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 <asm/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>
100 - if device has no dev->hard_header routine, it adds and removes ll header
101 inside itself. In this case ll header is invisible outside of device,
102 but higher levels still should reserve dev->hard_header_len.
103 Some devices are enough clever to reallocate skb, when header
104 will not fit to reserved space (tunnel), another ones are silly
106 - packet socket receives packets with pulled ll header,
107 so that SOCK_RAW should push it back.
112 Incoming, dev->hard_header!=NULL
113 mac_header -> ll header
116 Outgoing, dev->hard_header!=NULL
117 mac_header -> ll header
120 Incoming, dev->hard_header==NULL
121 mac_header -> UNKNOWN position. It is very likely, that it points to ll
122 header. PPP makes it, that is wrong, because introduce
123 assymetry between rx and tx paths.
126 Outgoing, dev->hard_header==NULL
127 mac_header -> data. ll header is still not built!
131 If dev->hard_header==NULL we are unlikely to restore sensible ll header.
137 dev->hard_header != NULL
138 mac_header -> ll header
141 dev->hard_header == NULL (ll header is added by device, we cannot control it)
145 We should set nh.raw on output to correct posistion,
146 packet classifier depends on it.
149 /* Private packet socket structures. */
151 /* identical to struct packet_mreq except it has
152 * a longer address field.
154 struct packet_mreq_max {
156 unsigned short mr_type;
157 unsigned short mr_alen;
158 unsigned char mr_address[MAX_ADDR_LEN];
162 struct tpacket_hdr *h1;
163 struct tpacket2_hdr *h2;
164 struct tpacket3_hdr *h3;
168 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
169 int closing, int tx_ring);
171 #define V3_ALIGNMENT (8)
173 #define BLK_HDR_LEN (ALIGN(sizeof(struct tpacket_block_desc), V3_ALIGNMENT))
175 #define BLK_PLUS_PRIV(sz_of_priv) \
176 (BLK_HDR_LEN + ALIGN((sz_of_priv), V3_ALIGNMENT))
178 #define PGV_FROM_VMALLOC 1
180 #define BLOCK_STATUS(x) ((x)->hdr.bh1.block_status)
181 #define BLOCK_NUM_PKTS(x) ((x)->hdr.bh1.num_pkts)
182 #define BLOCK_O2FP(x) ((x)->hdr.bh1.offset_to_first_pkt)
183 #define BLOCK_LEN(x) ((x)->hdr.bh1.blk_len)
184 #define BLOCK_SNUM(x) ((x)->hdr.bh1.seq_num)
185 #define BLOCK_O2PRIV(x) ((x)->offset_to_priv)
186 #define BLOCK_PRIV(x) ((void *)((char *)(x) + BLOCK_O2PRIV(x)))
189 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg);
190 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
191 struct packet_type *pt, struct net_device *orig_dev);
193 static void *packet_previous_frame(struct packet_sock *po,
194 struct packet_ring_buffer *rb,
196 static void packet_increment_head(struct packet_ring_buffer *buff);
197 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *,
198 struct tpacket_block_desc *);
199 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *,
200 struct packet_sock *);
201 static void prb_retire_current_block(struct tpacket_kbdq_core *,
202 struct packet_sock *, unsigned int status);
203 static int prb_queue_frozen(struct tpacket_kbdq_core *);
204 static void prb_open_block(struct tpacket_kbdq_core *,
205 struct tpacket_block_desc *);
206 static void prb_retire_rx_blk_timer_expired(unsigned long);
207 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *);
208 static void prb_init_blk_timer(struct packet_sock *,
209 struct tpacket_kbdq_core *,
210 void (*func) (unsigned long));
211 static void prb_fill_rxhash(struct tpacket_kbdq_core *, struct tpacket3_hdr *);
212 static void prb_clear_rxhash(struct tpacket_kbdq_core *,
213 struct tpacket3_hdr *);
214 static void prb_fill_vlan_info(struct tpacket_kbdq_core *,
215 struct tpacket3_hdr *);
216 static void packet_flush_mclist(struct sock *sk);
218 struct packet_skb_cb {
220 struct sockaddr_pkt pkt;
222 /* Trick: alias skb original length with
223 * ll.sll_family and ll.protocol in order
226 unsigned int origlen;
227 struct sockaddr_ll ll;
232 #define PACKET_SKB_CB(__skb) ((struct packet_skb_cb *)((__skb)->cb))
234 #define GET_PBDQC_FROM_RB(x) ((struct tpacket_kbdq_core *)(&(x)->prb_bdqc))
235 #define GET_PBLOCK_DESC(x, bid) \
236 ((struct tpacket_block_desc *)((x)->pkbdq[(bid)].buffer))
237 #define GET_CURR_PBLOCK_DESC_FROM_CORE(x) \
238 ((struct tpacket_block_desc *)((x)->pkbdq[(x)->kactive_blk_num].buffer))
239 #define GET_NEXT_PRB_BLK_NUM(x) \
240 (((x)->kactive_blk_num < ((x)->knum_blocks-1)) ? \
241 ((x)->kactive_blk_num+1) : 0)
243 static void __fanout_unlink(struct sock *sk, struct packet_sock *po);
244 static void __fanout_link(struct sock *sk, struct packet_sock *po);
246 static int packet_direct_xmit(struct sk_buff *skb)
248 struct net_device *dev = skb->dev;
249 netdev_features_t features;
250 struct netdev_queue *txq;
251 int ret = NETDEV_TX_BUSY;
253 if (unlikely(!netif_running(dev) ||
254 !netif_carrier_ok(dev)))
257 features = netif_skb_features(skb);
258 if (skb_needs_linearize(skb, features) &&
259 __skb_linearize(skb))
262 txq = skb_get_tx_queue(dev, skb);
266 HARD_TX_LOCK(dev, txq, smp_processor_id());
267 if (!netif_xmit_frozen_or_drv_stopped(txq))
268 ret = netdev_start_xmit(skb, dev, txq, false);
269 HARD_TX_UNLOCK(dev, txq);
273 if (!dev_xmit_complete(ret))
278 atomic_long_inc(&dev->tx_dropped);
280 return NET_XMIT_DROP;
283 static struct net_device *packet_cached_dev_get(struct packet_sock *po)
285 struct net_device *dev;
288 dev = rcu_dereference(po->cached_dev);
296 static void packet_cached_dev_assign(struct packet_sock *po,
297 struct net_device *dev)
299 rcu_assign_pointer(po->cached_dev, dev);
302 static void packet_cached_dev_reset(struct packet_sock *po)
304 RCU_INIT_POINTER(po->cached_dev, NULL);
307 static bool packet_use_direct_xmit(const struct packet_sock *po)
309 return po->xmit == packet_direct_xmit;
312 static u16 __packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
314 return (u16) raw_smp_processor_id() % dev->real_num_tx_queues;
317 static void packet_pick_tx_queue(struct net_device *dev, struct sk_buff *skb)
319 const struct net_device_ops *ops = dev->netdev_ops;
322 if (ops->ndo_select_queue) {
323 queue_index = ops->ndo_select_queue(dev, skb, NULL,
324 __packet_pick_tx_queue);
325 queue_index = netdev_cap_txqueue(dev, queue_index);
327 queue_index = __packet_pick_tx_queue(dev, skb);
330 skb_set_queue_mapping(skb, queue_index);
333 /* register_prot_hook must be invoked with the po->bind_lock held,
334 * or from a context in which asynchronous accesses to the packet
335 * socket is not possible (packet_create()).
337 static void register_prot_hook(struct sock *sk)
339 struct packet_sock *po = pkt_sk(sk);
343 __fanout_link(sk, po);
345 dev_add_pack(&po->prot_hook);
352 /* {,__}unregister_prot_hook() must be invoked with the po->bind_lock
353 * held. If the sync parameter is true, we will temporarily drop
354 * the po->bind_lock and do a synchronize_net to make sure no
355 * asynchronous packet processing paths still refer to the elements
356 * of po->prot_hook. If the sync parameter is false, it is the
357 * callers responsibility to take care of this.
359 static void __unregister_prot_hook(struct sock *sk, bool sync)
361 struct packet_sock *po = pkt_sk(sk);
366 __fanout_unlink(sk, po);
368 __dev_remove_pack(&po->prot_hook);
373 spin_unlock(&po->bind_lock);
375 spin_lock(&po->bind_lock);
379 static void unregister_prot_hook(struct sock *sk, bool sync)
381 struct packet_sock *po = pkt_sk(sk);
384 __unregister_prot_hook(sk, sync);
387 static inline struct page * __pure pgv_to_page(void *addr)
389 if (is_vmalloc_addr(addr))
390 return vmalloc_to_page(addr);
391 return virt_to_page(addr);
394 static void __packet_set_status(struct packet_sock *po, void *frame, int status)
396 union tpacket_uhdr h;
399 switch (po->tp_version) {
401 h.h1->tp_status = status;
402 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
405 h.h2->tp_status = status;
406 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
410 WARN(1, "TPACKET version not supported.\n");
417 static int __packet_get_status(struct packet_sock *po, void *frame)
419 union tpacket_uhdr h;
424 switch (po->tp_version) {
426 flush_dcache_page(pgv_to_page(&h.h1->tp_status));
427 return h.h1->tp_status;
429 flush_dcache_page(pgv_to_page(&h.h2->tp_status));
430 return h.h2->tp_status;
433 WARN(1, "TPACKET version not supported.\n");
439 static __u32 tpacket_get_timestamp(struct sk_buff *skb, struct timespec *ts,
442 struct skb_shared_hwtstamps *shhwtstamps = skb_hwtstamps(skb);
445 (flags & SOF_TIMESTAMPING_RAW_HARDWARE) &&
446 ktime_to_timespec_cond(shhwtstamps->hwtstamp, ts))
447 return TP_STATUS_TS_RAW_HARDWARE;
449 if (ktime_to_timespec_cond(skb->tstamp, ts))
450 return TP_STATUS_TS_SOFTWARE;
455 static __u32 __packet_set_timestamp(struct packet_sock *po, void *frame,
458 union tpacket_uhdr h;
462 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
466 switch (po->tp_version) {
468 h.h1->tp_sec = ts.tv_sec;
469 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
472 h.h2->tp_sec = ts.tv_sec;
473 h.h2->tp_nsec = ts.tv_nsec;
477 WARN(1, "TPACKET version not supported.\n");
481 /* one flush is safe, as both fields always lie on the same cacheline */
482 flush_dcache_page(pgv_to_page(&h.h1->tp_sec));
488 static void *packet_lookup_frame(struct packet_sock *po,
489 struct packet_ring_buffer *rb,
490 unsigned int position,
493 unsigned int pg_vec_pos, frame_offset;
494 union tpacket_uhdr h;
496 pg_vec_pos = position / rb->frames_per_block;
497 frame_offset = position % rb->frames_per_block;
499 h.raw = rb->pg_vec[pg_vec_pos].buffer +
500 (frame_offset * rb->frame_size);
502 if (status != __packet_get_status(po, h.raw))
508 static void *packet_current_frame(struct packet_sock *po,
509 struct packet_ring_buffer *rb,
512 return packet_lookup_frame(po, rb, rb->head, status);
515 static void prb_del_retire_blk_timer(struct tpacket_kbdq_core *pkc)
517 del_timer_sync(&pkc->retire_blk_timer);
520 static void prb_shutdown_retire_blk_timer(struct packet_sock *po,
522 struct sk_buff_head *rb_queue)
524 struct tpacket_kbdq_core *pkc;
526 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
527 GET_PBDQC_FROM_RB(&po->rx_ring);
529 spin_lock_bh(&rb_queue->lock);
530 pkc->delete_blk_timer = 1;
531 spin_unlock_bh(&rb_queue->lock);
533 prb_del_retire_blk_timer(pkc);
536 static void prb_init_blk_timer(struct packet_sock *po,
537 struct tpacket_kbdq_core *pkc,
538 void (*func) (unsigned long))
540 init_timer(&pkc->retire_blk_timer);
541 pkc->retire_blk_timer.data = (long)po;
542 pkc->retire_blk_timer.function = func;
543 pkc->retire_blk_timer.expires = jiffies;
546 static void prb_setup_retire_blk_timer(struct packet_sock *po, int tx_ring)
548 struct tpacket_kbdq_core *pkc;
553 pkc = tx_ring ? GET_PBDQC_FROM_RB(&po->tx_ring) :
554 GET_PBDQC_FROM_RB(&po->rx_ring);
555 prb_init_blk_timer(po, pkc, prb_retire_rx_blk_timer_expired);
558 static int prb_calc_retire_blk_tmo(struct packet_sock *po,
559 int blk_size_in_bytes)
561 struct net_device *dev;
562 unsigned int mbits = 0, msec = 0, div = 0, tmo = 0;
563 struct ethtool_cmd ecmd;
568 dev = __dev_get_by_index(sock_net(&po->sk), po->ifindex);
569 if (unlikely(!dev)) {
571 return DEFAULT_PRB_RETIRE_TOV;
573 err = __ethtool_get_settings(dev, &ecmd);
574 speed = ethtool_cmd_speed(&ecmd);
578 * If the link speed is so slow you don't really
579 * need to worry about perf anyways
581 if (speed < SPEED_1000 || speed == SPEED_UNKNOWN) {
582 return DEFAULT_PRB_RETIRE_TOV;
589 mbits = (blk_size_in_bytes * 8) / (1024 * 1024);
601 static void prb_init_ft_ops(struct tpacket_kbdq_core *p1,
602 union tpacket_req_u *req_u)
604 p1->feature_req_word = req_u->req3.tp_feature_req_word;
607 static void init_prb_bdqc(struct packet_sock *po,
608 struct packet_ring_buffer *rb,
610 union tpacket_req_u *req_u, int tx_ring)
612 struct tpacket_kbdq_core *p1 = GET_PBDQC_FROM_RB(rb);
613 struct tpacket_block_desc *pbd;
615 memset(p1, 0x0, sizeof(*p1));
617 p1->knxt_seq_num = 1;
619 pbd = (struct tpacket_block_desc *)pg_vec[0].buffer;
620 p1->pkblk_start = pg_vec[0].buffer;
621 p1->kblk_size = req_u->req3.tp_block_size;
622 p1->knum_blocks = req_u->req3.tp_block_nr;
623 p1->hdrlen = po->tp_hdrlen;
624 p1->version = po->tp_version;
625 p1->last_kactive_blk_num = 0;
626 po->stats.stats3.tp_freeze_q_cnt = 0;
627 if (req_u->req3.tp_retire_blk_tov)
628 p1->retire_blk_tov = req_u->req3.tp_retire_blk_tov;
630 p1->retire_blk_tov = prb_calc_retire_blk_tmo(po,
631 req_u->req3.tp_block_size);
632 p1->tov_in_jiffies = msecs_to_jiffies(p1->retire_blk_tov);
633 p1->blk_sizeof_priv = req_u->req3.tp_sizeof_priv;
635 p1->max_frame_len = p1->kblk_size - BLK_PLUS_PRIV(p1->blk_sizeof_priv);
636 prb_init_ft_ops(p1, req_u);
637 prb_setup_retire_blk_timer(po, tx_ring);
638 prb_open_block(p1, pbd);
641 /* Do NOT update the last_blk_num first.
642 * Assumes sk_buff_head lock is held.
644 static void _prb_refresh_rx_retire_blk_timer(struct tpacket_kbdq_core *pkc)
646 mod_timer(&pkc->retire_blk_timer,
647 jiffies + pkc->tov_in_jiffies);
648 pkc->last_kactive_blk_num = pkc->kactive_blk_num;
653 * 1) We refresh the timer only when we open a block.
654 * By doing this we don't waste cycles refreshing the timer
655 * on packet-by-packet basis.
657 * With a 1MB block-size, on a 1Gbps line, it will take
658 * i) ~8 ms to fill a block + ii) memcpy etc.
659 * In this cut we are not accounting for the memcpy time.
661 * So, if the user sets the 'tmo' to 10ms then the timer
662 * will never fire while the block is still getting filled
663 * (which is what we want). However, the user could choose
664 * to close a block early and that's fine.
666 * But when the timer does fire, we check whether or not to refresh it.
667 * Since the tmo granularity is in msecs, it is not too expensive
668 * to refresh the timer, lets say every '8' msecs.
669 * Either the user can set the 'tmo' or we can derive it based on
670 * a) line-speed and b) block-size.
671 * prb_calc_retire_blk_tmo() calculates the tmo.
674 static void prb_retire_rx_blk_timer_expired(unsigned long data)
676 struct packet_sock *po = (struct packet_sock *)data;
677 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
679 struct tpacket_block_desc *pbd;
681 spin_lock(&po->sk.sk_receive_queue.lock);
683 frozen = prb_queue_frozen(pkc);
684 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
686 if (unlikely(pkc->delete_blk_timer))
689 /* We only need to plug the race when the block is partially filled.
691 * lock(); increment BLOCK_NUM_PKTS; unlock()
692 * copy_bits() is in progress ...
693 * timer fires on other cpu:
694 * we can't retire the current block because copy_bits
698 if (BLOCK_NUM_PKTS(pbd)) {
699 while (atomic_read(&pkc->blk_fill_in_prog)) {
700 /* Waiting for skb_copy_bits to finish... */
705 if (pkc->last_kactive_blk_num == pkc->kactive_blk_num) {
707 if (!BLOCK_NUM_PKTS(pbd)) {
708 /* An empty block. Just refresh the timer. */
711 prb_retire_current_block(pkc, po, TP_STATUS_BLK_TMO);
712 if (!prb_dispatch_next_block(pkc, po))
717 /* Case 1. Queue was frozen because user-space was
720 if (prb_curr_blk_in_use(pkc, pbd)) {
722 * Ok, user-space is still behind.
723 * So just refresh the timer.
727 /* Case 2. queue was frozen,user-space caught up,
728 * now the link went idle && the timer fired.
729 * We don't have a block to close.So we open this
730 * block and restart the timer.
731 * opening a block thaws the queue,restarts timer
732 * Thawing/timer-refresh is a side effect.
734 prb_open_block(pkc, pbd);
741 _prb_refresh_rx_retire_blk_timer(pkc);
744 spin_unlock(&po->sk.sk_receive_queue.lock);
747 static void prb_flush_block(struct tpacket_kbdq_core *pkc1,
748 struct tpacket_block_desc *pbd1, __u32 status)
750 /* Flush everything minus the block header */
752 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
757 /* Skip the block header(we know header WILL fit in 4K) */
760 end = (u8 *)PAGE_ALIGN((unsigned long)pkc1->pkblk_end);
761 for (; start < end; start += PAGE_SIZE)
762 flush_dcache_page(pgv_to_page(start));
767 /* Now update the block status. */
769 BLOCK_STATUS(pbd1) = status;
771 /* Flush the block header */
773 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
775 flush_dcache_page(pgv_to_page(start));
785 * 2) Increment active_blk_num
787 * Note:We DONT refresh the timer on purpose.
788 * Because almost always the next block will be opened.
790 static void prb_close_block(struct tpacket_kbdq_core *pkc1,
791 struct tpacket_block_desc *pbd1,
792 struct packet_sock *po, unsigned int stat)
794 __u32 status = TP_STATUS_USER | stat;
796 struct tpacket3_hdr *last_pkt;
797 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
798 struct sock *sk = &po->sk;
800 if (po->stats.stats3.tp_drops)
801 status |= TP_STATUS_LOSING;
803 last_pkt = (struct tpacket3_hdr *)pkc1->prev;
804 last_pkt->tp_next_offset = 0;
806 /* Get the ts of the last pkt */
807 if (BLOCK_NUM_PKTS(pbd1)) {
808 h1->ts_last_pkt.ts_sec = last_pkt->tp_sec;
809 h1->ts_last_pkt.ts_nsec = last_pkt->tp_nsec;
811 /* Ok, we tmo'd - so get the current time.
813 * It shouldn't really happen as we don't close empty
814 * blocks. See prb_retire_rx_blk_timer_expired().
818 h1->ts_last_pkt.ts_sec = ts.tv_sec;
819 h1->ts_last_pkt.ts_nsec = ts.tv_nsec;
824 /* Flush the block */
825 prb_flush_block(pkc1, pbd1, status);
827 sk->sk_data_ready(sk);
829 pkc1->kactive_blk_num = GET_NEXT_PRB_BLK_NUM(pkc1);
832 static void prb_thaw_queue(struct tpacket_kbdq_core *pkc)
834 pkc->reset_pending_on_curr_blk = 0;
838 * Side effect of opening a block:
840 * 1) prb_queue is thawed.
841 * 2) retire_blk_timer is refreshed.
844 static void prb_open_block(struct tpacket_kbdq_core *pkc1,
845 struct tpacket_block_desc *pbd1)
848 struct tpacket_hdr_v1 *h1 = &pbd1->hdr.bh1;
852 /* We could have just memset this but we will lose the
853 * flexibility of making the priv area sticky
856 BLOCK_SNUM(pbd1) = pkc1->knxt_seq_num++;
857 BLOCK_NUM_PKTS(pbd1) = 0;
858 BLOCK_LEN(pbd1) = BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
862 h1->ts_first_pkt.ts_sec = ts.tv_sec;
863 h1->ts_first_pkt.ts_nsec = ts.tv_nsec;
865 pkc1->pkblk_start = (char *)pbd1;
866 pkc1->nxt_offset = pkc1->pkblk_start + BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
868 BLOCK_O2FP(pbd1) = (__u32)BLK_PLUS_PRIV(pkc1->blk_sizeof_priv);
869 BLOCK_O2PRIV(pbd1) = BLK_HDR_LEN;
871 pbd1->version = pkc1->version;
872 pkc1->prev = pkc1->nxt_offset;
873 pkc1->pkblk_end = pkc1->pkblk_start + pkc1->kblk_size;
875 prb_thaw_queue(pkc1);
876 _prb_refresh_rx_retire_blk_timer(pkc1);
882 * Queue freeze logic:
883 * 1) Assume tp_block_nr = 8 blocks.
884 * 2) At time 't0', user opens Rx ring.
885 * 3) Some time past 't0', kernel starts filling blocks starting from 0 .. 7
886 * 4) user-space is either sleeping or processing block '0'.
887 * 5) tpacket_rcv is currently filling block '7', since there is no space left,
888 * it will close block-7,loop around and try to fill block '0'.
890 * __packet_lookup_frame_in_block
891 * prb_retire_current_block()
892 * prb_dispatch_next_block()
893 * |->(BLOCK_STATUS == USER) evaluates to true
894 * 5.1) Since block-0 is currently in-use, we just freeze the queue.
895 * 6) Now there are two cases:
896 * 6.1) Link goes idle right after the queue is frozen.
897 * But remember, the last open_block() refreshed the timer.
898 * When this timer expires,it will refresh itself so that we can
899 * re-open block-0 in near future.
900 * 6.2) Link is busy and keeps on receiving packets. This is a simple
901 * case and __packet_lookup_frame_in_block will check if block-0
902 * is free and can now be re-used.
904 static void prb_freeze_queue(struct tpacket_kbdq_core *pkc,
905 struct packet_sock *po)
907 pkc->reset_pending_on_curr_blk = 1;
908 po->stats.stats3.tp_freeze_q_cnt++;
911 #define TOTAL_PKT_LEN_INCL_ALIGN(length) (ALIGN((length), V3_ALIGNMENT))
914 * If the next block is free then we will dispatch it
915 * and return a good offset.
916 * Else, we will freeze the queue.
917 * So, caller must check the return value.
919 static void *prb_dispatch_next_block(struct tpacket_kbdq_core *pkc,
920 struct packet_sock *po)
922 struct tpacket_block_desc *pbd;
926 /* 1. Get current block num */
927 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
929 /* 2. If this block is currently in_use then freeze the queue */
930 if (TP_STATUS_USER & BLOCK_STATUS(pbd)) {
931 prb_freeze_queue(pkc, po);
937 * open this block and return the offset where the first packet
938 * needs to get stored.
940 prb_open_block(pkc, pbd);
941 return (void *)pkc->nxt_offset;
944 static void prb_retire_current_block(struct tpacket_kbdq_core *pkc,
945 struct packet_sock *po, unsigned int status)
947 struct tpacket_block_desc *pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
949 /* retire/close the current block */
950 if (likely(TP_STATUS_KERNEL == BLOCK_STATUS(pbd))) {
952 * Plug the case where copy_bits() is in progress on
953 * cpu-0 and tpacket_rcv() got invoked on cpu-1, didn't
954 * have space to copy the pkt in the current block and
955 * called prb_retire_current_block()
957 * We don't need to worry about the TMO case because
958 * the timer-handler already handled this case.
960 if (!(status & TP_STATUS_BLK_TMO)) {
961 while (atomic_read(&pkc->blk_fill_in_prog)) {
962 /* Waiting for skb_copy_bits to finish... */
966 prb_close_block(pkc, pbd, po, status);
971 static int prb_curr_blk_in_use(struct tpacket_kbdq_core *pkc,
972 struct tpacket_block_desc *pbd)
974 return TP_STATUS_USER & BLOCK_STATUS(pbd);
977 static int prb_queue_frozen(struct tpacket_kbdq_core *pkc)
979 return pkc->reset_pending_on_curr_blk;
982 static void prb_clear_blk_fill_status(struct packet_ring_buffer *rb)
984 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
985 atomic_dec(&pkc->blk_fill_in_prog);
988 static void prb_fill_rxhash(struct tpacket_kbdq_core *pkc,
989 struct tpacket3_hdr *ppd)
991 ppd->hv1.tp_rxhash = skb_get_hash(pkc->skb);
994 static void prb_clear_rxhash(struct tpacket_kbdq_core *pkc,
995 struct tpacket3_hdr *ppd)
997 ppd->hv1.tp_rxhash = 0;
1000 static void prb_fill_vlan_info(struct tpacket_kbdq_core *pkc,
1001 struct tpacket3_hdr *ppd)
1003 if (skb_vlan_tag_present(pkc->skb)) {
1004 ppd->hv1.tp_vlan_tci = skb_vlan_tag_get(pkc->skb);
1005 ppd->hv1.tp_vlan_tpid = ntohs(pkc->skb->vlan_proto);
1006 ppd->tp_status = TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
1008 ppd->hv1.tp_vlan_tci = 0;
1009 ppd->hv1.tp_vlan_tpid = 0;
1010 ppd->tp_status = TP_STATUS_AVAILABLE;
1014 static void prb_run_all_ft_ops(struct tpacket_kbdq_core *pkc,
1015 struct tpacket3_hdr *ppd)
1017 ppd->hv1.tp_padding = 0;
1018 prb_fill_vlan_info(pkc, ppd);
1020 if (pkc->feature_req_word & TP_FT_REQ_FILL_RXHASH)
1021 prb_fill_rxhash(pkc, ppd);
1023 prb_clear_rxhash(pkc, ppd);
1026 static void prb_fill_curr_block(char *curr,
1027 struct tpacket_kbdq_core *pkc,
1028 struct tpacket_block_desc *pbd,
1031 struct tpacket3_hdr *ppd;
1033 ppd = (struct tpacket3_hdr *)curr;
1034 ppd->tp_next_offset = TOTAL_PKT_LEN_INCL_ALIGN(len);
1036 pkc->nxt_offset += TOTAL_PKT_LEN_INCL_ALIGN(len);
1037 BLOCK_LEN(pbd) += TOTAL_PKT_LEN_INCL_ALIGN(len);
1038 BLOCK_NUM_PKTS(pbd) += 1;
1039 atomic_inc(&pkc->blk_fill_in_prog);
1040 prb_run_all_ft_ops(pkc, ppd);
1043 /* Assumes caller has the sk->rx_queue.lock */
1044 static void *__packet_lookup_frame_in_block(struct packet_sock *po,
1045 struct sk_buff *skb,
1050 struct tpacket_kbdq_core *pkc;
1051 struct tpacket_block_desc *pbd;
1054 pkc = GET_PBDQC_FROM_RB(&po->rx_ring);
1055 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1057 /* Queue is frozen when user space is lagging behind */
1058 if (prb_queue_frozen(pkc)) {
1060 * Check if that last block which caused the queue to freeze,
1061 * is still in_use by user-space.
1063 if (prb_curr_blk_in_use(pkc, pbd)) {
1064 /* Can't record this packet */
1068 * Ok, the block was released by user-space.
1069 * Now let's open that block.
1070 * opening a block also thaws the queue.
1071 * Thawing is a side effect.
1073 prb_open_block(pkc, pbd);
1078 curr = pkc->nxt_offset;
1080 end = (char *)pbd + pkc->kblk_size;
1082 /* first try the current block */
1083 if (curr+TOTAL_PKT_LEN_INCL_ALIGN(len) < end) {
1084 prb_fill_curr_block(curr, pkc, pbd, len);
1085 return (void *)curr;
1088 /* Ok, close the current block */
1089 prb_retire_current_block(pkc, po, 0);
1091 /* Now, try to dispatch the next block */
1092 curr = (char *)prb_dispatch_next_block(pkc, po);
1094 pbd = GET_CURR_PBLOCK_DESC_FROM_CORE(pkc);
1095 prb_fill_curr_block(curr, pkc, pbd, len);
1096 return (void *)curr;
1100 * No free blocks are available.user_space hasn't caught up yet.
1101 * Queue was just frozen and now this packet will get dropped.
1106 static void *packet_current_rx_frame(struct packet_sock *po,
1107 struct sk_buff *skb,
1108 int status, unsigned int len)
1111 switch (po->tp_version) {
1114 curr = packet_lookup_frame(po, &po->rx_ring,
1115 po->rx_ring.head, status);
1118 return __packet_lookup_frame_in_block(po, skb, status, len);
1120 WARN(1, "TPACKET version not supported\n");
1126 static void *prb_lookup_block(struct packet_sock *po,
1127 struct packet_ring_buffer *rb,
1131 struct tpacket_kbdq_core *pkc = GET_PBDQC_FROM_RB(rb);
1132 struct tpacket_block_desc *pbd = GET_PBLOCK_DESC(pkc, idx);
1134 if (status != BLOCK_STATUS(pbd))
1139 static int prb_previous_blk_num(struct packet_ring_buffer *rb)
1142 if (rb->prb_bdqc.kactive_blk_num)
1143 prev = rb->prb_bdqc.kactive_blk_num-1;
1145 prev = rb->prb_bdqc.knum_blocks-1;
1149 /* Assumes caller has held the rx_queue.lock */
1150 static void *__prb_previous_block(struct packet_sock *po,
1151 struct packet_ring_buffer *rb,
1154 unsigned int previous = prb_previous_blk_num(rb);
1155 return prb_lookup_block(po, rb, previous, status);
1158 static void *packet_previous_rx_frame(struct packet_sock *po,
1159 struct packet_ring_buffer *rb,
1162 if (po->tp_version <= TPACKET_V2)
1163 return packet_previous_frame(po, rb, status);
1165 return __prb_previous_block(po, rb, status);
1168 static void packet_increment_rx_head(struct packet_sock *po,
1169 struct packet_ring_buffer *rb)
1171 switch (po->tp_version) {
1174 return packet_increment_head(rb);
1177 WARN(1, "TPACKET version not supported.\n");
1183 static void *packet_previous_frame(struct packet_sock *po,
1184 struct packet_ring_buffer *rb,
1187 unsigned int previous = rb->head ? rb->head - 1 : rb->frame_max;
1188 return packet_lookup_frame(po, rb, previous, status);
1191 static void packet_increment_head(struct packet_ring_buffer *buff)
1193 buff->head = buff->head != buff->frame_max ? buff->head+1 : 0;
1196 static void packet_inc_pending(struct packet_ring_buffer *rb)
1198 this_cpu_inc(*rb->pending_refcnt);
1201 static void packet_dec_pending(struct packet_ring_buffer *rb)
1203 this_cpu_dec(*rb->pending_refcnt);
1206 static unsigned int packet_read_pending(const struct packet_ring_buffer *rb)
1208 unsigned int refcnt = 0;
1211 /* We don't use pending refcount in rx_ring. */
1212 if (rb->pending_refcnt == NULL)
1215 for_each_possible_cpu(cpu)
1216 refcnt += *per_cpu_ptr(rb->pending_refcnt, cpu);
1221 static int packet_alloc_pending(struct packet_sock *po)
1223 po->rx_ring.pending_refcnt = NULL;
1225 po->tx_ring.pending_refcnt = alloc_percpu(unsigned int);
1226 if (unlikely(po->tx_ring.pending_refcnt == NULL))
1232 static void packet_free_pending(struct packet_sock *po)
1234 free_percpu(po->tx_ring.pending_refcnt);
1237 #define ROOM_POW_OFF 2
1238 #define ROOM_NONE 0x0
1239 #define ROOM_LOW 0x1
1240 #define ROOM_NORMAL 0x2
1242 static bool __tpacket_has_room(struct packet_sock *po, int pow_off)
1246 len = po->rx_ring.frame_max + 1;
1247 idx = po->rx_ring.head;
1249 idx += len >> pow_off;
1252 return packet_lookup_frame(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1255 static bool __tpacket_v3_has_room(struct packet_sock *po, int pow_off)
1259 len = po->rx_ring.prb_bdqc.knum_blocks;
1260 idx = po->rx_ring.prb_bdqc.kactive_blk_num;
1262 idx += len >> pow_off;
1265 return prb_lookup_block(po, &po->rx_ring, idx, TP_STATUS_KERNEL);
1268 static int packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1270 struct sock *sk = &po->sk;
1271 int ret = ROOM_NONE;
1273 if (po->prot_hook.func != tpacket_rcv) {
1274 int avail = sk->sk_rcvbuf - atomic_read(&sk->sk_rmem_alloc)
1276 if (avail > (sk->sk_rcvbuf >> ROOM_POW_OFF))
1284 spin_lock(&sk->sk_receive_queue.lock);
1285 if (po->tp_version == TPACKET_V3) {
1286 if (__tpacket_v3_has_room(po, ROOM_POW_OFF))
1288 else if (__tpacket_v3_has_room(po, 0))
1291 if (__tpacket_has_room(po, ROOM_POW_OFF))
1293 else if (__tpacket_has_room(po, 0))
1296 spin_unlock(&sk->sk_receive_queue.lock);
1301 static void packet_sock_destruct(struct sock *sk)
1303 skb_queue_purge(&sk->sk_error_queue);
1305 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1306 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1308 if (!sock_flag(sk, SOCK_DEAD)) {
1309 pr_err("Attempt to release alive packet socket: %p\n", sk);
1313 sk_refcnt_debug_dec(sk);
1316 static int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1318 int x = atomic_read(&f->rr_cur) + 1;
1326 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1327 struct sk_buff *skb,
1330 return reciprocal_scale(skb_get_hash(skb), num);
1333 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1334 struct sk_buff *skb,
1339 cur = atomic_read(&f->rr_cur);
1340 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1341 fanout_rr_next(f, num))) != cur)
1346 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1347 struct sk_buff *skb,
1350 return smp_processor_id() % num;
1353 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1354 struct sk_buff *skb,
1357 return prandom_u32_max(num);
1360 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1361 struct sk_buff *skb,
1362 unsigned int idx, bool try_self,
1365 struct packet_sock *po;
1368 po = pkt_sk(f->arr[idx]);
1369 if (try_self && packet_rcv_has_room(po, skb) != ROOM_NONE)
1372 i = j = min_t(int, po->rollover->sock, num - 1);
1375 packet_rcv_has_room(pkt_sk(f->arr[i]), skb) == ROOM_NORMAL) {
1377 po->rollover->sock = i;
1388 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1389 struct sk_buff *skb,
1392 return skb_get_queue_mapping(skb) % num;
1395 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1397 return f->flags & (flag >> 8);
1400 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1401 struct packet_type *pt, struct net_device *orig_dev)
1403 struct packet_fanout *f = pt->af_packet_priv;
1404 unsigned int num = f->num_members;
1405 struct packet_sock *po;
1408 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1414 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1415 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1420 case PACKET_FANOUT_HASH:
1422 idx = fanout_demux_hash(f, skb, num);
1424 case PACKET_FANOUT_LB:
1425 idx = fanout_demux_lb(f, skb, num);
1427 case PACKET_FANOUT_CPU:
1428 idx = fanout_demux_cpu(f, skb, num);
1430 case PACKET_FANOUT_RND:
1431 idx = fanout_demux_rnd(f, skb, num);
1433 case PACKET_FANOUT_QM:
1434 idx = fanout_demux_qm(f, skb, num);
1436 case PACKET_FANOUT_ROLLOVER:
1437 idx = fanout_demux_rollover(f, skb, 0, false, num);
1441 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1442 idx = fanout_demux_rollover(f, skb, idx, true, num);
1444 po = pkt_sk(f->arr[idx]);
1445 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1448 DEFINE_MUTEX(fanout_mutex);
1449 EXPORT_SYMBOL_GPL(fanout_mutex);
1450 static LIST_HEAD(fanout_list);
1452 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1454 struct packet_fanout *f = po->fanout;
1456 spin_lock(&f->lock);
1457 f->arr[f->num_members] = sk;
1460 spin_unlock(&f->lock);
1463 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1465 struct packet_fanout *f = po->fanout;
1468 spin_lock(&f->lock);
1469 for (i = 0; i < f->num_members; i++) {
1470 if (f->arr[i] == sk)
1473 BUG_ON(i >= f->num_members);
1474 f->arr[i] = f->arr[f->num_members - 1];
1476 spin_unlock(&f->lock);
1479 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1481 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1487 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1489 struct packet_sock *po = pkt_sk(sk);
1490 struct packet_fanout *f, *match;
1491 u8 type = type_flags & 0xff;
1492 u8 flags = type_flags >> 8;
1496 case PACKET_FANOUT_ROLLOVER:
1497 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1499 case PACKET_FANOUT_HASH:
1500 case PACKET_FANOUT_LB:
1501 case PACKET_FANOUT_CPU:
1502 case PACKET_FANOUT_RND:
1503 case PACKET_FANOUT_QM:
1515 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) {
1516 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1521 mutex_lock(&fanout_mutex);
1523 list_for_each_entry(f, &fanout_list, list) {
1525 read_pnet(&f->net) == sock_net(sk)) {
1531 if (match && match->flags != flags)
1535 match = kzalloc(sizeof(*match), GFP_KERNEL);
1538 write_pnet(&match->net, sock_net(sk));
1541 match->flags = flags;
1542 atomic_set(&match->rr_cur, 0);
1543 INIT_LIST_HEAD(&match->list);
1544 spin_lock_init(&match->lock);
1545 atomic_set(&match->sk_ref, 0);
1546 match->prot_hook.type = po->prot_hook.type;
1547 match->prot_hook.dev = po->prot_hook.dev;
1548 match->prot_hook.func = packet_rcv_fanout;
1549 match->prot_hook.af_packet_priv = match;
1550 match->prot_hook.id_match = match_fanout_group;
1551 dev_add_pack(&match->prot_hook);
1552 list_add(&match->list, &fanout_list);
1555 if (match->type == type &&
1556 match->prot_hook.type == po->prot_hook.type &&
1557 match->prot_hook.dev == po->prot_hook.dev) {
1559 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1560 __dev_remove_pack(&po->prot_hook);
1562 atomic_inc(&match->sk_ref);
1563 __fanout_link(sk, po);
1568 mutex_unlock(&fanout_mutex);
1570 kfree(po->rollover);
1571 po->rollover = NULL;
1576 static void fanout_release(struct sock *sk)
1578 struct packet_sock *po = pkt_sk(sk);
1579 struct packet_fanout *f;
1585 mutex_lock(&fanout_mutex);
1588 if (atomic_dec_and_test(&f->sk_ref)) {
1590 dev_remove_pack(&f->prot_hook);
1593 mutex_unlock(&fanout_mutex);
1595 kfree(po->rollover);
1598 static const struct proto_ops packet_ops;
1600 static const struct proto_ops packet_ops_spkt;
1602 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1603 struct packet_type *pt, struct net_device *orig_dev)
1606 struct sockaddr_pkt *spkt;
1609 * When we registered the protocol we saved the socket in the data
1610 * field for just this event.
1613 sk = pt->af_packet_priv;
1616 * Yank back the headers [hope the device set this
1617 * right or kerboom...]
1619 * Incoming packets have ll header pulled,
1622 * For outgoing ones skb->data == skb_mac_header(skb)
1623 * so that this procedure is noop.
1626 if (skb->pkt_type == PACKET_LOOPBACK)
1629 if (!net_eq(dev_net(dev), sock_net(sk)))
1632 skb = skb_share_check(skb, GFP_ATOMIC);
1636 /* drop any routing info */
1639 /* drop conntrack reference */
1642 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1644 skb_push(skb, skb->data - skb_mac_header(skb));
1647 * The SOCK_PACKET socket receives _all_ frames.
1650 spkt->spkt_family = dev->type;
1651 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1652 spkt->spkt_protocol = skb->protocol;
1655 * Charge the memory to the socket. This is done specifically
1656 * to prevent sockets using all the memory up.
1659 if (sock_queue_rcv_skb(sk, skb) == 0)
1670 * Output a raw packet to a device layer. This bypasses all the other
1671 * protocol layers and you must therefore supply it with a complete frame
1674 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1677 struct sock *sk = sock->sk;
1678 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1679 struct sk_buff *skb = NULL;
1680 struct net_device *dev;
1686 * Get and verify the address.
1690 if (msg->msg_namelen < sizeof(struct sockaddr))
1692 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1693 proto = saddr->spkt_protocol;
1695 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1698 * Find the device first to size check it
1701 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1704 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1710 if (!(dev->flags & IFF_UP))
1714 * You may not queue a frame bigger than the mtu. This is the lowest level
1715 * raw protocol and you must do your own fragmentation at this level.
1718 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1719 if (!netif_supports_nofcs(dev)) {
1720 err = -EPROTONOSUPPORT;
1723 extra_len = 4; /* We're doing our own CRC */
1727 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1731 size_t reserved = LL_RESERVED_SPACE(dev);
1732 int tlen = dev->needed_tailroom;
1733 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1736 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1739 /* FIXME: Save some space for broken drivers that write a hard
1740 * header at transmission time by themselves. PPP is the notable
1741 * one here. This should really be fixed at the driver level.
1743 skb_reserve(skb, reserved);
1744 skb_reset_network_header(skb);
1746 /* Try to align data part correctly */
1751 skb_reset_network_header(skb);
1753 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1759 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1760 /* Earlier code assumed this would be a VLAN pkt,
1761 * double-check this now that we have the actual
1764 struct ethhdr *ehdr;
1765 skb_reset_mac_header(skb);
1766 ehdr = eth_hdr(skb);
1767 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1773 skb->protocol = proto;
1775 skb->priority = sk->sk_priority;
1776 skb->mark = sk->sk_mark;
1778 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1780 if (unlikely(extra_len == 4))
1783 skb_probe_transport_header(skb, 0);
1785 dev_queue_xmit(skb);
1796 static unsigned int run_filter(const struct sk_buff *skb,
1797 const struct sock *sk,
1800 struct sk_filter *filter;
1803 filter = rcu_dereference(sk->sk_filter);
1805 res = SK_RUN_FILTER(filter, skb);
1812 * This function makes lazy skb cloning in hope that most of packets
1813 * are discarded by BPF.
1815 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1816 * and skb->cb are mangled. It works because (and until) packets
1817 * falling here are owned by current CPU. Output packets are cloned
1818 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1819 * sequencially, so that if we return skb to original state on exit,
1820 * we will not harm anyone.
1823 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1824 struct packet_type *pt, struct net_device *orig_dev)
1827 struct sockaddr_ll *sll;
1828 struct packet_sock *po;
1829 u8 *skb_head = skb->data;
1830 int skb_len = skb->len;
1831 unsigned int snaplen, res;
1833 if (skb->pkt_type == PACKET_LOOPBACK)
1836 sk = pt->af_packet_priv;
1839 if (!net_eq(dev_net(dev), sock_net(sk)))
1844 if (dev->header_ops) {
1845 /* The device has an explicit notion of ll header,
1846 * exported to higher levels.
1848 * Otherwise, the device hides details of its frame
1849 * structure, so that corresponding packet head is
1850 * never delivered to user.
1852 if (sk->sk_type != SOCK_DGRAM)
1853 skb_push(skb, skb->data - skb_mac_header(skb));
1854 else if (skb->pkt_type == PACKET_OUTGOING) {
1855 /* Special case: outgoing packets have ll header at head */
1856 skb_pull(skb, skb_network_offset(skb));
1862 res = run_filter(skb, sk, snaplen);
1864 goto drop_n_restore;
1868 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1871 if (skb_shared(skb)) {
1872 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1876 if (skb_head != skb->data) {
1877 skb->data = skb_head;
1884 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1886 sll = &PACKET_SKB_CB(skb)->sa.ll;
1887 sll->sll_hatype = dev->type;
1888 sll->sll_pkttype = skb->pkt_type;
1889 if (unlikely(po->origdev))
1890 sll->sll_ifindex = orig_dev->ifindex;
1892 sll->sll_ifindex = dev->ifindex;
1894 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1896 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1897 * Use their space for storing the original skb length.
1899 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1901 if (pskb_trim(skb, snaplen))
1904 skb_set_owner_r(skb, sk);
1908 /* drop conntrack reference */
1911 spin_lock(&sk->sk_receive_queue.lock);
1912 po->stats.stats1.tp_packets++;
1913 sock_skb_set_dropcount(sk, skb);
1914 __skb_queue_tail(&sk->sk_receive_queue, skb);
1915 spin_unlock(&sk->sk_receive_queue.lock);
1916 sk->sk_data_ready(sk);
1920 spin_lock(&sk->sk_receive_queue.lock);
1921 po->stats.stats1.tp_drops++;
1922 atomic_inc(&sk->sk_drops);
1923 spin_unlock(&sk->sk_receive_queue.lock);
1926 if (skb_head != skb->data && skb_shared(skb)) {
1927 skb->data = skb_head;
1935 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1936 struct packet_type *pt, struct net_device *orig_dev)
1939 struct packet_sock *po;
1940 struct sockaddr_ll *sll;
1941 union tpacket_uhdr h;
1942 u8 *skb_head = skb->data;
1943 int skb_len = skb->len;
1944 unsigned int snaplen, res;
1945 unsigned long status = TP_STATUS_USER;
1946 unsigned short macoff, netoff, hdrlen;
1947 struct sk_buff *copy_skb = NULL;
1951 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1952 * We may add members to them until current aligned size without forcing
1953 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1955 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1956 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1958 if (skb->pkt_type == PACKET_LOOPBACK)
1961 sk = pt->af_packet_priv;
1964 if (!net_eq(dev_net(dev), sock_net(sk)))
1967 if (dev->header_ops) {
1968 if (sk->sk_type != SOCK_DGRAM)
1969 skb_push(skb, skb->data - skb_mac_header(skb));
1970 else if (skb->pkt_type == PACKET_OUTGOING) {
1971 /* Special case: outgoing packets have ll header at head */
1972 skb_pull(skb, skb_network_offset(skb));
1978 res = run_filter(skb, sk, snaplen);
1980 goto drop_n_restore;
1982 if (skb->ip_summed == CHECKSUM_PARTIAL)
1983 status |= TP_STATUS_CSUMNOTREADY;
1984 else if (skb->pkt_type != PACKET_OUTGOING &&
1985 (skb->ip_summed == CHECKSUM_COMPLETE ||
1986 skb_csum_unnecessary(skb)))
1987 status |= TP_STATUS_CSUM_VALID;
1992 if (sk->sk_type == SOCK_DGRAM) {
1993 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1996 unsigned int maclen = skb_network_offset(skb);
1997 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1998 (maclen < 16 ? 16 : maclen)) +
2000 macoff = netoff - maclen;
2002 if (po->tp_version <= TPACKET_V2) {
2003 if (macoff + snaplen > po->rx_ring.frame_size) {
2004 if (po->copy_thresh &&
2005 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
2006 if (skb_shared(skb)) {
2007 copy_skb = skb_clone(skb, GFP_ATOMIC);
2009 copy_skb = skb_get(skb);
2010 skb_head = skb->data;
2013 skb_set_owner_r(copy_skb, sk);
2015 snaplen = po->rx_ring.frame_size - macoff;
2016 if ((int)snaplen < 0)
2019 } else if (unlikely(macoff + snaplen >
2020 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
2023 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
2024 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
2025 snaplen, nval, macoff);
2027 if (unlikely((int)snaplen < 0)) {
2029 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
2032 spin_lock(&sk->sk_receive_queue.lock);
2033 h.raw = packet_current_rx_frame(po, skb,
2034 TP_STATUS_KERNEL, (macoff+snaplen));
2037 if (po->tp_version <= TPACKET_V2) {
2038 packet_increment_rx_head(po, &po->rx_ring);
2040 * LOSING will be reported till you read the stats,
2041 * because it's COR - Clear On Read.
2042 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2045 if (po->stats.stats1.tp_drops)
2046 status |= TP_STATUS_LOSING;
2048 po->stats.stats1.tp_packets++;
2050 status |= TP_STATUS_COPY;
2051 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2053 spin_unlock(&sk->sk_receive_queue.lock);
2055 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2057 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2058 getnstimeofday(&ts);
2060 status |= ts_status;
2062 switch (po->tp_version) {
2064 h.h1->tp_len = skb->len;
2065 h.h1->tp_snaplen = snaplen;
2066 h.h1->tp_mac = macoff;
2067 h.h1->tp_net = netoff;
2068 h.h1->tp_sec = ts.tv_sec;
2069 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2070 hdrlen = sizeof(*h.h1);
2073 h.h2->tp_len = skb->len;
2074 h.h2->tp_snaplen = snaplen;
2075 h.h2->tp_mac = macoff;
2076 h.h2->tp_net = netoff;
2077 h.h2->tp_sec = ts.tv_sec;
2078 h.h2->tp_nsec = ts.tv_nsec;
2079 if (skb_vlan_tag_present(skb)) {
2080 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2081 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2082 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2084 h.h2->tp_vlan_tci = 0;
2085 h.h2->tp_vlan_tpid = 0;
2087 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2088 hdrlen = sizeof(*h.h2);
2091 /* tp_nxt_offset,vlan are already populated above.
2092 * So DONT clear those fields here
2094 h.h3->tp_status |= status;
2095 h.h3->tp_len = skb->len;
2096 h.h3->tp_snaplen = snaplen;
2097 h.h3->tp_mac = macoff;
2098 h.h3->tp_net = netoff;
2099 h.h3->tp_sec = ts.tv_sec;
2100 h.h3->tp_nsec = ts.tv_nsec;
2101 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2102 hdrlen = sizeof(*h.h3);
2108 sll = h.raw + TPACKET_ALIGN(hdrlen);
2109 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2110 sll->sll_family = AF_PACKET;
2111 sll->sll_hatype = dev->type;
2112 sll->sll_protocol = skb->protocol;
2113 sll->sll_pkttype = skb->pkt_type;
2114 if (unlikely(po->origdev))
2115 sll->sll_ifindex = orig_dev->ifindex;
2117 sll->sll_ifindex = dev->ifindex;
2121 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2122 if (po->tp_version <= TPACKET_V2) {
2125 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2128 for (start = h.raw; start < end; start += PAGE_SIZE)
2129 flush_dcache_page(pgv_to_page(start));
2134 if (po->tp_version <= TPACKET_V2) {
2135 __packet_set_status(po, h.raw, status);
2136 sk->sk_data_ready(sk);
2138 prb_clear_blk_fill_status(&po->rx_ring);
2142 if (skb_head != skb->data && skb_shared(skb)) {
2143 skb->data = skb_head;
2151 po->stats.stats1.tp_drops++;
2152 spin_unlock(&sk->sk_receive_queue.lock);
2154 sk->sk_data_ready(sk);
2155 kfree_skb(copy_skb);
2156 goto drop_n_restore;
2159 static void tpacket_destruct_skb(struct sk_buff *skb)
2161 struct packet_sock *po = pkt_sk(skb->sk);
2163 if (likely(po->tx_ring.pg_vec)) {
2167 ph = skb_shinfo(skb)->destructor_arg;
2168 packet_dec_pending(&po->tx_ring);
2170 ts = __packet_set_timestamp(po, ph, skb);
2171 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2177 static bool ll_header_truncated(const struct net_device *dev, int len)
2179 /* net device doesn't like empty head */
2180 if (unlikely(len <= dev->hard_header_len)) {
2181 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2182 current->comm, len, dev->hard_header_len);
2189 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2190 void *frame, struct net_device *dev, int size_max,
2191 __be16 proto, unsigned char *addr, int hlen)
2193 union tpacket_uhdr ph;
2194 int to_write, offset, len, tp_len, nr_frags, len_max;
2195 struct socket *sock = po->sk.sk_socket;
2202 skb->protocol = proto;
2204 skb->priority = po->sk.sk_priority;
2205 skb->mark = po->sk.sk_mark;
2206 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2207 skb_shinfo(skb)->destructor_arg = ph.raw;
2209 switch (po->tp_version) {
2211 tp_len = ph.h2->tp_len;
2214 tp_len = ph.h1->tp_len;
2217 if (unlikely(tp_len > size_max)) {
2218 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2222 skb_reserve(skb, hlen);
2223 skb_reset_network_header(skb);
2225 if (!packet_use_direct_xmit(po))
2226 skb_probe_transport_header(skb, 0);
2227 if (unlikely(po->tp_tx_has_off)) {
2228 int off_min, off_max, off;
2229 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2230 off_max = po->tx_ring.frame_size - tp_len;
2231 if (sock->type == SOCK_DGRAM) {
2232 switch (po->tp_version) {
2234 off = ph.h2->tp_net;
2237 off = ph.h1->tp_net;
2241 switch (po->tp_version) {
2243 off = ph.h2->tp_mac;
2246 off = ph.h1->tp_mac;
2250 if (unlikely((off < off_min) || (off_max < off)))
2252 data = ph.raw + off;
2254 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2258 if (sock->type == SOCK_DGRAM) {
2259 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2261 if (unlikely(err < 0))
2263 } else if (dev->hard_header_len) {
2264 if (ll_header_truncated(dev, tp_len))
2267 skb_push(skb, dev->hard_header_len);
2268 err = skb_store_bits(skb, 0, data,
2269 dev->hard_header_len);
2273 data += dev->hard_header_len;
2274 to_write -= dev->hard_header_len;
2277 offset = offset_in_page(data);
2278 len_max = PAGE_SIZE - offset;
2279 len = ((to_write > len_max) ? len_max : to_write);
2281 skb->data_len = to_write;
2282 skb->len += to_write;
2283 skb->truesize += to_write;
2284 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2286 while (likely(to_write)) {
2287 nr_frags = skb_shinfo(skb)->nr_frags;
2289 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2290 pr_err("Packet exceed the number of skb frags(%lu)\n",
2295 page = pgv_to_page(data);
2297 flush_dcache_page(page);
2299 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2302 len_max = PAGE_SIZE;
2303 len = ((to_write > len_max) ? len_max : to_write);
2309 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2311 struct sk_buff *skb;
2312 struct net_device *dev;
2314 int err, reserve = 0;
2316 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2317 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2318 int tp_len, size_max;
2319 unsigned char *addr;
2321 int status = TP_STATUS_AVAILABLE;
2324 mutex_lock(&po->pg_vec_lock);
2326 if (likely(saddr == NULL)) {
2327 dev = packet_cached_dev_get(po);
2332 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2334 if (msg->msg_namelen < (saddr->sll_halen
2335 + offsetof(struct sockaddr_ll,
2338 proto = saddr->sll_protocol;
2339 addr = saddr->sll_addr;
2340 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2344 if (unlikely(dev == NULL))
2347 if (unlikely(!(dev->flags & IFF_UP)))
2350 reserve = dev->hard_header_len + VLAN_HLEN;
2351 size_max = po->tx_ring.frame_size
2352 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2354 if (size_max > dev->mtu + reserve)
2355 size_max = dev->mtu + reserve;
2358 ph = packet_current_frame(po, &po->tx_ring,
2359 TP_STATUS_SEND_REQUEST);
2360 if (unlikely(ph == NULL)) {
2361 if (need_wait && need_resched())
2366 status = TP_STATUS_SEND_REQUEST;
2367 hlen = LL_RESERVED_SPACE(dev);
2368 tlen = dev->needed_tailroom;
2369 skb = sock_alloc_send_skb(&po->sk,
2370 hlen + tlen + sizeof(struct sockaddr_ll),
2373 if (unlikely(skb == NULL)) {
2374 /* we assume the socket was initially writeable ... */
2375 if (likely(len_sum > 0))
2379 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2381 if (tp_len > dev->mtu + dev->hard_header_len) {
2382 struct ethhdr *ehdr;
2383 /* Earlier code assumed this would be a VLAN pkt,
2384 * double-check this now that we have the actual
2388 skb_reset_mac_header(skb);
2389 ehdr = eth_hdr(skb);
2390 if (ehdr->h_proto != htons(ETH_P_8021Q))
2393 if (unlikely(tp_len < 0)) {
2395 __packet_set_status(po, ph,
2396 TP_STATUS_AVAILABLE);
2397 packet_increment_head(&po->tx_ring);
2401 status = TP_STATUS_WRONG_FORMAT;
2407 packet_pick_tx_queue(dev, skb);
2409 skb->destructor = tpacket_destruct_skb;
2410 __packet_set_status(po, ph, TP_STATUS_SENDING);
2411 packet_inc_pending(&po->tx_ring);
2413 status = TP_STATUS_SEND_REQUEST;
2414 err = po->xmit(skb);
2415 if (unlikely(err > 0)) {
2416 err = net_xmit_errno(err);
2417 if (err && __packet_get_status(po, ph) ==
2418 TP_STATUS_AVAILABLE) {
2419 /* skb was destructed already */
2424 * skb was dropped but not destructed yet;
2425 * let's treat it like congestion or err < 0
2429 packet_increment_head(&po->tx_ring);
2431 } while (likely((ph != NULL) ||
2432 /* Note: packet_read_pending() might be slow if we have
2433 * to call it as it's per_cpu variable, but in fast-path
2434 * we already short-circuit the loop with the first
2435 * condition, and luckily don't have to go that path
2438 (need_wait && packet_read_pending(&po->tx_ring))));
2444 __packet_set_status(po, ph, status);
2449 mutex_unlock(&po->pg_vec_lock);
2453 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2454 size_t reserve, size_t len,
2455 size_t linear, int noblock,
2458 struct sk_buff *skb;
2460 /* Under a page? Don't bother with paged skb. */
2461 if (prepad + len < PAGE_SIZE || !linear)
2464 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2469 skb_reserve(skb, reserve);
2470 skb_put(skb, linear);
2471 skb->data_len = len - linear;
2472 skb->len += len - linear;
2477 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2479 struct sock *sk = sock->sk;
2480 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2481 struct sk_buff *skb;
2482 struct net_device *dev;
2484 unsigned char *addr;
2485 int err, reserve = 0;
2486 struct virtio_net_hdr vnet_hdr = { 0 };
2489 struct packet_sock *po = pkt_sk(sk);
2490 unsigned short gso_type = 0;
2496 * Get and verify the address.
2499 if (likely(saddr == NULL)) {
2500 dev = packet_cached_dev_get(po);
2505 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2507 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2509 proto = saddr->sll_protocol;
2510 addr = saddr->sll_addr;
2511 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2515 if (unlikely(dev == NULL))
2518 if (unlikely(!(dev->flags & IFF_UP)))
2521 if (sock->type == SOCK_RAW)
2522 reserve = dev->hard_header_len;
2523 if (po->has_vnet_hdr) {
2524 vnet_hdr_len = sizeof(vnet_hdr);
2527 if (len < vnet_hdr_len)
2530 len -= vnet_hdr_len;
2533 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2534 if (n != vnet_hdr_len)
2537 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2538 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2539 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2540 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2541 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2542 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2543 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2546 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2549 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2550 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2551 case VIRTIO_NET_HDR_GSO_TCPV4:
2552 gso_type = SKB_GSO_TCPV4;
2554 case VIRTIO_NET_HDR_GSO_TCPV6:
2555 gso_type = SKB_GSO_TCPV6;
2557 case VIRTIO_NET_HDR_GSO_UDP:
2558 gso_type = SKB_GSO_UDP;
2564 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2565 gso_type |= SKB_GSO_TCP_ECN;
2567 if (vnet_hdr.gso_size == 0)
2573 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2574 if (!netif_supports_nofcs(dev)) {
2575 err = -EPROTONOSUPPORT;
2578 extra_len = 4; /* We're doing our own CRC */
2582 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2586 hlen = LL_RESERVED_SPACE(dev);
2587 tlen = dev->needed_tailroom;
2588 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2589 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2590 msg->msg_flags & MSG_DONTWAIT, &err);
2594 skb_set_network_header(skb, reserve);
2597 if (sock->type == SOCK_DGRAM) {
2598 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2599 if (unlikely(offset < 0))
2602 if (ll_header_truncated(dev, len))
2606 /* Returns -EFAULT on error */
2607 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2611 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2613 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2614 /* Earlier code assumed this would be a VLAN pkt,
2615 * double-check this now that we have the actual
2618 struct ethhdr *ehdr;
2619 skb_reset_mac_header(skb);
2620 ehdr = eth_hdr(skb);
2621 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2627 skb->protocol = proto;
2629 skb->priority = sk->sk_priority;
2630 skb->mark = sk->sk_mark;
2632 packet_pick_tx_queue(dev, skb);
2634 if (po->has_vnet_hdr) {
2635 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2636 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2637 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2638 if (!skb_partial_csum_set(skb, s, o)) {
2644 skb_shinfo(skb)->gso_size =
2645 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2646 skb_shinfo(skb)->gso_type = gso_type;
2648 /* Header must be checked, and gso_segs computed. */
2649 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2650 skb_shinfo(skb)->gso_segs = 0;
2652 len += vnet_hdr_len;
2655 if (!packet_use_direct_xmit(po))
2656 skb_probe_transport_header(skb, reserve);
2657 if (unlikely(extra_len == 4))
2660 err = po->xmit(skb);
2661 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2677 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2679 struct sock *sk = sock->sk;
2680 struct packet_sock *po = pkt_sk(sk);
2682 if (po->tx_ring.pg_vec)
2683 return tpacket_snd(po, msg);
2685 return packet_snd(sock, msg, len);
2689 * Close a PACKET socket. This is fairly simple. We immediately go
2690 * to 'closed' state and remove our protocol entry in the device list.
2693 static int packet_release(struct socket *sock)
2695 struct sock *sk = sock->sk;
2696 struct packet_sock *po;
2698 union tpacket_req_u req_u;
2706 mutex_lock(&net->packet.sklist_lock);
2707 sk_del_node_init_rcu(sk);
2708 mutex_unlock(&net->packet.sklist_lock);
2711 sock_prot_inuse_add(net, sk->sk_prot, -1);
2714 spin_lock(&po->bind_lock);
2715 unregister_prot_hook(sk, false);
2716 packet_cached_dev_reset(po);
2718 if (po->prot_hook.dev) {
2719 dev_put(po->prot_hook.dev);
2720 po->prot_hook.dev = NULL;
2722 spin_unlock(&po->bind_lock);
2724 packet_flush_mclist(sk);
2726 if (po->rx_ring.pg_vec) {
2727 memset(&req_u, 0, sizeof(req_u));
2728 packet_set_ring(sk, &req_u, 1, 0);
2731 if (po->tx_ring.pg_vec) {
2732 memset(&req_u, 0, sizeof(req_u));
2733 packet_set_ring(sk, &req_u, 1, 1);
2740 * Now the socket is dead. No more input will appear.
2747 skb_queue_purge(&sk->sk_receive_queue);
2748 packet_free_pending(po);
2749 sk_refcnt_debug_release(sk);
2756 * Attach a packet hook.
2759 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2761 struct packet_sock *po = pkt_sk(sk);
2762 const struct net_device *dev_curr;
2774 spin_lock(&po->bind_lock);
2776 proto_curr = po->prot_hook.type;
2777 dev_curr = po->prot_hook.dev;
2779 need_rehook = proto_curr != proto || dev_curr != dev;
2782 unregister_prot_hook(sk, true);
2785 po->prot_hook.type = proto;
2787 if (po->prot_hook.dev)
2788 dev_put(po->prot_hook.dev);
2790 po->prot_hook.dev = dev;
2792 po->ifindex = dev ? dev->ifindex : 0;
2793 packet_cached_dev_assign(po, dev);
2796 if (proto == 0 || !need_rehook)
2799 if (!dev || (dev->flags & IFF_UP)) {
2800 register_prot_hook(sk);
2802 sk->sk_err = ENETDOWN;
2803 if (!sock_flag(sk, SOCK_DEAD))
2804 sk->sk_error_report(sk);
2808 spin_unlock(&po->bind_lock);
2814 * Bind a packet socket to a device
2817 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2820 struct sock *sk = sock->sk;
2822 struct net_device *dev;
2829 if (addr_len != sizeof(struct sockaddr))
2831 strlcpy(name, uaddr->sa_data, sizeof(name));
2833 dev = dev_get_by_name(sock_net(sk), name);
2835 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2839 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2841 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2842 struct sock *sk = sock->sk;
2843 struct net_device *dev = NULL;
2851 if (addr_len < sizeof(struct sockaddr_ll))
2853 if (sll->sll_family != AF_PACKET)
2856 if (sll->sll_ifindex) {
2858 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2862 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2868 static struct proto packet_proto = {
2870 .owner = THIS_MODULE,
2871 .obj_size = sizeof(struct packet_sock),
2875 * Create a packet of type SOCK_PACKET.
2878 static int packet_create(struct net *net, struct socket *sock, int protocol,
2882 struct packet_sock *po;
2883 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2886 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2888 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2889 sock->type != SOCK_PACKET)
2890 return -ESOCKTNOSUPPORT;
2892 sock->state = SS_UNCONNECTED;
2895 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2899 sock->ops = &packet_ops;
2900 if (sock->type == SOCK_PACKET)
2901 sock->ops = &packet_ops_spkt;
2903 sock_init_data(sock, sk);
2906 sk->sk_family = PF_PACKET;
2908 po->xmit = dev_queue_xmit;
2910 err = packet_alloc_pending(po);
2914 packet_cached_dev_reset(po);
2916 sk->sk_destruct = packet_sock_destruct;
2917 sk_refcnt_debug_inc(sk);
2920 * Attach a protocol block
2923 spin_lock_init(&po->bind_lock);
2924 mutex_init(&po->pg_vec_lock);
2925 po->rollover = NULL;
2926 po->prot_hook.func = packet_rcv;
2928 if (sock->type == SOCK_PACKET)
2929 po->prot_hook.func = packet_rcv_spkt;
2931 po->prot_hook.af_packet_priv = sk;
2934 po->prot_hook.type = proto;
2935 register_prot_hook(sk);
2938 mutex_lock(&net->packet.sklist_lock);
2939 sk_add_node_rcu(sk, &net->packet.sklist);
2940 mutex_unlock(&net->packet.sklist_lock);
2943 sock_prot_inuse_add(net, &packet_proto, 1);
2954 * Pull a packet from our receive queue and hand it to the user.
2955 * If necessary we block.
2958 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2961 struct sock *sk = sock->sk;
2962 struct sk_buff *skb;
2964 int vnet_hdr_len = 0;
2965 unsigned int origlen = 0;
2968 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2972 /* What error should we return now? EUNATTACH? */
2973 if (pkt_sk(sk)->ifindex < 0)
2977 if (flags & MSG_ERRQUEUE) {
2978 err = sock_recv_errqueue(sk, msg, len,
2979 SOL_PACKET, PACKET_TX_TIMESTAMP);
2984 * Call the generic datagram receiver. This handles all sorts
2985 * of horrible races and re-entrancy so we can forget about it
2986 * in the protocol layers.
2988 * Now it will return ENETDOWN, if device have just gone down,
2989 * but then it will block.
2992 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2995 * An error occurred so return it. Because skb_recv_datagram()
2996 * handles the blocking we don't see and worry about blocking
3003 if (pkt_sk(sk)->has_vnet_hdr) {
3004 struct virtio_net_hdr vnet_hdr = { 0 };
3007 vnet_hdr_len = sizeof(vnet_hdr);
3008 if (len < vnet_hdr_len)
3011 len -= vnet_hdr_len;
3013 if (skb_is_gso(skb)) {
3014 struct skb_shared_info *sinfo = skb_shinfo(skb);
3016 /* This is a hint as to how much should be linear. */
3018 __cpu_to_virtio16(false, skb_headlen(skb));
3020 __cpu_to_virtio16(false, sinfo->gso_size);
3021 if (sinfo->gso_type & SKB_GSO_TCPV4)
3022 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
3023 else if (sinfo->gso_type & SKB_GSO_TCPV6)
3024 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
3025 else if (sinfo->gso_type & SKB_GSO_UDP)
3026 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
3027 else if (sinfo->gso_type & SKB_GSO_FCOE)
3031 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
3032 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
3034 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
3036 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3037 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
3038 vnet_hdr.csum_start = __cpu_to_virtio16(false,
3039 skb_checksum_start_offset(skb));
3040 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3042 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3043 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3044 } /* else everything is zero */
3046 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3051 /* You lose any data beyond the buffer you gave. If it worries
3052 * a user program they can ask the device for its MTU
3058 msg->msg_flags |= MSG_TRUNC;
3061 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3065 if (sock->type != SOCK_PACKET) {
3066 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3068 /* Original length was stored in sockaddr_ll fields */
3069 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3070 sll->sll_family = AF_PACKET;
3071 sll->sll_protocol = skb->protocol;
3074 sock_recv_ts_and_drops(msg, sk, skb);
3076 if (msg->msg_name) {
3077 /* If the address length field is there to be filled
3078 * in, we fill it in now.
3080 if (sock->type == SOCK_PACKET) {
3081 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3082 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3084 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3086 msg->msg_namelen = sll->sll_halen +
3087 offsetof(struct sockaddr_ll, sll_addr);
3089 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3093 if (pkt_sk(sk)->auxdata) {
3094 struct tpacket_auxdata aux;
3096 aux.tp_status = TP_STATUS_USER;
3097 if (skb->ip_summed == CHECKSUM_PARTIAL)
3098 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3099 else if (skb->pkt_type != PACKET_OUTGOING &&
3100 (skb->ip_summed == CHECKSUM_COMPLETE ||
3101 skb_csum_unnecessary(skb)))
3102 aux.tp_status |= TP_STATUS_CSUM_VALID;
3104 aux.tp_len = origlen;
3105 aux.tp_snaplen = skb->len;
3107 aux.tp_net = skb_network_offset(skb);
3108 if (skb_vlan_tag_present(skb)) {
3109 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3110 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3111 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3113 aux.tp_vlan_tci = 0;
3114 aux.tp_vlan_tpid = 0;
3116 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3120 * Free or return the buffer as appropriate. Again this
3121 * hides all the races and re-entrancy issues from us.
3123 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3126 skb_free_datagram(sk, skb);
3131 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3132 int *uaddr_len, int peer)
3134 struct net_device *dev;
3135 struct sock *sk = sock->sk;
3140 uaddr->sa_family = AF_PACKET;
3141 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3143 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3145 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3147 *uaddr_len = sizeof(*uaddr);
3152 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3153 int *uaddr_len, int peer)
3155 struct net_device *dev;
3156 struct sock *sk = sock->sk;
3157 struct packet_sock *po = pkt_sk(sk);
3158 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3163 sll->sll_family = AF_PACKET;
3164 sll->sll_ifindex = po->ifindex;
3165 sll->sll_protocol = po->num;
3166 sll->sll_pkttype = 0;
3168 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3170 sll->sll_hatype = dev->type;
3171 sll->sll_halen = dev->addr_len;
3172 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3174 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3178 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3183 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3187 case PACKET_MR_MULTICAST:
3188 if (i->alen != dev->addr_len)
3191 return dev_mc_add(dev, i->addr);
3193 return dev_mc_del(dev, i->addr);
3195 case PACKET_MR_PROMISC:
3196 return dev_set_promiscuity(dev, what);
3197 case PACKET_MR_ALLMULTI:
3198 return dev_set_allmulti(dev, what);
3199 case PACKET_MR_UNICAST:
3200 if (i->alen != dev->addr_len)
3203 return dev_uc_add(dev, i->addr);
3205 return dev_uc_del(dev, i->addr);
3213 static void packet_dev_mclist_delete(struct net_device *dev,
3214 struct packet_mclist **mlp)
3216 struct packet_mclist *ml;
3218 while ((ml = *mlp) != NULL) {
3219 if (ml->ifindex == dev->ifindex) {
3220 packet_dev_mc(dev, ml, -1);
3228 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3230 struct packet_sock *po = pkt_sk(sk);
3231 struct packet_mclist *ml, *i;
3232 struct net_device *dev;
3238 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3243 if (mreq->mr_alen > dev->addr_len)
3247 i = kmalloc(sizeof(*i), GFP_KERNEL);
3252 for (ml = po->mclist; ml; ml = ml->next) {
3253 if (ml->ifindex == mreq->mr_ifindex &&
3254 ml->type == mreq->mr_type &&
3255 ml->alen == mreq->mr_alen &&
3256 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3258 /* Free the new element ... */
3264 i->type = mreq->mr_type;
3265 i->ifindex = mreq->mr_ifindex;
3266 i->alen = mreq->mr_alen;
3267 memcpy(i->addr, mreq->mr_address, i->alen);
3269 i->next = po->mclist;
3271 err = packet_dev_mc(dev, i, 1);
3273 po->mclist = i->next;
3282 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3284 struct packet_mclist *ml, **mlp;
3288 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3289 if (ml->ifindex == mreq->mr_ifindex &&
3290 ml->type == mreq->mr_type &&
3291 ml->alen == mreq->mr_alen &&
3292 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3293 if (--ml->count == 0) {
3294 struct net_device *dev;
3296 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3298 packet_dev_mc(dev, ml, -1);
3308 static void packet_flush_mclist(struct sock *sk)
3310 struct packet_sock *po = pkt_sk(sk);
3311 struct packet_mclist *ml;
3317 while ((ml = po->mclist) != NULL) {
3318 struct net_device *dev;
3320 po->mclist = ml->next;
3321 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3323 packet_dev_mc(dev, ml, -1);
3330 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3332 struct sock *sk = sock->sk;
3333 struct packet_sock *po = pkt_sk(sk);
3336 if (level != SOL_PACKET)
3337 return -ENOPROTOOPT;
3340 case PACKET_ADD_MEMBERSHIP:
3341 case PACKET_DROP_MEMBERSHIP:
3343 struct packet_mreq_max mreq;
3345 memset(&mreq, 0, sizeof(mreq));
3346 if (len < sizeof(struct packet_mreq))
3348 if (len > sizeof(mreq))
3350 if (copy_from_user(&mreq, optval, len))
3352 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3354 if (optname == PACKET_ADD_MEMBERSHIP)
3355 ret = packet_mc_add(sk, &mreq);
3357 ret = packet_mc_drop(sk, &mreq);
3361 case PACKET_RX_RING:
3362 case PACKET_TX_RING:
3364 union tpacket_req_u req_u;
3367 switch (po->tp_version) {
3370 len = sizeof(req_u.req);
3374 len = sizeof(req_u.req3);
3379 if (pkt_sk(sk)->has_vnet_hdr)
3381 if (copy_from_user(&req_u.req, optval, len))
3383 return packet_set_ring(sk, &req_u, 0,
3384 optname == PACKET_TX_RING);
3386 case PACKET_COPY_THRESH:
3390 if (optlen != sizeof(val))
3392 if (copy_from_user(&val, optval, sizeof(val)))
3395 pkt_sk(sk)->copy_thresh = val;
3398 case PACKET_VERSION:
3402 if (optlen != sizeof(val))
3404 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3406 if (copy_from_user(&val, optval, sizeof(val)))
3412 po->tp_version = val;
3418 case PACKET_RESERVE:
3422 if (optlen != sizeof(val))
3424 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3426 if (copy_from_user(&val, optval, sizeof(val)))
3428 po->tp_reserve = val;
3435 if (optlen != sizeof(val))
3437 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3439 if (copy_from_user(&val, optval, sizeof(val)))
3441 po->tp_loss = !!val;
3444 case PACKET_AUXDATA:
3448 if (optlen < sizeof(val))
3450 if (copy_from_user(&val, optval, sizeof(val)))
3453 po->auxdata = !!val;
3456 case PACKET_ORIGDEV:
3460 if (optlen < sizeof(val))
3462 if (copy_from_user(&val, optval, sizeof(val)))
3465 po->origdev = !!val;
3468 case PACKET_VNET_HDR:
3472 if (sock->type != SOCK_RAW)
3474 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3476 if (optlen < sizeof(val))
3478 if (copy_from_user(&val, optval, sizeof(val)))
3481 po->has_vnet_hdr = !!val;
3484 case PACKET_TIMESTAMP:
3488 if (optlen != sizeof(val))
3490 if (copy_from_user(&val, optval, sizeof(val)))
3493 po->tp_tstamp = val;
3500 if (optlen != sizeof(val))
3502 if (copy_from_user(&val, optval, sizeof(val)))
3505 return fanout_add(sk, val & 0xffff, val >> 16);
3507 case PACKET_TX_HAS_OFF:
3511 if (optlen != sizeof(val))
3513 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3515 if (copy_from_user(&val, optval, sizeof(val)))
3517 po->tp_tx_has_off = !!val;
3520 case PACKET_QDISC_BYPASS:
3524 if (optlen != sizeof(val))
3526 if (copy_from_user(&val, optval, sizeof(val)))
3529 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3533 return -ENOPROTOOPT;
3537 static int packet_getsockopt(struct socket *sock, int level, int optname,
3538 char __user *optval, int __user *optlen)
3541 int val, lv = sizeof(val);
3542 struct sock *sk = sock->sk;
3543 struct packet_sock *po = pkt_sk(sk);
3545 union tpacket_stats_u st;
3547 if (level != SOL_PACKET)
3548 return -ENOPROTOOPT;
3550 if (get_user(len, optlen))
3557 case PACKET_STATISTICS:
3558 spin_lock_bh(&sk->sk_receive_queue.lock);
3559 memcpy(&st, &po->stats, sizeof(st));
3560 memset(&po->stats, 0, sizeof(po->stats));
3561 spin_unlock_bh(&sk->sk_receive_queue.lock);
3563 if (po->tp_version == TPACKET_V3) {
3564 lv = sizeof(struct tpacket_stats_v3);
3565 st.stats3.tp_packets += st.stats3.tp_drops;
3568 lv = sizeof(struct tpacket_stats);
3569 st.stats1.tp_packets += st.stats1.tp_drops;
3574 case PACKET_AUXDATA:
3577 case PACKET_ORIGDEV:
3580 case PACKET_VNET_HDR:
3581 val = po->has_vnet_hdr;
3583 case PACKET_VERSION:
3584 val = po->tp_version;
3587 if (len > sizeof(int))
3589 if (copy_from_user(&val, optval, len))
3593 val = sizeof(struct tpacket_hdr);
3596 val = sizeof(struct tpacket2_hdr);
3599 val = sizeof(struct tpacket3_hdr);
3605 case PACKET_RESERVE:
3606 val = po->tp_reserve;
3611 case PACKET_TIMESTAMP:
3612 val = po->tp_tstamp;
3616 ((u32)po->fanout->id |
3617 ((u32)po->fanout->type << 16) |
3618 ((u32)po->fanout->flags << 24)) :
3621 case PACKET_TX_HAS_OFF:
3622 val = po->tp_tx_has_off;
3624 case PACKET_QDISC_BYPASS:
3625 val = packet_use_direct_xmit(po);
3628 return -ENOPROTOOPT;
3633 if (put_user(len, optlen))
3635 if (copy_to_user(optval, data, len))
3641 static int packet_notifier(struct notifier_block *this,
3642 unsigned long msg, void *ptr)
3645 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3646 struct net *net = dev_net(dev);
3649 sk_for_each_rcu(sk, &net->packet.sklist) {
3650 struct packet_sock *po = pkt_sk(sk);
3653 case NETDEV_UNREGISTER:
3655 packet_dev_mclist_delete(dev, &po->mclist);
3659 if (dev->ifindex == po->ifindex) {
3660 spin_lock(&po->bind_lock);
3662 __unregister_prot_hook(sk, false);
3663 sk->sk_err = ENETDOWN;
3664 if (!sock_flag(sk, SOCK_DEAD))
3665 sk->sk_error_report(sk);
3667 if (msg == NETDEV_UNREGISTER) {
3668 packet_cached_dev_reset(po);
3670 if (po->prot_hook.dev)
3671 dev_put(po->prot_hook.dev);
3672 po->prot_hook.dev = NULL;
3674 spin_unlock(&po->bind_lock);
3678 if (dev->ifindex == po->ifindex) {
3679 spin_lock(&po->bind_lock);
3681 register_prot_hook(sk);
3682 spin_unlock(&po->bind_lock);
3692 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3695 struct sock *sk = sock->sk;
3700 int amount = sk_wmem_alloc_get(sk);
3702 return put_user(amount, (int __user *)arg);
3706 struct sk_buff *skb;
3709 spin_lock_bh(&sk->sk_receive_queue.lock);
3710 skb = skb_peek(&sk->sk_receive_queue);
3713 spin_unlock_bh(&sk->sk_receive_queue.lock);
3714 return put_user(amount, (int __user *)arg);
3717 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3719 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3729 case SIOCGIFBRDADDR:
3730 case SIOCSIFBRDADDR:
3731 case SIOCGIFNETMASK:
3732 case SIOCSIFNETMASK:
3733 case SIOCGIFDSTADDR:
3734 case SIOCSIFDSTADDR:
3736 return inet_dgram_ops.ioctl(sock, cmd, arg);
3740 return -ENOIOCTLCMD;
3745 static unsigned int packet_poll(struct file *file, struct socket *sock,
3748 struct sock *sk = sock->sk;
3749 struct packet_sock *po = pkt_sk(sk);
3750 unsigned int mask = datagram_poll(file, sock, wait);
3752 spin_lock_bh(&sk->sk_receive_queue.lock);
3753 if (po->rx_ring.pg_vec) {
3754 if (!packet_previous_rx_frame(po, &po->rx_ring,
3756 mask |= POLLIN | POLLRDNORM;
3758 spin_unlock_bh(&sk->sk_receive_queue.lock);
3759 spin_lock_bh(&sk->sk_write_queue.lock);
3760 if (po->tx_ring.pg_vec) {
3761 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3762 mask |= POLLOUT | POLLWRNORM;
3764 spin_unlock_bh(&sk->sk_write_queue.lock);
3769 /* Dirty? Well, I still did not learn better way to account
3773 static void packet_mm_open(struct vm_area_struct *vma)
3775 struct file *file = vma->vm_file;
3776 struct socket *sock = file->private_data;
3777 struct sock *sk = sock->sk;
3780 atomic_inc(&pkt_sk(sk)->mapped);
3783 static void packet_mm_close(struct vm_area_struct *vma)
3785 struct file *file = vma->vm_file;
3786 struct socket *sock = file->private_data;
3787 struct sock *sk = sock->sk;
3790 atomic_dec(&pkt_sk(sk)->mapped);
3793 static const struct vm_operations_struct packet_mmap_ops = {
3794 .open = packet_mm_open,
3795 .close = packet_mm_close,
3798 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3803 for (i = 0; i < len; i++) {
3804 if (likely(pg_vec[i].buffer)) {
3805 if (is_vmalloc_addr(pg_vec[i].buffer))
3806 vfree(pg_vec[i].buffer);
3808 free_pages((unsigned long)pg_vec[i].buffer,
3810 pg_vec[i].buffer = NULL;
3816 static char *alloc_one_pg_vec_page(unsigned long order)
3819 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3820 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3822 buffer = (char *) __get_free_pages(gfp_flags, order);
3826 /* __get_free_pages failed, fall back to vmalloc */
3827 buffer = vzalloc((1 << order) * PAGE_SIZE);
3831 /* vmalloc failed, lets dig into swap here */
3832 gfp_flags &= ~__GFP_NORETRY;
3833 buffer = (char *) __get_free_pages(gfp_flags, order);
3837 /* complete and utter failure */
3841 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3843 unsigned int block_nr = req->tp_block_nr;
3847 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3848 if (unlikely(!pg_vec))
3851 for (i = 0; i < block_nr; i++) {
3852 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3853 if (unlikely(!pg_vec[i].buffer))
3854 goto out_free_pgvec;
3861 free_pg_vec(pg_vec, order, block_nr);
3866 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3867 int closing, int tx_ring)
3869 struct pgv *pg_vec = NULL;
3870 struct packet_sock *po = pkt_sk(sk);
3871 int was_running, order = 0;
3872 struct packet_ring_buffer *rb;
3873 struct sk_buff_head *rb_queue;
3876 /* Added to avoid minimal code churn */
3877 struct tpacket_req *req = &req_u->req;
3879 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3880 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3881 WARN(1, "Tx-ring is not supported.\n");
3885 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3886 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3890 if (atomic_read(&po->mapped))
3892 if (packet_read_pending(rb))
3896 if (req->tp_block_nr) {
3897 /* Sanity tests and some calculations */
3899 if (unlikely(rb->pg_vec))
3902 switch (po->tp_version) {
3904 po->tp_hdrlen = TPACKET_HDRLEN;
3907 po->tp_hdrlen = TPACKET2_HDRLEN;
3910 po->tp_hdrlen = TPACKET3_HDRLEN;
3915 if (unlikely((int)req->tp_block_size <= 0))
3917 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3919 if (po->tp_version >= TPACKET_V3 &&
3920 (int)(req->tp_block_size -
3921 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3923 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3926 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3929 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3930 if (unlikely(rb->frames_per_block <= 0))
3932 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3937 order = get_order(req->tp_block_size);
3938 pg_vec = alloc_pg_vec(req, order);
3939 if (unlikely(!pg_vec))
3941 switch (po->tp_version) {
3943 /* Transmit path is not supported. We checked
3944 * it above but just being paranoid
3947 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3956 if (unlikely(req->tp_frame_nr))
3962 /* Detach socket from network */
3963 spin_lock(&po->bind_lock);
3964 was_running = po->running;
3968 __unregister_prot_hook(sk, false);
3970 spin_unlock(&po->bind_lock);
3975 mutex_lock(&po->pg_vec_lock);
3976 if (closing || atomic_read(&po->mapped) == 0) {
3978 spin_lock_bh(&rb_queue->lock);
3979 swap(rb->pg_vec, pg_vec);
3980 rb->frame_max = (req->tp_frame_nr - 1);
3982 rb->frame_size = req->tp_frame_size;
3983 spin_unlock_bh(&rb_queue->lock);
3985 swap(rb->pg_vec_order, order);
3986 swap(rb->pg_vec_len, req->tp_block_nr);
3988 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3989 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3990 tpacket_rcv : packet_rcv;
3991 skb_queue_purge(rb_queue);
3992 if (atomic_read(&po->mapped))
3993 pr_err("packet_mmap: vma is busy: %d\n",
3994 atomic_read(&po->mapped));
3996 mutex_unlock(&po->pg_vec_lock);
3998 spin_lock(&po->bind_lock);
4001 register_prot_hook(sk);
4003 spin_unlock(&po->bind_lock);
4004 if (closing && (po->tp_version > TPACKET_V2)) {
4005 /* Because we don't support block-based V3 on tx-ring */
4007 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
4012 free_pg_vec(pg_vec, order, req->tp_block_nr);
4017 static int packet_mmap(struct file *file, struct socket *sock,
4018 struct vm_area_struct *vma)
4020 struct sock *sk = sock->sk;
4021 struct packet_sock *po = pkt_sk(sk);
4022 unsigned long size, expected_size;
4023 struct packet_ring_buffer *rb;
4024 unsigned long start;
4031 mutex_lock(&po->pg_vec_lock);
4034 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4036 expected_size += rb->pg_vec_len
4042 if (expected_size == 0)
4045 size = vma->vm_end - vma->vm_start;
4046 if (size != expected_size)
4049 start = vma->vm_start;
4050 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4051 if (rb->pg_vec == NULL)
4054 for (i = 0; i < rb->pg_vec_len; i++) {
4056 void *kaddr = rb->pg_vec[i].buffer;
4059 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4060 page = pgv_to_page(kaddr);
4061 err = vm_insert_page(vma, start, page);
4070 atomic_inc(&po->mapped);
4071 vma->vm_ops = &packet_mmap_ops;
4075 mutex_unlock(&po->pg_vec_lock);
4079 static const struct proto_ops packet_ops_spkt = {
4080 .family = PF_PACKET,
4081 .owner = THIS_MODULE,
4082 .release = packet_release,
4083 .bind = packet_bind_spkt,
4084 .connect = sock_no_connect,
4085 .socketpair = sock_no_socketpair,
4086 .accept = sock_no_accept,
4087 .getname = packet_getname_spkt,
4088 .poll = datagram_poll,
4089 .ioctl = packet_ioctl,
4090 .listen = sock_no_listen,
4091 .shutdown = sock_no_shutdown,
4092 .setsockopt = sock_no_setsockopt,
4093 .getsockopt = sock_no_getsockopt,
4094 .sendmsg = packet_sendmsg_spkt,
4095 .recvmsg = packet_recvmsg,
4096 .mmap = sock_no_mmap,
4097 .sendpage = sock_no_sendpage,
4100 static const struct proto_ops packet_ops = {
4101 .family = PF_PACKET,
4102 .owner = THIS_MODULE,
4103 .release = packet_release,
4104 .bind = packet_bind,
4105 .connect = sock_no_connect,
4106 .socketpair = sock_no_socketpair,
4107 .accept = sock_no_accept,
4108 .getname = packet_getname,
4109 .poll = packet_poll,
4110 .ioctl = packet_ioctl,
4111 .listen = sock_no_listen,
4112 .shutdown = sock_no_shutdown,
4113 .setsockopt = packet_setsockopt,
4114 .getsockopt = packet_getsockopt,
4115 .sendmsg = packet_sendmsg,
4116 .recvmsg = packet_recvmsg,
4117 .mmap = packet_mmap,
4118 .sendpage = sock_no_sendpage,
4121 static const struct net_proto_family packet_family_ops = {
4122 .family = PF_PACKET,
4123 .create = packet_create,
4124 .owner = THIS_MODULE,
4127 static struct notifier_block packet_netdev_notifier = {
4128 .notifier_call = packet_notifier,
4131 #ifdef CONFIG_PROC_FS
4133 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4136 struct net *net = seq_file_net(seq);
4139 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4142 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4144 struct net *net = seq_file_net(seq);
4145 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4148 static void packet_seq_stop(struct seq_file *seq, void *v)
4154 static int packet_seq_show(struct seq_file *seq, void *v)
4156 if (v == SEQ_START_TOKEN)
4157 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4159 struct sock *s = sk_entry(v);
4160 const struct packet_sock *po = pkt_sk(s);
4163 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4165 atomic_read(&s->sk_refcnt),
4170 atomic_read(&s->sk_rmem_alloc),
4171 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4178 static const struct seq_operations packet_seq_ops = {
4179 .start = packet_seq_start,
4180 .next = packet_seq_next,
4181 .stop = packet_seq_stop,
4182 .show = packet_seq_show,
4185 static int packet_seq_open(struct inode *inode, struct file *file)
4187 return seq_open_net(inode, file, &packet_seq_ops,
4188 sizeof(struct seq_net_private));
4191 static const struct file_operations packet_seq_fops = {
4192 .owner = THIS_MODULE,
4193 .open = packet_seq_open,
4195 .llseek = seq_lseek,
4196 .release = seq_release_net,
4201 static int __net_init packet_net_init(struct net *net)
4203 mutex_init(&net->packet.sklist_lock);
4204 INIT_HLIST_HEAD(&net->packet.sklist);
4206 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4212 static void __net_exit packet_net_exit(struct net *net)
4214 remove_proc_entry("packet", net->proc_net);
4217 static struct pernet_operations packet_net_ops = {
4218 .init = packet_net_init,
4219 .exit = packet_net_exit,
4223 static void __exit packet_exit(void)
4225 unregister_netdevice_notifier(&packet_netdev_notifier);
4226 unregister_pernet_subsys(&packet_net_ops);
4227 sock_unregister(PF_PACKET);
4228 proto_unregister(&packet_proto);
4231 static int __init packet_init(void)
4233 int rc = proto_register(&packet_proto, 0);
4238 sock_register(&packet_family_ops);
4239 register_pernet_subsys(&packet_net_ops);
4240 register_netdevice_notifier(&packet_netdev_notifier);
4245 module_init(packet_init);
4246 module_exit(packet_exit);
4247 MODULE_LICENSE("GPL");
4248 MODULE_ALIAS_NETPROTO(PF_PACKET);