143 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1146 if (rb->prb_bdqc.kactive_blk_num)
1147 prev = rb->prb_bdqc.kactive_blk_num-1;
1149 prev = rb->prb_bdqc.knum_blocks-1;
1153 /* Assumes caller has held the rx_queue.lock */
1154 static void *__prb_previous_block(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1158 unsigned int previous = prb_previous_blk_num(rb);
1159 return prb_lookup_block(po, rb, previous, status);
1162 static void *packet_previous_rx_frame(struct packet_sock *po,
1163 struct packet_ring_buffer *rb,
1166 if (po->tp_version <= TPACKET_V2)
1167 return packet_previous_frame(po, rb, status);
1169 return __prb_previous_block(po, rb, status);
1172 static void packet_increment_rx_head(struct packet_sock *po,
1173 struct packet_ring_buffer *rb)
1175 switch (po->tp_version) {
1178 return packet_increment_head(rb);
1181 WARN(1, "TPACKET version not supported.\n");
1187 static void *packet_previous_frame(struct packet_sock *po,
1188 struct packet_ring_buffer *rb,
1191 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1192 return packet_lookup_frame(po, rb, previous, status);
1195 static void packet_increment_head(struct packet_ring_buffer *buff)
1197 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1200 static void packet_inc_pending(struct packet_ring_buffer *rb)
1202 this_cpu_inc(*rb->pending_refcnt);
1205 static void packet_dec_pending(struct packet_ring_buffer *rb)
1207 this_cpu_dec(*rb->pending_refcnt);
1210 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1212 unsigned int refcnt = 0;
1215 /* We don't use pending refcount in rx_ring. */
1216 if (rb->pending_refcnt == NULL)
1219 for_each_possible_cpu(cpu)
1220 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1225 static int packet_alloc_pending(struct packet_sock *po)
1227 po->rx_ring.pending_refcnt = NULL;
1229 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1230 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1236 static void packet_free_pending(struct packet_sock *po)
1238 free_percpu(po->tx_ring.pending_refcnt);
1241 #define ROOM_POW_OFF 2
1242 #define ROOM_NONE 0x0
1243 #define ROOM_LOW 0x1
1244 #define ROOM_NORMAL 0x2
1246 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1250 len = po->rx_ring.frame_max + 1;
1251 idx = po->rx_ring.head;
1253 idx += len >> pow_off;
1256 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1259 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1263 len = po->rx_ring.prb_bdqc.knum_blocks;
1264 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1266 idx += len >> pow_off;
1269 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1272 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1274 struct sock *sk = &po->sk;
1275 int ret = ROOM_NONE;
1277 if (po->prot_hook.func != tpacket_rcv) {
1278 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1279 - (skb ? skb->truesize : 0);
1280 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1288 if (po->tp_version == TPACKET_V3) {
1289 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1291 else if (__tpacket_v3_has_room(po, 0))
1294 if (__tpacket_has_room(po, ROOM_POW_OFF))
1296 else if (__tpacket_has_room(po, 0))
1303 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1308 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1309 ret = __packet_rcv_has_room(po, skb);
1310 has_room = ret == ROOM_NORMAL;
1311 if (po->pressure == has_room)
1312 po->pressure = !has_room;
1313 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1318 static void packet_sock_destruct(struct sock *sk)
1320 skb_queue_purge(&sk->sk_error_queue);
1322 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1323 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1325 if (!sock_flag(sk, SOCK_DEAD)) {
1326 pr_err("Attempt to release alive packet socket: %p\n", sk);
1330 sk_refcnt_debug_dec(sk);
1333 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1338 rxhash = skb_get_hash(skb);
1339 for (i = 0; i < ROLLOVER_HLEN; i++)
1340 if (po->rollover->history[i] == rxhash)
1343 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1344 return count > (ROLLOVER_HLEN >> 1);
1347 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1348 struct sk_buff *skb,
1351 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1354 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1355 struct sk_buff *skb,
1358 unsigned int val = atomic_inc_return(&f->rr_cur);
1363 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1364 struct sk_buff *skb,
1367 return smp_processor_id() % num;
1370 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1371 struct sk_buff *skb,
1374 return prandom_u32_max(num);
1377 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1378 struct sk_buff *skb,
1379 unsigned int idx, bool try_self,
1382 struct packet_sock *po, *po_next, *po_skip = NULL;
1383 unsigned int i, j, room = ROOM_NONE;
1385 po = pkt_sk(f->arr[idx]);
1388 room = packet_rcv_has_room(po, skb);
1389 if (room == ROOM_NORMAL ||
1390 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1395 i = j = min_t(int, po->rollover->sock, num - 1);
1397 po_next = pkt_sk(f->arr[i]);
1398 if (po_next != po_skip && !po_next->pressure &&
1399 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1401 po->rollover->sock = i;
1402 atomic_long_inc(&po->rollover->num);
1403 if (room == ROOM_LOW)
1404 atomic_long_inc(&po->rollover->num_huge);
1412 atomic_long_inc(&po->rollover->num_failed);
1416 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1417 struct sk_buff *skb,
1420 return skb_get_queue_mapping(skb) % num;
1423 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1424 struct sk_buff *skb,
1427 struct bpf_prog *prog;
1428 unsigned int ret = 0;
1431 prog = rcu_dereference(f->bpf_prog);
1433 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1439 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1441 return f->flags & (flag >> 8);
1444 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1445 struct packet_type *pt, struct net_device *orig_dev)
1447 struct packet_fanout *f = pt->af_packet_priv;
1448 unsigned int num = READ_ONCE(f->num_members);
1449 struct net *net = read_pnet(&f->net);
1450 struct packet_sock *po;
1453 if (!net_eq(dev_net(dev), net) || !num) {
1458 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1459 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1464 case PACKET_FANOUT_HASH:
1466 idx = fanout_demux_hash(f, skb, num);
1468 case PACKET_FANOUT_LB:
1469 idx = fanout_demux_lb(f, skb, num);
1471 case PACKET_FANOUT_CPU:
1472 idx = fanout_demux_cpu(f, skb, num);
1474 case PACKET_FANOUT_RND:
1475 idx = fanout_demux_rnd(f, skb, num);
1477 case PACKET_FANOUT_QM:
1478 idx = fanout_demux_qm(f, skb, num);
1480 case PACKET_FANOUT_ROLLOVER:
1481 idx = fanout_demux_rollover(f, skb, 0, false, num);
1483 case PACKET_FANOUT_CBPF:
1484 case PACKET_FANOUT_EBPF:
1485 idx = fanout_demux_bpf(f, skb, num);
1489 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1490 idx = fanout_demux_rollover(f, skb, idx, true, num);
1492 po = pkt_sk(f->arr[idx]);
1493 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1496 DEFINE_MUTEX(fanout_mutex);
1497 EXPORT_SYMBOL_GPL(fanout_mutex);
1498 static LIST_HEAD(fanout_list);
1500 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1502 struct packet_fanout *f = po->fanout;
1504 spin_lock(&f->lock);
1505 f->arr[f->num_members] = sk;
1508 if (f->num_members == 1)
1509 dev_add_pack(&f->prot_hook);
1510 spin_unlock(&f->lock);
1513 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1515 struct packet_fanout *f = po->fanout;
1518 spin_lock(&f->lock);
1519 for (i = 0; i < f->num_members; i++) {
1520 if (f->arr[i] == sk)
1523 BUG_ON(i >= f->num_members);
1524 f->arr[i] = f->arr[f->num_members - 1];
1526 if (f->num_members == 0)
1527 __dev_remove_pack(&f->prot_hook);
1528 spin_unlock(&f->lock);
1531 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1533 if (sk->sk_family != PF_PACKET)
1536 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1539 static void fanout_init_data(struct packet_fanout *f)
1542 case PACKET_FANOUT_LB:
1543 atomic_set(&f->rr_cur, 0);
1545 case PACKET_FANOUT_CBPF:
1546 case PACKET_FANOUT_EBPF:
1547 RCU_INIT_POINTER(f->bpf_prog, NULL);
1552 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1554 struct bpf_prog *old;
1556 spin_lock(&f->lock);
1557 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1558 rcu_assign_pointer(f->bpf_prog, new);
1559 spin_unlock(&f->lock);
1563 bpf_prog_destroy(old);
1567 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1570 struct bpf_prog *new;
1571 struct sock_fprog fprog;
1574 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1576 if (len != sizeof(fprog))
1578 if (copy_from_user(&fprog, data, len))
1581 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1585 __fanout_set_data_bpf(po->fanout, new);
1589 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1592 struct bpf_prog *new;
1595 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1597 if (len != sizeof(fd))
1599 if (copy_from_user(&fd, data, len))
1602 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1604 return PTR_ERR(new);
1606 __fanout_set_data_bpf(po->fanout, new);
1610 static int fanout_set_data(struct packet_sock *po, char __user *data,
1613 switch (po->fanout->type) {
1614 case PACKET_FANOUT_CBPF:
1615 return fanout_set_data_cbpf(po, data, len);
1616 case PACKET_FANOUT_EBPF:
1617 return fanout_set_data_ebpf(po, data, len);
1623 static void fanout_release_data(struct packet_fanout *f)
1626 case PACKET_FANOUT_CBPF:
1627 case PACKET_FANOUT_EBPF:
1628 __fanout_set_data_bpf(f, NULL);
1632 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1634 struct packet_rollover *rollover = NULL;
1635 struct packet_sock *po = pkt_sk(sk);
1636 struct packet_fanout *f, *match;
1637 u8 type = type_flags & 0xff;
1638 u8 flags = type_flags >> 8;
1642 case PACKET_FANOUT_ROLLOVER:
1643 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1645 case PACKET_FANOUT_HASH:
1646 case PACKET_FANOUT_LB:
1647 case PACKET_FANOUT_CPU:
1648 case PACKET_FANOUT_RND:
1649 case PACKET_FANOUT_QM:
1650 case PACKET_FANOUT_CBPF:
1651 case PACKET_FANOUT_EBPF:
1657 mutex_lock(&fanout_mutex);
1667 if (type == PACKET_FANOUT_ROLLOVER ||
1668 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1670 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1673 atomic_long_set(&rollover->num, 0);
1674 atomic_long_set(&rollover->num_huge, 0);
1675 atomic_long_set(&rollover->num_failed, 0);
1676 po->rollover = rollover;
1680 list_for_each_entry(f, &fanout_list, list) {
1682 read_pnet(&f->net) == sock_net(sk)) {
1688 if (match && match->flags != flags)
1692 match = kzalloc(sizeof(*match), GFP_KERNEL);
1695 write_pnet(&match->net, sock_net(sk));
1698 match->flags = flags;
1699 INIT_LIST_HEAD(&match->list);
1700 spin_lock_init(&match->lock);
1701 atomic_set(&match->sk_ref, 0);
1702 fanout_init_data(match);
1703 match->prot_hook.type = po->prot_hook.type;
1704 match->prot_hook.dev = po->prot_hook.dev;
1705 match->prot_hook.func = packet_rcv_fanout;
1706 match->prot_hook.af_packet_priv = match;
1707 match->prot_hook.id_match = match_fanout_group;
1708 list_add(&match->list, &fanout_list);
1711 if (match->type == type &&
1712 match->prot_hook.type == po->prot_hook.type &&
1713 match->prot_hook.dev == po->prot_hook.dev) {
1715 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1716 __dev_remove_pack(&po->prot_hook);
1718 atomic_inc(&match->sk_ref);
1719 __fanout_link(sk, po);
1724 if (err && rollover) {
1726 po->rollover = NULL;
1728 mutex_unlock(&fanout_mutex);
1732 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1733 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1734 * It is the responsibility of the caller to call fanout_release_data() and
1735 * free the returned packet_fanout (after synchronize_net())
1737 static struct packet_fanout *fanout_release(struct sock *sk)
1739 struct packet_sock *po = pkt_sk(sk);
1740 struct packet_fanout *f;
1742 mutex_lock(&fanout_mutex);
1747 if (atomic_dec_and_test(&f->sk_ref))
1753 kfree_rcu(po->rollover, rcu);
1755 mutex_unlock(&fanout_mutex);
1760 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1761 struct sk_buff *skb)
1763 /* Earlier code assumed this would be a VLAN pkt, double-check
1764 * this now that we have the actual packet in hand. We can only
1765 * do this check on Ethernet devices.
1767 if (unlikely(dev->type != ARPHRD_ETHER))
1770 skb_reset_mac_header(skb);
1771 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1774 static const struct proto_ops packet_ops;
1776 static const struct proto_ops packet_ops_spkt;
1778 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1779 struct packet_type *pt, struct net_device *orig_dev)
1782 struct sockaddr_pkt *spkt;
1785 * When we registered the protocol we saved the socket in the data
1786 * field for just this event.
1789 sk = pt->af_packet_priv;
1792 * Yank back the headers [hope the device set this
1793 * right or kerboom...]
1795 * Incoming packets have ll header pulled,
1798 * For outgoing ones skb->data == skb_mac_header(skb)
1799 * so that this procedure is noop.
1802 if (skb->pkt_type == PACKET_LOOPBACK)
1805 if (!net_eq(dev_net(dev), sock_net(sk)))
1808 skb = skb_share_check(skb, GFP_ATOMIC);
1812 /* drop any routing info */
1815 /* drop conntrack reference */
1818 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1820 skb_push(skb, skb->data - skb_mac_header(skb));
1823 * The SOCK_PACKET socket receives _all_ frames.
1826 spkt->spkt_family = dev->type;
1827 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1828 spkt->spkt_protocol = skb->protocol;
1831 * Charge the memory to the socket. This is done specifically
1832 * to prevent sockets using all the memory up.
1835 if (sock_queue_rcv_skb(sk, skb) == 0)
1846 * Output a raw packet to a device layer. This bypasses all the other
1847 * protocol layers and you must therefore supply it with a complete frame
1850 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1853 struct sock *sk = sock->sk;
1854 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1855 struct sk_buff *skb = NULL;
1856 struct net_device *dev;
1857 struct sockcm_cookie sockc;
1863 * Get and verify the address.
1867 if (msg->msg_namelen < sizeof(struct sockaddr))
1869 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1870 proto = saddr->spkt_protocol;
1872 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1875 * Find the device first to size check it
1878 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1881 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1887 if (!(dev->flags & IFF_UP))
1891 * You may not queue a frame bigger than the mtu. This is the lowest level
1892 * raw protocol and you must do your own fragmentation at this level.
1895 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1896 if (!netif_supports_nofcs(dev)) {
1897 err = -EPROTONOSUPPORT;
1900 extra_len = 4; /* We're doing our own CRC */
1904 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1908 size_t reserved = LL_RESERVED_SPACE(dev);
1909 int tlen = dev->needed_tailroom;
1910 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1913 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1916 /* FIXME: Save some space for broken drivers that write a hard
1917 * header at transmission time by themselves. PPP is the notable
1918 * one here. This should really be fixed at the driver level.
1920 skb_reserve(skb, reserved);
1921 skb_reset_network_header(skb);
1923 /* Try to align data part correctly */
1928 skb_reset_network_header(skb);
1930 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1936 if (!dev_validate_header(dev, skb->data, len)) {
1940 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1941 !packet_extra_vlan_len_allowed(dev, skb)) {
1946 sockc.tsflags = sk->sk_tsflags;
1947 if (msg->msg_controllen) {
1948 err = sock_cmsg_send(sk, msg, &sockc);
1953 skb->protocol = proto;
1955 skb->priority = sk->sk_priority;
1956 skb->mark = sk->sk_mark;
1958 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1960 if (unlikely(extra_len == 4))
1963 skb_probe_transport_header(skb, 0);
1965 dev_queue_xmit(skb);
1976 static unsigned int run_filter(struct sk_buff *skb,
1977 const struct sock *sk,
1980 struct sk_filter *filter;
1983 filter = rcu_dereference(sk->sk_filter);
1985 res = bpf_prog_run_clear_cb(filter->prog, skb);
1991 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
1994 struct virtio_net_hdr vnet_hdr;
1996 if (*len < sizeof(vnet_hdr))
1998 *len -= sizeof(vnet_hdr);
2000 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
2003 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
2007 * This function makes lazy skb cloning in hope that most of packets
2008 * are discarded by BPF.
2010 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2011 * and skb->cb are mangled. It works because (and until) packets
2012 * falling here are owned by current CPU. Output packets are cloned
2013 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2014 * sequencially, so that if we return skb to original state on exit,
2015 * we will not harm anyone.
2018 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2019 struct packet_type *pt, struct net_device *orig_dev)
2022 struct sockaddr_ll *sll;
2023 struct packet_sock *po;
2024 u8 *skb_head = skb->data;
2025 int skb_len = skb->len;
2026 unsigned int snaplen, res;
2027 bool is_drop_n_account = false;
2029 if (skb->pkt_type == PACKET_LOOPBACK)
2032 sk = pt->af_packet_priv;
2035 if (!net_eq(dev_net(dev), sock_net(sk)))
2040 if (dev->header_ops) {
2041 /* The device has an explicit notion of ll header,
2042 * exported to higher levels.
2044 * Otherwise, the device hides details of its frame
2045 * structure, so that corresponding packet head is
2046 * never delivered to user.
2048 if (sk->sk_type != SOCK_DGRAM)
2049 skb_push(skb, skb->data - skb_mac_header(skb));
2050 else if (skb->pkt_type == PACKET_OUTGOING) {
2051 /* Special case: outgoing packets have ll header at head */
2052 skb_pull(skb, skb_network_offset(skb));
2058 res = run_filter(skb, sk, snaplen);
2060 goto drop_n_restore;
2064 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2067 if (skb_shared(skb)) {
2068 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2072 if (skb_head != skb->data) {
2073 skb->data = skb_head;
2080 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2082 sll = &PACKET_SKB_CB(skb)->sa.ll;
2083 sll->sll_hatype = dev->type;
2084 sll->sll_pkttype = skb->pkt_type;
2085 if (unlikely(po->origdev))
2086 sll->sll_ifindex = orig_dev->ifindex;
2088 sll->sll_ifindex = dev->ifindex;
2090 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2092 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2093 * Use their space for storing the original skb length.
2095 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2097 if (pskb_trim(skb, snaplen))
2100 skb_set_owner_r(skb, sk);
2104 /* drop conntrack reference */
2107 spin_lock(&sk->sk_receive_queue.lock);
2108 po->stats.stats1.tp_packets++;
2109 sock_skb_set_dropcount(sk, skb);
2110 __skb_queue_tail(&sk->sk_receive_queue, skb);
2111 spin_unlock(&sk->sk_receive_queue.lock);
2112 sk->sk_data_ready(sk);
2116 is_drop_n_account = true;
2117 spin_lock(&sk->sk_receive_queue.lock);
2118 po->stats.stats1.tp_drops++;
2119 atomic_inc(&sk->sk_drops);
2120 spin_unlock(&sk->sk_receive_queue.lock);
2123 if (skb_head != skb->data && skb_shared(skb)) {
2124 skb->data = skb_head;
2128 if (!is_drop_n_account)
2135 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2136 struct packet_type *pt, struct net_device *orig_dev)
2139 struct packet_sock *po;
2140 struct sockaddr_ll *sll;
2141 union tpacket_uhdr h;
2142 u8 *skb_head = skb->data;
2143 int skb_len = skb->len;
2144 unsigned int snaplen, res;
2145 unsigned long status = TP_STATUS_USER;
2146 unsigned short macoff, netoff, hdrlen;
2147 struct sk_buff *copy_skb = NULL;
2150 bool is_drop_n_account = false;
2152 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2153 * We may add members to them until current aligned size without forcing
2154 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2156 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2157 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2159 if (skb->pkt_type == PACKET_LOOPBACK)
2162 sk = pt->af_packet_priv;
2165 if (!net_eq(dev_net(dev), sock_net(sk)))
2168 if (dev->header_ops) {
2169 if (sk->sk_type != SOCK_DGRAM)
2170 skb_push(skb, skb->data - skb_mac_header(skb));
2171 else if (skb->pkt_type == PACKET_OUTGOING) {
2172 /* Special case: outgoing packets have ll header at head */
2173 skb_pull(skb, skb_network_offset(skb));
2179 res = run_filter(skb, sk, snaplen);
2181 goto drop_n_restore;
2183 if (skb->ip_summed == CHECKSUM_PARTIAL)
2184 status |= TP_STATUS_CSUMNOTREADY;
2185 else if (skb->pkt_type != PACKET_OUTGOING &&
2186 (skb->ip_summed == CHECKSUM_COMPLETE ||
2187 skb_csum_unnecessary(skb)))
2188 status |= TP_STATUS_CSUM_VALID;
2193 if (sk->sk_type == SOCK_DGRAM) {
2194 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2197 unsigned int maclen = skb_network_offset(skb);
2198 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2199 (maclen < 16 ? 16 : maclen)) +
2201 if (po->has_vnet_hdr)
2202 netoff += sizeof(struct virtio_net_hdr);
2203 macoff = netoff - maclen;
2205 if (po->tp_version <= TPACKET_V2) {
2206 if (macoff + snaplen > po->rx_ring.frame_size) {
2207 if (po->copy_thresh &&
2208 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2209 if (skb_shared(skb)) {
2210 copy_skb = skb_clone(skb, GFP_ATOMIC);
2212 copy_skb = skb_get(skb);
2213 skb_head = skb->data;
2216 skb_set_owner_r(copy_skb, sk);
2218 snaplen = po->rx_ring.frame_size - macoff;
2219 if ((int)snaplen < 0)
2222 } else if (unlikely(macoff + snaplen >
2223 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2226 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2227 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2228 snaplen, nval, macoff);
2230 if (unlikely((int)snaplen < 0)) {
2232 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2235 spin_lock(&sk->sk_receive_queue.lock);
2236 h.raw = packet_current_rx_frame(po, skb,
2237 TP_STATUS_KERNEL, (macoff+snaplen));
2239 goto drop_n_account;
2240 if (po->tp_version <= TPACKET_V2) {
2241 packet_increment_rx_head(po, &po->rx_ring);
2243 * LOSING will be reported till you read the stats,
2244 * because it's COR - Clear On Read.
2245 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2248 if (po->stats.stats1.tp_drops)
2249 status |= TP_STATUS_LOSING;
2251 po->stats.stats1.tp_packets++;
2253 status |= TP_STATUS_COPY;
2254 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2256 spin_unlock(&sk->sk_receive_queue.lock);
2258 if (po->has_vnet_hdr) {
2259 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2260 sizeof(struct virtio_net_hdr),
2262 spin_lock(&sk->sk_receive_queue.lock);
2263 goto drop_n_account;
2267 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2269 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2270 getnstimeofday(&ts);
2272 status |= ts_status;
2274 switch (po->tp_version) {
2276 h.h1->tp_len = skb->len;
2277 h.h1->tp_snaplen = snaplen;
2278 h.h1->tp_mac = macoff;
2279 h.h1->tp_net = netoff;
2280 h.h1->tp_sec = ts.tv_sec;
2281 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2282 hdrlen = sizeof(*h.h1);
2285 h.h2->tp_len = skb->len;
2286 h.h2->tp_snaplen = snaplen;
2287 h.h2->tp_mac = macoff;
2288 h.h2->tp_net = netoff;
2289 h.h2->tp_sec = ts.tv_sec;
2290 h.h2->tp_nsec = ts.tv_nsec;
2291 if (skb_vlan_tag_present(skb)) {
2292 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2293 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2294 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2296 h.h2->tp_vlan_tci = 0;
2297 h.h2->tp_vlan_tpid = 0;
2299 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2300 hdrlen = sizeof(*h.h2);
2303 /* tp_nxt_offset,vlan are already populated above.
2304 * So DONT clear those fields here
2306 h.h3->tp_status |= status;
2307 h.h3->tp_len = skb->len;
2308 h.h3->tp_snaplen = snaplen;
2309 h.h3->tp_mac = macoff;
2310 h.h3->tp_net = netoff;
2311 h.h3->tp_sec = ts.tv_sec;
2312 h.h3->tp_nsec = ts.tv_nsec;
2313 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2314 hdrlen = sizeof(*h.h3);
2320 sll = h.raw + TPACKET_ALIGN(hdrlen);
2321 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2322 sll->sll_family = AF_PACKET;
2323 sll->sll_hatype = dev->type;
2324 sll->sll_protocol = skb->protocol;
2325 sll->sll_pkttype = skb->pkt_type;
2326 if (unlikely(po->origdev))
2327 sll->sll_ifindex = orig_dev->ifindex;
2329 sll->sll_ifindex = dev->ifindex;
2333 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2334 if (po->tp_version <= TPACKET_V2) {
2337 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2340 for (start = h.raw; start < end; start += PAGE_SIZE)
2341 flush_dcache_page(pgv_to_page(start));
2346 if (po->tp_version <= TPACKET_V2) {
2347 __packet_set_status(po, h.raw, status);
2348 sk->sk_data_ready(sk);
2350 prb_clear_blk_fill_status(&po->rx_ring);
2354 if (skb_head != skb->data && skb_shared(skb)) {
2355 skb->data = skb_head;
2359 if (!is_drop_n_account)
2366 is_drop_n_account = true;
2367 po->stats.stats1.tp_drops++;
2368 spin_unlock(&sk->sk_receive_queue.lock);
2370 sk->sk_data_ready(sk);
2371 kfree_skb(copy_skb);
2372 goto drop_n_restore;
2375 static void tpacket_destruct_skb(struct sk_buff *skb)
2377 struct packet_sock *po = pkt_sk(skb->sk);
2379 if (likely(po->tx_ring.pg_vec)) {
2383 ph = skb_shinfo(skb)->destructor_arg;
2384 packet_dec_pending(&po->tx_ring);
2386 ts = __packet_set_timestamp(po, ph, skb);
2387 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2393 static void tpacket_set_protocol(const struct net_device *dev,
2394 struct sk_buff *skb)
2396 if (dev->type == ARPHRD_ETHER) {
2397 skb_reset_mac_header(skb);
2398 skb->protocol = eth_hdr(skb)->h_proto;
2402 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2404 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2405 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2406 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2407 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2408 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2409 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2410 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2412 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2418 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2419 struct virtio_net_hdr *vnet_hdr)
2421 if (*len < sizeof(*vnet_hdr))
2423 *len -= sizeof(*vnet_hdr);
2425 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2428 return __packet_snd_vnet_parse(vnet_hdr, *len);
2431 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2432 void *frame, struct net_device *dev, void *data, int tp_len,
2433 __be16 proto, unsigned char *addr, int hlen, int copylen,
2434 const struct sockcm_cookie *sockc)
2436 union tpacket_uhdr ph;
2437 int to_write, offset, len, nr_frags, len_max;
2438 struct socket *sock = po->sk.sk_socket;
2444 skb->protocol = proto;
2446 skb->priority = po->sk.sk_priority;
2447 skb->mark = po->sk.sk_mark;
2448 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2449 skb_shinfo(skb)->destructor_arg = ph.raw;
2451 skb_reserve(skb, hlen);
2452 skb_reset_network_header(skb);
2456 if (sock->type == SOCK_DGRAM) {
2457 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2459 if (unlikely(err < 0))
2461 } else if (copylen) {
2462 int hdrlen = min_t(int, copylen, tp_len);
2464 skb_push(skb, dev->hard_header_len);
2465 skb_put(skb, copylen - dev->hard_header_len);
2466 err = skb_store_bits(skb, 0, data, hdrlen);
2469 if (!dev_validate_header(dev, skb->data, hdrlen))
2472 tpacket_set_protocol(dev, skb);
2478 offset = offset_in_page(data);
2479 len_max = PAGE_SIZE - offset;
2480 len = ((to_write > len_max) ? len_max : to_write);
2482 skb->data_len = to_write;
2483 skb->len += to_write;
2484 skb->truesize += to_write;
2485 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2487 while (likely(to_write)) {
2488 nr_frags = skb_shinfo(skb)->nr_frags;
2490 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2491 pr_err("Packet exceed the number of skb frags(%lu)\n",
2496 page = pgv_to_page(data);
2498 flush_dcache_page(page);
2500 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2503 len_max = PAGE_SIZE;
2504 len = ((to_write > len_max) ? len_max : to_write);
2507 skb_probe_transport_header(skb, 0);
2512 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2513 int size_max, void **data)
2515 union tpacket_uhdr ph;
2520 switch (po->tp_version) {
2522 if (ph.h3->tp_next_offset != 0) {
2523 pr_warn_once("variable sized slot not supported");
2526 tp_len = ph.h3->tp_len;
2529 tp_len = ph.h2->tp_len;
2532 tp_len = ph.h1->tp_len;
2535 if (unlikely(tp_len > size_max)) {
2536 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2540 if (unlikely(po->tp_tx_has_off)) {
2541 int off_min, off_max;
2543 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2544 off_max = po->tx_ring.frame_size - tp_len;
2545 if (po->sk.sk_type == SOCK_DGRAM) {
2546 switch (po->tp_version) {
2548 off = ph.h3->tp_net;
2551 off = ph.h2->tp_net;
2554 off = ph.h1->tp_net;
2558 switch (po->tp_version) {
2560 off = ph.h3->tp_mac;
2563 off = ph.h2->tp_mac;
2566 off = ph.h1->tp_mac;
2570 if (unlikely((off < off_min) || (off_max < off)))
2573 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2576 *data = frame + off;
2580 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2582 struct sk_buff *skb;
2583 struct net_device *dev;
2584 struct virtio_net_hdr *vnet_hdr = NULL;
2585 struct sockcm_cookie sockc;
2587 int err, reserve = 0;
2589 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2590 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2591 int tp_len, size_max;
2592 unsigned char *addr;
2595 int status = TP_STATUS_AVAILABLE;
2596 int hlen, tlen, copylen = 0;
2598 mutex_lock(&po->pg_vec_lock);
2600 if (likely(saddr == NULL)) {
2601 dev = packet_cached_dev_get(po);
2606 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2608 if (msg->msg_namelen < (saddr->sll_halen
2609 + offsetof(struct sockaddr_ll,
2612 proto = saddr->sll_protocol;
2613 addr = saddr->sll_addr;
2614 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2617 sockc.tsflags = po->sk.sk_tsflags;
2618 if (msg->msg_controllen) {
2619 err = sock_cmsg_send(&po->sk, msg, &sockc);
2625 if (unlikely(dev == NULL))
2628 if (unlikely(!(dev->flags & IFF_UP)))
2631 if (po->sk.sk_socket->type == SOCK_RAW)
2632 reserve = dev->hard_header_len;
2633 size_max = po->tx_ring.frame_size
2634 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2636 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2637 size_max = dev->mtu + reserve + VLAN_HLEN;
2640 ph = packet_current_frame(po, &po->tx_ring,
2641 TP_STATUS_SEND_REQUEST);
2642 if (unlikely(ph == NULL)) {
2643 if (need_wait && need_resched())
2649 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2653 status = TP_STATUS_SEND_REQUEST;
2654 hlen = LL_RESERVED_SPACE(dev);
2655 tlen = dev->needed_tailroom;
2656 if (po->has_vnet_hdr) {
2658 data += sizeof(*vnet_hdr);
2659 tp_len -= sizeof(*vnet_hdr);
2661 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2665 copylen = __virtio16_to_cpu(vio_le(),
2668 copylen = max_t(int, copylen, dev->hard_header_len);
2669 skb = sock_alloc_send_skb(&po->sk,
2670 hlen + tlen + sizeof(struct sockaddr_ll) +
2671 (copylen - dev->hard_header_len),
2674 if (unlikely(skb == NULL)) {
2675 /* we assume the socket was initially writeable ... */
2676 if (likely(len_sum > 0))
2680 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2681 addr, hlen, copylen, &sockc);
2682 if (likely(tp_len >= 0) &&
2683 tp_len > dev->mtu + reserve &&
2684 !po->has_vnet_hdr &&
2685 !packet_extra_vlan_len_allowed(dev, skb))
2688 if (unlikely(tp_len < 0)) {
2691 __packet_set_status(po, ph,
2692 TP_STATUS_AVAILABLE);
2693 packet_increment_head(&po->tx_ring);
2697 status = TP_STATUS_WRONG_FORMAT;
2703 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2709 packet_pick_tx_queue(dev, skb);
2711 skb->destructor = tpacket_destruct_skb;
2712 __packet_set_status(po, ph, TP_STATUS_SENDING);
2713 packet_inc_pending(&po->tx_ring);
2715 status = TP_STATUS_SEND_REQUEST;
2716 err = po->xmit(skb);
2717 if (unlikely(err > 0)) {
2718 err = net_xmit_errno(err);
2719 if (err && __packet_get_status(po, ph) ==
2720 TP_STATUS_AVAILABLE) {
2721 /* skb was destructed already */
2726 * skb was dropped but not destructed yet;
2727 * let's treat it like congestion or err < 0
2731 packet_increment_head(&po->tx_ring);
2733 } while (likely((ph != NULL) ||
2734 /* Note: packet_read_pending() might be slow if we have
2735 * to call it as it's per_cpu variable, but in fast-path
2736 * we already short-circuit the loop with the first
2737 * condition, and luckily don't have to go that path
2740 (need_wait && packet_read_pending(&po->tx_ring))));
2746 __packet_set_status(po, ph, status);
2751 mutex_unlock(&po->pg_vec_lock);
2755 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2756 size_t reserve, size_t len,
2757 size_t linear, int noblock,
2760 struct sk_buff *skb;
2762 /* Under a page? Don't bother with paged skb. */
2763 if (prepad + len < PAGE_SIZE || !linear)
2766 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2771 skb_reserve(skb, reserve);
2772 skb_put(skb, linear);
2773 skb->data_len = len - linear;
2774 skb->len += len - linear;
2779 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2781 struct sock *sk = sock->sk;
2782 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2783 struct sk_buff *skb;
2784 struct net_device *dev;
2786 unsigned char *addr;
2787 int err, reserve = 0;
2788 struct sockcm_cookie sockc;
2789 struct virtio_net_hdr vnet_hdr = { 0 };
2791 struct packet_sock *po = pkt_sk(sk);
2792 int hlen, tlen, linear;
2796 * Get and verify the address.
2799 if (likely(saddr == NULL)) {
2800 dev = packet_cached_dev_get(po);
2805 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2807 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2809 proto = saddr->sll_protocol;
2810 addr = saddr->sll_addr;
2811 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2815 if (unlikely(dev == NULL))
2818 if (unlikely(!(dev->flags & IFF_UP)))
2821 sockc.tsflags = sk->sk_tsflags;
2822 sockc.mark = sk->sk_mark;
2823 if (msg->msg_controllen) {
2824 err = sock_cmsg_send(sk, msg, &sockc);
2829 if (sock->type == SOCK_RAW)
2830 reserve = dev->hard_header_len;
2831 if (po->has_vnet_hdr) {
2832 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2837 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2838 if (!netif_supports_nofcs(dev)) {
2839 err = -EPROTONOSUPPORT;
2842 extra_len = 4; /* We're doing our own CRC */
2846 if (!vnet_hdr.gso_type &&
2847 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2851 hlen = LL_RESERVED_SPACE(dev);
2852 tlen = dev->needed_tailroom;
2853 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2854 linear = max(linear, min_t(int, len, dev->hard_header_len));
2855 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2856 msg->msg_flags & MSG_DONTWAIT, &err);
2860 skb_set_network_header(skb, reserve);
2863 if (sock->type == SOCK_DGRAM) {
2864 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2865 if (unlikely(offset < 0))
2869 /* Returns -EFAULT on error */
2870 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2874 if (sock->type == SOCK_RAW &&
2875 !dev_validate_header(dev, skb->data, len)) {
2880 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2882 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2883 !packet_extra_vlan_len_allowed(dev, skb)) {
2888 skb->protocol = proto;
2890 skb->priority = sk->sk_priority;
2891 skb->mark = sockc.mark;
2893 packet_pick_tx_queue(dev, skb);
2895 if (po->has_vnet_hdr) {
2896 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2899 len += sizeof(vnet_hdr);
2902 skb_probe_transport_header(skb, reserve);
2904 if (unlikely(extra_len == 4))
2907 err = po->xmit(skb);
2908 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2924 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2926 struct sock *sk = sock->sk;
2927 struct packet_sock *po = pkt_sk(sk);
2929 if (po->tx_ring.pg_vec)
2930 return tpacket_snd(po, msg);
2932 return packet_snd(sock, msg, len);
2936 * Close a PACKET socket. This is fairly simple. We immediately go
2937 * to 'closed' state and remove our protocol entry in the device list.
2940 static int packet_release(struct socket *sock)
2942 struct sock *sk = sock->sk;
2943 struct packet_sock *po;
2944 struct packet_fanout *f;
2946 union tpacket_req_u req_u;
2954 mutex_lock(&net->packet.sklist_lock);
2955 sk_del_node_init_rcu(sk);
2956 mutex_unlock(&net->packet.sklist_lock);
2959 sock_prot_inuse_add(net, sk->sk_prot, -1);
2962 spin_lock(&po->bind_lock);
2963 unregister_prot_hook(sk, false);
2964 packet_cached_dev_reset(po);
2966 if (po->prot_hook.dev) {
2967 dev_put(po->prot_hook.dev);
2968 po->prot_hook.dev = NULL;
2970 spin_unlock(&po->bind_lock);
2972 packet_flush_mclist(sk);
2974 if (po->rx_ring.pg_vec) {
2975 memset(&req_u, 0, sizeof(req_u));
2976 packet_set_ring(sk, &req_u, 1, 0);
2979 if (po->tx_ring.pg_vec) {
2980 memset(&req_u, 0, sizeof(req_u));
2981 packet_set_ring(sk, &req_u, 1, 1);
2984 f = fanout_release(sk);
2989 fanout_release_data(f);
2993 * Now the socket is dead. No more input will appear.
3000 skb_queue_purge(&sk->sk_receive_queue);
3001 packet_free_pending(po);
3002 sk_refcnt_debug_release(sk);
3009 * Attach a packet hook.
3012 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
3015 struct packet_sock *po = pkt_sk(sk);
3016 struct net_device *dev_curr;
3019 struct net_device *dev = NULL;
3021 bool unlisted = false;
3027 spin_lock(&po->bind_lock);
3031 dev = dev_get_by_name_rcu(sock_net(sk), name);
3036 } else if (ifindex) {
3037 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3047 proto_curr = po->prot_hook.type;
3048 dev_curr = po->prot_hook.dev;
3050 need_rehook = proto_curr != proto || dev_curr != dev;
3055 __unregister_prot_hook(sk, true);
3057 dev_curr = po->prot_hook.dev;
3059 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3064 po->prot_hook.type = proto;
3066 if (unlikely(unlisted)) {
3068 po->prot_hook.dev = NULL;
3070 packet_cached_dev_reset(po);
3072 po->prot_hook.dev = dev;
3073 po->ifindex = dev ? dev->ifindex : 0;
3074 packet_cached_dev_assign(po, dev);
3080 if (proto == 0 || !need_rehook)
3083 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3084 register_prot_hook(sk);
3086 sk->sk_err = ENETDOWN;
3087 if (!sock_flag(sk, SOCK_DEAD))
3088 sk->sk_error_report(sk);
3093 spin_unlock(&po->bind_lock);
3099 * Bind a packet socket to a device
3102 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3105 struct sock *sk = sock->sk;
3112 if (addr_len != sizeof(struct sockaddr))
3114 strlcpy(name, uaddr->sa_data, sizeof(name));
3116 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3119 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3121 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3122 struct sock *sk = sock->sk;
3128 if (addr_len < sizeof(struct sockaddr_ll))
3130 if (sll->sll_family != AF_PACKET)
3133 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3134 sll->sll_protocol ? : pkt_sk(sk)->num);
3137 static struct proto packet_proto = {
3139 .owner = THIS_MODULE,
3140 .obj_size = sizeof(struct packet_sock),
3144 * Create a packet of type SOCK_PACKET.
3147 static int packet_create(struct net *net, struct socket *sock, int protocol,
3151 struct packet_sock *po;
3152 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3155 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3157 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3158 sock->type != SOCK_PACKET)
3159 return -ESOCKTNOSUPPORT;
3161 sock->state = SS_UNCONNECTED;
3164 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3168 sock->ops = &packet_ops;
3169 if (sock->type == SOCK_PACKET)
3170 sock->ops = &packet_ops_spkt;
3172 sock_init_data(sock, sk);
3175 sk->sk_family = PF_PACKET;
3177 po->xmit = dev_queue_xmit;
3179 err = packet_alloc_pending(po);
3183 packet_cached_dev_reset(po);
3185 sk->sk_destruct = packet_sock_destruct;
3186 sk_refcnt_debug_inc(sk);
3189 * Attach a protocol block
3192 spin_lock_init(&po->bind_lock);
3193 mutex_init(&po->pg_vec_lock);
3194 po->rollover = NULL;
3195 po->prot_hook.func = packet_rcv;
3197 if (sock->type == SOCK_PACKET)
3198 po->prot_hook.func = packet_rcv_spkt;
3200 po->prot_hook.af_packet_priv = sk;
3203 po->prot_hook.type = proto;
3204 register_prot_hook(sk);
3207 mutex_lock(&net->packet.sklist_lock);
3208 sk_add_node_rcu(sk, &net->packet.sklist);
3209 mutex_unlock(&net->packet.sklist_lock);
3212 sock_prot_inuse_add(net, &packet_proto, 1);
3223 * Pull a packet from our receive queue and hand it to the user.
3224 * If necessary we block.
3227 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3230 struct sock *sk = sock->sk;
3231 struct sk_buff *skb;
3233 int vnet_hdr_len = 0;
3234 unsigned int origlen = 0;
3237 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3241 /* What error should we return now? EUNATTACH? */
3242 if (pkt_sk(sk)->ifindex < 0)
3246 if (flags & MSG_ERRQUEUE) {
3247 err = sock_recv_errqueue(sk, msg, len,
3248 SOL_PACKET, PACKET_TX_TIMESTAMP);
3253 * Call the generic datagram receiver. This handles all sorts
3254 * of horrible races and re-entrancy so we can forget about it
3255 * in the protocol layers.
3257 * Now it will return ENETDOWN, if device have just gone down,
3258 * but then it will block.
3261 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3264 * An error occurred so return it. Because skb_recv_datagram()
3265 * handles the blocking we don't see and worry about blocking
3272 if (pkt_sk(sk)->pressure)
3273 packet_rcv_has_room(pkt_sk(sk), NULL);
3275 if (pkt_sk(sk)->has_vnet_hdr) {
3276 err = packet_rcv_vnet(msg, skb, &len);
3279 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3282 /* You lose any data beyond the buffer you gave. If it worries
3283 * a user program they can ask the device for its MTU
3289 msg->msg_flags |= MSG_TRUNC;
3292 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3296 if (sock->type != SOCK_PACKET) {
3297 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3299 /* Original length was stored in sockaddr_ll fields */
3300 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3301 sll->sll_family = AF_PACKET;
3302 sll->sll_protocol = skb->protocol;
3305 sock_recv_ts_and_drops(msg, sk, skb);
3307 if (msg->msg_name) {
3308 /* If the address length field is there to be filled
3309 * in, we fill it in now.
3311 if (sock->type == SOCK_PACKET) {
3312 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3313 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3315 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3317 msg->msg_namelen = sll->sll_halen +
3318 offsetof(struct sockaddr_ll, sll_addr);
3320 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3324 if (pkt_sk(sk)->auxdata) {
3325 struct tpacket_auxdata aux;
3327 aux.tp_status = TP_STATUS_USER;
3328 if (skb->ip_summed == CHECKSUM_PARTIAL)
3329 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3330 else if (skb->pkt_type != PACKET_OUTGOING &&
3331 (skb->ip_summed == CHECKSUM_COMPLETE ||
3332 skb_csum_unnecessary(skb)))
3333 aux.tp_status |= TP_STATUS_CSUM_VALID;
3335 aux.tp_len = origlen;
3336 aux.tp_snaplen = skb->len;
3338 aux.tp_net = skb_network_offset(skb);
3339 if (skb_vlan_tag_present(skb)) {
3340 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3341 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3342 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3344 aux.tp_vlan_tci = 0;
3345 aux.tp_vlan_tpid = 0;
3347 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3351 * Free or return the buffer as appropriate. Again this
3352 * hides all the races and re-entrancy issues from us.
3354 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3357 skb_free_datagram(sk, skb);
3362 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3363 int *uaddr_len, int peer)
3365 struct net_device *dev;
3366 struct sock *sk = sock->sk;
3371 uaddr->sa_family = AF_PACKET;
3372 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3374 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3376 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3378 *uaddr_len = sizeof(*uaddr);
3383 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3384 int *uaddr_len, int peer)
3386 struct net_device *dev;
3387 struct sock *sk = sock->sk;
3388 struct packet_sock *po = pkt_sk(sk);
3389 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3394 sll->sll_family = AF_PACKET;
3395 sll->sll_ifindex = po->ifindex;
3396 sll->sll_protocol = po->num;
3397 sll->sll_pkttype = 0;
3399 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3401 sll->sll_hatype = dev->type;
3402 sll->sll_halen = dev->addr_len;
3403 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3405 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3409 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3414 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3418 case PACKET_MR_MULTICAST:
3419 if (i->alen != dev->addr_len)
3422 return dev_mc_add(dev, i->addr);
3424 return dev_mc_del(dev, i->addr);
3426 case PACKET_MR_PROMISC:
3427 return dev_set_promiscuity(dev, what);
3428 case PACKET_MR_ALLMULTI:
3429 return dev_set_allmulti(dev, what);
3430 case PACKET_MR_UNICAST:
3431 if (i->alen != dev->addr_len)
3434 return dev_uc_add(dev, i->addr);
3436 return dev_uc_del(dev, i->addr);
3444 static void packet_dev_mclist_delete(struct net_device *dev,
3445 struct packet_mclist **mlp)
3447 struct packet_mclist *ml;
3449 while ((ml = *mlp) != NULL) {
3450 if (ml->ifindex == dev->ifindex) {
3451 packet_dev_mc(dev, ml, -1);
3459 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3461 struct packet_sock *po = pkt_sk(sk);
3462 struct packet_mclist *ml, *i;
3463 struct net_device *dev;
3469 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3474 if (mreq->mr_alen > dev->addr_len)
3478 i = kmalloc(sizeof(*i), GFP_KERNEL);
3483 for (ml = po->mclist; ml; ml = ml->next) {
3484 if (ml->ifindex == mreq->mr_ifindex &&
3485 ml->type == mreq->mr_type &&
3486 ml->alen == mreq->mr_alen &&
3487 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3489 /* Free the new element ... */
3495 i->type = mreq->mr_type;
3496 i->ifindex = mreq->mr_ifindex;
3497 i->alen = mreq->mr_alen;
3498 memcpy(i->addr, mreq->mr_address, i->alen);
3499 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3501 i->next = po->mclist;
3503 err = packet_dev_mc(dev, i, 1);
3505 po->mclist = i->next;
3514 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3516 struct packet_mclist *ml, **mlp;
3520 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3521 if (ml->ifindex == mreq->mr_ifindex &&
3522 ml->type == mreq->mr_type &&
3523 ml->alen == mreq->mr_alen &&
3524 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3525 if (--ml->count == 0) {
3526 struct net_device *dev;
3528 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3530 packet_dev_mc(dev, ml, -1);
3540 static void packet_flush_mclist(struct sock *sk)
3542 struct packet_sock *po = pkt_sk(sk);
3543 struct packet_mclist *ml;
3549 while ((ml = po->mclist) != NULL) {
3550 struct net_device *dev;
3552 po->mclist = ml->next;
3553 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3555 packet_dev_mc(dev, ml, -1);
3562 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3564 struct sock *sk = sock->sk;
3565 struct packet_sock *po = pkt_sk(sk);
3568 if (level != SOL_PACKET)
3569 return -ENOPROTOOPT;
3572 case PACKET_ADD_MEMBERSHIP:
3573 case PACKET_DROP_MEMBERSHIP:
3575 struct packet_mreq_max mreq;
3577 memset(&mreq, 0, sizeof(mreq));
3578 if (len < sizeof(struct packet_mreq))
3580 if (len > sizeof(mreq))
3582 if (copy_from_user(&mreq, optval, len))
3584 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3586 if (optname == PACKET_ADD_MEMBERSHIP)
3587 ret = packet_mc_add(sk, &mreq);
3589 ret = packet_mc_drop(sk, &mreq);
3593 case PACKET_RX_RING:
3594 case PACKET_TX_RING:
3596 union tpacket_req_u req_u;
3599 switch (po->tp_version) {
3602 len = sizeof(req_u.req);
3606 len = sizeof(req_u.req3);
3611 if (copy_from_user(&req_u.req, optval, len))
3613 return packet_set_ring(sk, &req_u, 0,
3614 optname == PACKET_TX_RING);
3616 case PACKET_COPY_THRESH:
3620 if (optlen != sizeof(val))
3622 if (copy_from_user(&val, optval, sizeof(val)))
3625 pkt_sk(sk)->copy_thresh = val;
3628 case PACKET_VERSION:
3632 if (optlen != sizeof(val))
3634 if (copy_from_user(&val, optval, sizeof(val)))
3645 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3648 po->tp_version = val;
3654 case PACKET_RESERVE:
3658 if (optlen != sizeof(val))
3660 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3662 if (copy_from_user(&val, optval, sizeof(val)))
3664 po->tp_reserve = val;
3671 if (optlen != sizeof(val))
3673 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3675 if (copy_from_user(&val, optval, sizeof(val)))
3677 po->tp_loss = !!val;
3680 case PACKET_AUXDATA:
3684 if (optlen < sizeof(val))
3686 if (copy_from_user(&val, optval, sizeof(val)))
3689 po->auxdata = !!val;
3692 case PACKET_ORIGDEV:
3696 if (optlen < sizeof(val))
3698 if (copy_from_user(&val, optval, sizeof(val)))
3701 po->origdev = !!val;
3704 case PACKET_VNET_HDR:
3708 if (sock->type != SOCK_RAW)
3710 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3712 if (optlen < sizeof(val))
3714 if (copy_from_user(&val, optval, sizeof(val)))
3717 po->has_vnet_hdr = !!val;
3720 case PACKET_TIMESTAMP:
3724 if (optlen != sizeof(val))
3726 if (copy_from_user(&val, optval, sizeof(val)))
3729 po->tp_tstamp = val;
3736 if (optlen != sizeof(val))
3738 if (copy_from_user(&val, optval, sizeof(val)))
3741 return fanout_add(sk, val & 0xffff, val >> 16);
3743 case PACKET_FANOUT_DATA:
3748 return fanout_set_data(po, optval, optlen);
3750 case PACKET_TX_HAS_OFF:
3754 if (optlen != sizeof(val))
3756 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3758 if (copy_from_user(&val, optval, sizeof(val)))
3760 po->tp_tx_has_off = !!val;
3763 case PACKET_QDISC_BYPASS:
3767 if (optlen != sizeof(val))
3769 if (copy_from_user(&val, optval, sizeof(val)))
3772 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3776 return -ENOPROTOOPT;
3780 static int packet_getsockopt(struct socket *sock, int level, int optname,
3781 char __user *optval, int __user *optlen)
3784 int val, lv = sizeof(val);
3785 struct sock *sk = sock->sk;
3786 struct packet_sock *po = pkt_sk(sk);
3788 union tpacket_stats_u st;
3789 struct tpacket_rollover_stats rstats;
3791 if (level != SOL_PACKET)
3792 return -ENOPROTOOPT;
3794 if (get_user(len, optlen))
3801 case PACKET_STATISTICS:
3802 spin_lock_bh(&sk->sk_receive_queue.lock);
3803 memcpy(&st, &po->stats, sizeof(st));
3804 memset(&po->stats, 0, sizeof(po->stats));
3805 spin_unlock_bh(&sk->sk_receive_queue.lock);
3807 if (po->tp_version == TPACKET_V3) {
3808 lv = sizeof(struct tpacket_stats_v3);
3809 st.stats3.tp_packets += st.stats3.tp_drops;
3812 lv = sizeof(struct tpacket_stats);
3813 st.stats1.tp_packets += st.stats1.tp_drops;
3818 case PACKET_AUXDATA:
3821 case PACKET_ORIGDEV:
3824 case PACKET_VNET_HDR:
3825 val = po->has_vnet_hdr;
3827 case PACKET_VERSION:
3828 val = po->tp_version;
3831 if (len > sizeof(int))
3833 if (copy_from_user(&val, optval, len))
3837 val = sizeof(struct tpacket_hdr);
3840 val = sizeof(struct tpacket2_hdr);
3843 val = sizeof(struct tpacket3_hdr);
3849 case PACKET_RESERVE:
3850 val = po->tp_reserve;
3855 case PACKET_TIMESTAMP:
3856 val = po->tp_tstamp;
3860 ((u32)po->fanout->id |
3861 ((u32)po->fanout->type << 16) |
3862 ((u32)po->fanout->flags << 24)) :
3865 case PACKET_ROLLOVER_STATS:
3868 rstats.tp_all = atomic_long_read(&po->rollover->num);
3869 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3870 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3872 lv = sizeof(rstats);
3874 case PACKET_TX_HAS_OFF:
3875 val = po->tp_tx_has_off;
3877 case PACKET_QDISC_BYPASS:
3878 val = packet_use_direct_xmit(po);
3881 return -ENOPROTOOPT;
3886 if (put_user(len, optlen))
3888 if (copy_to_user(optval, data, len))
3894 #ifdef CONFIG_COMPAT
3895 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3896 char __user *optval, unsigned int optlen)
3898 struct packet_sock *po = pkt_sk(sock->sk);
3900 if (level != SOL_PACKET)
3901 return -ENOPROTOOPT;
3903 if (optname == PACKET_FANOUT_DATA &&
3904 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3905 optval = (char __user *)get_compat_bpf_fprog(optval);
3908 optlen = sizeof(struct sock_fprog);
3911 return packet_setsockopt(sock, level, optname, optval, optlen);
3915 static int packet_notifier(struct notifier_block *this,
3916 unsigned long msg, void *ptr)
3919 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3920 struct net *net = dev_net(dev);
3923 sk_for_each_rcu(sk, &net->packet.sklist) {
3924 struct packet_sock *po = pkt_sk(sk);
3927 case NETDEV_UNREGISTER:
3929 packet_dev_mclist_delete(dev, &po->mclist);
3933 if (dev->ifindex == po->ifindex) {
3934 spin_lock(&po->bind_lock);
3936 __unregister_prot_hook(sk, false);
3937 sk->sk_err = ENETDOWN;
3938 if (!sock_flag(sk, SOCK_DEAD))
3939 sk->sk_error_report(sk);
3941 if (msg == NETDEV_UNREGISTER) {
3942 packet_cached_dev_reset(po);
3944 if (po->prot_hook.dev)
3945 dev_put(po->prot_hook.dev);
3946 po->prot_hook.dev = NULL;
3948 spin_unlock(&po->bind_lock);
3952 if (dev->ifindex == po->ifindex) {
3953 spin_lock(&po->bind_lock);
3955 register_prot_hook(sk);
3956 spin_unlock(&po->bind_lock);
3966 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3969 struct sock *sk = sock->sk;
3974 int amount = sk_wmem_alloc_get(sk);
3976 return put_user(amount, (int __user *)arg);
3980 struct sk_buff *skb;
3983 spin_lock_bh(&sk->sk_receive_queue.lock);
3984 skb = skb_peek(&sk->sk_receive_queue);
3987 spin_unlock_bh(&sk->sk_receive_queue.lock);
3988 return put_user(amount, (int __user *)arg);
3991 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3993 return sock_get_timestampns(sk, (struct timespec __user *)arg);
4003 case SIOCGIFBRDADDR:
4004 case SIOCSIFBRDADDR:
4005 case SIOCGIFNETMASK:
4006 case SIOCSIFNETMASK:
4007 case SIOCGIFDSTADDR:
4008 case SIOCSIFDSTADDR:
4010 return inet_dgram_ops.ioctl(sock, cmd, arg);
4014 return -ENOIOCTLCMD;
4019 static unsigned int packet_poll(struct file *file, struct socket *sock,
4022 struct sock *sk = sock->sk;
4023 struct packet_sock *po = pkt_sk(sk);
4024 unsigned int mask = datagram_poll(file, sock, wait);
4026 spin_lock_bh(&sk->sk_receive_queue.lock);
4027 if (po->rx_ring.pg_vec) {
4028 if (!packet_previous_rx_frame(po, &po->rx_ring,
4030 mask |= POLLIN | POLLRDNORM;
4032 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4034 spin_unlock_bh(&sk->sk_receive_queue.lock);
4035 spin_lock_bh(&sk->sk_write_queue.lock);
4036 if (po->tx_ring.pg_vec) {
4037 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4038 mask |= POLLOUT | POLLWRNORM;
4040 spin_unlock_bh(&sk->sk_write_queue.lock);
4045 /* Dirty? Well, I still did not learn better way to account
4049 static void packet_mm_open(struct vm_area_struct *vma)
4051 struct file *file = vma->vm_file;
4052 struct socket *sock = file->private_data;
4053 struct sock *sk = sock->sk;
4056 atomic_inc(&pkt_sk(sk)->mapped);
4059 static void packet_mm_close(struct vm_area_struct *vma)
4061 struct file *file = vma->vm_file;
4062 struct socket *sock = file->private_data;
4063 struct sock *sk = sock->sk;
4066 atomic_dec(&pkt_sk(sk)->mapped);
4069 static const struct vm_operations_struct packet_mmap_ops = {
4070 .open = packet_mm_open,
4071 .close = packet_mm_close,
4074 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4079 for (i = 0; i < len; i++) {
4080 if (likely(pg_vec[i].buffer)) {
4081 if (is_vmalloc_addr(pg_vec[i].buffer))
4082 vfree(pg_vec[i].buffer);
4084 free_pages((unsigned long)pg_vec[i].buffer,
4086 pg_vec[i].buffer = NULL;
4092 static char *alloc_one_pg_vec_page(unsigned long order)
4095 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4096 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4098 buffer = (char *) __get_free_pages(gfp_flags, order);
4102 /* __get_free_pages failed, fall back to vmalloc */
4103 buffer = vzalloc((1 << order) * PAGE_SIZE);
4107 /* vmalloc failed, lets dig into swap here */
4108 gfp_flags &= ~__GFP_NORETRY;
4109 buffer = (char *) __get_free_pages(gfp_flags, order);
4113 /* complete and utter failure */
4117 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4119 unsigned int block_nr = req->tp_block_nr;
4123 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4124 if (unlikely(!pg_vec))
4127 for (i = 0; i < block_nr; i++) {
4128 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4129 if (unlikely(!pg_vec[i].buffer))
4130 goto out_free_pgvec;
4137 free_pg_vec(pg_vec, order, block_nr);
4142 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4143 int closing, int tx_ring)
4145 struct pgv *pg_vec = NULL;
4146 struct packet_sock *po = pkt_sk(sk);
4147 int was_running, order = 0;
4148 struct packet_ring_buffer *rb;
4149 struct sk_buff_head *rb_queue;
4152 /* Added to avoid minimal code churn */
4153 struct tpacket_req *req = &req_u->req;
4157 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4158 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4162 if (atomic_read(&po->mapped))
4164 if (packet_read_pending(rb))
4168 if (req->tp_block_nr) {
4169 /* Sanity tests and some calculations */
4171 if (unlikely(rb->pg_vec))
4174 switch (po->tp_version) {
4176 po->tp_hdrlen = TPACKET_HDRLEN;
4179 po->tp_hdrlen = TPACKET2_HDRLEN;
4182 po->tp_hdrlen = TPACKET3_HDRLEN;
4187 if (unlikely((int)req->tp_block_size <= 0))
4189 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4191 if (po->tp_version >= TPACKET_V3 &&
4192 (int)(req->tp_block_size -
4193 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4195 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4198 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4201 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4202 if (unlikely(rb->frames_per_block == 0))
4204 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4209 order = get_order(req->tp_block_size);
4210 pg_vec = alloc_pg_vec(req, order);
4211 if (unlikely(!pg_vec))
4213 switch (po->tp_version) {
4215 /* Block transmit is not supported yet */
4217 init_prb_bdqc(po, rb, pg_vec, req_u);
4219 struct tpacket_req3 *req3 = &req_u->req3;
4221 if (req3->tp_retire_blk_tov ||
4222 req3->tp_sizeof_priv ||
4223 req3->tp_feature_req_word) {
4236 if (unlikely(req->tp_frame_nr))
4241 /* Detach socket from network */
4242 spin_lock(&po->bind_lock);
4243 was_running = po->running;
4247 __unregister_prot_hook(sk, false);
4249 spin_unlock(&po->bind_lock);
4254 mutex_lock(&po->pg_vec_lock);
4255 if (closing || atomic_read(&po->mapped) == 0) {
4257 spin_lock_bh(&rb_queue->lock);
4258 swap(rb->pg_vec, pg_vec);
4259 rb->frame_max = (req->tp_frame_nr - 1);
4261 rb->frame_size = req->tp_frame_size;
4262 spin_unlock_bh(&rb_queue->lock);
4264 swap(rb->pg_vec_order, order);
4265 swap(rb->pg_vec_len, req->tp_block_nr);
4267 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4268 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4269 tpacket_rcv : packet_rcv;
4270 skb_queue_purge(rb_queue);
4271 if (atomic_read(&po->mapped))
4272 pr_err("packet_mmap: vma is busy: %d\n",
4273 atomic_read(&po->mapped));
4275 mutex_unlock(&po->pg_vec_lock);
4277 spin_lock(&po->bind_lock);
4280 register_prot_hook(sk);
4282 spin_unlock(&po->bind_lock);
4283 if (closing && (po->tp_version > TPACKET_V2)) {
4284 /* Because we don't support block-based V3 on tx-ring */
4286 prb_shutdown_retire_blk_timer(po, rb_queue);
4290 free_pg_vec(pg_vec, order, req->tp_block_nr);
4296 static int packet_mmap(struct file *file, struct socket *sock,
4297 struct vm_area_struct *vma)
4299 struct sock *sk = sock->sk;
4300 struct packet_sock *po = pkt_sk(sk);
4301 unsigned long size, expected_size;
4302 struct packet_ring_buffer *rb;
4303 unsigned long start;
4310 mutex_lock(&po->pg_vec_lock);
4313 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4315 expected_size += rb->pg_vec_len
4321 if (expected_size == 0)
4324 size = vma->vm_end - vma->vm_start;
4325 if (size != expected_size)
4328 start = vma->vm_start;
4329 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4330 if (rb->pg_vec == NULL)
4333 for (i = 0; i < rb->pg_vec_len; i++) {
4335 void *kaddr = rb->pg_vec[i].buffer;
4338 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4339 page = pgv_to_page(kaddr);
4340 err = vm_insert_page(vma, start, page);
4349 atomic_inc(&po->mapped);
4350 vma->vm_ops = &packet_mmap_ops;
4354 mutex_unlock(&po->pg_vec_lock);
4358 static const struct proto_ops packet_ops_spkt = {
4359 .family = PF_PACKET,
4360 .owner = THIS_MODULE,
4361 .release = packet_release,
4362 .bind = packet_bind_spkt,
4363 .connect = sock_no_connect,
4364 .socketpair = sock_no_socketpair,
4365 .accept = sock_no_accept,
4366 .getname = packet_getname_spkt,
4367 .poll = datagram_poll,
4368 .ioctl = packet_ioctl,
4369 .listen = sock_no_listen,
4370 .shutdown = sock_no_shutdown,
4371 .setsockopt = sock_no_setsockopt,
4372 .getsockopt = sock_no_getsockopt,
4373 .sendmsg = packet_sendmsg_spkt,
4374 .recvmsg = packet_recvmsg,
4375 .mmap = sock_no_mmap,
4376 .sendpage = sock_no_sendpage,
4379 static const struct proto_ops packet_ops = {
4380 .family = PF_PACKET,
4381 .owner = THIS_MODULE,
4382 .release = packet_release,
4383 .bind = packet_bind,
4384 .connect = sock_no_connect,
4385 .socketpair = sock_no_socketpair,
4386 .accept = sock_no_accept,
4387 .getname = packet_getname,
4388 .poll = packet_poll,
4389 .ioctl = packet_ioctl,
4390 .listen = sock_no_listen,
4391 .shutdown = sock_no_shutdown,
4392 .setsockopt = packet_setsockopt,
4393 .getsockopt = packet_getsockopt,
4394 #ifdef CONFIG_COMPAT
4395 .compat_setsockopt = compat_packet_setsockopt,
4397 .sendmsg = packet_sendmsg,
4398 .recvmsg = packet_recvmsg,
4399 .mmap = packet_mmap,
4400 .sendpage = sock_no_sendpage,
4403 static const struct net_proto_family packet_family_ops = {
4404 .family = PF_PACKET,
4405 .create = packet_create,
4406 .owner = THIS_MODULE,
4409 static struct notifier_block packet_netdev_notifier = {
4410 .notifier_call = packet_notifier,
4413 #ifdef CONFIG_PROC_FS
4415 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4418 struct net *net = seq_file_net(seq);
4421 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4424 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4426 struct net *net = seq_file_net(seq);
4427 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4430 static void packet_seq_stop(struct seq_file *seq, void *v)
4436 static int packet_seq_show(struct seq_file *seq, void *v)
4438 if (v == SEQ_START_TOKEN)
4439 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4441 struct sock *s = sk_entry(v);
4442 const struct packet_sock *po = pkt_sk(s);
4445 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4447 atomic_read(&s->sk_refcnt),
4452 atomic_read(&s->sk_rmem_alloc),
4453 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4460 static const struct seq_operations packet_seq_ops = {
4461 .start = packet_seq_start,
4462 .next = packet_seq_next,
4463 .stop = packet_seq_stop,
4464 .show = packet_seq_show,
4467 static int packet_seq_open(struct inode *inode, struct file *file)
4469 return seq_open_net(inode, file, &packet_seq_ops,
4470 sizeof(struct seq_net_private));
4473 static const struct file_operations packet_seq_fops = {
4474 .owner = THIS_MODULE,
4475 .open = packet_seq_open,
4477 .llseek = seq_lseek,
4478 .release = seq_release_net,
4483 static int __net_init packet_net_init(struct net *net)
4485 mutex_init(&net->packet.sklist_lock);
4486 INIT_HLIST_HEAD(&net->packet.sklist);
4488 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4494 static void __net_exit packet_net_exit(struct net *net)
4496 remove_proc_entry("packet", net->proc_net);
4499 static struct pernet_operations packet_net_ops = {
4500 .init = packet_net_init,
4501 .exit = packet_net_exit,
4505 static void __exit packet_exit(void)
4507 unregister_netdevice_notifier(&packet_netdev_notifier);
4508 unregister_pernet_subsys(&packet_net_ops);
4509 sock_unregister(PF_PACKET);
4510 proto_unregister(&packet_proto);
4513 static int __init packet_init(void)
4515 int rc = proto_register(&packet_proto, 0);
4520 sock_register(&packet_family_ops);
4521 register_pernet_subsys(&packet_net_ops);
4522 register_netdevice_notifier(&packet_netdev_notifier);
4527 module_init(packet_init);
4528 module_exit(packet_exit);
4529 MODULE_LICENSE("GPL");
4530 MODULE_ALIAS_NETPROTO(PF_PACKET);