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 int fanout_rr_next(struct packet_fanout *f, unsigned int num)
1277 int x = atomic_read(&f->rr_cur) + 1;
1285 static unsigned int fanout_demux_hash(struct packet_fanout *f,
1286 struct sk_buff *skb,
1289 return reciprocal_scale(skb_get_hash(skb), num);
1292 static unsigned int fanout_demux_lb(struct packet_fanout *f,
1293 struct sk_buff *skb,
1298 cur = atomic_read(&f->rr_cur);
1299 while ((old = atomic_cmpxchg(&f->rr_cur, cur,
1300 fanout_rr_next(f, num))) != cur)
1305 static unsigned int fanout_demux_cpu(struct packet_fanout *f,
1306 struct sk_buff *skb,
1309 return smp_processor_id() % num;
1312 static unsigned int fanout_demux_rnd(struct packet_fanout *f,
1313 struct sk_buff *skb,
1316 return prandom_u32_max(num);
1319 static unsigned int fanout_demux_rollover(struct packet_fanout *f,
1320 struct sk_buff *skb,
1321 unsigned int idx, unsigned int skip,
1326 i = j = min_t(int, f->next[idx], num - 1);
1328 if (i != skip && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1340 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1341 struct sk_buff *skb,
1344 return skb_get_queue_mapping(skb) % num;
1347 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1349 return f->flags & (flag >> 8);
1352 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1353 struct packet_type *pt, struct net_device *orig_dev)
1355 struct packet_fanout *f = pt->af_packet_priv;
1356 unsigned int num = f->num_members;
1357 struct packet_sock *po;
1360 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1366 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1367 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1372 case PACKET_FANOUT_HASH:
1374 idx = fanout_demux_hash(f, skb, num);
1376 case PACKET_FANOUT_LB:
1377 idx = fanout_demux_lb(f, skb, num);
1379 case PACKET_FANOUT_CPU:
1380 idx = fanout_demux_cpu(f, skb, num);
1382 case PACKET_FANOUT_RND:
1383 idx = fanout_demux_rnd(f, skb, num);
1385 case PACKET_FANOUT_QM:
1386 idx = fanout_demux_qm(f, skb, num);
1388 case PACKET_FANOUT_ROLLOVER:
1389 idx = fanout_demux_rollover(f, skb, 0, (unsigned int) -1, num);
1393 po = pkt_sk(f->arr[idx]);
1394 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER) &&
1395 unlikely(!packet_rcv_has_room(po, skb))) {
1396 idx = fanout_demux_rollover(f, skb, idx, idx, num);
1397 po = pkt_sk(f->arr[idx]);
1400 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1403 DEFINE_MUTEX(fanout_mutex);
1404 EXPORT_SYMBOL_GPL(fanout_mutex);
1405 static LIST_HEAD(fanout_list);
1407 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1409 struct packet_fanout *f = po->fanout;
1411 spin_lock(&f->lock);
1412 f->arr[f->num_members] = sk;
1415 spin_unlock(&f->lock);
1418 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1420 struct packet_fanout *f = po->fanout;
1423 spin_lock(&f->lock);
1424 for (i = 0; i < f->num_members; i++) {
1425 if (f->arr[i] == sk)
1428 BUG_ON(i >= f->num_members);
1429 f->arr[i] = f->arr[f->num_members - 1];
1431 spin_unlock(&f->lock);
1434 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1436 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1442 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1444 struct packet_sock *po = pkt_sk(sk);
1445 struct packet_fanout *f, *match;
1446 u8 type = type_flags & 0xff;
1447 u8 flags = type_flags >> 8;
1451 case PACKET_FANOUT_ROLLOVER:
1452 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1454 case PACKET_FANOUT_HASH:
1455 case PACKET_FANOUT_LB:
1456 case PACKET_FANOUT_CPU:
1457 case PACKET_FANOUT_RND:
1458 case PACKET_FANOUT_QM:
1470 mutex_lock(&fanout_mutex);
1472 list_for_each_entry(f, &fanout_list, list) {
1474 read_pnet(&f->net) == sock_net(sk)) {
1480 if (match && match->flags != flags)
1484 match = kzalloc(sizeof(*match), GFP_KERNEL);
1487 write_pnet(&match->net, sock_net(sk));
1490 match->flags = flags;
1491 atomic_set(&match->rr_cur, 0);
1492 INIT_LIST_HEAD(&match->list);
1493 spin_lock_init(&match->lock);
1494 atomic_set(&match->sk_ref, 0);
1495 match->prot_hook.type = po->prot_hook.type;
1496 match->prot_hook.dev = po->prot_hook.dev;
1497 match->prot_hook.func = packet_rcv_fanout;
1498 match->prot_hook.af_packet_priv = match;
1499 match->prot_hook.id_match = match_fanout_group;
1500 dev_add_pack(&match->prot_hook);
1501 list_add(&match->list, &fanout_list);
1504 if (match->type == type &&
1505 match->prot_hook.type == po->prot_hook.type &&
1506 match->prot_hook.dev == po->prot_hook.dev) {
1508 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1509 __dev_remove_pack(&po->prot_hook);
1511 atomic_inc(&match->sk_ref);
1512 __fanout_link(sk, po);
1517 mutex_unlock(&fanout_mutex);
1521 static void fanout_release(struct sock *sk)
1523 struct packet_sock *po = pkt_sk(sk);
1524 struct packet_fanout *f;
1530 mutex_lock(&fanout_mutex);
1533 if (atomic_dec_and_test(&f->sk_ref)) {
1535 dev_remove_pack(&f->prot_hook);
1538 mutex_unlock(&fanout_mutex);
1541 static const struct proto_ops packet_ops;
1543 static const struct proto_ops packet_ops_spkt;
1545 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1546 struct packet_type *pt, struct net_device *orig_dev)
1549 struct sockaddr_pkt *spkt;
1552 * When we registered the protocol we saved the socket in the data
1553 * field for just this event.
1556 sk = pt->af_packet_priv;
1559 * Yank back the headers [hope the device set this
1560 * right or kerboom...]
1562 * Incoming packets have ll header pulled,
1565 * For outgoing ones skb->data == skb_mac_header(skb)
1566 * so that this procedure is noop.
1569 if (skb->pkt_type == PACKET_LOOPBACK)
1572 if (!net_eq(dev_net(dev), sock_net(sk)))
1575 skb = skb_share_check(skb, GFP_ATOMIC);
1579 /* drop any routing info */
1582 /* drop conntrack reference */
1585 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1587 skb_push(skb, skb->data - skb_mac_header(skb));
1590 * The SOCK_PACKET socket receives _all_ frames.
1593 spkt->spkt_family = dev->type;
1594 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1595 spkt->spkt_protocol = skb->protocol;
1598 * Charge the memory to the socket. This is done specifically
1599 * to prevent sockets using all the memory up.
1602 if (sock_queue_rcv_skb(sk, skb) == 0)
1613 * Output a raw packet to a device layer. This bypasses all the other
1614 * protocol layers and you must therefore supply it with a complete frame
1617 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1620 struct sock *sk = sock->sk;
1621 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1622 struct sk_buff *skb = NULL;
1623 struct net_device *dev;
1629 * Get and verify the address.
1633 if (msg->msg_namelen < sizeof(struct sockaddr))
1635 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1636 proto = saddr->spkt_protocol;
1638 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1641 * Find the device first to size check it
1644 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1647 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1653 if (!(dev->flags & IFF_UP))
1657 * You may not queue a frame bigger than the mtu. This is the lowest level
1658 * raw protocol and you must do your own fragmentation at this level.
1661 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1662 if (!netif_supports_nofcs(dev)) {
1663 err = -EPROTONOSUPPORT;
1666 extra_len = 4; /* We're doing our own CRC */
1670 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1674 size_t reserved = LL_RESERVED_SPACE(dev);
1675 int tlen = dev->needed_tailroom;
1676 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1679 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1682 /* FIXME: Save some space for broken drivers that write a hard
1683 * header at transmission time by themselves. PPP is the notable
1684 * one here. This should really be fixed at the driver level.
1686 skb_reserve(skb, reserved);
1687 skb_reset_network_header(skb);
1689 /* Try to align data part correctly */
1694 skb_reset_network_header(skb);
1696 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1702 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1703 /* Earlier code assumed this would be a VLAN pkt,
1704 * double-check this now that we have the actual
1707 struct ethhdr *ehdr;
1708 skb_reset_mac_header(skb);
1709 ehdr = eth_hdr(skb);
1710 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1716 skb->protocol = proto;
1718 skb->priority = sk->sk_priority;
1719 skb->mark = sk->sk_mark;
1721 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1723 if (unlikely(extra_len == 4))
1726 skb_probe_transport_header(skb, 0);
1728 dev_queue_xmit(skb);
1739 static unsigned int run_filter(const struct sk_buff *skb,
1740 const struct sock *sk,
1743 struct sk_filter *filter;
1746 filter = rcu_dereference(sk->sk_filter);
1748 res = SK_RUN_FILTER(filter, skb);
1755 * This function makes lazy skb cloning in hope that most of packets
1756 * are discarded by BPF.
1758 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1759 * and skb->cb are mangled. It works because (and until) packets
1760 * falling here are owned by current CPU. Output packets are cloned
1761 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1762 * sequencially, so that if we return skb to original state on exit,
1763 * we will not harm anyone.
1766 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1767 struct packet_type *pt, struct net_device *orig_dev)
1770 struct sockaddr_ll *sll;
1771 struct packet_sock *po;
1772 u8 *skb_head = skb->data;
1773 int skb_len = skb->len;
1774 unsigned int snaplen, res;
1776 if (skb->pkt_type == PACKET_LOOPBACK)
1779 sk = pt->af_packet_priv;
1782 if (!net_eq(dev_net(dev), sock_net(sk)))
1787 if (dev->header_ops) {
1788 /* The device has an explicit notion of ll header,
1789 * exported to higher levels.
1791 * Otherwise, the device hides details of its frame
1792 * structure, so that corresponding packet head is
1793 * never delivered to user.
1795 if (sk->sk_type != SOCK_DGRAM)
1796 skb_push(skb, skb->data - skb_mac_header(skb));
1797 else if (skb->pkt_type == PACKET_OUTGOING) {
1798 /* Special case: outgoing packets have ll header at head */
1799 skb_pull(skb, skb_network_offset(skb));
1805 res = run_filter(skb, sk, snaplen);
1807 goto drop_n_restore;
1811 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1814 if (skb_shared(skb)) {
1815 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1819 if (skb_head != skb->data) {
1820 skb->data = skb_head;
1827 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1829 sll = &PACKET_SKB_CB(skb)->sa.ll;
1830 sll->sll_hatype = dev->type;
1831 sll->sll_pkttype = skb->pkt_type;
1832 if (unlikely(po->origdev))
1833 sll->sll_ifindex = orig_dev->ifindex;
1835 sll->sll_ifindex = dev->ifindex;
1837 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1839 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1840 * Use their space for storing the original skb length.
1842 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1844 if (pskb_trim(skb, snaplen))
1847 skb_set_owner_r(skb, sk);
1851 /* drop conntrack reference */
1854 spin_lock(&sk->sk_receive_queue.lock);
1855 po->stats.stats1.tp_packets++;
1856 sock_skb_set_dropcount(sk, skb);
1857 __skb_queue_tail(&sk->sk_receive_queue, skb);
1858 spin_unlock(&sk->sk_receive_queue.lock);
1859 sk->sk_data_ready(sk);
1863 spin_lock(&sk->sk_receive_queue.lock);
1864 po->stats.stats1.tp_drops++;
1865 atomic_inc(&sk->sk_drops);
1866 spin_unlock(&sk->sk_receive_queue.lock);
1869 if (skb_head != skb->data && skb_shared(skb)) {
1870 skb->data = skb_head;
1878 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1879 struct packet_type *pt, struct net_device *orig_dev)
1882 struct packet_sock *po;
1883 struct sockaddr_ll *sll;
1884 union tpacket_uhdr h;
1885 u8 *skb_head = skb->data;
1886 int skb_len = skb->len;
1887 unsigned int snaplen, res;
1888 unsigned long status = TP_STATUS_USER;
1889 unsigned short macoff, netoff, hdrlen;
1890 struct sk_buff *copy_skb = NULL;
1894 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1895 * We may add members to them until current aligned size without forcing
1896 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1898 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1899 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1901 if (skb->pkt_type == PACKET_LOOPBACK)
1904 sk = pt->af_packet_priv;
1907 if (!net_eq(dev_net(dev), sock_net(sk)))
1910 if (dev->header_ops) {
1911 if (sk->sk_type != SOCK_DGRAM)
1912 skb_push(skb, skb->data - skb_mac_header(skb));
1913 else if (skb->pkt_type == PACKET_OUTGOING) {
1914 /* Special case: outgoing packets have ll header at head */
1915 skb_pull(skb, skb_network_offset(skb));
1921 res = run_filter(skb, sk, snaplen);
1923 goto drop_n_restore;
1925 if (skb->ip_summed == CHECKSUM_PARTIAL)
1926 status |= TP_STATUS_CSUMNOTREADY;
1927 else if (skb->pkt_type != PACKET_OUTGOING &&
1928 (skb->ip_summed == CHECKSUM_COMPLETE ||
1929 skb_csum_unnecessary(skb)))
1930 status |= TP_STATUS_CSUM_VALID;
1935 if (sk->sk_type == SOCK_DGRAM) {
1936 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1939 unsigned int maclen = skb_network_offset(skb);
1940 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1941 (maclen < 16 ? 16 : maclen)) +
1943 macoff = netoff - maclen;
1945 if (po->tp_version <= TPACKET_V2) {
1946 if (macoff + snaplen > po->rx_ring.frame_size) {
1947 if (po->copy_thresh &&
1948 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1949 if (skb_shared(skb)) {
1950 copy_skb = skb_clone(skb, GFP_ATOMIC);
1952 copy_skb = skb_get(skb);
1953 skb_head = skb->data;
1956 skb_set_owner_r(copy_skb, sk);
1958 snaplen = po->rx_ring.frame_size - macoff;
1959 if ((int)snaplen < 0)
1962 } else if (unlikely(macoff + snaplen >
1963 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1966 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1967 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1968 snaplen, nval, macoff);
1970 if (unlikely((int)snaplen < 0)) {
1972 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1975 spin_lock(&sk->sk_receive_queue.lock);
1976 h.raw = packet_current_rx_frame(po, skb,
1977 TP_STATUS_KERNEL, (macoff+snaplen));
1980 if (po->tp_version <= TPACKET_V2) {
1981 packet_increment_rx_head(po, &po->rx_ring);
1983 * LOSING will be reported till you read the stats,
1984 * because it's COR - Clear On Read.
1985 * Anyways, moving it for V1/V2 only as V3 doesn't need this
1988 if (po->stats.stats1.tp_drops)
1989 status |= TP_STATUS_LOSING;
1991 po->stats.stats1.tp_packets++;
1993 status |= TP_STATUS_COPY;
1994 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
1996 spin_unlock(&sk->sk_receive_queue.lock);
1998 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2000 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2001 getnstimeofday(&ts);
2003 status |= ts_status;
2005 switch (po->tp_version) {
2007 h.h1->tp_len = skb->len;
2008 h.h1->tp_snaplen = snaplen;
2009 h.h1->tp_mac = macoff;
2010 h.h1->tp_net = netoff;
2011 h.h1->tp_sec = ts.tv_sec;
2012 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2013 hdrlen = sizeof(*h.h1);
2016 h.h2->tp_len = skb->len;
2017 h.h2->tp_snaplen = snaplen;
2018 h.h2->tp_mac = macoff;
2019 h.h2->tp_net = netoff;
2020 h.h2->tp_sec = ts.tv_sec;
2021 h.h2->tp_nsec = ts.tv_nsec;
2022 if (skb_vlan_tag_present(skb)) {
2023 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2024 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2025 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2027 h.h2->tp_vlan_tci = 0;
2028 h.h2->tp_vlan_tpid = 0;
2030 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2031 hdrlen = sizeof(*h.h2);
2034 /* tp_nxt_offset,vlan are already populated above.
2035 * So DONT clear those fields here
2037 h.h3->tp_status |= status;
2038 h.h3->tp_len = skb->len;
2039 h.h3->tp_snaplen = snaplen;
2040 h.h3->tp_mac = macoff;
2041 h.h3->tp_net = netoff;
2042 h.h3->tp_sec = ts.tv_sec;
2043 h.h3->tp_nsec = ts.tv_nsec;
2044 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2045 hdrlen = sizeof(*h.h3);
2051 sll = h.raw + TPACKET_ALIGN(hdrlen);
2052 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2053 sll->sll_family = AF_PACKET;
2054 sll->sll_hatype = dev->type;
2055 sll->sll_protocol = skb->protocol;
2056 sll->sll_pkttype = skb->pkt_type;
2057 if (unlikely(po->origdev))
2058 sll->sll_ifindex = orig_dev->ifindex;
2060 sll->sll_ifindex = dev->ifindex;
2064 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2065 if (po->tp_version <= TPACKET_V2) {
2068 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2071 for (start = h.raw; start < end; start += PAGE_SIZE)
2072 flush_dcache_page(pgv_to_page(start));
2077 if (po->tp_version <= TPACKET_V2) {
2078 __packet_set_status(po, h.raw, status);
2079 sk->sk_data_ready(sk);
2081 prb_clear_blk_fill_status(&po->rx_ring);
2085 if (skb_head != skb->data && skb_shared(skb)) {
2086 skb->data = skb_head;
2094 po->stats.stats1.tp_drops++;
2095 spin_unlock(&sk->sk_receive_queue.lock);
2097 sk->sk_data_ready(sk);
2098 kfree_skb(copy_skb);
2099 goto drop_n_restore;
2102 static void tpacket_destruct_skb(struct sk_buff *skb)
2104 struct packet_sock *po = pkt_sk(skb->sk);
2106 if (likely(po->tx_ring.pg_vec)) {
2110 ph = skb_shinfo(skb)->destructor_arg;
2111 packet_dec_pending(&po->tx_ring);
2113 ts = __packet_set_timestamp(po, ph, skb);
2114 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2120 static bool ll_header_truncated(const struct net_device *dev, int len)
2122 /* net device doesn't like empty head */
2123 if (unlikely(len <= dev->hard_header_len)) {
2124 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2125 current->comm, len, dev->hard_header_len);
2132 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2133 void *frame, struct net_device *dev, int size_max,
2134 __be16 proto, unsigned char *addr, int hlen)
2136 union tpacket_uhdr ph;
2137 int to_write, offset, len, tp_len, nr_frags, len_max;
2138 struct socket *sock = po->sk.sk_socket;
2145 skb->protocol = proto;
2147 skb->priority = po->sk.sk_priority;
2148 skb->mark = po->sk.sk_mark;
2149 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2150 skb_shinfo(skb)->destructor_arg = ph.raw;
2152 switch (po->tp_version) {
2154 tp_len = ph.h2->tp_len;
2157 tp_len = ph.h1->tp_len;
2160 if (unlikely(tp_len > size_max)) {
2161 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2165 skb_reserve(skb, hlen);
2166 skb_reset_network_header(skb);
2168 if (!packet_use_direct_xmit(po))
2169 skb_probe_transport_header(skb, 0);
2170 if (unlikely(po->tp_tx_has_off)) {
2171 int off_min, off_max, off;
2172 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2173 off_max = po->tx_ring.frame_size - tp_len;
2174 if (sock->type == SOCK_DGRAM) {
2175 switch (po->tp_version) {
2177 off = ph.h2->tp_net;
2180 off = ph.h1->tp_net;
2184 switch (po->tp_version) {
2186 off = ph.h2->tp_mac;
2189 off = ph.h1->tp_mac;
2193 if (unlikely((off < off_min) || (off_max < off)))
2195 data = ph.raw + off;
2197 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2201 if (sock->type == SOCK_DGRAM) {
2202 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2204 if (unlikely(err < 0))
2206 } else if (dev->hard_header_len) {
2207 if (ll_header_truncated(dev, tp_len))
2210 skb_push(skb, dev->hard_header_len);
2211 err = skb_store_bits(skb, 0, data,
2212 dev->hard_header_len);
2216 data += dev->hard_header_len;
2217 to_write -= dev->hard_header_len;
2220 offset = offset_in_page(data);
2221 len_max = PAGE_SIZE - offset;
2222 len = ((to_write > len_max) ? len_max : to_write);
2224 skb->data_len = to_write;
2225 skb->len += to_write;
2226 skb->truesize += to_write;
2227 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2229 while (likely(to_write)) {
2230 nr_frags = skb_shinfo(skb)->nr_frags;
2232 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2233 pr_err("Packet exceed the number of skb frags(%lu)\n",
2238 page = pgv_to_page(data);
2240 flush_dcache_page(page);
2242 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2245 len_max = PAGE_SIZE;
2246 len = ((to_write > len_max) ? len_max : to_write);
2252 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2254 struct sk_buff *skb;
2255 struct net_device *dev;
2257 int err, reserve = 0;
2259 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2260 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2261 int tp_len, size_max;
2262 unsigned char *addr;
2264 int status = TP_STATUS_AVAILABLE;
2267 mutex_lock(&po->pg_vec_lock);
2269 if (likely(saddr == NULL)) {
2270 dev = packet_cached_dev_get(po);
2275 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2277 if (msg->msg_namelen < (saddr->sll_halen
2278 + offsetof(struct sockaddr_ll,
2281 proto = saddr->sll_protocol;
2282 addr = saddr->sll_addr;
2283 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2287 if (unlikely(dev == NULL))
2290 if (unlikely(!(dev->flags & IFF_UP)))
2293 reserve = dev->hard_header_len + VLAN_HLEN;
2294 size_max = po->tx_ring.frame_size
2295 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2297 if (size_max > dev->mtu + reserve)
2298 size_max = dev->mtu + reserve;
2301 ph = packet_current_frame(po, &po->tx_ring,
2302 TP_STATUS_SEND_REQUEST);
2303 if (unlikely(ph == NULL)) {
2304 if (need_wait && need_resched())
2309 status = TP_STATUS_SEND_REQUEST;
2310 hlen = LL_RESERVED_SPACE(dev);
2311 tlen = dev->needed_tailroom;
2312 skb = sock_alloc_send_skb(&po->sk,
2313 hlen + tlen + sizeof(struct sockaddr_ll),
2316 if (unlikely(skb == NULL))
2319 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2321 if (tp_len > dev->mtu + dev->hard_header_len) {
2322 struct ethhdr *ehdr;
2323 /* Earlier code assumed this would be a VLAN pkt,
2324 * double-check this now that we have the actual
2328 skb_reset_mac_header(skb);
2329 ehdr = eth_hdr(skb);
2330 if (ehdr->h_proto != htons(ETH_P_8021Q))
2333 if (unlikely(tp_len < 0)) {
2335 __packet_set_status(po, ph,
2336 TP_STATUS_AVAILABLE);
2337 packet_increment_head(&po->tx_ring);
2341 status = TP_STATUS_WRONG_FORMAT;
2347 packet_pick_tx_queue(dev, skb);
2349 skb->destructor = tpacket_destruct_skb;
2350 __packet_set_status(po, ph, TP_STATUS_SENDING);
2351 packet_inc_pending(&po->tx_ring);
2353 status = TP_STATUS_SEND_REQUEST;
2354 err = po->xmit(skb);
2355 if (unlikely(err > 0)) {
2356 err = net_xmit_errno(err);
2357 if (err && __packet_get_status(po, ph) ==
2358 TP_STATUS_AVAILABLE) {
2359 /* skb was destructed already */
2364 * skb was dropped but not destructed yet;
2365 * let's treat it like congestion or err < 0
2369 packet_increment_head(&po->tx_ring);
2371 } while (likely((ph != NULL) ||
2372 /* Note: packet_read_pending() might be slow if we have
2373 * to call it as it's per_cpu variable, but in fast-path
2374 * we already short-circuit the loop with the first
2375 * condition, and luckily don't have to go that path
2378 (need_wait && packet_read_pending(&po->tx_ring))));
2384 __packet_set_status(po, ph, status);
2389 mutex_unlock(&po->pg_vec_lock);
2393 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2394 size_t reserve, size_t len,
2395 size_t linear, int noblock,
2398 struct sk_buff *skb;
2400 /* Under a page? Don't bother with paged skb. */
2401 if (prepad + len < PAGE_SIZE || !linear)
2404 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2409 skb_reserve(skb, reserve);
2410 skb_put(skb, linear);
2411 skb->data_len = len - linear;
2412 skb->len += len - linear;
2417 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2419 struct sock *sk = sock->sk;
2420 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2421 struct sk_buff *skb;
2422 struct net_device *dev;
2424 unsigned char *addr;
2425 int err, reserve = 0;
2426 struct virtio_net_hdr vnet_hdr = { 0 };
2429 struct packet_sock *po = pkt_sk(sk);
2430 unsigned short gso_type = 0;
2436 * Get and verify the address.
2439 if (likely(saddr == NULL)) {
2440 dev = packet_cached_dev_get(po);
2445 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2447 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2449 proto = saddr->sll_protocol;
2450 addr = saddr->sll_addr;
2451 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2455 if (unlikely(dev == NULL))
2458 if (unlikely(!(dev->flags & IFF_UP)))
2461 if (sock->type == SOCK_RAW)
2462 reserve = dev->hard_header_len;
2463 if (po->has_vnet_hdr) {
2464 vnet_hdr_len = sizeof(vnet_hdr);
2467 if (len < vnet_hdr_len)
2470 len -= vnet_hdr_len;
2473 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2474 if (n != vnet_hdr_len)
2477 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2478 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2479 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2480 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2481 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2482 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2483 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2486 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2489 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2490 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2491 case VIRTIO_NET_HDR_GSO_TCPV4:
2492 gso_type = SKB_GSO_TCPV4;
2494 case VIRTIO_NET_HDR_GSO_TCPV6:
2495 gso_type = SKB_GSO_TCPV6;
2497 case VIRTIO_NET_HDR_GSO_UDP:
2498 gso_type = SKB_GSO_UDP;
2504 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2505 gso_type |= SKB_GSO_TCP_ECN;
2507 if (vnet_hdr.gso_size == 0)
2513 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2514 if (!netif_supports_nofcs(dev)) {
2515 err = -EPROTONOSUPPORT;
2518 extra_len = 4; /* We're doing our own CRC */
2522 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2526 hlen = LL_RESERVED_SPACE(dev);
2527 tlen = dev->needed_tailroom;
2528 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2529 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2530 msg->msg_flags & MSG_DONTWAIT, &err);
2534 skb_set_network_header(skb, reserve);
2537 if (sock->type == SOCK_DGRAM) {
2538 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2539 if (unlikely(offset < 0))
2542 if (ll_header_truncated(dev, len))
2546 /* Returns -EFAULT on error */
2547 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2551 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2553 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2554 /* Earlier code assumed this would be a VLAN pkt,
2555 * double-check this now that we have the actual
2558 struct ethhdr *ehdr;
2559 skb_reset_mac_header(skb);
2560 ehdr = eth_hdr(skb);
2561 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2567 skb->protocol = proto;
2569 skb->priority = sk->sk_priority;
2570 skb->mark = sk->sk_mark;
2572 packet_pick_tx_queue(dev, skb);
2574 if (po->has_vnet_hdr) {
2575 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2576 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2577 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2578 if (!skb_partial_csum_set(skb, s, o)) {
2584 skb_shinfo(skb)->gso_size =
2585 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2586 skb_shinfo(skb)->gso_type = gso_type;
2588 /* Header must be checked, and gso_segs computed. */
2589 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2590 skb_shinfo(skb)->gso_segs = 0;
2592 len += vnet_hdr_len;
2595 if (!packet_use_direct_xmit(po))
2596 skb_probe_transport_header(skb, reserve);
2597 if (unlikely(extra_len == 4))
2600 err = po->xmit(skb);
2601 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2617 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2619 struct sock *sk = sock->sk;
2620 struct packet_sock *po = pkt_sk(sk);
2622 if (po->tx_ring.pg_vec)
2623 return tpacket_snd(po, msg);
2625 return packet_snd(sock, msg, len);
2629 * Close a PACKET socket. This is fairly simple. We immediately go
2630 * to 'closed' state and remove our protocol entry in the device list.
2633 static int packet_release(struct socket *sock)
2635 struct sock *sk = sock->sk;
2636 struct packet_sock *po;
2638 union tpacket_req_u req_u;
2646 mutex_lock(&net->packet.sklist_lock);
2647 sk_del_node_init_rcu(sk);
2648 mutex_unlock(&net->packet.sklist_lock);
2651 sock_prot_inuse_add(net, sk->sk_prot, -1);
2654 spin_lock(&po->bind_lock);
2655 unregister_prot_hook(sk, false);
2656 packet_cached_dev_reset(po);
2658 if (po->prot_hook.dev) {
2659 dev_put(po->prot_hook.dev);
2660 po->prot_hook.dev = NULL;
2662 spin_unlock(&po->bind_lock);
2664 packet_flush_mclist(sk);
2666 if (po->rx_ring.pg_vec) {
2667 memset(&req_u, 0, sizeof(req_u));
2668 packet_set_ring(sk, &req_u, 1, 0);
2671 if (po->tx_ring.pg_vec) {
2672 memset(&req_u, 0, sizeof(req_u));
2673 packet_set_ring(sk, &req_u, 1, 1);
2680 * Now the socket is dead. No more input will appear.
2687 skb_queue_purge(&sk->sk_receive_queue);
2688 packet_free_pending(po);
2689 sk_refcnt_debug_release(sk);
2696 * Attach a packet hook.
2699 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2701 struct packet_sock *po = pkt_sk(sk);
2702 const struct net_device *dev_curr;
2714 spin_lock(&po->bind_lock);
2716 proto_curr = po->prot_hook.type;
2717 dev_curr = po->prot_hook.dev;
2719 need_rehook = proto_curr != proto || dev_curr != dev;
2722 unregister_prot_hook(sk, true);
2725 po->prot_hook.type = proto;
2727 if (po->prot_hook.dev)
2728 dev_put(po->prot_hook.dev);
2730 po->prot_hook.dev = dev;
2732 po->ifindex = dev ? dev->ifindex : 0;
2733 packet_cached_dev_assign(po, dev);
2736 if (proto == 0 || !need_rehook)
2739 if (!dev || (dev->flags & IFF_UP)) {
2740 register_prot_hook(sk);
2742 sk->sk_err = ENETDOWN;
2743 if (!sock_flag(sk, SOCK_DEAD))
2744 sk->sk_error_report(sk);
2748 spin_unlock(&po->bind_lock);
2754 * Bind a packet socket to a device
2757 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2760 struct sock *sk = sock->sk;
2762 struct net_device *dev;
2769 if (addr_len != sizeof(struct sockaddr))
2771 strlcpy(name, uaddr->sa_data, sizeof(name));
2773 dev = dev_get_by_name(sock_net(sk), name);
2775 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2779 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2781 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2782 struct sock *sk = sock->sk;
2783 struct net_device *dev = NULL;
2791 if (addr_len < sizeof(struct sockaddr_ll))
2793 if (sll->sll_family != AF_PACKET)
2796 if (sll->sll_ifindex) {
2798 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2802 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2808 static struct proto packet_proto = {
2810 .owner = THIS_MODULE,
2811 .obj_size = sizeof(struct packet_sock),
2815 * Create a packet of type SOCK_PACKET.
2818 static int packet_create(struct net *net, struct socket *sock, int protocol,
2822 struct packet_sock *po;
2823 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2826 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2828 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2829 sock->type != SOCK_PACKET)
2830 return -ESOCKTNOSUPPORT;
2832 sock->state = SS_UNCONNECTED;
2835 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto);
2839 sock->ops = &packet_ops;
2840 if (sock->type == SOCK_PACKET)
2841 sock->ops = &packet_ops_spkt;
2843 sock_init_data(sock, sk);
2846 sk->sk_family = PF_PACKET;
2848 po->xmit = dev_queue_xmit;
2850 err = packet_alloc_pending(po);
2854 packet_cached_dev_reset(po);
2856 sk->sk_destruct = packet_sock_destruct;
2857 sk_refcnt_debug_inc(sk);
2860 * Attach a protocol block
2863 spin_lock_init(&po->bind_lock);
2864 mutex_init(&po->pg_vec_lock);
2865 po->prot_hook.func = packet_rcv;
2867 if (sock->type == SOCK_PACKET)
2868 po->prot_hook.func = packet_rcv_spkt;
2870 po->prot_hook.af_packet_priv = sk;
2873 po->prot_hook.type = proto;
2874 register_prot_hook(sk);
2877 mutex_lock(&net->packet.sklist_lock);
2878 sk_add_node_rcu(sk, &net->packet.sklist);
2879 mutex_unlock(&net->packet.sklist_lock);
2882 sock_prot_inuse_add(net, &packet_proto, 1);
2893 * Pull a packet from our receive queue and hand it to the user.
2894 * If necessary we block.
2897 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2900 struct sock *sk = sock->sk;
2901 struct sk_buff *skb;
2903 int vnet_hdr_len = 0;
2904 unsigned int origlen = 0;
2907 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2911 /* What error should we return now? EUNATTACH? */
2912 if (pkt_sk(sk)->ifindex < 0)
2916 if (flags & MSG_ERRQUEUE) {
2917 err = sock_recv_errqueue(sk, msg, len,
2918 SOL_PACKET, PACKET_TX_TIMESTAMP);
2923 * Call the generic datagram receiver. This handles all sorts
2924 * of horrible races and re-entrancy so we can forget about it
2925 * in the protocol layers.
2927 * Now it will return ENETDOWN, if device have just gone down,
2928 * but then it will block.
2931 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2934 * An error occurred so return it. Because skb_recv_datagram()
2935 * handles the blocking we don't see and worry about blocking
2942 if (pkt_sk(sk)->has_vnet_hdr) {
2943 struct virtio_net_hdr vnet_hdr = { 0 };
2946 vnet_hdr_len = sizeof(vnet_hdr);
2947 if (len < vnet_hdr_len)
2950 len -= vnet_hdr_len;
2952 if (skb_is_gso(skb)) {
2953 struct skb_shared_info *sinfo = skb_shinfo(skb);
2955 /* This is a hint as to how much should be linear. */
2957 __cpu_to_virtio16(false, skb_headlen(skb));
2959 __cpu_to_virtio16(false, sinfo->gso_size);
2960 if (sinfo->gso_type & SKB_GSO_TCPV4)
2961 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2962 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2963 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2964 else if (sinfo->gso_type & SKB_GSO_UDP)
2965 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2966 else if (sinfo->gso_type & SKB_GSO_FCOE)
2970 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2971 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2973 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2975 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2976 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2977 vnet_hdr.csum_start = __cpu_to_virtio16(false,
2978 skb_checksum_start_offset(skb));
2979 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
2981 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2982 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
2983 } /* else everything is zero */
2985 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
2990 /* You lose any data beyond the buffer you gave. If it worries
2991 * a user program they can ask the device for its MTU
2997 msg->msg_flags |= MSG_TRUNC;
3000 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3004 if (sock->type != SOCK_PACKET) {
3005 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3007 /* Original length was stored in sockaddr_ll fields */
3008 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3009 sll->sll_family = AF_PACKET;
3010 sll->sll_protocol = skb->protocol;
3013 sock_recv_ts_and_drops(msg, sk, skb);
3015 if (msg->msg_name) {
3016 /* If the address length field is there to be filled
3017 * in, we fill it in now.
3019 if (sock->type == SOCK_PACKET) {
3020 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3021 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3023 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3025 msg->msg_namelen = sll->sll_halen +
3026 offsetof(struct sockaddr_ll, sll_addr);
3028 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3032 if (pkt_sk(sk)->auxdata) {
3033 struct tpacket_auxdata aux;
3035 aux.tp_status = TP_STATUS_USER;
3036 if (skb->ip_summed == CHECKSUM_PARTIAL)
3037 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3038 else if (skb->pkt_type != PACKET_OUTGOING &&
3039 (skb->ip_summed == CHECKSUM_COMPLETE ||
3040 skb_csum_unnecessary(skb)))
3041 aux.tp_status |= TP_STATUS_CSUM_VALID;
3043 aux.tp_len = origlen;
3044 aux.tp_snaplen = skb->len;
3046 aux.tp_net = skb_network_offset(skb);
3047 if (skb_vlan_tag_present(skb)) {
3048 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3049 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3050 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3052 aux.tp_vlan_tci = 0;
3053 aux.tp_vlan_tpid = 0;
3055 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3059 * Free or return the buffer as appropriate. Again this
3060 * hides all the races and re-entrancy issues from us.
3062 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3065 skb_free_datagram(sk, skb);
3070 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3071 int *uaddr_len, int peer)
3073 struct net_device *dev;
3074 struct sock *sk = sock->sk;
3079 uaddr->sa_family = AF_PACKET;
3080 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3082 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3084 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3086 *uaddr_len = sizeof(*uaddr);
3091 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3092 int *uaddr_len, int peer)
3094 struct net_device *dev;
3095 struct sock *sk = sock->sk;
3096 struct packet_sock *po = pkt_sk(sk);
3097 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3102 sll->sll_family = AF_PACKET;
3103 sll->sll_ifindex = po->ifindex;
3104 sll->sll_protocol = po->num;
3105 sll->sll_pkttype = 0;
3107 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3109 sll->sll_hatype = dev->type;
3110 sll->sll_halen = dev->addr_len;
3111 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3113 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3117 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3122 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3126 case PACKET_MR_MULTICAST:
3127 if (i->alen != dev->addr_len)
3130 return dev_mc_add(dev, i->addr);
3132 return dev_mc_del(dev, i->addr);
3134 case PACKET_MR_PROMISC:
3135 return dev_set_promiscuity(dev, what);
3136 case PACKET_MR_ALLMULTI:
3137 return dev_set_allmulti(dev, what);
3138 case PACKET_MR_UNICAST:
3139 if (i->alen != dev->addr_len)
3142 return dev_uc_add(dev, i->addr);
3144 return dev_uc_del(dev, i->addr);
3152 static void packet_dev_mclist_delete(struct net_device *dev,
3153 struct packet_mclist **mlp)
3155 struct packet_mclist *ml;
3157 while ((ml = *mlp) != NULL) {
3158 if (ml->ifindex == dev->ifindex) {
3159 packet_dev_mc(dev, ml, -1);
3167 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3169 struct packet_sock *po = pkt_sk(sk);
3170 struct packet_mclist *ml, *i;
3171 struct net_device *dev;
3177 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3182 if (mreq->mr_alen > dev->addr_len)
3186 i = kmalloc(sizeof(*i), GFP_KERNEL);
3191 for (ml = po->mclist; ml; ml = ml->next) {
3192 if (ml->ifindex == mreq->mr_ifindex &&
3193 ml->type == mreq->mr_type &&
3194 ml->alen == mreq->mr_alen &&
3195 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3197 /* Free the new element ... */
3203 i->type = mreq->mr_type;
3204 i->ifindex = mreq->mr_ifindex;
3205 i->alen = mreq->mr_alen;
3206 memcpy(i->addr, mreq->mr_address, i->alen);
3208 i->next = po->mclist;
3210 err = packet_dev_mc(dev, i, 1);
3212 po->mclist = i->next;
3221 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3223 struct packet_mclist *ml, **mlp;
3227 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3228 if (ml->ifindex == mreq->mr_ifindex &&
3229 ml->type == mreq->mr_type &&
3230 ml->alen == mreq->mr_alen &&
3231 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3232 if (--ml->count == 0) {
3233 struct net_device *dev;
3235 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3237 packet_dev_mc(dev, ml, -1);
3247 static void packet_flush_mclist(struct sock *sk)
3249 struct packet_sock *po = pkt_sk(sk);
3250 struct packet_mclist *ml;
3256 while ((ml = po->mclist) != NULL) {
3257 struct net_device *dev;
3259 po->mclist = ml->next;
3260 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3262 packet_dev_mc(dev, ml, -1);
3269 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3271 struct sock *sk = sock->sk;
3272 struct packet_sock *po = pkt_sk(sk);
3275 if (level != SOL_PACKET)
3276 return -ENOPROTOOPT;
3279 case PACKET_ADD_MEMBERSHIP:
3280 case PACKET_DROP_MEMBERSHIP:
3282 struct packet_mreq_max mreq;
3284 memset(&mreq, 0, sizeof(mreq));
3285 if (len < sizeof(struct packet_mreq))
3287 if (len > sizeof(mreq))
3289 if (copy_from_user(&mreq, optval, len))
3291 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3293 if (optname == PACKET_ADD_MEMBERSHIP)
3294 ret = packet_mc_add(sk, &mreq);
3296 ret = packet_mc_drop(sk, &mreq);
3300 case PACKET_RX_RING:
3301 case PACKET_TX_RING:
3303 union tpacket_req_u req_u;
3306 switch (po->tp_version) {
3309 len = sizeof(req_u.req);
3313 len = sizeof(req_u.req3);
3318 if (pkt_sk(sk)->has_vnet_hdr)
3320 if (copy_from_user(&req_u.req, optval, len))
3322 return packet_set_ring(sk, &req_u, 0,
3323 optname == PACKET_TX_RING);
3325 case PACKET_COPY_THRESH:
3329 if (optlen != sizeof(val))
3331 if (copy_from_user(&val, optval, sizeof(val)))
3334 pkt_sk(sk)->copy_thresh = val;
3337 case PACKET_VERSION:
3341 if (optlen != sizeof(val))
3343 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3345 if (copy_from_user(&val, optval, sizeof(val)))
3351 po->tp_version = val;
3357 case PACKET_RESERVE:
3361 if (optlen != sizeof(val))
3363 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3365 if (copy_from_user(&val, optval, sizeof(val)))
3367 po->tp_reserve = val;
3374 if (optlen != sizeof(val))
3376 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3378 if (copy_from_user(&val, optval, sizeof(val)))
3380 po->tp_loss = !!val;
3383 case PACKET_AUXDATA:
3387 if (optlen < sizeof(val))
3389 if (copy_from_user(&val, optval, sizeof(val)))
3392 po->auxdata = !!val;
3395 case PACKET_ORIGDEV:
3399 if (optlen < sizeof(val))
3401 if (copy_from_user(&val, optval, sizeof(val)))
3404 po->origdev = !!val;
3407 case PACKET_VNET_HDR:
3411 if (sock->type != SOCK_RAW)
3413 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3415 if (optlen < sizeof(val))
3417 if (copy_from_user(&val, optval, sizeof(val)))
3420 po->has_vnet_hdr = !!val;
3423 case PACKET_TIMESTAMP:
3427 if (optlen != sizeof(val))
3429 if (copy_from_user(&val, optval, sizeof(val)))
3432 po->tp_tstamp = val;
3439 if (optlen != sizeof(val))
3441 if (copy_from_user(&val, optval, sizeof(val)))
3444 return fanout_add(sk, val & 0xffff, val >> 16);
3446 case PACKET_TX_HAS_OFF:
3450 if (optlen != sizeof(val))
3452 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3454 if (copy_from_user(&val, optval, sizeof(val)))
3456 po->tp_tx_has_off = !!val;
3459 case PACKET_QDISC_BYPASS:
3463 if (optlen != sizeof(val))
3465 if (copy_from_user(&val, optval, sizeof(val)))
3468 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3472 return -ENOPROTOOPT;
3476 static int packet_getsockopt(struct socket *sock, int level, int optname,
3477 char __user *optval, int __user *optlen)
3480 int val, lv = sizeof(val);
3481 struct sock *sk = sock->sk;
3482 struct packet_sock *po = pkt_sk(sk);
3484 union tpacket_stats_u st;
3486 if (level != SOL_PACKET)
3487 return -ENOPROTOOPT;
3489 if (get_user(len, optlen))
3496 case PACKET_STATISTICS:
3497 spin_lock_bh(&sk->sk_receive_queue.lock);
3498 memcpy(&st, &po->stats, sizeof(st));
3499 memset(&po->stats, 0, sizeof(po->stats));
3500 spin_unlock_bh(&sk->sk_receive_queue.lock);
3502 if (po->tp_version == TPACKET_V3) {
3503 lv = sizeof(struct tpacket_stats_v3);
3504 st.stats3.tp_packets += st.stats3.tp_drops;
3507 lv = sizeof(struct tpacket_stats);
3508 st.stats1.tp_packets += st.stats1.tp_drops;
3513 case PACKET_AUXDATA:
3516 case PACKET_ORIGDEV:
3519 case PACKET_VNET_HDR:
3520 val = po->has_vnet_hdr;
3522 case PACKET_VERSION:
3523 val = po->tp_version;
3526 if (len > sizeof(int))
3528 if (copy_from_user(&val, optval, len))
3532 val = sizeof(struct tpacket_hdr);
3535 val = sizeof(struct tpacket2_hdr);
3538 val = sizeof(struct tpacket3_hdr);
3544 case PACKET_RESERVE:
3545 val = po->tp_reserve;
3550 case PACKET_TIMESTAMP:
3551 val = po->tp_tstamp;
3555 ((u32)po->fanout->id |
3556 ((u32)po->fanout->type << 16) |
3557 ((u32)po->fanout->flags << 24)) :
3560 case PACKET_TX_HAS_OFF:
3561 val = po->tp_tx_has_off;
3563 case PACKET_QDISC_BYPASS:
3564 val = packet_use_direct_xmit(po);
3567 return -ENOPROTOOPT;
3572 if (put_user(len, optlen))
3574 if (copy_to_user(optval, data, len))
3580 static int packet_notifier(struct notifier_block *this,
3581 unsigned long msg, void *ptr)
3584 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3585 struct net *net = dev_net(dev);
3588 sk_for_each_rcu(sk, &net->packet.sklist) {
3589 struct packet_sock *po = pkt_sk(sk);
3592 case NETDEV_UNREGISTER:
3594 packet_dev_mclist_delete(dev, &po->mclist);
3598 if (dev->ifindex == po->ifindex) {
3599 spin_lock(&po->bind_lock);
3601 __unregister_prot_hook(sk, false);
3602 sk->sk_err = ENETDOWN;
3603 if (!sock_flag(sk, SOCK_DEAD))
3604 sk->sk_error_report(sk);
3606 if (msg == NETDEV_UNREGISTER) {
3607 packet_cached_dev_reset(po);
3609 if (po->prot_hook.dev)
3610 dev_put(po->prot_hook.dev);
3611 po->prot_hook.dev = NULL;
3613 spin_unlock(&po->bind_lock);
3617 if (dev->ifindex == po->ifindex) {
3618 spin_lock(&po->bind_lock);
3620 register_prot_hook(sk);
3621 spin_unlock(&po->bind_lock);
3631 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3634 struct sock *sk = sock->sk;
3639 int amount = sk_wmem_alloc_get(sk);
3641 return put_user(amount, (int __user *)arg);
3645 struct sk_buff *skb;
3648 spin_lock_bh(&sk->sk_receive_queue.lock);
3649 skb = skb_peek(&sk->sk_receive_queue);
3652 spin_unlock_bh(&sk->sk_receive_queue.lock);
3653 return put_user(amount, (int __user *)arg);
3656 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3658 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3668 case SIOCGIFBRDADDR:
3669 case SIOCSIFBRDADDR:
3670 case SIOCGIFNETMASK:
3671 case SIOCSIFNETMASK:
3672 case SIOCGIFDSTADDR:
3673 case SIOCSIFDSTADDR:
3675 return inet_dgram_ops.ioctl(sock, cmd, arg);
3679 return -ENOIOCTLCMD;
3684 static unsigned int packet_poll(struct file *file, struct socket *sock,
3687 struct sock *sk = sock->sk;
3688 struct packet_sock *po = pkt_sk(sk);
3689 unsigned int mask = datagram_poll(file, sock, wait);
3691 spin_lock_bh(&sk->sk_receive_queue.lock);
3692 if (po->rx_ring.pg_vec) {
3693 if (!packet_previous_rx_frame(po, &po->rx_ring,
3695 mask |= POLLIN | POLLRDNORM;
3697 spin_unlock_bh(&sk->sk_receive_queue.lock);
3698 spin_lock_bh(&sk->sk_write_queue.lock);
3699 if (po->tx_ring.pg_vec) {
3700 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3701 mask |= POLLOUT | POLLWRNORM;
3703 spin_unlock_bh(&sk->sk_write_queue.lock);
3708 /* Dirty? Well, I still did not learn better way to account
3712 static void packet_mm_open(struct vm_area_struct *vma)
3714 struct file *file = vma->vm_file;
3715 struct socket *sock = file->private_data;
3716 struct sock *sk = sock->sk;
3719 atomic_inc(&pkt_sk(sk)->mapped);
3722 static void packet_mm_close(struct vm_area_struct *vma)
3724 struct file *file = vma->vm_file;
3725 struct socket *sock = file->private_data;
3726 struct sock *sk = sock->sk;
3729 atomic_dec(&pkt_sk(sk)->mapped);
3732 static const struct vm_operations_struct packet_mmap_ops = {
3733 .open = packet_mm_open,
3734 .close = packet_mm_close,
3737 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3742 for (i = 0; i < len; i++) {
3743 if (likely(pg_vec[i].buffer)) {
3744 if (is_vmalloc_addr(pg_vec[i].buffer))
3745 vfree(pg_vec[i].buffer);
3747 free_pages((unsigned long)pg_vec[i].buffer,
3749 pg_vec[i].buffer = NULL;
3755 static char *alloc_one_pg_vec_page(unsigned long order)
3758 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3759 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3761 buffer = (char *) __get_free_pages(gfp_flags, order);
3765 /* __get_free_pages failed, fall back to vmalloc */
3766 buffer = vzalloc((1 << order) * PAGE_SIZE);
3770 /* vmalloc failed, lets dig into swap here */
3771 gfp_flags &= ~__GFP_NORETRY;
3772 buffer = (char *) __get_free_pages(gfp_flags, order);
3776 /* complete and utter failure */
3780 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3782 unsigned int block_nr = req->tp_block_nr;
3786 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3787 if (unlikely(!pg_vec))
3790 for (i = 0; i < block_nr; i++) {
3791 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3792 if (unlikely(!pg_vec[i].buffer))
3793 goto out_free_pgvec;
3800 free_pg_vec(pg_vec, order, block_nr);
3805 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3806 int closing, int tx_ring)
3808 struct pgv *pg_vec = NULL;
3809 struct packet_sock *po = pkt_sk(sk);
3810 int was_running, order = 0;
3811 struct packet_ring_buffer *rb;
3812 struct sk_buff_head *rb_queue;
3815 /* Added to avoid minimal code churn */
3816 struct tpacket_req *req = &req_u->req;
3818 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3819 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3820 WARN(1, "Tx-ring is not supported.\n");
3824 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3825 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3829 if (atomic_read(&po->mapped))
3831 if (packet_read_pending(rb))
3835 if (req->tp_block_nr) {
3836 /* Sanity tests and some calculations */
3838 if (unlikely(rb->pg_vec))
3841 switch (po->tp_version) {
3843 po->tp_hdrlen = TPACKET_HDRLEN;
3846 po->tp_hdrlen = TPACKET2_HDRLEN;
3849 po->tp_hdrlen = TPACKET3_HDRLEN;
3854 if (unlikely((int)req->tp_block_size <= 0))
3856 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3858 if (po->tp_version >= TPACKET_V3 &&
3859 (int)(req->tp_block_size -
3860 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3862 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3865 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3868 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3869 if (unlikely(rb->frames_per_block <= 0))
3871 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3876 order = get_order(req->tp_block_size);
3877 pg_vec = alloc_pg_vec(req, order);
3878 if (unlikely(!pg_vec))
3880 switch (po->tp_version) {
3882 /* Transmit path is not supported. We checked
3883 * it above but just being paranoid
3886 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3895 if (unlikely(req->tp_frame_nr))
3901 /* Detach socket from network */
3902 spin_lock(&po->bind_lock);
3903 was_running = po->running;
3907 __unregister_prot_hook(sk, false);
3909 spin_unlock(&po->bind_lock);
3914 mutex_lock(&po->pg_vec_lock);
3915 if (closing || atomic_read(&po->mapped) == 0) {
3917 spin_lock_bh(&rb_queue->lock);
3918 swap(rb->pg_vec, pg_vec);
3919 rb->frame_max = (req->tp_frame_nr - 1);
3921 rb->frame_size = req->tp_frame_size;
3922 spin_unlock_bh(&rb_queue->lock);
3924 swap(rb->pg_vec_order, order);
3925 swap(rb->pg_vec_len, req->tp_block_nr);
3927 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3928 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3929 tpacket_rcv : packet_rcv;
3930 skb_queue_purge(rb_queue);
3931 if (atomic_read(&po->mapped))
3932 pr_err("packet_mmap: vma is busy: %d\n",
3933 atomic_read(&po->mapped));
3935 mutex_unlock(&po->pg_vec_lock);
3937 spin_lock(&po->bind_lock);
3940 register_prot_hook(sk);
3942 spin_unlock(&po->bind_lock);
3943 if (closing && (po->tp_version > TPACKET_V2)) {
3944 /* Because we don't support block-based V3 on tx-ring */
3946 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3951 free_pg_vec(pg_vec, order, req->tp_block_nr);
3956 static int packet_mmap(struct file *file, struct socket *sock,
3957 struct vm_area_struct *vma)
3959 struct sock *sk = sock->sk;
3960 struct packet_sock *po = pkt_sk(sk);
3961 unsigned long size, expected_size;
3962 struct packet_ring_buffer *rb;
3963 unsigned long start;
3970 mutex_lock(&po->pg_vec_lock);
3973 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3975 expected_size += rb->pg_vec_len
3981 if (expected_size == 0)
3984 size = vma->vm_end - vma->vm_start;
3985 if (size != expected_size)
3988 start = vma->vm_start;
3989 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3990 if (rb->pg_vec == NULL)
3993 for (i = 0; i < rb->pg_vec_len; i++) {
3995 void *kaddr = rb->pg_vec[i].buffer;
3998 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
3999 page = pgv_to_page(kaddr);
4000 err = vm_insert_page(vma, start, page);
4009 atomic_inc(&po->mapped);
4010 vma->vm_ops = &packet_mmap_ops;
4014 mutex_unlock(&po->pg_vec_lock);
4018 static const struct proto_ops packet_ops_spkt = {
4019 .family = PF_PACKET,
4020 .owner = THIS_MODULE,
4021 .release = packet_release,
4022 .bind = packet_bind_spkt,
4023 .connect = sock_no_connect,
4024 .socketpair = sock_no_socketpair,
4025 .accept = sock_no_accept,
4026 .getname = packet_getname_spkt,
4027 .poll = datagram_poll,
4028 .ioctl = packet_ioctl,
4029 .listen = sock_no_listen,
4030 .shutdown = sock_no_shutdown,
4031 .setsockopt = sock_no_setsockopt,
4032 .getsockopt = sock_no_getsockopt,
4033 .sendmsg = packet_sendmsg_spkt,
4034 .recvmsg = packet_recvmsg,
4035 .mmap = sock_no_mmap,
4036 .sendpage = sock_no_sendpage,
4039 static const struct proto_ops packet_ops = {
4040 .family = PF_PACKET,
4041 .owner = THIS_MODULE,
4042 .release = packet_release,
4043 .bind = packet_bind,
4044 .connect = sock_no_connect,
4045 .socketpair = sock_no_socketpair,
4046 .accept = sock_no_accept,
4047 .getname = packet_getname,
4048 .poll = packet_poll,
4049 .ioctl = packet_ioctl,
4050 .listen = sock_no_listen,
4051 .shutdown = sock_no_shutdown,
4052 .setsockopt = packet_setsockopt,
4053 .getsockopt = packet_getsockopt,
4054 .sendmsg = packet_sendmsg,
4055 .recvmsg = packet_recvmsg,
4056 .mmap = packet_mmap,
4057 .sendpage = sock_no_sendpage,
4060 static const struct net_proto_family packet_family_ops = {
4061 .family = PF_PACKET,
4062 .create = packet_create,
4063 .owner = THIS_MODULE,
4066 static struct notifier_block packet_netdev_notifier = {
4067 .notifier_call = packet_notifier,
4070 #ifdef CONFIG_PROC_FS
4072 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4075 struct net *net = seq_file_net(seq);
4078 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4081 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4083 struct net *net = seq_file_net(seq);
4084 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4087 static void packet_seq_stop(struct seq_file *seq, void *v)
4093 static int packet_seq_show(struct seq_file *seq, void *v)
4095 if (v == SEQ_START_TOKEN)
4096 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4098 struct sock *s = sk_entry(v);
4099 const struct packet_sock *po = pkt_sk(s);
4102 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4104 atomic_read(&s->sk_refcnt),
4109 atomic_read(&s->sk_rmem_alloc),
4110 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4117 static const struct seq_operations packet_seq_ops = {
4118 .start = packet_seq_start,
4119 .next = packet_seq_next,
4120 .stop = packet_seq_stop,
4121 .show = packet_seq_show,
4124 static int packet_seq_open(struct inode *inode, struct file *file)
4126 return seq_open_net(inode, file, &packet_seq_ops,
4127 sizeof(struct seq_net_private));
4130 static const struct file_operations packet_seq_fops = {
4131 .owner = THIS_MODULE,
4132 .open = packet_seq_open,
4134 .llseek = seq_lseek,
4135 .release = seq_release_net,
4140 static int __net_init packet_net_init(struct net *net)
4142 mutex_init(&net->packet.sklist_lock);
4143 INIT_HLIST_HEAD(&net->packet.sklist);
4145 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4151 static void __net_exit packet_net_exit(struct net *net)
4153 remove_proc_entry("packet", net->proc_net);
4156 static struct pernet_operations packet_net_ops = {
4157 .init = packet_net_init,
4158 .exit = packet_net_exit,
4162 static void __exit packet_exit(void)
4164 unregister_netdevice_notifier(&packet_netdev_notifier);
4165 unregister_pernet_subsys(&packet_net_ops);
4166 sock_unregister(PF_PACKET);
4167 proto_unregister(&packet_proto);
4170 static int __init packet_init(void)
4172 int rc = proto_register(&packet_proto, 0);
4177 sock_register(&packet_family_ops);
4178 register_pernet_subsys(&packet_net_ops);
4179 register_netdevice_notifier(&packet_netdev_notifier);
4184 module_init(packet_init);
4185 module_exit(packet_exit);
4186 MODULE_LICENSE("GPL");
4187 MODULE_ALIAS_NETPROTO(PF_PACKET);