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));
413 WARN(1, "TPACKET version not supported.\n");
420 static int __packet_get_status(struct packet_sock *po, void *frame)
422 union tpacket_uhdr h;
427 switch (po->tp_version) {
429 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
430 return h.h1->tp_status;
432 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
433 return h.h2->tp_status;
436 WARN(1, "TPACKET version not supported.\n");
442 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
445 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
448 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
449 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
450 return TP_STATUS_TS_RAW_HARDWARE;
452 if (ktime_to_timespec_cond(skb->tstamp, ts))
453 return TP_STATUS_TS_SOFTWARE;
458 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
461 union tpacket_uhdr h;
465 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
469 switch (po->tp_version) {
471 h.h1->tp_sec = ts.tv_sec;
472 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
475 h.h2->tp_sec = ts.tv_sec;
476 h.h2->tp_nsec = ts.tv_nsec;
480 WARN(1, "TPACKET version not supported.\n");
484 /* one flush is safe, as both fields always lie on the same cacheline */
485 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
491 static void *packet_lookup_frame(struct packet_sock *po,
492 struct packet_ring_buffer *rb,
493 unsigned int position,
496 unsigned int pg_vec_pos, frame_offset;
497 union tpacket_uhdr h;
499 pg_vec_pos = position / rb->frames_per_block;
500 frame_offset = position % rb->frames_per_block;
502 h.raw = rb->pg_vec[pg_vec_pos].buffer +
503 (frame_offset * rb->frame_size);
505 if (status != __packet_get_status(po, h.raw))
511 static void *packet_current_frame(struct packet_sock *po,
512 struct packet_ring_buffer *rb,
515 return packet_lookup_frame(po, rb, rb->head, status);
518 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
520 del_timer_sync(&pkc->retire_blk_timer);
523 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
524 struct sk_buff_head *rb_queue)
526 struct tpacket_kbdq_core *pkc;
528 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
530 spin_lock_bh(&rb_queue->lock);
531 pkc->delete_blk_timer = 1;
532 spin_unlock_bh(&rb_queue->lock);
534 prb_del_retire_blk_timer(pkc);
537 static void prb_init_blk_timer(struct packet_sock *po,
538 struct tpacket_kbdq_core *pkc,
539 void (*func) (unsigned long))
541 init_timer(&pkc->retire_blk_timer);
542 pkc->retire_blk_timer.data = (long)po;
543 pkc->retire_blk_timer.function = func;
544 pkc->retire_blk_timer.expires = jiffies;
547 static void prb_setup_retire_blk_timer(struct packet_sock *po)
549 struct tpacket_kbdq_core *pkc;
551 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
552 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
555 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
556 int blk_size_in_bytes)
558 struct net_device *dev;
559 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
560 struct ethtool_link_ksettings ecmd;
564 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
565 if (unlikely(!dev)) {
567 return DEFAULT_PRB_RETIRE_TOV;
569 err = __ethtool_get_link_ksettings(dev, &ecmd);
573 * If the link speed is so slow you don't really
574 * need to worry about perf anyways
576 if (ecmd.base.speed < SPEED_1000 ||
577 ecmd.base.speed == SPEED_UNKNOWN) {
578 return DEFAULT_PRB_RETIRE_TOV;
581 div = ecmd.base.speed / 1000;
585 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
597 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
598 union tpacket_req_u *req_u)
600 p1->feature_req_word = req_u->req3.tp_feature_req_word;
603 static void init_prb_bdqc(struct packet_sock *po,
604 struct packet_ring_buffer *rb,
606 union tpacket_req_u *req_u)
608 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
609 struct tpacket_block_desc *pbd;
611 memset(p1, 0x0, sizeof(*p1));
613 p1->knxt_seq_num = 1;
615 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
616 p1->pkblk_start = pg_vec[0].buffer;
617 p1->kblk_size = req_u->req3.tp_block_size;
618 p1->knum_blocks = req_u->req3.tp_block_nr;
619 p1->hdrlen = po->tp_hdrlen;
620 p1->version = po->tp_version;
621 p1->last_kactive_blk_num = 0;
622 po->stats.stats3.tp_freeze_q_cnt = 0;
623 if (req_u->req3.tp_retire_blk_tov)
624 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
626 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
627 req_u->req3.tp_block_size);
628 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
629 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
631 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
632 prb_init_ft_ops(p1, req_u);
633 prb_setup_retire_blk_timer(po);
634 prb_open_block(p1, pbd);
637 /* Do NOT update the last_blk_num first.
638 * Assumes sk_buff_head lock is held.
640 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
642 mod_timer(&pkc->retire_blk_timer,
643 jiffies + pkc->tov_in_jiffies);
644 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
649 * 1) We refresh the timer only when we open a block.
650 * By doing this we don't waste cycles refreshing the timer
651 * on packet-by-packet basis.
653 * With a 1MB block-size, on a 1Gbps line, it will take
654 * i) ~8 ms to fill a block + ii) memcpy etc.
655 * In this cut we are not accounting for the memcpy time.
657 * So, if the user sets the 'tmo' to 10ms then the timer
658 * will never fire while the block is still getting filled
659 * (which is what we want). However, the user could choose
660 * to close a block early and that's fine.
662 * But when the timer does fire, we check whether or not to refresh it.
663 * Since the tmo granularity is in msecs, it is not too expensive
664 * to refresh the timer, lets say every '8' msecs.
665 * Either the user can set the 'tmo' or we can derive it based on
666 * a) line-speed and b) block-size.
667 * prb_calc_retire_blk_tmo() calculates the tmo.
670 static void prb_retire_rx_blk_timer_expired(unsigned long data)
672 struct packet_sock *po = (struct packet_sock *)data;
673 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
675 struct tpacket_block_desc *pbd;
677 spin_lock(&po->sk.sk_receive_queue.lock);
679 frozen = prb_queue_frozen(pkc);
680 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
682 if (unlikely(pkc->delete_blk_timer))
685 /* We only need to plug the race when the block is partially filled.
687 * lock(); increment BLOCK_NUM_PKTS; unlock()
688 * copy_bits() is in progress ...
689 * timer fires on other cpu:
690 * we can't retire the current block because copy_bits
694 if (BLOCK_NUM_PKTS(pbd)) {
695 while (atomic_read(&pkc->blk_fill_in_prog)) {
696 /* Waiting for skb_copy_bits to finish... */
701 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
703 if (!BLOCK_NUM_PKTS(pbd)) {
704 /* An empty block. Just refresh the timer. */
707 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
708 if (!prb_dispatch_next_block(pkc, po))
713 /* Case 1. Queue was frozen because user-space was
716 if (prb_curr_blk_in_use(pkc, pbd)) {
718 * Ok, user-space is still behind.
719 * So just refresh the timer.
723 /* Case 2. queue was frozen,user-space caught up,
724 * now the link went idle && the timer fired.
725 * We don't have a block to close.So we open this
726 * block and restart the timer.
727 * opening a block thaws the queue,restarts timer
728 * Thawing/timer-refresh is a side effect.
730 prb_open_block(pkc, pbd);
737 _prb_refresh_rx_retire_blk_timer(pkc);
740 spin_unlock(&po->sk.sk_receive_queue.lock);
743 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
744 struct tpacket_block_desc *pbd1, __u32 status)
746 /* Flush everything minus the block header */
748 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
753 /* Skip the block header(we know header WILL fit in 4K) */
756 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
757 for (; start < end; start += PAGE_SIZE)
758 flush_dcache_page(pgv_to_page(start));
763 /* Now update the block status. */
765 BLOCK_STATUS(pbd1) = status;
767 /* Flush the block header */
769 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
771 flush_dcache_page(pgv_to_page(start));
781 * 2) Increment active_blk_num
783 * Note:We DONT refresh the timer on purpose.
784 * Because almost always the next block will be opened.
786 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
787 struct tpacket_block_desc *pbd1,
788 struct packet_sock *po, unsigned int stat)
790 __u32 status = TP_STATUS_USER | stat;
792 struct tpacket3_hdr *last_pkt;
793 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
794 struct sock *sk = &po->sk;
796 if (po->stats.stats3.tp_drops)
797 status |= TP_STATUS_LOSING;
799 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
800 last_pkt->tp_next_offset = 0;
802 /* Get the ts of the last pkt */
803 if (BLOCK_NUM_PKTS(pbd1)) {
804 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
805 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
807 /* Ok, we tmo'd - so get the current time.
809 * It shouldn't really happen as we don't close empty
810 * blocks. See prb_retire_rx_blk_timer_expired().
814 h1->ts_last_pkt.ts_sec = ts.tv_sec;
815 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
820 /* Flush the block */
821 prb_flush_block(pkc1, pbd1, status);
823 sk->sk_data_ready(sk);
825 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
828 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
830 pkc->reset_pending_on_curr_blk = 0;
834 * Side effect of opening a block:
836 * 1) prb_queue is thawed.
837 * 2) retire_blk_timer is refreshed.
840 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
841 struct tpacket_block_desc *pbd1)
844 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
848 /* We could have just memset this but we will lose the
849 * flexibility of making the priv area sticky
852 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
853 BLOCK_NUM_PKTS(pbd1) = 0;
854 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
858 h1->ts_first_pkt.ts_sec = ts.tv_sec;
859 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
861 pkc1->pkblk_start = (char *)pbd1;
862 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
864 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
865 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
867 pbd1->version = pkc1->version;
868 pkc1->prev = pkc1->nxt_offset;
869 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
871 prb_thaw_queue(pkc1);
872 _prb_refresh_rx_retire_blk_timer(pkc1);
878 * Queue freeze logic:
879 * 1) Assume tp_block_nr = 8 blocks.
880 * 2) At time 't0', user opens Rx ring.
881 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
882 * 4) user-space is either sleeping or processing block '0'.
883 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
884 * it will close block-7,loop around and try to fill block '0'.
886 * __packet_lookup_frame_in_block
887 * prb_retire_current_block()
888 * prb_dispatch_next_block()
889 * |->(BLOCK_STATUS == USER) evaluates to true
890 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
891 * 6) Now there are two cases:
892 * 6.1) Link goes idle right after the queue is frozen.
893 * But remember, the last open_block() refreshed the timer.
894 * When this timer expires,it will refresh itself so that we can
895 * re-open block-0 in near future.
896 * 6.2) Link is busy and keeps on receiving packets. This is a simple
897 * case and __packet_lookup_frame_in_block will check if block-0
898 * is free and can now be re-used.
900 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
901 struct packet_sock *po)
903 pkc->reset_pending_on_curr_blk = 1;
904 po->stats.stats3.tp_freeze_q_cnt++;
907 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
910 * If the next block is free then we will dispatch it
911 * and return a good offset.
912 * Else, we will freeze the queue.
913 * So, caller must check the return value.
915 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
916 struct packet_sock *po)
918 struct tpacket_block_desc *pbd;
922 /* 1. Get current block num */
923 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
925 /* 2. If this block is currently in_use then freeze the queue */
926 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
927 prb_freeze_queue(pkc, po);
933 * open this block and return the offset where the first packet
934 * needs to get stored.
936 prb_open_block(pkc, pbd);
937 return (void *)pkc->nxt_offset;
940 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
941 struct packet_sock *po, unsigned int status)
943 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
945 /* retire/close the current block */
946 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
948 * Plug the case where copy_bits() is in progress on
949 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
950 * have space to copy the pkt in the current block and
951 * called prb_retire_current_block()
953 * We don't need to worry about the TMO case because
954 * the timer-handler already handled this case.
956 if (!(status & TP_STATUS_BLK_TMO)) {
957 while (atomic_read(&pkc->blk_fill_in_prog)) {
958 /* Waiting for skb_copy_bits to finish... */
962 prb_close_block(pkc, pbd, po, status);
967 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
968 struct tpacket_block_desc *pbd)
970 return TP_STATUS_USER & BLOCK_STATUS(pbd);
973 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
975 return pkc->reset_pending_on_curr_blk;
978 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
980 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
981 atomic_dec(&pkc->blk_fill_in_prog);
984 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
985 struct tpacket3_hdr *ppd)
987 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
990 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
991 struct tpacket3_hdr *ppd)
993 ppd->hv1.tp_rxhash = 0;
996 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
997 struct tpacket3_hdr *ppd)
999 if (skb_vlan_tag_present(pkc->skb)) {
1000 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1001 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1002 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1004 ppd->hv1.tp_vlan_tci = 0;
1005 ppd->hv1.tp_vlan_tpid = 0;
1006 ppd->tp_status = TP_STATUS_AVAILABLE;
1010 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1011 struct tpacket3_hdr *ppd)
1013 ppd->hv1.tp_padding = 0;
1014 prb_fill_vlan_info(pkc, ppd);
1016 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1017 prb_fill_rxhash(pkc, ppd);
1019 prb_clear_rxhash(pkc, ppd);
1022 static void prb_fill_curr_block(char *curr,
1023 struct tpacket_kbdq_core *pkc,
1024 struct tpacket_block_desc *pbd,
1027 struct tpacket3_hdr *ppd;
1029 ppd = (struct tpacket3_hdr *)curr;
1030 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1032 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1033 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1034 BLOCK_NUM_PKTS(pbd) += 1;
1035 atomic_inc(&pkc->blk_fill_in_prog);
1036 prb_run_all_ft_ops(pkc, ppd);
1039 /* Assumes caller has the sk->rx_queue.lock */
1040 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1041 struct sk_buff *skb,
1046 struct tpacket_kbdq_core *pkc;
1047 struct tpacket_block_desc *pbd;
1050 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1051 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1053 /* Queue is frozen when user space is lagging behind */
1054 if (prb_queue_frozen(pkc)) {
1056 * Check if that last block which caused the queue to freeze,
1057 * is still in_use by user-space.
1059 if (prb_curr_blk_in_use(pkc, pbd)) {
1060 /* Can't record this packet */
1064 * Ok, the block was released by user-space.
1065 * Now let's open that block.
1066 * opening a block also thaws the queue.
1067 * Thawing is a side effect.
1069 prb_open_block(pkc, pbd);
1074 curr = pkc->nxt_offset;
1076 end = (char *)pbd + pkc->kblk_size;
1078 /* first try the current block */
1079 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1080 prb_fill_curr_block(curr, pkc, pbd, len);
1081 return (void *)curr;
1084 /* Ok, close the current block */
1085 prb_retire_current_block(pkc, po, 0);
1087 /* Now, try to dispatch the next block */
1088 curr = (char *)prb_dispatch_next_block(pkc, po);
1090 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1091 prb_fill_curr_block(curr, pkc, pbd, len);
1092 return (void *)curr;
1096 * No free blocks are available.user_space hasn't caught up yet.
1097 * Queue was just frozen and now this packet will get dropped.
1102 static void *packet_current_rx_frame(struct packet_sock *po,
1103 struct sk_buff *skb,
1104 int status, unsigned int len)
1107 switch (po->tp_version) {
1110 curr = packet_lookup_frame(po, &po->rx_ring,
1111 po->rx_ring.head, status);
1114 return __packet_lookup_frame_in_block(po, skb, status, len);
1116 WARN(1, "TPACKET version not supported\n");
1122 static void *prb_lookup_block(struct packet_sock *po,
1123 struct packet_ring_buffer *rb,
1127 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1128 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1130 if (status != BLOCK_STATUS(pbd))
1135 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1138 if (rb->prb_bdqc.kactive_blk_num)
1139 prev = rb->prb_bdqc.kactive_blk_num-1;
1141 prev = rb->prb_bdqc.knum_blocks-1;
1145 /* Assumes caller has held the rx_queue.lock */
1146 static void *__prb_previous_block(struct packet_sock *po,
1147 struct packet_ring_buffer *rb,
1150 unsigned int previous = prb_previous_blk_num(rb);
1151 return prb_lookup_block(po, rb, previous, status);
1154 static void *packet_previous_rx_frame(struct packet_sock *po,
1155 struct packet_ring_buffer *rb,
1158 if (po->tp_version <= TPACKET_V2)
1159 return packet_previous_frame(po, rb, status);
1161 return __prb_previous_block(po, rb, status);
1164 static void packet_increment_rx_head(struct packet_sock *po,
1165 struct packet_ring_buffer *rb)
1167 switch (po->tp_version) {
1170 return packet_increment_head(rb);
1173 WARN(1, "TPACKET version not supported.\n");
1179 static void *packet_previous_frame(struct packet_sock *po,
1180 struct packet_ring_buffer *rb,
1183 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1184 return packet_lookup_frame(po, rb, previous, status);
1187 static void packet_increment_head(struct packet_ring_buffer *buff)
1189 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1192 static void packet_inc_pending(struct packet_ring_buffer *rb)
1194 this_cpu_inc(*rb->pending_refcnt);
1197 static void packet_dec_pending(struct packet_ring_buffer *rb)
1199 this_cpu_dec(*rb->pending_refcnt);
1202 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1204 unsigned int refcnt = 0;
1207 /* We don't use pending refcount in rx_ring. */
1208 if (rb->pending_refcnt == NULL)
1211 for_each_possible_cpu(cpu)
1212 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1217 static int packet_alloc_pending(struct packet_sock *po)
1219 po->rx_ring.pending_refcnt = NULL;
1221 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1222 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1228 static void packet_free_pending(struct packet_sock *po)
1230 free_percpu(po->tx_ring.pending_refcnt);
1233 #define ROOM_POW_OFF 2
1234 #define ROOM_NONE 0x0
1235 #define ROOM_LOW 0x1
1236 #define ROOM_NORMAL 0x2
1238 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1242 len = po->rx_ring.frame_max + 1;
1243 idx = po->rx_ring.head;
1245 idx += len >> pow_off;
1248 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1251 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1255 len = po->rx_ring.prb_bdqc.knum_blocks;
1256 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1258 idx += len >> pow_off;
1261 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1264 static int __packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1266 struct sock *sk = &po->sk;
1267 int ret = ROOM_NONE;
1269 if (po->prot_hook.func != tpacket_rcv) {
1270 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1271 - (skb ? skb->truesize : 0);
1272 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1280 if (po->tp_version == TPACKET_V3) {
1281 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1283 else if (__tpacket_v3_has_room(po, 0))
1286 if (__tpacket_has_room(po, ROOM_POW_OFF))
1288 else if (__tpacket_has_room(po, 0))
1295 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1300 spin_lock_bh(&po->sk.sk_receive_queue.lock);
1301 ret = __packet_rcv_has_room(po, skb);
1302 has_room = ret == ROOM_NORMAL;
1303 if (po->pressure == has_room)
1304 po->pressure = !has_room;
1305 spin_unlock_bh(&po->sk.sk_receive_queue.lock);
1310 static void packet_sock_destruct(struct sock *sk)
1312 skb_queue_purge(&sk->sk_error_queue);
1314 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1315 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1317 if (!sock_flag(sk, SOCK_DEAD)) {
1318 pr_err("Attempt to release alive packet socket: %p\n", sk);
1322 sk_refcnt_debug_dec(sk);
1325 static bool fanout_flow_is_huge(struct packet_sock *po, struct sk_buff *skb)
1330 rxhash = skb_get_hash(skb);
1331 for (i = 0; i < ROLLOVER_HLEN; i++)
1332 if (po->rollover->history[i] == rxhash)
1335 po->rollover->history[prandom_u32() % ROLLOVER_HLEN] = rxhash;
1336 return count > (ROLLOVER_HLEN >> 1);
1339 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1340 struct sk_buff *skb,
1343 return reciprocal_scale(__skb_get_hash_symmetric(skb), num);
1346 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1347 struct sk_buff *skb,
1350 unsigned int val = atomic_inc_return(&f->rr_cur);
1355 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1356 struct sk_buff *skb,
1359 return smp_processor_id() % num;
1362 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1363 struct sk_buff *skb,
1366 return prandom_u32_max(num);
1369 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1370 struct sk_buff *skb,
1371 unsigned int idx, bool try_self,
1374 struct packet_sock *po, *po_next, *po_skip = NULL;
1375 unsigned int i, j, room = ROOM_NONE;
1377 po = pkt_sk(f->arr[idx]);
1380 room = packet_rcv_has_room(po, skb);
1381 if (room == ROOM_NORMAL ||
1382 (room == ROOM_LOW && !fanout_flow_is_huge(po, skb)))
1387 i = j = min_t(int, po->rollover->sock, num - 1);
1389 po_next = pkt_sk(f->arr[i]);
1390 if (po_next != po_skip && !po_next->pressure &&
1391 packet_rcv_has_room(po_next, skb) == ROOM_NORMAL) {
1393 po->rollover->sock = i;
1394 atomic_long_inc(&po->rollover->num);
1395 if (room == ROOM_LOW)
1396 atomic_long_inc(&po->rollover->num_huge);
1404 atomic_long_inc(&po->rollover->num_failed);
1408 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1409 struct sk_buff *skb,
1412 return skb_get_queue_mapping(skb) % num;
1415 static unsigned int fanout_demux_bpf(struct packet_fanout *f,
1416 struct sk_buff *skb,
1419 struct bpf_prog *prog;
1420 unsigned int ret = 0;
1423 prog = rcu_dereference(f->bpf_prog);
1425 ret = bpf_prog_run_clear_cb(prog, skb) % num;
1431 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1433 return f->flags & (flag >> 8);
1436 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1437 struct packet_type *pt, struct net_device *orig_dev)
1439 struct packet_fanout *f = pt->af_packet_priv;
1440 unsigned int num = READ_ONCE(f->num_members);
1441 struct net *net = read_pnet(&f->net);
1442 struct packet_sock *po;
1445 if (!net_eq(dev_net(dev), net) || !num) {
1450 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1451 skb = ip_check_defrag(net, skb, IP_DEFRAG_AF_PACKET);
1456 case PACKET_FANOUT_HASH:
1458 idx = fanout_demux_hash(f, skb, num);
1460 case PACKET_FANOUT_LB:
1461 idx = fanout_demux_lb(f, skb, num);
1463 case PACKET_FANOUT_CPU:
1464 idx = fanout_demux_cpu(f, skb, num);
1466 case PACKET_FANOUT_RND:
1467 idx = fanout_demux_rnd(f, skb, num);
1469 case PACKET_FANOUT_QM:
1470 idx = fanout_demux_qm(f, skb, num);
1472 case PACKET_FANOUT_ROLLOVER:
1473 idx = fanout_demux_rollover(f, skb, 0, false, num);
1475 case PACKET_FANOUT_CBPF:
1476 case PACKET_FANOUT_EBPF:
1477 idx = fanout_demux_bpf(f, skb, num);
1481 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1482 idx = fanout_demux_rollover(f, skb, idx, true, num);
1484 po = pkt_sk(f->arr[idx]);
1485 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1488 DEFINE_MUTEX(fanout_mutex);
1489 EXPORT_SYMBOL_GPL(fanout_mutex);
1490 static LIST_HEAD(fanout_list);
1492 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1494 struct packet_fanout *f = po->fanout;
1496 spin_lock(&f->lock);
1497 f->arr[f->num_members] = sk;
1500 if (f->num_members == 1)
1501 dev_add_pack(&f->prot_hook);
1502 spin_unlock(&f->lock);
1505 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1507 struct packet_fanout *f = po->fanout;
1510 spin_lock(&f->lock);
1511 for (i = 0; i < f->num_members; i++) {
1512 if (f->arr[i] == sk)
1515 BUG_ON(i >= f->num_members);
1516 f->arr[i] = f->arr[f->num_members - 1];
1518 if (f->num_members == 0)
1519 __dev_remove_pack(&f->prot_hook);
1520 spin_unlock(&f->lock);
1523 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1525 if (sk->sk_family != PF_PACKET)
1528 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1531 static void fanout_init_data(struct packet_fanout *f)
1534 case PACKET_FANOUT_LB:
1535 atomic_set(&f->rr_cur, 0);
1537 case PACKET_FANOUT_CBPF:
1538 case PACKET_FANOUT_EBPF:
1539 RCU_INIT_POINTER(f->bpf_prog, NULL);
1544 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1546 struct bpf_prog *old;
1548 spin_lock(&f->lock);
1549 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1550 rcu_assign_pointer(f->bpf_prog, new);
1551 spin_unlock(&f->lock);
1555 bpf_prog_destroy(old);
1559 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1562 struct bpf_prog *new;
1563 struct sock_fprog fprog;
1566 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1568 if (len != sizeof(fprog))
1570 if (copy_from_user(&fprog, data, len))
1573 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1577 __fanout_set_data_bpf(po->fanout, new);
1581 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1584 struct bpf_prog *new;
1587 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1589 if (len != sizeof(fd))
1591 if (copy_from_user(&fd, data, len))
1594 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1596 return PTR_ERR(new);
1598 __fanout_set_data_bpf(po->fanout, new);
1602 static int fanout_set_data(struct packet_sock *po, char __user *data,
1605 switch (po->fanout->type) {
1606 case PACKET_FANOUT_CBPF:
1607 return fanout_set_data_cbpf(po, data, len);
1608 case PACKET_FANOUT_EBPF:
1609 return fanout_set_data_ebpf(po, data, len);
1615 static void fanout_release_data(struct packet_fanout *f)
1618 case PACKET_FANOUT_CBPF:
1619 case PACKET_FANOUT_EBPF:
1620 __fanout_set_data_bpf(f, NULL);
1624 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1626 struct packet_rollover *rollover = NULL;
1627 struct packet_sock *po = pkt_sk(sk);
1628 struct packet_fanout *f, *match;
1629 u8 type = type_flags & 0xff;
1630 u8 flags = type_flags >> 8;
1634 case PACKET_FANOUT_ROLLOVER:
1635 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1637 case PACKET_FANOUT_HASH:
1638 case PACKET_FANOUT_LB:
1639 case PACKET_FANOUT_CPU:
1640 case PACKET_FANOUT_RND:
1641 case PACKET_FANOUT_QM:
1642 case PACKET_FANOUT_CBPF:
1643 case PACKET_FANOUT_EBPF:
1649 mutex_lock(&fanout_mutex);
1659 if (type == PACKET_FANOUT_ROLLOVER ||
1660 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1662 rollover = kzalloc(sizeof(*rollover), GFP_KERNEL);
1665 atomic_long_set(&rollover->num, 0);
1666 atomic_long_set(&rollover->num_huge, 0);
1667 atomic_long_set(&rollover->num_failed, 0);
1668 po->rollover = rollover;
1672 list_for_each_entry(f, &fanout_list, list) {
1674 read_pnet(&f->net) == sock_net(sk)) {
1680 if (match && match->flags != flags)
1684 match = kzalloc(sizeof(*match), GFP_KERNEL);
1687 write_pnet(&match->net, sock_net(sk));
1690 match->flags = flags;
1691 INIT_LIST_HEAD(&match->list);
1692 spin_lock_init(&match->lock);
1693 atomic_set(&match->sk_ref, 0);
1694 fanout_init_data(match);
1695 match->prot_hook.type = po->prot_hook.type;
1696 match->prot_hook.dev = po->prot_hook.dev;
1697 match->prot_hook.func = packet_rcv_fanout;
1698 match->prot_hook.af_packet_priv = match;
1699 match->prot_hook.id_match = match_fanout_group;
1700 list_add(&match->list, &fanout_list);
1703 if (match->type == type &&
1704 match->prot_hook.type == po->prot_hook.type &&
1705 match->prot_hook.dev == po->prot_hook.dev) {
1707 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1708 __dev_remove_pack(&po->prot_hook);
1710 atomic_inc(&match->sk_ref);
1711 __fanout_link(sk, po);
1716 if (err && rollover) {
1718 po->rollover = NULL;
1720 mutex_unlock(&fanout_mutex);
1724 /* If pkt_sk(sk)->fanout->sk_ref is zero, this function removes
1725 * pkt_sk(sk)->fanout from fanout_list and returns pkt_sk(sk)->fanout.
1726 * It is the responsibility of the caller to call fanout_release_data() and
1727 * free the returned packet_fanout (after synchronize_net())
1729 static struct packet_fanout *fanout_release(struct sock *sk)
1731 struct packet_sock *po = pkt_sk(sk);
1732 struct packet_fanout *f;
1734 mutex_lock(&fanout_mutex);
1739 if (atomic_dec_and_test(&f->sk_ref))
1745 kfree_rcu(po->rollover, rcu);
1747 mutex_unlock(&fanout_mutex);
1752 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1753 struct sk_buff *skb)
1755 /* Earlier code assumed this would be a VLAN pkt, double-check
1756 * this now that we have the actual packet in hand. We can only
1757 * do this check on Ethernet devices.
1759 if (unlikely(dev->type != ARPHRD_ETHER))
1762 skb_reset_mac_header(skb);
1763 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1766 static const struct proto_ops packet_ops;
1768 static const struct proto_ops packet_ops_spkt;
1770 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1771 struct packet_type *pt, struct net_device *orig_dev)
1774 struct sockaddr_pkt *spkt;
1777 * When we registered the protocol we saved the socket in the data
1778 * field for just this event.
1781 sk = pt->af_packet_priv;
1784 * Yank back the headers [hope the device set this
1785 * right or kerboom...]
1787 * Incoming packets have ll header pulled,
1790 * For outgoing ones skb->data == skb_mac_header(skb)
1791 * so that this procedure is noop.
1794 if (skb->pkt_type == PACKET_LOOPBACK)
1797 if (!net_eq(dev_net(dev), sock_net(sk)))
1800 skb = skb_share_check(skb, GFP_ATOMIC);
1804 /* drop any routing info */
1807 /* drop conntrack reference */
1810 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1812 skb_push(skb, skb->data - skb_mac_header(skb));
1815 * The SOCK_PACKET socket receives _all_ frames.
1818 spkt->spkt_family = dev->type;
1819 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1820 spkt->spkt_protocol = skb->protocol;
1823 * Charge the memory to the socket. This is done specifically
1824 * to prevent sockets using all the memory up.
1827 if (sock_queue_rcv_skb(sk, skb) == 0)
1838 * Output a raw packet to a device layer. This bypasses all the other
1839 * protocol layers and you must therefore supply it with a complete frame
1842 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1845 struct sock *sk = sock->sk;
1846 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1847 struct sk_buff *skb = NULL;
1848 struct net_device *dev;
1849 struct sockcm_cookie sockc;
1855 * Get and verify the address.
1859 if (msg->msg_namelen < sizeof(struct sockaddr))
1861 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1862 proto = saddr->spkt_protocol;
1864 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1867 * Find the device first to size check it
1870 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1873 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1879 if (!(dev->flags & IFF_UP))
1883 * You may not queue a frame bigger than the mtu. This is the lowest level
1884 * raw protocol and you must do your own fragmentation at this level.
1887 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1888 if (!netif_supports_nofcs(dev)) {
1889 err = -EPROTONOSUPPORT;
1892 extra_len = 4; /* We're doing our own CRC */
1896 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1900 size_t reserved = LL_RESERVED_SPACE(dev);
1901 int tlen = dev->needed_tailroom;
1902 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1905 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1908 /* FIXME: Save some space for broken drivers that write a hard
1909 * header at transmission time by themselves. PPP is the notable
1910 * one here. This should really be fixed at the driver level.
1912 skb_reserve(skb, reserved);
1913 skb_reset_network_header(skb);
1915 /* Try to align data part correctly */
1920 skb_reset_network_header(skb);
1922 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1928 if (!dev_validate_header(dev, skb->data, len)) {
1932 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1933 !packet_extra_vlan_len_allowed(dev, skb)) {
1938 sockc.tsflags = sk->sk_tsflags;
1939 if (msg->msg_controllen) {
1940 err = sock_cmsg_send(sk, msg, &sockc);
1945 skb->protocol = proto;
1947 skb->priority = sk->sk_priority;
1948 skb->mark = sk->sk_mark;
1950 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1952 if (unlikely(extra_len == 4))
1955 skb_probe_transport_header(skb, 0);
1957 dev_queue_xmit(skb);
1968 static unsigned int run_filter(struct sk_buff *skb,
1969 const struct sock *sk,
1972 struct sk_filter *filter;
1975 filter = rcu_dereference(sk->sk_filter);
1977 res = bpf_prog_run_clear_cb(filter->prog, skb);
1983 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
1986 struct virtio_net_hdr vnet_hdr;
1988 if (*len < sizeof(vnet_hdr))
1990 *len -= sizeof(vnet_hdr);
1992 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
1995 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
1999 * This function makes lazy skb cloning in hope that most of packets
2000 * are discarded by BPF.
2002 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
2003 * and skb->cb are mangled. It works because (and until) packets
2004 * falling here are owned by current CPU. Output packets are cloned
2005 * by dev_queue_xmit_nit(), input packets are processed by net_bh
2006 * sequencially, so that if we return skb to original state on exit,
2007 * we will not harm anyone.
2010 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
2011 struct packet_type *pt, struct net_device *orig_dev)
2014 struct sockaddr_ll *sll;
2015 struct packet_sock *po;
2016 u8 *skb_head = skb->data;
2017 int skb_len = skb->len;
2018 unsigned int snaplen, res;
2019 bool is_drop_n_account = false;
2021 if (skb->pkt_type == PACKET_LOOPBACK)
2024 sk = pt->af_packet_priv;
2027 if (!net_eq(dev_net(dev), sock_net(sk)))
2032 if (dev->header_ops) {
2033 /* The device has an explicit notion of ll header,
2034 * exported to higher levels.
2036 * Otherwise, the device hides details of its frame
2037 * structure, so that corresponding packet head is
2038 * never delivered to user.
2040 if (sk->sk_type != SOCK_DGRAM)
2041 skb_push(skb, skb->data - skb_mac_header(skb));
2042 else if (skb->pkt_type == PACKET_OUTGOING) {
2043 /* Special case: outgoing packets have ll header at head */
2044 skb_pull(skb, skb_network_offset(skb));
2050 res = run_filter(skb, sk, snaplen);
2052 goto drop_n_restore;
2056 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2059 if (skb_shared(skb)) {
2060 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2064 if (skb_head != skb->data) {
2065 skb->data = skb_head;
2072 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2074 sll = &PACKET_SKB_CB(skb)->sa.ll;
2075 sll->sll_hatype = dev->type;
2076 sll->sll_pkttype = skb->pkt_type;
2077 if (unlikely(po->origdev))
2078 sll->sll_ifindex = orig_dev->ifindex;
2080 sll->sll_ifindex = dev->ifindex;
2082 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2084 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2085 * Use their space for storing the original skb length.
2087 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2089 if (pskb_trim(skb, snaplen))
2092 skb_set_owner_r(skb, sk);
2096 /* drop conntrack reference */
2099 spin_lock(&sk->sk_receive_queue.lock);
2100 po->stats.stats1.tp_packets++;
2101 sock_skb_set_dropcount(sk, skb);
2102 __skb_queue_tail(&sk->sk_receive_queue, skb);
2103 spin_unlock(&sk->sk_receive_queue.lock);
2104 sk->sk_data_ready(sk);
2108 is_drop_n_account = true;
2109 spin_lock(&sk->sk_receive_queue.lock);
2110 po->stats.stats1.tp_drops++;
2111 atomic_inc(&sk->sk_drops);
2112 spin_unlock(&sk->sk_receive_queue.lock);
2115 if (skb_head != skb->data && skb_shared(skb)) {
2116 skb->data = skb_head;
2120 if (!is_drop_n_account)
2127 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2128 struct packet_type *pt, struct net_device *orig_dev)
2131 struct packet_sock *po;
2132 struct sockaddr_ll *sll;
2133 union tpacket_uhdr h;
2134 u8 *skb_head = skb->data;
2135 int skb_len = skb->len;
2136 unsigned int snaplen, res;
2137 unsigned long status = TP_STATUS_USER;
2138 unsigned short macoff, netoff, hdrlen;
2139 struct sk_buff *copy_skb = NULL;
2142 bool is_drop_n_account = false;
2144 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2145 * We may add members to them until current aligned size without forcing
2146 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2148 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2149 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2151 if (skb->pkt_type == PACKET_LOOPBACK)
2154 sk = pt->af_packet_priv;
2157 if (!net_eq(dev_net(dev), sock_net(sk)))
2160 if (dev->header_ops) {
2161 if (sk->sk_type != SOCK_DGRAM)
2162 skb_push(skb, skb->data - skb_mac_header(skb));
2163 else if (skb->pkt_type == PACKET_OUTGOING) {
2164 /* Special case: outgoing packets have ll header at head */
2165 skb_pull(skb, skb_network_offset(skb));
2171 res = run_filter(skb, sk, snaplen);
2173 goto drop_n_restore;
2175 if (skb->ip_summed == CHECKSUM_PARTIAL)
2176 status |= TP_STATUS_CSUMNOTREADY;
2177 else if (skb->pkt_type != PACKET_OUTGOING &&
2178 (skb->ip_summed == CHECKSUM_COMPLETE ||
2179 skb_csum_unnecessary(skb)))
2180 status |= TP_STATUS_CSUM_VALID;
2185 if (sk->sk_type == SOCK_DGRAM) {
2186 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2189 unsigned int maclen = skb_network_offset(skb);
2190 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2191 (maclen < 16 ? 16 : maclen)) +
2193 if (po->has_vnet_hdr)
2194 netoff += sizeof(struct virtio_net_hdr);
2195 macoff = netoff - maclen;
2197 if (po->tp_version <= TPACKET_V2) {
2198 if (macoff + snaplen > po->rx_ring.frame_size) {
2199 if (po->copy_thresh &&
2200 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2201 if (skb_shared(skb)) {
2202 copy_skb = skb_clone(skb, GFP_ATOMIC);
2204 copy_skb = skb_get(skb);
2205 skb_head = skb->data;
2208 skb_set_owner_r(copy_skb, sk);
2210 snaplen = po->rx_ring.frame_size - macoff;
2211 if ((int)snaplen < 0)
2214 } else if (unlikely(macoff + snaplen >
2215 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2218 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2219 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2220 snaplen, nval, macoff);
2222 if (unlikely((int)snaplen < 0)) {
2224 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2227 spin_lock(&sk->sk_receive_queue.lock);
2228 h.raw = packet_current_rx_frame(po, skb,
2229 TP_STATUS_KERNEL, (macoff+snaplen));
2231 goto drop_n_account;
2232 if (po->tp_version <= TPACKET_V2) {
2233 packet_increment_rx_head(po, &po->rx_ring);
2235 * LOSING will be reported till you read the stats,
2236 * because it's COR - Clear On Read.
2237 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2240 if (po->stats.stats1.tp_drops)
2241 status |= TP_STATUS_LOSING;
2243 po->stats.stats1.tp_packets++;
2245 status |= TP_STATUS_COPY;
2246 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2248 spin_unlock(&sk->sk_receive_queue.lock);
2250 if (po->has_vnet_hdr) {
2251 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2252 sizeof(struct virtio_net_hdr),
2254 spin_lock(&sk->sk_receive_queue.lock);
2255 goto drop_n_account;
2259 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2261 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2262 getnstimeofday(&ts);
2264 status |= ts_status;
2266 switch (po->tp_version) {
2268 h.h1->tp_len = skb->len;
2269 h.h1->tp_snaplen = snaplen;
2270 h.h1->tp_mac = macoff;
2271 h.h1->tp_net = netoff;
2272 h.h1->tp_sec = ts.tv_sec;
2273 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2274 hdrlen = sizeof(*h.h1);
2277 h.h2->tp_len = skb->len;
2278 h.h2->tp_snaplen = snaplen;
2279 h.h2->tp_mac = macoff;
2280 h.h2->tp_net = netoff;
2281 h.h2->tp_sec = ts.tv_sec;
2282 h.h2->tp_nsec = ts.tv_nsec;
2283 if (skb_vlan_tag_present(skb)) {
2284 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2285 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2286 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2288 h.h2->tp_vlan_tci = 0;
2289 h.h2->tp_vlan_tpid = 0;
2291 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2292 hdrlen = sizeof(*h.h2);
2295 /* tp_nxt_offset,vlan are already populated above.
2296 * So DONT clear those fields here
2298 h.h3->tp_status |= status;
2299 h.h3->tp_len = skb->len;
2300 h.h3->tp_snaplen = snaplen;
2301 h.h3->tp_mac = macoff;
2302 h.h3->tp_net = netoff;
2303 h.h3->tp_sec = ts.tv_sec;
2304 h.h3->tp_nsec = ts.tv_nsec;
2305 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2306 hdrlen = sizeof(*h.h3);
2312 sll = h.raw + TPACKET_ALIGN(hdrlen);
2313 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2314 sll->sll_family = AF_PACKET;
2315 sll->sll_hatype = dev->type;
2316 sll->sll_protocol = skb->protocol;
2317 sll->sll_pkttype = skb->pkt_type;
2318 if (unlikely(po->origdev))
2319 sll->sll_ifindex = orig_dev->ifindex;
2321 sll->sll_ifindex = dev->ifindex;
2325 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2326 if (po->tp_version <= TPACKET_V2) {
2329 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2332 for (start = h.raw; start < end; start += PAGE_SIZE)
2333 flush_dcache_page(pgv_to_page(start));
2338 if (po->tp_version <= TPACKET_V2) {
2339 __packet_set_status(po, h.raw, status);
2340 sk->sk_data_ready(sk);
2342 prb_clear_blk_fill_status(&po->rx_ring);
2346 if (skb_head != skb->data && skb_shared(skb)) {
2347 skb->data = skb_head;
2351 if (!is_drop_n_account)
2358 is_drop_n_account = true;
2359 po->stats.stats1.tp_drops++;
2360 spin_unlock(&sk->sk_receive_queue.lock);
2362 sk->sk_data_ready(sk);
2363 kfree_skb(copy_skb);
2364 goto drop_n_restore;
2367 static void tpacket_destruct_skb(struct sk_buff *skb)
2369 struct packet_sock *po = pkt_sk(skb->sk);
2371 if (likely(po->tx_ring.pg_vec)) {
2375 ph = skb_shinfo(skb)->destructor_arg;
2376 packet_dec_pending(&po->tx_ring);
2378 ts = __packet_set_timestamp(po, ph, skb);
2379 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2385 static void tpacket_set_protocol(const struct net_device *dev,
2386 struct sk_buff *skb)
2388 if (dev->type == ARPHRD_ETHER) {
2389 skb_reset_mac_header(skb);
2390 skb->protocol = eth_hdr(skb)->h_proto;
2394 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2396 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2397 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2398 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2399 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2400 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2401 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2402 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2404 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2410 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2411 struct virtio_net_hdr *vnet_hdr)
2413 if (*len < sizeof(*vnet_hdr))
2415 *len -= sizeof(*vnet_hdr);
2417 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2420 return __packet_snd_vnet_parse(vnet_hdr, *len);
2423 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2424 void *frame, struct net_device *dev, void *data, int tp_len,
2425 __be16 proto, unsigned char *addr, int hlen, int copylen,
2426 const struct sockcm_cookie *sockc)
2428 union tpacket_uhdr ph;
2429 int to_write, offset, len, nr_frags, len_max;
2430 struct socket *sock = po->sk.sk_socket;
2436 skb->protocol = proto;
2438 skb->priority = po->sk.sk_priority;
2439 skb->mark = po->sk.sk_mark;
2440 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2441 skb_shinfo(skb)->destructor_arg = ph.raw;
2443 skb_reserve(skb, hlen);
2444 skb_reset_network_header(skb);
2448 if (sock->type == SOCK_DGRAM) {
2449 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2451 if (unlikely(err < 0))
2453 } else if (copylen) {
2454 int hdrlen = min_t(int, copylen, tp_len);
2456 skb_push(skb, dev->hard_header_len);
2457 skb_put(skb, copylen - dev->hard_header_len);
2458 err = skb_store_bits(skb, 0, data, hdrlen);
2461 if (!dev_validate_header(dev, skb->data, hdrlen))
2464 tpacket_set_protocol(dev, skb);
2470 offset = offset_in_page(data);
2471 len_max = PAGE_SIZE - offset;
2472 len = ((to_write > len_max) ? len_max : to_write);
2474 skb->data_len = to_write;
2475 skb->len += to_write;
2476 skb->truesize += to_write;
2477 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2479 while (likely(to_write)) {
2480 nr_frags = skb_shinfo(skb)->nr_frags;
2482 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2483 pr_err("Packet exceed the number of skb frags(%lu)\n",
2488 page = pgv_to_page(data);
2490 flush_dcache_page(page);
2492 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2495 len_max = PAGE_SIZE;
2496 len = ((to_write > len_max) ? len_max : to_write);
2499 skb_probe_transport_header(skb, 0);
2504 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2505 int size_max, void **data)
2507 union tpacket_uhdr ph;
2512 switch (po->tp_version) {
2514 tp_len = ph.h2->tp_len;
2517 tp_len = ph.h1->tp_len;
2520 if (unlikely(tp_len > size_max)) {
2521 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2525 if (unlikely(po->tp_tx_has_off)) {
2526 int off_min, off_max;
2528 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2529 off_max = po->tx_ring.frame_size - tp_len;
2530 if (po->sk.sk_type == SOCK_DGRAM) {
2531 switch (po->tp_version) {
2533 off = ph.h2->tp_net;
2536 off = ph.h1->tp_net;
2540 switch (po->tp_version) {
2542 off = ph.h2->tp_mac;
2545 off = ph.h1->tp_mac;
2549 if (unlikely((off < off_min) || (off_max < off)))
2552 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2555 *data = frame + off;
2559 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2561 struct sk_buff *skb;
2562 struct net_device *dev;
2563 struct virtio_net_hdr *vnet_hdr = NULL;
2564 struct sockcm_cookie sockc;
2566 int err, reserve = 0;
2568 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2569 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2570 int tp_len, size_max;
2571 unsigned char *addr;
2574 int status = TP_STATUS_AVAILABLE;
2575 int hlen, tlen, copylen = 0;
2577 mutex_lock(&po->pg_vec_lock);
2579 if (likely(saddr == NULL)) {
2580 dev = packet_cached_dev_get(po);
2585 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2587 if (msg->msg_namelen < (saddr->sll_halen
2588 + offsetof(struct sockaddr_ll,
2591 proto = saddr->sll_protocol;
2592 addr = saddr->sll_addr;
2593 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2596 sockc.tsflags = po->sk.sk_tsflags;
2597 if (msg->msg_controllen) {
2598 err = sock_cmsg_send(&po->sk, msg, &sockc);
2604 if (unlikely(dev == NULL))
2607 if (unlikely(!(dev->flags & IFF_UP)))
2610 if (po->sk.sk_socket->type == SOCK_RAW)
2611 reserve = dev->hard_header_len;
2612 size_max = po->tx_ring.frame_size
2613 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2615 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2616 size_max = dev->mtu + reserve + VLAN_HLEN;
2619 ph = packet_current_frame(po, &po->tx_ring,
2620 TP_STATUS_SEND_REQUEST);
2621 if (unlikely(ph == NULL)) {
2622 if (need_wait && need_resched())
2628 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2632 status = TP_STATUS_SEND_REQUEST;
2633 hlen = LL_RESERVED_SPACE(dev);
2634 tlen = dev->needed_tailroom;
2635 if (po->has_vnet_hdr) {
2637 data += sizeof(*vnet_hdr);
2638 tp_len -= sizeof(*vnet_hdr);
2640 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2644 copylen = __virtio16_to_cpu(vio_le(),
2647 copylen = max_t(int, copylen, dev->hard_header_len);
2648 skb = sock_alloc_send_skb(&po->sk,
2649 hlen + tlen + sizeof(struct sockaddr_ll) +
2650 (copylen - dev->hard_header_len),
2653 if (unlikely(skb == NULL)) {
2654 /* we assume the socket was initially writeable ... */
2655 if (likely(len_sum > 0))
2659 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2660 addr, hlen, copylen, &sockc);
2661 if (likely(tp_len >= 0) &&
2662 tp_len > dev->mtu + reserve &&
2663 !po->has_vnet_hdr &&
2664 !packet_extra_vlan_len_allowed(dev, skb))
2667 if (unlikely(tp_len < 0)) {
2670 __packet_set_status(po, ph,
2671 TP_STATUS_AVAILABLE);
2672 packet_increment_head(&po->tx_ring);
2676 status = TP_STATUS_WRONG_FORMAT;
2682 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2688 packet_pick_tx_queue(dev, skb);
2690 skb->destructor = tpacket_destruct_skb;
2691 __packet_set_status(po, ph, TP_STATUS_SENDING);
2692 packet_inc_pending(&po->tx_ring);
2694 status = TP_STATUS_SEND_REQUEST;
2695 err = po->xmit(skb);
2696 if (unlikely(err > 0)) {
2697 err = net_xmit_errno(err);
2698 if (err && __packet_get_status(po, ph) ==
2699 TP_STATUS_AVAILABLE) {
2700 /* skb was destructed already */
2705 * skb was dropped but not destructed yet;
2706 * let's treat it like congestion or err < 0
2710 packet_increment_head(&po->tx_ring);
2712 } while (likely((ph != NULL) ||
2713 /* Note: packet_read_pending() might be slow if we have
2714 * to call it as it's per_cpu variable, but in fast-path
2715 * we already short-circuit the loop with the first
2716 * condition, and luckily don't have to go that path
2719 (need_wait && packet_read_pending(&po->tx_ring))));
2725 __packet_set_status(po, ph, status);
2730 mutex_unlock(&po->pg_vec_lock);
2734 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2735 size_t reserve, size_t len,
2736 size_t linear, int noblock,
2739 struct sk_buff *skb;
2741 /* Under a page? Don't bother with paged skb. */
2742 if (prepad + len < PAGE_SIZE || !linear)
2745 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2750 skb_reserve(skb, reserve);
2751 skb_put(skb, linear);
2752 skb->data_len = len - linear;
2753 skb->len += len - linear;
2758 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2760 struct sock *sk = sock->sk;
2761 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2762 struct sk_buff *skb;
2763 struct net_device *dev;
2765 unsigned char *addr;
2766 int err, reserve = 0;
2767 struct sockcm_cookie sockc;
2768 struct virtio_net_hdr vnet_hdr = { 0 };
2770 struct packet_sock *po = pkt_sk(sk);
2771 int hlen, tlen, linear;
2775 * Get and verify the address.
2778 if (likely(saddr == NULL)) {
2779 dev = packet_cached_dev_get(po);
2784 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2786 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2788 proto = saddr->sll_protocol;
2789 addr = saddr->sll_addr;
2790 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2794 if (unlikely(dev == NULL))
2797 if (unlikely(!(dev->flags & IFF_UP)))
2800 sockc.tsflags = sk->sk_tsflags;
2801 sockc.mark = sk->sk_mark;
2802 if (msg->msg_controllen) {
2803 err = sock_cmsg_send(sk, msg, &sockc);
2808 if (sock->type == SOCK_RAW)
2809 reserve = dev->hard_header_len;
2810 if (po->has_vnet_hdr) {
2811 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2816 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2817 if (!netif_supports_nofcs(dev)) {
2818 err = -EPROTONOSUPPORT;
2821 extra_len = 4; /* We're doing our own CRC */
2825 if (!vnet_hdr.gso_type &&
2826 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2830 hlen = LL_RESERVED_SPACE(dev);
2831 tlen = dev->needed_tailroom;
2832 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2833 linear = max(linear, min_t(int, len, dev->hard_header_len));
2834 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2835 msg->msg_flags & MSG_DONTWAIT, &err);
2839 skb_set_network_header(skb, reserve);
2842 if (sock->type == SOCK_DGRAM) {
2843 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2844 if (unlikely(offset < 0))
2848 /* Returns -EFAULT on error */
2849 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2853 if (sock->type == SOCK_RAW &&
2854 !dev_validate_header(dev, skb->data, len)) {
2859 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2861 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2862 !packet_extra_vlan_len_allowed(dev, skb)) {
2867 skb->protocol = proto;
2869 skb->priority = sk->sk_priority;
2870 skb->mark = sockc.mark;
2872 packet_pick_tx_queue(dev, skb);
2874 if (po->has_vnet_hdr) {
2875 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2878 len += sizeof(vnet_hdr);
2881 skb_probe_transport_header(skb, reserve);
2883 if (unlikely(extra_len == 4))
2886 err = po->xmit(skb);
2887 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2903 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2905 struct sock *sk = sock->sk;
2906 struct packet_sock *po = pkt_sk(sk);
2908 if (po->tx_ring.pg_vec)
2909 return tpacket_snd(po, msg);
2911 return packet_snd(sock, msg, len);
2915 * Close a PACKET socket. This is fairly simple. We immediately go
2916 * to 'closed' state and remove our protocol entry in the device list.
2919 static int packet_release(struct socket *sock)
2921 struct sock *sk = sock->sk;
2922 struct packet_sock *po;
2923 struct packet_fanout *f;
2925 union tpacket_req_u req_u;
2933 mutex_lock(&net->packet.sklist_lock);
2934 sk_del_node_init_rcu(sk);
2935 mutex_unlock(&net->packet.sklist_lock);
2938 sock_prot_inuse_add(net, sk->sk_prot, -1);
2941 spin_lock(&po->bind_lock);
2942 unregister_prot_hook(sk, false);
2943 packet_cached_dev_reset(po);
2945 if (po->prot_hook.dev) {
2946 dev_put(po->prot_hook.dev);
2947 po->prot_hook.dev = NULL;
2949 spin_unlock(&po->bind_lock);
2951 packet_flush_mclist(sk);
2953 if (po->rx_ring.pg_vec) {
2954 memset(&req_u, 0, sizeof(req_u));
2955 packet_set_ring(sk, &req_u, 1, 0);
2958 if (po->tx_ring.pg_vec) {
2959 memset(&req_u, 0, sizeof(req_u));
2960 packet_set_ring(sk, &req_u, 1, 1);
2963 f = fanout_release(sk);
2968 fanout_release_data(f);
2972 * Now the socket is dead. No more input will appear.
2979 skb_queue_purge(&sk->sk_receive_queue);
2980 packet_free_pending(po);
2981 sk_refcnt_debug_release(sk);
2988 * Attach a packet hook.
2991 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2994 struct packet_sock *po = pkt_sk(sk);
2995 struct net_device *dev_curr;
2998 struct net_device *dev = NULL;
3000 bool unlisted = false;
3006 spin_lock(&po->bind_lock);
3010 dev = dev_get_by_name_rcu(sock_net(sk), name);
3015 } else if (ifindex) {
3016 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3026 proto_curr = po->prot_hook.type;
3027 dev_curr = po->prot_hook.dev;
3029 need_rehook = proto_curr != proto || dev_curr != dev;
3034 __unregister_prot_hook(sk, true);
3036 dev_curr = po->prot_hook.dev;
3038 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3043 po->prot_hook.type = proto;
3045 if (unlikely(unlisted)) {
3047 po->prot_hook.dev = NULL;
3049 packet_cached_dev_reset(po);
3051 po->prot_hook.dev = dev;
3052 po->ifindex = dev ? dev->ifindex : 0;
3053 packet_cached_dev_assign(po, dev);
3059 if (proto == 0 || !need_rehook)
3062 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3063 register_prot_hook(sk);
3065 sk->sk_err = ENETDOWN;
3066 if (!sock_flag(sk, SOCK_DEAD))
3067 sk->sk_error_report(sk);
3072 spin_unlock(&po->bind_lock);
3078 * Bind a packet socket to a device
3081 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3084 struct sock *sk = sock->sk;
3091 if (addr_len != sizeof(struct sockaddr))
3093 strlcpy(name, uaddr->sa_data, sizeof(name));
3095 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3098 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3100 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3101 struct sock *sk = sock->sk;
3107 if (addr_len < sizeof(struct sockaddr_ll))
3109 if (sll->sll_family != AF_PACKET)
3112 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3113 sll->sll_protocol ? : pkt_sk(sk)->num);
3116 static struct proto packet_proto = {
3118 .owner = THIS_MODULE,
3119 .obj_size = sizeof(struct packet_sock),
3123 * Create a packet of type SOCK_PACKET.
3126 static int packet_create(struct net *net, struct socket *sock, int protocol,
3130 struct packet_sock *po;
3131 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3134 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3136 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3137 sock->type != SOCK_PACKET)
3138 return -ESOCKTNOSUPPORT;
3140 sock->state = SS_UNCONNECTED;
3143 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3147 sock->ops = &packet_ops;
3148 if (sock->type == SOCK_PACKET)
3149 sock->ops = &packet_ops_spkt;
3151 sock_init_data(sock, sk);
3154 sk->sk_family = PF_PACKET;
3156 po->xmit = dev_queue_xmit;
3158 err = packet_alloc_pending(po);
3162 packet_cached_dev_reset(po);
3164 sk->sk_destruct = packet_sock_destruct;
3165 sk_refcnt_debug_inc(sk);
3168 * Attach a protocol block
3171 spin_lock_init(&po->bind_lock);
3172 mutex_init(&po->pg_vec_lock);
3173 po->rollover = NULL;
3174 po->prot_hook.func = packet_rcv;
3176 if (sock->type == SOCK_PACKET)
3177 po->prot_hook.func = packet_rcv_spkt;
3179 po->prot_hook.af_packet_priv = sk;
3182 po->prot_hook.type = proto;
3183 register_prot_hook(sk);
3186 mutex_lock(&net->packet.sklist_lock);
3187 sk_add_node_rcu(sk, &net->packet.sklist);
3188 mutex_unlock(&net->packet.sklist_lock);
3191 sock_prot_inuse_add(net, &packet_proto, 1);
3202 * Pull a packet from our receive queue and hand it to the user.
3203 * If necessary we block.
3206 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3209 struct sock *sk = sock->sk;
3210 struct sk_buff *skb;
3212 int vnet_hdr_len = 0;
3213 unsigned int origlen = 0;
3216 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3220 /* What error should we return now? EUNATTACH? */
3221 if (pkt_sk(sk)->ifindex < 0)
3225 if (flags & MSG_ERRQUEUE) {
3226 err = sock_recv_errqueue(sk, msg, len,
3227 SOL_PACKET, PACKET_TX_TIMESTAMP);
3232 * Call the generic datagram receiver. This handles all sorts
3233 * of horrible races and re-entrancy so we can forget about it
3234 * in the protocol layers.
3236 * Now it will return ENETDOWN, if device have just gone down,
3237 * but then it will block.
3240 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3243 * An error occurred so return it. Because skb_recv_datagram()
3244 * handles the blocking we don't see and worry about blocking
3251 if (pkt_sk(sk)->pressure)
3252 packet_rcv_has_room(pkt_sk(sk), NULL);
3254 if (pkt_sk(sk)->has_vnet_hdr) {
3255 err = packet_rcv_vnet(msg, skb, &len);
3258 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3261 /* You lose any data beyond the buffer you gave. If it worries
3262 * a user program they can ask the device for its MTU
3268 msg->msg_flags |= MSG_TRUNC;
3271 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3275 if (sock->type != SOCK_PACKET) {
3276 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3278 /* Original length was stored in sockaddr_ll fields */
3279 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3280 sll->sll_family = AF_PACKET;
3281 sll->sll_protocol = skb->protocol;
3284 sock_recv_ts_and_drops(msg, sk, skb);
3286 if (msg->msg_name) {
3287 /* If the address length field is there to be filled
3288 * in, we fill it in now.
3290 if (sock->type == SOCK_PACKET) {
3291 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3292 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3294 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3296 msg->msg_namelen = sll->sll_halen +
3297 offsetof(struct sockaddr_ll, sll_addr);
3299 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3303 if (pkt_sk(sk)->auxdata) {
3304 struct tpacket_auxdata aux;
3306 aux.tp_status = TP_STATUS_USER;
3307 if (skb->ip_summed == CHECKSUM_PARTIAL)
3308 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3309 else if (skb->pkt_type != PACKET_OUTGOING &&
3310 (skb->ip_summed == CHECKSUM_COMPLETE ||
3311 skb_csum_unnecessary(skb)))
3312 aux.tp_status |= TP_STATUS_CSUM_VALID;
3314 aux.tp_len = origlen;
3315 aux.tp_snaplen = skb->len;
3317 aux.tp_net = skb_network_offset(skb);
3318 if (skb_vlan_tag_present(skb)) {
3319 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3320 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3321 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3323 aux.tp_vlan_tci = 0;
3324 aux.tp_vlan_tpid = 0;
3326 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3330 * Free or return the buffer as appropriate. Again this
3331 * hides all the races and re-entrancy issues from us.
3333 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3336 skb_free_datagram(sk, skb);
3341 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3342 int *uaddr_len, int peer)
3344 struct net_device *dev;
3345 struct sock *sk = sock->sk;
3350 uaddr->sa_family = AF_PACKET;
3351 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3353 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3355 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3357 *uaddr_len = sizeof(*uaddr);
3362 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3363 int *uaddr_len, int peer)
3365 struct net_device *dev;
3366 struct sock *sk = sock->sk;
3367 struct packet_sock *po = pkt_sk(sk);
3368 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3373 sll->sll_family = AF_PACKET;
3374 sll->sll_ifindex = po->ifindex;
3375 sll->sll_protocol = po->num;
3376 sll->sll_pkttype = 0;
3378 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3380 sll->sll_hatype = dev->type;
3381 sll->sll_halen = dev->addr_len;
3382 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3384 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3388 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3393 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3397 case PACKET_MR_MULTICAST:
3398 if (i->alen != dev->addr_len)
3401 return dev_mc_add(dev, i->addr);
3403 return dev_mc_del(dev, i->addr);
3405 case PACKET_MR_PROMISC:
3406 return dev_set_promiscuity(dev, what);
3407 case PACKET_MR_ALLMULTI:
3408 return dev_set_allmulti(dev, what);
3409 case PACKET_MR_UNICAST:
3410 if (i->alen != dev->addr_len)
3413 return dev_uc_add(dev, i->addr);
3415 return dev_uc_del(dev, i->addr);
3423 static void packet_dev_mclist_delete(struct net_device *dev,
3424 struct packet_mclist **mlp)
3426 struct packet_mclist *ml;
3428 while ((ml = *mlp) != NULL) {
3429 if (ml->ifindex == dev->ifindex) {
3430 packet_dev_mc(dev, ml, -1);
3438 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3440 struct packet_sock *po = pkt_sk(sk);
3441 struct packet_mclist *ml, *i;
3442 struct net_device *dev;
3448 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3453 if (mreq->mr_alen > dev->addr_len)
3457 i = kmalloc(sizeof(*i), GFP_KERNEL);
3462 for (ml = po->mclist; ml; ml = ml->next) {
3463 if (ml->ifindex == mreq->mr_ifindex &&
3464 ml->type == mreq->mr_type &&
3465 ml->alen == mreq->mr_alen &&
3466 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3468 /* Free the new element ... */
3474 i->type = mreq->mr_type;
3475 i->ifindex = mreq->mr_ifindex;
3476 i->alen = mreq->mr_alen;
3477 memcpy(i->addr, mreq->mr_address, i->alen);
3478 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3480 i->next = po->mclist;
3482 err = packet_dev_mc(dev, i, 1);
3484 po->mclist = i->next;
3493 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3495 struct packet_mclist *ml, **mlp;
3499 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3500 if (ml->ifindex == mreq->mr_ifindex &&
3501 ml->type == mreq->mr_type &&
3502 ml->alen == mreq->mr_alen &&
3503 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3504 if (--ml->count == 0) {
3505 struct net_device *dev;
3507 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3509 packet_dev_mc(dev, ml, -1);
3519 static void packet_flush_mclist(struct sock *sk)
3521 struct packet_sock *po = pkt_sk(sk);
3522 struct packet_mclist *ml;
3528 while ((ml = po->mclist) != NULL) {
3529 struct net_device *dev;
3531 po->mclist = ml->next;
3532 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3534 packet_dev_mc(dev, ml, -1);
3541 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3543 struct sock *sk = sock->sk;
3544 struct packet_sock *po = pkt_sk(sk);
3547 if (level != SOL_PACKET)
3548 return -ENOPROTOOPT;
3551 case PACKET_ADD_MEMBERSHIP:
3552 case PACKET_DROP_MEMBERSHIP:
3554 struct packet_mreq_max mreq;
3556 memset(&mreq, 0, sizeof(mreq));
3557 if (len < sizeof(struct packet_mreq))
3559 if (len > sizeof(mreq))
3561 if (copy_from_user(&mreq, optval, len))
3563 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3565 if (optname == PACKET_ADD_MEMBERSHIP)
3566 ret = packet_mc_add(sk, &mreq);
3568 ret = packet_mc_drop(sk, &mreq);
3572 case PACKET_RX_RING:
3573 case PACKET_TX_RING:
3575 union tpacket_req_u req_u;
3578 switch (po->tp_version) {
3581 len = sizeof(req_u.req);
3585 len = sizeof(req_u.req3);
3590 if (copy_from_user(&req_u.req, optval, len))
3592 return packet_set_ring(sk, &req_u, 0,
3593 optname == PACKET_TX_RING);
3595 case PACKET_COPY_THRESH:
3599 if (optlen != sizeof(val))
3601 if (copy_from_user(&val, optval, sizeof(val)))
3604 pkt_sk(sk)->copy_thresh = val;
3607 case PACKET_VERSION:
3611 if (optlen != sizeof(val))
3613 if (copy_from_user(&val, optval, sizeof(val)))
3624 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3627 po->tp_version = val;
3633 case PACKET_RESERVE:
3637 if (optlen != sizeof(val))
3639 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3641 if (copy_from_user(&val, optval, sizeof(val)))
3643 po->tp_reserve = val;
3650 if (optlen != sizeof(val))
3652 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3654 if (copy_from_user(&val, optval, sizeof(val)))
3656 po->tp_loss = !!val;
3659 case PACKET_AUXDATA:
3663 if (optlen < sizeof(val))
3665 if (copy_from_user(&val, optval, sizeof(val)))
3668 po->auxdata = !!val;
3671 case PACKET_ORIGDEV:
3675 if (optlen < sizeof(val))
3677 if (copy_from_user(&val, optval, sizeof(val)))
3680 po->origdev = !!val;
3683 case PACKET_VNET_HDR:
3687 if (sock->type != SOCK_RAW)
3689 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3691 if (optlen < sizeof(val))
3693 if (copy_from_user(&val, optval, sizeof(val)))
3696 po->has_vnet_hdr = !!val;
3699 case PACKET_TIMESTAMP:
3703 if (optlen != sizeof(val))
3705 if (copy_from_user(&val, optval, sizeof(val)))
3708 po->tp_tstamp = val;
3715 if (optlen != sizeof(val))
3717 if (copy_from_user(&val, optval, sizeof(val)))
3720 return fanout_add(sk, val & 0xffff, val >> 16);
3722 case PACKET_FANOUT_DATA:
3727 return fanout_set_data(po, optval, optlen);
3729 case PACKET_TX_HAS_OFF:
3733 if (optlen != sizeof(val))
3735 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3737 if (copy_from_user(&val, optval, sizeof(val)))
3739 po->tp_tx_has_off = !!val;
3742 case PACKET_QDISC_BYPASS:
3746 if (optlen != sizeof(val))
3748 if (copy_from_user(&val, optval, sizeof(val)))
3751 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3755 return -ENOPROTOOPT;
3759 static int packet_getsockopt(struct socket *sock, int level, int optname,
3760 char __user *optval, int __user *optlen)
3763 int val, lv = sizeof(val);
3764 struct sock *sk = sock->sk;
3765 struct packet_sock *po = pkt_sk(sk);
3767 union tpacket_stats_u st;
3768 struct tpacket_rollover_stats rstats;
3770 if (level != SOL_PACKET)
3771 return -ENOPROTOOPT;
3773 if (get_user(len, optlen))
3780 case PACKET_STATISTICS:
3781 spin_lock_bh(&sk->sk_receive_queue.lock);
3782 memcpy(&st, &po->stats, sizeof(st));
3783 memset(&po->stats, 0, sizeof(po->stats));
3784 spin_unlock_bh(&sk->sk_receive_queue.lock);
3786 if (po->tp_version == TPACKET_V3) {
3787 lv = sizeof(struct tpacket_stats_v3);
3788 st.stats3.tp_packets += st.stats3.tp_drops;
3791 lv = sizeof(struct tpacket_stats);
3792 st.stats1.tp_packets += st.stats1.tp_drops;
3797 case PACKET_AUXDATA:
3800 case PACKET_ORIGDEV:
3803 case PACKET_VNET_HDR:
3804 val = po->has_vnet_hdr;
3806 case PACKET_VERSION:
3807 val = po->tp_version;
3810 if (len > sizeof(int))
3812 if (copy_from_user(&val, optval, len))
3816 val = sizeof(struct tpacket_hdr);
3819 val = sizeof(struct tpacket2_hdr);
3822 val = sizeof(struct tpacket3_hdr);
3828 case PACKET_RESERVE:
3829 val = po->tp_reserve;
3834 case PACKET_TIMESTAMP:
3835 val = po->tp_tstamp;
3839 ((u32)po->fanout->id |
3840 ((u32)po->fanout->type << 16) |
3841 ((u32)po->fanout->flags << 24)) :
3844 case PACKET_ROLLOVER_STATS:
3847 rstats.tp_all = atomic_long_read(&po->rollover->num);
3848 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3849 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3851 lv = sizeof(rstats);
3853 case PACKET_TX_HAS_OFF:
3854 val = po->tp_tx_has_off;
3856 case PACKET_QDISC_BYPASS:
3857 val = packet_use_direct_xmit(po);
3860 return -ENOPROTOOPT;
3865 if (put_user(len, optlen))
3867 if (copy_to_user(optval, data, len))
3873 #ifdef CONFIG_COMPAT
3874 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3875 char __user *optval, unsigned int optlen)
3877 struct packet_sock *po = pkt_sk(sock->sk);
3879 if (level != SOL_PACKET)
3880 return -ENOPROTOOPT;
3882 if (optname == PACKET_FANOUT_DATA &&
3883 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3884 optval = (char __user *)get_compat_bpf_fprog(optval);
3887 optlen = sizeof(struct sock_fprog);
3890 return packet_setsockopt(sock, level, optname, optval, optlen);
3894 static int packet_notifier(struct notifier_block *this,
3895 unsigned long msg, void *ptr)
3898 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3899 struct net *net = dev_net(dev);
3902 sk_for_each_rcu(sk, &net->packet.sklist) {
3903 struct packet_sock *po = pkt_sk(sk);
3906 case NETDEV_UNREGISTER:
3908 packet_dev_mclist_delete(dev, &po->mclist);
3912 if (dev->ifindex == po->ifindex) {
3913 spin_lock(&po->bind_lock);
3915 __unregister_prot_hook(sk, false);
3916 sk->sk_err = ENETDOWN;
3917 if (!sock_flag(sk, SOCK_DEAD))
3918 sk->sk_error_report(sk);
3920 if (msg == NETDEV_UNREGISTER) {
3921 packet_cached_dev_reset(po);
3923 if (po->prot_hook.dev)
3924 dev_put(po->prot_hook.dev);
3925 po->prot_hook.dev = NULL;
3927 spin_unlock(&po->bind_lock);
3931 if (dev->ifindex == po->ifindex) {
3932 spin_lock(&po->bind_lock);
3934 register_prot_hook(sk);
3935 spin_unlock(&po->bind_lock);
3945 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3948 struct sock *sk = sock->sk;
3953 int amount = sk_wmem_alloc_get(sk);
3955 return put_user(amount, (int __user *)arg);
3959 struct sk_buff *skb;
3962 spin_lock_bh(&sk->sk_receive_queue.lock);
3963 skb = skb_peek(&sk->sk_receive_queue);
3966 spin_unlock_bh(&sk->sk_receive_queue.lock);
3967 return put_user(amount, (int __user *)arg);
3970 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3972 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3982 case SIOCGIFBRDADDR:
3983 case SIOCSIFBRDADDR:
3984 case SIOCGIFNETMASK:
3985 case SIOCSIFNETMASK:
3986 case SIOCGIFDSTADDR:
3987 case SIOCSIFDSTADDR:
3989 return inet_dgram_ops.ioctl(sock, cmd, arg);
3993 return -ENOIOCTLCMD;
3998 static unsigned int packet_poll(struct file *file, struct socket *sock,
4001 struct sock *sk = sock->sk;
4002 struct packet_sock *po = pkt_sk(sk);
4003 unsigned int mask = datagram_poll(file, sock, wait);
4005 spin_lock_bh(&sk->sk_receive_queue.lock);
4006 if (po->rx_ring.pg_vec) {
4007 if (!packet_previous_rx_frame(po, &po->rx_ring,
4009 mask |= POLLIN | POLLRDNORM;
4011 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
4013 spin_unlock_bh(&sk->sk_receive_queue.lock);
4014 spin_lock_bh(&sk->sk_write_queue.lock);
4015 if (po->tx_ring.pg_vec) {
4016 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
4017 mask |= POLLOUT | POLLWRNORM;
4019 spin_unlock_bh(&sk->sk_write_queue.lock);
4024 /* Dirty? Well, I still did not learn better way to account
4028 static void packet_mm_open(struct vm_area_struct *vma)
4030 struct file *file = vma->vm_file;
4031 struct socket *sock = file->private_data;
4032 struct sock *sk = sock->sk;
4035 atomic_inc(&pkt_sk(sk)->mapped);
4038 static void packet_mm_close(struct vm_area_struct *vma)
4040 struct file *file = vma->vm_file;
4041 struct socket *sock = file->private_data;
4042 struct sock *sk = sock->sk;
4045 atomic_dec(&pkt_sk(sk)->mapped);
4048 static const struct vm_operations_struct packet_mmap_ops = {
4049 .open = packet_mm_open,
4050 .close = packet_mm_close,
4053 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4058 for (i = 0; i < len; i++) {
4059 if (likely(pg_vec[i].buffer)) {
4060 if (is_vmalloc_addr(pg_vec[i].buffer))
4061 vfree(pg_vec[i].buffer);
4063 free_pages((unsigned long)pg_vec[i].buffer,
4065 pg_vec[i].buffer = NULL;
4071 static char *alloc_one_pg_vec_page(unsigned long order)
4074 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4075 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4077 buffer = (char *) __get_free_pages(gfp_flags, order);
4081 /* __get_free_pages failed, fall back to vmalloc */
4082 buffer = vzalloc((1 << order) * PAGE_SIZE);
4086 /* vmalloc failed, lets dig into swap here */
4087 gfp_flags &= ~__GFP_NORETRY;
4088 buffer = (char *) __get_free_pages(gfp_flags, order);
4092 /* complete and utter failure */
4096 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4098 unsigned int block_nr = req->tp_block_nr;
4102 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4103 if (unlikely(!pg_vec))
4106 for (i = 0; i < block_nr; i++) {
4107 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4108 if (unlikely(!pg_vec[i].buffer))
4109 goto out_free_pgvec;
4116 free_pg_vec(pg_vec, order, block_nr);
4121 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4122 int closing, int tx_ring)
4124 struct pgv *pg_vec = NULL;
4125 struct packet_sock *po = pkt_sk(sk);
4126 int was_running, order = 0;
4127 struct packet_ring_buffer *rb;
4128 struct sk_buff_head *rb_queue;
4131 /* Added to avoid minimal code churn */
4132 struct tpacket_req *req = &req_u->req;
4135 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4136 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4137 net_warn_ratelimited("Tx-ring is not supported.\n");
4141 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4142 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4146 if (atomic_read(&po->mapped))
4148 if (packet_read_pending(rb))
4152 if (req->tp_block_nr) {
4153 /* Sanity tests and some calculations */
4155 if (unlikely(rb->pg_vec))
4158 switch (po->tp_version) {
4160 po->tp_hdrlen = TPACKET_HDRLEN;
4163 po->tp_hdrlen = TPACKET2_HDRLEN;
4166 po->tp_hdrlen = TPACKET3_HDRLEN;
4171 if (unlikely((int)req->tp_block_size <= 0))
4173 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4175 if (po->tp_version >= TPACKET_V3 &&
4176 (int)(req->tp_block_size -
4177 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4179 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4182 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4185 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4186 if (unlikely(rb->frames_per_block == 0))
4188 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4193 order = get_order(req->tp_block_size);
4194 pg_vec = alloc_pg_vec(req, order);
4195 if (unlikely(!pg_vec))
4197 switch (po->tp_version) {
4199 /* Transmit path is not supported. We checked
4200 * it above but just being paranoid
4203 init_prb_bdqc(po, rb, pg_vec, req_u);
4212 if (unlikely(req->tp_frame_nr))
4217 /* Detach socket from network */
4218 spin_lock(&po->bind_lock);
4219 was_running = po->running;
4223 __unregister_prot_hook(sk, false);
4225 spin_unlock(&po->bind_lock);
4230 mutex_lock(&po->pg_vec_lock);
4231 if (closing || atomic_read(&po->mapped) == 0) {
4233 spin_lock_bh(&rb_queue->lock);
4234 swap(rb->pg_vec, pg_vec);
4235 rb->frame_max = (req->tp_frame_nr - 1);
4237 rb->frame_size = req->tp_frame_size;
4238 spin_unlock_bh(&rb_queue->lock);
4240 swap(rb->pg_vec_order, order);
4241 swap(rb->pg_vec_len, req->tp_block_nr);
4243 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4244 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4245 tpacket_rcv : packet_rcv;
4246 skb_queue_purge(rb_queue);
4247 if (atomic_read(&po->mapped))
4248 pr_err("packet_mmap: vma is busy: %d\n",
4249 atomic_read(&po->mapped));
4251 mutex_unlock(&po->pg_vec_lock);
4253 spin_lock(&po->bind_lock);
4256 register_prot_hook(sk);
4258 spin_unlock(&po->bind_lock);
4259 if (closing && (po->tp_version > TPACKET_V2)) {
4260 /* Because we don't support block-based V3 on tx-ring */
4262 prb_shutdown_retire_blk_timer(po, rb_queue);
4266 free_pg_vec(pg_vec, order, req->tp_block_nr);
4272 static int packet_mmap(struct file *file, struct socket *sock,
4273 struct vm_area_struct *vma)
4275 struct sock *sk = sock->sk;
4276 struct packet_sock *po = pkt_sk(sk);
4277 unsigned long size, expected_size;
4278 struct packet_ring_buffer *rb;
4279 unsigned long start;
4286 mutex_lock(&po->pg_vec_lock);
4289 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4291 expected_size += rb->pg_vec_len
4297 if (expected_size == 0)
4300 size = vma->vm_end - vma->vm_start;
4301 if (size != expected_size)
4304 start = vma->vm_start;
4305 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4306 if (rb->pg_vec == NULL)
4309 for (i = 0; i < rb->pg_vec_len; i++) {
4311 void *kaddr = rb->pg_vec[i].buffer;
4314 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4315 page = pgv_to_page(kaddr);
4316 err = vm_insert_page(vma, start, page);
4325 atomic_inc(&po->mapped);
4326 vma->vm_ops = &packet_mmap_ops;
4330 mutex_unlock(&po->pg_vec_lock);
4334 static const struct proto_ops packet_ops_spkt = {
4335 .family = PF_PACKET,
4336 .owner = THIS_MODULE,
4337 .release = packet_release,
4338 .bind = packet_bind_spkt,
4339 .connect = sock_no_connect,
4340 .socketpair = sock_no_socketpair,
4341 .accept = sock_no_accept,
4342 .getname = packet_getname_spkt,
4343 .poll = datagram_poll,
4344 .ioctl = packet_ioctl,
4345 .listen = sock_no_listen,
4346 .shutdown = sock_no_shutdown,
4347 .setsockopt = sock_no_setsockopt,
4348 .getsockopt = sock_no_getsockopt,
4349 .sendmsg = packet_sendmsg_spkt,
4350 .recvmsg = packet_recvmsg,
4351 .mmap = sock_no_mmap,
4352 .sendpage = sock_no_sendpage,
4355 static const struct proto_ops packet_ops = {
4356 .family = PF_PACKET,
4357 .owner = THIS_MODULE,
4358 .release = packet_release,
4359 .bind = packet_bind,
4360 .connect = sock_no_connect,
4361 .socketpair = sock_no_socketpair,
4362 .accept = sock_no_accept,
4363 .getname = packet_getname,
4364 .poll = packet_poll,
4365 .ioctl = packet_ioctl,
4366 .listen = sock_no_listen,
4367 .shutdown = sock_no_shutdown,
4368 .setsockopt = packet_setsockopt,
4369 .getsockopt = packet_getsockopt,
4370 #ifdef CONFIG_COMPAT
4371 .compat_setsockopt = compat_packet_setsockopt,
4373 .sendmsg = packet_sendmsg,
4374 .recvmsg = packet_recvmsg,
4375 .mmap = packet_mmap,
4376 .sendpage = sock_no_sendpage,
4379 static const struct net_proto_family packet_family_ops = {
4380 .family = PF_PACKET,
4381 .create = packet_create,
4382 .owner = THIS_MODULE,
4385 static struct notifier_block packet_netdev_notifier = {
4386 .notifier_call = packet_notifier,
4389 #ifdef CONFIG_PROC_FS
4391 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4394 struct net *net = seq_file_net(seq);
4397 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4400 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4402 struct net *net = seq_file_net(seq);
4403 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4406 static void packet_seq_stop(struct seq_file *seq, void *v)
4412 static int packet_seq_show(struct seq_file *seq, void *v)
4414 if (v == SEQ_START_TOKEN)
4415 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4417 struct sock *s = sk_entry(v);
4418 const struct packet_sock *po = pkt_sk(s);
4421 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4423 atomic_read(&s->sk_refcnt),
4428 atomic_read(&s->sk_rmem_alloc),
4429 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4436 static const struct seq_operations packet_seq_ops = {
4437 .start = packet_seq_start,
4438 .next = packet_seq_next,
4439 .stop = packet_seq_stop,
4440 .show = packet_seq_show,
4443 static int packet_seq_open(struct inode *inode, struct file *file)
4445 return seq_open_net(inode, file, &packet_seq_ops,
4446 sizeof(struct seq_net_private));
4449 static const struct file_operations packet_seq_fops = {
4450 .owner = THIS_MODULE,
4451 .open = packet_seq_open,
4453 .llseek = seq_lseek,
4454 .release = seq_release_net,
4459 static int __net_init packet_net_init(struct net *net)
4461 mutex_init(&net->packet.sklist_lock);
4462 INIT_HLIST_HEAD(&net->packet.sklist);
4464 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4470 static void __net_exit packet_net_exit(struct net *net)
4472 remove_proc_entry("packet", net->proc_net);
4475 static struct pernet_operations packet_net_ops = {
4476 .init = packet_net_init,
4477 .exit = packet_net_exit,
4481 static void __exit packet_exit(void)
4483 unregister_netdevice_notifier(&packet_netdev_notifier);
4484 unregister_pernet_subsys(&packet_net_ops);
4485 sock_unregister(PF_PACKET);
4486 proto_unregister(&packet_proto);
4489 static int __init packet_init(void)
4491 int rc = proto_register(&packet_proto, 0);
4496 sock_register(&packet_family_ops);
4497 register_pernet_subsys(&packet_net_ops);
4498 register_netdevice_notifier(&packet_netdev_notifier);
4503 module_init(packet_init);
4504 module_exit(packet_exit);
4505 MODULE_LICENSE("GPL");
4506 MODULE_ALIAS_NETPROTO(PF_PACKET);