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 static bool packet_rcv_has_room(struct packet_sock *po, struct sk_buff *skb)
1239 struct sock *sk = &po->sk;
1242 if (po->prot_hook.func != tpacket_rcv)
1243 return (atomic_read(&sk->sk_rmem_alloc) + skb->truesize)
1246 spin_lock(&sk->sk_receive_queue.lock);
1247 if (po->tp_version == TPACKET_V3)
1248 has_room = prb_lookup_block(po, &po->rx_ring,
1249 po->rx_ring.prb_bdqc.kactive_blk_num,
1252 has_room = packet_lookup_frame(po, &po->rx_ring,
1255 spin_unlock(&sk->sk_receive_queue.lock);
1260 static void packet_sock_destruct(struct sock *sk)
1262 skb_queue_purge(&sk->sk_error_queue);
1264 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
1265 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
1267 if (!sock_flag(sk, SOCK_DEAD)) {
1268 pr_err("Attempt to release alive packet socket: %p\n", sk);
1272 sk_refcnt_debug_dec(sk);
1275 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1276 struct sk_buff *skb,
1279 return reciprocal_scale(skb_get_hash(skb), num);
1282 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1283 struct sk_buff *skb,
1286 unsigned int val = atomic_inc_return(&f->rr_cur);
1291 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1292 struct sk_buff *skb,
1295 return smp_processor_id() % num;
1298 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1299 struct sk_buff *skb,
1302 return prandom_u32_max(num);
1305 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1306 struct sk_buff *skb,
1307 unsigned int idx, unsigned int skip,
1312 i = j = min_t(int, f->next[idx], num - 1);
1314 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1326 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1327 struct sk_buff *skb,
1330 return skb_get_queue_mapping(skb) % num;
1333 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1335 return f->flags & (flag >> 8);
1338 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1339 struct packet_type *pt, struct net_device *orig_dev)
1341 struct packet_fanout *f = pt->af_packet_priv;
1342 unsigned int num = READ_ONCE(f->num_members);
1343 struct packet_sock *po;
1346 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1352 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1353 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1358 case PACKET_FANOUT_HASH:
1360 idx = fanout_demux_hash(f, skb, num);
1362 case PACKET_FANOUT_LB:
1363 idx = fanout_demux_lb(f, skb, num);
1365 case PACKET_FANOUT_CPU:
1366 idx = fanout_demux_cpu(f, skb, num);
1368 case PACKET_FANOUT_RND:
1369 idx = fanout_demux_rnd(f, skb, num);
1371 case PACKET_FANOUT_QM:
1372 idx = fanout_demux_qm(f, skb, num);
1374 case PACKET_FANOUT_ROLLOVER:
1375 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1379 po = pkt_sk(f->arr[idx]);
1380 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1381 unlikely(!packet_rcv_has_room(po, skb))) {
1382 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1383 po = pkt_sk(f->arr[idx]);
1386 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1389 DEFINE_MUTEX(fanout_mutex);
1390 EXPORT_SYMBOL_GPL(fanout_mutex);
1391 static LIST_HEAD(fanout_list);
1393 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1395 struct packet_fanout *f = po->fanout;
1397 spin_lock(&f->lock);
1398 f->arr[f->num_members] = sk;
1401 spin_unlock(&f->lock);
1404 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1406 struct packet_fanout *f = po->fanout;
1409 spin_lock(&f->lock);
1410 for (i = 0; i < f->num_members; i++) {
1411 if (f->arr[i] == sk)
1414 BUG_ON(i >= f->num_members);
1415 f->arr[i] = f->arr[f->num_members - 1];
1417 spin_unlock(&f->lock);
1420 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1422 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1428 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1430 struct packet_sock *po = pkt_sk(sk);
1431 struct packet_fanout *f, *match;
1432 u8 type = type_flags & 0xff;
1433 u8 flags = type_flags >> 8;
1437 case PACKET_FANOUT_ROLLOVER:
1438 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1440 case PACKET_FANOUT_HASH:
1441 case PACKET_FANOUT_LB:
1442 case PACKET_FANOUT_CPU:
1443 case PACKET_FANOUT_RND:
1444 case PACKET_FANOUT_QM:
1456 mutex_lock(&fanout_mutex);
1458 list_for_each_entry(f, &fanout_list, list) {
1460 read_pnet(&f->net) == sock_net(sk)) {
1466 if (match && match->flags != flags)
1470 match = kzalloc(sizeof(*match), GFP_KERNEL);
1473 write_pnet(&match->net, sock_net(sk));
1476 match->flags = flags;
1477 atomic_set(&match->rr_cur, 0);
1478 INIT_LIST_HEAD(&match->list);
1479 spin_lock_init(&match->lock);
1480 atomic_set(&match->sk_ref, 0);
1481 match->prot_hook.type = po->prot_hook.type;
1482 match->prot_hook.dev = po->prot_hook.dev;
1483 match->prot_hook.func = packet_rcv_fanout;
1484 match->prot_hook.af_packet_priv = match;
1485 match->prot_hook.id_match = match_fanout_group;
1486 dev_add_pack(&match->prot_hook);
1487 list_add(&match->list, &fanout_list);
1490 if (match->type == type &&
1491 match->prot_hook.type == po->prot_hook.type &&
1492 match->prot_hook.dev == po->prot_hook.dev) {
1494 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1495 __dev_remove_pack(&po->prot_hook);
1497 atomic_inc(&match->sk_ref);
1498 __fanout_link(sk, po);
1503 mutex_unlock(&fanout_mutex);
1507 static void fanout_release(struct sock *sk)
1509 struct packet_sock *po = pkt_sk(sk);
1510 struct packet_fanout *f;
1516 mutex_lock(&fanout_mutex);
1519 if (atomic_dec_and_test(&f->sk_ref)) {
1521 dev_remove_pack(&f->prot_hook);
1524 mutex_unlock(&fanout_mutex);
1527 static const struct proto_ops packet_ops;
1529 static const struct proto_ops packet_ops_spkt;
1531 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1532 struct packet_type *pt, struct net_device *orig_dev)
1535 struct sockaddr_pkt *spkt;
1538 * When we registered the protocol we saved the socket in the data
1539 * field for just this event.
1542 sk = pt->af_packet_priv;
1545 * Yank back the headers [hope the device set this
1546 * right or kerboom...]
1548 * Incoming packets have ll header pulled,
1551 * For outgoing ones skb->data == skb_mac_header(skb)
1552 * so that this procedure is noop.
1555 if (skb->pkt_type == PACKET_LOOPBACK)
1558 if (!net_eq(dev_net(dev), sock_net(sk)))
1561 skb = skb_share_check(skb, GFP_ATOMIC);
1565 /* drop any routing info */
1568 /* drop conntrack reference */
1571 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1573 skb_push(skb, skb->data - skb_mac_header(skb));
1576 * The SOCK_PACKET socket receives _all_ frames.
1579 spkt->spkt_family = dev->type;
1580 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1581 spkt->spkt_protocol = skb->protocol;
1584 * Charge the memory to the socket. This is done specifically
1585 * to prevent sockets using all the memory up.
1588 if (sock_queue_rcv_skb(sk, skb) == 0)
1599 * Output a raw packet to a device layer. This bypasses all the other
1600 * protocol layers and you must therefore supply it with a complete frame
1603 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1606 struct sock *sk = sock->sk;
1607 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1608 struct sk_buff *skb = NULL;
1609 struct net_device *dev;
1615 * Get and verify the address.
1619 if (msg->msg_namelen < sizeof(struct sockaddr))
1621 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1622 proto = saddr->spkt_protocol;
1624 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1627 * Find the device first to size check it
1630 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1633 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1639 if (!(dev->flags & IFF_UP))
1643 * You may not queue a frame bigger than the mtu. This is the lowest level
1644 * raw protocol and you must do your own fragmentation at this level.
1647 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1648 if (!netif_supports_nofcs(dev)) {
1649 err = -EPROTONOSUPPORT;
1652 extra_len = 4; /* We're doing our own CRC */
1656 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1660 size_t reserved = LL_RESERVED_SPACE(dev);
1661 int tlen = dev->needed_tailroom;
1662 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1665 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1668 /* FIXME: Save some space for broken drivers that write a hard
1669 * header at transmission time by themselves. PPP is the notable
1670 * one here. This should really be fixed at the driver level.
1672 skb_reserve(skb, reserved);
1673 skb_reset_network_header(skb);
1675 /* Try to align data part correctly */
1680 skb_reset_network_header(skb);
1682 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1688 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1689 /* Earlier code assumed this would be a VLAN pkt,
1690 * double-check this now that we have the actual
1693 struct ethhdr *ehdr;
1694 skb_reset_mac_header(skb);
1695 ehdr = eth_hdr(skb);
1696 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1702 skb->protocol = proto;
1704 skb->priority = sk->sk_priority;
1705 skb->mark = sk->sk_mark;
1707 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1709 if (unlikely(extra_len == 4))
1712 skb_probe_transport_header(skb, 0);
1714 dev_queue_xmit(skb);
1725 static unsigned int run_filter(const struct sk_buff *skb,
1726 const struct sock *sk,
1729 struct sk_filter *filter;
1732 filter = rcu_dereference(sk->sk_filter);
1734 res = SK_RUN_FILTER(filter, skb);
1741 * This function makes lazy skb cloning in hope that most of packets
1742 * are discarded by BPF.
1744 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1745 * and skb->cb are mangled. It works because (and until) packets
1746 * falling here are owned by current CPU. Output packets are cloned
1747 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1748 * sequencially, so that if we return skb to original state on exit,
1749 * we will not harm anyone.
1752 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1753 struct packet_type *pt, struct net_device *orig_dev)
1756 struct sockaddr_ll *sll;
1757 struct packet_sock *po;
1758 u8 *skb_head = skb->data;
1759 int skb_len = skb->len;
1760 unsigned int snaplen, res;
1762 if (skb->pkt_type == PACKET_LOOPBACK)
1765 sk = pt->af_packet_priv;
1768 if (!net_eq(dev_net(dev), sock_net(sk)))
1773 if (dev->header_ops) {
1774 /* The device has an explicit notion of ll header,
1775 * exported to higher levels.
1777 * Otherwise, the device hides details of its frame
1778 * structure, so that corresponding packet head is
1779 * never delivered to user.
1781 if (sk->sk_type != SOCK_DGRAM)
1782 skb_push(skb, skb->data - skb_mac_header(skb));
1783 else if (skb->pkt_type == PACKET_OUTGOING) {
1784 /* Special case: outgoing packets have ll header at head */
1785 skb_pull(skb, skb_network_offset(skb));
1791 res = run_filter(skb, sk, snaplen);
1793 goto drop_n_restore;
1797 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1800 if (skb_shared(skb)) {
1801 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1805 if (skb_head != skb->data) {
1806 skb->data = skb_head;
1813 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1815 sll = &PACKET_SKB_CB(skb)->sa.ll;
1816 sll->sll_hatype = dev->type;
1817 sll->sll_pkttype = skb->pkt_type;
1818 if (unlikely(po->origdev))
1819 sll->sll_ifindex = orig_dev->ifindex;
1821 sll->sll_ifindex = dev->ifindex;
1823 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1825 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1826 * Use their space for storing the original skb length.
1828 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1830 if (pskb_trim(skb, snaplen))
1833 skb_set_owner_r(skb, sk);
1837 /* drop conntrack reference */
1840 spin_lock(&sk->sk_receive_queue.lock);
1841 po->stats.stats1.tp_packets++;
1842 sock_skb_set_dropcount(sk, skb);
1843 __skb_queue_tail(&sk->sk_receive_queue, skb);
1844 spin_unlock(&sk->sk_receive_queue.lock);
1845 sk->sk_data_ready(sk);
1849 spin_lock(&sk->sk_receive_queue.lock);
1850 po->stats.stats1.tp_drops++;
1851 atomic_inc(&sk->sk_drops);
1852 spin_unlock(&sk->sk_receive_queue.lock);
1855 if (skb_head != skb->data && skb_shared(skb)) {
1856 skb->data = skb_head;
1864 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1865 struct packet_type *pt, struct net_device *orig_dev)
1868 struct packet_sock *po;
1869 struct sockaddr_ll *sll;
1870 union tpacket_uhdr h;
1871 u8 *skb_head = skb->data;
1872 int skb_len = skb->len;
1873 unsigned int snaplen, res;
1874 unsigned long status = TP_STATUS_USER;
1875 unsigned short macoff, netoff, hdrlen;
1876 struct sk_buff *copy_skb = NULL;
1880 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1881 * We may add members to them until current aligned size without forcing
1882 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1884 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1885 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1887 if (skb->pkt_type == PACKET_LOOPBACK)
1890 sk = pt->af_packet_priv;
1893 if (!net_eq(dev_net(dev), sock_net(sk)))
1896 if (dev->header_ops) {
1897 if (sk->sk_type != SOCK_DGRAM)
1898 skb_push(skb, skb->data - skb_mac_header(skb));
1899 else if (skb->pkt_type == PACKET_OUTGOING) {
1900 /* Special case: outgoing packets have ll header at head */
1901 skb_pull(skb, skb_network_offset(skb));
1907 res = run_filter(skb, sk, snaplen);
1909 goto drop_n_restore;
1911 if (skb->ip_summed == CHECKSUM_PARTIAL)
1912 status |= TP_STATUS_CSUMNOTREADY;
1913 else if (skb->pkt_type != PACKET_OUTGOING &&
1914 (skb->ip_summed == CHECKSUM_COMPLETE ||
1915 skb_csum_unnecessary(skb)))
1916 status |= TP_STATUS_CSUM_VALID;
1921 if (sk->sk_type == SOCK_DGRAM) {
1922 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1925 unsigned int maclen = skb_network_offset(skb);
1926 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1927 (maclen < 16 ? 16 : maclen)) +
1929 macoff = netoff - maclen;
1931 if (po->tp_version <= TPACKET_V2) {
1932 if (macoff + snaplen > po->rx_ring.frame_size) {
1933 if (po->copy_thresh &&
1934 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1935 if (skb_shared(skb)) {
1936 copy_skb = skb_clone(skb, GFP_ATOMIC);
1938 copy_skb = skb_get(skb);
1939 skb_head = skb->data;
1942 skb_set_owner_r(copy_skb, sk);
1944 snaplen = po->rx_ring.frame_size - macoff;
1945 if ((int)snaplen < 0)
1948 } else if (unlikely(macoff + snaplen >
1949 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1952 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1953 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1954 snaplen, nval, macoff);
1956 if (unlikely((int)snaplen < 0)) {
1958 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1961 spin_lock(&sk->sk_receive_queue.lock);
1962 h.raw = packet_current_rx_frame(po, skb,
1963 TP_STATUS_KERNEL, (macoff+snaplen));
1966 if (po->tp_version <= TPACKET_V2) {
1967 packet_increment_rx_head(po, &po->rx_ring);
1969 * LOSING will be reported till you read the stats,
1970 * because it's COR - Clear On Read.
1971 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1974 if (po->stats.stats1.tp_drops)
1975 status |= TP_STATUS_LOSING;
1977 po->stats.stats1.tp_packets++;
1979 status |= TP_STATUS_COPY;
1980 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1982 spin_unlock(&sk->sk_receive_queue.lock);
1984 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
1986 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
1987 getnstimeofday(&ts);
1989 status |= ts_status;
1991 switch (po->tp_version) {
1993 h.h1->tp_len = skb->len;
1994 h.h1->tp_snaplen = snaplen;
1995 h.h1->tp_mac = macoff;
1996 h.h1->tp_net = netoff;
1997 h.h1->tp_sec = ts.tv_sec;
1998 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
1999 hdrlen = sizeof(*h.h1);
2002 h.h2->tp_len = skb->len;
2003 h.h2->tp_snaplen = snaplen;
2004 h.h2->tp_mac = macoff;
2005 h.h2->tp_net = netoff;
2006 h.h2->tp_sec = ts.tv_sec;
2007 h.h2->tp_nsec = ts.tv_nsec;
2008 if (skb_vlan_tag_present(skb)) {
2009 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2010 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2011 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2013 h.h2->tp_vlan_tci = 0;
2014 h.h2->tp_vlan_tpid = 0;
2016 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2017 hdrlen = sizeof(*h.h2);
2020 /* tp_nxt_offset,vlan are already populated above.
2021 * So DONT clear those fields here
2023 h.h3->tp_status |= status;
2024 h.h3->tp_len = skb->len;
2025 h.h3->tp_snaplen = snaplen;
2026 h.h3->tp_mac = macoff;
2027 h.h3->tp_net = netoff;
2028 h.h3->tp_sec = ts.tv_sec;
2029 h.h3->tp_nsec = ts.tv_nsec;
2030 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2031 hdrlen = sizeof(*h.h3);
2037 sll = h.raw + TPACKET_ALIGN(hdrlen);
2038 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2039 sll->sll_family = AF_PACKET;
2040 sll->sll_hatype = dev->type;
2041 sll->sll_protocol = skb->protocol;
2042 sll->sll_pkttype = skb->pkt_type;
2043 if (unlikely(po->origdev))
2044 sll->sll_ifindex = orig_dev->ifindex;
2046 sll->sll_ifindex = dev->ifindex;
2050 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2051 if (po->tp_version <= TPACKET_V2) {
2054 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2057 for (start = h.raw; start < end; start += PAGE_SIZE)
2058 flush_dcache_page(pgv_to_page(start));
2063 if (po->tp_version <= TPACKET_V2) {
2064 __packet_set_status(po, h.raw, status);
2065 sk->sk_data_ready(sk);
2067 prb_clear_blk_fill_status(&po->rx_ring);
2071 if (skb_head != skb->data && skb_shared(skb)) {
2072 skb->data = skb_head;
2080 po->stats.stats1.tp_drops++;
2081 spin_unlock(&sk->sk_receive_queue.lock);
2083 sk->sk_data_ready(sk);
2084 kfree_skb(copy_skb);
2085 goto drop_n_restore;
2088 static void tpacket_destruct_skb(struct sk_buff *skb)
2090 struct packet_sock *po = pkt_sk(skb->sk);
2092 if (likely(po->tx_ring.pg_vec)) {
2096 ph = skb_shinfo(skb)->destructor_arg;
2097 packet_dec_pending(&po->tx_ring);
2099 ts = __packet_set_timestamp(po, ph, skb);
2100 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2106 static bool ll_header_truncated(const struct net_device *dev, int len)
2108 /* net device doesn't like empty head */
2109 if (unlikely(len <= dev->hard_header_len)) {
2110 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2111 current->comm, len, dev->hard_header_len);
2118 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2119 void *frame, struct net_device *dev, int size_max,
2120 __be16 proto, unsigned char *addr, int hlen)
2122 union tpacket_uhdr ph;
2123 int to_write, offset, len, tp_len, nr_frags, len_max;
2124 struct socket *sock = po->sk.sk_socket;
2131 skb->protocol = proto;
2133 skb->priority = po->sk.sk_priority;
2134 skb->mark = po->sk.sk_mark;
2135 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2136 skb_shinfo(skb)->destructor_arg = ph.raw;
2138 switch (po->tp_version) {
2140 tp_len = ph.h2->tp_len;
2143 tp_len = ph.h1->tp_len;
2146 if (unlikely(tp_len > size_max)) {
2147 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2151 skb_reserve(skb, hlen);
2152 skb_reset_network_header(skb);
2154 if (!packet_use_direct_xmit(po))
2155 skb_probe_transport_header(skb, 0);
2156 if (unlikely(po->tp_tx_has_off)) {
2157 int off_min, off_max, off;
2158 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2159 off_max = po->tx_ring.frame_size - tp_len;
2160 if (sock->type == SOCK_DGRAM) {
2161 switch (po->tp_version) {
2163 off = ph.h2->tp_net;
2166 off = ph.h1->tp_net;
2170 switch (po->tp_version) {
2172 off = ph.h2->tp_mac;
2175 off = ph.h1->tp_mac;
2179 if (unlikely((off < off_min) || (off_max < off)))
2181 data = ph.raw + off;
2183 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2187 if (sock->type == SOCK_DGRAM) {
2188 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2190 if (unlikely(err < 0))
2192 } else if (dev->hard_header_len) {
2193 if (ll_header_truncated(dev, tp_len))
2196 skb_push(skb, dev->hard_header_len);
2197 err = skb_store_bits(skb, 0, data,
2198 dev->hard_header_len);
2202 data += dev->hard_header_len;
2203 to_write -= dev->hard_header_len;
2206 offset = offset_in_page(data);
2207 len_max = PAGE_SIZE - offset;
2208 len = ((to_write > len_max) ? len_max : to_write);
2210 skb->data_len = to_write;
2211 skb->len += to_write;
2212 skb->truesize += to_write;
2213 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2215 while (likely(to_write)) {
2216 nr_frags = skb_shinfo(skb)->nr_frags;
2218 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2219 pr_err("Packet exceed the number of skb frags(%lu)\n",
2224 page = pgv_to_page(data);
2226 flush_dcache_page(page);
2228 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2231 len_max = PAGE_SIZE;
2232 len = ((to_write > len_max) ? len_max : to_write);
2238 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2240 struct sk_buff *skb;
2241 struct net_device *dev;
2243 int err, reserve = 0;
2245 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2246 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2247 int tp_len, size_max;
2248 unsigned char *addr;
2250 int status = TP_STATUS_AVAILABLE;
2253 mutex_lock(&po->pg_vec_lock);
2255 if (likely(saddr == NULL)) {
2256 dev = packet_cached_dev_get(po);
2261 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2263 if (msg->msg_namelen < (saddr->sll_halen
2264 + offsetof(struct sockaddr_ll,
2267 proto = saddr->sll_protocol;
2268 addr = saddr->sll_addr;
2269 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2273 if (unlikely(dev == NULL))
2276 if (unlikely(!(dev->flags & IFF_UP)))
2279 reserve = dev->hard_header_len + VLAN_HLEN;
2280 size_max = po->tx_ring.frame_size
2281 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2283 if (size_max > dev->mtu + reserve)
2284 size_max = dev->mtu + reserve;
2287 ph = packet_current_frame(po, &po->tx_ring,
2288 TP_STATUS_SEND_REQUEST);
2289 if (unlikely(ph == NULL)) {
2290 if (need_wait && need_resched())
2295 status = TP_STATUS_SEND_REQUEST;
2296 hlen = LL_RESERVED_SPACE(dev);
2297 tlen = dev->needed_tailroom;
2298 skb = sock_alloc_send_skb(&po->sk,
2299 hlen + tlen + sizeof(struct sockaddr_ll),
2302 if (unlikely(skb == NULL)) {
2303 /* we assume the socket was initially writeable ... */
2304 if (likely(len_sum > 0))
2308 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2310 if (tp_len > dev->mtu + dev->hard_header_len) {
2311 struct ethhdr *ehdr;
2312 /* Earlier code assumed this would be a VLAN pkt,
2313 * double-check this now that we have the actual
2317 skb_reset_mac_header(skb);
2318 ehdr = eth_hdr(skb);
2319 if (ehdr->h_proto != htons(ETH_P_8021Q))
2322 if (unlikely(tp_len < 0)) {
2324 __packet_set_status(po, ph,
2325 TP_STATUS_AVAILABLE);
2326 packet_increment_head(&po->tx_ring);
2330 status = TP_STATUS_WRONG_FORMAT;
2336 packet_pick_tx_queue(dev, skb);
2338 skb->destructor = tpacket_destruct_skb;
2339 __packet_set_status(po, ph, TP_STATUS_SENDING);
2340 packet_inc_pending(&po->tx_ring);
2342 status = TP_STATUS_SEND_REQUEST;
2343 err = po->xmit(skb);
2344 if (unlikely(err > 0)) {
2345 err = net_xmit_errno(err);
2346 if (err && __packet_get_status(po, ph) ==
2347 TP_STATUS_AVAILABLE) {
2348 /* skb was destructed already */
2353 * skb was dropped but not destructed yet;
2354 * let's treat it like congestion or err < 0
2358 packet_increment_head(&po->tx_ring);
2360 } while (likely((ph != NULL) ||
2361 /* Note: packet_read_pending() might be slow if we have
2362 * to call it as it's per_cpu variable, but in fast-path
2363 * we already short-circuit the loop with the first
2364 * condition, and luckily don't have to go that path
2367 (need_wait && packet_read_pending(&po->tx_ring))));
2373 __packet_set_status(po, ph, status);
2378 mutex_unlock(&po->pg_vec_lock);
2382 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2383 size_t reserve, size_t len,
2384 size_t linear, int noblock,
2387 struct sk_buff *skb;
2389 /* Under a page? Don't bother with paged skb. */
2390 if (prepad + len < PAGE_SIZE || !linear)
2393 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2398 skb_reserve(skb, reserve);
2399 skb_put(skb, linear);
2400 skb->data_len = len - linear;
2401 skb->len += len - linear;
2406 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2408 struct sock *sk = sock->sk;
2409 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2410 struct sk_buff *skb;
2411 struct net_device *dev;
2413 unsigned char *addr;
2414 int err, reserve = 0;
2415 struct virtio_net_hdr vnet_hdr = { 0 };
2418 struct packet_sock *po = pkt_sk(sk);
2419 unsigned short gso_type = 0;
2425 * Get and verify the address.
2428 if (likely(saddr == NULL)) {
2429 dev = packet_cached_dev_get(po);
2434 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2436 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2438 proto = saddr->sll_protocol;
2439 addr = saddr->sll_addr;
2440 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2444 if (unlikely(dev == NULL))
2447 if (unlikely(!(dev->flags & IFF_UP)))
2450 if (sock->type == SOCK_RAW)
2451 reserve = dev->hard_header_len;
2452 if (po->has_vnet_hdr) {
2453 vnet_hdr_len = sizeof(vnet_hdr);
2456 if (len < vnet_hdr_len)
2459 len -= vnet_hdr_len;
2462 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2463 if (n != vnet_hdr_len)
2466 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2467 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2468 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2469 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2470 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2471 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2472 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2475 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2478 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2479 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2480 case VIRTIO_NET_HDR_GSO_TCPV4:
2481 gso_type = SKB_GSO_TCPV4;
2483 case VIRTIO_NET_HDR_GSO_TCPV6:
2484 gso_type = SKB_GSO_TCPV6;
2486 case VIRTIO_NET_HDR_GSO_UDP:
2487 gso_type = SKB_GSO_UDP;
2493 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2494 gso_type |= SKB_GSO_TCP_ECN;
2496 if (vnet_hdr.gso_size == 0)
2502 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2503 if (!netif_supports_nofcs(dev)) {
2504 err = -EPROTONOSUPPORT;
2507 extra_len = 4; /* We're doing our own CRC */
2511 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2515 hlen = LL_RESERVED_SPACE(dev);
2516 tlen = dev->needed_tailroom;
2517 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2518 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2519 msg->msg_flags & MSG_DONTWAIT, &err);
2523 skb_set_network_header(skb, reserve);
2526 if (sock->type == SOCK_DGRAM) {
2527 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2528 if (unlikely(offset < 0))
2531 if (ll_header_truncated(dev, len))
2535 /* Returns -EFAULT on error */
2536 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2540 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2542 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2543 /* Earlier code assumed this would be a VLAN pkt,
2544 * double-check this now that we have the actual
2547 struct ethhdr *ehdr;
2548 skb_reset_mac_header(skb);
2549 ehdr = eth_hdr(skb);
2550 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2556 skb->protocol = proto;
2558 skb->priority = sk->sk_priority;
2559 skb->mark = sk->sk_mark;
2561 packet_pick_tx_queue(dev, skb);
2563 if (po->has_vnet_hdr) {
2564 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2565 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2566 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2567 if (!skb_partial_csum_set(skb, s, o)) {
2573 skb_shinfo(skb)->gso_size =
2574 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2575 skb_shinfo(skb)->gso_type = gso_type;
2577 /* Header must be checked, and gso_segs computed. */
2578 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2579 skb_shinfo(skb)->gso_segs = 0;
2581 len += vnet_hdr_len;
2584 if (!packet_use_direct_xmit(po))
2585 skb_probe_transport_header(skb, reserve);
2586 if (unlikely(extra_len == 4))
2589 err = po->xmit(skb);
2590 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2606 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2608 struct sock *sk = sock->sk;
2609 struct packet_sock *po = pkt_sk(sk);
2611 if (po->tx_ring.pg_vec)
2612 return tpacket_snd(po, msg);
2614 return packet_snd(sock, msg, len);
2618 * Close a PACKET socket. This is fairly simple. We immediately go
2619 * to 'closed' state and remove our protocol entry in the device list.
2622 static int packet_release(struct socket *sock)
2624 struct sock *sk = sock->sk;
2625 struct packet_sock *po;
2627 union tpacket_req_u req_u;
2635 mutex_lock(&net->packet.sklist_lock);
2636 sk_del_node_init_rcu(sk);
2637 mutex_unlock(&net->packet.sklist_lock);
2640 sock_prot_inuse_add(net, sk->sk_prot, -1);
2643 spin_lock(&po->bind_lock);
2644 unregister_prot_hook(sk, false);
2645 packet_cached_dev_reset(po);
2647 if (po->prot_hook.dev) {
2648 dev_put(po->prot_hook.dev);
2649 po->prot_hook.dev = NULL;
2651 spin_unlock(&po->bind_lock);
2653 packet_flush_mclist(sk);
2655 if (po->rx_ring.pg_vec) {
2656 memset(&req_u, 0, sizeof(req_u));
2657 packet_set_ring(sk, &req_u, 1, 0);
2660 if (po->tx_ring.pg_vec) {
2661 memset(&req_u, 0, sizeof(req_u));
2662 packet_set_ring(sk, &req_u, 1, 1);
2669 * Now the socket is dead. No more input will appear.
2676 skb_queue_purge(&sk->sk_receive_queue);
2677 packet_free_pending(po);
2678 sk_refcnt_debug_release(sk);
2685 * Attach a packet hook.
2688 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2690 struct packet_sock *po = pkt_sk(sk);
2691 const struct net_device *dev_curr;
2703 spin_lock(&po->bind_lock);
2705 proto_curr = po->prot_hook.type;
2706 dev_curr = po->prot_hook.dev;
2708 need_rehook = proto_curr != proto || dev_curr != dev;
2711 unregister_prot_hook(sk, true);
2714 po->prot_hook.type = proto;
2716 if (po->prot_hook.dev)
2717 dev_put(po->prot_hook.dev);
2719 po->prot_hook.dev = dev;
2721 po->ifindex = dev ? dev->ifindex : 0;
2722 packet_cached_dev_assign(po, dev);
2725 if (proto == 0 || !need_rehook)
2728 if (!dev || (dev->flags & IFF_UP)) {
2729 register_prot_hook(sk);
2731 sk->sk_err = ENETDOWN;
2732 if (!sock_flag(sk, SOCK_DEAD))
2733 sk->sk_error_report(sk);
2737 spin_unlock(&po->bind_lock);
2743 * Bind a packet socket to a device
2746 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2749 struct sock *sk = sock->sk;
2751 struct net_device *dev;
2758 if (addr_len != sizeof(struct sockaddr))
2760 strlcpy(name, uaddr->sa_data, sizeof(name));
2762 dev = dev_get_by_name(sock_net(sk), name);
2764 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2768 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2770 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2771 struct sock *sk = sock->sk;
2772 struct net_device *dev = NULL;
2780 if (addr_len < sizeof(struct sockaddr_ll))
2782 if (sll->sll_family != AF_PACKET)
2785 if (sll->sll_ifindex) {
2787 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2791 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2797 static struct proto packet_proto = {
2799 .owner = THIS_MODULE,
2800 .obj_size = sizeof(struct packet_sock),
2804 * Create a packet of type SOCK_PACKET.
2807 static int packet_create(struct net *net, struct socket *sock, int protocol,
2811 struct packet_sock *po;
2812 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2815 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2817 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2818 sock->type != SOCK_PACKET)
2819 return -ESOCKTNOSUPPORT;
2821 sock->state = SS_UNCONNECTED;
2824 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2828 sock->ops = &packet_ops;
2829 if (sock->type == SOCK_PACKET)
2830 sock->ops = &packet_ops_spkt;
2832 sock_init_data(sock, sk);
2835 sk->sk_family = PF_PACKET;
2837 po->xmit = dev_queue_xmit;
2839 err = packet_alloc_pending(po);
2843 packet_cached_dev_reset(po);
2845 sk->sk_destruct = packet_sock_destruct;
2846 sk_refcnt_debug_inc(sk);
2849 * Attach a protocol block
2852 spin_lock_init(&po->bind_lock);
2853 mutex_init(&po->pg_vec_lock);
2854 po->prot_hook.func = packet_rcv;
2856 if (sock->type == SOCK_PACKET)
2857 po->prot_hook.func = packet_rcv_spkt;
2859 po->prot_hook.af_packet_priv = sk;
2862 po->prot_hook.type = proto;
2863 register_prot_hook(sk);
2866 mutex_lock(&net->packet.sklist_lock);
2867 sk_add_node_rcu(sk, &net->packet.sklist);
2868 mutex_unlock(&net->packet.sklist_lock);
2871 sock_prot_inuse_add(net, &packet_proto, 1);
2882 * Pull a packet from our receive queue and hand it to the user.
2883 * If necessary we block.
2886 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2889 struct sock *sk = sock->sk;
2890 struct sk_buff *skb;
2892 int vnet_hdr_len = 0;
2893 unsigned int origlen = 0;
2896 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2900 /* What error should we return now? EUNATTACH? */
2901 if (pkt_sk(sk)->ifindex < 0)
2905 if (flags & MSG_ERRQUEUE) {
2906 err = sock_recv_errqueue(sk, msg, len,
2907 SOL_PACKET, PACKET_TX_TIMESTAMP);
2912 * Call the generic datagram receiver. This handles all sorts
2913 * of horrible races and re-entrancy so we can forget about it
2914 * in the protocol layers.
2916 * Now it will return ENETDOWN, if device have just gone down,
2917 * but then it will block.
2920 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2923 * An error occurred so return it. Because skb_recv_datagram()
2924 * handles the blocking we don't see and worry about blocking
2931 if (pkt_sk(sk)->has_vnet_hdr) {
2932 struct virtio_net_hdr vnet_hdr = { 0 };
2935 vnet_hdr_len = sizeof(vnet_hdr);
2936 if (len < vnet_hdr_len)
2939 len -= vnet_hdr_len;
2941 if (skb_is_gso(skb)) {
2942 struct skb_shared_info *sinfo = skb_shinfo(skb);
2944 /* This is a hint as to how much should be linear. */
2946 __cpu_to_virtio16(false, skb_headlen(skb));
2948 __cpu_to_virtio16(false, sinfo->gso_size);
2949 if (sinfo->gso_type & SKB_GSO_TCPV4)
2950 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2951 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2952 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2953 else if (sinfo->gso_type & SKB_GSO_UDP)
2954 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2955 else if (sinfo->gso_type & SKB_GSO_FCOE)
2959 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2960 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2962 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2964 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2965 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2966 vnet_hdr.csum_start = __cpu_to_virtio16(false,
2967 skb_checksum_start_offset(skb));
2968 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
2970 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2971 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2972 } /* else everything is zero */
2974 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
2979 /* You lose any data beyond the buffer you gave. If it worries
2980 * a user program they can ask the device for its MTU
2986 msg->msg_flags |= MSG_TRUNC;
2989 err = skb_copy_datagram_msg(skb, 0, msg, copied);
2993 if (sock->type != SOCK_PACKET) {
2994 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
2996 /* Original length was stored in sockaddr_ll fields */
2997 origlen = PACKET_SKB_CB(skb)->sa.origlen;
2998 sll->sll_family = AF_PACKET;
2999 sll->sll_protocol = skb->protocol;
3002 sock_recv_ts_and_drops(msg, sk, skb);
3004 if (msg->msg_name) {
3005 /* If the address length field is there to be filled
3006 * in, we fill it in now.
3008 if (sock->type == SOCK_PACKET) {
3009 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3010 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3012 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3014 msg->msg_namelen = sll->sll_halen +
3015 offsetof(struct sockaddr_ll, sll_addr);
3017 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3021 if (pkt_sk(sk)->auxdata) {
3022 struct tpacket_auxdata aux;
3024 aux.tp_status = TP_STATUS_USER;
3025 if (skb->ip_summed == CHECKSUM_PARTIAL)
3026 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3027 else if (skb->pkt_type != PACKET_OUTGOING &&
3028 (skb->ip_summed == CHECKSUM_COMPLETE ||
3029 skb_csum_unnecessary(skb)))
3030 aux.tp_status |= TP_STATUS_CSUM_VALID;
3032 aux.tp_len = origlen;
3033 aux.tp_snaplen = skb->len;
3035 aux.tp_net = skb_network_offset(skb);
3036 if (skb_vlan_tag_present(skb)) {
3037 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3038 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3039 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3041 aux.tp_vlan_tci = 0;
3042 aux.tp_vlan_tpid = 0;
3044 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3048 * Free or return the buffer as appropriate. Again this
3049 * hides all the races and re-entrancy issues from us.
3051 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3054 skb_free_datagram(sk, skb);
3059 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3060 int *uaddr_len, int peer)
3062 struct net_device *dev;
3063 struct sock *sk = sock->sk;
3068 uaddr->sa_family = AF_PACKET;
3069 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3071 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3073 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3075 *uaddr_len = sizeof(*uaddr);
3080 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3081 int *uaddr_len, int peer)
3083 struct net_device *dev;
3084 struct sock *sk = sock->sk;
3085 struct packet_sock *po = pkt_sk(sk);
3086 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3091 sll->sll_family = AF_PACKET;
3092 sll->sll_ifindex = po->ifindex;
3093 sll->sll_protocol = po->num;
3094 sll->sll_pkttype = 0;
3096 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3098 sll->sll_hatype = dev->type;
3099 sll->sll_halen = dev->addr_len;
3100 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3102 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3106 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3111 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3115 case PACKET_MR_MULTICAST:
3116 if (i->alen != dev->addr_len)
3119 return dev_mc_add(dev, i->addr);
3121 return dev_mc_del(dev, i->addr);
3123 case PACKET_MR_PROMISC:
3124 return dev_set_promiscuity(dev, what);
3125 case PACKET_MR_ALLMULTI:
3126 return dev_set_allmulti(dev, what);
3127 case PACKET_MR_UNICAST:
3128 if (i->alen != dev->addr_len)
3131 return dev_uc_add(dev, i->addr);
3133 return dev_uc_del(dev, i->addr);
3141 static void packet_dev_mclist_delete(struct net_device *dev,
3142 struct packet_mclist **mlp)
3144 struct packet_mclist *ml;
3146 while ((ml = *mlp) != NULL) {
3147 if (ml->ifindex == dev->ifindex) {
3148 packet_dev_mc(dev, ml, -1);
3156 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3158 struct packet_sock *po = pkt_sk(sk);
3159 struct packet_mclist *ml, *i;
3160 struct net_device *dev;
3166 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3171 if (mreq->mr_alen > dev->addr_len)
3175 i = kmalloc(sizeof(*i), GFP_KERNEL);
3180 for (ml = po->mclist; ml; ml = ml->next) {
3181 if (ml->ifindex == mreq->mr_ifindex &&
3182 ml->type == mreq->mr_type &&
3183 ml->alen == mreq->mr_alen &&
3184 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3186 /* Free the new element ... */
3192 i->type = mreq->mr_type;
3193 i->ifindex = mreq->mr_ifindex;
3194 i->alen = mreq->mr_alen;
3195 memcpy(i->addr, mreq->mr_address, i->alen);
3197 i->next = po->mclist;
3199 err = packet_dev_mc(dev, i, 1);
3201 po->mclist = i->next;
3210 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3212 struct packet_mclist *ml, **mlp;
3216 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3217 if (ml->ifindex == mreq->mr_ifindex &&
3218 ml->type == mreq->mr_type &&
3219 ml->alen == mreq->mr_alen &&
3220 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3221 if (--ml->count == 0) {
3222 struct net_device *dev;
3224 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3226 packet_dev_mc(dev, ml, -1);
3236 static void packet_flush_mclist(struct sock *sk)
3238 struct packet_sock *po = pkt_sk(sk);
3239 struct packet_mclist *ml;
3245 while ((ml = po->mclist) != NULL) {
3246 struct net_device *dev;
3248 po->mclist = ml->next;
3249 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3251 packet_dev_mc(dev, ml, -1);
3258 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3260 struct sock *sk = sock->sk;
3261 struct packet_sock *po = pkt_sk(sk);
3264 if (level != SOL_PACKET)
3265 return -ENOPROTOOPT;
3268 case PACKET_ADD_MEMBERSHIP:
3269 case PACKET_DROP_MEMBERSHIP:
3271 struct packet_mreq_max mreq;
3273 memset(&mreq, 0, sizeof(mreq));
3274 if (len < sizeof(struct packet_mreq))
3276 if (len > sizeof(mreq))
3278 if (copy_from_user(&mreq, optval, len))
3280 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3282 if (optname == PACKET_ADD_MEMBERSHIP)
3283 ret = packet_mc_add(sk, &mreq);
3285 ret = packet_mc_drop(sk, &mreq);
3289 case PACKET_RX_RING:
3290 case PACKET_TX_RING:
3292 union tpacket_req_u req_u;
3295 switch (po->tp_version) {
3298 len = sizeof(req_u.req);
3302 len = sizeof(req_u.req3);
3307 if (pkt_sk(sk)->has_vnet_hdr)
3309 if (copy_from_user(&req_u.req, optval, len))
3311 return packet_set_ring(sk, &req_u, 0,
3312 optname == PACKET_TX_RING);
3314 case PACKET_COPY_THRESH:
3318 if (optlen != sizeof(val))
3320 if (copy_from_user(&val, optval, sizeof(val)))
3323 pkt_sk(sk)->copy_thresh = val;
3326 case PACKET_VERSION:
3330 if (optlen != sizeof(val))
3332 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3334 if (copy_from_user(&val, optval, sizeof(val)))
3340 po->tp_version = val;
3346 case PACKET_RESERVE:
3350 if (optlen != sizeof(val))
3352 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3354 if (copy_from_user(&val, optval, sizeof(val)))
3356 po->tp_reserve = val;
3363 if (optlen != sizeof(val))
3365 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3367 if (copy_from_user(&val, optval, sizeof(val)))
3369 po->tp_loss = !!val;
3372 case PACKET_AUXDATA:
3376 if (optlen < sizeof(val))
3378 if (copy_from_user(&val, optval, sizeof(val)))
3381 po->auxdata = !!val;
3384 case PACKET_ORIGDEV:
3388 if (optlen < sizeof(val))
3390 if (copy_from_user(&val, optval, sizeof(val)))
3393 po->origdev = !!val;
3396 case PACKET_VNET_HDR:
3400 if (sock->type != SOCK_RAW)
3402 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3404 if (optlen < sizeof(val))
3406 if (copy_from_user(&val, optval, sizeof(val)))
3409 po->has_vnet_hdr = !!val;
3412 case PACKET_TIMESTAMP:
3416 if (optlen != sizeof(val))
3418 if (copy_from_user(&val, optval, sizeof(val)))
3421 po->tp_tstamp = val;
3428 if (optlen != sizeof(val))
3430 if (copy_from_user(&val, optval, sizeof(val)))
3433 return fanout_add(sk, val & 0xffff, val >> 16);
3435 case PACKET_TX_HAS_OFF:
3439 if (optlen != sizeof(val))
3441 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3443 if (copy_from_user(&val, optval, sizeof(val)))
3445 po->tp_tx_has_off = !!val;
3448 case PACKET_QDISC_BYPASS:
3452 if (optlen != sizeof(val))
3454 if (copy_from_user(&val, optval, sizeof(val)))
3457 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3461 return -ENOPROTOOPT;
3465 static int packet_getsockopt(struct socket *sock, int level, int optname,
3466 char __user *optval, int __user *optlen)
3469 int val, lv = sizeof(val);
3470 struct sock *sk = sock->sk;
3471 struct packet_sock *po = pkt_sk(sk);
3473 union tpacket_stats_u st;
3475 if (level != SOL_PACKET)
3476 return -ENOPROTOOPT;
3478 if (get_user(len, optlen))
3485 case PACKET_STATISTICS:
3486 spin_lock_bh(&sk->sk_receive_queue.lock);
3487 memcpy(&st, &po->stats, sizeof(st));
3488 memset(&po->stats, 0, sizeof(po->stats));
3489 spin_unlock_bh(&sk->sk_receive_queue.lock);
3491 if (po->tp_version == TPACKET_V3) {
3492 lv = sizeof(struct tpacket_stats_v3);
3493 st.stats3.tp_packets += st.stats3.tp_drops;
3496 lv = sizeof(struct tpacket_stats);
3497 st.stats1.tp_packets += st.stats1.tp_drops;
3502 case PACKET_AUXDATA:
3505 case PACKET_ORIGDEV:
3508 case PACKET_VNET_HDR:
3509 val = po->has_vnet_hdr;
3511 case PACKET_VERSION:
3512 val = po->tp_version;
3515 if (len > sizeof(int))
3517 if (copy_from_user(&val, optval, len))
3521 val = sizeof(struct tpacket_hdr);
3524 val = sizeof(struct tpacket2_hdr);
3527 val = sizeof(struct tpacket3_hdr);
3533 case PACKET_RESERVE:
3534 val = po->tp_reserve;
3539 case PACKET_TIMESTAMP:
3540 val = po->tp_tstamp;
3544 ((u32)po->fanout->id |
3545 ((u32)po->fanout->type << 16) |
3546 ((u32)po->fanout->flags << 24)) :
3549 case PACKET_TX_HAS_OFF:
3550 val = po->tp_tx_has_off;
3552 case PACKET_QDISC_BYPASS:
3553 val = packet_use_direct_xmit(po);
3556 return -ENOPROTOOPT;
3561 if (put_user(len, optlen))
3563 if (copy_to_user(optval, data, len))
3569 static int packet_notifier(struct notifier_block *this,
3570 unsigned long msg, void *ptr)
3573 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3574 struct net *net = dev_net(dev);
3577 sk_for_each_rcu(sk, &net->packet.sklist) {
3578 struct packet_sock *po = pkt_sk(sk);
3581 case NETDEV_UNREGISTER:
3583 packet_dev_mclist_delete(dev, &po->mclist);
3587 if (dev->ifindex == po->ifindex) {
3588 spin_lock(&po->bind_lock);
3590 __unregister_prot_hook(sk, false);
3591 sk->sk_err = ENETDOWN;
3592 if (!sock_flag(sk, SOCK_DEAD))
3593 sk->sk_error_report(sk);
3595 if (msg == NETDEV_UNREGISTER) {
3596 packet_cached_dev_reset(po);
3598 if (po->prot_hook.dev)
3599 dev_put(po->prot_hook.dev);
3600 po->prot_hook.dev = NULL;
3602 spin_unlock(&po->bind_lock);
3606 if (dev->ifindex == po->ifindex) {
3607 spin_lock(&po->bind_lock);
3609 register_prot_hook(sk);
3610 spin_unlock(&po->bind_lock);
3620 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3623 struct sock *sk = sock->sk;
3628 int amount = sk_wmem_alloc_get(sk);
3630 return put_user(amount, (int __user *)arg);
3634 struct sk_buff *skb;
3637 spin_lock_bh(&sk->sk_receive_queue.lock);
3638 skb = skb_peek(&sk->sk_receive_queue);
3641 spin_unlock_bh(&sk->sk_receive_queue.lock);
3642 return put_user(amount, (int __user *)arg);
3645 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3647 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3657 case SIOCGIFBRDADDR:
3658 case SIOCSIFBRDADDR:
3659 case SIOCGIFNETMASK:
3660 case SIOCSIFNETMASK:
3661 case SIOCGIFDSTADDR:
3662 case SIOCSIFDSTADDR:
3664 return inet_dgram_ops.ioctl(sock, cmd, arg);
3668 return -ENOIOCTLCMD;
3673 static unsigned int packet_poll(struct file *file, struct socket *sock,
3676 struct sock *sk = sock->sk;
3677 struct packet_sock *po = pkt_sk(sk);
3678 unsigned int mask = datagram_poll(file, sock, wait);
3680 spin_lock_bh(&sk->sk_receive_queue.lock);
3681 if (po->rx_ring.pg_vec) {
3682 if (!packet_previous_rx_frame(po, &po->rx_ring,
3684 mask |= POLLIN | POLLRDNORM;
3686 spin_unlock_bh(&sk->sk_receive_queue.lock);
3687 spin_lock_bh(&sk->sk_write_queue.lock);
3688 if (po->tx_ring.pg_vec) {
3689 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3690 mask |= POLLOUT | POLLWRNORM;
3692 spin_unlock_bh(&sk->sk_write_queue.lock);
3697 /* Dirty? Well, I still did not learn better way to account
3701 static void packet_mm_open(struct vm_area_struct *vma)
3703 struct file *file = vma->vm_file;
3704 struct socket *sock = file->private_data;
3705 struct sock *sk = sock->sk;
3708 atomic_inc(&pkt_sk(sk)->mapped);
3711 static void packet_mm_close(struct vm_area_struct *vma)
3713 struct file *file = vma->vm_file;
3714 struct socket *sock = file->private_data;
3715 struct sock *sk = sock->sk;
3718 atomic_dec(&pkt_sk(sk)->mapped);
3721 static const struct vm_operations_struct packet_mmap_ops = {
3722 .open = packet_mm_open,
3723 .close = packet_mm_close,
3726 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3731 for (i = 0; i < len; i++) {
3732 if (likely(pg_vec[i].buffer)) {
3733 if (is_vmalloc_addr(pg_vec[i].buffer))
3734 vfree(pg_vec[i].buffer);
3736 free_pages((unsigned long)pg_vec[i].buffer,
3738 pg_vec[i].buffer = NULL;
3744 static char *alloc_one_pg_vec_page(unsigned long order)
3747 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3748 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3750 buffer = (char *) __get_free_pages(gfp_flags, order);
3754 /* __get_free_pages failed, fall back to vmalloc */
3755 buffer = vzalloc((1 << order) * PAGE_SIZE);
3759 /* vmalloc failed, lets dig into swap here */
3760 gfp_flags &= ~__GFP_NORETRY;
3761 buffer = (char *) __get_free_pages(gfp_flags, order);
3765 /* complete and utter failure */
3769 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3771 unsigned int block_nr = req->tp_block_nr;
3775 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3776 if (unlikely(!pg_vec))
3779 for (i = 0; i < block_nr; i++) {
3780 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3781 if (unlikely(!pg_vec[i].buffer))
3782 goto out_free_pgvec;
3789 free_pg_vec(pg_vec, order, block_nr);
3794 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3795 int closing, int tx_ring)
3797 struct pgv *pg_vec = NULL;
3798 struct packet_sock *po = pkt_sk(sk);
3799 int was_running, order = 0;
3800 struct packet_ring_buffer *rb;
3801 struct sk_buff_head *rb_queue;
3804 /* Added to avoid minimal code churn */
3805 struct tpacket_req *req = &req_u->req;
3807 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3808 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3809 WARN(1, "Tx-ring is not supported.\n");
3813 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3814 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3818 if (atomic_read(&po->mapped))
3820 if (packet_read_pending(rb))
3824 if (req->tp_block_nr) {
3825 /* Sanity tests and some calculations */
3827 if (unlikely(rb->pg_vec))
3830 switch (po->tp_version) {
3832 po->tp_hdrlen = TPACKET_HDRLEN;
3835 po->tp_hdrlen = TPACKET2_HDRLEN;
3838 po->tp_hdrlen = TPACKET3_HDRLEN;
3843 if (unlikely((int)req->tp_block_size <= 0))
3845 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3847 if (po->tp_version >= TPACKET_V3 &&
3848 (int)(req->tp_block_size -
3849 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3851 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3854 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3857 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3858 if (unlikely(rb->frames_per_block <= 0))
3860 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3865 order = get_order(req->tp_block_size);
3866 pg_vec = alloc_pg_vec(req, order);
3867 if (unlikely(!pg_vec))
3869 switch (po->tp_version) {
3871 /* Transmit path is not supported. We checked
3872 * it above but just being paranoid
3875 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3884 if (unlikely(req->tp_frame_nr))
3890 /* Detach socket from network */
3891 spin_lock(&po->bind_lock);
3892 was_running = po->running;
3896 __unregister_prot_hook(sk, false);
3898 spin_unlock(&po->bind_lock);
3903 mutex_lock(&po->pg_vec_lock);
3904 if (closing || atomic_read(&po->mapped) == 0) {
3906 spin_lock_bh(&rb_queue->lock);
3907 swap(rb->pg_vec, pg_vec);
3908 rb->frame_max = (req->tp_frame_nr - 1);
3910 rb->frame_size = req->tp_frame_size;
3911 spin_unlock_bh(&rb_queue->lock);
3913 swap(rb->pg_vec_order, order);
3914 swap(rb->pg_vec_len, req->tp_block_nr);
3916 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3917 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3918 tpacket_rcv : packet_rcv;
3919 skb_queue_purge(rb_queue);
3920 if (atomic_read(&po->mapped))
3921 pr_err("packet_mmap: vma is busy: %d\n",
3922 atomic_read(&po->mapped));
3924 mutex_unlock(&po->pg_vec_lock);
3926 spin_lock(&po->bind_lock);
3929 register_prot_hook(sk);
3931 spin_unlock(&po->bind_lock);
3932 if (closing && (po->tp_version > TPACKET_V2)) {
3933 /* Because we don't support block-based V3 on tx-ring */
3935 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3940 free_pg_vec(pg_vec, order, req->tp_block_nr);
3945 static int packet_mmap(struct file *file, struct socket *sock,
3946 struct vm_area_struct *vma)
3948 struct sock *sk = sock->sk;
3949 struct packet_sock *po = pkt_sk(sk);
3950 unsigned long size, expected_size;
3951 struct packet_ring_buffer *rb;
3952 unsigned long start;
3959 mutex_lock(&po->pg_vec_lock);
3962 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3964 expected_size += rb->pg_vec_len
3970 if (expected_size == 0)
3973 size = vma->vm_end - vma->vm_start;
3974 if (size != expected_size)
3977 start = vma->vm_start;
3978 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3979 if (rb->pg_vec == NULL)
3982 for (i = 0; i < rb->pg_vec_len; i++) {
3984 void *kaddr = rb->pg_vec[i].buffer;
3987 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3988 page = pgv_to_page(kaddr);
3989 err = vm_insert_page(vma, start, page);
3998 atomic_inc(&po->mapped);
3999 vma->vm_ops = &packet_mmap_ops;
4003 mutex_unlock(&po->pg_vec_lock);
4007 static const struct proto_ops packet_ops_spkt = {
4008 .family = PF_PACKET,
4009 .owner = THIS_MODULE,
4010 .release = packet_release,
4011 .bind = packet_bind_spkt,
4012 .connect = sock_no_connect,
4013 .socketpair = sock_no_socketpair,
4014 .accept = sock_no_accept,
4015 .getname = packet_getname_spkt,
4016 .poll = datagram_poll,
4017 .ioctl = packet_ioctl,
4018 .listen = sock_no_listen,
4019 .shutdown = sock_no_shutdown,
4020 .setsockopt = sock_no_setsockopt,
4021 .getsockopt = sock_no_getsockopt,
4022 .sendmsg = packet_sendmsg_spkt,
4023 .recvmsg = packet_recvmsg,
4024 .mmap = sock_no_mmap,
4025 .sendpage = sock_no_sendpage,
4028 static const struct proto_ops packet_ops = {
4029 .family = PF_PACKET,
4030 .owner = THIS_MODULE,
4031 .release = packet_release,
4032 .bind = packet_bind,
4033 .connect = sock_no_connect,
4034 .socketpair = sock_no_socketpair,
4035 .accept = sock_no_accept,
4036 .getname = packet_getname,
4037 .poll = packet_poll,
4038 .ioctl = packet_ioctl,
4039 .listen = sock_no_listen,
4040 .shutdown = sock_no_shutdown,
4041 .setsockopt = packet_setsockopt,
4042 .getsockopt = packet_getsockopt,
4043 .sendmsg = packet_sendmsg,
4044 .recvmsg = packet_recvmsg,
4045 .mmap = packet_mmap,
4046 .sendpage = sock_no_sendpage,
4049 static const struct net_proto_family packet_family_ops = {
4050 .family = PF_PACKET,
4051 .create = packet_create,
4052 .owner = THIS_MODULE,
4055 static struct notifier_block packet_netdev_notifier = {
4056 .notifier_call = packet_notifier,
4059 #ifdef CONFIG_PROC_FS
4061 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4064 struct net *net = seq_file_net(seq);
4067 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4070 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4072 struct net *net = seq_file_net(seq);
4073 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4076 static void packet_seq_stop(struct seq_file *seq, void *v)
4082 static int packet_seq_show(struct seq_file *seq, void *v)
4084 if (v == SEQ_START_TOKEN)
4085 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4087 struct sock *s = sk_entry(v);
4088 const struct packet_sock *po = pkt_sk(s);
4091 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4093 atomic_read(&s->sk_refcnt),
4098 atomic_read(&s->sk_rmem_alloc),
4099 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4106 static const struct seq_operations packet_seq_ops = {
4107 .start = packet_seq_start,
4108 .next = packet_seq_next,
4109 .stop = packet_seq_stop,
4110 .show = packet_seq_show,
4113 static int packet_seq_open(struct inode *inode, struct file *file)
4115 return seq_open_net(inode, file, &packet_seq_ops,
4116 sizeof(struct seq_net_private));
4119 static const struct file_operations packet_seq_fops = {
4120 .owner = THIS_MODULE,
4121 .open = packet_seq_open,
4123 .llseek = seq_lseek,
4124 .release = seq_release_net,
4129 static int __net_init packet_net_init(struct net *net)
4131 mutex_init(&net->packet.sklist_lock);
4132 INIT_HLIST_HEAD(&net->packet.sklist);
4134 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4140 static void __net_exit packet_net_exit(struct net *net)
4142 remove_proc_entry("packet", net->proc_net);
4145 static struct pernet_operations packet_net_ops = {
4146 .init = packet_net_init,
4147 .exit = packet_net_exit,
4151 static void __exit packet_exit(void)
4153 unregister_netdevice_notifier(&packet_netdev_notifier);
4154 unregister_pernet_subsys(&packet_net_ops);
4155 sock_unregister(PF_PACKET);
4156 proto_unregister(&packet_proto);
4159 static int __init packet_init(void)
4161 int rc = proto_register(&packet_proto, 0);
4166 sock_register(&packet_family_ops);
4167 register_pernet_subsys(&packet_net_ops);
4168 register_netdevice_notifier(&packet_netdev_notifier);
4173 module_init(packet_init);
4174 module_exit(packet_exit);
4175 MODULE_LICENSE("GPL");
4176 MODULE_ALIAS_NETPROTO(PF_PACKET);