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, bool try_self,
1324 struct packet_sock *po;
1327 po = pkt_sk(f->arr[idx]);
1328 if (try_self && packet_rcv_has_room(po, skb))
1331 i = j = min_t(int, po->rollover->sock, num - 1);
1333 if (i != idx && packet_rcv_has_room(pkt_sk(f->arr[i]), skb)) {
1335 po->rollover->sock = i;
1346 static unsigned int fanout_demux_qm(struct packet_fanout *f,
1347 struct sk_buff *skb,
1350 return skb_get_queue_mapping(skb) % num;
1353 static bool fanout_has_flag(struct packet_fanout *f, u16 flag)
1355 return f->flags & (flag >> 8);
1358 static int packet_rcv_fanout(struct sk_buff *skb, struct net_device *dev,
1359 struct packet_type *pt, struct net_device *orig_dev)
1361 struct packet_fanout *f = pt->af_packet_priv;
1362 unsigned int num = f->num_members;
1363 struct packet_sock *po;
1366 if (!net_eq(dev_net(dev), read_pnet(&f->net)) ||
1372 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_DEFRAG)) {
1373 skb = ip_check_defrag(skb, IP_DEFRAG_AF_PACKET);
1378 case PACKET_FANOUT_HASH:
1380 idx = fanout_demux_hash(f, skb, num);
1382 case PACKET_FANOUT_LB:
1383 idx = fanout_demux_lb(f, skb, num);
1385 case PACKET_FANOUT_CPU:
1386 idx = fanout_demux_cpu(f, skb, num);
1388 case PACKET_FANOUT_RND:
1389 idx = fanout_demux_rnd(f, skb, num);
1391 case PACKET_FANOUT_QM:
1392 idx = fanout_demux_qm(f, skb, num);
1394 case PACKET_FANOUT_ROLLOVER:
1395 idx = fanout_demux_rollover(f, skb, 0, false, num);
1399 if (fanout_has_flag(f, PACKET_FANOUT_FLAG_ROLLOVER))
1400 idx = fanout_demux_rollover(f, skb, idx, true, num);
1402 po = pkt_sk(f->arr[idx]);
1403 return po->prot_hook.func(skb, dev, &po->prot_hook, orig_dev);
1406 DEFINE_MUTEX(fanout_mutex);
1407 EXPORT_SYMBOL_GPL(fanout_mutex);
1408 static LIST_HEAD(fanout_list);
1410 static void __fanout_link(struct sock *sk, struct packet_sock *po)
1412 struct packet_fanout *f = po->fanout;
1414 spin_lock(&f->lock);
1415 f->arr[f->num_members] = sk;
1418 spin_unlock(&f->lock);
1421 static void __fanout_unlink(struct sock *sk, struct packet_sock *po)
1423 struct packet_fanout *f = po->fanout;
1426 spin_lock(&f->lock);
1427 for (i = 0; i < f->num_members; i++) {
1428 if (f->arr[i] == sk)
1431 BUG_ON(i >= f->num_members);
1432 f->arr[i] = f->arr[f->num_members - 1];
1434 spin_unlock(&f->lock);
1437 static bool match_fanout_group(struct packet_type *ptype, struct sock *sk)
1439 if (ptype->af_packet_priv == (void *)((struct packet_sock *)sk)->fanout)
1445 static int fanout_add(struct sock *sk, u16 id, u16 type_flags)
1447 struct packet_sock *po = pkt_sk(sk);
1448 struct packet_fanout *f, *match;
1449 u8 type = type_flags & 0xff;
1450 u8 flags = type_flags >> 8;
1454 case PACKET_FANOUT_ROLLOVER:
1455 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER)
1457 case PACKET_FANOUT_HASH:
1458 case PACKET_FANOUT_LB:
1459 case PACKET_FANOUT_CPU:
1460 case PACKET_FANOUT_RND:
1461 case PACKET_FANOUT_QM:
1473 if (type_flags & PACKET_FANOUT_FLAG_ROLLOVER) {
1474 po->rollover = kzalloc(sizeof(*po->rollover), GFP_KERNEL);
1479 mutex_lock(&fanout_mutex);
1481 list_for_each_entry(f, &fanout_list, list) {
1483 read_pnet(&f->net) == sock_net(sk)) {
1489 if (match && match->flags != flags)
1493 match = kzalloc(sizeof(*match), GFP_KERNEL);
1496 write_pnet(&match->net, sock_net(sk));
1499 match->flags = flags;
1500 atomic_set(&match->rr_cur, 0);
1501 INIT_LIST_HEAD(&match->list);
1502 spin_lock_init(&match->lock);
1503 atomic_set(&match->sk_ref, 0);
1504 match->prot_hook.type = po->prot_hook.type;
1505 match->prot_hook.dev = po->prot_hook.dev;
1506 match->prot_hook.func = packet_rcv_fanout;
1507 match->prot_hook.af_packet_priv = match;
1508 match->prot_hook.id_match = match_fanout_group;
1509 dev_add_pack(&match->prot_hook);
1510 list_add(&match->list, &fanout_list);
1513 if (match->type == type &&
1514 match->prot_hook.type == po->prot_hook.type &&
1515 match->prot_hook.dev == po->prot_hook.dev) {
1517 if (atomic_read(&match->sk_ref) < PACKET_FANOUT_MAX) {
1518 __dev_remove_pack(&po->prot_hook);
1520 atomic_inc(&match->sk_ref);
1521 __fanout_link(sk, po);
1526 mutex_unlock(&fanout_mutex);
1528 kfree(po->rollover);
1529 po->rollover = NULL;
1534 static void fanout_release(struct sock *sk)
1536 struct packet_sock *po = pkt_sk(sk);
1537 struct packet_fanout *f;
1543 mutex_lock(&fanout_mutex);
1546 if (atomic_dec_and_test(&f->sk_ref)) {
1548 dev_remove_pack(&f->prot_hook);
1551 mutex_unlock(&fanout_mutex);
1553 kfree(po->rollover);
1556 static const struct proto_ops packet_ops;
1558 static const struct proto_ops packet_ops_spkt;
1560 static int packet_rcv_spkt(struct sk_buff *skb, struct net_device *dev,
1561 struct packet_type *pt, struct net_device *orig_dev)
1564 struct sockaddr_pkt *spkt;
1567 * When we registered the protocol we saved the socket in the data
1568 * field for just this event.
1571 sk = pt->af_packet_priv;
1574 * Yank back the headers [hope the device set this
1575 * right or kerboom...]
1577 * Incoming packets have ll header pulled,
1580 * For outgoing ones skb->data == skb_mac_header(skb)
1581 * so that this procedure is noop.
1584 if (skb->pkt_type == PACKET_LOOPBACK)
1587 if (!net_eq(dev_net(dev), sock_net(sk)))
1590 skb = skb_share_check(skb, GFP_ATOMIC);
1594 /* drop any routing info */
1597 /* drop conntrack reference */
1600 spkt = &PACKET_SKB_CB(skb)->sa.pkt;
1602 skb_push(skb, skb->data - skb_mac_header(skb));
1605 * The SOCK_PACKET socket receives _all_ frames.
1608 spkt->spkt_family = dev->type;
1609 strlcpy(spkt->spkt_device, dev->name, sizeof(spkt->spkt_device));
1610 spkt->spkt_protocol = skb->protocol;
1613 * Charge the memory to the socket. This is done specifically
1614 * to prevent sockets using all the memory up.
1617 if (sock_queue_rcv_skb(sk, skb) == 0)
1628 * Output a raw packet to a device layer. This bypasses all the other
1629 * protocol layers and you must therefore supply it with a complete frame
1632 static int packet_sendmsg_spkt(struct socket *sock, struct msghdr *msg,
1635 struct sock *sk = sock->sk;
1636 DECLARE_SOCKADDR(struct sockaddr_pkt *, saddr, msg->msg_name);
1637 struct sk_buff *skb = NULL;
1638 struct net_device *dev;
1644 * Get and verify the address.
1648 if (msg->msg_namelen < sizeof(struct sockaddr))
1650 if (msg->msg_namelen == sizeof(struct sockaddr_pkt))
1651 proto = saddr->spkt_protocol;
1653 return -ENOTCONN; /* SOCK_PACKET must be sent giving an address */
1656 * Find the device first to size check it
1659 saddr->spkt_device[sizeof(saddr->spkt_device) - 1] = 0;
1662 dev = dev_get_by_name_rcu(sock_net(sk), saddr->spkt_device);
1668 if (!(dev->flags & IFF_UP))
1672 * You may not queue a frame bigger than the mtu. This is the lowest level
1673 * raw protocol and you must do your own fragmentation at this level.
1676 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
1677 if (!netif_supports_nofcs(dev)) {
1678 err = -EPROTONOSUPPORT;
1681 extra_len = 4; /* We're doing our own CRC */
1685 if (len > dev->mtu + dev->hard_header_len + VLAN_HLEN + extra_len)
1689 size_t reserved = LL_RESERVED_SPACE(dev);
1690 int tlen = dev->needed_tailroom;
1691 unsigned int hhlen = dev->header_ops ? dev->hard_header_len : 0;
1694 skb = sock_wmalloc(sk, len + reserved + tlen, 0, GFP_KERNEL);
1697 /* FIXME: Save some space for broken drivers that write a hard
1698 * header at transmission time by themselves. PPP is the notable
1699 * one here. This should really be fixed at the driver level.
1701 skb_reserve(skb, reserved);
1702 skb_reset_network_header(skb);
1704 /* Try to align data part correctly */
1709 skb_reset_network_header(skb);
1711 err = memcpy_from_msg(skb_put(skb, len), msg, len);
1717 if (len > (dev->mtu + dev->hard_header_len + extra_len)) {
1718 /* Earlier code assumed this would be a VLAN pkt,
1719 * double-check this now that we have the actual
1722 struct ethhdr *ehdr;
1723 skb_reset_mac_header(skb);
1724 ehdr = eth_hdr(skb);
1725 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
1731 skb->protocol = proto;
1733 skb->priority = sk->sk_priority;
1734 skb->mark = sk->sk_mark;
1736 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
1738 if (unlikely(extra_len == 4))
1741 skb_probe_transport_header(skb, 0);
1743 dev_queue_xmit(skb);
1754 static unsigned int run_filter(const struct sk_buff *skb,
1755 const struct sock *sk,
1758 struct sk_filter *filter;
1761 filter = rcu_dereference(sk->sk_filter);
1763 res = SK_RUN_FILTER(filter, skb);
1770 * This function makes lazy skb cloning in hope that most of packets
1771 * are discarded by BPF.
1773 * Note tricky part: we DO mangle shared skb! skb->data, skb->len
1774 * and skb->cb are mangled. It works because (and until) packets
1775 * falling here are owned by current CPU. Output packets are cloned
1776 * by dev_queue_xmit_nit(), input packets are processed by net_bh
1777 * sequencially, so that if we return skb to original state on exit,
1778 * we will not harm anyone.
1781 static int packet_rcv(struct sk_buff *skb, struct net_device *dev,
1782 struct packet_type *pt, struct net_device *orig_dev)
1785 struct sockaddr_ll *sll;
1786 struct packet_sock *po;
1787 u8 *skb_head = skb->data;
1788 int skb_len = skb->len;
1789 unsigned int snaplen, res;
1791 if (skb->pkt_type == PACKET_LOOPBACK)
1794 sk = pt->af_packet_priv;
1797 if (!net_eq(dev_net(dev), sock_net(sk)))
1802 if (dev->header_ops) {
1803 /* The device has an explicit notion of ll header,
1804 * exported to higher levels.
1806 * Otherwise, the device hides details of its frame
1807 * structure, so that corresponding packet head is
1808 * never delivered to user.
1810 if (sk->sk_type != SOCK_DGRAM)
1811 skb_push(skb, skb->data - skb_mac_header(skb));
1812 else if (skb->pkt_type == PACKET_OUTGOING) {
1813 /* Special case: outgoing packets have ll header at head */
1814 skb_pull(skb, skb_network_offset(skb));
1820 res = run_filter(skb, sk, snaplen);
1822 goto drop_n_restore;
1826 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf)
1829 if (skb_shared(skb)) {
1830 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1834 if (skb_head != skb->data) {
1835 skb->data = skb_head;
1842 sock_skb_cb_check_size(sizeof(*PACKET_SKB_CB(skb)) + MAX_ADDR_LEN - 8);
1844 sll = &PACKET_SKB_CB(skb)->sa.ll;
1845 sll->sll_hatype = dev->type;
1846 sll->sll_pkttype = skb->pkt_type;
1847 if (unlikely(po->origdev))
1848 sll->sll_ifindex = orig_dev->ifindex;
1850 sll->sll_ifindex = dev->ifindex;
1852 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
1854 /* sll->sll_family and sll->sll_protocol are set in packet_recvmsg().
1855 * Use their space for storing the original skb length.
1857 PACKET_SKB_CB(skb)->sa.origlen = skb->len;
1859 if (pskb_trim(skb, snaplen))
1862 skb_set_owner_r(skb, sk);
1866 /* drop conntrack reference */
1869 spin_lock(&sk->sk_receive_queue.lock);
1870 po->stats.stats1.tp_packets++;
1871 sock_skb_set_dropcount(sk, skb);
1872 __skb_queue_tail(&sk->sk_receive_queue, skb);
1873 spin_unlock(&sk->sk_receive_queue.lock);
1874 sk->sk_data_ready(sk);
1878 spin_lock(&sk->sk_receive_queue.lock);
1879 po->stats.stats1.tp_drops++;
1880 atomic_inc(&sk->sk_drops);
1881 spin_unlock(&sk->sk_receive_queue.lock);
1884 if (skb_head != skb->data && skb_shared(skb)) {
1885 skb->data = skb_head;
1893 static int tpacket_rcv(struct sk_buff *skb, struct net_device *dev,
1894 struct packet_type *pt, struct net_device *orig_dev)
1897 struct packet_sock *po;
1898 struct sockaddr_ll *sll;
1899 union tpacket_uhdr h;
1900 u8 *skb_head = skb->data;
1901 int skb_len = skb->len;
1902 unsigned int snaplen, res;
1903 unsigned long status = TP_STATUS_USER;
1904 unsigned short macoff, netoff, hdrlen;
1905 struct sk_buff *copy_skb = NULL;
1909 /* struct tpacket{2,3}_hdr is aligned to a multiple of TPACKET_ALIGNMENT.
1910 * We may add members to them until current aligned size without forcing
1911 * userspace to call getsockopt(..., PACKET_HDRLEN, ...).
1913 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h2)) != 32);
1914 BUILD_BUG_ON(TPACKET_ALIGN(sizeof(*h.h3)) != 48);
1916 if (skb->pkt_type == PACKET_LOOPBACK)
1919 sk = pt->af_packet_priv;
1922 if (!net_eq(dev_net(dev), sock_net(sk)))
1925 if (dev->header_ops) {
1926 if (sk->sk_type != SOCK_DGRAM)
1927 skb_push(skb, skb->data - skb_mac_header(skb));
1928 else if (skb->pkt_type == PACKET_OUTGOING) {
1929 /* Special case: outgoing packets have ll header at head */
1930 skb_pull(skb, skb_network_offset(skb));
1936 res = run_filter(skb, sk, snaplen);
1938 goto drop_n_restore;
1940 if (skb->ip_summed == CHECKSUM_PARTIAL)
1941 status |= TP_STATUS_CSUMNOTREADY;
1942 else if (skb->pkt_type != PACKET_OUTGOING &&
1943 (skb->ip_summed == CHECKSUM_COMPLETE ||
1944 skb_csum_unnecessary(skb)))
1945 status |= TP_STATUS_CSUM_VALID;
1950 if (sk->sk_type == SOCK_DGRAM) {
1951 macoff = netoff = TPACKET_ALIGN(po->tp_hdrlen) + 16 +
1954 unsigned int maclen = skb_network_offset(skb);
1955 netoff = TPACKET_ALIGN(po->tp_hdrlen +
1956 (maclen < 16 ? 16 : maclen)) +
1958 macoff = netoff - maclen;
1960 if (po->tp_version <= TPACKET_V2) {
1961 if (macoff + snaplen > po->rx_ring.frame_size) {
1962 if (po->copy_thresh &&
1963 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1964 if (skb_shared(skb)) {
1965 copy_skb = skb_clone(skb, GFP_ATOMIC);
1967 copy_skb = skb_get(skb);
1968 skb_head = skb->data;
1971 skb_set_owner_r(copy_skb, sk);
1973 snaplen = po->rx_ring.frame_size - macoff;
1974 if ((int)snaplen < 0)
1977 } else if (unlikely(macoff + snaplen >
1978 GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len)) {
1981 nval = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len - macoff;
1982 pr_err_once("tpacket_rcv: packet too big, clamped from %u to %u. macoff=%u\n",
1983 snaplen, nval, macoff);
1985 if (unlikely((int)snaplen < 0)) {
1987 macoff = GET_PBDQC_FROM_RB(&po->rx_ring)->max_frame_len;
1990 spin_lock(&sk->sk_receive_queue.lock);
1991 h.raw = packet_current_rx_frame(po, skb,
1992 TP_STATUS_KERNEL, (macoff+snaplen));
1995 if (po->tp_version <= TPACKET_V2) {
1996 packet_increment_rx_head(po, &po->rx_ring);
1998 * LOSING will be reported till you read the stats,
1999 * because it's COR - Clear On Read.
2000 * Anyways, moving it for V1/V2 only as V3 doesn't need this
2003 if (po->stats.stats1.tp_drops)
2004 status |= TP_STATUS_LOSING;
2006 po->stats.stats1.tp_packets++;
2008 status |= TP_STATUS_COPY;
2009 __skb_queue_tail(&sk->sk_receive_queue, copy_skb);
2011 spin_unlock(&sk->sk_receive_queue.lock);
2013 skb_copy_bits(skb, 0, h.raw + macoff, snaplen);
2015 if (!(ts_status = tpacket_get_timestamp(skb, &ts, po->tp_tstamp)))
2016 getnstimeofday(&ts);
2018 status |= ts_status;
2020 switch (po->tp_version) {
2022 h.h1->tp_len = skb->len;
2023 h.h1->tp_snaplen = snaplen;
2024 h.h1->tp_mac = macoff;
2025 h.h1->tp_net = netoff;
2026 h.h1->tp_sec = ts.tv_sec;
2027 h.h1->tp_usec = ts.tv_nsec / NSEC_PER_USEC;
2028 hdrlen = sizeof(*h.h1);
2031 h.h2->tp_len = skb->len;
2032 h.h2->tp_snaplen = snaplen;
2033 h.h2->tp_mac = macoff;
2034 h.h2->tp_net = netoff;
2035 h.h2->tp_sec = ts.tv_sec;
2036 h.h2->tp_nsec = ts.tv_nsec;
2037 if (skb_vlan_tag_present(skb)) {
2038 h.h2->tp_vlan_tci = skb_vlan_tag_get(skb);
2039 h.h2->tp_vlan_tpid = ntohs(skb->vlan_proto);
2040 status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
2042 h.h2->tp_vlan_tci = 0;
2043 h.h2->tp_vlan_tpid = 0;
2045 memset(h.h2->tp_padding, 0, sizeof(h.h2->tp_padding));
2046 hdrlen = sizeof(*h.h2);
2049 /* tp_nxt_offset,vlan are already populated above.
2050 * So DONT clear those fields here
2052 h.h3->tp_status |= status;
2053 h.h3->tp_len = skb->len;
2054 h.h3->tp_snaplen = snaplen;
2055 h.h3->tp_mac = macoff;
2056 h.h3->tp_net = netoff;
2057 h.h3->tp_sec = ts.tv_sec;
2058 h.h3->tp_nsec = ts.tv_nsec;
2059 memset(h.h3->tp_padding, 0, sizeof(h.h3->tp_padding));
2060 hdrlen = sizeof(*h.h3);
2066 sll = h.raw + TPACKET_ALIGN(hdrlen);
2067 sll->sll_halen = dev_parse_header(skb, sll->sll_addr);
2068 sll->sll_family = AF_PACKET;
2069 sll->sll_hatype = dev->type;
2070 sll->sll_protocol = skb->protocol;
2071 sll->sll_pkttype = skb->pkt_type;
2072 if (unlikely(po->origdev))
2073 sll->sll_ifindex = orig_dev->ifindex;
2075 sll->sll_ifindex = dev->ifindex;
2079 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE == 1
2080 if (po->tp_version <= TPACKET_V2) {
2083 end = (u8 *) PAGE_ALIGN((unsigned long) h.raw +
2086 for (start = h.raw; start < end; start += PAGE_SIZE)
2087 flush_dcache_page(pgv_to_page(start));
2092 if (po->tp_version <= TPACKET_V2) {
2093 __packet_set_status(po, h.raw, status);
2094 sk->sk_data_ready(sk);
2096 prb_clear_blk_fill_status(&po->rx_ring);
2100 if (skb_head != skb->data && skb_shared(skb)) {
2101 skb->data = skb_head;
2109 po->stats.stats1.tp_drops++;
2110 spin_unlock(&sk->sk_receive_queue.lock);
2112 sk->sk_data_ready(sk);
2113 kfree_skb(copy_skb);
2114 goto drop_n_restore;
2117 static void tpacket_destruct_skb(struct sk_buff *skb)
2119 struct packet_sock *po = pkt_sk(skb->sk);
2121 if (likely(po->tx_ring.pg_vec)) {
2125 ph = skb_shinfo(skb)->destructor_arg;
2126 packet_dec_pending(&po->tx_ring);
2128 ts = __packet_set_timestamp(po, ph, skb);
2129 __packet_set_status(po, ph, TP_STATUS_AVAILABLE | ts);
2135 static bool ll_header_truncated(const struct net_device *dev, int len)
2137 /* net device doesn't like empty head */
2138 if (unlikely(len <= dev->hard_header_len)) {
2139 net_warn_ratelimited("%s: packet size is too short (%d <= %d)\n",
2140 current->comm, len, dev->hard_header_len);
2147 static int tpacket_fill_skb(struct packet_sock *po, struct sk_buff *skb,
2148 void *frame, struct net_device *dev, int size_max,
2149 __be16 proto, unsigned char *addr, int hlen)
2151 union tpacket_uhdr ph;
2152 int to_write, offset, len, tp_len, nr_frags, len_max;
2153 struct socket *sock = po->sk.sk_socket;
2160 skb->protocol = proto;
2162 skb->priority = po->sk.sk_priority;
2163 skb->mark = po->sk.sk_mark;
2164 sock_tx_timestamp(&po->sk, &skb_shinfo(skb)->tx_flags);
2165 skb_shinfo(skb)->destructor_arg = ph.raw;
2167 switch (po->tp_version) {
2169 tp_len = ph.h2->tp_len;
2172 tp_len = ph.h1->tp_len;
2175 if (unlikely(tp_len > size_max)) {
2176 pr_err("packet size is too long (%d > %d)\n", tp_len, size_max);
2180 skb_reserve(skb, hlen);
2181 skb_reset_network_header(skb);
2183 if (!packet_use_direct_xmit(po))
2184 skb_probe_transport_header(skb, 0);
2185 if (unlikely(po->tp_tx_has_off)) {
2186 int off_min, off_max, off;
2187 off_min = po->tp_hdrlen - sizeof(struct sockaddr_ll);
2188 off_max = po->tx_ring.frame_size - tp_len;
2189 if (sock->type == SOCK_DGRAM) {
2190 switch (po->tp_version) {
2192 off = ph.h2->tp_net;
2195 off = ph.h1->tp_net;
2199 switch (po->tp_version) {
2201 off = ph.h2->tp_mac;
2204 off = ph.h1->tp_mac;
2208 if (unlikely((off < off_min) || (off_max < off)))
2210 data = ph.raw + off;
2212 data = ph.raw + po->tp_hdrlen - sizeof(struct sockaddr_ll);
2216 if (sock->type == SOCK_DGRAM) {
2217 err = dev_hard_header(skb, dev, ntohs(proto), addr,
2219 if (unlikely(err < 0))
2221 } else if (dev->hard_header_len) {
2222 if (ll_header_truncated(dev, tp_len))
2225 skb_push(skb, dev->hard_header_len);
2226 err = skb_store_bits(skb, 0, data,
2227 dev->hard_header_len);
2231 data += dev->hard_header_len;
2232 to_write -= dev->hard_header_len;
2235 offset = offset_in_page(data);
2236 len_max = PAGE_SIZE - offset;
2237 len = ((to_write > len_max) ? len_max : to_write);
2239 skb->data_len = to_write;
2240 skb->len += to_write;
2241 skb->truesize += to_write;
2242 atomic_add(to_write, &po->sk.sk_wmem_alloc);
2244 while (likely(to_write)) {
2245 nr_frags = skb_shinfo(skb)->nr_frags;
2247 if (unlikely(nr_frags >= MAX_SKB_FRAGS)) {
2248 pr_err("Packet exceed the number of skb frags(%lu)\n",
2253 page = pgv_to_page(data);
2255 flush_dcache_page(page);
2257 skb_fill_page_desc(skb, nr_frags, page, offset, len);
2260 len_max = PAGE_SIZE;
2261 len = ((to_write > len_max) ? len_max : to_write);
2267 static int tpacket_snd(struct packet_sock *po, struct msghdr *msg)
2269 struct sk_buff *skb;
2270 struct net_device *dev;
2272 int err, reserve = 0;
2274 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2275 bool need_wait = !(msg->msg_flags & MSG_DONTWAIT);
2276 int tp_len, size_max;
2277 unsigned char *addr;
2279 int status = TP_STATUS_AVAILABLE;
2282 mutex_lock(&po->pg_vec_lock);
2284 if (likely(saddr == NULL)) {
2285 dev = packet_cached_dev_get(po);
2290 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2292 if (msg->msg_namelen < (saddr->sll_halen
2293 + offsetof(struct sockaddr_ll,
2296 proto = saddr->sll_protocol;
2297 addr = saddr->sll_addr;
2298 dev = dev_get_by_index(sock_net(&po->sk), saddr->sll_ifindex);
2302 if (unlikely(dev == NULL))
2305 if (unlikely(!(dev->flags & IFF_UP)))
2308 reserve = dev->hard_header_len + VLAN_HLEN;
2309 size_max = po->tx_ring.frame_size
2310 - (po->tp_hdrlen - sizeof(struct sockaddr_ll));
2312 if (size_max > dev->mtu + reserve)
2313 size_max = dev->mtu + reserve;
2316 ph = packet_current_frame(po, &po->tx_ring,
2317 TP_STATUS_SEND_REQUEST);
2318 if (unlikely(ph == NULL)) {
2319 if (need_wait && need_resched())
2324 status = TP_STATUS_SEND_REQUEST;
2325 hlen = LL_RESERVED_SPACE(dev);
2326 tlen = dev->needed_tailroom;
2327 skb = sock_alloc_send_skb(&po->sk,
2328 hlen + tlen + sizeof(struct sockaddr_ll),
2331 if (unlikely(skb == NULL)) {
2332 /* we assume the socket was initially writeable ... */
2333 if (likely(len_sum > 0))
2337 tp_len = tpacket_fill_skb(po, skb, ph, dev, size_max, proto,
2339 if (tp_len > dev->mtu + dev->hard_header_len) {
2340 struct ethhdr *ehdr;
2341 /* Earlier code assumed this would be a VLAN pkt,
2342 * double-check this now that we have the actual
2346 skb_reset_mac_header(skb);
2347 ehdr = eth_hdr(skb);
2348 if (ehdr->h_proto != htons(ETH_P_8021Q))
2351 if (unlikely(tp_len < 0)) {
2353 __packet_set_status(po, ph,
2354 TP_STATUS_AVAILABLE);
2355 packet_increment_head(&po->tx_ring);
2359 status = TP_STATUS_WRONG_FORMAT;
2365 packet_pick_tx_queue(dev, skb);
2367 skb->destructor = tpacket_destruct_skb;
2368 __packet_set_status(po, ph, TP_STATUS_SENDING);
2369 packet_inc_pending(&po->tx_ring);
2371 status = TP_STATUS_SEND_REQUEST;
2372 err = po->xmit(skb);
2373 if (unlikely(err > 0)) {
2374 err = net_xmit_errno(err);
2375 if (err && __packet_get_status(po, ph) ==
2376 TP_STATUS_AVAILABLE) {
2377 /* skb was destructed already */
2382 * skb was dropped but not destructed yet;
2383 * let's treat it like congestion or err < 0
2387 packet_increment_head(&po->tx_ring);
2389 } while (likely((ph != NULL) ||
2390 /* Note: packet_read_pending() might be slow if we have
2391 * to call it as it's per_cpu variable, but in fast-path
2392 * we already short-circuit the loop with the first
2393 * condition, and luckily don't have to go that path
2396 (need_wait && packet_read_pending(&po->tx_ring))));
2402 __packet_set_status(po, ph, status);
2407 mutex_unlock(&po->pg_vec_lock);
2411 static struct sk_buff *packet_alloc_skb(struct sock *sk, size_t prepad,
2412 size_t reserve, size_t len,
2413 size_t linear, int noblock,
2416 struct sk_buff *skb;
2418 /* Under a page? Don't bother with paged skb. */
2419 if (prepad + len < PAGE_SIZE || !linear)
2422 skb = sock_alloc_send_pskb(sk, prepad + linear, len - linear, noblock,
2427 skb_reserve(skb, reserve);
2428 skb_put(skb, linear);
2429 skb->data_len = len - linear;
2430 skb->len += len - linear;
2435 static int packet_snd(struct socket *sock, struct msghdr *msg, size_t len)
2437 struct sock *sk = sock->sk;
2438 DECLARE_SOCKADDR(struct sockaddr_ll *, saddr, msg->msg_name);
2439 struct sk_buff *skb;
2440 struct net_device *dev;
2442 unsigned char *addr;
2443 int err, reserve = 0;
2444 struct virtio_net_hdr vnet_hdr = { 0 };
2447 struct packet_sock *po = pkt_sk(sk);
2448 unsigned short gso_type = 0;
2454 * Get and verify the address.
2457 if (likely(saddr == NULL)) {
2458 dev = packet_cached_dev_get(po);
2463 if (msg->msg_namelen < sizeof(struct sockaddr_ll))
2465 if (msg->msg_namelen < (saddr->sll_halen + offsetof(struct sockaddr_ll, sll_addr)))
2467 proto = saddr->sll_protocol;
2468 addr = saddr->sll_addr;
2469 dev = dev_get_by_index(sock_net(sk), saddr->sll_ifindex);
2473 if (unlikely(dev == NULL))
2476 if (unlikely(!(dev->flags & IFF_UP)))
2479 if (sock->type == SOCK_RAW)
2480 reserve = dev->hard_header_len;
2481 if (po->has_vnet_hdr) {
2482 vnet_hdr_len = sizeof(vnet_hdr);
2485 if (len < vnet_hdr_len)
2488 len -= vnet_hdr_len;
2491 n = copy_from_iter(&vnet_hdr, vnet_hdr_len, &msg->msg_iter);
2492 if (n != vnet_hdr_len)
2495 if ((vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) &&
2496 (__virtio16_to_cpu(false, vnet_hdr.csum_start) +
2497 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2 >
2498 __virtio16_to_cpu(false, vnet_hdr.hdr_len)))
2499 vnet_hdr.hdr_len = __cpu_to_virtio16(false,
2500 __virtio16_to_cpu(false, vnet_hdr.csum_start) +
2501 __virtio16_to_cpu(false, vnet_hdr.csum_offset) + 2);
2504 if (__virtio16_to_cpu(false, vnet_hdr.hdr_len) > len)
2507 if (vnet_hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
2508 switch (vnet_hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
2509 case VIRTIO_NET_HDR_GSO_TCPV4:
2510 gso_type = SKB_GSO_TCPV4;
2512 case VIRTIO_NET_HDR_GSO_TCPV6:
2513 gso_type = SKB_GSO_TCPV6;
2515 case VIRTIO_NET_HDR_GSO_UDP:
2516 gso_type = SKB_GSO_UDP;
2522 if (vnet_hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
2523 gso_type |= SKB_GSO_TCP_ECN;
2525 if (vnet_hdr.gso_size == 0)
2531 if (unlikely(sock_flag(sk, SOCK_NOFCS))) {
2532 if (!netif_supports_nofcs(dev)) {
2533 err = -EPROTONOSUPPORT;
2536 extra_len = 4; /* We're doing our own CRC */
2540 if (!gso_type && (len > dev->mtu + reserve + VLAN_HLEN + extra_len))
2544 hlen = LL_RESERVED_SPACE(dev);
2545 tlen = dev->needed_tailroom;
2546 skb = packet_alloc_skb(sk, hlen + tlen, hlen, len,
2547 __virtio16_to_cpu(false, vnet_hdr.hdr_len),
2548 msg->msg_flags & MSG_DONTWAIT, &err);
2552 skb_set_network_header(skb, reserve);
2555 if (sock->type == SOCK_DGRAM) {
2556 offset = dev_hard_header(skb, dev, ntohs(proto), addr, NULL, len);
2557 if (unlikely(offset < 0))
2560 if (ll_header_truncated(dev, len))
2564 /* Returns -EFAULT on error */
2565 err = skb_copy_datagram_from_iter(skb, offset, &msg->msg_iter, len);
2569 sock_tx_timestamp(sk, &skb_shinfo(skb)->tx_flags);
2571 if (!gso_type && (len > dev->mtu + reserve + extra_len)) {
2572 /* Earlier code assumed this would be a VLAN pkt,
2573 * double-check this now that we have the actual
2576 struct ethhdr *ehdr;
2577 skb_reset_mac_header(skb);
2578 ehdr = eth_hdr(skb);
2579 if (ehdr->h_proto != htons(ETH_P_8021Q)) {
2585 skb->protocol = proto;
2587 skb->priority = sk->sk_priority;
2588 skb->mark = sk->sk_mark;
2590 packet_pick_tx_queue(dev, skb);
2592 if (po->has_vnet_hdr) {
2593 if (vnet_hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
2594 u16 s = __virtio16_to_cpu(false, vnet_hdr.csum_start);
2595 u16 o = __virtio16_to_cpu(false, vnet_hdr.csum_offset);
2596 if (!skb_partial_csum_set(skb, s, o)) {
2602 skb_shinfo(skb)->gso_size =
2603 __virtio16_to_cpu(false, vnet_hdr.gso_size);
2604 skb_shinfo(skb)->gso_type = gso_type;
2606 /* Header must be checked, and gso_segs computed. */
2607 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2608 skb_shinfo(skb)->gso_segs = 0;
2610 len += vnet_hdr_len;
2613 if (!packet_use_direct_xmit(po))
2614 skb_probe_transport_header(skb, reserve);
2615 if (unlikely(extra_len == 4))
2618 err = po->xmit(skb);
2619 if (err > 0 && (err = net_xmit_errno(err)) != 0)
2635 static int packet_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2637 struct sock *sk = sock->sk;
2638 struct packet_sock *po = pkt_sk(sk);
2640 if (po->tx_ring.pg_vec)
2641 return tpacket_snd(po, msg);
2643 return packet_snd(sock, msg, len);
2647 * Close a PACKET socket. This is fairly simple. We immediately go
2648 * to 'closed' state and remove our protocol entry in the device list.
2651 static int packet_release(struct socket *sock)
2653 struct sock *sk = sock->sk;
2654 struct packet_sock *po;
2656 union tpacket_req_u req_u;
2664 mutex_lock(&net->packet.sklist_lock);
2665 sk_del_node_init_rcu(sk);
2666 mutex_unlock(&net->packet.sklist_lock);
2669 sock_prot_inuse_add(net, sk->sk_prot, -1);
2672 spin_lock(&po->bind_lock);
2673 unregister_prot_hook(sk, false);
2674 packet_cached_dev_reset(po);
2676 if (po->prot_hook.dev) {
2677 dev_put(po->prot_hook.dev);
2678 po->prot_hook.dev = NULL;
2680 spin_unlock(&po->bind_lock);
2682 packet_flush_mclist(sk);
2684 if (po->rx_ring.pg_vec) {
2685 memset(&req_u, 0, sizeof(req_u));
2686 packet_set_ring(sk, &req_u, 1, 0);
2689 if (po->tx_ring.pg_vec) {
2690 memset(&req_u, 0, sizeof(req_u));
2691 packet_set_ring(sk, &req_u, 1, 1);
2698 * Now the socket is dead. No more input will appear.
2705 skb_queue_purge(&sk->sk_receive_queue);
2706 packet_free_pending(po);
2707 sk_refcnt_debug_release(sk);
2714 * Attach a packet hook.
2717 static int packet_do_bind(struct sock *sk, struct net_device *dev, __be16 proto)
2719 struct packet_sock *po = pkt_sk(sk);
2720 const struct net_device *dev_curr;
2732 spin_lock(&po->bind_lock);
2734 proto_curr = po->prot_hook.type;
2735 dev_curr = po->prot_hook.dev;
2737 need_rehook = proto_curr != proto || dev_curr != dev;
2740 unregister_prot_hook(sk, true);
2743 po->prot_hook.type = proto;
2745 if (po->prot_hook.dev)
2746 dev_put(po->prot_hook.dev);
2748 po->prot_hook.dev = dev;
2750 po->ifindex = dev ? dev->ifindex : 0;
2751 packet_cached_dev_assign(po, dev);
2754 if (proto == 0 || !need_rehook)
2757 if (!dev || (dev->flags & IFF_UP)) {
2758 register_prot_hook(sk);
2760 sk->sk_err = ENETDOWN;
2761 if (!sock_flag(sk, SOCK_DEAD))
2762 sk->sk_error_report(sk);
2766 spin_unlock(&po->bind_lock);
2772 * Bind a packet socket to a device
2775 static int packet_bind_spkt(struct socket *sock, struct sockaddr *uaddr,
2778 struct sock *sk = sock->sk;
2780 struct net_device *dev;
2787 if (addr_len != sizeof(struct sockaddr))
2789 strlcpy(name, uaddr->sa_data, sizeof(name));
2791 dev = dev_get_by_name(sock_net(sk), name);
2793 err = packet_do_bind(sk, dev, pkt_sk(sk)->num);
2797 static int packet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
2799 struct sockaddr_ll *sll = (struct sockaddr_ll *)uaddr;
2800 struct sock *sk = sock->sk;
2801 struct net_device *dev = NULL;
2809 if (addr_len < sizeof(struct sockaddr_ll))
2811 if (sll->sll_family != AF_PACKET)
2814 if (sll->sll_ifindex) {
2816 dev = dev_get_by_index(sock_net(sk), sll->sll_ifindex);
2820 err = packet_do_bind(sk, dev, sll->sll_protocol ? : pkt_sk(sk)->num);
2826 static struct proto packet_proto = {
2828 .owner = THIS_MODULE,
2829 .obj_size = sizeof(struct packet_sock),
2833 * Create a packet of type SOCK_PACKET.
2836 static int packet_create(struct net *net, struct socket *sock, int protocol,
2840 struct packet_sock *po;
2841 __be16 proto = (__force __be16)protocol; /* weird, but documented */
2844 if (!ns_capable(net->user_ns, CAP_NET_RAW))
2846 if (sock->type != SOCK_DGRAM && sock->type != SOCK_RAW &&
2847 sock->type != SOCK_PACKET)
2848 return -ESOCKTNOSUPPORT;
2850 sock->state = SS_UNCONNECTED;
2853 sk = sk_alloc(net, PF_PACKET, GFP_KERNEL, &packet_proto, kern);
2857 sock->ops = &packet_ops;
2858 if (sock->type == SOCK_PACKET)
2859 sock->ops = &packet_ops_spkt;
2861 sock_init_data(sock, sk);
2864 sk->sk_family = PF_PACKET;
2866 po->xmit = dev_queue_xmit;
2868 err = packet_alloc_pending(po);
2872 packet_cached_dev_reset(po);
2874 sk->sk_destruct = packet_sock_destruct;
2875 sk_refcnt_debug_inc(sk);
2878 * Attach a protocol block
2881 spin_lock_init(&po->bind_lock);
2882 mutex_init(&po->pg_vec_lock);
2883 po->rollover = NULL;
2884 po->prot_hook.func = packet_rcv;
2886 if (sock->type == SOCK_PACKET)
2887 po->prot_hook.func = packet_rcv_spkt;
2889 po->prot_hook.af_packet_priv = sk;
2892 po->prot_hook.type = proto;
2893 register_prot_hook(sk);
2896 mutex_lock(&net->packet.sklist_lock);
2897 sk_add_node_rcu(sk, &net->packet.sklist);
2898 mutex_unlock(&net->packet.sklist_lock);
2901 sock_prot_inuse_add(net, &packet_proto, 1);
2912 * Pull a packet from our receive queue and hand it to the user.
2913 * If necessary we block.
2916 static int packet_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2919 struct sock *sk = sock->sk;
2920 struct sk_buff *skb;
2922 int vnet_hdr_len = 0;
2923 unsigned int origlen = 0;
2926 if (flags & ~(MSG_PEEK|MSG_DONTWAIT|MSG_TRUNC|MSG_CMSG_COMPAT|MSG_ERRQUEUE))
2930 /* What error should we return now? EUNATTACH? */
2931 if (pkt_sk(sk)->ifindex < 0)
2935 if (flags & MSG_ERRQUEUE) {
2936 err = sock_recv_errqueue(sk, msg, len,
2937 SOL_PACKET, PACKET_TX_TIMESTAMP);
2942 * Call the generic datagram receiver. This handles all sorts
2943 * of horrible races and re-entrancy so we can forget about it
2944 * in the protocol layers.
2946 * Now it will return ENETDOWN, if device have just gone down,
2947 * but then it will block.
2950 skb = skb_recv_datagram(sk, flags, flags & MSG_DONTWAIT, &err);
2953 * An error occurred so return it. Because skb_recv_datagram()
2954 * handles the blocking we don't see and worry about blocking
2961 if (pkt_sk(sk)->has_vnet_hdr) {
2962 struct virtio_net_hdr vnet_hdr = { 0 };
2965 vnet_hdr_len = sizeof(vnet_hdr);
2966 if (len < vnet_hdr_len)
2969 len -= vnet_hdr_len;
2971 if (skb_is_gso(skb)) {
2972 struct skb_shared_info *sinfo = skb_shinfo(skb);
2974 /* This is a hint as to how much should be linear. */
2976 __cpu_to_virtio16(false, skb_headlen(skb));
2978 __cpu_to_virtio16(false, sinfo->gso_size);
2979 if (sinfo->gso_type & SKB_GSO_TCPV4)
2980 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
2981 else if (sinfo->gso_type & SKB_GSO_TCPV6)
2982 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
2983 else if (sinfo->gso_type & SKB_GSO_UDP)
2984 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP;
2985 else if (sinfo->gso_type & SKB_GSO_FCOE)
2989 if (sinfo->gso_type & SKB_GSO_TCP_ECN)
2990 vnet_hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN;
2992 vnet_hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE;
2994 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2995 vnet_hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
2996 vnet_hdr.csum_start = __cpu_to_virtio16(false,
2997 skb_checksum_start_offset(skb));
2998 vnet_hdr.csum_offset = __cpu_to_virtio16(false,
3000 } else if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3001 vnet_hdr.flags = VIRTIO_NET_HDR_F_DATA_VALID;
3002 } /* else everything is zero */
3004 err = memcpy_to_msg(msg, (void *)&vnet_hdr, vnet_hdr_len);
3009 /* You lose any data beyond the buffer you gave. If it worries
3010 * a user program they can ask the device for its MTU
3016 msg->msg_flags |= MSG_TRUNC;
3019 err = skb_copy_datagram_msg(skb, 0, msg, copied);
3023 if (sock->type != SOCK_PACKET) {
3024 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3026 /* Original length was stored in sockaddr_ll fields */
3027 origlen = PACKET_SKB_CB(skb)->sa.origlen;
3028 sll->sll_family = AF_PACKET;
3029 sll->sll_protocol = skb->protocol;
3032 sock_recv_ts_and_drops(msg, sk, skb);
3034 if (msg->msg_name) {
3035 /* If the address length field is there to be filled
3036 * in, we fill it in now.
3038 if (sock->type == SOCK_PACKET) {
3039 __sockaddr_check_size(sizeof(struct sockaddr_pkt));
3040 msg->msg_namelen = sizeof(struct sockaddr_pkt);
3042 struct sockaddr_ll *sll = &PACKET_SKB_CB(skb)->sa.ll;
3044 msg->msg_namelen = sll->sll_halen +
3045 offsetof(struct sockaddr_ll, sll_addr);
3047 memcpy(msg->msg_name, &PACKET_SKB_CB(skb)->sa,
3051 if (pkt_sk(sk)->auxdata) {
3052 struct tpacket_auxdata aux;
3054 aux.tp_status = TP_STATUS_USER;
3055 if (skb->ip_summed == CHECKSUM_PARTIAL)
3056 aux.tp_status |= TP_STATUS_CSUMNOTREADY;
3057 else if (skb->pkt_type != PACKET_OUTGOING &&
3058 (skb->ip_summed == CHECKSUM_COMPLETE ||
3059 skb_csum_unnecessary(skb)))
3060 aux.tp_status |= TP_STATUS_CSUM_VALID;
3062 aux.tp_len = origlen;
3063 aux.tp_snaplen = skb->len;
3065 aux.tp_net = skb_network_offset(skb);
3066 if (skb_vlan_tag_present(skb)) {
3067 aux.tp_vlan_tci = skb_vlan_tag_get(skb);
3068 aux.tp_vlan_tpid = ntohs(skb->vlan_proto);
3069 aux.tp_status |= TP_STATUS_VLAN_VALID | TP_STATUS_VLAN_TPID_VALID;
3071 aux.tp_vlan_tci = 0;
3072 aux.tp_vlan_tpid = 0;
3074 put_cmsg(msg, SOL_PACKET, PACKET_AUXDATA, sizeof(aux), &aux);
3078 * Free or return the buffer as appropriate. Again this
3079 * hides all the races and re-entrancy issues from us.
3081 err = vnet_hdr_len + ((flags&MSG_TRUNC) ? skb->len : copied);
3084 skb_free_datagram(sk, skb);
3089 static int packet_getname_spkt(struct socket *sock, struct sockaddr *uaddr,
3090 int *uaddr_len, int peer)
3092 struct net_device *dev;
3093 struct sock *sk = sock->sk;
3098 uaddr->sa_family = AF_PACKET;
3099 memset(uaddr->sa_data, 0, sizeof(uaddr->sa_data));
3101 dev = dev_get_by_index_rcu(sock_net(sk), pkt_sk(sk)->ifindex);
3103 strlcpy(uaddr->sa_data, dev->name, sizeof(uaddr->sa_data));
3105 *uaddr_len = sizeof(*uaddr);
3110 static int packet_getname(struct socket *sock, struct sockaddr *uaddr,
3111 int *uaddr_len, int peer)
3113 struct net_device *dev;
3114 struct sock *sk = sock->sk;
3115 struct packet_sock *po = pkt_sk(sk);
3116 DECLARE_SOCKADDR(struct sockaddr_ll *, sll, uaddr);
3121 sll->sll_family = AF_PACKET;
3122 sll->sll_ifindex = po->ifindex;
3123 sll->sll_protocol = po->num;
3124 sll->sll_pkttype = 0;
3126 dev = dev_get_by_index_rcu(sock_net(sk), po->ifindex);
3128 sll->sll_hatype = dev->type;
3129 sll->sll_halen = dev->addr_len;
3130 memcpy(sll->sll_addr, dev->dev_addr, dev->addr_len);
3132 sll->sll_hatype = 0; /* Bad: we have no ARPHRD_UNSPEC */
3136 *uaddr_len = offsetof(struct sockaddr_ll, sll_addr) + sll->sll_halen;
3141 static int packet_dev_mc(struct net_device *dev, struct packet_mclist *i,
3145 case PACKET_MR_MULTICAST:
3146 if (i->alen != dev->addr_len)
3149 return dev_mc_add(dev, i->addr);
3151 return dev_mc_del(dev, i->addr);
3153 case PACKET_MR_PROMISC:
3154 return dev_set_promiscuity(dev, what);
3155 case PACKET_MR_ALLMULTI:
3156 return dev_set_allmulti(dev, what);
3157 case PACKET_MR_UNICAST:
3158 if (i->alen != dev->addr_len)
3161 return dev_uc_add(dev, i->addr);
3163 return dev_uc_del(dev, i->addr);
3171 static void packet_dev_mclist_delete(struct net_device *dev,
3172 struct packet_mclist **mlp)
3174 struct packet_mclist *ml;
3176 while ((ml = *mlp) != NULL) {
3177 if (ml->ifindex == dev->ifindex) {
3178 packet_dev_mc(dev, ml, -1);
3186 static int packet_mc_add(struct sock *sk, struct packet_mreq_max *mreq)
3188 struct packet_sock *po = pkt_sk(sk);
3189 struct packet_mclist *ml, *i;
3190 struct net_device *dev;
3196 dev = __dev_get_by_index(sock_net(sk), mreq->mr_ifindex);
3201 if (mreq->mr_alen > dev->addr_len)
3205 i = kmalloc(sizeof(*i), GFP_KERNEL);
3210 for (ml = po->mclist; ml; ml = ml->next) {
3211 if (ml->ifindex == mreq->mr_ifindex &&
3212 ml->type == mreq->mr_type &&
3213 ml->alen == mreq->mr_alen &&
3214 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3216 /* Free the new element ... */
3222 i->type = mreq->mr_type;
3223 i->ifindex = mreq->mr_ifindex;
3224 i->alen = mreq->mr_alen;
3225 memcpy(i->addr, mreq->mr_address, i->alen);
3227 i->next = po->mclist;
3229 err = packet_dev_mc(dev, i, 1);
3231 po->mclist = i->next;
3240 static int packet_mc_drop(struct sock *sk, struct packet_mreq_max *mreq)
3242 struct packet_mclist *ml, **mlp;
3246 for (mlp = &pkt_sk(sk)->mclist; (ml = *mlp) != NULL; mlp = &ml->next) {
3247 if (ml->ifindex == mreq->mr_ifindex &&
3248 ml->type == mreq->mr_type &&
3249 ml->alen == mreq->mr_alen &&
3250 memcmp(ml->addr, mreq->mr_address, ml->alen) == 0) {
3251 if (--ml->count == 0) {
3252 struct net_device *dev;
3254 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3256 packet_dev_mc(dev, ml, -1);
3266 static void packet_flush_mclist(struct sock *sk)
3268 struct packet_sock *po = pkt_sk(sk);
3269 struct packet_mclist *ml;
3275 while ((ml = po->mclist) != NULL) {
3276 struct net_device *dev;
3278 po->mclist = ml->next;
3279 dev = __dev_get_by_index(sock_net(sk), ml->ifindex);
3281 packet_dev_mc(dev, ml, -1);
3288 packet_setsockopt(struct socket *sock, int level, int optname, char __user *optval, unsigned int optlen)
3290 struct sock *sk = sock->sk;
3291 struct packet_sock *po = pkt_sk(sk);
3294 if (level != SOL_PACKET)
3295 return -ENOPROTOOPT;
3298 case PACKET_ADD_MEMBERSHIP:
3299 case PACKET_DROP_MEMBERSHIP:
3301 struct packet_mreq_max mreq;
3303 memset(&mreq, 0, sizeof(mreq));
3304 if (len < sizeof(struct packet_mreq))
3306 if (len > sizeof(mreq))
3308 if (copy_from_user(&mreq, optval, len))
3310 if (len < (mreq.mr_alen + offsetof(struct packet_mreq, mr_address)))
3312 if (optname == PACKET_ADD_MEMBERSHIP)
3313 ret = packet_mc_add(sk, &mreq);
3315 ret = packet_mc_drop(sk, &mreq);
3319 case PACKET_RX_RING:
3320 case PACKET_TX_RING:
3322 union tpacket_req_u req_u;
3325 switch (po->tp_version) {
3328 len = sizeof(req_u.req);
3332 len = sizeof(req_u.req3);
3337 if (pkt_sk(sk)->has_vnet_hdr)
3339 if (copy_from_user(&req_u.req, optval, len))
3341 return packet_set_ring(sk, &req_u, 0,
3342 optname == PACKET_TX_RING);
3344 case PACKET_COPY_THRESH:
3348 if (optlen != sizeof(val))
3350 if (copy_from_user(&val, optval, sizeof(val)))
3353 pkt_sk(sk)->copy_thresh = val;
3356 case PACKET_VERSION:
3360 if (optlen != sizeof(val))
3362 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3364 if (copy_from_user(&val, optval, sizeof(val)))
3370 po->tp_version = val;
3376 case PACKET_RESERVE:
3380 if (optlen != sizeof(val))
3382 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3384 if (copy_from_user(&val, optval, sizeof(val)))
3386 po->tp_reserve = val;
3393 if (optlen != sizeof(val))
3395 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3397 if (copy_from_user(&val, optval, sizeof(val)))
3399 po->tp_loss = !!val;
3402 case PACKET_AUXDATA:
3406 if (optlen < sizeof(val))
3408 if (copy_from_user(&val, optval, sizeof(val)))
3411 po->auxdata = !!val;
3414 case PACKET_ORIGDEV:
3418 if (optlen < sizeof(val))
3420 if (copy_from_user(&val, optval, sizeof(val)))
3423 po->origdev = !!val;
3426 case PACKET_VNET_HDR:
3430 if (sock->type != SOCK_RAW)
3432 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3434 if (optlen < sizeof(val))
3436 if (copy_from_user(&val, optval, sizeof(val)))
3439 po->has_vnet_hdr = !!val;
3442 case PACKET_TIMESTAMP:
3446 if (optlen != sizeof(val))
3448 if (copy_from_user(&val, optval, sizeof(val)))
3451 po->tp_tstamp = val;
3458 if (optlen != sizeof(val))
3460 if (copy_from_user(&val, optval, sizeof(val)))
3463 return fanout_add(sk, val & 0xffff, val >> 16);
3465 case PACKET_TX_HAS_OFF:
3469 if (optlen != sizeof(val))
3471 if (po->rx_ring.pg_vec || po->tx_ring.pg_vec)
3473 if (copy_from_user(&val, optval, sizeof(val)))
3475 po->tp_tx_has_off = !!val;
3478 case PACKET_QDISC_BYPASS:
3482 if (optlen != sizeof(val))
3484 if (copy_from_user(&val, optval, sizeof(val)))
3487 po->xmit = val ? packet_direct_xmit : dev_queue_xmit;
3491 return -ENOPROTOOPT;
3495 static int packet_getsockopt(struct socket *sock, int level, int optname,
3496 char __user *optval, int __user *optlen)
3499 int val, lv = sizeof(val);
3500 struct sock *sk = sock->sk;
3501 struct packet_sock *po = pkt_sk(sk);
3503 union tpacket_stats_u st;
3505 if (level != SOL_PACKET)
3506 return -ENOPROTOOPT;
3508 if (get_user(len, optlen))
3515 case PACKET_STATISTICS:
3516 spin_lock_bh(&sk->sk_receive_queue.lock);
3517 memcpy(&st, &po->stats, sizeof(st));
3518 memset(&po->stats, 0, sizeof(po->stats));
3519 spin_unlock_bh(&sk->sk_receive_queue.lock);
3521 if (po->tp_version == TPACKET_V3) {
3522 lv = sizeof(struct tpacket_stats_v3);
3523 st.stats3.tp_packets += st.stats3.tp_drops;
3526 lv = sizeof(struct tpacket_stats);
3527 st.stats1.tp_packets += st.stats1.tp_drops;
3532 case PACKET_AUXDATA:
3535 case PACKET_ORIGDEV:
3538 case PACKET_VNET_HDR:
3539 val = po->has_vnet_hdr;
3541 case PACKET_VERSION:
3542 val = po->tp_version;
3545 if (len > sizeof(int))
3547 if (copy_from_user(&val, optval, len))
3551 val = sizeof(struct tpacket_hdr);
3554 val = sizeof(struct tpacket2_hdr);
3557 val = sizeof(struct tpacket3_hdr);
3563 case PACKET_RESERVE:
3564 val = po->tp_reserve;
3569 case PACKET_TIMESTAMP:
3570 val = po->tp_tstamp;
3574 ((u32)po->fanout->id |
3575 ((u32)po->fanout->type << 16) |
3576 ((u32)po->fanout->flags << 24)) :
3579 case PACKET_TX_HAS_OFF:
3580 val = po->tp_tx_has_off;
3582 case PACKET_QDISC_BYPASS:
3583 val = packet_use_direct_xmit(po);
3586 return -ENOPROTOOPT;
3591 if (put_user(len, optlen))
3593 if (copy_to_user(optval, data, len))
3599 static int packet_notifier(struct notifier_block *this,
3600 unsigned long msg, void *ptr)
3603 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
3604 struct net *net = dev_net(dev);
3607 sk_for_each_rcu(sk, &net->packet.sklist) {
3608 struct packet_sock *po = pkt_sk(sk);
3611 case NETDEV_UNREGISTER:
3613 packet_dev_mclist_delete(dev, &po->mclist);
3617 if (dev->ifindex == po->ifindex) {
3618 spin_lock(&po->bind_lock);
3620 __unregister_prot_hook(sk, false);
3621 sk->sk_err = ENETDOWN;
3622 if (!sock_flag(sk, SOCK_DEAD))
3623 sk->sk_error_report(sk);
3625 if (msg == NETDEV_UNREGISTER) {
3626 packet_cached_dev_reset(po);
3628 if (po->prot_hook.dev)
3629 dev_put(po->prot_hook.dev);
3630 po->prot_hook.dev = NULL;
3632 spin_unlock(&po->bind_lock);
3636 if (dev->ifindex == po->ifindex) {
3637 spin_lock(&po->bind_lock);
3639 register_prot_hook(sk);
3640 spin_unlock(&po->bind_lock);
3650 static int packet_ioctl(struct socket *sock, unsigned int cmd,
3653 struct sock *sk = sock->sk;
3658 int amount = sk_wmem_alloc_get(sk);
3660 return put_user(amount, (int __user *)arg);
3664 struct sk_buff *skb;
3667 spin_lock_bh(&sk->sk_receive_queue.lock);
3668 skb = skb_peek(&sk->sk_receive_queue);
3671 spin_unlock_bh(&sk->sk_receive_queue.lock);
3672 return put_user(amount, (int __user *)arg);
3675 return sock_get_timestamp(sk, (struct timeval __user *)arg);
3677 return sock_get_timestampns(sk, (struct timespec __user *)arg);
3687 case SIOCGIFBRDADDR:
3688 case SIOCSIFBRDADDR:
3689 case SIOCGIFNETMASK:
3690 case SIOCSIFNETMASK:
3691 case SIOCGIFDSTADDR:
3692 case SIOCSIFDSTADDR:
3694 return inet_dgram_ops.ioctl(sock, cmd, arg);
3698 return -ENOIOCTLCMD;
3703 static unsigned int packet_poll(struct file *file, struct socket *sock,
3706 struct sock *sk = sock->sk;
3707 struct packet_sock *po = pkt_sk(sk);
3708 unsigned int mask = datagram_poll(file, sock, wait);
3710 spin_lock_bh(&sk->sk_receive_queue.lock);
3711 if (po->rx_ring.pg_vec) {
3712 if (!packet_previous_rx_frame(po, &po->rx_ring,
3714 mask |= POLLIN | POLLRDNORM;
3716 spin_unlock_bh(&sk->sk_receive_queue.lock);
3717 spin_lock_bh(&sk->sk_write_queue.lock);
3718 if (po->tx_ring.pg_vec) {
3719 if (packet_current_frame(po, &po->tx_ring, TP_STATUS_AVAILABLE))
3720 mask |= POLLOUT | POLLWRNORM;
3722 spin_unlock_bh(&sk->sk_write_queue.lock);
3727 /* Dirty? Well, I still did not learn better way to account
3731 static void packet_mm_open(struct vm_area_struct *vma)
3733 struct file *file = vma->vm_file;
3734 struct socket *sock = file->private_data;
3735 struct sock *sk = sock->sk;
3738 atomic_inc(&pkt_sk(sk)->mapped);
3741 static void packet_mm_close(struct vm_area_struct *vma)
3743 struct file *file = vma->vm_file;
3744 struct socket *sock = file->private_data;
3745 struct sock *sk = sock->sk;
3748 atomic_dec(&pkt_sk(sk)->mapped);
3751 static const struct vm_operations_struct packet_mmap_ops = {
3752 .open = packet_mm_open,
3753 .close = packet_mm_close,
3756 static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
3761 for (i = 0; i < len; i++) {
3762 if (likely(pg_vec[i].buffer)) {
3763 if (is_vmalloc_addr(pg_vec[i].buffer))
3764 vfree(pg_vec[i].buffer);
3766 free_pages((unsigned long)pg_vec[i].buffer,
3768 pg_vec[i].buffer = NULL;
3774 static char *alloc_one_pg_vec_page(unsigned long order)
3777 gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
3778 __GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
3780 buffer = (char *) __get_free_pages(gfp_flags, order);
3784 /* __get_free_pages failed, fall back to vmalloc */
3785 buffer = vzalloc((1 << order) * PAGE_SIZE);
3789 /* vmalloc failed, lets dig into swap here */
3790 gfp_flags &= ~__GFP_NORETRY;
3791 buffer = (char *) __get_free_pages(gfp_flags, order);
3795 /* complete and utter failure */
3799 static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
3801 unsigned int block_nr = req->tp_block_nr;
3805 pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
3806 if (unlikely(!pg_vec))
3809 for (i = 0; i < block_nr; i++) {
3810 pg_vec[i].buffer = alloc_one_pg_vec_page(order);
3811 if (unlikely(!pg_vec[i].buffer))
3812 goto out_free_pgvec;
3819 free_pg_vec(pg_vec, order, block_nr);
3824 static int packet_set_ring(struct sock *sk, union tpacket_req_u *req_u,
3825 int closing, int tx_ring)
3827 struct pgv *pg_vec = NULL;
3828 struct packet_sock *po = pkt_sk(sk);
3829 int was_running, order = 0;
3830 struct packet_ring_buffer *rb;
3831 struct sk_buff_head *rb_queue;
3834 /* Added to avoid minimal code churn */
3835 struct tpacket_req *req = &req_u->req;
3837 /* Opening a Tx-ring is NOT supported in TPACKET_V3 */
3838 if (!closing && tx_ring && (po->tp_version > TPACKET_V2)) {
3839 WARN(1, "Tx-ring is not supported.\n");
3843 rb = tx_ring ? &po->tx_ring : &po->rx_ring;
3844 rb_queue = tx_ring ? &sk->sk_write_queue : &sk->sk_receive_queue;
3848 if (atomic_read(&po->mapped))
3850 if (packet_read_pending(rb))
3854 if (req->tp_block_nr) {
3855 /* Sanity tests and some calculations */
3857 if (unlikely(rb->pg_vec))
3860 switch (po->tp_version) {
3862 po->tp_hdrlen = TPACKET_HDRLEN;
3865 po->tp_hdrlen = TPACKET2_HDRLEN;
3868 po->tp_hdrlen = TPACKET3_HDRLEN;
3873 if (unlikely((int)req->tp_block_size <= 0))
3875 if (unlikely(req->tp_block_size & (PAGE_SIZE - 1)))
3877 if (po->tp_version >= TPACKET_V3 &&
3878 (int)(req->tp_block_size -
3879 BLK_PLUS_PRIV(req_u->req3.tp_sizeof_priv)) <= 0)
3881 if (unlikely(req->tp_frame_size < po->tp_hdrlen +
3884 if (unlikely(req->tp_frame_size & (TPACKET_ALIGNMENT - 1)))
3887 rb->frames_per_block = req->tp_block_size/req->tp_frame_size;
3888 if (unlikely(rb->frames_per_block <= 0))
3890 if (unlikely((rb->frames_per_block * req->tp_block_nr) !=
3895 order = get_order(req->tp_block_size);
3896 pg_vec = alloc_pg_vec(req, order);
3897 if (unlikely(!pg_vec))
3899 switch (po->tp_version) {
3901 /* Transmit path is not supported. We checked
3902 * it above but just being paranoid
3905 init_prb_bdqc(po, rb, pg_vec, req_u, tx_ring);
3914 if (unlikely(req->tp_frame_nr))
3920 /* Detach socket from network */
3921 spin_lock(&po->bind_lock);
3922 was_running = po->running;
3926 __unregister_prot_hook(sk, false);
3928 spin_unlock(&po->bind_lock);
3933 mutex_lock(&po->pg_vec_lock);
3934 if (closing || atomic_read(&po->mapped) == 0) {
3936 spin_lock_bh(&rb_queue->lock);
3937 swap(rb->pg_vec, pg_vec);
3938 rb->frame_max = (req->tp_frame_nr - 1);
3940 rb->frame_size = req->tp_frame_size;
3941 spin_unlock_bh(&rb_queue->lock);
3943 swap(rb->pg_vec_order, order);
3944 swap(rb->pg_vec_len, req->tp_block_nr);
3946 rb->pg_vec_pages = req->tp_block_size/PAGE_SIZE;
3947 po->prot_hook.func = (po->rx_ring.pg_vec) ?
3948 tpacket_rcv : packet_rcv;
3949 skb_queue_purge(rb_queue);
3950 if (atomic_read(&po->mapped))
3951 pr_err("packet_mmap: vma is busy: %d\n",
3952 atomic_read(&po->mapped));
3954 mutex_unlock(&po->pg_vec_lock);
3956 spin_lock(&po->bind_lock);
3959 register_prot_hook(sk);
3961 spin_unlock(&po->bind_lock);
3962 if (closing && (po->tp_version > TPACKET_V2)) {
3963 /* Because we don't support block-based V3 on tx-ring */
3965 prb_shutdown_retire_blk_timer(po, tx_ring, rb_queue);
3970 free_pg_vec(pg_vec, order, req->tp_block_nr);
3975 static int packet_mmap(struct file *file, struct socket *sock,
3976 struct vm_area_struct *vma)
3978 struct sock *sk = sock->sk;
3979 struct packet_sock *po = pkt_sk(sk);
3980 unsigned long size, expected_size;
3981 struct packet_ring_buffer *rb;
3982 unsigned long start;
3989 mutex_lock(&po->pg_vec_lock);
3992 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
3994 expected_size += rb->pg_vec_len
4000 if (expected_size == 0)
4003 size = vma->vm_end - vma->vm_start;
4004 if (size != expected_size)
4007 start = vma->vm_start;
4008 for (rb = &po->rx_ring; rb <= &po->tx_ring; rb++) {
4009 if (rb->pg_vec == NULL)
4012 for (i = 0; i < rb->pg_vec_len; i++) {
4014 void *kaddr = rb->pg_vec[i].buffer;
4017 for (pg_num = 0; pg_num < rb->pg_vec_pages; pg_num++) {
4018 page = pgv_to_page(kaddr);
4019 err = vm_insert_page(vma, start, page);
4028 atomic_inc(&po->mapped);
4029 vma->vm_ops = &packet_mmap_ops;
4033 mutex_unlock(&po->pg_vec_lock);
4037 static const struct proto_ops packet_ops_spkt = {
4038 .family = PF_PACKET,
4039 .owner = THIS_MODULE,
4040 .release = packet_release,
4041 .bind = packet_bind_spkt,
4042 .connect = sock_no_connect,
4043 .socketpair = sock_no_socketpair,
4044 .accept = sock_no_accept,
4045 .getname = packet_getname_spkt,
4046 .poll = datagram_poll,
4047 .ioctl = packet_ioctl,
4048 .listen = sock_no_listen,
4049 .shutdown = sock_no_shutdown,
4050 .setsockopt = sock_no_setsockopt,
4051 .getsockopt = sock_no_getsockopt,
4052 .sendmsg = packet_sendmsg_spkt,
4053 .recvmsg = packet_recvmsg,
4054 .mmap = sock_no_mmap,
4055 .sendpage = sock_no_sendpage,
4058 static const struct proto_ops packet_ops = {
4059 .family = PF_PACKET,
4060 .owner = THIS_MODULE,
4061 .release = packet_release,
4062 .bind = packet_bind,
4063 .connect = sock_no_connect,
4064 .socketpair = sock_no_socketpair,
4065 .accept = sock_no_accept,
4066 .getname = packet_getname,
4067 .poll = packet_poll,
4068 .ioctl = packet_ioctl,
4069 .listen = sock_no_listen,
4070 .shutdown = sock_no_shutdown,
4071 .setsockopt = packet_setsockopt,
4072 .getsockopt = packet_getsockopt,
4073 .sendmsg = packet_sendmsg,
4074 .recvmsg = packet_recvmsg,
4075 .mmap = packet_mmap,
4076 .sendpage = sock_no_sendpage,
4079 static const struct net_proto_family packet_family_ops = {
4080 .family = PF_PACKET,
4081 .create = packet_create,
4082 .owner = THIS_MODULE,
4085 static struct notifier_block packet_netdev_notifier = {
4086 .notifier_call = packet_notifier,
4089 #ifdef CONFIG_PROC_FS
4091 static void *packet_seq_start(struct seq_file *seq, loff_t *pos)
4094 struct net *net = seq_file_net(seq);
4097 return seq_hlist_start_head_rcu(&net->packet.sklist, *pos);
4100 static void *packet_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4102 struct net *net = seq_file_net(seq);
4103 return seq_hlist_next_rcu(v, &net->packet.sklist, pos);
4106 static void packet_seq_stop(struct seq_file *seq, void *v)
4112 static int packet_seq_show(struct seq_file *seq, void *v)
4114 if (v == SEQ_START_TOKEN)
4115 seq_puts(seq, "sk RefCnt Type Proto Iface R Rmem User Inode\n");
4117 struct sock *s = sk_entry(v);
4118 const struct packet_sock *po = pkt_sk(s);
4121 "%pK %-6d %-4d %04x %-5d %1d %-6u %-6u %-6lu\n",
4123 atomic_read(&s->sk_refcnt),
4128 atomic_read(&s->sk_rmem_alloc),
4129 from_kuid_munged(seq_user_ns(seq), sock_i_uid(s)),
4136 static const struct seq_operations packet_seq_ops = {
4137 .start = packet_seq_start,
4138 .next = packet_seq_next,
4139 .stop = packet_seq_stop,
4140 .show = packet_seq_show,
4143 static int packet_seq_open(struct inode *inode, struct file *file)
4145 return seq_open_net(inode, file, &packet_seq_ops,
4146 sizeof(struct seq_net_private));
4149 static const struct file_operations packet_seq_fops = {
4150 .owner = THIS_MODULE,
4151 .open = packet_seq_open,
4153 .llseek = seq_lseek,
4154 .release = seq_release_net,
4159 static int __net_init packet_net_init(struct net *net)
4161 mutex_init(&net->packet.sklist_lock);
4162 INIT_HLIST_HEAD(&net->packet.sklist);
4164 if (!proc_create("packet", 0, net->proc_net, &packet_seq_fops))
4170 static void __net_exit packet_net_exit(struct net *net)
4172 remove_proc_entry("packet", net->proc_net);
4175 static struct pernet_operations packet_net_ops = {
4176 .init = packet_net_init,
4177 .exit = packet_net_exit,
4181 static void __exit packet_exit(void)
4183 unregister_netdevice_notifier(&packet_netdev_notifier);
4184 unregister_pernet_subsys(&packet_net_ops);
4185 sock_unregister(PF_PACKET);
4186 proto_unregister(&packet_proto);
4189 static int __init packet_init(void)
4191 int rc = proto_register(&packet_proto, 0);
4196 sock_register(&packet_family_ops);
4197 register_pernet_subsys(&packet_net_ops);
4198 register_netdevice_notifier(&packet_netdev_notifier);
4203 module_init(packet_init);
4204 module_exit(packet_exit);
4205 MODULE_LICENSE("GPL");
4206 MODULE_ALIAS_NETPROTO(PF_PACKET);