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 spin_unlock(&f->lock);
1503 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1505 struct packet_fanout *f = po->fanout;
1508 spin_lock(&f->lock);
1509 for (i = 0; i < f->num_members; i++) {
1510 if (f->arr[i] == sk)
1513 BUG_ON(i >= f->num_members);
1514 f->arr[i] = f->arr[f->num_members - 1];
1516 spin_unlock(&f->lock);
1519 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1521 if (sk->sk_family != PF_PACKET)
1524 return ptype->af_packet_priv == pkt_sk(sk)->fanout;
1527 static void fanout_init_data(struct packet_fanout *f)
1530 case PACKET_FANOUT_LB:
1531 atomic_set(&f->rr_cur, 0);
1533 case PACKET_FANOUT_CBPF:
1534 case PACKET_FANOUT_EBPF:
1535 RCU_INIT_POINTER(f->bpf_prog, NULL);
1540 static void __fanout_set_data_bpf(struct packet_fanout *f, struct bpf_prog *new)
1542 struct bpf_prog *old;
1544 spin_lock(&f->lock);
1545 old = rcu_dereference_protected(f->bpf_prog, lockdep_is_held(&f->lock));
1546 rcu_assign_pointer(f->bpf_prog, new);
1547 spin_unlock(&f->lock);
1551 bpf_prog_destroy(old);
1555 static int fanout_set_data_cbpf(struct packet_sock *po, char __user *data,
1558 struct bpf_prog *new;
1559 struct sock_fprog fprog;
1562 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1564 if (len != sizeof(fprog))
1566 if (copy_from_user(&fprog, data, len))
1569 ret = bpf_prog_create_from_user(&new, &fprog, NULL, false);
1573 __fanout_set_data_bpf(po->fanout, new);
1577 static int fanout_set_data_ebpf(struct packet_sock *po, char __user *data,
1580 struct bpf_prog *new;
1583 if (sock_flag(&po->sk, SOCK_FILTER_LOCKED))
1585 if (len != sizeof(fd))
1587 if (copy_from_user(&fd, data, len))
1590 new = bpf_prog_get_type(fd, BPF_PROG_TYPE_SOCKET_FILTER);
1592 return PTR_ERR(new);
1594 __fanout_set_data_bpf(po->fanout, new);
1598 static int fanout_set_data(struct packet_sock *po, char __user *data,
1601 switch (po->fanout->type) {
1602 case PACKET_FANOUT_CBPF:
1603 return fanout_set_data_cbpf(po, data, len);
1604 case PACKET_FANOUT_EBPF:
1605 return fanout_set_data_ebpf(po, data, len);
1611 static void fanout_release_data(struct packet_fanout *f)
1614 case PACKET_FANOUT_CBPF:
1615 case PACKET_FANOUT_EBPF:
1616 __fanout_set_data_bpf(f, NULL);
1620 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1622 struct packet_sock *po = pkt_sk(sk);
1623 struct packet_fanout *f, *match;
1624 u8 type = type_flags & 0xff;
1625 u8 flags = type_flags >> 8;
1629 case PACKET_FANOUT_ROLLOVER:
1630 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1632 case PACKET_FANOUT_HASH:
1633 case PACKET_FANOUT_LB:
1634 case PACKET_FANOUT_CPU:
1635 case PACKET_FANOUT_RND:
1636 case PACKET_FANOUT_QM:
1637 case PACKET_FANOUT_CBPF:
1638 case PACKET_FANOUT_EBPF:
1650 if (type == PACKET_FANOUT_ROLLOVER ||
1651 (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)) {
1652 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1655 atomic_long_set(&po->rollover->num, 0);
1656 atomic_long_set(&po->rollover->num_huge, 0);
1657 atomic_long_set(&po->rollover->num_failed, 0);
1660 mutex_lock(&fanout_mutex);
1662 list_for_each_entry(f, &fanout_list, list) {
1664 read_pnet(&f->net) == sock_net(sk)) {
1670 if (match && match->flags != flags)
1674 match = kzalloc(sizeof(*match), GFP_KERNEL);
1677 write_pnet(&match->net, sock_net(sk));
1680 match->flags = flags;
1681 INIT_LIST_HEAD(&match->list);
1682 spin_lock_init(&match->lock);
1683 atomic_set(&match->sk_ref, 0);
1684 fanout_init_data(match);
1685 match->prot_hook.type = po->prot_hook.type;
1686 match->prot_hook.dev = po->prot_hook.dev;
1687 match->prot_hook.func = packet_rcv_fanout;
1688 match->prot_hook.af_packet_priv = match;
1689 match->prot_hook.id_match = match_fanout_group;
1690 dev_add_pack(&match->prot_hook);
1691 list_add(&match->list, &fanout_list);
1694 if (match->type == type &&
1695 match->prot_hook.type == po->prot_hook.type &&
1696 match->prot_hook.dev == po->prot_hook.dev) {
1698 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1699 __dev_remove_pack(&po->prot_hook);
1701 atomic_inc(&match->sk_ref);
1702 __fanout_link(sk, po);
1707 mutex_unlock(&fanout_mutex);
1709 kfree(po->rollover);
1710 po->rollover = NULL;
1715 static void fanout_release(struct sock *sk)
1717 struct packet_sock *po = pkt_sk(sk);
1718 struct packet_fanout *f;
1724 mutex_lock(&fanout_mutex);
1727 if (atomic_dec_and_test(&f->sk_ref)) {
1729 dev_remove_pack(&f->prot_hook);
1730 fanout_release_data(f);
1733 mutex_unlock(&fanout_mutex);
1736 kfree_rcu(po->rollover, rcu);
1739 static bool packet_extra_vlan_len_allowed(const struct net_device *dev,
1740 struct sk_buff *skb)
1742 /* Earlier code assumed this would be a VLAN pkt, double-check
1743 * this now that we have the actual packet in hand. We can only
1744 * do this check on Ethernet devices.
1746 if (unlikely(dev->type != ARPHRD_ETHER))
1749 skb_reset_mac_header(skb);
1750 return likely(eth_hdr(skb)->h_proto == htons(ETH_P_8021Q));
1753 static const struct proto_ops packet_ops;
1755 static const struct proto_ops packet_ops_spkt;
1757 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1758 struct packet_type *pt, struct net_device *orig_dev)
1761 struct sockaddr_pkt *spkt;
1764 * When we registered the protocol we saved the socket in the data
1765 * field for just this event.
1768 sk = pt->af_packet_priv;
1771 * Yank back the headers [hope the device set this
1772 * right or kerboom...]
1774 * Incoming packets have ll header pulled,
1777 * For outgoing ones skb->data == skb_mac_header(skb)
1778 * so that this procedure is noop.
1781 if (skb->pkt_type == PACKET_LOOPBACK)
1784 if (!net_eq(dev_net(dev), sock_net(sk)))
1787 skb = skb_share_check(skb, GFP_ATOMIC);
1791 /* drop any routing info */
1794 /* drop conntrack reference */
1797 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1799 skb_push(skb, skb->data - skb_mac_header(skb));
1802 * The SOCK_PACKET socket receives _all_ frames.
1805 spkt->spkt_family = dev->type;
1806 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1807 spkt->spkt_protocol = skb->protocol;
1810 * Charge the memory to the socket. This is done specifically
1811 * to prevent sockets using all the memory up.
1814 if (sock_queue_rcv_skb(sk, skb) == 0)
1825 * Output a raw packet to a device layer. This bypasses all the other
1826 * protocol layers and you must therefore supply it with a complete frame
1829 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1832 struct sock *sk = sock->sk;
1833 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1834 struct sk_buff *skb = NULL;
1835 struct net_device *dev;
1836 struct sockcm_cookie sockc;
1842 * Get and verify the address.
1846 if (msg->msg_namelen < sizeof(struct sockaddr))
1848 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1849 proto = saddr->spkt_protocol;
1851 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1854 * Find the device first to size check it
1857 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1860 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1866 if (!(dev->flags & IFF_UP))
1870 * You may not queue a frame bigger than the mtu. This is the lowest level
1871 * raw protocol and you must do your own fragmentation at this level.
1874 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1875 if (!netif_supports_nofcs(dev)) {
1876 err = -EPROTONOSUPPORT;
1879 extra_len = 4; /* We're doing our own CRC */
1883 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1887 size_t reserved = LL_RESERVED_SPACE(dev);
1888 int tlen = dev->needed_tailroom;
1889 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1892 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1895 /* FIXME: Save some space for broken drivers that write a hard
1896 * header at transmission time by themselves. PPP is the notable
1897 * one here. This should really be fixed at the driver level.
1899 skb_reserve(skb, reserved);
1900 skb_reset_network_header(skb);
1902 /* Try to align data part correctly */
1907 skb_reset_network_header(skb);
1909 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1915 if (!dev_validate_header(dev, skb->data, len)) {
1919 if (len > (dev->mtu + dev->hard_header_len + extra_len) &&
1920 !packet_extra_vlan_len_allowed(dev, skb)) {
1925 sockc.tsflags = sk->sk_tsflags;
1926 if (msg->msg_controllen) {
1927 err = sock_cmsg_send(sk, msg, &sockc);
1932 skb->protocol = proto;
1934 skb->priority = sk->sk_priority;
1935 skb->mark = sk->sk_mark;
1937 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
1939 if (unlikely(extra_len == 4))
1942 skb_probe_transport_header(skb, 0);
1944 dev_queue_xmit(skb);
1955 static unsigned int run_filter(struct sk_buff *skb,
1956 const struct sock *sk,
1959 struct sk_filter *filter;
1962 filter = rcu_dereference(sk->sk_filter);
1964 res = bpf_prog_run_clear_cb(filter->prog, skb);
1970 static int packet_rcv_vnet(struct msghdr *msg, const struct sk_buff *skb,
1973 struct virtio_net_hdr vnet_hdr;
1975 if (*len < sizeof(vnet_hdr))
1977 *len -= sizeof(vnet_hdr);
1979 if (virtio_net_hdr_from_skb(skb, &vnet_hdr, vio_le(), true))
1982 return memcpy_to_msg(msg, (void *)&vnet_hdr, sizeof(vnet_hdr));
1986 * This function makes lazy skb cloning in hope that most of packets
1987 * are discarded by BPF.
1989 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1990 * and skb->cb are mangled. It works because (and until) packets
1991 * falling here are owned by current CPU. Output packets are cloned
1992 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1993 * sequencially, so that if we return skb to original state on exit,
1994 * we will not harm anyone.
1997 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1998 struct packet_type *pt, struct net_device *orig_dev)
2001 struct sockaddr_ll *sll;
2002 struct packet_sock *po;
2003 u8 *skb_head = skb->data;
2004 int skb_len = skb->len;
2005 unsigned int snaplen, res;
2006 bool is_drop_n_account = false;
2008 if (skb->pkt_type == PACKET_LOOPBACK)
2011 sk = pt->af_packet_priv;
2014 if (!net_eq(dev_net(dev), sock_net(sk)))
2019 if (dev->header_ops) {
2020 /* The device has an explicit notion of ll header,
2021 * exported to higher levels.
2023 * Otherwise, the device hides details of its frame
2024 * structure, so that corresponding packet head is
2025 * never delivered to user.
2027 if (sk->sk_type != SOCK_DGRAM)
2028 skb_push(skb, skb->data - skb_mac_header(skb));
2029 else if (skb->pkt_type == PACKET_OUTGOING) {
2030 /* Special case: outgoing packets have ll header at head */
2031 skb_pull(skb, skb_network_offset(skb));
2037 res = run_filter(skb, sk, snaplen);
2039 goto drop_n_restore;
2043 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
2046 if (skb_shared(skb)) {
2047 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2051 if (skb_head != skb->data) {
2052 skb->data = skb_head;
2059 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
2061 sll = &PACKET_SKB_CB(skb)->sa.ll;
2062 sll->sll_hatype = dev->type;
2063 sll->sll_pkttype = skb->pkt_type;
2064 if (unlikely(po->origdev))
2065 sll->sll_ifindex = orig_dev->ifindex;
2067 sll->sll_ifindex = dev->ifindex;
2069 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2071 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
2072 * Use their space for storing the original skb length.
2074 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
2076 if (pskb_trim(skb, snaplen))
2079 skb_set_owner_r(skb, sk);
2083 /* drop conntrack reference */
2086 spin_lock(&sk->sk_receive_queue.lock);
2087 po->stats.stats1.tp_packets++;
2088 sock_skb_set_dropcount(sk, skb);
2089 __skb_queue_tail(&sk->sk_receive_queue, skb);
2090 spin_unlock(&sk->sk_receive_queue.lock);
2091 sk->sk_data_ready(sk);
2095 is_drop_n_account = true;
2096 spin_lock(&sk->sk_receive_queue.lock);
2097 po->stats.stats1.tp_drops++;
2098 atomic_inc(&sk->sk_drops);
2099 spin_unlock(&sk->sk_receive_queue.lock);
2102 if (skb_head != skb->data && skb_shared(skb)) {
2103 skb->data = skb_head;
2107 if (!is_drop_n_account)
2114 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
2115 struct packet_type *pt, struct net_device *orig_dev)
2118 struct packet_sock *po;
2119 struct sockaddr_ll *sll;
2120 union tpacket_uhdr h;
2121 u8 *skb_head = skb->data;
2122 int skb_len = skb->len;
2123 unsigned int snaplen, res;
2124 unsigned long status = TP_STATUS_USER;
2125 unsigned short macoff, netoff, hdrlen;
2126 struct sk_buff *copy_skb = NULL;
2129 bool is_drop_n_account = false;
2131 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
2132 * We may add members to them until current aligned size without forcing
2133 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
2135 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
2136 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
2138 if (skb->pkt_type == PACKET_LOOPBACK)
2141 sk = pt->af_packet_priv;
2144 if (!net_eq(dev_net(dev), sock_net(sk)))
2147 if (dev->header_ops) {
2148 if (sk->sk_type != SOCK_DGRAM)
2149 skb_push(skb, skb->data - skb_mac_header(skb));
2150 else if (skb->pkt_type == PACKET_OUTGOING) {
2151 /* Special case: outgoing packets have ll header at head */
2152 skb_pull(skb, skb_network_offset(skb));
2158 res = run_filter(skb, sk, snaplen);
2160 goto drop_n_restore;
2162 if (skb->ip_summed == CHECKSUM_PARTIAL)
2163 status |= TP_STATUS_CSUMNOTREADY;
2164 else if (skb->pkt_type != PACKET_OUTGOING &&
2165 (skb->ip_summed == CHECKSUM_COMPLETE ||
2166 skb_csum_unnecessary(skb)))
2167 status |= TP_STATUS_CSUM_VALID;
2172 if (sk->sk_type == SOCK_DGRAM) {
2173 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
2176 unsigned int maclen = skb_network_offset(skb);
2177 netoff = TPACKET_ALIGN(po->tp_hdrlen +
2178 (maclen < 16 ? 16 : maclen)) +
2180 if (po->has_vnet_hdr)
2181 netoff += sizeof(struct virtio_net_hdr);
2182 macoff = netoff - maclen;
2184 if (po->tp_version <= TPACKET_V2) {
2185 if (macoff + snaplen > po->rx_ring.frame_size) {
2186 if (po->copy_thresh &&
2187 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2188 if (skb_shared(skb)) {
2189 copy_skb = skb_clone(skb, GFP_ATOMIC);
2191 copy_skb = skb_get(skb);
2192 skb_head = skb->data;
2195 skb_set_owner_r(copy_skb, sk);
2197 snaplen = po->rx_ring.frame_size - macoff;
2198 if ((int)snaplen < 0)
2201 } else if (unlikely(macoff + snaplen >
2202 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2205 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2206 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2207 snaplen, nval, macoff);
2209 if (unlikely((int)snaplen < 0)) {
2211 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2214 spin_lock(&sk->sk_receive_queue.lock);
2215 h.raw = packet_current_rx_frame(po, skb,
2216 TP_STATUS_KERNEL, (macoff+snaplen));
2218 goto drop_n_account;
2219 if (po->tp_version <= TPACKET_V2) {
2220 packet_increment_rx_head(po, &po->rx_ring);
2222 * LOSING will be reported till you read the stats,
2223 * because it's COR - Clear On Read.
2224 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2227 if (po->stats.stats1.tp_drops)
2228 status |= TP_STATUS_LOSING;
2230 po->stats.stats1.tp_packets++;
2232 status |= TP_STATUS_COPY;
2233 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2235 spin_unlock(&sk->sk_receive_queue.lock);
2237 if (po->has_vnet_hdr) {
2238 if (virtio_net_hdr_from_skb(skb, h.raw + macoff -
2239 sizeof(struct virtio_net_hdr),
2241 spin_lock(&sk->sk_receive_queue.lock);
2242 goto drop_n_account;
2246 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2248 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2249 getnstimeofday(&ts);
2251 status |= ts_status;
2253 switch (po->tp_version) {
2255 h.h1->tp_len = skb->len;
2256 h.h1->tp_snaplen = snaplen;
2257 h.h1->tp_mac = macoff;
2258 h.h1->tp_net = netoff;
2259 h.h1->tp_sec = ts.tv_sec;
2260 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2261 hdrlen = sizeof(*h.h1);
2264 h.h2->tp_len = skb->len;
2265 h.h2->tp_snaplen = snaplen;
2266 h.h2->tp_mac = macoff;
2267 h.h2->tp_net = netoff;
2268 h.h2->tp_sec = ts.tv_sec;
2269 h.h2->tp_nsec = ts.tv_nsec;
2270 if (skb_vlan_tag_present(skb)) {
2271 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2272 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2273 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2275 h.h2->tp_vlan_tci = 0;
2276 h.h2->tp_vlan_tpid = 0;
2278 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2279 hdrlen = sizeof(*h.h2);
2282 /* tp_nxt_offset,vlan are already populated above.
2283 * So DONT clear those fields here
2285 h.h3->tp_status |= status;
2286 h.h3->tp_len = skb->len;
2287 h.h3->tp_snaplen = snaplen;
2288 h.h3->tp_mac = macoff;
2289 h.h3->tp_net = netoff;
2290 h.h3->tp_sec = ts.tv_sec;
2291 h.h3->tp_nsec = ts.tv_nsec;
2292 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2293 hdrlen = sizeof(*h.h3);
2299 sll = h.raw + TPACKET_ALIGN(hdrlen);
2300 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2301 sll->sll_family = AF_PACKET;
2302 sll->sll_hatype = dev->type;
2303 sll->sll_protocol = skb->protocol;
2304 sll->sll_pkttype = skb->pkt_type;
2305 if (unlikely(po->origdev))
2306 sll->sll_ifindex = orig_dev->ifindex;
2308 sll->sll_ifindex = dev->ifindex;
2312 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2313 if (po->tp_version <= TPACKET_V2) {
2316 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2319 for (start = h.raw; start < end; start += PAGE_SIZE)
2320 flush_dcache_page(pgv_to_page(start));
2325 if (po->tp_version <= TPACKET_V2) {
2326 __packet_set_status(po, h.raw, status);
2327 sk->sk_data_ready(sk);
2329 prb_clear_blk_fill_status(&po->rx_ring);
2333 if (skb_head != skb->data && skb_shared(skb)) {
2334 skb->data = skb_head;
2338 if (!is_drop_n_account)
2345 is_drop_n_account = true;
2346 po->stats.stats1.tp_drops++;
2347 spin_unlock(&sk->sk_receive_queue.lock);
2349 sk->sk_data_ready(sk);
2350 kfree_skb(copy_skb);
2351 goto drop_n_restore;
2354 static void tpacket_destruct_skb(struct sk_buff *skb)
2356 struct packet_sock *po = pkt_sk(skb->sk);
2358 if (likely(po->tx_ring.pg_vec)) {
2362 ph = skb_shinfo(skb)->destructor_arg;
2363 packet_dec_pending(&po->tx_ring);
2365 ts = __packet_set_timestamp(po, ph, skb);
2366 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2372 static void tpacket_set_protocol(const struct net_device *dev,
2373 struct sk_buff *skb)
2375 if (dev->type == ARPHRD_ETHER) {
2376 skb_reset_mac_header(skb);
2377 skb->protocol = eth_hdr(skb)->h_proto;
2381 static int __packet_snd_vnet_parse(struct virtio_net_hdr *vnet_hdr, size_t len)
2383 if ((vnet_hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2384 (__virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2385 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2 >
2386 __virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len)))
2387 vnet_hdr->hdr_len = __cpu_to_virtio16(vio_le(),
2388 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_start) +
2389 __virtio16_to_cpu(vio_le(), vnet_hdr->csum_offset) + 2);
2391 if (__virtio16_to_cpu(vio_le(), vnet_hdr->hdr_len) > len)
2397 static int packet_snd_vnet_parse(struct msghdr *msg, size_t *len,
2398 struct virtio_net_hdr *vnet_hdr)
2400 if (*len < sizeof(*vnet_hdr))
2402 *len -= sizeof(*vnet_hdr);
2404 if (!copy_from_iter_full(vnet_hdr, sizeof(*vnet_hdr), &msg->msg_iter))
2407 return __packet_snd_vnet_parse(vnet_hdr, *len);
2410 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2411 void *frame, struct net_device *dev, void *data, int tp_len,
2412 __be16 proto, unsigned char *addr, int hlen, int copylen,
2413 const struct sockcm_cookie *sockc)
2415 union tpacket_uhdr ph;
2416 int to_write, offset, len, nr_frags, len_max;
2417 struct socket *sock = po->sk.sk_socket;
2423 skb->protocol = proto;
2425 skb->priority = po->sk.sk_priority;
2426 skb->mark = po->sk.sk_mark;
2427 sock_tx_timestamp(&po->sk, sockc->tsflags, &skb_shinfo(skb)->tx_flags);
2428 skb_shinfo(skb)->destructor_arg = ph.raw;
2430 skb_reserve(skb, hlen);
2431 skb_reset_network_header(skb);
2435 if (sock->type == SOCK_DGRAM) {
2436 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2438 if (unlikely(err < 0))
2440 } else if (copylen) {
2441 int hdrlen = min_t(int, copylen, tp_len);
2443 skb_push(skb, dev->hard_header_len);
2444 skb_put(skb, copylen - dev->hard_header_len);
2445 err = skb_store_bits(skb, 0, data, hdrlen);
2448 if (!dev_validate_header(dev, skb->data, hdrlen))
2451 tpacket_set_protocol(dev, skb);
2457 offset = offset_in_page(data);
2458 len_max = PAGE_SIZE - offset;
2459 len = ((to_write > len_max) ? len_max : to_write);
2461 skb->data_len = to_write;
2462 skb->len += to_write;
2463 skb->truesize += to_write;
2464 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2466 while (likely(to_write)) {
2467 nr_frags = skb_shinfo(skb)->nr_frags;
2469 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2470 pr_err("Packet exceed the number of skb frags(%lu)\n",
2475 page = pgv_to_page(data);
2477 flush_dcache_page(page);
2479 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2482 len_max = PAGE_SIZE;
2483 len = ((to_write > len_max) ? len_max : to_write);
2486 skb_probe_transport_header(skb, 0);
2491 static int tpacket_parse_header(struct packet_sock *po, void *frame,
2492 int size_max, void **data)
2494 union tpacket_uhdr ph;
2499 switch (po->tp_version) {
2501 tp_len = ph.h2->tp_len;
2504 tp_len = ph.h1->tp_len;
2507 if (unlikely(tp_len > size_max)) {
2508 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2512 if (unlikely(po->tp_tx_has_off)) {
2513 int off_min, off_max;
2515 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2516 off_max = po->tx_ring.frame_size - tp_len;
2517 if (po->sk.sk_type == SOCK_DGRAM) {
2518 switch (po->tp_version) {
2520 off = ph.h2->tp_net;
2523 off = ph.h1->tp_net;
2527 switch (po->tp_version) {
2529 off = ph.h2->tp_mac;
2532 off = ph.h1->tp_mac;
2536 if (unlikely((off < off_min) || (off_max < off)))
2539 off = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2542 *data = frame + off;
2546 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2548 struct sk_buff *skb;
2549 struct net_device *dev;
2550 struct virtio_net_hdr *vnet_hdr = NULL;
2551 struct sockcm_cookie sockc;
2553 int err, reserve = 0;
2555 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2556 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2557 int tp_len, size_max;
2558 unsigned char *addr;
2561 int status = TP_STATUS_AVAILABLE;
2562 int hlen, tlen, copylen = 0;
2564 mutex_lock(&po->pg_vec_lock);
2566 if (likely(saddr == NULL)) {
2567 dev = packet_cached_dev_get(po);
2572 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2574 if (msg->msg_namelen < (saddr->sll_halen
2575 + offsetof(struct sockaddr_ll,
2578 proto = saddr->sll_protocol;
2579 addr = saddr->sll_addr;
2580 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2583 sockc.tsflags = po->sk.sk_tsflags;
2584 if (msg->msg_controllen) {
2585 err = sock_cmsg_send(&po->sk, msg, &sockc);
2591 if (unlikely(dev == NULL))
2594 if (unlikely(!(dev->flags & IFF_UP)))
2597 if (po->sk.sk_socket->type == SOCK_RAW)
2598 reserve = dev->hard_header_len;
2599 size_max = po->tx_ring.frame_size
2600 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2602 if ((size_max > dev->mtu + reserve + VLAN_HLEN) && !po->has_vnet_hdr)
2603 size_max = dev->mtu + reserve + VLAN_HLEN;
2606 ph = packet_current_frame(po, &po->tx_ring,
2607 TP_STATUS_SEND_REQUEST);
2608 if (unlikely(ph == NULL)) {
2609 if (need_wait && need_resched())
2615 tp_len = tpacket_parse_header(po, ph, size_max, &data);
2619 status = TP_STATUS_SEND_REQUEST;
2620 hlen = LL_RESERVED_SPACE(dev);
2621 tlen = dev->needed_tailroom;
2622 if (po->has_vnet_hdr) {
2624 data += sizeof(*vnet_hdr);
2625 tp_len -= sizeof(*vnet_hdr);
2627 __packet_snd_vnet_parse(vnet_hdr, tp_len)) {
2631 copylen = __virtio16_to_cpu(vio_le(),
2634 copylen = max_t(int, copylen, dev->hard_header_len);
2635 skb = sock_alloc_send_skb(&po->sk,
2636 hlen + tlen + sizeof(struct sockaddr_ll) +
2637 (copylen - dev->hard_header_len),
2640 if (unlikely(skb == NULL)) {
2641 /* we assume the socket was initially writeable ... */
2642 if (likely(len_sum > 0))
2646 tp_len = tpacket_fill_skb(po, skb, ph, dev, data, tp_len, proto,
2647 addr, hlen, copylen, &sockc);
2648 if (likely(tp_len >= 0) &&
2649 tp_len > dev->mtu + reserve &&
2650 !po->has_vnet_hdr &&
2651 !packet_extra_vlan_len_allowed(dev, skb))
2654 if (unlikely(tp_len < 0)) {
2657 __packet_set_status(po, ph,
2658 TP_STATUS_AVAILABLE);
2659 packet_increment_head(&po->tx_ring);
2663 status = TP_STATUS_WRONG_FORMAT;
2669 if (po->has_vnet_hdr && virtio_net_hdr_to_skb(skb, vnet_hdr,
2675 packet_pick_tx_queue(dev, skb);
2677 skb->destructor = tpacket_destruct_skb;
2678 __packet_set_status(po, ph, TP_STATUS_SENDING);
2679 packet_inc_pending(&po->tx_ring);
2681 status = TP_STATUS_SEND_REQUEST;
2682 err = po->xmit(skb);
2683 if (unlikely(err > 0)) {
2684 err = net_xmit_errno(err);
2685 if (err && __packet_get_status(po, ph) ==
2686 TP_STATUS_AVAILABLE) {
2687 /* skb was destructed already */
2692 * skb was dropped but not destructed yet;
2693 * let's treat it like congestion or err < 0
2697 packet_increment_head(&po->tx_ring);
2699 } while (likely((ph != NULL) ||
2700 /* Note: packet_read_pending() might be slow if we have
2701 * to call it as it's per_cpu variable, but in fast-path
2702 * we already short-circuit the loop with the first
2703 * condition, and luckily don't have to go that path
2706 (need_wait && packet_read_pending(&po->tx_ring))));
2712 __packet_set_status(po, ph, status);
2717 mutex_unlock(&po->pg_vec_lock);
2721 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2722 size_t reserve, size_t len,
2723 size_t linear, int noblock,
2726 struct sk_buff *skb;
2728 /* Under a page? Don't bother with paged skb. */
2729 if (prepad + len < PAGE_SIZE || !linear)
2732 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2737 skb_reserve(skb, reserve);
2738 skb_put(skb, linear);
2739 skb->data_len = len - linear;
2740 skb->len += len - linear;
2745 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2747 struct sock *sk = sock->sk;
2748 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2749 struct sk_buff *skb;
2750 struct net_device *dev;
2752 unsigned char *addr;
2753 int err, reserve = 0;
2754 struct sockcm_cookie sockc;
2755 struct virtio_net_hdr vnet_hdr = { 0 };
2757 struct packet_sock *po = pkt_sk(sk);
2758 int hlen, tlen, linear;
2762 * Get and verify the address.
2765 if (likely(saddr == NULL)) {
2766 dev = packet_cached_dev_get(po);
2771 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2773 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2775 proto = saddr->sll_protocol;
2776 addr = saddr->sll_addr;
2777 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2781 if (unlikely(dev == NULL))
2784 if (unlikely(!(dev->flags & IFF_UP)))
2787 sockc.tsflags = sk->sk_tsflags;
2788 sockc.mark = sk->sk_mark;
2789 if (msg->msg_controllen) {
2790 err = sock_cmsg_send(sk, msg, &sockc);
2795 if (sock->type == SOCK_RAW)
2796 reserve = dev->hard_header_len;
2797 if (po->has_vnet_hdr) {
2798 err = packet_snd_vnet_parse(msg, &len, &vnet_hdr);
2803 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2804 if (!netif_supports_nofcs(dev)) {
2805 err = -EPROTONOSUPPORT;
2808 extra_len = 4; /* We're doing our own CRC */
2812 if (!vnet_hdr.gso_type &&
2813 (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2817 hlen = LL_RESERVED_SPACE(dev);
2818 tlen = dev->needed_tailroom;
2819 linear = __virtio16_to_cpu(vio_le(), vnet_hdr.hdr_len);
2820 linear = max(linear, min_t(int, len, dev->hard_header_len));
2821 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len, linear,
2822 msg->msg_flags & MSG_DONTWAIT, &err);
2826 skb_set_network_header(skb, reserve);
2829 if (sock->type == SOCK_DGRAM) {
2830 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2831 if (unlikely(offset < 0))
2835 /* Returns -EFAULT on error */
2836 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2840 if (sock->type == SOCK_RAW &&
2841 !dev_validate_header(dev, skb->data, len)) {
2846 sock_tx_timestamp(sk, sockc.tsflags, &skb_shinfo(skb)->tx_flags);
2848 if (!vnet_hdr.gso_type && (len > dev->mtu + reserve + extra_len) &&
2849 !packet_extra_vlan_len_allowed(dev, skb)) {
2854 skb->protocol = proto;
2856 skb->priority = sk->sk_priority;
2857 skb->mark = sockc.mark;
2859 packet_pick_tx_queue(dev, skb);
2861 if (po->has_vnet_hdr) {
2862 err = virtio_net_hdr_to_skb(skb, &vnet_hdr, vio_le());
2865 len += sizeof(vnet_hdr);
2868 skb_probe_transport_header(skb, reserve);
2870 if (unlikely(extra_len == 4))
2873 err = po->xmit(skb);
2874 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2890 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2892 struct sock *sk = sock->sk;
2893 struct packet_sock *po = pkt_sk(sk);
2895 if (po->tx_ring.pg_vec)
2896 return tpacket_snd(po, msg);
2898 return packet_snd(sock, msg, len);
2902 * Close a PACKET socket. This is fairly simple. We immediately go
2903 * to 'closed' state and remove our protocol entry in the device list.
2906 static int packet_release(struct socket *sock)
2908 struct sock *sk = sock->sk;
2909 struct packet_sock *po;
2911 union tpacket_req_u req_u;
2919 mutex_lock(&net->packet.sklist_lock);
2920 sk_del_node_init_rcu(sk);
2921 mutex_unlock(&net->packet.sklist_lock);
2924 sock_prot_inuse_add(net, sk->sk_prot, -1);
2927 spin_lock(&po->bind_lock);
2928 unregister_prot_hook(sk, false);
2929 packet_cached_dev_reset(po);
2931 if (po->prot_hook.dev) {
2932 dev_put(po->prot_hook.dev);
2933 po->prot_hook.dev = NULL;
2935 spin_unlock(&po->bind_lock);
2937 packet_flush_mclist(sk);
2939 if (po->rx_ring.pg_vec) {
2940 memset(&req_u, 0, sizeof(req_u));
2941 packet_set_ring(sk, &req_u, 1, 0);
2944 if (po->tx_ring.pg_vec) {
2945 memset(&req_u, 0, sizeof(req_u));
2946 packet_set_ring(sk, &req_u, 1, 1);
2953 * Now the socket is dead. No more input will appear.
2960 skb_queue_purge(&sk->sk_receive_queue);
2961 packet_free_pending(po);
2962 sk_refcnt_debug_release(sk);
2969 * Attach a packet hook.
2972 static int packet_do_bind(struct sock *sk, const char *name, int ifindex,
2975 struct packet_sock *po = pkt_sk(sk);
2976 struct net_device *dev_curr;
2979 struct net_device *dev = NULL;
2981 bool unlisted = false;
2987 spin_lock(&po->bind_lock);
2991 dev = dev_get_by_name_rcu(sock_net(sk), name);
2996 } else if (ifindex) {
2997 dev = dev_get_by_index_rcu(sock_net(sk), ifindex);
3007 proto_curr = po->prot_hook.type;
3008 dev_curr = po->prot_hook.dev;
3010 need_rehook = proto_curr != proto || dev_curr != dev;
3015 __unregister_prot_hook(sk, true);
3017 dev_curr = po->prot_hook.dev;
3019 unlisted = !dev_get_by_index_rcu(sock_net(sk),
3024 po->prot_hook.type = proto;
3026 if (unlikely(unlisted)) {
3028 po->prot_hook.dev = NULL;
3030 packet_cached_dev_reset(po);
3032 po->prot_hook.dev = dev;
3033 po->ifindex = dev ? dev->ifindex : 0;
3034 packet_cached_dev_assign(po, dev);
3040 if (proto == 0 || !need_rehook)
3043 if (!unlisted && (!dev || (dev->flags & IFF_UP))) {
3044 register_prot_hook(sk);
3046 sk->sk_err = ENETDOWN;
3047 if (!sock_flag(sk, SOCK_DEAD))
3048 sk->sk_error_report(sk);
3053 spin_unlock(&po->bind_lock);
3059 * Bind a packet socket to a device
3062 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
3065 struct sock *sk = sock->sk;
3072 if (addr_len != sizeof(struct sockaddr))
3074 strlcpy(name, uaddr->sa_data, sizeof(name));
3076 return packet_do_bind(sk, name, 0, pkt_sk(sk)->num);
3079 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
3081 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
3082 struct sock *sk = sock->sk;
3088 if (addr_len < sizeof(struct sockaddr_ll))
3090 if (sll->sll_family != AF_PACKET)
3093 return packet_do_bind(sk, NULL, sll->sll_ifindex,
3094 sll->sll_protocol ? : pkt_sk(sk)->num);
3097 static struct proto packet_proto = {
3099 .owner = THIS_MODULE,
3100 .obj_size = sizeof(struct packet_sock),
3104 * Create a packet of type SOCK_PACKET.
3107 static int packet_create(struct net *net, struct socket *sock, int protocol,
3111 struct packet_sock *po;
3112 __be16 proto = (__force __be16)protocol; /* weird, but documented */
3115 if (!ns_capable(net->user_ns, CAP_NET_RAW))
3117 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
3118 sock->type != SOCK_PACKET)
3119 return -ESOCKTNOSUPPORT;
3121 sock->state = SS_UNCONNECTED;
3124 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
3128 sock->ops = &packet_ops;
3129 if (sock->type == SOCK_PACKET)
3130 sock->ops = &packet_ops_spkt;
3132 sock_init_data(sock, sk);
3135 sk->sk_family = PF_PACKET;
3137 po->xmit = dev_queue_xmit;
3139 err = packet_alloc_pending(po);
3143 packet_cached_dev_reset(po);
3145 sk->sk_destruct = packet_sock_destruct;
3146 sk_refcnt_debug_inc(sk);
3149 * Attach a protocol block
3152 spin_lock_init(&po->bind_lock);
3153 mutex_init(&po->pg_vec_lock);
3154 po->rollover = NULL;
3155 po->prot_hook.func = packet_rcv;
3157 if (sock->type == SOCK_PACKET)
3158 po->prot_hook.func = packet_rcv_spkt;
3160 po->prot_hook.af_packet_priv = sk;
3163 po->prot_hook.type = proto;
3164 register_prot_hook(sk);
3167 mutex_lock(&net->packet.sklist_lock);
3168 sk_add_node_rcu(sk, &net->packet.sklist);
3169 mutex_unlock(&net->packet.sklist_lock);
3172 sock_prot_inuse_add(net, &packet_proto, 1);
3183 * Pull a packet from our receive queue and hand it to the user.
3184 * If necessary we block.
3187 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
3190 struct sock *sk = sock->sk;
3191 struct sk_buff *skb;
3193 int vnet_hdr_len = 0;
3194 unsigned int origlen = 0;
3197 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
3201 /* What error should we return now? EUNATTACH? */
3202 if (pkt_sk(sk)->ifindex < 0)
3206 if (flags & MSG_ERRQUEUE) {
3207 err = sock_recv_errqueue(sk, msg, len,
3208 SOL_PACKET, PACKET_TX_TIMESTAMP);
3213 * Call the generic datagram receiver. This handles all sorts
3214 * of horrible races and re-entrancy so we can forget about it
3215 * in the protocol layers.
3217 * Now it will return ENETDOWN, if device have just gone down,
3218 * but then it will block.
3221 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
3224 * An error occurred so return it. Because skb_recv_datagram()
3225 * handles the blocking we don't see and worry about blocking
3232 if (pkt_sk(sk)->pressure)
3233 packet_rcv_has_room(pkt_sk(sk), NULL);
3235 if (pkt_sk(sk)->has_vnet_hdr) {
3236 err = packet_rcv_vnet(msg, skb, &len);
3239 vnet_hdr_len = sizeof(struct virtio_net_hdr);
3242 /* You lose any data beyond the buffer you gave. If it worries
3243 * a user program they can ask the device for its MTU
3249 msg->msg_flags |= MSG_TRUNC;
3252 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3256 if (sock->type != SOCK_PACKET) {
3257 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3259 /* Original length was stored in sockaddr_ll fields */
3260 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3261 sll->sll_family = AF_PACKET;
3262 sll->sll_protocol = skb->protocol;
3265 sock_recv_ts_and_drops(msg, sk, skb);
3267 if (msg->msg_name) {
3268 /* If the address length field is there to be filled
3269 * in, we fill it in now.
3271 if (sock->type == SOCK_PACKET) {
3272 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3273 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3275 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3277 msg->msg_namelen = sll->sll_halen +
3278 offsetof(struct sockaddr_ll, sll_addr);
3280 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3284 if (pkt_sk(sk)->auxdata) {
3285 struct tpacket_auxdata aux;
3287 aux.tp_status = TP_STATUS_USER;
3288 if (skb->ip_summed == CHECKSUM_PARTIAL)
3289 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3290 else if (skb->pkt_type != PACKET_OUTGOING &&
3291 (skb->ip_summed == CHECKSUM_COMPLETE ||
3292 skb_csum_unnecessary(skb)))
3293 aux.tp_status |= TP_STATUS_CSUM_VALID;
3295 aux.tp_len = origlen;
3296 aux.tp_snaplen = skb->len;
3298 aux.tp_net = skb_network_offset(skb);
3299 if (skb_vlan_tag_present(skb)) {
3300 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3301 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3302 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3304 aux.tp_vlan_tci = 0;
3305 aux.tp_vlan_tpid = 0;
3307 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3311 * Free or return the buffer as appropriate. Again this
3312 * hides all the races and re-entrancy issues from us.
3314 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3317 skb_free_datagram(sk, skb);
3322 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3323 int *uaddr_len, int peer)
3325 struct net_device *dev;
3326 struct sock *sk = sock->sk;
3331 uaddr->sa_family = AF_PACKET;
3332 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3334 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3336 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3338 *uaddr_len = sizeof(*uaddr);
3343 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3344 int *uaddr_len, int peer)
3346 struct net_device *dev;
3347 struct sock *sk = sock->sk;
3348 struct packet_sock *po = pkt_sk(sk);
3349 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3354 sll->sll_family = AF_PACKET;
3355 sll->sll_ifindex = po->ifindex;
3356 sll->sll_protocol = po->num;
3357 sll->sll_pkttype = 0;
3359 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3361 sll->sll_hatype = dev->type;
3362 sll->sll_halen = dev->addr_len;
3363 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3365 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3369 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3374 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3378 case PACKET_MR_MULTICAST:
3379 if (i->alen != dev->addr_len)
3382 return dev_mc_add(dev, i->addr);
3384 return dev_mc_del(dev, i->addr);
3386 case PACKET_MR_PROMISC:
3387 return dev_set_promiscuity(dev, what);
3388 case PACKET_MR_ALLMULTI:
3389 return dev_set_allmulti(dev, what);
3390 case PACKET_MR_UNICAST:
3391 if (i->alen != dev->addr_len)
3394 return dev_uc_add(dev, i->addr);
3396 return dev_uc_del(dev, i->addr);
3404 static void packet_dev_mclist_delete(struct net_device *dev,
3405 struct packet_mclist **mlp)
3407 struct packet_mclist *ml;
3409 while ((ml = *mlp) != NULL) {
3410 if (ml->ifindex == dev->ifindex) {
3411 packet_dev_mc(dev, ml, -1);
3419 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3421 struct packet_sock *po = pkt_sk(sk);
3422 struct packet_mclist *ml, *i;
3423 struct net_device *dev;
3429 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3434 if (mreq->mr_alen > dev->addr_len)
3438 i = kmalloc(sizeof(*i), GFP_KERNEL);
3443 for (ml = po->mclist; ml; ml = ml->next) {
3444 if (ml->ifindex == mreq->mr_ifindex &&
3445 ml->type == mreq->mr_type &&
3446 ml->alen == mreq->mr_alen &&
3447 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3449 /* Free the new element ... */
3455 i->type = mreq->mr_type;
3456 i->ifindex = mreq->mr_ifindex;
3457 i->alen = mreq->mr_alen;
3458 memcpy(i->addr, mreq->mr_address, i->alen);
3459 memset(i->addr + i->alen, 0, sizeof(i->addr) - i->alen);
3461 i->next = po->mclist;
3463 err = packet_dev_mc(dev, i, 1);
3465 po->mclist = i->next;
3474 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3476 struct packet_mclist *ml, **mlp;
3480 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3481 if (ml->ifindex == mreq->mr_ifindex &&
3482 ml->type == mreq->mr_type &&
3483 ml->alen == mreq->mr_alen &&
3484 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3485 if (--ml->count == 0) {
3486 struct net_device *dev;
3488 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3490 packet_dev_mc(dev, ml, -1);
3500 static void packet_flush_mclist(struct sock *sk)
3502 struct packet_sock *po = pkt_sk(sk);
3503 struct packet_mclist *ml;
3509 while ((ml = po->mclist) != NULL) {
3510 struct net_device *dev;
3512 po->mclist = ml->next;
3513 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3515 packet_dev_mc(dev, ml, -1);
3522 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3524 struct sock *sk = sock->sk;
3525 struct packet_sock *po = pkt_sk(sk);
3528 if (level != SOL_PACKET)
3529 return -ENOPROTOOPT;
3532 case PACKET_ADD_MEMBERSHIP:
3533 case PACKET_DROP_MEMBERSHIP:
3535 struct packet_mreq_max mreq;
3537 memset(&mreq, 0, sizeof(mreq));
3538 if (len < sizeof(struct packet_mreq))
3540 if (len > sizeof(mreq))
3542 if (copy_from_user(&mreq, optval, len))
3544 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3546 if (optname == PACKET_ADD_MEMBERSHIP)
3547 ret = packet_mc_add(sk, &mreq);
3549 ret = packet_mc_drop(sk, &mreq);
3553 case PACKET_RX_RING:
3554 case PACKET_TX_RING:
3556 union tpacket_req_u req_u;
3559 switch (po->tp_version) {
3562 len = sizeof(req_u.req);
3566 len = sizeof(req_u.req3);
3571 if (copy_from_user(&req_u.req, optval, len))
3573 return packet_set_ring(sk, &req_u, 0,
3574 optname == PACKET_TX_RING);
3576 case PACKET_COPY_THRESH:
3580 if (optlen != sizeof(val))
3582 if (copy_from_user(&val, optval, sizeof(val)))
3585 pkt_sk(sk)->copy_thresh = val;
3588 case PACKET_VERSION:
3592 if (optlen != sizeof(val))
3594 if (copy_from_user(&val, optval, sizeof(val)))
3605 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec) {
3608 po->tp_version = val;
3614 case PACKET_RESERVE:
3618 if (optlen != sizeof(val))
3620 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3622 if (copy_from_user(&val, optval, sizeof(val)))
3624 po->tp_reserve = val;
3631 if (optlen != sizeof(val))
3633 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3635 if (copy_from_user(&val, optval, sizeof(val)))
3637 po->tp_loss = !!val;
3640 case PACKET_AUXDATA:
3644 if (optlen < sizeof(val))
3646 if (copy_from_user(&val, optval, sizeof(val)))
3649 po->auxdata = !!val;
3652 case PACKET_ORIGDEV:
3656 if (optlen < sizeof(val))
3658 if (copy_from_user(&val, optval, sizeof(val)))
3661 po->origdev = !!val;
3664 case PACKET_VNET_HDR:
3668 if (sock->type != SOCK_RAW)
3670 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3672 if (optlen < sizeof(val))
3674 if (copy_from_user(&val, optval, sizeof(val)))
3677 po->has_vnet_hdr = !!val;
3680 case PACKET_TIMESTAMP:
3684 if (optlen != sizeof(val))
3686 if (copy_from_user(&val, optval, sizeof(val)))
3689 po->tp_tstamp = val;
3696 if (optlen != sizeof(val))
3698 if (copy_from_user(&val, optval, sizeof(val)))
3701 return fanout_add(sk, val & 0xffff, val >> 16);
3703 case PACKET_FANOUT_DATA:
3708 return fanout_set_data(po, optval, optlen);
3710 case PACKET_TX_HAS_OFF:
3714 if (optlen != sizeof(val))
3716 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3718 if (copy_from_user(&val, optval, sizeof(val)))
3720 po->tp_tx_has_off = !!val;
3723 case PACKET_QDISC_BYPASS:
3727 if (optlen != sizeof(val))
3729 if (copy_from_user(&val, optval, sizeof(val)))
3732 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3736 return -ENOPROTOOPT;
3740 static int packet_getsockopt(struct socket *sock, int level, int optname,
3741 char __user *optval, int __user *optlen)
3744 int val, lv = sizeof(val);
3745 struct sock *sk = sock->sk;
3746 struct packet_sock *po = pkt_sk(sk);
3748 union tpacket_stats_u st;
3749 struct tpacket_rollover_stats rstats;
3751 if (level != SOL_PACKET)
3752 return -ENOPROTOOPT;
3754 if (get_user(len, optlen))
3761 case PACKET_STATISTICS:
3762 spin_lock_bh(&sk->sk_receive_queue.lock);
3763 memcpy(&st, &po->stats, sizeof(st));
3764 memset(&po->stats, 0, sizeof(po->stats));
3765 spin_unlock_bh(&sk->sk_receive_queue.lock);
3767 if (po->tp_version == TPACKET_V3) {
3768 lv = sizeof(struct tpacket_stats_v3);
3769 st.stats3.tp_packets += st.stats3.tp_drops;
3772 lv = sizeof(struct tpacket_stats);
3773 st.stats1.tp_packets += st.stats1.tp_drops;
3778 case PACKET_AUXDATA:
3781 case PACKET_ORIGDEV:
3784 case PACKET_VNET_HDR:
3785 val = po->has_vnet_hdr;
3787 case PACKET_VERSION:
3788 val = po->tp_version;
3791 if (len > sizeof(int))
3793 if (copy_from_user(&val, optval, len))
3797 val = sizeof(struct tpacket_hdr);
3800 val = sizeof(struct tpacket2_hdr);
3803 val = sizeof(struct tpacket3_hdr);
3809 case PACKET_RESERVE:
3810 val = po->tp_reserve;
3815 case PACKET_TIMESTAMP:
3816 val = po->tp_tstamp;
3820 ((u32)po->fanout->id |
3821 ((u32)po->fanout->type << 16) |
3822 ((u32)po->fanout->flags << 24)) :
3825 case PACKET_ROLLOVER_STATS:
3828 rstats.tp_all = atomic_long_read(&po->rollover->num);
3829 rstats.tp_huge = atomic_long_read(&po->rollover->num_huge);
3830 rstats.tp_failed = atomic_long_read(&po->rollover->num_failed);
3832 lv = sizeof(rstats);
3834 case PACKET_TX_HAS_OFF:
3835 val = po->tp_tx_has_off;
3837 case PACKET_QDISC_BYPASS:
3838 val = packet_use_direct_xmit(po);
3841 return -ENOPROTOOPT;
3846 if (put_user(len, optlen))
3848 if (copy_to_user(optval, data, len))
3854 #ifdef CONFIG_COMPAT
3855 static int compat_packet_setsockopt(struct socket *sock, int level, int optname,
3856 char __user *optval, unsigned int optlen)
3858 struct packet_sock *po = pkt_sk(sock->sk);
3860 if (level != SOL_PACKET)
3861 return -ENOPROTOOPT;
3863 if (optname == PACKET_FANOUT_DATA &&
3864 po->fanout && po->fanout->type == PACKET_FANOUT_CBPF) {
3865 optval = (char __user *)get_compat_bpf_fprog(optval);
3868 optlen = sizeof(struct sock_fprog);
3871 return packet_setsockopt(sock, level, optname, optval, optlen);
3875 static int packet_notifier(struct notifier_block *this,
3876 unsigned long msg, void *ptr)
3879 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3880 struct net *net = dev_net(dev);
3883 sk_for_each_rcu(sk, &net->packet.sklist) {
3884 struct packet_sock *po = pkt_sk(sk);
3887 case NETDEV_UNREGISTER:
3889 packet_dev_mclist_delete(dev, &po->mclist);
3893 if (dev->ifindex == po->ifindex) {
3894 spin_lock(&po->bind_lock);
3896 __unregister_prot_hook(sk, false);
3897 sk->sk_err = ENETDOWN;
3898 if (!sock_flag(sk, SOCK_DEAD))
3899 sk->sk_error_report(sk);
3901 if (msg == NETDEV_UNREGISTER) {
3902 packet_cached_dev_reset(po);
3905 if (po->prot_hook.dev)
3906 dev_put(po->prot_hook.dev);
3907 po->prot_hook.dev = NULL;
3909 spin_unlock(&po->bind_lock);
3913 if (dev->ifindex == po->ifindex) {
3914 spin_lock(&po->bind_lock);
3916 register_prot_hook(sk);
3917 spin_unlock(&po->bind_lock);
3927 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3930 struct sock *sk = sock->sk;
3935 int amount = sk_wmem_alloc_get(sk);
3937 return put_user(amount, (int __user *)arg);
3941 struct sk_buff *skb;
3944 spin_lock_bh(&sk->sk_receive_queue.lock);
3945 skb = skb_peek(&sk->sk_receive_queue);
3948 spin_unlock_bh(&sk->sk_receive_queue.lock);
3949 return put_user(amount, (int __user *)arg);
3952 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3954 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3964 case SIOCGIFBRDADDR:
3965 case SIOCSIFBRDADDR:
3966 case SIOCGIFNETMASK:
3967 case SIOCSIFNETMASK:
3968 case SIOCGIFDSTADDR:
3969 case SIOCSIFDSTADDR:
3971 return inet_dgram_ops.ioctl(sock, cmd, arg);
3975 return -ENOIOCTLCMD;
3980 static unsigned int packet_poll(struct file *file, struct socket *sock,
3983 struct sock *sk = sock->sk;
3984 struct packet_sock *po = pkt_sk(sk);
3985 unsigned int mask = datagram_poll(file, sock, wait);
3987 spin_lock_bh(&sk->sk_receive_queue.lock);
3988 if (po->rx_ring.pg_vec) {
3989 if (!packet_previous_rx_frame(po, &po->rx_ring,
3991 mask |= POLLIN | POLLRDNORM;
3993 if (po->pressure && __packet_rcv_has_room(po, NULL) == ROOM_NORMAL)
3995 spin_unlock_bh(&sk->sk_receive_queue.lock);
3996 spin_lock_bh(&sk->sk_write_queue.lock);
3997 if (po->tx_ring.pg_vec) {
3998 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3999 mask |= POLLOUT | POLLWRNORM;
4001 spin_unlock_bh(&sk->sk_write_queue.lock);
4006 /* Dirty? Well, I still did not learn better way to account
4010 static void packet_mm_open(struct vm_area_struct *vma)
4012 struct file *file = vma->vm_file;
4013 struct socket *sock = file->private_data;
4014 struct sock *sk = sock->sk;
4017 atomic_inc(&pkt_sk(sk)->mapped);
4020 static void packet_mm_close(struct vm_area_struct *vma)
4022 struct file *file = vma->vm_file;
4023 struct socket *sock = file->private_data;
4024 struct sock *sk = sock->sk;
4027 atomic_dec(&pkt_sk(sk)->mapped);
4030 static const struct vm_operations_struct packet_mmap_ops = {
4031 .open = packet_mm_open,
4032 .close = packet_mm_close,
4035 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
4040 for (i = 0; i < len; i++) {
4041 if (likely(pg_vec[i].buffer)) {
4042 if (is_vmalloc_addr(pg_vec[i].buffer))
4043 vfree(pg_vec[i].buffer);
4045 free_pages((unsigned long)pg_vec[i].buffer,
4047 pg_vec[i].buffer = NULL;
4053 static char *alloc_one_pg_vec_page(unsigned long order)
4056 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
4057 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
4059 buffer = (char *) __get_free_pages(gfp_flags, order);
4063 /* __get_free_pages failed, fall back to vmalloc */
4064 buffer = vzalloc((1 << order) * PAGE_SIZE);
4068 /* vmalloc failed, lets dig into swap here */
4069 gfp_flags &= ~__GFP_NORETRY;
4070 buffer = (char *) __get_free_pages(gfp_flags, order);
4074 /* complete and utter failure */
4078 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
4080 unsigned int block_nr = req->tp_block_nr;
4084 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
4085 if (unlikely(!pg_vec))
4088 for (i = 0; i < block_nr; i++) {
4089 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
4090 if (unlikely(!pg_vec[i].buffer))
4091 goto out_free_pgvec;
4098 free_pg_vec(pg_vec, order, block_nr);
4103 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
4104 int closing, int tx_ring)
4106 struct pgv *pg_vec = NULL;
4107 struct packet_sock *po = pkt_sk(sk);
4108 int was_running, order = 0;
4109 struct packet_ring_buffer *rb;
4110 struct sk_buff_head *rb_queue;
4113 /* Added to avoid minimal code churn */
4114 struct tpacket_req *req = &req_u->req;
4117 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
4118 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
4119 net_warn_ratelimited("Tx-ring is not supported.\n");
4123 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
4124 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
4128 if (atomic_read(&po->mapped))
4130 if (packet_read_pending(rb))
4134 if (req->tp_block_nr) {
4135 /* Sanity tests and some calculations */
4137 if (unlikely(rb->pg_vec))
4140 switch (po->tp_version) {
4142 po->tp_hdrlen = TPACKET_HDRLEN;
4145 po->tp_hdrlen = TPACKET2_HDRLEN;
4148 po->tp_hdrlen = TPACKET3_HDRLEN;
4153 if (unlikely((int)req->tp_block_size <= 0))
4155 if (unlikely(!PAGE_ALIGNED(req->tp_block_size)))
4157 if (po->tp_version >= TPACKET_V3 &&
4158 (int)(req->tp_block_size -
4159 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
4161 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
4164 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
4167 rb->frames_per_block = req->tp_block_size / req->tp_frame_size;
4168 if (unlikely(rb->frames_per_block == 0))
4170 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
4175 order = get_order(req->tp_block_size);
4176 pg_vec = alloc_pg_vec(req, order);
4177 if (unlikely(!pg_vec))
4179 switch (po->tp_version) {
4181 /* Transmit path is not supported. We checked
4182 * it above but just being paranoid
4185 init_prb_bdqc(po, rb, pg_vec, req_u);
4194 if (unlikely(req->tp_frame_nr))
4199 /* Detach socket from network */
4200 spin_lock(&po->bind_lock);
4201 was_running = po->running;
4205 __unregister_prot_hook(sk, false);
4207 spin_unlock(&po->bind_lock);
4212 mutex_lock(&po->pg_vec_lock);
4213 if (closing || atomic_read(&po->mapped) == 0) {
4215 spin_lock_bh(&rb_queue->lock);
4216 swap(rb->pg_vec, pg_vec);
4217 rb->frame_max = (req->tp_frame_nr - 1);
4219 rb->frame_size = req->tp_frame_size;
4220 spin_unlock_bh(&rb_queue->lock);
4222 swap(rb->pg_vec_order, order);
4223 swap(rb->pg_vec_len, req->tp_block_nr);
4225 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
4226 po->prot_hook.func = (po->rx_ring.pg_vec) ?
4227 tpacket_rcv : packet_rcv;
4228 skb_queue_purge(rb_queue);
4229 if (atomic_read(&po->mapped))
4230 pr_err("packet_mmap: vma is busy: %d\n",
4231 atomic_read(&po->mapped));
4233 mutex_unlock(&po->pg_vec_lock);
4235 spin_lock(&po->bind_lock);
4238 register_prot_hook(sk);
4240 spin_unlock(&po->bind_lock);
4241 if (closing && (po->tp_version > TPACKET_V2)) {
4242 /* Because we don't support block-based V3 on tx-ring */
4244 prb_shutdown_retire_blk_timer(po, rb_queue);
4248 free_pg_vec(pg_vec, order, req->tp_block_nr);
4254 static int packet_mmap(struct file *file, struct socket *sock,
4255 struct vm_area_struct *vma)
4257 struct sock *sk = sock->sk;
4258 struct packet_sock *po = pkt_sk(sk);
4259 unsigned long size, expected_size;
4260 struct packet_ring_buffer *rb;
4261 unsigned long start;
4268 mutex_lock(&po->pg_vec_lock);
4271 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4273 expected_size += rb->pg_vec_len
4279 if (expected_size == 0)
4282 size = vma->vm_end - vma->vm_start;
4283 if (size != expected_size)
4286 start = vma->vm_start;
4287 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4288 if (rb->pg_vec == NULL)
4291 for (i = 0; i < rb->pg_vec_len; i++) {
4293 void *kaddr = rb->pg_vec[i].buffer;
4296 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4297 page = pgv_to_page(kaddr);
4298 err = vm_insert_page(vma, start, page);
4307 atomic_inc(&po->mapped);
4308 vma->vm_ops = &packet_mmap_ops;
4312 mutex_unlock(&po->pg_vec_lock);
4316 static const struct proto_ops packet_ops_spkt = {
4317 .family = PF_PACKET,
4318 .owner = THIS_MODULE,
4319 .release = packet_release,
4320 .bind = packet_bind_spkt,
4321 .connect = sock_no_connect,
4322 .socketpair = sock_no_socketpair,
4323 .accept = sock_no_accept,
4324 .getname = packet_getname_spkt,
4325 .poll = datagram_poll,
4326 .ioctl = packet_ioctl,
4327 .listen = sock_no_listen,
4328 .shutdown = sock_no_shutdown,
4329 .setsockopt = sock_no_setsockopt,
4330 .getsockopt = sock_no_getsockopt,
4331 .sendmsg = packet_sendmsg_spkt,
4332 .recvmsg = packet_recvmsg,
4333 .mmap = sock_no_mmap,
4334 .sendpage = sock_no_sendpage,
4337 static const struct proto_ops packet_ops = {
4338 .family = PF_PACKET,
4339 .owner = THIS_MODULE,
4340 .release = packet_release,
4341 .bind = packet_bind,
4342 .connect = sock_no_connect,
4343 .socketpair = sock_no_socketpair,
4344 .accept = sock_no_accept,
4345 .getname = packet_getname,
4346 .poll = packet_poll,
4347 .ioctl = packet_ioctl,
4348 .listen = sock_no_listen,
4349 .shutdown = sock_no_shutdown,
4350 .setsockopt = packet_setsockopt,
4351 .getsockopt = packet_getsockopt,
4352 #ifdef CONFIG_COMPAT
4353 .compat_setsockopt = compat_packet_setsockopt,
4355 .sendmsg = packet_sendmsg,
4356 .recvmsg = packet_recvmsg,
4357 .mmap = packet_mmap,
4358 .sendpage = sock_no_sendpage,
4361 static const struct net_proto_family packet_family_ops = {
4362 .family = PF_PACKET,
4363 .create = packet_create,
4364 .owner = THIS_MODULE,
4367 static struct notifier_block packet_netdev_notifier = {
4368 .notifier_call = packet_notifier,
4371 #ifdef CONFIG_PROC_FS
4373 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4376 struct net *net = seq_file_net(seq);
4379 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4382 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4384 struct net *net = seq_file_net(seq);
4385 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4388 static void packet_seq_stop(struct seq_file *seq, void *v)
4394 static int packet_seq_show(struct seq_file *seq, void *v)
4396 if (v == SEQ_START_TOKEN)
4397 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4399 struct sock *s = sk_entry(v);
4400 const struct packet_sock *po = pkt_sk(s);
4403 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4405 atomic_read(&s->sk_refcnt),
4410 atomic_read(&s->sk_rmem_alloc),
4411 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4418 static const struct seq_operations packet_seq_ops = {
4419 .start = packet_seq_start,
4420 .next = packet_seq_next,
4421 .stop = packet_seq_stop,
4422 .show = packet_seq_show,
4425 static int packet_seq_open(struct inode *inode, struct file *file)
4427 return seq_open_net(inode, file, &packet_seq_ops,
4428 sizeof(struct seq_net_private));
4431 static const struct file_operations packet_seq_fops = {
4432 .owner = THIS_MODULE,
4433 .open = packet_seq_open,
4435 .llseek = seq_lseek,
4436 .release = seq_release_net,
4441 static int __net_init packet_net_init(struct net *net)
4443 mutex_init(&net->packet.sklist_lock);
4444 INIT_HLIST_HEAD(&net->packet.sklist);
4446 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4452 static void __net_exit packet_net_exit(struct net *net)
4454 remove_proc_entry("packet", net->proc_net);
4457 static struct pernet_operations packet_net_ops = {
4458 .init = packet_net_init,
4459 .exit = packet_net_exit,
4463 static void __exit packet_exit(void)
4465 unregister_netdevice_notifier(&packet_netdev_notifier);
4466 unregister_pernet_subsys(&packet_net_ops);
4467 sock_unregister(PF_PACKET);
4468 proto_unregister(&packet_proto);
4471 static int __init packet_init(void)
4473 int rc = proto_register(&packet_proto, 0);
4478 sock_register(&packet_family_ops);
4479 register_pernet_subsys(&packet_net_ops);
4480 register_netdevice_notifier(&packet_netdev_notifier);
4485 module_init(packet_init);
4486 module_exit(packet_exit);
4487 MODULE_LICENSE("GPL");
4488 MODULE_ALIAS_NETPROTO(PF_PACKET);